HIPPOCAMPAL PLACE FIELDS
This page intentionally left blank
Hippocampal Place Fields Relevance to Learning and Memo...
55 downloads
1094 Views
12MB Size
Report
This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form
HIPPOCAMPAL PLACE FIELDS
This page intentionally left blank
Hippocampal Place Fields Relevance to Learning and Memory
Edited by
Sheri J. Y. Mizumori
1 2008
1 Oxford University Press, Inc., publishes works that further Oxford University’s objective of excellence in research, scholarship, and education. Oxford New York Auckland Cape Town Dar es Salaam Hong Kong Karachi Kuala Lumpur Madrid Melbourne Mexico City Nairobi New Delhi Shanghai Taipei Toronto With offices in Argentina Austria Brazil Chile Czech Republic France Greece Guatemala Hungary Italy Japan Poland Portugal Singapore South Korea Switzerland Thailand Turkey Ukraine Vietnam
Copyright # 2008 by Sheri J. Y. Mizumori Published by Oxford University Press, Inc. 198 Madison Avenue, New York, New York 10016 www.oup.com Oxford is a registered trademark of Oxford University Press All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, without the prior permission of Oxford University Press. Library of Congress Cataloging-in-Publication Data Hippocampal place fields: relevance to learning and memory / edited by Sheri J.Y. Mizumori. p. cm. Includes bibliographical references and index. ISBN 978-0-19-532324-5 1. Hippocampus (Brain) 2. Learning—Physiological aspects. 3. Memory—Physiological aspects. I. Mizumori, Sheri Jane. [DNLM: 1. Hippocampus—physiology. 2. Learning—physiology. 3. Memory—physiology. 4. Space Perception. WL 314 H6655 2007] QP383.25.H52 2007 153.1'5—dc22 2007021138
987654321 Printed in China on acid-free paper
This book is dedicated to my beloved parents (Lillian and George) and family (Jim, Ryan, and Aiko).
This page intentionally left blank
Preface: A Place for Place Cells in Hippocampal-Dependent Memory?
The experiences of our lives, whether good or bad, continually shape or modify the knowledge base that we call our memories. These memories define who we are and which purposeful behaviors become engaged in any given situation. Without these memories, we have no personal identity or purpose. Therefore, studies of neural signaling mechanisms that underlie memory formation and retrieval reflect attempts to understand what are likely to be evolutionarily conserved and fundamental biological mechanisms of behavioral plasticity. Over the years, technological advances have generated waves of excitement as new discoveries are reported about the role that the hippocampus plays in defining our memories. The chapters of this book review and analyze the progress made during the wave of research that followed the initial report of place cells in the hippocampus (O’Keefe and Dostrovsky, 1971), a discovery based on the development of methods by which one can record extracellular signals of single neurons in freely behaving rats. An historical perspective (below) provided by Jim Ranck and John Kubie takes us back to those days of initial discovery. They remind us that what seems obvious to some is not always so obvious to others, and that new discoveries, especially those that challenge current dogma, are often met with strong resistance. Place cells exhibit high-frequency discharge when animals traverse specific locations in an environment, referred to as the cell’s place field. A common goal of the following chapters is to assess the extent to which the study of place cells has enlightened our view of
how hippocampus contributes to learning and memory. A currently popular hypothesis is that place fields represent context information (Part I: Place Cells and Spatial Context, and Part IV: Theoretical Significance of Place Fields). Although the definition and purpose of such context representations continues to be debated (see Chapters 1, 2, 3, and 11 as examples), exciting and new insights about the link between place fields and learning have been obtained by studying the relationship between place fields and hypothesized elements of the context analysis. As examples, place cells have been studied in terms of hippocampal subregional specializations (Chapters 6 and 20), whether the context analysis is guided by memory or attention (Chapters 2, 7, 13, 14, and 19), the extent to which place fields are directly related to ongoing behaviors of the animal (Chapters 9 and 12), and the role of novelty (i.e., match–mismatch comparisons) in hippocampal context analyses (Chapters 1, 2, 3, 4, 8, 11, and 21). Comparison of rodent place cell findings with hippocampal single-unit findings in nonhuman primates (Part II: Primate Hippocampus and Place Representation) supports the view that the wave of research started by O’Keefe and Dostrovsky (1971) deals with an evolutionarily conserved neural mechanism of behavioral plasticity. That is, primate research indicates that important adaptations may have evolved from the rodent-like system (Chapters 12, 13, and 14). Moreover, as suggested in Chapter 10, we may in the future see a strong link between human hippocampaldependent function and rodent place fields.
viii
PREFACE
Recent identification of context-relevant neural codes in brain areas that are connected to hippocampal place cells (Part III: Neural Systems Perspective on the Significance of Place Fields) leads the way to our understanding of how hippocampal processing interacts within a more broad neural circuit (Chapters 15, 16, 17, and 18). Such systems analyses are absolutely necessary to address many central issues, such as where location information is actually computed. Our understanding of the dynamic mnemonic code represented by place fields is further enhanced by studying how age-related changes in hippocampal place fields relate to age-related changes in neural plasticity mechanisms and behavior (Part V: Place Fields and Age-Related Changes in Memory; Chapters 22 and 23). The collective review and analysis of the literature that is provided by these chapters demonstrates the impressive detail of our current understanding of place fields. These chapters also offer provocative and exciting new insights in terms of how we should proceed in the near future to achieve greater closure on the relationship between place fields and learning and memory. An example of one such important issue is how place field information contributes to the functional architecture that defines context processing with the hippocampus. Another issue begging to be resolved is the familiar question of whether place fields represent more than information about the external environment. A significant amount of converging evidence suggests that place cells do in fact represent more than just spatial information, at least under many conditions. The current question, then, is what other information is reflected by place cell firing? Is it goal or reward information? Is it behavioral information? Is it both? More perplexing, is it the input of the contextdefining information that drives place fields, or is it the behavioral repertoire that is expected at different locations?
Although somewhat obvious, it must also be recognized that the mission to understand the relationship between place fields and learning or memory relies not only on the findings of place cell studies but also on the continued development of theories regarding the specific role of the hippocampus in learning and memory, and perhaps even reconsideration of our views on the organization of memories in the brain (see Chapter 1). As an example, if the hippocampus is central to the processing of episodic memories, does that mean that the hippocampus does not process semantic memories? Are these two kinds of memory mutually exclusive or interactive? Answers to such questions will have important implications for the interpretation of place field reorganization (or remapping) during learning. The existence of unresolved issues does not mean that we have not made progress. Rather, the existence of these complex issues clearly indicates that we as a field are closer to being able to answer the question of how hippocampal place fields relate to learning and memory. (Notice the assumption that place fields are in fact related to learning and memory. This assumption is based on the simple and consistent observation that location-selective firing is perhaps the most prominent and immediately present firing pattern by hippocampal neurons in freely behaving animals, regardless of the spatial or nonspatial nature of the task, and findings that place fields change dramatically when subjects engage in hippocampal-dependent learning.) In fact, it is anticipated that with continued growth in the sophistication of our behavioral and neurobiological methods, dramatic new insights into the dynamic nature of place cells are on the horizon. It is hoped that consideration of the theoretical and empirical evidence summarized in this volume will facilitate movement toward that next horizon, for this will bring us closer to a true understanding of the hippocampal contribution to learning and memory.
Acknowledgments
The work described in this book reflects the culmination of the tireless effort by generations of creative students of the behavioral neurosciences. We are all
grateful for their dedication. This particular volume was supported by NIMH grant 58755.
This page intentionally left blank
Contents
Contributors
xiii
Formation: Implications for Episodic Memory 73
Historical Perspective: Place Cells in Ann Arbor and Brooklyn xvii
PATRICIA E. SHARP
6. The Roles of Hippocampal Subfields in Processing Spatial Contexts of Events: Neurophysiological and Behavioral Analyses 82
JAMES B. RANCK, JR. AND JOHN KUBIE
I. Place Cells and Spatial Context
INAH LEE, RAYMOND P. KESNER, AND
1. The Hippocampus and Context Revisited 3
JAMES J. KNIERIM
7. Plasticity, Attention, and the Stabilization of Hippocampal Representations 107
LYNN NADEL
2. A Context for Hippocampal Place Cells during Learning 16
DAVID C. ROWLAND AND CLIFFORD G. KENTROS
SHERI J. Y. MIZUMORI
8. What Do Place Cells Tell Us about Learning to Associate and Learning to Segregate? 127
3. Context-Dependent Firing of Hippocampal Place Cells: Does It Underlie Memory? 44
EDUARD KELEMEN AND ANDRE´ A. FENTON
JAMES A. AINGE, PAUL A. DUDCHENKO,
9. Do Place Cells Guide Spatial Behaviors? 138
AND EMMA R. WOOD
4. The Place Cells—Cognitive Map or Memory System? 59
ETIENNE SAVE AND BRUNO POUCET
10. Place Cells Identify Hippocampus with Location-Specific Construction of Mental Images 150
KATHRYN J. JEFFERY
5. Context-Specific Versus Context-Invariant Spatial Coding in the Hippocampal
NEIL BURGESS AND CHRIS M. BIRD
xi
xii CONTENTS
11. Hippocampal Neuronal Activity and Memory: Should We Still Be Talking about Place Cells? 161 HOWARD EICHENBAUM
18. Spatial Decisions and Neuronal Activity in Hippocampal Projection Zones in Prefrontal Cortex and Striatum 289 FRANCESCO P. BATTAGLIA, ADRIEN PEYRACHE, MEHDI KHAMASSI, AND SIDNEY I. WIENER
II. Primate Hippocampus and Place Representation IV. Theoretical Significance of Place Fields 12. Place-Differential Neural Responses in the Monkey Hippocampal Formation during Real and Virtual Navigation 177 HISAO NISHIJO, ETSURO HORI,
19. Hippocampal Theta Rhythm and Memory-Guided Behavior
313
AMY L. GRIFFIN, HOWARD EICHENBAUM, AND MICHAEL E. HASSELMO
AND TAKETOSHI ONO
13. Spatial View Cells in the Primate Hippocampus, and Memory 192
20. Network Analysis of the Significance of Hippocampal Subfields 328 GERGELY PAPP AND ALESSANDRO TREVES
EDMUND T. ROLLS
14. Learning, Memory and the Monkey Hippocampus 218 WENDY A. SUZUKI
21. Storage of the Distance between Place Cell Firing Fields in the Strength of Plastic Synapses with a Novel Learning Rule 343 ¨ RGY CSIZMADIA AND ROBERT U. MULLER GYO
III. Neural Systems Perspective on the Significance of Place Fields 15. Entorhinal Grid Cells and the Neural Basis of Navigation 237 MARIANNE FYHN, TRYGVE SOLSTAD, AND TORKEL HAFTING
V. Place Fields and Age-Related Changes in Memory 22. Hippocampal Place Cells as a Window into Age-Related Memory Impairments 353 IAIN WILSON AND HEIKKI TANILA
16. Neocortical Influences on Hippocampal Place Cells 253 DAVID K. BILKEY
23. Aging Ensembles: Circuit Contributions to Memory Deficits SARA N. BURKE AND CAROL A. BARNES
17. Spatial Learning and the Selectivity of Hippocampal Place Fields: Modulation by Dopamine 271 KATHRYN M. GILL AND SHERI J. Y. MIZUMORI
Author Index
385
Subject Index
397
364
Contributors
James A. Ainge Department of Psychology University of Stirling Carol A. Barnes Neural Systems, Memory, and Aging Division University of Arizona Francesco P. Battaglia Graduate School of Neurosciences University of Amsterdam
State University of New York Downstate Medical Center Paul A. Dudchenko Department of Psychology University of Stirling Howard Eichenbaum Center for Memory and Brain Boston University
David K. Bilkey Department of Psychology University of Otago
Andre´ A. Fenton Department of Physiology and Pharmacology State University of New York Downstate Medical Center
Chris M. Bird Institute of Cognitive Neuroscience and Department of Anatomy University College London
Marianne Fyhn Centre for the Biology of Memory Norwegian University of Science and Technology
Neil Burgess Institute of Cognitive Neuroscience and Department of Anatomy University College London
Kathryn M. Gill University of Washington Department of Psychology
Sara N. Burke Neural Systems, Memory, and Aging Division University of Arizona Gyo¨rgy Csizmadia Department of Physiology and Pharmacology
Amy L. Griffin Center for Memory and Brain Boston University Torkel Hafting Centre for the Biology of Memory Norwegian University of Science and Technology xiii
xiv
CONTRIBUTORS
Michael E. Hasselmo Center for Memory and Brain Boston University
Hisao Nishijo Graduate School of Medicine University of Toyama
Etsuro Hori Graduate School of Medicine University of Toyama
Taketoshi Ono Graduate School of Medicine University of Toyama
Kathryn J. Jeffery Department of Psychology University College London
Gergely Papp Cognitive Neuroscience Sector Scoula Internazionale Superiore di Studi Avanzati (SISSA) International School for Advanced Studies
Eduard Kelemen Department of Physiology and Pharmacology State University of New York Downstate Medical Center Clifford G. Kentros Institute of Neuroscience University of Oregon Raymond P. Kesner Department of Psychology University of Utah Mehdi Khamassi Laboratoire de Physiologie de la Perception et de l’Action CNRS Colle`ge de France James J. Knierim Department of Neurobiology and Anatomy University of Texas Medical School at Houston John Kubie Department of Anatomy State University of New York Downstate Medical Center Inah Lee Department of Psychology University of Iowa Sheri J. Y. Mizumori Department of Psychology University of Washington
Adrien Peyrache Laboratoire de Physiologie de la Perception et de l’Action CNRS Colle`ge de France Bruno Poucet Laboratory of Neurobiology and Cognition CNRS Aix-Marseille Universite´s James B. Ranck, Jr. Department of Physiology and Pharmacology State University of New York Downstate Medical Center Edmund T. Rolls Department of Experimental Psychology University of Oxford David C. Rowland Institute of Neuroscience University of Oregon Etienne Save Laboratory of Neurobiology and Cognition CNRS Aix-Marseille Universite´s Patricia E. Sharp Department of Psychology Bowling Green State University Trygve Solstad Centre for the Biology of Memory Norwegian University of Science and Technology
Robert U. Muller Department of Physiology and Pharmacology State University of New York Downstate Medical Center
Wendy A. Suzuki Center for Neural Science New York University
Lynn Nadel Department of Psychology University of Arizona
Heikki Tanila Department of Neurobiology University of Kuopio
CONTRIBUTORS
Alessandro Treves Cognitive Neuroscience Sector Scoula Internazionale Superiore di Studi Avanzati (SISSA) International School for Advanced Studies Sidney I. Wiener Laboratoire de Physiologie de la Perception et de l’Action CNRS Colle`ge de France
xv
Iain Wilson Centre for Cognitive and Neural Systems University of Edinburgh Emma R. Wood Laboratory for Cognitive Neuroscience University of Edinburgh
This page intentionally left blank
Historical Perspective: Place Cells in Ann Arbor and Brooklyn
A history of the discovery of place cells is simple. O’Keefe and Dostrovsky discovered them in the hippocampus and published their findings in 1971. But it took over 15 years for the truth of the discovery to be recognized. Here I (Ranck) will describe my own personal path through the intellectual and technological world up to about 1980. Phil Best and Aaron White have written two wonderful histories of place cells (Best and White, 1998, 1999) that I agree with and enjoy, including the things they have to say about me. Renshaw, Forbes, and Morrison (1940) were the first to record the firing of a single neuron in the central nervous system in a mammal, and one of the neurons they recorded was a complex-spike cell in the hippocampus. In 1960 Jasper, Ricci, and Doane reported on recording from single neurons in an awake monkey with its head fixed. Although they did not follow up on this work, it was picked up by Evarts to study sleep and the motor system. Hubel (1957) reported using tungsten electrodes with a sharpened tip to record from singled neurons, and in 1959 Stumwasser showed that he could record from single neurons in the brain by using a cut-off 25-mm wire. In the mid-1960s Olds at Michigan used Strumwasser’s method to record from single neurons in behaving rats, but he did it without a head stage. Therefore, because of microphonics, the method did not allow recording while the rats moved. The rats were trained to remain motionless, thus all of the recording was in motionless rats. O’Keefe and Bouma
(1969) used the Olds method to record the sensory properties of single neurons in the amygdala of the awake cat. I (Ranck) was at Michigan and knew Olds and his work well. It was clear that in order to record from a moving animal, it would be necessary to put a head stage with a low-output impedance on the animal’s head. The only devices available at the time for head stages were vacuum tubes, which were too big, or field effect transistors (FET), which were very temperamental and blew out if their gates were left open. In 1967–68 FETs became available that did not blow out if their gates were left open and rugged enough to be banged around on the head of a rat. Influenced by Olds, O’Keefe and I (Ranck) independently began to use these FETs at about the same time to record from single neurons in moving rats. We both used sharpened tungsten electrodes, not cut-off wires. I (Ranck) had been working on the impedance of brain as a way to study extracellular space and electrophysiological properties of glia. I had found a big impedance change in the subicular region during REM sleep and wanted to study the firing of those neurons during REM sleep. When I started to record in moving rats, I was swept away with the firing of hippocampal single neurons and behavior and never studied the firing of subicular neurons during sleep, nor has anyone else. In 1968 O’Keefe went to University College London to work with Patrick Wall, who is the real hero in the discovery and final acceptance of place cells. Wall supported O’Keefe and Nadel for years in
xvii
xviii HISTORICAL PERSPECTIVE a field that was remote from his own field, while place cells were ignored or actively opposed. Wall was not only a believer in place cells, he and O’Keefe also gave financial support to this belief, in spite of much opposition. In the late 1960s O’Keefe and I (Ranck) were both looking at the firing of single hippocampal neurons in freely moving rats. O’Keefe saw place cells and recognized what they were. In a 1973 article describing the firing of single hippocampal neurons, I described a neuron ‘‘that fired when the rat sniffed at the empty hole where the water spout was put. This cell also fired if the rat simply stood near the empty water-spout hole, if he licked the floor under the water hole, if he went to the site where the water jar was usually put, if he went to the dipper, or stood near the dipper.’’ I said that ‘‘each rat usually carried a pellet to the same site and ate it there. Whenever the rat was near this site the cell fired.’’ I (Ranck) had finished the experimental work before I became aware of O’Keefe and Dostrovsky’s (1971) work. Evidently I saw place cells, but did not understand what they were. One cannot see something unless one knows what to look for. As Pasteur said, ‘‘chance favors the prepared mind.’’ O’Keefe’s mind was prepared; mine was not. I did not use quantitative methods, but described an assortment of kinds of neurons—approach consummate, mismatch, motion punctuate. As Best and White (1999) have stated, this work generated a lot of words. In the fall of 1972, with my wife and daughter, I (Ranck) spent four months in London, where I worked with O’Keefe and Nadel. The moment I arrived they handed me a manuscript of a book. It was first time I had been exposed to Tolman and cognitive maps. I spent that fall learning cognitive psychology. Unfortunately, that fall there were no rats with place cells, but my mind was nonetheless being prepared. Phil Best spent the summer of 1973 in my lab in Ann Arbor. He tells the story well (and with humor) in an article by Best and White (1999). Suffice it to say that I saw place cells for the first time and knew what I was seeing that summer. To provide a context for the remainder of this historical perspective, it is worth noting the hippocampal people who were in Ann Arbor from 1957 to 1979. The first time I (Ranck) ever saw a hippocampus of a live animal was in the summer of 1956 at the University of Chicago, when I watched Bob Moore ablate a hippocampus from a cat. Moore then interned in Ann Arbor and introduced Bob Isaacson to hippocampal lesions. Isaacson worked on hippocampal lesions until he left Michigan in about 1969. Among his students and postdoctoral fellows were Rob Douglas, David Olton, and Len Jarrard. I taught a course in
electrophysiology, which was taken by Phil Best (then a postdoctoral student of Olds), David Olton (a graduate student of Isaacson), and Howard Eichenbaum (a graduate student of Charlie Butter). Eichenbaum and I got to know each other especially well, and Eichenbaum spent a few weeks in my lab in Ann Arbor before going off on his postdoctoral research at MIT. Case Vanderwolf was at the University of Western Ontario in London, about 100 miles away, and he came over to the Ann Arbor lab twice. My article on the behavioral correlates of the firing of hippocampal neurons came out in 1973, and for a time, I am embarrassed to say, my work was sometimes cited as being opposed to the place cell idea. I also contributed to the delay in acceptance of place cells. My acceptance of cognitive maps was gradually increasing, but except for the report of the Best and Ranck experiments at the Society of Neuroscience meetings in 1975, I did not change what I said in public until about 1976. I (Ranck) have no training in psychology. I was trained in medicine (I interned and then retired from clinical work), and then did a postdoctoral fellowship in electrophysiology and biophysics at the University of Washington. My knowledge of psychology was a once over lightly of the standard behaviorism of the time. My lack of training in psychology meant that I did not know about cognitive maps, but my lack of training also meant that I did not consider cognitive maps a heresy. It took me a while to catch on, but I did not have to unlearn as much as those with ‘‘proper’’ training. The O’Keefe and Nadel book was not published until 1978, but it circulated in the hippocampal underground from about 1974. In the eyes of many scientists at the time, place cells were not only wrong but blasphemous. In 2007, 30 years later, it is hard to convey the strength of feeling and anger that some felt toward the idea. The 1971 article of O’Keefe and Dostrovsky was easy to criticize; anyone who did not like place cells could rightly say that these neurons had not been properly described. The O’Keefe (1976) and O’Keefe and Conway (1978) studies were, however, good science. One of the lessons of these years (beyond having a prepared mind) is that when analyzing the relation of the brain to behavior, one should watch the behavior of the animal. Vanderwolf’s study of the behavioral correlates of field potentials from the hippocampus is a marvelous example of watching the behavior (1969). His work influenced me a great deal. At the time, much behavioral neuroscience was done by narrowly testing theory (e.g., via bar presses) rather than broadly looking for correlates. Even a weak correlate was taken as support of a theory. O’Keefe, Best, and
HISTORICAL PERSPECTIVE
I (Ranck) took a different tack: we watched the animal’s behavior and looked for strong neuronal firing correlates. This approach was the right one to take and, for many questions, remains appropriate today. The ‘‘watch-the-behavior’’ approach has the problem that it is difficult to quantitate, but is greatly aided by videotape (or digital video). It should be borne in mind that not all neuronal correlates are codes, but strong correlates are good candidates. Watching the behavior is the only way to screen for strong correlates. Our group in Brooklyn discussed this approach (Ranck et al., 1983). I (Ranck) went to Brooklyn in 1975 and in 1978 started working with John Kubie on place cells. We recorded the same place cell in lactating female rats in three different situations: running a differential reinforcement for low rates of responding (DRL) in a Skinner box, retrieving rat pups, and running an Olton radial-arm maze. The behaviors were chosen because they were all disrupted by hippocampal lesions, but the pup retrieval used a different motivation. All of the tasks were run in the same location in the same experimental room—i.e., the distal cues were the same, but the apparatus of each task was different. The operant chamber was placed on the radial-arm maze. The maternal chamber was a large box with clear walls, also placed on the radial-arm maze. We saw spatial correlates as described by O’Keefe in virtually every complex-spike cell that was well isolated. An individual cell’s place firing was environment specific. A cell could have firing fields in one, two, or three environments. We saw many clear examples of cells that had a clear place field in one environment, but were dead off in another. The firing of the same place cell was different in all three tasks. The results were summarized in two book chapters (Kubie and Ranck, 1983, 1984), in which we speculated that the environment-specific firing was a context signal. In 1983 Bob Muller joined the group. Muller and Kubie developed a computer-based method of recording location and place cell firing. Together, we also developed a strategy of using a single behavioral task (pellet retrieval) in simple geometric enclosures. The work was thoroughly quantitative and the falsecolor figures were striking. From the publication of those studies (Muller and Kubie, 1987; Muller et al., 1987) it seemed that no one doubted the existence of place cells any more. These studies clearly documented the phenomenon of environment-specific firing and introduced the concept of partial and complete remapping. In 1984 I (Ranck) discovered head direction cells. This time my mind was prepared and I saw and understood them immediately. Another computer-based system was developed for head direction cells. In 1986
xix
Jeff Taube came to our lab and did the work of documenting them. When the first article was published describing head direction cells (Taube et al., 1990), the world was also prepared to believe and accept them. James B. Ranck, Jr. John Kubie
References Best PJ, White AM (1998) Hippocampal cellular activity: a brief history of space. Proc Natl Acad Sci USA 95: 2717–2719. Best PJ, White AM (1999) Placing hippocampal singleunits studies in a historical context. Hippocampus 9: 346–351. Best PJ, Ranck JB Jr (1975) Reliability of the relationship between hippocampal unit activity and sensorybehavioral events in the rat. Soc Neurosci Abstr. Best PJ, Ranck JB Jr (1982) The reliability of the relationship between hippocampal unit activity and behavior in the rat. Exp Neurol 75:655–664. Evarts EV (1969) Pyramidal tract activity associated with conditioned hand movement in the monkey. J Neurophysiol 29:1011–1027. Hubel DH (1957) Tungsten microelectrodes for recording single units. Science 125:549–550. Jasper H, Ricci GF, Doane B (1960) Microelectrode analysis of cortical cell discharge during avoidance conditioning in the monkey. Electroencephal Clin Neurophysiol 27:205–208. Kubie JL, Ranck JB Jr (1983) Sensory-behavioral correlates in individual hippocampus neurons in three situations: space and context. In: Neurobiology of the Hippocampus (Seifert W, ed.), pp. 303–319. San Diego: Academic Press. Kubie JL, Ranck JB Jr (1984) Hippocampal neuronal firing, context, and learning. In: The Neurobiology of Memory (Butters N, Squire LR, eds.), pp 417–423. New York: Guilford Press. Muller RU, Kubie JL (1987) The effects of changes in the environment on the spatial firing of hippocampal complex-spike cells. J Neurosci 7:1951–1968. Muller RU, Kubie JL, Ranck JB Jr (1987) Spatial firing patterns of hippocampal complex spike cells in a fixed environment. J Neurosci 7:1935–1950. O’Keefe J, Bouma H (1969) Complex sensory properties of certain amygdala units in the freely moving cat. Exp Neurol 23:384–398. O’Keefe J, Conway DH (1978) Hippocampal place units in the freely moving rat: why they fire where they fire. Exp Brain Res 31:573–590. O’Keefe J, Dostrovsky J (1971) The hippocampus as a spatial map. Preliminary evidence from unit activity in the freely moving rat. Brain Res 34:171–175.
xx
HISTORICAL PERSPECTIVE
Olds J (1965) Operant conditioning of single unit responses. Proceedings of the XXIII International Congress of Physiological Science, Tokyo. Ranck JB Jr. (1973) Studies on single neurons in dorsal hippocampal formation and septum in unrestrained rats. Part I. Behavioral correlates and firing repertoires. Exp Neurol 41:461–555. Ranck JB Jr, Kubie JL, Fox SE, Wolfson S, Muller RU (1983) Single neuron recording in behaving animals: bridging the gap between neuronal events and sensorybehavioral variables. In: Behavioral Contributions to Brain Research (Robinson TE, ed.), pp 62–93. New York: Oxford University Press.
Renshaw BA, Forbes BR, Morrison A (1940) Activity of isocortex and hippocampus: electrical studies with microelectrodes. J Neurophysiol 3:74–105. Stumwasser F (1958) Long term recording in brains of unrestrained mammals. Science 127: 469–470. Taube JS, Muller RU, Ranck JB Jr (1990) Head-direction cells recorded from the postsubiculum in freely moving rats. I. Description and quantitative analysis. J Neurosci 10:420–435. Vanderwolf CH (1969) Hippocampal electrical activity and voluntary movement in the rat. Electroenceph Clin Neurophysiol 26:407–418.
I PLACE CELLS AND SPATIAL CONTEXT
This page intentionally left blank
1 The Hippocampus and Context Revisited LYNN NADEL
EXACTLY WHAT ROLE DOES THE HIPPOCAMPUS PLAY IN CONTEXT?
Cognitive map theory (O’Keefe and Nadel, 1978, 1979), based on the discovery of place cells in the hippocampus (O’Keefe and Dostrovsky, 1971; O’Keefe, 1976), asserted that the hippocampus is ‘‘central to a particular form of memory: that concerned with the representation of experiences within a specific context’’ (O’Keefe and Nadel, 1978, p. 381). This in turn led to the suggestion that the hippocampal system plays a central role in representing spatial context (Nadel and Willner, 1980; Nadel et al., 1985). We asserted that (1) the hippocampus is critical specifically for spatial context (O’Keefe et al., 1979), and (2) this contextual function accounts for the hippocampal role in episodic memory (O’Keefe and Nadel, 1978). Although there was early agreement that the functions of the hippocampus were somehow related to context, and data from the study of patients with medial temporal lobe damage supported this view, there was little systematic empirical work exploring this idea until the 1990s. More recently, however, the critical role of context in understanding the basic functions of the hippocampus has become clearer. This chapter revisits ideas first promulgated over 20 years ago, and then suggests some new avenues of investigation that might bring us closer to understanding the role of the hippocampal system in mediating context. In it I attempt to clarify how the kind of contextual knowledge represented in hippocampal circuits is central to the construction of both the experienced past and the imagined future, thus shedding light on how episodic memory works, as well as the putative processes of memory ‘‘consolidation’’ and ‘‘reconsolidation.’’1
In our early writing on the role of the hippocampus in context we stressed a few critical ideas. Most important perhaps was the idea that context was not simply a ‘‘cue’’ like other cues an organism might learn about. We argued that context representations formed in the hippocampus were fundamentally configural, being based on relations among the environmental features that comprised the physical layout of space. We further argued that because of this configural nature, learning about spatial context was different in important ways from learning about isolated cues, or elements, in the organism’s world.2 This idea led us to suggest that ‘‘the traditional view which treats context as merely another cue is incorrect’’ (Nadel and Willner, 1980, p. 226), and that the laws of associative learning applied to a wide range of learning paradigms, based, among others, on the formulations of Rescorla and Wagner (1972), did not apply to the kind of spatial learning at the heart of hippocampal function. One direct prediction of this idea is that phenomena such as ‘‘blocking’’ and ‘‘overshadowing’’ might not be observed in spatial learning involving the hippocampus. Recent data suggest that this prediction holds, at least under some conditions (e.g., Hayward et al., 2003). A second idea we emphasized was that hippocampal damage would manifest as a lack of context specificity—learning would be inappropriately generalized to novel contexts. This followed from the idea 3
4
PLACE CELLS AND SPATIAL CONTEXT
that one of the major functions of contextual coding is to allow organisms to respond appropriately to ambiguous signals. That is, a particular stimulus (or object) can be rewarding (or punishing) in one context, but have a different meaning in another. Behavior that is appropriate in one place is often not appropriate elsewhere. It is highly adaptive to be able to code the meaning of experiences with reference to contexts, which then allows an organism to predict, based on prior experience, what a given situation is likely to bring, which then allows the organism to respond selectively and adaptively. In support of this view, we noted that learning in various situations is context dependent, e.g., latent-inhibition, habituation, extinction, and reinstatement. That is, what an organism learns about entities and events in the world, and their meaning, is typically tied to the context in which the learning occurs. Recent work expands this list and shows, for example, that such contextual control, and hence disambiguation, occurs widely throughout the animal kingdom (e.g., in cockroaches; see Sato et al., 2006). The prediction that hippocampal damage in mammals should lead to abnormal generalization across context has also been largely confirmed, as discussed at greater length below. Our approach to context was by no means novel. A long tradition of work in Eastern Europe had emphasized the importance of learning about contexts (e.g., Asratyan, 1965; Beritoff, 1965; Konorski, 1967). In North America, Bolles (1985) also emphasized the importance of context, noting that ‘‘contextual stimuli, which are customarily assumed to enter into association with USs [unconditional stimuli] just as the primary CSs [conditional stimuli] do, do not seem to behave like that’’ (p. 364). Bolles went on to note that contextual cues become important under ambiguous conditions, and that this is particularly important with respect to conditioned inhibition, which arises only in reaction to prior excitatory conditioning, hence describes an inherently ambiguous situation. Notwithstanding these various claims that context was something rather different than a typical CS, this idea had only a modest impact until the 1990s, when evidence began to accumulate in support of two fundamental claims: (1) spatial context is indeed different than a CS; and (2) the hippocampus is critically important in representing spatial context. One of the earliest such demonstrations was provided by Penick and Solomon (1991), who trained rabbits on eyeblink conditioning after either hippocampal or neocortical lesions. Following training, animals were tested for retention in either the same or new context. Whereas unoperated control animals and animals with neocortical lesions showed impaired retention when tested in the new context, animals with
hippocampal lesions were unaffected by this context shift. The authors noted that assigning a role to the hippocampus in learning about context might resolve a long-standing debate about why the hippocampus is critical for some forms of classical conditioning (e.g., trace conditioning) but not others (e.g., delay conditioning). Complications arose, however, when investigators shifted to using a fear-conditioning paradigm. In this situation training involves exposing a rat (or mouse) to an electric shock (the US), which in the early studies was typically paired with a CS. Selden et al. (1991), Phillips and LeDoux (1992) and Kim and Fanselow (1992) all showed that hippocampal lesions interfered with context conditioning in this paradigm. However, one can also demonstrate context conditioning of fear in the absence of a CS, or with the CS and US explicitly unpaired. When there is a CS explicitly paired with the US, as in the early studies just noted, it is said that the context is ‘‘in the background.’’ When there is no CS, or a CS explicitly unpaired with the US, then it is said that context is ‘‘in the foreground.’’ Phillips and LeDoux (1994) showed that hippocampal lesions in rats interfered with ‘‘background’’ but not ‘‘foreground’’ context conditioning. Understanding this pattern of results requires an unpacking of what is meant by context, and how animals learn about it.
A PERSPECTIVE ON THE HIPPOCAMPUS— CONTEXT LITERATURE There have been several recent reviews of work on hippocampus and context (e.g., Holland and Bouton, 1999; Rudy et al., 2004; Smith and Mizumori, 2006). Although there is general agreement that the hippocampus plays a role in context effects, confusion arises from continued misunderstanding of the nature of context itself, in particular the fact that there are multiple forms, and hence representations, of context, only one of which depends upon the hippocampus. In many ways this assertion about multiple forms of context parallels assertions about multiple forms of memory. This should not be surprising since memory is nothing but acquired knowledge, and knowledge about contexts is what these multiple contextual representations provide (see Nadel, in press, for further discussion of this ‘‘memory as knowledge’’ perspective). The possibility that animals might learn about contexts in more than one way was made explicit by Nadel et al. (1985), as noted above, and also discussed by Phillips and LeDoux (1994) in accounting for their results with background and foreground context conditioning. They noted the possibility that ‘‘in the absence of a functional hippocampal system, the subjects
THE HIPPOCAMPUS AND CONTEXT REVISITED
may be reduced to forming single cue associations’’ (p. 41), which can be mediated in their view by the amygdala. These associations will suffice in the case of foreground conditioning, since the absence of a CS paired with the US leads the animal to associate the US with salient individual static stimuli. In the case of background conditioning, however, the paired CS is the most salient stimulus, and context conditioning, if it is to happen at all, would be with reference to the configuration of static cues ‘‘in the background.’’ And since the hippocampus is critical for learning about such configurations, lesions interfere with context conditioning in this case.3 The fact that contextual knowledge is multiply represented has been discussed at some length by Jerry Rudy and his colleagues. In a series of important empirical and conceptual studies (e.g., Rudy, 1996; Rudy and Pugh, 1998; Rudy and O’Reilly, 1999, 2001; Rudy et al., 2002, 2004; Matus-Amat et al., 2004), they have built on our earlier discussion of contexts coded both as ‘‘elements’’ and ‘‘configurations’’ (Nadel and Willner, 1980; Nadel et al., 1985). I outline the story briefly here; interested readers should consult Rudy’s studies for more detail. Research using conditioned fear from studies by Rudy’s group as well as Fanselow’s group (Kim and Fanselow, 1992; Anagnostaras et al., 1999, 2001; Fanselow, 1999, 2000) has demonstrated two very important things: (1) the hippocampus seems to be necessary for acquisition of context fear, and for retrieval of such fear for some days (or weeks) after initial training, but not for retrieval 28 days after training; and (2) the acquisition of context fear itself depends upon the animal having had some exposure to the context before fear training starts. Absent such experience, context fear does not develop. Although it is widely appreciated that some exposure to the context prior to introduction of the US is necessary for fear of the context to be conditioned, the connection between this finding and the data showing that hippocampal involvement in contextual fear is time limited has not been made explicit. These two findings are, in my view, related. Consider the second effect, which is typically taken to reflect the ‘‘systemslevel consolidation’’ of context fear over time, such that the memory trace involved becomes independent of the hippocampus and dependent instead on circuits in cortex (or elsewhere). My colleagues and I have argued at some length that this standard view of memory consolidation is misguided (e.g., Nadel and Moscovitch, 1997, 1998; Moscovitch et al., 2006; Nadel et al., 2007) in assuming that memories ‘‘transfer’’ from hippocampus to neocortex over time. We have also argued that a more sophisticated version of the standard consolidation story—that memories are
5
always stored only in neocortex but initially hippocampal involvement is required for retrieval—is also wrong. The major problem with this story, which can be seen quite clearly in the work on conditioned fear, lies in its assumption that the content of the memory remains constant as the neural system responsible for it shifts across the 28 days of the study. Instead, we have argued (e.g., Nadel and Moscovitch, 1997) that some aspects of memory change over time, as consolidation proceeds. More specifically, memories can lose their ‘‘context’’ dependence, becoming less ‘‘episodic’’ and more ‘‘semantic’’ in nature. This is one reason why we feel it would be better to use the term memory transformation than to use consolidation to label this process. Evidence supporting this transformational view has been provided by a number of recent studies, many of them using the conditioned-fear paradigm (MacArdy and Riccio, 1995; Houston et al., 1999; Balogh et al., 2002; McAllister and McAllister, 2006; Wiltgen and Silva, 2007; Winocur et al., 2007). The main result is that shortly after acquisition, conditioned fear is manifest strongly in the training context, but not in other contexts; some weeks later (in the rat or mouse), conditioned fear can be expressed in both training and novel contexts. These results suggest an altogether different interpretation of the retrograde amnesia gradient observed with hippocampal lesions than that offered by standard consolidation theory. Instead of assuming that ‘‘memory’’ is either transferred from hippocampus to neocortex, or given independent status within neocortex after a period of requiring hippocampal help in retrieval, one can best account for the data by assuming that the ‘‘context’’ representation that supports conditioned fear after several weeks is a representation based on elements in the test situation rather than a configural representation of the whole. Given this control by elements rather than a configuration, generalization to contexts sharing some of these elements (but not the configuration) becomes understandable. Further, this shift in control provides ready explanation for several facts that until now have been interpreted within the framework of standard consolidation theory. First, hippocampal lesions fail to impair conditioned fear after this transformational process because the neural representations supporting fear at this point are elemental representations dependent upon the neocortex. These representations play an increasingly important role in supporting fear over time, regardless of whether the hippocampus is surgically removed. Second, neural activation studies show that, after some weeks, expression of conditioned fear is accompanied by decreased hippocampal activation and a concomitant
6
PLACE CELLS AND SPATIAL CONTEXT
increase in neocortical activation (e.g., Bontempi et al., 1999), reflecting this shift in control from a configural to elemental representation of the context. In this regard it is critical to note that in Bontempi et al.’s original study two things happen with time: first, as the activation data show, the hippocampus plays a less critical role; and second, performance itself weakens. This is exactly what one would expect if the representation supporting behavior changes over time, from one that represents the configural nature of the context to one that represents isolated elements in that context. Both kinds of representation can support the expression of fear, but the hippocampus-dependent configural representation supports fear restricted to the training context, while the neocortex-dependent elemental representations support fear that is both weaker than and generalized to other contexts that share some of the same environmental elements. This story connects with what has been learned from studying the pre-exposure effect (see Fanselow’s work and Rudy’s work). When an animal is given fear training without some exposure to the training context before US introduction, it fails to learn to associate shock with the ‘‘context’’ understood as the configuration of elements (and their spatial relations) in the chamber. This happens because exposure to the shock chamber is essential for the animal to acquire a configural representation of the context in the first place— what we called a ‘‘cognitive map’’ (O’Keefe and Nadel, 1978), or a ‘‘contextual representation’’ (Nadel and Willner, 1980). I believe that this parallels what happens over time in the consolidation domain. Initial training (with preexposure) leads the animal to associate fear with the configurally represented context. As such, the behavior depends upon the hippocampus as well as the amygdala. Over time, and as a direct consequence of what has been called consolidation, the contextual binding weakens, leaving behind only linkages between elements of the chamber and the shock. These considerations make it much easier to understand the existing literature on hippocampal lesion effects and context and why doubts still exist about hippocampal involvement in context learning (e.g., Gewirtz et al., 2000). When ‘‘normal’’ behavior depends upon a configural representation of context, hippocampal lesions will lead to impairment. This should be the case in both acquisition and retention. When a task is used that can be solved with either a configural or an elemental representation of context, hippocampal lesions will not cause an obvious impairment; rather, special testing methods will have to be used to show that the basis of performance differs between animals with hippocampal lesions and control animals. The most obvious such method would be
to shift the test context. Paradoxically, animals with hippocampal lesions should be less affected by such a shift than intact animals. In the case of conditioned fear, for example, hippocampal-lesioned rats should show greater than normal fear in an out-of-context test. In my doctoral work (Nadel, 1968) I showed exactly this effect, but did not at that time understand what it was telling me. Rats with dorsal hippocampal lesions tested in context B for fear of a CS paired with shock in context A actually showed more fear than did control rats. This finding parallels the Penick and Solomon (1991) result noted above, and is consistent with the report by Good and Honey (1991) showing that hippocampal lesions impaired rats’ ability to learn that a stimulus was reinforced in one context but not in another (see also Lehmann et al., 2005; but see Hall et al., 1996). It is also consistent with the recent findings that hippocampal inactivation impairs the context specificity of latent inhibition (Maren and Holt, 2000), and extinction (Corcoran et al., 2005; Hobin et al., 2006), and that reinstatement of conditioned fear in humans is context specific (LaBar and Phelps, 2005). The impact of hippocampal lesions on retention of a context-based task will depend on when retention is tested, and on whether or not the animal was reminded of the context before retention was tested. As we argued above, after some weeks during which a rat is not returned to the training context, its configural representation of that context weakens, and elemental contextual representations take over. In the absence of reminders that bring the context back into the picture, hippocampal lesions have little or no effect. But, as Land et al. (2000) have shown, if animals are reminded of the context before lesions are made, these lesions can impair retention (for a similar result see Debiec et al., 2002). This result, of course, reflects what has now come to be called ‘‘reconsolidation,’’ a phenomenon I return to after some consideration of the underlying neural mechanisms at play.
SOME BRIEF COMMENTS ON THE PHYSIOLOGY OF CONFIGURAL CONTEXT Assuming hippocampal circuits represent configural context, one needs to ask how they do it. Perhaps the simplest answer would be that a collection of place cells comprises a ‘‘cognitive map’’ of an environment, and that this cognitive map simply is the configural representation of that context. An animal (or human) recognizes a context by calling up the appropriate cognitive map. Within this simple view, we can understand the relation between configural and elemental
THE HIPPOCAMPUS AND CONTEXT REVISITED
contexts as follows: the hippocampal system4 ‘‘represents’’ context in that it provides the spatial scaffold within which representations of the elements comprising the context are related to one another. These elemental representations are not contained within the hippocampus, but the machinery responsible for relating them spatially depends critically on hippocampus. In this way, context construed configurally requires the hippocampus, but ‘‘context’’ construed elementally does not.
How Animals ‘‘Recognize’’ a Context When an animal is placed into a context, neural ensembles previously associated with that context are activated—that is, the place cells representing locations in the environment fire at their appropriate sites. Something more than mere recognition of local places must be happening in this situation to account for context recognition. Samsonovich and McNaughton (1997) talked about the calling up of the correct ‘‘chart’’ within the hippocampus, where a chart is a representation of context (Samsonovich and Nadel, 2005). But how does the system (and the animal) recognize the context such that the correct chart is retrieved? It must be the case that attention to a number of environmental features, and their relation to each other, triggers the activation of a part of a particular hippocampal chart, or representation, and this part-activation in turn triggers the retrieval of the remainder of that representation. What is meant by ‘‘retrieval’’ here is critical: it cannot mean that all the elements representing all the features in that environment are simultaneously fully activated, since the neurons that actually ‘‘fire’’ appear to be largely limited to those representing the local place the animal is in at the moment. At the same time, this partial retrieval, accompanied by a limited set of activations, in some way constrains the subsequent neural activations that will occur as the organism moves through the environment. By calling up a particular ‘‘chart’’ the system ‘‘decides’’ that it is in a particular environment and that certain neurons will be activated in certain places, should the organism get to those places.5 This process reflects the fact that the hippocampal representation, perhaps because it embodies ‘‘attractor’’ dynamics (cf. Wills et al., 2005; but see Leutgeb et al., 2005), can produce ‘‘pattern completion’’ when incomplete inputs are provided to the hippocampal system. These dynamics mean that small changes in the environment will be ‘‘compensated’’ and will not lead to the conclusion that the animal is in a different context. They will, however, be noticed, and frequently an animal will go out of its way to explore such novelty so as to incorporate information about it into its
7
representation going forward. At the physiological level this adjustment of the representation appears to be reflected in what has been called ‘‘rate remapping’’— place cells within the hippocampus continue to be activated in the same locations, but with somewhat changed firing rates. This automatic update process was postulated as a critical part of the cognitive mapping system (O’Keefe and Nadel, 1978), and there have been numerous demonstrations that it happens in intact animals and that the hippocampus is critical to it. Lesions in the hippocampal system have the effect of rendering animals incapable of either noticing changes in the configuration of environmental features (Save et al., 1992a,b; Lenck-Santini et al., 2005) or of reacting selectively to them, or both. The fact that updating is triggered only after the recognition of a familiar context, within which some change has occurred, turns out to be critical, as we discuss below. If, however, the context is changed in a ‘‘major’’ way, such pattern completion becomes impossible, updating an old representation does not occur, and a new representation is formed instead. In this case one could say that the system/animal has ‘‘decided’’ that it is in a new context. So-called global remapping of hippocampal place cells (and related grid cells and head direction cells; cf. Knierim et al., 1998; Fyhn et al., 2007) after substantial situational change tells us both that the animal has decided it is in a new context and that it has formed a new map of that context.6 I would argue that this apparently simple process of ‘‘deciding’’ whether the context is old or new has profound implications for how the brain learns and what it remembers. This decision has several consequences: 1. An animal placed in an old context within which minor changes have transpired will selectively explore the site of those changes, allowing it to update its representation of the context. 2. An animal placed in what it perceives to be a new context will leave its old contextual representations intact (i.e., updating will not occur) and will form a new representation instead based on exploration of the entire context. That is, it will ‘‘remap.’’ Stated in this way, the decision as to whether one is in the same or a new context is fundamental—it determines whether one should update an old memory file or create a new one. We know that part of the hippocampal system (the dentate gyrus) serves to ‘‘orthogonalize’’ inputs, such that relatively similar inputs to this level of the system are ‘‘mapped onto’’ quite different elements at the next level (the CA3 pyramidal cells). This process reflects the all-or-none character
8
PLACE CELLS AND SPATIAL CONTEXT
of the decision about context—either it is a familiar, albeit modestly changed context, or it is a new context. Up to some point the system accepts change as reflecting nothing more than minor variations in a familiar context, and the collection of place cells retain their ‘‘fields.’’ Beyond that point, the system undergoes a phase transition, and the place cells ‘‘remap.’’ It is important to note that our best measure of when an animal ‘‘thinks’’ it is in a new context is this very process of ‘‘global remapping.’’ When it occurs, the animal, or, more properly, the animal’s brain, has decided that it is in a new context. When it does not occur, the animal decides it is on familiar ground, and only ‘‘rate remapping’’ occurs. At the behavioral level this ‘‘decision’’ can be reflected in the nature of the exploration the animal engages in. In an old, changed, context exploration focuses on the site of the change. In a new context exploration takes in the entire environment.
WHAT DEFINES A CONTEXT? This discussion of context change, and the ways in which animals determine when contexts have changed, has largely avoided one very important question: which features of an environment does an animal use to define context? This question is critical for a variety of reasons, although answering it in depth is beyond the scope of this chapter. Nonetheless, a few things can be said, while admitting that this is an area ripe for further study. Features that are relatively stable first of all define contexts. Things that move around, or come and go, less reliably define a context. For much the same reason, internal states that change frequently over time, such as one’s state of hunger or thirst, are also less reliable indicators of specific contexts. Not only does one feel variably hungry or thirsty in a given context, but one also feels quite hungry or thirsty in multiple contexts. For these reasons internal states are not very useful in helping organisms disambiguate one context from another, except in unusual cases (seeKennedy and Shapiro, 2004). The enormous variety of spatial configurations in the natural world makes the use of space a particularly good choice in defining context, precisely because this variety affords the possibility of reliable discrimination between contexts (see Smith and Mizumori, 2006, for a somewhat different take on what defines context). Given these considerations it is not surprising that spatial representation and context-dependent (episodic) memory use the same neural substrate. Though it seems clear that static features are favored in defining contexts, one can still ask if all static features are equally potent, and the answer seems to be no. More salient things are likely to play a greater role
in defining a context, if only because they are more likely to be attended. The question, of course, is what makes a feature salient such that it attracts attention. James (1890, p. 403) said a long time ago that ‘‘everyone knows what attention is,’’ and that such things as loud bangs, blood, shiny objects and the like attract it, but this is not so obvious in the domain of determining which environmental features define a context. Recent work from Jeffery’s group raises the possibility that animals use metric and nonmetric features of environments in two different ways. The metric features are used to enable orientation and navigation, while the nonmetric features are used to define the context for purposes of distinguishing one place (or context) from another. Metric features tend to be distal ones, such as boundaries and geometric shape (Anderson et al., 2006), while nonmetric features can be either proximal or distal. It has long been known that animals prefer to use distal features for orientation (see Nadel and Hupbach, 2005), and the work of Jeffery and her colleagues (e.g., Jeffery and Anderson, 2003; Jeffery et al., 2003, 2004) raises the intriguing possibility that by using different features for two distinct purposes the same system can be used to ‘‘tell the animal both which environment it is in, and where it is within it’’ (Anderson et al., 2006, p. 730). This would allow an animal to both orient successfully in an environmental space and attach distinctive meanings to distinguishable ‘‘contexts’’ within that space. In a recent study Lever et al. (2006) suggested that a specific behavior pattern—rearing—can be used to distinguish between various forms of an animal’s reaction to novelty. In line with Jeffery’s work, Lever et al. note that when placed in a novel environment rats will rear extensively, perhaps to gain greater information about distal features of that environment. In contrast, when placed in a familiar context containing a few changes, such as new or displaced objects, rats will not rear but will instead focus their exploration on the specific sites of change. Lever et al. also note that maximal rearing is obtained in a situation they call ‘‘mismatch novelty’’—placing the animal in a strange local context within a familiar environment (e.g., a new box in an old room). These patterns of behavior parallel the distinction between global remapping and rate remapping noted earlier, and it is intriguing that extensive rearing is typically accompanied by global remapping. New analyses of the subtleties of exploration and an animal’s reaction to contextual change are quite important, as too little attention has been paid in the recent past to the critically important function of exploration. As Lever et al. point out, and as we noted years ago (O’Keefe and Nadel, 1978), exploration is how animals create internal models of the external world. These promising approaches deserve
THE HIPPOCAMPUS AND CONTEXT REVISITED
careful attention in the future, given how important context is in understanding the core functions of the hippocampal system.
CONTEXT RECOGNITION—THE KEY TO RECONSOLIDATION Creating an entirely new representation in response to deciding that one is in a new environment is quite different than updating an existing representation on the basis of some local change. I believe this difference is fundamental to the distinction between memory ‘‘consolidation’’ and memory ‘‘reconsolidation.’’ In what follows I will try to use the terms suggested above, namely memory transformation and memory updating, in place of memory consolidation and reconsolidation. I have discussed various aspects of these processes elsewhere (Nadel, in press), so I will be concise here. It has long been assumed that a time-dependent stabilization process unfolds after initial acquisition of a memory (Muller and Pilzecker, 1900). During this time period, termed the ‘‘consolidation’’ interval, memories can be disrupted by new learning experiences, blows to the head, hypothermia, electroconvulsive shock, etc. This idea was initially couched in both physiological and psychological terms, and included the possibility that the content of the memory might itself be transformed during consolidation (cf. Burnham, 1903). Although these early writers assumed that consolidation involved a physiological process, the first detailed proposal came from Hebb (1949), who provided a way of understanding how memories could become stabilized. Hebb assumed that memories are captured in the brain through changes in synaptic efficacy, and that these changes depend upon complex cellular and molecular mechanisms that lead to structural alterations underpinning potentiated synaptic function. In Hebb’s view, these changes unfolded within the same cell assemblies initially activated by the experience, possibly through reverberations within these assemblies. Study of patient H. M. (Scoville and Milner, 1957), however, suggested that, at least for episodic memory, consolidation involves a shift in which brain structures are critical for memory retrieval. Thus began a long tradition of linking what has come to be called ‘‘systemslevel memory consolidation’’ to a shift from hippocampal to neocortical dominance in memory retrieval. The consolidation period was assumed to end when the hippocampal system was no longer essential in retrieval. It was in this context that the idea of memory reconsolidation first emerged. A number of investigators
9
in the 1960s and 1970s, unconvinced by the consolidation idea, argued instead that memories were always open to alteration and/or disruption so long as they were in an active state (see Misanin et al., 1968; Lewis, 1979). Memories could be brought back to an active state through ‘‘reminders’’ such as exposing the organism to the CS used in the learning task, or the context in which learning took place. These ideas, though backed by several well-replicated findings, were pushed aside by the consolidation bandwagon, for reasons that would be of interest in an article on the history of science but that are beyond the scope of this effort. The notion of reconsolidation re-emerged in two labs: Sara and her colleagues (Przybyslawski and Sara, 1997; Sara, 2000) and Nader, LeDoux and their colleagues (Nader et al., 2000) showed that reminders could bring well-consolidated memories for maze learning and fear conditioning, respectively, back to a fragile state, and that these newly fragile memories could be disrupted by the systemic injection of MK-801 (an NMDA receptor antagonist) or protein synthesis inhibitors into the amygdala, respectively. There has followed a torrent of studies demonstrating the robust nature of reconsolidation, its presence in a wide variety of species and learning situations, the ways in which it is differentiated from consolidation, and some of the boundary conditions that constrain it. The focus of this work has been almost entirely on manipulations that can disrupt memory retrieval once a reminder has brought it back to a fragile state, giving many the impression that what was at stake was largely a destructive process. At the same time, a tradition of research using human subjects has demonstrated apparently similar malleability in supposedly consolidated memories (see Loftus, 2005). Much of this work uses a standard procedure: subjects are exposed to a complex event, then some time later they are given misinformation about some detail of that event. When subsequently asked to recall the event, quite often the misinformation rather than the original detail is remembered. One thing that distinguishes this important work on human memory from the animal work just discussed is the absence of any systematic manipulation of specific reminders. In the hope of bringing these literatures together, we have recently developed a paradigm to study reconsolidation in human episodic memory that depends on reminding the subjects about what they previously learned (Hupbach et al., 2007). Subjects are initially trained on a ‘‘list’’ of objects. These objects—such things as a pencil, comb, or other similarly sized common object—are kept in a blue basket and presented one by one to the subject. After all 20 objects are presented the subject is asked to verbally ‘‘recall’’ the
10 PLACE CELLS AND SPATIAL CONTEXT list. This training sequence is continued until the subject recalls at least 17 of the 20 objects (in any order) or for a maximum of four learning trials. Two days later subjects return to the laboratory and are divided into two groups. Subjects in one group are reminded of their previous training experience, subjects in the other group are not. Then, a second list of objects is learned, but in a different way. The objects on this second list are arrayed on a table instead of being contained in a basket. Following the learning of this second list we test for recall of both lists either immediately or 2 days later. In one study we tested recall of list 1, and in another study we tested recall of list 2. In both studies we contrasted subjects who had been reminded with subjects who had not. The results can be summarized as follows (see Hupbach et al., 2007, for a full description of this study): if, and only if, a reminder is given prior to the learning of list 2, subjects will ‘‘intrude’’ items from list 2 into list 1 when asked to recall list 1 (Fig. 1–1A). Additionally, we showed that this result occurs only when recall is tested 2 days later (Fig. 1–1B). It is not observed when recall is tested immediately. Intrusions from list 1 into list 2 recall are never seen, regardless of whether list 2 is recalled immediately or 2 days later (see Fig. 1–1C). We interpret these results as showing that the reminder prior to list 2 learning reactivates the memory of list 1, and triggers an ‘‘update’’ mechanism that causes the subject to conflate the list 2 and list 1 objects. Absent the reminder the subjects treat list 2 learning and list 1 learning as separate episodes and intrusions do not occur. We have more recently started to explore exactly what kinds of reminders play a critical role in initiating this ‘‘update’’ mechanism (Hupbach et al., submitted). In the original study the reminder involved bringing the subject back to the same context, with the same experimenter, who asked a leading question about the list 1 training experience. The no-reminder group was
Figure 1–1. A. Mean percentage of items correctly recalled from list 1 and incorrectly recalled from list 2 in the reminder, the no-reminder, and the interference control conditions. Participants were asked to recall objects from list 1. This test was done 2 days after learning list 2. B. Mean percentage of items correctly
recalled from list 1 and incorrectly recalled from list 2 in the reminder, the no-reminder, and the interference control conditions. Participants were asked to recall objects from list 1. This test was done immediately after learning list 2. C. Mean percentage of items correctly recalled from list 2 and incorrectly recalled from list 1 in the reminder, the no-reminder, and the interference control conditions. Participants were asked to recall objects from list 2. This test was done 2 days after learning list 2. Error bars represent standard errors of means.
THE HIPPOCAMPUS AND CONTEXT REVISITED
Figure 1–2. A. Mean percentage of items correctly recalled from list 1 and incorrectly recalled from list 2 in the reminder conditions in which participants were reminded about context, the experimenter, or the listlearning task (question). Participants were asked to recall objects from list 1. This test was done 2 days after learning list 2. B. Mean percentage of items correctly recalled from list 1 and incorrectly recalled from list 2 in the reminder conditions in which participants were reminded about context and experimenter, context and the list-learning task, or the experimenter and the list-learning task. Participants were asked to recall objects from list 1. This test was done 2 days after learning list 2. Error bars represent standard errors of means.
11
brought to a different context, with a different experimenter, and was not asked about list 1 training. In our most recent work we systematically manipulated the nature of the ‘‘reminders’’ available to the subjects prior to learning list 2. In one set of studies we provided only one of the three reminder cues: the original training context, the original experimenter, or the leading question about the basket in which list 1 objects were kept. Figure 1–2A shows the results of these manipulations: only the group that received a context reminder showed the memory-updating effect. The other two groups showed few if any intrusions of list 2 items into list 1 memory, indicating that updating had not occurred in these groups. In a second set of studies we provided two of the three cues, either context plus experimenter, context plus question, or experimenter plus question. We wanted to explore the possibility that the failure of the experimenter or question to initiate an updating process might have reflected the fact that these are weak cues compared to context, and that by combining these two weaker cues we would be able to demonstrate updating. Figure 1–2B shows that this was not the case. Once again, only the provision of a context reminder, in combination with either the experimenter or the question, elicited updating. That these conditions are effective is hardly surprising since the previous study had shown that context alone is enough to trigger updating. This set of experiments demonstrates that reconsolidation, as reflected in memory-updating effects dependent upon reminding, can be observed in human learning. They further show that such updating only occurs when the context is part of the reminder manipulation, at least in the experimental conditions of our studies. We believe these results support the idea that context plays a unique role in determining how the memory system behaves. Put most directly: when in the same context, an existing representation is updated and transformed, but when the organism is in a new context, an entirely new representation is created.
CONCLUSIONS Context is a critical component of episodic memory. It is, however, more than just a component of such memory. It also seems to play a determining role in the dynamics of the episodic memory system as a whole. To the extent to which this is the case, further study of how context is represented physiologically should greatly enhance our understanding of human memory. The implications of such study can scarcely be exaggerated. Recent work has shown that the episodic
12 PLACE CELLS AND SPATIAL CONTEXT memory system is not only engaged in creating representations of the past but also critical to using these representations as the basis for imagining the future (e.g., Hassabis et al., 2007). When Tulving first speculated about episodic memory his emphasis was on the past, but in recent writings (e.g., Tulving, 2002) he has pointed out that a system capable of mental time travel backward should also be capable of time travel forward. When one thinks about memory from a functional perspective (see Klein et al., 2002), this extension into the future makes a great deal of sense. We use knowledge derived from prior experience to think about possible future states. Indeed, one might argue that such future-thinking is the most important function of the episodic memory system. It allows us to simulate what we might expect to happen should we behave in certain ways, and to use these predictions adaptively. In combination with systems that allow us to simulate the future behavior of others, no doubt based on our memories of how others have behaved in the past, this future-thinking capacity is critical to organizing our lives going forward, much as memory allows us to organize our past into a narrative that provides coherent structure to our life. Context plays a central role in all this by providing a scaffold that gives structure to our experiences that, after all, must always occur somewhere in space and time.
acknowledgments I gratefully acknowledge the contributions to this work made by many others over the years. My collaborator on the cognitive map theory of hippocampal function, John O’Keefe, got it all started. Jeff Willner collaborated in formulating original ideas about configural and elemental context that have withstood the test of time. Work on memory consolidation with Morris Moscovitch and Lee Ryan (funded by an NINDS grant to L.R. and L.N.) and on memory reconsolidation with Oliver Hardt, Almut Hupbach, and Rebecca Gomez (funded by an NSF CAREER award to R.G.) played an important part in my recent thinking. I am grateful to Oliver Hardt for reading and commenting on an earlier draft. Although I did not accept all his suggestions, what I did accept improved the chapter. I am grateful to Almut Hupbach for her comments on an earlier draft and for creating the figures. Finally, I am grateful to Kate Jeffery for pointing out my too-simple analysis of the states that place cell ensembles could achieve.
Notes 1. The term reconsolidation refers to the recent rediscovery that certain memory reminders have the effect of returning a previously ‘‘consolidated’’ memory to a
labile state. While I will have more to say about this later, it is critical at the outset to note that this term inadvertently leads to the wrong way of thinking about what is actually happening, both during ‘‘systems consolidation’’ and after a reminder. It would be better to use the terms memory transformation (rather than systems consolidation) and memory updating (rather than reconsolidation). The term consolidation should be reserved for what is now termed cellular consolidation. 2. ‘‘ . . . environmental contexts exist both as integrated ensembles (in the hippocampal map) and as collections of individual cues (in the neocortex)’’ (Nadel et al., 1985, p. 398). ‘‘In the (largely) neocortical systems, all the cues to be found within an environment are represented simply as things that have certain features, and these representations provide one means by which cues can be associated with each other. . . . Places, on the other hand, are higher order constructs elaborated in the hippocampal system. Here, all the cues in an environment are linked together such that their spatial relations are represented, and such that they act in concert’’ (Nadel et al., 1985, pp. 393–394). 3. Recent work shows that background context conditioning is also disrupted by lesions to the entorhinal cortex (Majchrzak et al., 2006). 4. The question of where within the hippocampal system this representation resides, and which part of the system provides the spatial frame, remains unclear. Recent work on so-called grid cells in the medial entorhinal cortex (Hafting et al., 2005) strongly suggests that the fixed spatial frame might be contributed by this entorhinal system, and that conjoining this frame with other inputs to the hippocampus proper leads to the creation of what we have called cognitive maps in the hippocampus itself. 5. ‘‘Upon entering a previously mapped environment, an organism identifies the place it is in by detecting a small set of things in a particular spatial arrangement. Appropriate spatial arrangements activate enough elements in one particular map ensemble to initiate place recognition. This recognition process involves the activation of a specific set of place cells, ultimately affecting activity within the entire ensemble of neural elements involved in the representation of that environment. This is not to say that all these elements are literally excited— merely that they are in a different state than when the organism is not in that environment. This altered state disposes the organism to interpret or act upon the environment in certain ways, reflecting the ‘expectations’ embodied in the environment-specific activation’’ (Nadel et al., 1985, p. 391). 6. The discussion here is overly simple, as a number of studies have shown that the hippocampus is not monolithic in its response to environmental change. Vazdarjanova and Guzowski (2004) have shown, for example,
THE HIPPOCAMPUS AND CONTEXT REVISITED
that the hippocampal subfields CA1 and CA3 have rather different responses to the same contextual change—in their case CA1 responded to change in a graded fashion whereas CA3 responded in a discontinuous fashion. It seems likely that within the hippocampus itself there exist mechanisms for both pattern completion and pattern separation, and that both of these dynamical responses can be observed in each of the primary CA fields, depending on the circumstance (see Guzowski et al., 2004). The discussion is also overly simple in implying that only two states of the system are possible: pattern completion or pattern separation. K. Jeffery and colleagues (see below for some discussion of her work) have elegantly shown that intermediate states are possible.
References Anagnostaras SG, Maren S, Fanselow, MS (1999) Temporally graded retrograde amnesia of contextual fear after hippocampal damage in rats: within-subjects examination. J Neurosci 19:1106–1114. Anagnostaras SG, Gale GD, Fanselow MS (2001) Hippocampus and contextual fear conditioning: recent controversies and advances. Hippocampus 11:8–17. Anderson MI, Killing S, Morris C, O’Donoghue A, Onyiagha D, Stevenson R, Verriotis M, Jeffery KJ (2006) Behavioral correlates of the distributed coding of spatial context. Hippocampus 16:730–742. Asratyan EA (1965) Compensatory Adaptations, Reflex Activity, and the Brain. Oxford: Pergamon Press. Balogh SA, Radcliffe RA, Logue SF, Wehner JM (2002) Contextual and cued fear conditioning in C57BL/6J and DBA/2J mice: context discrimination and the effects of retention interval. Behav Neurosci 116:947– 957. Beritoff JS (1965) Neural Mechanisms of Higher Vertebrate Behavior. Boston: Little Brown. Bolles RC (1985) A cognitive, nonassociative view of inhibition. In: Information Processing in Animals: Conditioned Inhibition (Miller RR, Spear NE, eds.), pp 355–367. Hillsdale, NJ: Lawrence Erlbaum Associates. Bontempi B, Laurent-Demir C, Destrade C, Jaffard R (1999) Time-dependent reorganization of brain circuitry underlying long-term memory storage. Nature 400:671–675. Burnham WH (1903) Retroactive amnesia: illustrative cases and tentative explanation. Am J Psychol 14: 382–396. Corcoran KA, Desmond TJ, Frey KA, Maren S (2005) Hippocampal inactivation disrupts the acquisition and contextual encoding of fear extinction. J Neurosci 25: 8978–8987. Debiec J, LeDoux JE, Nader K (2002) Cellular and systems reconsolidation in the hippocampus. Neuron 36: 527–538.
13
Fanselow MS (1999) Learning theory and neuropsychology: configuring their disparate elements in the hippocampus. J Exp Psychol Anim Behav Process 25: 275–283. Fanselow MS (2000) Contextual fear gestalt memories, and the hippocampus. Behav Brain Res 110:73–81. Fyhn M, Hafting T, Treves A, Moser EI, Moser M-B (2007) Hippocampal remapping and grid realignment in entorhinal cortex. Nature 446(7132):190–194. Gewirtz JC, McNish KA, Davis M (2000) Is the hippocampus necessary for contextual fear conditioning? Behav Brain Res 110:83–95. Good M, Honey RC (1991) Conditioning and contextual retrieval in hippocampal rats. Behav Neurosci 105: 499–509. Guzowski JF, Knierim JJ, Moser EI (2004) Ensemble dynamics of hippocampal regions CA3 and CA1. Neuron 44:581–584. Hafting T, Fyhn M, Molden S, Moser MB, Moser EI (2005) Microstructure of a spatial map in the entorhinal cortex. Nature 436:801–806. Hall G, Purves D, Bonardi C (1996) Contextual control of conditioned responding in rats with dorsal hippocampal lesions. Behav Neurosci 110:933–945. Hassabis D, Kumaran D, Vann SD, Maguire EA (2007) Patients with hippocampal amnesia can’t imagine new experiences. Proc Natl Acad Sci USA 104:1726–1731. Hayward A, McGregor A, Good MA, Pearce JM (2003) Absence of overshadowing and blocking between landmarks and the geometric cues provided by the shape of a test arena. Q J Exp Psychol 56B:114–126. Hebb DO (1949) The Organization of Behavior. New York: Wiley-Interscience. Hobin JA, Ji J, Maren S (2006) Ventral hippocampal muscimol disrupts context-specific fear memory retrieval after extinction in rats. Hippocampus 16:174–182. Holland PC, Bouton M (1999) Hippocampus and context in classical conditioning. Curr Opin Neurobiol 9:195– 202. Houston FP, Stevenson GD, McNaughton BL, Barnes CA (1999) Effects of age on the generalization and incubation of memory in the F344 rat. Learn Mem 6:111– 119. Hupbach A, Gomez R, Nadel L, Hardt O (2007) Reconsolidation of children’s episodic memories: a subtle reminder triggers integration of new information. Learn Mem 14:47–53. Hupbach A, Hardt O, Gomez R, Nadel L The dynamics of memory: context-dependent updating (Science, under review). James W (1890) The Principles of Psychology. Vol. 1. New York: Henry Holt. Jeffery KJ, Anderson MI (2003) Dissociation of the geometric and contextual influences on place cells. Hippocampus 13:868–872.
14 PLACE CELLS AND SPATIAL CONTEXT Jeffery KJ, Gilbert A, Burton S, Strudwick A (2003) Preserved performance in a hippocampal dependent spatial task despite complete place cell remapping. Hippocampus 13:175–189. Jeffery KJ, Anderson MI, Hayman R, Chakraborty S (2004) A proposed architecture for the neural representation of spatial context. Neurosci Biobehav Rev 28: 201–218. Kennedy PJ, Shapiro ML (2004) Retrieving memories via internal context requires the hippocampus. J Neurosci 24:6979–6985. Kim JJ, Fanselow MS (1992) Modality-specific retrograde amnesia of fear. Science 256:675–677. Klein SB, Cosmides L, Tooby J, Chance S (2002) Decisions and the evolution of memory: multiple systems, multiple function. Psychol Rev 109:306–329. Knierim JJ, Kudrimoti H, McNaughton BL (1998) Interaction between idiothetic cues and external landmarks in the control of place cells and head direction cells. J Neurosci 18:425–446. Konorski J (1967) Integrative Activity of the Brain: An Interdisciplinary Approach. Chicago: University of Chicago Press. LaBar KS, Phelps EA (2005) Reinstatement of conditioned fear in humans is context dependent and impaired in humans. Behav Neurosci 119:677–686. Land C, Bunsey M, Riccio DC (2000) Anomalous properties of hippocampal lesion-induced retrograde amnesia. Psychobiology 28:476–485. Lehmann H, Carfagnini A, Yamin S, Mumby DG (2005) Context-dependent effects of hippocampal damage on memory in the shock-probe test. Hippocampus 15: 18–25. Lenck-Santini P-P, Rivard B, Muller RU, Poucet B (2005) Study of CA1 place cell activity and exploratory behavior following spatial and nonspatial changes in the environment. Hippocampus 15:356–369. Leutgeb JK, Leutgeb S, Treves A, Meyer R, Barnes CA, McNaughton BL, Moser M-B, Moser EI (2005) Progressive transformation of hippocampal neuronal representations in ‘‘morphed’’ environments. Neuron 48:345–358. Lever C, Burton S, O’Keefe J (2006) Rearing on hind legs, environmental novelty, and the hippocampal formation. Rev Neurosci 17:111–133. Lewis DJ (1979) Psychobiology of active and inactive memory. Psychol Bull 86:1054–1083. Loftus EF (2005) Planting misinformation in the human mind: a 30-year investigation of the malleability of memory. Learn Mem 12:361–366. MacArdy EA, Riccio DC (1995) Time-dependent changes in the effectiveness of a non-contingent footshock reminder. Learn Motiv 26:29–42. Majchrzak M, Ferry B, Marchand AR, Herbeaux K, Seillier A, Barbelivien A (2006) Entorhinal cortex
lesions disrupt fear conditioning to background context but spare fear conditioning to a tone in the rat. Hippocampus 16:114–124. Maren S, Holt W (2000) The hippocampus and contextual memory retrieval in Pavlovian conditioning. Behav Brain Res 110:97–108. Matus-Amat P, Higgins EA, Barrientos RM, Rudy JW (2004) The role of the dorsal hippocampus in the acquisition and retrieval of context memory representations. J Neurosci 24:2431–2439. McAllister WR, McAllister DE (2006) Recovery of conditioned fear by a single postextinction shock: effect of similarity of shock contexts and of time following extinction. Learn Behav 34:44–49. Misanin JR, Miller RR, Lewis DJ (1968) Retrograde amnesia produced by electroconvulsive shock after reactivation of a consolidated memory trace. Science 160:554–555. Moscovitch M, Nadel L, Winocur G, Gilboa A, Rosenbaum S (2006) The cognitive neuroscience of remote episodic, semantic and spatial memory. Curr Opin Neurobiol 16:179–190. Muller GE, Pilzecker A (1900) Experimentale Beitrage zur Lehre vom Gedaechtnis. Zeit Psychol 1:1–288. Nadel L (1968) Dorsal and ventral hippocampal lesions and behavior. Physiol Behav 3:891–900. Nadel L (2007) Memory consolidation: in search of a paradigm. In: Science of Memory: Concepts (Roediger HL, Dudai Y, Fitzpatrick S, eds.). Oxford: Oxford University Press. Nadel L (in press) Multiple memory systems: a new view. In: Learning and Memory: A Comprehensive Reference (Byrne J, ed). London: Elsevier Press. Nadel L, Hupbach A (2005) Cross-species comparisons in development: the case of the spatial ‘‘module.’’ In Attention and Performance XXI (Johnson, MH, Munakata Y, eds.). Oxford: Oxford University Press. Nadel L, Moscovitch M (1997) Consolidation, retrograde amnesia and the hippocampal formation. Curr Opin Neurobiol 7:217–227. Nadel L, Moscovitch M (1998) Hippocampal contribution to cortical plasticity. Neuropharmacology 37: 431–440. Nadel L, Willner J (1980) Context and conditioning: a place for space. Physiol Psychol 8:218–228. Nadel L, Willner J, Kurz EM (1985) Cognitive maps and environmental context. In: Context and Learning (Balsam P, Tomie A, eds.), pp 385–406, Hillsdale, NJ: Lawrence Erlbaum Associates. Nadel L, Winocur G, Ryan L, Moscovitch, M (2007) Systems consolidation and hippocampus: two views Debates in Neuroscience, online. Nader K, Schafe GE, LeDoux JE (2000) Fear memories require protein synthesis in the amygdala for reconsolidation after retrieval. Nature 406:722–726.
THE HIPPOCAMPUS AND CONTEXT REVISITED
O’Keefe J (1976) Place units in the hippocampus of the freely moving rat. Exp Neurol 51:78–109. O’Keefe J, Dostrovsky J (1971) The hippocampus as a spatial map. Preliminary evidence from unit activity in the freely moving rat. Brain Res 34:171–175. O’Keefe J, Nadel L (1978) The Hippocampus as a Cognitive Map. Oxford: Clarendon Press. O’Keefe J, Nadel L (1979) Pre´cis of O’Keefe and Nadel’s The Hippocampus as a Cognitive Map, and author’s response to commentaries. Behav Brain Sci 2 487– 534. O’Keefe, J., Nadel, L. and Willner, J. (1979) Tuning out irrelevancy? Comments on Solomon’s temporal mapping view of hippocampus. Psych Bull 86: 1280–1289. Penick S, Solomon PR (1991) Hippocampus, context, and conditioning. Behav Neurosci 105 611–617. Phillips RG, LeDoux JE (1992) Differential contribution of amygdala and hippocampus to cued and contextual fear conditioning. Behav Neurosci 106:274–285. Phillips RG, LeDoux JE (1994) Lesions of the dorsal hippocampal formation interfere with background but not foreground contextual fear conditioning. Learn Mem 1:34–44. Przybyslawski J, Sara SJ (1997) Reconsolidation after reactivation of memory. Behav Brain Res 84:241– 246. Rescorla RA, Wagner AR (1972) A theory of Pavlovian conditioning: variations in the effectiveness of reinforcement and nonreinforcement. In: Classical Conditioning II: Current Research and Theory (Black AH, Prokasy WF, eds.), pp 64–99. New York: Appleton-Century-Crofts. Rudy JW (1996) Post-conditioning isolation disrupts contextual fear conditioning: an experimental analysis. Behav Neurosci 110:238–246. Rudy JW, Barrientos RM, O’Reilly RC (2002) The hippocampal formation supports conditioning to memory of a context. Behav Neurosci 116:530–538. Rudy JW, Huff NC, Matus-Amat P (2004) Understanding contextual fear conditioning: insights from a twoprocess model. Neurosci Biobehav Rev 28:675–685. Rudy JW, O’Reilly RC (1999) Contextual fear conditioning, conjunctive representations, pattern completion, and the hippocampus. Behav Neurosci 113:867– 880. Rudy JW, O’Reilly RC (2001) Conjunctive representations, the hippocampus, and contextual fear conditioning. Cogn Affect Behav Neurosci 1:66–82. Rudy JW, Pugh CR (1998) Time of conditioning selectively influences contextual fear conditioning: further
15
support for a multiple-memory systems view of fear conditioning. J Exp Psychol Anim Behav Process 24:316–324. Samsonovich A, McNaughton BL (1997) Path integration and cognitive mapping in a continuous attractor neural network model. J Neurosci 17:5900–5920. Samsonovich A, Nadel L (2005) Fundamental principles and mechanisms of the conscious self. Cortex 41:669– 689. Sara SJ (2000) Retrieval and reconsolidation: toward a neurobiology of remembering. Learn Mem 7:73–84. Sato C, Matsumoto Y, Sakura M, Mizunami M (2006) Contextual olfactory learning in cockroaches. NeuroReport 17:553–557. Save E, Buhot MC, Foreman N, Thinus-Blanc C (1992a) Exploratory activity and response to a spatial change in rats with hippocampal or posterior parietal cortical lesions. Behav Br Res 47:113–127. Save E, Poucet B, Foreman N, Buhot MC (1992b) Object exploration and reactions to spatial and nonspatial changes in hooded rats following damage to parietal cortex or hippocampal formation. Behav Neurosci 106: 447–456. Scoville WB, Milner B (1957) Loss of recent memory after bilateral hippocampal lesion. J Neurol Neurosurg Psychiatry 20:11–21. Selden, N.R.W., Everitt, B.J., Jarrard, L.E. & Robbins, T.W. (1991) Complementary roles for the amygdala and hippocampus in aversive conditioning to explicit and contextual cues. Neuroscience 42:335–350. Smith DM, Mizumori SJY (2006) Hippocampal place cells, context, and episodic memory. Hippocampus 16: 716–729. Tulving E (2002) Episodic memory: from mind to brain. Annu Rev Psychol 53:1–25. Vazdarjanova A, Guzowski JF (2004) Differences in hippocampal neuronal population responses to modifications of an environmental context: evidence for distinct, yet complementary, functions of CA3 and CA1 ensembles. J Neurosci 24:6489–6496. Wills TJ, Lever C, Cacucci F, Burgess N, O’Keefe J (2005) Attractor dynamics in the hippocampal representation of the local environment. Science 308: 873–876. Wiltgen BJ, Silva AJ (2007) Memory for context becomes less specific with time. Learn Mem 14:313–317. Winocur G, Moscovitch M, Sekeres M (2007) Memory consolidation or transformation: context manipulation and hippocampal representations of memory. Nat Neurosci 10:555–557.
2 A Context for Hippocampal Place Cells during Learning SHERI J. Y. MIZUMORI
cell and interneuron discharge is modulated by the speed of movement through the environment (e.g., McNaughton et al., 1983). Thus, it was clear from early on that while visual input is important, it is by no means the only determinant of place fields. As evidenced by most of the chapters in this volume, a popular current view is that place fields are also shaped by very specific prior experiences of animals. That is, memory (or memory retrieval) plays an important role in determining the way place fields respond to current sensory and behavioral information. Here, a context discrimination hypothesis postulates that memory inputs are important since they define what one expects to find in a given situation, or context. Moreover, memory biases one’s perception of the current contextual features. By comparing the expected and current contextual features, the hippocampus can identify changes in context, resulting in increased exploration and/ or attention that facilitates new learning.
It has been decades since the initial description of hippocampal place fields (O’Keefe and Dostrovsky, 1971; Ranck, 1973), and a number of perspectives have been offered to account for their functional significance (e.g., O’Keefe, 1976; O’Keefe and Nadel, 1978; O’Mara, 1995; McNaughton et al., 1996; Muller et al., 1996; Wiener, 1996; Redish and Touretzky, 1997; Mizumori et al., 1999; Poucet et al., 2000; Best et al., 2001; Sharp, 2002; Jeffery, 2003, Jeffery et al., 2004; Smith and Mizumori, 2006). Understandably, researchers initially sought to characterize place fields according to metrics of the visual–spatial environment. These experiments typically involved recording from freely behaving rats as they repeatedly foraged for food in constrained environments such as open arenas or elevated mazes. A dominating influence of visual information over place fields was demonstrated when the rotation of distal cues resulted in a corresponding shift in place fields. (e.g., O’Keefe and Conway, 1978; Muller and Kubie, 1987; Lenck-Santini et al., 2001). Despite the clear sensitivity of place fields to visual information, there was mounting evidence that place fields represent information beyond the visual domain. For example, pyramidal (place) cells firing can be tuned to nonspatial information such as olfactory (Eichenbaum et al., 1999; Save et al., 2000), auditory (O’Keefe and Conway, 1978; Sakurai, 1994; McEchron and Disterhoft, 1999), and somatosensory (Young et al., 1994; Tanila et al., 1997) cue information. Hippocampal neural firing is also differentially modified by a variety of behavioral conditions, e.g., pyramidal
HIPPOCAMPAL PLACE FIELDS REPRESENT INTEGRATED, EXPERIENCE-DEPENDENT SENSORY AND BEHAVIORAL INFORMATION Place fields exhibit a heterogeneous collection of neural responses to changes in sensory input. While many place fields rapidly shift locations or in-field firing rates in response to changes in familiar visual environments, others appear unchanged even when visual input is eliminated (Hill and Best, 1981, Quirk et al., 16
HIPPOCAMPAL PLACE CELLS DURING LEARNING
1990; Muller et al., 1991; Markus et al., 1994). This collective pattern of responses is referred to as a partial reorganization of neural activity, since only a subpopulation of cells changed their responses. (Complete reorganization occurs when the entire population of cells shows significantly changed responses.) That place fields were found to persist after environmental changes, yet initially required visual input to become established (e.g., Mizumori et al., 1999), provided some of the first evidence that past experience (i.e., memories) guides place cell firing. That is, place fields reflect learned associations between visual and nonvisual information. In cases of either partial or complete reorganization, the new population activity pattern presumably reflects the processing of information relevant to new experimental conditions. A consequence of the association between visual and nonvisual information is that place fields are able to quickly shift their reliance between different types of cues as needed. As an example, place fields rapidly shift from being dependent on visual cues to being dependent on nonvisual cues (Wiener et al., 1989). Such a feature could certainly contribute to the adaptive and flexible processing attributed to the hippocampus. Another example that place fields reflect multisensory associations is that internal sensory cues may become associated with visual information. As an example, the location-selective firing by place cells often varies according to nonvisual, egocentric movement variables, such as velocity and acceleration of an animal’s movement (Hill and Best, 1981; McNaughton et al., 1983; Markus et al., 1994; Knierim et al., 1995, 1998; Wiener et al., 1995; Gavrilov et al., 1998). Since place fields represent polymodal information, it is possible for place fields to be maintained in the absence of familiar visual cues based on calculations of the present position from information about recent velocity, direction of movement, and distance traveled from a known start position, a process called path integration, or dead reckoning (Mittelstaedt and Mittelstaedt, 1982; McNaughton et al., 1996; Whishaw and Gorny, 1999; Etienne and Jeffery, 2004). Some of the directional and angular velocity information needed for path integration appears to be derived from neural circuitry involving the tegmentum, mammillary nuclei, the anterior thalamus, subiculum, and retrosplenial cortex (Mizumori and Williams, 1993; Taube and Muller, 1998; Sharp et al., 2001). It is within this circuitry that a neural code for the directional heading of animals is generated and represented by head direction cells. Inhibitory interneurons of the hippocampus are commonly referred to as theta cells because many of them fire with temporal regularity according to the ongoing theta rhythm. Like place cells, these neurons are also sensitive to the velocity and acceleration of
17
translational movements by the rat (Vanderwolf, 1969; Ranck, 1973; Fox and Ranck, 1981; Buzsaki, 2002). Thus, velocity coding appears more commonplace than location or directional codes. Interestingly, the extent to which place cells are sensitive to idiothetic cues is reported to systematically decline from the septal pole to the temporal pole of the hippocampus (Maurer et al., 2005). Coincident with this trend is the finding that place fields become increasingly larger for place cells recorded along the dorsal–ventral axis (e.g., Jung et al., 1994). Thus, place field selectivity appears related to the degree of sensitivity to idiothetic cues, which implies that movement ‘‘sharpens’’ the spatial code in the hippocampus. Since place cells are impacted by specific aspects of ongoing behavior, behavioral responses should be ‘‘clamped’’ when studying the relationship between place fields and memory functions per se. That is, identical behaviors should be required for the phases of the task that are to be compared. In such test situations, it has been shown that manipulating cognitive demand per se can alter place fields such that their expression is conditional upon the recent (i.e., retrospective coding) or upcoming (i.e., prospective coding) behavioral sequences, or response trajectories (Frank et al., 2000; Wood et al., 2000; Ferbinteanu and Shapiro, 2003). The expectation of a future goal location appeared to influence some place fields, while the exhibition of place fields by other cells depended on recent trajectories taken by a rat on the maze. Eschenko and Mizumori (2007) showed an explicit influence of cognitive/memory demand that was independent of behavioral trajectories. They showed that place fields are also guided by the cognitive strategy (e.g., spatial or nonspatial) used to solve a plus-maze task when behavioral and sensory conditions are held constant. Shapiro and Ferbinteanu (2006) further showed that temporally correlated firing is related to the prospective codes of hippocampal neurons. Thus, a memory-related code occurs within not only the spatial character of cell discharge but also its temporal character. The correlated firing may be mediated by a spike timing-dependent mechanism (Markram et al., 1997; Bi and Poo, 1998; Dan and Poo, 2006) that varies with task phase (i.e., trial type, or ‘‘journeys’’). Such temporally regulated neural codes (described in more detail below) are likely to be an essential component of a hippocampal contribution to episodic learning and memory. Despite the fascinating response properties that have been reported for place cells, understanding the specific contributions of these responses to hippocampaldependent learning remains a significant challenge. The following argues that (similar to what was initially described in 1978 by O’Keefe and Nadel) it may be
18 PLACE CELLS AND SPATIAL CONTEXT possible to bridge place field results and theories of hippocampal-dependent learning by focusing on what appears to be a critical role for the hippocampus in context processing.
CONTEXT DISCRIMINATION REGULATES ENCODING AND RETRIEVAL FUNCTIONS BY HIPPOCAMPUS Significant animal research implicates the hippocampus in processing contextual information (for excellent reviews see Hirsh, 1974; Myers and Gluck, 1994; Anagnostaras et al., 2001; Maren, 2001; Fanselow and Poulos, 2005). Rats with hippocampal damage do not exhibit conditioned fear responses to contextual stimuli, even though responses to discrete conditional stimuli remain intact (Kim and Fanselow, 1992; Phillips and LeDoux, 1992, 1994). Also, hippocampal or entorhinal cortical damage produces an insensitivity to changes in the context, as evidenced by the fact that lesioned animals do not show the normal decrement in conditioned responding when the context is altered (Penick and Solomon, 1991; Freeman et al., 1996a,b). Fornix-lesioned subjects showed impaired learning of two different auditory discrimination tasks when they occurred in separate contexts (Smith et al., 2004). In the same animals, context-specific neural firing patterns were degraded in structures that normally receive hippocampal input via the fornix (e.g., anterior thalamus and cingulate cortex). Furthermore, manipulations that impact hippocampal synaptic plasticity (e.g., long-term potentiation [LTP]) also affect context learning (e.g., Shors and Matzel, 1997). Findings such as these converge on a hypothesis that hippocampal processing serves to facilitate the discrimination of meaningful contexts. Views that the hippocampus mediates the flexible use of conjunctive, sequential, relational, and spatial information (e.g., O’Keefe and Nadel, 1978; Foster et al., 1987; Eichenbaum et al., 1999; Wood et al., 2000; Eichenbaum and Cohen, 2001; O’Reilly and Rudy, 2001; Fortin et al., 2002) are not inconsistent with a context-processing account. In fact, these flexible, spatial, sequential, and relational operations likely enable the hippocampus to make accurate context discriminations, since central to the ability to make context discriminations is determination of the extent to which a learned (or expected) context has changed, or the extent to which contexts are sufficiently distinct that they should be considered different. In both cases, flexible consideration of sequences of sensory and motor information and/or the significance of (or relationships among) these types of information is very critical. A larger challenge is to understand how a
context-processing theory of hippocampus accounts for the numerous hippocampal place cell responses that have been reported. A context discrimination hypothesis (CDH) provides a theoretical framework that can account for a significant amount of place cell findings as well as help bridge the growing body of work on neuroplasticity of place representation to current ideas about the specific role of the hippocampus in learning and episodic memory (Tulving, 2002). While many features of this hypothesis build on concepts discussed by other investigators (as pointed out below), its consideration here provides an opportunity to elaborate on specifically the functional networks that should exist within the neural organization of hippocampus. Furthermore, the CDH makes clear predictions about how hippocampal efferent messages ultimately come to impact ongoing behavior. The CDH postulates that a fundamental operation of the hippocampus, one that ultimately defines its contribution to complex forms of learning, is the discrimination of meaningful contexts (Mizumori et al., 1999, 2000a, 2007; Smith and Mizumori, 2006). Hippocampal place cells may represent aspects of the current context (O’Keefe and Nadel, 1978; Nadel and Wilner, 1980; Nadel and Payne, 2002) for the purpose of determining the extent to which familiar contexts have changed, perhaps by performing a match–mismatch comparison (e.g., Gray, 1982, 2000; Vinogradova, 1995; Mizumori et al., 1999, 2000a; Lisman and Otmakhova, 2001; Hasselmo et al., 2002; Anderson and Jeffery, 2003; Jeffery et al., 2004, Hasselmo, 2005b; Manns et al., 2007) whereby the present context is evaluated according to how similar it is to the context that an animal is expecting, based on past experience. Detected mismatches can be used to identify novel contexts and distinguish different contexts, functions that are necessary to define significant events or episodes. As described above, a growing literature involving a wide range of behavioral paradigms (e.g., classical conditioning, instrumental conditioning, navigationbased learning) has revealed the importance of the hippocampus for context-dependent learning. More specifically, converging evidence indicates that the hippocampus serves to identify changes in context, presumably because of the need to engage cellular mechanisms for new learning when contexts change (Paulsen and Moser, 1998). Smith and Mizumori (2006) showed that hippocampal neurons develop context-specific responses, but only if the task required rats to discriminate contexts. Discriminating neural responses were not observed when rats were allowed to randomly forage for the same amount of time. Most recently Manns and colleagues (2007) showed that relative to match trials in an odor cue or object recognition task, CA1 neurons preferentially
HIPPOCAMPAL PLACE CELLS DURING LEARNING
discharged when animals experienced a nonmatch situation in these same tasks. Moreover, the firing tended to occur during the encoding phase of the ongoing theta rhythm. Thus, detection of a nonmatch situation changes the relationship between cell discharge and the local theta rhythm such that encoding functions are enhanced. Detection of matches may result in efferent messages that continue to retrieve and use the currently active memory network. Also supporting the CDH, disconnecting hippocampus by fornix lesions impairs context discrimination (Smith et al., 2004), and hippocampal lesions reduce animals’ ability to respond to changes in a familiar environment (Good and Honey, 1991; Save et al., 1992a,b). Spatial novelty detection corresponds to selective elevation of the immediate early gene c-fos in hippocampus, and not in surrounding parahippocampal cortical regions (Jenkins et al., 2004). Hippocampal neurons show significantly altered firing patterns when rats experience spatial or nonspatial changes in a familiar environment (e.g., O’Keefe, 1976; Wible et al., 1986; Muller and Kubie, 1987; Wood et al., 1999; Fyhn et al., 2002; Ferbinteanu and Shapiro, 2003; Moita et al., 2004; Yeshenko et al., 2004; Leutgeb et al., 2005a,b; Puryear et al., 2006; Smith and Mizumori, 2006; Eschenko and Mizumori, 2007).The specific patterns of neural change following alterations in different features of an experimental context tend to be multidimensional and complex. After considering the diversity of neural responses and test conditions that have been reported, we suggest that there are (at least) three types of influences on hippocampal neural codes that occur automatically regardless of the specific task demands: memory, spatial, and temporal. This proposal is based in part on past suggestions that hippocampal-dependent memory involves associations of spatial and temporal features (e.g., Mizumori et al., 2000a; Redish et al., 2000; Burgess et al., 2001; Eichenbaum and Cohen, 2001; Morris, 2001; O’Reilly and Rudy, 2001; Buzsa´ki, 2005). A slight variation of these proposals, however, is the suggestion that the default mode of hippocampal processing is to continually integrate memory-guided perceptions of sensory, movement, and motivational information, or memory (M), within a spatial (S) reference framework as a function of time (T). An important point here is that navigation predisposes the hippocampus to receive highly preprocessed M information for the purpose of integration with spatial and temporal domains. The integration may occur within a hierarchically organized framework that incorporates M, S, and T information as fundamental inputs. As a result, during unrestrained navigation, place fields emerge as neural representations of different types of sensory, behavioral, and intrinsic
19
information that have strong spatial and temporal features—that is, spatial context representations. The CDH, then, suggests a more integrated definition of context than is typically used in the learning literature—a definition that emphasizes the integration of sensory, motivational, response, and memorial input for all hippocampal cell types and not just the passive reflection of the static background external sensory environment by pyramidal neurons
HIERARCHICALLY ORGANIZED CONTEXT PROCESSING WITHIN HIPPOCAMPUS Previous hierarchical models of hippocampal processing consider the fundamental elements of the hierarchy to be either stimulus-defining information (e.g., Shapiro et al., 1997) or individual events (e.g., Eichenbaum et al., 1999; Shapiro and Eichenbaum, 1999). Temporal organization plays a large role in connecting sequences of events to generate representations of complete episodes. Thus, the primary goal for the hippocampus is to identify sequences of events. Flexibility of processing emerges from proposed interconnections between representations. The top of the hierarchy reflects a process whereby common features across episodes are extracted to determine the static features (Wood et al., 1999). Noticeably absent in such hierarchical schemes is the influence of spatial information. Also, a role for memory influences is not clearly defined. Figures 2–1 and 2–2 present a hierarchical model that includes spatial information (S) as a form of organization rather than specific environmental features that individual cells discriminate. Furthermore, we propose that it is the memory input (M) from extrahippocampal systems that brings preselected task-relevant sensory, behavioral, motivational, and learned information into hippocampus. Finally, a temporal organization of information (T) emerges as a natural consequence of synaptic plasticity mechanisms that assist in the discrimination of contexts. That is, according to our model, temporal organization is not the ultimate goal of hippocampal processing but rather is a fundamental organizing property that emerges via the intrinsic synaptic plasticity properties of neurons. Other neural networks in brain likely also organize information temporally, but for different reasons. According to CDH, then, meaningful epochs are defined by feedback from activated memory circuits. Consideration of the details and significance of place field representations within a hierarchically organized scheme reveals possible relationships among three general levels of integration (Fig. 2–1). The basic features are summarized first, followed by a more detailed explanation of these features.
Figure 2–1. A hierarchical representation of context processing within the hippocampus. Each square corresponds to a context matrix that reflects multiple types of information coded by single cells (level 1), local ensembles (level 2), or large regions of hippocampus (level 3). Place cell studies reveal that a number of specific features (e.g., sensory, behavioral, motivation, and knowledge of task rules; see Fig. 2–2) are represented in hippocampal networks according to past experience or memory (M). The M input to place cells may vary in strength depending on the learned significance of the information that M represents. Stronger weights (influences) are depicted by thicker lines that connect the different levels. As a result, some M inputs may be biased to present current context information to hippocampal networks, while other M inputs may be influenced more by expectations for a given context, such as memory for task rules. Another factor that continually shapes the organization of information in the hippocampus is the spatial reference framework (S) provided by entorhinal cortex grid cells (described in text). A third continual influence on hippocampal place fields is one that strives to organize incoming information as a function of time (T), both retrospectively and prospectively. The location of the peak of the cylinder within the three-dimensional matrix identifies for each cell the relative contributions of M, S, and T input. Level 2 neural integration reflects ensemble activity of local networks of cells. Presumably this is where representations by single place cells are integrated to define portions of the expected and current context within local ensembles. In this example, cells 1 and 3 combine to define a part of what will become a component of the 20
HIPPOCAMPAL PLACE CELLS DURING LEARNING
Level 1 Organization: Place Cell Codes 1. Place cells receive highly preprocessed information in the form of representations of memory-guided perceptions of external and internal states (M). 2. Place cells consider M information within a spatial framework (S) that reflects the current spatially extended environment, giving rise to the common observation of location-selective firing in freely-behaving animals. 3. Intrinsic and extrinsic hippocampal neural mechanisms also organize M information as a function of time (T) by regulating spike timing relative to sensory stimulation, the discharge of other cells, or the EEG. Such temporal processing likely underlies sequence coding. It also may contribute to a higher level of temporal function that predicts future behaviors (a prospective code) or codes recent behaviors (a retrospective code).
Level 2 Organization: Local Circuit Code 1. Local neural circuits code for and further integrate context information from individual place cells in preparation for context comparison computations at level 3. 2. Different local circuits may code for different aspects of the expected or current context.
Level 3 Organization: Population Efferent Code 1. Local circuit information provides input to a larger population computation that determines
21
the degree of similarity between what is expected to occur in a given context and what is actually happening. 2. Hippocampal output impacts neocortical memory systems, which in turn impact subsequent M input to place cells. Detected mismatches engage cellular mechanisms that support new learning, while a match may effectively strengthen synaptic connections within currently active memory circuits.
Level 1 Organization: Place Cell Codes Memory-Guided Perceptions In addition to reflecting sensory, behavioral, and reinforcement information, there were early indications that memory plays a role in shaping place field characteristics. Mentioned earlier were the reports of visually sensitive place fields that persist when familiar visual information was no longer present. Also, O’Keefe and Speakman (1987) described place fields that appeared to shift to locations where a rat made errors, suggesting that the expectation of rewards guided place field location. Wood et al. (2000) showed that the exhibition of place fields can be made conditional upon an animal’s past or planned behavioral actions. More recently, Ferbinteanu and Shapiro (2003) showed a relationship between prospective coding and correct choices on a maze. According to the CDH, the combined activity profiles of hippocampal pyramidal cells (and interneurons) represent an experience-dependent context code that is unique for a particular test situation. Individual place fields have been shown to reflect simultaneously combinations of context-defining features such as spatial information (i.e., location and heading direction
3 definition of the expected features of a context by local circuit 1. Cell 3 has stronger influence than cell 2, perhaps reflecting a stronger history of synaptic activation. Cells 2 and 4 similarly combine to define another portion of the expected context. Cells 5 and 7 combine to define a segment of the current context, with cell 7 contributing to a greater extent than cell 5. The greater influence at level 2 is depicted by the taller cylinder. Finally, cells 6 and 8 combine to form another local ensemble that processes current context information. Presumably there are many such local circuits throughout hippocampus, and many more than two individual place cells contribute to each local ensemble. Also, given the known massive interconnection between cells in a specific hippocampal subregion, it is likely that the activity of one ensemble impacts the activity of connected ensembles (not shown). Level 3 integration compares expected and current context ensemble representations to produce an efferent code that indicates the extent to which the expected context has changed. If the ensemble codes for the expected and current context are the same, the expected and current context matrices should overlap as in the present case (match). (See text for discussion of a role for pattern completion in the process of context comparison). When a match is detected, the output message of the hippocampus may signal the continuation of the ongoing behavior, and it may strengthen synaptic connections within the neural network that defines the current activated memory.
Figure 2–2. Place cells are known to be responsive to multiple types of information. The context discrimination hypothesis postulates that the hippocampus receives highly preprocessed information that reflects a perception of sensory stimuli (external and internal), appropriate actions, and task rules according to past experience (or memory, M). As animals explore their environment, the entorhinal cortex grid cells automatically provide the hippocampus with a spatial reference frame (S) within which contextual information can be placed. A third organizational influence on place field codes is temporal (T) in nature. T factors organize spike firing relative to the firing of other cells, salient stimuli, or behavioral responses. As a result, hippocampal neurons appear to code sequences of sensory and response information, and relay information about stimulus or response duration. Because of these three organizational factors (M, S, and T), hippocampal neurons are predisposed (at least upon initial exposure to a new situation or context) to display place fields regardless of the whether the task is hippocampal dependent or not. As in Figure 2–1, the width of the lines reflects the relative strengths of the different input. Cell 1 shows a strong retrospective (T) component and moderately strong influence from S and M. Cell 2 shows strong S influences, with comparatively weaker M input. Together with moderate prospective coding, cell 2 seems biased to represent spatial information in the near future. Cell 3 shows a cell with strong M input that could promote the observed strong prospective coding. The degree of spatial coding is only moderate, suggesting stronger influences of nonenvironmental factors in M. Finally, cell 4 reflects a code for learned, nonspatial information in a retrospective manner.
22
HIPPOCAMPAL PLACE CELLS DURING LEARNING
within a spatial reference framework), consequential information (i.e., reward), movement-related information (i.e., velocity and acceleration—determinants of response trajectory), external (nonspatial) sensory information, the nature of the currently active memory (defined operationally in terms of task strategy and/or task rules), and the current motivational state (Fig. 2– 2). The relative strengths of the different features may vary from cell to cell and from situation to situation. As discussed below, with sufficient training, specific inputs may be strengthened while others are made weaker. Experimental results provide evidence for this multidimensional conceptualization of place fields. For example, individual cells have been shown to exhibit a place correlate during the performance of one task and a nonspatial correlate during the performance of a different task (e.g., Wiener et al., 1989). Also, cell firing that was correlated with egocentric movement (e.g., movement velocity) during behavioral performance in one task no longer reflected egocentric movement when the mnemonic (not behavioral) response demands of the task changed (e.g., Yeshenko et al., 2004; Eschenko and Mizumori, 2007). Further evidence that hippocampal neurons encode simultaneously different types of information is that reversible inactivation of afferent information from the lateral dorsal thalamus (where head direction cells are found) did not merely eliminate directional firing within place fields in the hippocampus (Mizumori et al., 1994, 2004a). Rather, place fields changed by relocating, by revealing themselves for the first time in a session, or simply disappearing. This implies that place fields reflect such highly integrated information that loss of a single type of input changes the nature of the relationship among the other types of input. Also supporting the view that hippocampal pyramidal (place) neurons code different kinds of information is that they are sensitive to a variety of types of sensory and response information when rats are tested in different tasks. These cells respond in task-relevant ways during olfactory (e.g., Wiener et al., 1989) or auditory discrimination testing (e.g., Sakurai, 1994). Finally, a variety of correlate types are also observed for hippocampal pyramidal cells during the performance of a single task (e.g., Gothard et al., 1996; Smith and Mizumori, 2006). The varied categories of input that combine to form context-specific M-based perceptions of a given situation are illustrated in Figure 2–2. Research has shown that each one of these types of information has different degrees, or weights, of influence on the firing of simultaneously recorded single cells. For example, some cells may show striking place fields as the primary behavioral correlate with firing that is secondarily modulated movement velocity. Other cells may
23
show significant modulation by velocity regardless of the animal’s location, while also showing significant though less impressive location selectivity. Either one of these variables can be modified in terms of strength according to cognitive (not behavioral) demands (e.g., Eschenko and Mizumori, 2007). Regardless of the particular combination of inputs, it appears that Minformation is continuously and automatically processed in the hippocampus irrespective of the task.
M Inputs within a Spatial Reference Framework A long-standing issue is the extent to which spatial features dominate hippocampal processing. In addition to a strong influence of long-term memory circuitry, there is a clear strong predisposition to organize M information spatially in the form of spatially localized firing (i.e., place fields). The prominence of the place field correlate led to the somewhat intuitively appealing idea that the hippocampus supports cognitive map functions (O’Keefe and Nadel, 1978). However, findings of nonspatial correlations to hippocampal cell discharge, combined with evidence that place fields are modulated by nonspatial factors, supported an alternative hypothesis that a fundamental organizing principle is the temporal relationship among sensory or behavioral events (Olton et al., 1979; Eichenbaum et al., 1999). Both views are likely to be correct to some extent. However, either view alone appears too exclusive. The fact that spatial representations have been reported as the predominant form of representation in the hippocampus may be due in part to the way in which place cell experiments with rodents are structured. These experiments invariably involve exploration of the environment by subjects. Voluntary movement may direct spatial and movement signals through the medial entorhinal cortex such that networks of spatial grid cells (Hafting et al., 2005; Leutgeb et al., 2005b; Sargolini et al., 2006) become activated. Grid cells selectively discharge when rats traverse positions that coincide with the vertices of a triangular tessellating grid that spans the environment. The grid fields appear to represent conjunctions of location, direction, and movement information within a two-dimensional coordinate representation of the environment. Thus, the entorhinal cortex may pass on to the hippocampus a spatial framework (McNaughton et al., 2006) in which to organize and process context-specific M and T information. For this reason, the term spatial context is appropriate when discussing context processing during navigation (Nadel et al., 1985; Mizumori et al., 1999, 2000a; Jeffery et al., 2004). To summarize, voluntary movement may predispose a significant number
24 PLACE CELLS AND SPATIAL CONTEXT of hippocampal neurons to organize context-defining information in the form of location-selective firing. The notion of exploration-induced spatial organization of hippocampal representations is consistent with the finding that hippocampal place fields are observed upon first exposure to a new environment (Muller and Kubie, 1987; Wilson and McNaughton, 1993; Markus et al., 1995; O’Keefe and Burgess, 1996a,b; Hetherington and Shapiro, 1997; Frank et al., 2004). Also, compared to passive movement conditions in which rats are made to go through a place field by being either held by the experimenter or placed on a moveable robotic device, active and unrestrained movement seems to generate more selective and reliable place fields (Foster et al., 1989; Gavrilov et al., 1998, Song et al., 2005). Furthermore, pyramidal cells fire more robustly when rats run faster across a given location (i.e., running in a running wheel (Czurko et al., 1999). Voluntary locomotion, then, appears to have the effect of sharpening the neural image of spatial representations, at least in dorsal hippocampus. A sharper code should relay more specific spatial information to intrinsic and extrinsic computational circuits. The finding that unrestrained movement produced sharper codes than did passive movement (under conditions that allow rats to experience the same directional and velocity information) suggests that learned behavioral responses have more meaning for, and impact on, place fields than movement of the animal per se. This may be one way in which learned information helps to define a context code. The tendency for voluntary navigation to impose a spatial organization of contextual information may also explain why place fields have not been seen as the predominant form of coding in the primate hippocampus. Monkey hippocampal neurons respond primarily when the subject directs its gaze at a particular part of the environment (Rolls, 1999; Chapter 13 this volume). Whereas rodents explore the environment by active locomotion, primates accomplish much of their exploration visually, by directing their gaze about the environment. Place cells had also not been found in human recording studies (e.g., Fried et al., 1997; Heit et al., 1998;), perhaps because subjects were not freely navigating. In more recent studies using virtual navigation methods, location-selective firing has not only been reported but seems selective to contexts in which subjects must search for and identify meaningful locations (Ekstrom et al., 2003). Thus, the apparently discrepant findings between primate, human, and rodent studies may not indicate fundamentally different computations by the hippocampus, but rather may result from different combinations of information passed on to the hippocampus, information that is dictated by the task conditions.
Consideration of the organization of other neural systems provides support for the expanded view that a fundamental and generally applicable organizing principle when analyzing information from our environments is that this is done according to a spatial reference framework. For example, a striking feature of the neural organization of most sensory systems is that there is a clear spatial organization (or topography) to neural representations from the sensory receptor to sensory cortex. Presumably, this form of organization facilitates adaptive responding since motor output systems (from motor cortical areas through spinal cord) are also organized topographically. Given that the use of a spatial reference frame to organize sensory and response information is highly efficient, it seems reasonable to speculate that there was strong evolutionary pressure to process and retain sensory information spatially as sensory association (cortical) regions evolved. Maintaining the same reference framework for fundamentally important tasks such as accurate navigation may have facilitated an organism’s ability to rapidly adapt to environmental changes. For this reason, phylogenetically old structures such as the hippocampus may be initially predisposed to process information within a spatial framework during navigation-based tasks. Indeed, hippocampal place fields are observed with similar abundance in spatial and nonspatial tasks (Yeshenko et al., 2004). The fact that it is possible to break out of the spatial framework if needed (see Chapter 11) may reflect a more recently evolved adaptation. The CDH model, then, considers the spatial nature of the hippocampal representations to reflect a basic form of organization of the M inputs.
Temporally Organized Hippocampal Processing In addition to the automatic influence of M features and spatial analyses, hippocampal context representations appear to be continually subject to some form of temporal organization by local and distal neural networks. That is, the timing of place cell discharge can be precisely regulated in different ways. This temporal organization may take different forms, perhaps for different purposes. Many years ago, it was shown that movement through place fields is associated with dynamic changes in spike timing relative to the ongoing theta oscillations in the EEG (O’Keefe and Recce, 1993). On a single pass through a field, the first spike of successive bursts occurs at progressively earlier phases of the theta cycle. The discovery of this socalled phase precession effect is considered significant, for it was the first clear evidence that place cells are part of a temporal code that could contribute to the mnemonic processes of hippocampus. Such tempo-
HIPPOCAMPAL PLACE CELLS DURING LEARNING
rally dynamic changes in spike timing may be a key mechanism by which place fields provide a link between temporally extended behaviors of an animal and the comparatively rapid synaptic plasticity mechanisms (e.g., LTP) that are thought to subserve learning (e.g., Skaggs et al., 1996). Theoretical considerations have generated detailed models of how phase precession could explain the link between behavior and neural plasticity mechanisms (e.g., Buzsaki, 2005; Zugaro et al., 2005). Another form of temporal-based neuroplasticity involves a change in the timing of spike discharge by one cell relative to spiking of other cells, or relative to prior spike discharge by the same cell. For example, it has been shown that the temporal coherence of place cell discharge is greater in mice with an intact hippocampus than that in mice with deficient NMDA systems (McHugh et al., 1996). Greater synchronization could offer a stronger output signal to efferent structures. Similarly, experience-dependent temporal codes may be found in terms of the temporal relationships between the firing of cells with adjacent place fields. With continued exposure to a new environment, place fields begin to expand asymmetrically; the peak firing rate occurs with shorter latency upon entrance into the field (Mehta et al., 1997, 2000). It was postulated that with continued exposure to a spatial task, place cells are repeatedly activated in a particular sequence. As a result, synaptic connections between cells with adjacent fields become stronger, so that entry into one place field begins to activate the cell with the adjacent place field at shorter and shorter latency. The backward expansion of place fields may result from asymmetric Hebbian mechanisms such as LTP, since NMDA antagonism blocks LTP and the expansion effect (Ekstrom et al., 2001). The asymmetric backward expansion of place fields is thought to provide a neural mechanism for learning directional sequences. Moreover, it has been suggested that the backward-expansion phenomenon may contribute to the transformation of a rate code to a temporal code such as that illustrated in phase precession (Mehta et al., 2002). Perhaps the backward-expansion phenomenon could help to explain other place field phenomena, such as the tendency for place cells to fire in anticipation of entering a specified location within a familiar environment (Muller and Kubie, 1989). While the dynamic changes in place field shape are intriguing, it remains a challenge to determine whether, and then how, changes in the temporal distribution of cell firing are directly related to spatial learning. Recent studies certainly suggest that they are related. Shapiro and Ferbinteanu (2006) reported that the temporal relationship between firing of simultaneously recorded place cells discriminated task phase,
25
suggesting that past experience determines the relative spike timing of individual neurons. Lee and colleagues (2006) described place fields whose center of mass moved in the forward direction (toward a goal) with each successive trial. The direction of place field movement was in the direction opposite that predicted by spike timing–dependent plasticity mechanisms. However, it may be that place fields will move in the forward or backward directions for different purposes, and their relative direction of movement may depend on the task conditions. For example, forward movement of place fields may be more prevalent when choices must be made between multiple goal locations. Backward movement may be more prominent when merely knowing the direction of movement is sufficient to predict future reinforcement. In these cases, a distinguishing feature would be not only the direction of place field movement but also the temporal resolution of the movement itself. Indeed, forward movement of place fields seem to occur over a protracted time scale relative to the backward-expansion effect. If the above analysis is correct, then a critical issue is to determine the mechanism that regulates place field shifts in the forward and backward directions, and whether these are independent mechanisms. It is also possible that the more protracted forward movement of place fields reflects changes in the spatial code, instead of or in addition to a temporal shift. Experience-dependent changes in spike timing are evident when considering not only cell discharge of one cell relative to another or to itself, but also timing relative to task-specific events. Already mentioned is the finding that place fields tend to move toward goal locations (Lee et al., 2006). Ferbinteanu and Shapiro (2003) described place fields that were consistently located on the goal arm of a plus maze regardless of the start location. These cells were considered to represent prospective information about the upcoming reward event. An earlier example of prospective coding by place cells was the demonstration that place fields seemed to anticipate the future location of an animal by some 90 ms (Muller and Kubie, 1989). Again, the time scales for the latter two studies are very different, so it remains to be determined if they are derived from the same or different mechanism. One may also consider the backward-expansion effect (described above) to reflect a neural code that anticipates task-specific events. It is possible, then, that at least a subpopulation of place cells conveys information about what to expect in the future within the milliseconds to seconds range. Place cells may also code information retrospectively. Place fields have been found at consistent distances from a start location on linear tracks and plus mazes (e.g., Gothard et al., 1996; Ferbinteanu and Shapiro, 2003). In summary,
26 PLACE CELLS AND SPATIAL CONTEXT these findings are consistent with the view that the hippocampus encodes perhaps simultaneously information about what to expect in a given context and information about current or recent events. Another striking example of experience-dependent temporal organization of hippocampal neural codes is what has been described to occur during the slowwave sleep period after exploration. It was shown that the temporal sequence of neural firing recorded during exploration is essentially ‘‘replayed’’ during slowwave sleep, but in a temporally condensed fashion (Wilson and McNaughton, 1994). This raises the interesting possibility that, with experience, neural discharge comes to repeat a temporally ordered sequence during sleep that is necessary for effective memory consolidation (Hoffman and McNaughton, 2002; Pennartz et al., 2002). Recently, another form of replay of neural activation was reported when rats paused between traversals along an elevated track (Foster and Wilson, 2006). This differed from the replay observed during sleep because the temporal order of neural activation was reversed relative to the order in which locations were visited during locomotion. This reverse replay was interpreted as being useful for a different mnemonic function, one that allows the evaluation of recent behaviors (trajectories in this case) in terms of their reinforcement outcomes. This particular form of temporal organization may reflect a retrospective mechanism that is needed to detect changes in the expected situation. Continued investigation of the role of spike timing in neuronal communication will also provide new insight into the impact that the hippocampus has on memory processing in efferent structures. Prefrontal cortical neurons can be phase locked to CA1 hippocampal theta rhythms such that phase precession is observed (Hyman et al., 2005; Jones and Wilson, 2005). The entrained prefrontal cortex neurons tended to be those whose firing is correlated with specific behavioral acts (Hyman et al., 2005). Siapas et al. (2005) verified the directional importance of the hippocampal–prefrontal cortical connection by showing that the phase locking of hippocampal theta and prefrontal cortical unit activity is best when theta is delayed by about 50 ms. To summarize, temporally coordinated activity can take place separately within each cell, locally between cells, or between single cells and the larger population signal as reflected in the EEG. It may also take place between cell ensembles. Each of these forms of coordination can be used to predict behaviors and events in a prospective manner or to recall recent behaviors or events in a retrospective manner. Thus, T can be represented in the multidimensional context matrix according to the degree to which the neural code is
prospective or retrospective. Cell-to-cell spike timing changes likely emerge from a number of natural synaptic sources that regulate coordinated spike activity within the hippocampus. More broadly speaking, rhythmic activity may be regulated by plastic changes within intrahippocampal circuitry and by extrinsic systems responsible for generating rhythmic activity in the hippocampus (e.g., brain stem; Buzsaki, 2002, 2005; Vertes et al., 2004; Hasselmo, 2005b). The varied sources of temporal regulation may allow for coding M information as stimulus sequences or for determining stimulus duration.
Integrated M, S, and T Information A multidimensional context matrix model can be used to illustrate that information coded by place cells reflects M information that has been organized spatially and temporally. The particular piece of information coded by each place cell could be described as a single value in the matrix. Level 1 of the hierarchical organization of context processing, then, is meant to reflect the convergence of M, S, and T information onto single hippocampal cells. M input to place fields represents a predetermined configuration of input variables, and the contribution of each input variable is differentially weighted according to past experience (Fig. 2.2). The value of M may vary across cells, indicating that each cell’s M factor varies according to memory strength. M, then, could be thought of as weak to strong memory-guided perception. Regardless of the specific configuration of the input variables, they combine to form only one of three organizational factors that are continually imposed on each place cell (level 1). Space (S) and time (T) represent two additional organizational gradients. The relative strengths of M, S, and T input can differ as a function of experience for each cell. That each cell receives input from a different combination of M, S, and T inputs may result in a range of response profiles. Information coded by individual place fields can be defined according to its degree of memory involvement (from completely driven by memory system to not impacted by memory at all; range 1.0 to 0.0), the strength of the impact of a spatial reference framework (from strong to weak; range 1.0 to 0.0), and temporal organization (from retrospective to prospective codes; range 1.0 to 1.0). The final strength of an individual cell’s signal may additionally be affected by factors such as the inherent biological ‘‘noise’’ in the neural system (Koulakov et al., 2002) at the time of recording, by the synaptic weight distribution between neurons based on past experience, and by modulatory input from subcortical areas.
HIPPOCAMPAL PLACE CELLS DURING LEARNING
Modulation by Subcortical Systems An animal’s interpretation of its current internal and external sensory environments depends not only on how it interacts with or behaves in the environment but also on the current motivational state. When one is hungry, preferential attention will be directed toward cues and behaviors that might otherwise be ignored if one is in the same environment but is searching for a mate. Traditionally, the effects of motivational states on behavior have been studied by either considering the consequences of varying hunger or thirst or investigating the effects of appetitive or aversive events (e.g., Kennedy and Shapiro, 2004). The hypothalamus has long been considered central to the regulation of homeostatic systems such as hunger and thirst. There is substantial evidence that several subcortical structures (in addition to the hypothalamus) exert powerful control over the excitability of limbic system neurons. The traditional interpretation of these subcortical influences has been that they somehow gate, or filter, cortical information arriving in the hippocampus (e.g., Winson, 1984). The gating hypothesis is supported by findings that electrical stimulation of numerous subcortical structures facilitates synaptic transmission through the hippocampus (AlvarezLeefmans and Gardiner-Medwin, 1975; Assaf and Miller, 1978; Segal, 1979; Bilkey and Goddard, 1985; Dahl and Winson, 1985; Mizumori et al., 1989a). Of these, the medial septum appears to be strategically located to provide the hippocampus with information concerning the animal’s motivational state, since the septum receives direct input from hypothalamic nuclei (Swanson and Cowan, 1979; Jakab and Leranth, 1995) and it projects directly to the hippocampus. The hippocampal effects (discussed below) are mediated by powerful GABAergic and cholinergic septal afferents onto both pyramidal and nonpyramidal neurons within multiple subregions of hippocampus (Freund and Antal, 1988; Risold and Swanson, 1995; Freund and Buzsaki, 1996). Disruption of septal function, by either permanent lesions or reversible inactivation, impairs hippocampaldependent learning (Winson, 1978; Mizumori et al., 1989b; Harzi and Jarrard, 1992) and the patterned activity of hippocampal neurons. In an intact, behaving animal, recordings of the hippocampal EEG show a rhythmic oscillation around the theta frequency (about 7–9 Hz). Compromise of the integrity of the medial septum significantly attenuates the hippocampal theta rhythm (Winson, 1978; Mizumori et al., 1989b). Septal lesions or reversible inactivation prevents hippocampal place fields from responding appropriately to changing environments (Mizumori et al., 1989b; Leutgeb and Mizumori, 1999; Ikonen et al.,
27
2002). Other evidence shows that cholinergic input (presumably from the septum) significantly modulates hippocampal LTP (a synaptic model of plasticity). Also, it appears that there is an increase in acetylcholine release during new learning (Ragozzino et al., 1998; Gold, 2003). Taken together, these data indicate that the septum is in a key position to regulate hippocampal plasticity and presumably the efficiency of intrahippocampal network functions. One interpretation of the septal influence is that it identifies for the hippocampus the appropriate internal state (or motivational state; Mizumori et al., 2000a). In theory, such information can bias the strength of other types of M input. Relevant spatial-context information could arrive in septum via the hippocampal CA3 efferent projection system. Indeed, it has been shown that the lateral septum (the main cortical input area of the septal region) contains neurons whose firing is correlated with the location of animals within their environment (Leutgeb and Mizumori, 1999; Zhou et al., 1999). That is, lateral septal neurons show place fields and, similar to hippocampal place fields, they respond to changes in the spatial context. Our current working hypothesis (Mizumori et al., 2000a) is that the lateral septum, via its extensive projection to various hypothalamic and mammillary nuclei, informs motivational systems of the brain about the current spatial context. In doing so, it may bias the firing properties of hypothalamic neurons to reflect the appropriate motivational state. Such a bias could in turn dramatically alter hypothalamic influences over efferent structures, such as the medial septum. Changes in medial septal activity could then refine hippocampal neural plasticity. In short, the hypothalamoseptal informational system could be thought of as reflecting the motivational perspective within which the hippocampus should interpret M, S, and T information. Consistent with this view, it has been shown that the motivational state of an animal can greatly impact hippocampal-dependent learning (Kennedy and Shapiro, 2004). Also, there are preliminary findings that hippocampal neural codes appear to change according to the current motivational state (Shapiro and Ferbinteanu, 2006).
Level 2 Organization: Local Circuit Codes Assuming that the M, S, and T variables in fact build context representations at the single-cell level, a challenging question is to understand how the hippocampus uses this context information. The common finding of partial reorganization of place fields following context manipulations suggests that a subpopulation of place cells is quite sensitive to changes in context, while others are not. Context sensitivity is demonstrated by changes in firing rates, altered temporal patterns
28 PLACE CELLS AND SPATIAL CONTEXT of unit activity, or place fields that rapidly reorganize (i.e., change field location and/or firing rate) when the environment changes. The responsive place fields may have the function of monitoring current contextual features, while the persistent fields may reflect expected contextual features. If the current context is determined to be different from the expected context (i.e., the two contexts are discriminated), then an appropriately changed message may be sent to update cortical memory circuits. It is reasonable to consider, then, that a number of individual place cells contribute to functionally defined local networks (level 2) that generate distinct ensemble codes for expected and current context information. Local circuits for expected information may vary in terms of the relative strengths of its inputs. For example, one local circuit may receive information from place cells with strong reward codes relative to other types of information, while another local circuit may receive strong spatial codes compared to other types of information. The former local circuit, then, contributes primarily reinforcement expectation information to level 3 analyses, while the latter local circuit provides mainly information about the expected spatial features of the context. Local circuits for current context information may also be biased to selectively process one type of information relative to another. Such biases are likely based on past uses of the neural circuitry. There may also be interactions between local circuits that further bias the relative distribution of local circuit output strengths. Communication between individual cells (level 1) and between the local circuits of level 2 could have the effect of conferring a level of flexible processing to the system, a function often attributed to hippocampaldependent learning and memory systems. Most recently, Dragoi and Buzsaki (2006) provided evidence that local cell ensembles in CA3 predicted a rat’s location about a half a theta cycle before CA1 cell assemblies. This finding led the authors to speculate that environmental stimuli trigger the replay of local CA3– CA1 ensembles in a time-compressed fashion. This interaction may reflect how local circuit codes make use of spike timing–dependent plasticity from level 1 organization.
Level 3 Organization: Population Efferent Code At the next level (level 3) of the hierarchy, an integrated efferent code should be generated that reflects the results of a match–mismatch computation based on inputs from level 2. To permit a certain degree of flexibility and specificity, there may be different de-
grees of level 3 context comparisons. For example, initially, level 3 integration may occur within specific subregions of hippocampus (e.g., sectors of dorsal CA1, sectors of dorsal CA3, etc). The output of this first round of context comparisons may inform additional context comparators that span larger regions (e.g., dorsal CA1, dorsal CA3, etc.), which in turn feed information to computational circuits that evaluate even larger sectors of hippocampus. For the sake of efficiency, it is assumed that the structure of the context matrix for single cells also applies to local circuits and to population codes as well. That is, information contained within any level of population code is also multidimensional in that it includes temporal gradients (from past to present to future expectations), spatial gradients (from weak to strong spatial information), and memory impact gradients (from low to high). Associative memories are generally thought to be represented as a stable network of neural activity, or attractors (Hopfield, 1982; McNaughton and Morris, 1987). An updated cortical long-term memory, then, could be represented as a moderate to severe shift in the attractor basin, or stable state, of a memory network, depending on the degree of mismatch detected. The resultant activated memory network will then inform the most recent hippocampal expectation for a given context. The latter process should result in a subsequent reorganization of neural activity patterns in the hippocampus. If a context is defined by a unique and integrated array of inputs, in theory a change in any one or combination of features (i.e., input variables) should produce an ‘‘error’’ signal that reflects a mismatch (Mizumori et al., 1999, 2000a). Indeed, changes in a variety of single features have been shown to have dramatic effects on place fields (e.g., Quirk et al., 1990; Bostock et al., 1991; Knierim et al., 1995; Markus et al., 1995; Skaggs and McNaughton, 1998; Tanila, 1998; Lever et al., 2002; Wirth et al., 2003; Leutgeb et al., 2005a). The detection of a mismatch between expected and current contextual conditions could serve as a signal to efferent structures that regulate ongoing behaviors. In particular, a mismatch may increase exploration, which in turn engages greater sensitivity to novelty. The latter should have the effect of facilitating the encoding process. Inhibition of the GABAergic system is likely to be an important component of this novelty detection mechanism (Morris, 2006), as is release of dopamine from the ventral tegmental area (Lisman and Grace, 2005) and acetylcholine release from medial septum (Hasselmo and McGaughy, 2004). Each cell, then, has great potential to be affected by endogenous regulators of synaptic activity. In cases when the context has not changed (i.e., there is no place field reorganization), a consistent
HIPPOCAMPAL PLACE CELLS DURING LEARNING
hippocampal output could signal that the most recently engaged memory system and behavioral expression patterns will be maintained.
Mechanisms of Context Comparisons: The Relative Contributions of CA3 and CA1 Place Fields The hippocampus must be able to engage in a number of operations if effective context comparisons are to occur. Initially, relevant stimuli need to be selected for comparison. The dentate gyrus is thought to engage in pattern separation functions that might serve this purpose by distinguishing (or orthogonalizing) potentially important and relevant input (Marr, 1969; O’Reilly and McClelland, 1994; Rolls, 1996). Information regarding both expected and current contexts needs to be represented, and there is ample evidence to support this requirement. As described above, place fields may reorganize following context manipulations, indicating that some place cells are sensitive to the status of the current definition of context. Other place fields may not reorganize following the same context manipulation, and these place fields may be driven more by an expectation of the current context based on past experience. Context information needs to be held ‘‘online,’’ in order for a comparison to be made. Many investigators (Marr, 1971; O’Keefe and Nadel, 1978; McNaughton and Morris, 1987; Rolls, 1996; Eichenbaum, 2000; Guzowski et al., 2004; Treves, 2004) have provided theoretical evidence that the CA3 recurrent network likely provides a shortterm buffer that is necessary for comparisons to be made, perhaps via pattern completion, a function whereby missing information is filled in when there is incomplete but familiar input (e.g., Marr, 1971; Rolls and Treves, 1998; Gold and Kesner, 2005). Presumably, pattern completion mechanisms come into play when input patterns have been judged to be similar, which in turn lead to constant behavioral output. The tendency to separate or complete incoming stimulus patterns within the dentate gyrus–CA3 circuit may represent opposite tendencies of a gradient of network plasticity (Martin et al., 2000). Whether the network identifies input patterns (i.e., context information) as being similar or different may depend on a Bienenstock-Cooper-Munro (BCM) sliding modification threshold (Bienenstock et al., 1982) whereby the threshold for the identification of a mismatch varies with modulating circumstances or conditions and/or with past experiences (Barry et al., 2006). Other suggestions that hippocampus can store information temporarily (presumably to allow context comparisons) are findings of retrospective and pro-
29
spective neural codes (e.g., Ferbinteanu and Shapiro, 2003). These may provide a platform on which to compare recent, current, and anticipated contextual information. Dragoi and Buzsaki (2006) further suggest that temporal coordination of local CA3–CA1 networks may provide a mechanism to retain information about past, current, and future information ‘‘online.’’ Of note, however, Bower et al. (2005) have suggested that place field reorganization may reduce the demand for such temporary memory storage. Computational models provide suggestions for ways in which error analyses (i.e., match–mismatch computations) might be implemented within hippocampal memory-related circuitry and then modulated by neurotransmitters such as dopamine and acetylcholine (e.g., Lisman, 1999; Hasselmo and McGaughty, 2004; Treves, 2004; Hasselmo, 2005a,b; Lisman and Grace, 2005). While more specific mechanisms underlying context comparisons remain to be worked out, the outcome of the comparison analysis can be used to identify distinct events or episodes. The results can be forwarded to an anatomically extended memory system to impact long-term memory representations in, for example, cortex, to engage or change behaviors, re-evaluate reinforcement outcomes relative to behavior (e.g., via striatal processing), or update the motivational context (via subcortical projections that affect hypothalamic regions). It is also possible that CA3 and CA1 place cells make different contributions to the context comparison process. When rats perform at asymptotic levels on a hippocampal-dependent, spatial working memory task, CA3 place fields are smaller (i.e., more spatially selective) than CA1 place fields (Mizumori et al., 1989b, 1995, 1996; Barnes et al., 1990) and more easily disrupted following cue manipulations (Mizumori et al., 1999). Also, CA3 place fields are more sensitive than CA1 place fields to disruption following reversible inactivation of the medial septum (Mizumori et al., 1989b). The greater sensitivity of CA3 fields to context manipulation seems to occur regardless of the type of task (Lee et al., 2004; Leutgeb et al., 2004), a finding consistent with the view that the CA3 region is continually and specifically engaged in the analysis of contexts. It is important to note that despite the greater overall sensitivity of CA3 place fields to changes in contextual information, about 40% of CA3 place fields continued to persist when faced with contextual changes. CA1 also has the potential for representing current and expected contextual information, but compared to CA3, a greater proportion of cells (about 60%) show persistent place fields despite context change (e.g., Mizumori et al., 1999; Lee et al., 2004; Leutgeb et al., 2004). One interpretation of the differences observed
30 PLACE CELLS AND SPATIAL CONTEXT between CA1 and CA3 place field stability in the face of context changes is that CA1 neurons are preferentially driven by memory representations of the context, perhaps via entorhinal afferents that bypass CA3 (Witter et al., 2000). For this reason, it may be predicted that the effects of context manipulations on choice accuracy will be more directly correlated with the responses of CA3 place cells than with those of CA1 place cells. Although a direct test of this prediction has yet to be carried out, it is becoming more common to find reports of a lack of correlation between CA1 place field reorganization and behavioral responses (e.g., Cooper and Mizumori, 2001; Jeffery et al., 2003). If the dentate gyrus–CA3 region has a greater role during the comparison of contextual information, then CA1 may have a special role in the temporal sequencing of information (e.g., Olton et al., 1979; Rawlins, 1985; Kesner, 1991; Hampson et al., 1993; Wiener et al., 1995; Gilbert et al., 2001; Kesner et al., 2004; Treves, 2004). More specifically, CA1 place cells may temporally organize, or group, CA3 output to allow for meaningful sequences to be passed on to efferent targets (Mizumori et al., 1999), such as the prefrontal cortex (Jay et al., 1989) and subiculum. Direct and selective entorhinal input to CA1 (Witter et al., 2000) may provide the strong memory-based M, S, and T information needed to accomplish this task. Although the precise nature of the temporal organization remains to be determined, one possibility is that CA1 is more tightly coupled to the rhythmic oscillations of hippocampal EEG, a phenomenon thought to contribute to the temporal organization of hippocampal output (Buzsaki, 2005). Understanding the mechanism of sequence coding is importantly connected to understanding context discrimination by the hippocampus, since sequence coding is thought to result in a fundamental unit of information that becomes subject to analysis via the context comparison algorithm.
THE HIERARCHICAL MODEL OF CONTEXT DISCRIMINATION RELATIVE TO DIFFERENT BEHAVIORAL AND LEARNING SITUATIONS Representation within the hierarchical model can vary depending on whether an animal finds itself in a familiar situation with familiar information, a familiar situation that has new information embedded within it, or a novel situation with new cues.
Experience within a Familiar Context When animals enter a familiar situation, activated long-term memory circuitry will make available to the
hippocampus knowledge (manifested here as a particular distribution of M information weights) about what to expect in terms of the significant sensory cues and appropriate behavioral responses. In this sense, details of hippocampal representations are guided by what the animal has learned in terms of the specific task rules and established associations. Therefore, it would be expected that at least some of the place cells would represent prospective information on what to expect in a given task (Fig. 2–1; cells 2 and 3). Retrospective coding may also be strong (as in cells 1 and 4) since animals will have learned which information to track during task performance. From prior experience, at least some of the connections from individual place cells should be stronger than others, reflecting a context-specific bias in the distribution of weighted input to a (level 2) local circuit from individual, activated neurons. As an example, note that cell 3 is strongly connected to the local network (indicated by thicker lines), resulting in a comparatively stronger representation of the same information at level 2. Consideration of both the pattern of synaptic strengths and the pattern of neural activation will increase the combinatorial power of intrahippocampal circuitry. If all cues are as expected, the patterns of synaptic strengths and neural activation within local circuitry that represents the current context should match the neural activity pattern seen in local circuits that represent the expected context. In this situation, hippocampal output will signal that the situation was as expected, which in turn may provide added strength to the currently active long-term memory circuitry.
Experience with a Familiar Context That Has Changed In the event that a component of a familiar situation changes, different information will be forwarded to local circuits from place cells that represent expected information, compared to information forwarded by cellular representations of current context information (Fig. 2–3). As a result, the local circuit definition of the expected spatial-context conditions and those actually experienced will differ, leading to a mismatch output. In the example shown in Figure 2–3, cells 5 and 8 may code familiar information that is different from what was expected. Cells 6 and 7 are low on the vertical M scale, indicating they represent less familiar information. Note that the synaptic strengths vary across cells. The stronger connection of cell 6 results in a local circuit representation that is accentuated in accordance with the strength of cell 6 (i.e., taller cylinder). In contrast, the synaptic strength of cell 4’s connection is weak, resulting in an attenuated signal at level 2.
HIPPOCAMPAL PLACE CELLS DURING LEARNING
31
Figure 2–3. If a mismatch is identified as a result of the context comparison, the context matrix at level 3 will appear as a blend of ensemble activity from the expected and current context matrices. Consequently, the output message from this computation will signal the need for a change in ongoing behavior, and it may induce a stable memory network to become unstable, allowing a new but related memory to be activated. Indeed, changed familiar environments can result in novelty detection, which in turn tends to increase exploratory behavior.
Experience with a Novel Context When accounting for neural responses during exposure to novel environments (Fig. 2–4), it should first be noted that if an animal finds itself in a new situation, past memories may still impact the nature of hippocampal representations, although to a lesser degree, because it is difficult to present a completely novel situation. More often than not, a rat will have experienced some aspect of a test situation, or some-
thing similar to it. This may bias a rat’s perception of elements of the sensory environment that are similar to those that have been experienced previously (e.g., for rats, this may be narrow paths or dark holes). When we find ourselves in ‘‘new’’ environments, we also tend to focus as best we can on information that seems familiar so that we can make our best guess as to an appropriate response. Perhaps familiar components of a test environment activate the most closely related, pre-existing schema in memory in an attempt to
32 PLACE CELLS AND SPATIAL CONTEXT
Figure 2–4. Context matrix during exposure to a novel task situation. Since the situation is novel, it is expected that one’s expectations would not match the currently experienced context. Under this test condition, one would expect that the M influence should be generally weak. There should be a bias to code context information spatially since, it is argued, that is the default mode of operation in behaving animals. Since M is weak, the ability of the cells to code prospectively should be relatively weak. Retrospective coding may be stronger if certain features of the novel context remind the subject of prior encounters with similar features. The informational locus within the context matrix is expected to be typically toward the center, or perhaps biased toward a stronger S.
organize the novel information (Morris, 2006). The more similar the ‘‘novel’’ situation is to a previously experienced one, the more quickly ‘‘new’’ learning will progress and neural representations become stabilized. An animal’s interpretation of components of a test situation as being familiar may constrain the initial selection of responses to those that have proven
successful in the past when the animal experienced similar sensory cues. This pattern exemplifies the constant and automatic impact of memory representations even in ‘‘new learning’’ situations. The degree of influence by truly novel aspects of a situation should increase as the situational novelty declines. Correspondingly, although behavioral responses will be slower
HIPPOCAMPAL PLACE CELLS DURING LEARNING
initially when memory impact is weak, behaviors should speed up as memory influences strengthen— that is, behaviors will appear automatic. During initial exposure to novel situations, the population code (level 3) should be relatively weak in terms of its impact on long-term memory representations, since comparatively little is expected (i.e., M is between 0.0 and 0.5), and the S and T conditions are not yet task specific. We would expect that T inputs would not be prospective in nature because the animal has not yet learned the significance of the current context. Rather, there may be occasionally strong retrospective coding as the animal attempts to keep track of its recent history so that it can develop plans for future behaviors. T inputs, then, are expected to be between 1.0 and 0.0. During initial learning, it is expected that inputs along the S gradient will be between 0.5 and 1.0, since the hippocampus is thought to be predisposed to organize inputs within a spatial framework in freely behaving rats. Thus, single neuron and local circuit signals will tend to occupy the center portion of the three-dimensional context matrices (Fig. 2–5). At early stages of learning, mismatches between expected and experienced contexts are likely to be many, resulting in the continual and rapid shaping of long-term memory representations (McClelland et al., 1995; Morris, 2006). As cortical memory representations become more specific, so should the feedback to hippocampal cells that encode the expected contextual features. Consequently, the signals of individual place cells should become stronger and more distinct. When animals are first placed into an experimental chamber or situation, they do not yet know the task demands. Typically, exploration is a predominant initial behavior, followed by explicit training that effectively ‘‘shapes’’ the animal’s behavior in ways that are more conducive to learning the specific task at hand. Consistent with the view that exploration engages the entorhinal cortical grid system are the numerous observations of place fields immediately after being placed in a novel environment. After learning, place cell representations should occupy more extreme locations in the context matrix (Fig. 2–5): M inputs should approach 1.0; T inputs should also approach 1.0 or 1.0 as prospective coding and stronger retrospective coding develop; S inputs may become more distinct with values approaching 1.0 or 1.0. The latter is predicted because for animals that learn a nonspatial task, the firing correlates of at least a proportion of the cells should be observed to shift to represent nonspatial, task-relevant information. This phenomenon remains to be tested systematically. Place fields (level 1) should become more specific and reliable with continued training as one gradually
33
Figure 2–5. Level 3 context matrices as a function of learning. When animals explore a familiar environment, the influence of memory systems (M) should be high, and there should be greater evidence of strong retrospective and prospective coding (T). In cases of random exploration or during spatially directed behaviors, the spatial (S) influence should be high. When animals enter new environments or situations, the M influence should be less. There could be slightly strong S coding since, it is argued, animals have a natural tendency to organize information spatially in new situations. T values should be mid-range because in novel situations it is not clear what type of retrospective or prospective information one might need.
learns about associations relevant to the test environment. While many studies have indeed shown that place fields become more specific and/or reliable with short-term exposure to novel environments (e.g., Muller and Kubie, 1987; Wilson and McNaughton, 1993, Markus et al., 1995; O’Keefe and Burgess, 1996b; Hetherington and Shapiro, 1997; Frank et al., 2004), there are fewer reports of place field changes during explicit training on hippocampal-dependent tasks (but see Mizumori and Kalyani, 1997). Recent evidence, however, suggests that as animals learn to discriminate contexts, characteristics of place fields shift accordingly (Smith and Mizumori, 2006). Learning can be considered complete when mismatches no longer occur and consistent memory representations (and associated neural activity patterns) are maintained during behavior. That is, there should be a preponderance of pattern completion (rather than pattern separation) outcomes from the context analysis. It is worth noting that to ensure adaptive behav-
34 PLACE CELLS AND SPATIAL CONTEXT ioral functions, the hippocampus must continue to engage in context comparisons in the event that the context shifts. Similarly, the hippocampus should process contextual information even for tasks that do not explicitly require contextual knowledge in the event that contextual information becomes relevant. Indeed, we have shown that specific neural codes in the hippocampus remain responsive to changes in context even though contextual learning was not necessary to solve the (response) task (Yeshenko et al., 2004; Eschenko and Mizumori, 2007). These data show that hippocampal processing of contextual information is automatic and continuous (Morris and Frey, 1997).
SUMMARY AND IMPLICATIONS OF THE CDH HIERARCHICAL MODEL OF HIPPOCAMPAL PROCESSING Memory (M) influences, and their organization by spatial (S) and temporal (T) factors, appear part of a seemingly continuous and automatic form of activity within the hippocampus. Evidence shows that past experience (or memories) alters the precise constellation of place cells that are active in a given environment. The spatial arrangement of the place fields have been shown to change according to learned task demands in hippocampal-dependent and hippocampal-independent tasks. Many of these changes are observed in terms of both location and temporal characteristics of individual place fields. Elimination of a single input source has often led to unpredictable changes in the place fields (i.e., it engages a pattern separation process), indicating that place fields normally represent a highly integrative bit of information whose integrity depends on a contribution from all inputs. If hippocampal place cells represent a wide range of context-specific information, then why are place fields the most frequently reported correlate of hippocampal neurons? The recent finding of grid cells in entorhinal cortex (Hafting et al., 2005) supports the view that the multiple forms of context information may be organized according to location-selective nodes of a spatial reference frame. It is postulated that from voluntary locomotion grid fields emerge, resulting in a preponderance of place fields in the hippocampus during tasks involving the navigation of spatially extended environments. The term spatial context should be used to describe context processing when animals navigate their environment. Numerous temporal organizational schemes exist in the hippocampus and these are presumably generated as a result of natural synaptic plasticity mechanisms. These temporal algorithms are key to the hip-
pocampus’ ability to form conditional associations across extended periods of time. These temporal codes come to reflect future (expected) information as memories become strengthened. If hippocampal processing is continuous, regardless of the task at hand, what determines its apparently specific contributions to learning? Much of the support for a separate memory function of hippocampus is based on studies of the mnemonic consequences of brain damage (in animals and humans). Lesion effects may be observed only when the intrinsic processing by the structure of interest is unique and essential for learning to take place (Mizumori et al., 2004b). If other neural circuits compensate for lost function after a hippocampal lesion (causing, for example, animals to adopt a different learning algorithm), no behavioral impairment will be observed. Thus, stimulus–response learning is not impaired following hippocampal lesions as striatal computations are sufficient to support such learning. This does not mean that the hippocampus does not normally play a role in stimulus–response performance. The hippocampus may contribute by defining the context for the learning, which in turn may serve to make the learned information more adaptive to new situations in the future. This view is in line with the argument that ‘‘hippocampal-independent’’ memories may benefit from hippocampal processing even though such memories may be formed in the absence of hippocampus (e.g., Westmacott et al., 2004). A growing literature from neuroimaging studies shows that hippocampal regions are preferentially activated during the use of specific cognitive strategies, such as sequence learning (Schendan et al., 2003) or spatial navigation (e.g., Maguire et al., 2006). As compelling as these data are, these studies do not address the possibility that accurate performance requires the operation of a network of brain regions that differ in their metabolic activity, but nonetheless are important for normal performance. It is possible, then, that the apparently selective importance of hippocampal processing in memory is not determined solely by intrahippocampal computations but also by virtue of the patterns of connections and the impact hippocampal processing has on other neural circuits (see additional discussion in Chapter 17 of this volume). The prefrontal cortex is likely one of those extrahippocampal circuits that may critically determine whether hippocampal processing has a direct role in guiding behavior at any given point in time, since it is in a position to integrate striatal, amygdalar, and limbic information (Jay and Witter, 1991) and because of its comparatively close proximity to motor output systems. Permanent prefrontal cortical lesions in rats result in reliable deficits on tasks that require the flexible use of location
HIPPOCAMPAL PLACE CELLS DURING LEARNING
information (Poucet and Herrmann, 1990; Granon et al., 1996; Gemmell and O’Mara, 1999). Also, reversible inactivation of the prefrontal cortex impairs spatial working memory (e.g., Ragozzino et al., 1998). Attempts to identify the neural codes of prefrontal cortex have led to the surprising finding of a paucity of spatial representation (Poucet, 1997; Jung et al., 1998; Pratt and Mizumori, 2001). Instead, the most consistent behavioral correlate identified for prefrontal cortex neurons was reward- and movement-related discharge. Similar to amygdala neurons, prefrontal neurons changed firing rates in anticipation of encounters with rewards of different magnitudes (Pratt and Mizumori, 2001). The combined results of the lesion and neural recording studies are consistent with the view that the prefrontal cortex may provide a prospective representation of the incentive value associated with different locations in a context-specific manner (Seamans et al., 1995). It is important to distinguish the significance of prefrontal-cortex coding and hippocampal place cell codes. Although both are sensitive to similar types of information (e.g., spatial context, reward, and movement), the representations are considered to have different functions. In the hippocampus, these different inputs are necessary to define the expected or current context conditions. Prefrontal cortex, by contrast, actively integrates hippocampal information to generate prospective codes (Baeg et al., 2003) that impact neural response patterns in efferent structures traditionally thought to control ongoing voluntary movements (e.g., striatum and motor cortex). In sum, the present view considers a role for the hippocampus in memory in terms of the processes it contributes to a broader neural system. This perspective assumes that the hippocampus potentially contributes to multiple forms of learning and memory. The extent to which memory relies on hippocampal processing per se depends on the extent to which context discrimination (i.e., context-based pattern completion and pattern separation) is required to perform the task. This view contrasts with the traditional view of assigning specific types of memories to individual brain structures. It is more consistent, however, with emerging views that there may be ‘‘multiple memory traces’’ that operate in parallel during a particular learning situation (e.g., Moscovitch et al., 2005). Hippocampal neural correlates with specific behavioral acts are not necessarily only a reflection of the ongoing behavior; it may also reflect a learned association between expected contextual information and the relevant response (or aspect of behavior such as velocity). Such an integrated representation could be useful to provide information to the local computational network about the expected behavioral con-
35
text of a task, a variable known to impact movementrelated responses of parietal cortex neurons in primates (Colby and Goldberg, 1999). The broad presence of behavioral context information across brain regions in addition to the hippocampus may reveal one of the functional architectures, or global domains, through which different neural processing systems are orchestrated (Mizumori et al., 2004b). If the behavioral context changes (resulting in altered firing patterns of the context-sensitive movement cells), information is fed back to the functional neuroarchitecture underlying behavioral expression. The behavioral expression domain refers to a processing network responsible for behavioral selection, planning of actions, and memory of behavioral acts. Frontal and parietal cortices are likely to be centrally involved in the operation of this functional domain. Feedback indicating a change in behavioral context may cause the neural-activity landscape within the behavioral expression domain to reconfigure, which in turn provides adaptive feedback that updates movement-sensitive representations in multiple neural systems. Parallel to the behavioral expression domain, we (Mizumori et al., 2004b) have postulated the existence of a distributed network that corresponds to the functional domain of context memory. In the special case of spatial context, the network likely includes the hippocampus, parietal cortex, and temporal cortex. This memory network may serve to coordinate context codes across different neural systems, such as between hippocampus and striatum (Mizumori et al., 2000b, 2004b). That is, the spatial-context memory network could define for different neural systems an expectation of sensory, behavioral, and reward elements of a learning situation. As noted above, this information could then be used in different ways to coordinate local network functions across different brain structures. Feedback to the spatial-context memory network from connected neural systems may be required to update memory as the learning situation changes. This process may in turn update memory representations within other functional domains, such as the behavioral expression system. There may be other functional domains that interact with the context and behavioral-control domains. One fundamentally important implication of the pattern of unit responses observed following context changes is that different categories of correlated neurons recorded within different neural systems respond most often to changes in context with only partial reorganization of firing patterns. That is, only a subset of place, movement, and reward-related cells respond to context change, while others do not. If contextindependent firing reflects expected information based
36 PLACE CELLS AND SPATIAL CONTEXT on past experience (e.g., expected spatial contexts, learned responses, or reinforcement outcomes), and if context-dependent neural codes reflect ongoing features of a current situation, then a fundamental operating principle that applies to diverse neural systems could be engagement in error-driven (match–mismatch) computations. That is, pattern completion and pattern separation may be basic computational algorithms that are common to most (if not all) neural structures. Such a conclusion is consistent with the prediction of computational models of striatal and hippocampal function (e.g., Houk, 1995; Manns et al., 2007). This conclusion is also consistent with the fact that many fields in neuroscience (from hindbrain analysis of sensory information to neocortex-mediated ‘‘decision making’’) employ a match–mismatch rule to interpret their findings. This type of computation may have been conserved across evolution, given that it is highly adaptive to detect changes in any type of information-processing network. Novelty detection may be a universal consequence of neural network function, one that emerges from common synaptic plasticity mechanisms that have been found in different brain areas. That neurons from different neural systems of multiple species exhibit representational neural reorganization in response to a mismatch in the expected context (or novelty detection; Merzenich and deCharms, 1996; Mizumori et al., 2000a; Kumaran and Maguire, 2006) further supports this view. Studies of hippocampal place fields have entered an exciting new chapter. There is growing new evidence that place fields make functional contributions to learning and memory. A significant challenge, however, is understanding precisely how this contribution is made. Our immediate goal is to rigorously test accounts, such as the CDH hierarchical model, that attempt to incorporate the multidimensional, dynamic, and conditional nature of hippocampal mnemonic processing. Only then can we move forward to understand how abnormal internal state conditions (e.g., degenerative diseases, poor motivation, or physiological conditions such as stress) impair hippocampal-dependent learning.
acknowledgments I would like to express special appreciation to the many outstanding past graduate and postdoctoral students (notably Brent Cooper, Oxana Eschenko, Stefan Leutgeb, Wayne Pratt, and David Smith) who have contributed significant data and stimulating discussions throughout the years. More recently, I thank Katy Gill, Min Jung Kim, Adria Martig, and Corey Puryear for their very thoughtful contributions. This work was supported by NIMH grant 58755.
References Alvarez-Leefmans FJ, Gardiner-Medwin AR (1975) Influences of the septum on the hippocampal dentate area which are unaccompanied by field potentials. J Physiol 249:14P–16P. Anagnostaras SG, Gale GD, Fanselow MS (2001) Hippocampus and contextual fear conditioning: recent controversies and advances. Hippocampus 11:8–17. Anderson MI, Jeffery KJ (2003) Heterogeneous modulation of place cell firing by changes in context. J Neurosci 23:8827–8835. Assaf SY, Miller JJ (1978) Neuronal transmission in the dentate gyrus: role of inhibitory mechanisms. Brain Res 151:587–592. Baeg EH, Kim YB, Huh K, Mook-Jung I, Kim HT, Jung MW (2003) Dynamics of population code for working memory in the prefrontal cortex. Neuron 40:177–188. Barnes CA, McNaughton BL, Mizumori SJY, Leonard BW, Lin L-H (1990) Comparison of spatial and temporal characteristics of neuronal activity in sequential stages of hippocampal processing. Prog Brain Res 83:287–300. Barry C, Lever C, Hayman R, Hartley T, Burton S, O’Keefe J, Jeffery K, Burgess N (2006) The boundary vector cell model of place cell firing and spatial memory. Rev Neurosci 17:71–97. Best PJ, White AM, Minai A (2001) Spatial processing in the brain: the activity of hippocampal place cells. Annu Rev Neurosci 24:459–486. Bi GQ, Poo MM (1998) Synaptic modifications in cultured hippocampal neurons: dependence on spike timing, synaptic strength, and postsynaptic cell type. J Neurosci 18:10464–10472. Bienenstock EL, Cooper LN, Munro PW (1982) Theory for the development of neuron selectivity: orientation specificity and binocular interaction in visual cortex. J Neurosci 2:32–48. Bilkey DK, Goddard GV (1985) Medial septal facilitation of hippocampal granule cell activity is mediated by inhibition of inhibitory interneurones. Brain Res 361: 99–106. Bostock E, Muller RU, Kubie JL (1991) Experiencedependent modifications of hippocampal place cell firing. Hippocampus 1:193–205. Bower MR, Duston AR, McNaughton BL (2005) Sequential-context-dependent hippocampal activity is not necessary to learn sequences with repeated elements. J Neurosci 25:1313–1323. Burgess N, Becker S, King JA, O’Keefe J (2001) Memory for events and their spatial context: models and experiments. Phil Trans R Soc Lond Biol Sci 356:1493– 1503. Buzsa´ki G (2002) Theta oscillations in the hippocampus. Neuron 33:325–340.
HIPPOCAMPAL PLACE CELLS DURING LEARNING
Buzsa´ki G (2005) Theta rhythm of navigation: link between path integration and landmark navigation, episodic and semantic memory. Hippocampus 15:827– 840. Colby CL, Goldberg ME (1999) Space and attention in parietal cortex. Annu Rev Neurosci 22:319–349. Cooper BG, Mizumori SJY (2001) Temporary inactivation of retrosplenial cortex causes a transient reorganization of spatial coding in hippocampus. J Neurosci 21:3986–4001. Czurko´ A, Hirase H, Csicsvari J, Buzsa´ki G (1999) Sustained activation of hippocampal pyramidal cells by ‘space clamping’ in a running wheel. Eur J Neurosci 11:344–352. Dahl D, Winson J (1985) Action of norepinephrine in the dentate gyrus. I. Stimulation of locus coeruleus. Exp Brain Res 59:491–496. Dan Y, Poo MM (2006) Spike timing-dependent plasticity: from synapse to perception. Physiol Rev 86: 1033–1048. Dragoi G, Buzsa´ki G (2006) Temporal encoding of place sequences by hippocampal cell assemblies. Neuron 50:145–57. Eichenbaum H (2000) A cortical–hippocampal system for declarative memory. Nat Rev Neurosci 1:41–50. Eichenbaum H, Cohen NJ (2001) From Conditioning to Conscious Recollection: Memory Systems of the Brain. New York: Oxford University Press. Eichenbaum H, Dudchenko P, Wood E, Shapiro M, Tanila H (1999) The hippocampus, memory, and place cells: is it spatial memory or a memory space? Neuron 425:184–188, Ekstrom AD, Kahana MJ, Caplan JB, Fields TA, Isham EA, Newman EL, Fried I (2003) Cellular networks underlying human spatial navigation. Nature 425: 184–188. Ekstrom AD, Meltzer J, McNaughton BL, Barnes CA (2001) NMDA receptor antagonism blocks experiencedependent expansion of hippocampal ‘‘place fields.’’ Neuron 31:631–638. Eschenko O, Mizumori SJY (2007) Memory influences on hippocampal and striatal neural codes: effects of a shift between task rules. Neurobiol Learn Mem 87: 495–509. Etienne AS, Jeffery KJ (2004) Path integration in mammals. Hippocampus 14:180–192. Fanselow MS, Poulos AM (2005) The neuroscience of mammalian associative learning. Annu Rev Psychol 56:207–234. Ferbinteanu J, Shapiro ML (2003) Prospective and retrospective memory coding in the hippocampus. Neuron 40:1227–1239. Fortin NJ, Agster KL, Eichenbaum HB (2002) Critical role of the hippocampus in memory for sequences of events. Nat Neurosci 5:458–462.
37
Foster DJ, Wilson MA (2006) Reverse replay of behavioural sequences in hippocampal place cells during the awake state. Nature 440:680–683. Foster TC, Castro CA, McNaughton BL (1989) Spatial selectivity of rat hippocampal neurons: dependence on preparedness for movement. Science 244:1580– 1582. Foster TC, Christian EP, Hampson RE, Campbell KA, Deadwyler SA (1987) Sequential dependencies regulate sensory evoked responses of single units in the rat hippocampus. Brain Res 408:86–96. Fox SE, Ranck JB Jr (1981) Electrophysiological characteristics of hippocampal complex-spike cells and theta cells. Exp Brain Res 41:399–410. Frank LM, Brown EM, Wilson MA (2000) Trajectory encoding in the hippocampus and entorhinal cortex. Neuron 27:169–178. Frank LM, Stanley GB, Brown EN (2004) Hippocampal plasticity across multiple days of exposure to novel environments. J Neurosci 24:7681–7689. Freeman JH Jr, Cuppernell C, Flannery K, Gabriel M (1996a) Context-specific multi-site cingulate cortical, limbic thalamic, and hippocampal neuronal activity during concurrent discriminative approach and avoidance training in rabbits. J Neurosci 16:1538–1549. Freeman JH Jr, Cuppernell C, Flannery K, Gabriel M (1996b) Limbic thalamic, cingulate cortical and hippocampal neuronal correlates of discriminative approach learning in rabbits. Behav Brain Res 80:123– 136. Freund TF, Antal M (1988) GABA-containing neurons in the septum control inhibitory interneurons in the hippocampus. Nature 336:170–173. Freund TF, Buzsaki G (1996) Interneurons of the hippocampus. Hippocampus 6:347–470. Fried I, MacDonald KA, Wilson CL (1997) Single neuron activity in human hippocampus and amygdala during recognition of faces and objects. Neuron 18:753–765. Fyhn M, Molden S, Hollup S, Moser MB, Moser E (2002) Hippocampal neurons responding to first-time dislocation of a target object. Neuron 35:555–566. Gavrilov VV, Wiener SI, Berthoz A (1998) Discharge correlates of hippocampal complex spike neurons in behaving rats passively displaced on a mobile robot. Hippocampus 8:475–490. Gemmell C, O’Mara SM (1999) Medial prefrontal cortex lesions cause deficits in a variable-goal location task but not in object exploration. Behav Neurosci 113: 465–474. Gilbert PE, Kesner RP, Lee I (2001) Dissociating hippocampal subregions: double dissociation between dentate gyrus and CA1. Hippocampus 11:626–636 Gold E, Kesner RP (2005) The role of the CA3 subregion of the dorsal hippocampus in spatial pattern completion in the rat. Hippocampus 15:808–814.
38 PLACE CELLS AND SPATIAL CONTEXT Gold PE (2003) Acetylcholine modulation of neural systems involved in learning and memory. Neurobiol Learn Mem 80:194–210. Good M, Honey RC (1991) Conditioning and contextual retrieval in hippocampal rats. Behav Neurosci 105: 499–509. Gothard KM, Skaggs WE, Moore KM, McNaughton BL (1996) Binding of hippocampal CA1 neural activity to multiple reference frames in a landmark-based navigation task. J Neurosci 16:825–835. Granon S, Save E, Buhot MC, Poucet B (1996) Effortful information processing in a spontaneous spatial situation by rats with medial prefrontal lesions. Behav Brain Res 78:147–154. Gray JA (1982) The Neuropsychology of Anxiety: An Enquiry into the Functions of the Septo-hippocamal System. Oxford: Oxford University Press. Gray JA (2000) The Neuropsychology of Anxiety: An Enquiry into the Functions of the Septo-hippocamal System, 2nd edition. Oxford: Oxford University Press. Guzowski JF, Knierim JJ, Moser EI (2004) Ensemble dynamics of hippocampal regions CA3 and CA1. Neuron 44:581–584. Hafting T, Fyhn M, Molden S, Moser MB, Moser EI (2005) Microstructure of a spatial map in the entorhinal cortex. Nature 436:801–806. Hampson RE, Heyser CJ, Deadwyler SA (1993) Hippocampal cell firing correlates of delayed-match-tosample performance in the rat. Behav Neurosci 107: 715–739. Harzi MM, Jarrard LE (1992) Effects of medial and lateral septal lesions on acquisition of a place and cue radial maze task. Behav Brain Res 49:159–165. Hasselmo ME (2005a) The role of hippocampal regions CA3 and CA1 in matching entorhinal input with retrieval of associations between objects and context: theoretical comment on Lee et al. (2005). Behav Neurosci 119:342–345. Hasselmo ME (2005b) What is the function of hippocampal theta rhythm?—Linking behavioral data to phasic properties of field potential and unit recording data. Hippocampus 15:936–949. Hasselmo ME, Hay J, Ilyn M, Gorchetchnikov A (2002) Neuromodulation, theta rhythm and rat spatial navigation. Neural Networks 15:689–707. Hasselmo ME, McGaughy J (2004) High acetylcholine levels set circuit dynamics for attention and encoding and low acetylcholine levels set dynamics for consolidation. Prog Brain Res 145:207–231. Heit G, Smith ME, Halgren E (1998) Neural encoding of individual words and faces by the human hippocampus and amygdala. Nature 333:773–5. Hetherington PA, Shapiro ML (1997) Hippocampal place fields are altered by the removal of single visual cues
in a distance-dependent manner. Behav Neurosci 11: 20–34. Hill AJ, Best PJ (1981) Effects of deafness and blindness on the spatial correlates of hippocampal unit activity in the rat. Exp Neurol 74:204–217. Hirsh R (1974) The hippocampus and contextual retrieval of information from memory: a theory. Behav Biol 12: 421–444. Hoffman KL, McNaughton BL (2002) Sleep on it: cortical reorganization after-the-fact. Trends Neurosci 25:1–2. Hopfield JJ (1982) Neural networks and physical systems with emergent collective computational abilities. Proc Natl Acad Sci USA 79:2554–2558. Houk JC (1995) Information processing in modular circuits linking basal ganglia and cerebral cortex. In: Models of Information Processing in the Basal Ganglia (Houk JC, Davis JL, Beiser DG, eds.), pp 3–10. Cambridge, MA: MIT Press. Hyman JM, Zilli EA, Paley AM, Hasselmo ME (2005) Medial prefrontal cortex cells show dynamic modulation with the hippocampal theta rhythm dependent on behavior. Hippocampus 15:739–749. Ikonen S, McMahan R, Gallagher M, Eichenbaum H, Tanila H (2002) Cholinergic system regulation of spatial representation by the hippocampus. Hippocampus 12:386–397. Jakab RL, Leranth C (1995) Septum. In: The Rat Nervous System, 2nd edition (Paxinos G, ed.), pp 405–442. San Diego: Academic Press. Jay TM, Glowinski J, Thierry AM (1989) Selectivity of the hippocampal projection of the prelimbic area of the prefrontal cortex in the rat. Brain Res 505:337– 340. Jay TM, Witter MP (1991) Distribution of hippocampal CA1 and subicular efferents in the prefrontal cortex of the rat studied by means of anterograde transport of Phaseolus vulgaris-leucoagglutinin. J Comp Neurol 313:574–586. Jeffery KJ (2003) The Neurobiology of Spatial Behaviour. New York: Oxford University Press. Jeffery KJ, Anderson MI, Hayman R, Chakraborty S (2004) A proposed architecture for the neural representation of spatial context. Neurosci Biobehav Rev 28:201–218. Jeffery KJ, Gilbert A, Burton S, Strudwick A (2003) Preserved performance in a hippocampal-dependent spatial task despite complete place cell remapping. Hippocampus 13:175–189. Jenkins TA, Amin E, Pearce JM, Brown MW, Aggleton JP (2004) Novel spatial arrangements of familiar visual stimuli promote activity in the rat hippocampal formation but not the parahippocampal cortices: a cfos expression study. Neuroscience 124:43–52.
HIPPOCAMPAL PLACE CELLS DURING LEARNING
Jones MW, Wilson MA (2005) Theta rhythms coordinate hippocampal-prefrontal interactions in a spatial memory task. PLoS Biol 3:e402. Jung MW, Qin Y, McNaughton BL, Barnes CA (1998) Firing characteristics of deep layer neurons in prefrontal cortex in rats performing spatial working memory tasks. Cereb Cortex 8:437–450. Jung MW, Wiener SI, McNaughton BL (1994) Comparison of spatial firing characteristics of units in dorsal and ventral hippocampus of the rat. J Neurosci 14:7347–7356. Kennedy PJ, Shapiro ML (2004) Retrieving memories via internal context requires the hippocampus. J Neurosci 24:6979–6985. Kesner RP (1991) Neurobiological views of memory. In: Learning and Memory: A Biological View (Martinez JL Jr, Kesner RP, eds.), pp 499–547. San Diego: Academic Press. Kesner RP, Lee I, Gilbert P (2004) A behavioral assessment of hippocampal function based on a subregional analysis. Rev Neurosci 15:333–351. Kim JJ, Fanselow MS (1992) Modality-specific retrograde amnesia of fear. Science 256:675–677. Knierim JJ, Kudrimoti H, McNaughton BL (1995) Hippocampal place fields, the internal compass, and the learning of landmark stability. J Neurosci 15:1648– 1659. Knierim JJ, Kudrimoti HS, McNaughton BL (1998) Interactions between idiothetic cues and external landmarks in the control of place cells and head direction cells. J Neurophysiol 80:425–446. Koulakov AA, Raqhavachari S, Kepecs A, Lisman JE (2002) Model for a robust neural integrator. Nat Neurosci 5:775–782. Kumaran D, Maguire EA (2006) An unexpected sequence of events: mismatch detection in the human hippocampus. PLoS Biol 4:2372–2382. Lee I, Griffin AL, Zilli EA, Eichenbaum H, Hasselmo ME (2006) Gradual translocation of spatial correlates of neuronal firing in the hippocampus toward prospective reward locations. Neuron 51:639–650. Lee I, Yoganarasimha D, Rao G, Knierim JJ (2004) Comparison of population coherence of place cells in hippocampal subfields CA1 and CA3. Nature 430: 456–459. Lenck-Santini PP, Save E Poucet B (2001) Evidence for a relationship between place-cell spatial firing and spatial memory performance. Hippocampus 11:377– 390. Leutgeb JK, Leutgeb S, Treves A, Meyer R, Barnes CA, McNaughton BL, Moser MB, Moser EI (2005a) Progressive transformation of hippocampal neuronal representations in ‘‘morphed’’ environments. Neuron 48:345–58.
39
Leutgeb S, Leutgeb JK, Barnes CA, Moser EI, McNaughton BL, Moser MB (2005b) Independent codes for spatial and episodic memory in hippocampal neuronal ensembles. Science 309:619–623. Leutgeb S, Leutgeb JK, Treves A, Moser MB, Moser EI (2004) Distinct ensemble codes in hippocampal areas CA3 and CA1. Science 305:1295–1298. Leutgeb S, Mizumori SJY (1999) Excitotoxic septal lesions result in spatial memory deficits and altered flexibility of hippocampal single-unit representations. J Neurosci 19:6661–6672. Lever C, Wills T, Caccucci F, Burgess N, O’Keefe J (2002) Long-term plasticity in hippocampal placecell representation of environmental geometry. Nature 416:90–94. Lisman JE (1999) Relating hippocampal circuitry to function: recall of memory sequences by reciprocal dentate-CA3 interactions. Neuron 22:233–242. Lisman JE, Grace AA (2005) The hippocampal-VTA loop: controlling the entry of information into longterm memory. Neuron 46:703–713. Lisman JE, Otmakhova NA (2001) Storage, recall, and novelty detection of sequences by the hippocampus: elaborating on the SOCRATIC model to account for normal and aberrant effects of dopamine. Hippocampus 11:551–68. Maguire EA, Woollett K, Spiers HJ (2006) London taxi drivers and bus drivers: a structural MRI and neuropsychological analysis. Hippocampus 16:1091–101. Manns JR, Zilli EA, Ong KC, Hasselmo ME, Eichenbaum H (2007) Hippocampal CA1 spiking during encoding and retrieval: relation to theta phase. Neurobiol Learn Mem 87:9–20. Maren S (2001) Neurobiology of Pavlovian fear conditioning. Annu Rev Neurosci 2:897–931. Markram H, Lubke J, Frotscher M, Sakmann B (1997) Regulation of synaptic efficacy by coincidence of postsynaptic APs and EPSPs. Science 275:213–215. Markus EJ, Barnes CA, McNaughton BL, Gladden VL, Skaggs WE (1994) Spatial information content and reliability of hippocampal CA1 neurons: effects of visual input. Hippocampus 4:410–421. Markus EJ, Qin YL, Leonard B, Skaggs WE, McNaughton BL, Barnes CA (1995) Interactions between location and task affect the spatial and directional firing of hippocampal neurons. J Neurosci 15:7079–7094. Marr D (1969) A theory of cerebellar cortex. J Physiol 202:437–70 Marr D (1971) Simple memory: a theory for archicortex. Philos Trans R Soc Lond B Biol Sci 262:23–81. Martin SJ, Grimwood PD, Morris RG (2000) Synaptic plasticity and memory: an evaluation of the hypothesis. Annu Rev Neurosci 23:649–711.
40 PLACE CELLS AND SPATIAL CONTEXT Maurer AP, VanRhoads SR, Sutherland GR, Lipa P, McNaughton BL (2005) Self-motion and the origin of differential spatial scaling along the septo-temporal axis of the hippocampus. Hippocampus 15:841–852. McClelland JL, McNaughton BL, O’Reilly RC (1995) Why there are complementary learning systems in the hippocampus and neocortex: insights from the success and failures of connectionist models of learning and memory. Psychol Rev 102:419–457. McEchron MD, Disterhoft JG (1999) Hippocampal encoding of non-spatial trace conditioning. Hippocampus 9:385–396. McHugh TJ, Blum KI, Tsien JZ, Tonegawa S, Wilson MA (1996) Impaired hippocampal representation of space in CA1-specific NMDAR1 knockout mice. Cell 87:1339–1349. McNaughton BL, Barnes CA, Gerrard JL, Gothard K, Jung MW, Knierim JJ, Kudrimoti H, Qin Y, Skaggs WE, Suster M, Weaver KL (1996) Deciphering the hippocampal polyglot: the hippocampus as a path integration system. J Exp Biol 199:173–185. McNaughton BL, Barnes CA, O’Keefe J (1983) The contributions of position, direction, and velocity to single unit activity in the hippocampus of freely moving rats. Exp Brain Res 52:41–49. McNaughton BL, Battaglia FP, Jensen O, Moser EI, Moser MB (2006) Path integration and the neural basis of the ‘cognitive map.’ Nat Rev Neurosci 7:663– 678. McNaughton BL, Morris RGM (1987) Hippocampal synaptic enhancement and information storage within a distributed memory system. Trends Neurosci 10: 408–415. Mehta MR, Barnes CA, McNaughton BL (1997) Experience-dependent, asymmetric expansion of hippocampal place fields. Proc Natl Acad Sci USA 94: 8918–8921. Mehta MR, Lee AK, Wilson MA (2002) Role of experience and oscillations in transforming a rate code into a temporal code. Nature 417:741–746. Mehta MR, Quirk MC, Wilson MA (2000) Experiencedependent asymmetric shape of hippocampal receptive fields. Neuron 25:707–715. Merzenich MM, deCharms RC (1996) Neural representation, experience, and change. In: Mind–Brain Continuum: Sensory Processes (Llinas R, Churchland PS, eds.), pp 61–81. Cambridge, MA: MIT Press. Mittelstaedt MK, Mittelstaedt H (1982) Homing by path integration. In: Avian Navigation (Papi F, Wallraff HG, eds.), pp 290–297. Berlin: Springer-Verlag. Mizumori SJY, Cooper BG, Leutgeb S, Pratt WE (2000a) A neural systems analysis of adaptive navigation. Mol Neurobiol 21:57–82. Mizumori SJY, Garcia PA, Volpe BT, Raja M (1995) Reorganization of spatial representation in hippo-
campus following ischemia. Behav Neurosci 109: 1081–1094. Mizumori SJY, Kalyani A (1997) Age and experiencedependent representational reorganization during spatial learning. Neurobiol Aging 18:651–659. Mizumori SJY, Lavoie AM, Kalyani A (1996) Redistribution of spatial representation in the hippocampus of aged rats performing a spatial memory task. Behav Neurosci 110:1006–1016. Mizumori SJY, McNaughton BL, Barnes CA (1989a) A comparison of supramammillary and medial septal influences on hippocampal field potentials and singleunit activity. J Neurophysiol 61:15–31. Mizumori SJY, McNaughton BL, Barnes CA, Fox KB (1989b) Preserved spatial coding in hippocampal CA1 pyramidal cells during reversible suppression of CA3 output: evidence for pattern completion in HPC. J Neurosci 9:3915–3928. Mizumori SJY, Miya DY, Ward KE (1994) Reversible inactivation of the lateral dorsal thalamus disrupts hippocampal place representation and impairs spatial learning. Brain Res 644:168–174. Mizumori SJY, Puryear CM, Gill KM, Guazzelli A (2004a) Head direction codes in hippocampal afferent and efferent systems: what functions do they serve? In: Head Direction Cells and the Neural Mechanisms Underlying Directional Orientation (Wiener SL, Taube JS, eds.), pp 203–220. Cambridge, MA: MIT Press. Mizumori SJY, Ragozzino KE, Cooper BG (2000b) Location and head direction representation in the dorsal striatum of rats. Psychobiology 28:441–462. Mizumori SJY, Ragozzino KE, Cooper BG, Leutgeb S (1999) Hippocampal representational organization and spatial context. Hippocampus 9:444–451. Mizumori SJY, Smith DM, Puryear CB (2007) Mnemonic contributions of hippocampal place cells. In: Neurobiology of Learning and Memory, second edition (Martinez JL, Kesner RP, eds.), pp 155–190. San Diego: Academic Press. Mizumori SJY, Williams JD (1993) Directionally selective mnemonic properties of neurons in the lateral dorsal nucleus of the thalamus of rats. J Neurosci 13: 4015–4028. Mizumori SJY, Yeshenko O, Gill K, Davis D (2004b) Parallel processing across neural systems: implications for a multiple memory systems hypothesis. Neurobiol Learn Mem 82:278–298. Moita MA, Rosis S, Zhou Y, LeDoux JE, Blair HT (2004) Putting fear in its place: remapping of hippocampal place cells during fear conditioning. J Neurosci 24: 7015–7023. Morris RGM (2001) Episodic-like memory in animals psychological criteria, neural mechanisms and the value of episodic-like tasks to investigate animal
HIPPOCAMPAL PLACE CELLS DURING LEARNING
models of neurodegenerative disease. Phil Trans R Soc Lond Biol Sci 356:1453–1465. Morris RGM (2006) Elements of a neurobiological theory of hippocampal function: the role of synaptic plasticity, synaptic tagging and schemas. Eur J Neurosci 23:2829–46 Morris RGM, Frey U (1997) Hippocampal synaptic plasticity: role in spatial learning or the automatic recording of attended experience? Philos Trans R Soc Lond B Biol Sci 352:489–503. Moscovitch M, Rosenbaum RS, Gilboa A, Addis DR, Westmacott R, Grady C, McAndrews MP, Levine B, Black S, Winocur G, Nadel L (2005) Functional neuroanatomy of remote episodic, semantic and spatial memory: a unified account based on multiple trace theory. J Anat 207:35–66. Muller RU, Kubie JL (1987) The effects of changes in the environment on the spatial firing of hippocampal complex-spike cells. J Neurosci 7:1951–1968. Muller RU, Kubie JL (1989) The firing of hippocampal place cells predicts the future position of freely moving rats. J Neurosci 9:4101–4110. Muller RU, Kubie JL, Saypoff R (1991) The hippocampus as a cognitive graph (abridged version). Hippocampus 1:243–246. Muller RU, Stead M, Pach J (1996) The hippocampus as a cognitive graph. J Gen Physiol 107:663–694. Myers CE, Gluck M (1994) Context, conditioning, and hippocampal representation in animal learning. Behav Neurosci 108:835–847. Nadel L, Payne JD (2002) The hippocampus, wayfinding and episodic memory. In: The Neural Basis of Navigation: Evidence from Single Cell Recording (Sharp PE, ed.), pp 235–248. New York: Kluwer Academic Publishing. Nadel L, Wilner J (1980) Context and conditioning: a place for space. Physiol Psychol 8:218–228. Nadel L, Willner J, Kurz EM (1985) Cognitive maps and environmental context. In: Context and Learning (Balsam P, Tomie A, eds.), pp 385–406. Hillside, NJ: Lawrence Erlbaum Associates. O’Keefe J (1976) Place units in the HPC of the freely moving rat. Exp Neurol 51:78–109. O’Keefe J, Burgess N (1996a) Geometric determinants of the place fields of hippocampal neurons. Nature 381:425–428. O’Keefe J, Burgess N (1996b) Spatial and temporal determinants of hippocampal place cell activity. In: Perception, Memory and Emotion: Frontiers in Neuroscience (Ono T, McNaughton BL, Molotchnikoff S, Rolls ET, Nishijo H, eds.), pp 359–373. New York: Pergamon Press. O’Keefe J, Conway DH (1978) Hippocampal place units in the freely moving rat: why they fire where they fire. Exp Brain Res 31:573–590.
41
O’Keefe J, Dostrovsky J (1971) The hippocampus as a spatial map. Preliminary evidence from unit activity in the freely moving rat. Brain Res 34:171–175. O’Keefe J, Nadel L (1978) The Hippocampus as a Cognitive Map. Oxford: Oxford University Press. O’Keefe J, Recce ML (1993) Phase relationship between hippocampal place units and the EEG theta rhythm. Hippocampus 3:317–330. O’Keefe J, Speakman A (1987) Single unit activity in the rat hippocampus during a spatial memory task. Exp Brain Res 68:1–27. Olton DS, Becker JT, Handelmann GE (1979) Hippocampus, space, and memory. Brain Behav Sci 2:313–365. O’Mara SM (1995) Spatially selective firing properties of hippocampal formation neurons in rodents and primates. Prog Neurobiol 45:253–274. O’Reilly RC, McClelland JL (1994) Hippocampal conjunctive encoding, storage, and recall: avoiding a trade-off. Hippocampus 4:661–682. O’Reilly RC, Rudy JW (2001) Conjunctive representations in learning and memory: Principles of cortical and hippocampal function. Psychol Rev 108:311–345. Paulsen O, Moser EI (1998) A model of hippocampal memory encoding and retrieval: GABAergic control of synaptic plasticity. Trends Neurosci 21:273–278. Penick S, Solomon PR (1991) Hippocampus, context, and conditioning. Behav Neurosci 105:611–617. Pennartz CM, Uylings HB, Barnes CA, McNaughton BL (2002) Memory reactivation and consolidation during sleep: from cellular mechanisms to human performance. Prog Brain Res 138:143–66. Phillips RG, LeDoux JE (1992) Differential contribution of amygdala and hippocampus to cued and contextual fear conditioning. Behav Neurosci 106:274–285. Phillips RG, LeDoux JE (1994) Lesions of the dorsal hippocampal formation interfere with background but not foreground contextual fear conditioning. Learn Mem 1:34–44. Poucet B (1997) Searching for spatial unit firing in the prelimbic area of the rat medial prefrontal cortex. Behav Brain Res 84:151–159. Poucet B, Herrmann T (1990) Septum and medial frontal cortex contribution to spatial problem-solving. Behav Brain Res 37:269–280. Poucet B, Save E, Lenck-Santini PP (2000) Sensory and memory properties of hippocampal place cells. Rev Neurosci 11:95–111. Pratt WE, Mizumori SJY (2001) Neurons in rat medial prefrontal cortex show anticipatory rate changes to predictable differential rewards in a spatial memory task. Behav Brain Res 123:165–183. Puryear CB, King M, Mizumori SJY (2006) Specific changes in hippocampal spatial codes predict spatial working memory performance. Behav Brain Res 169: 168–175.
42 PLACE CELLS AND SPATIAL CONTEXT Quirk GJ, Muller RU, Kubie JL (1990) The firing of hippocampal place cells in the dark depends on the rat’s recent experience. J Neurosci 10:2008–2017. Ragozzino ME, Pal SN, Unick K, Stefani MR, Gold PE (1998) Modulation of hippocampal acetylcholine release and spontaneous alternation scores by intrahippocampal glucose injections. J Neurosci 18:1595– 1601. Ranck Jr JR (1973) Studies on single neurons in dorsal hippocampus formation and septum in unrestrained rats. Part I. Behavioral correlates and firing repertoires. Exp Neurol 41:461–535. Rawlins JNP (1985) Associations across time: the hippocampus as a temporary memory store. Brain Behav Sci 8:479–496. Redish AD, Rosesnzweig ES, Bohanick JD, McNaughton BL, Barnes CA (2000) Dynamics of hippocampal ensemble activity realignment: time versus space. J Neurosci 20:9298–9309. Redish AD, Touretzky DS (1997) Cognitive maps beyond the hippocampus. Hippocampus 7:15–35. Risold PY, Swanson LW (1995) Evidence for a hypothalamothalamocortical circuit mediating pheromonal influences on eye and head movements. Proc Nat Acad Sci USA 92:3898–3902. Rolls ET (1996) A theory of hippocampal function in memory. Hippocampus 6:601–620. Rolls ET (1999) Spatial view cells and the representation of place in the primate hippocampus. Hippocampus 9:467–480. Rolls ET, Treves A (1998) Neural Networks and Brain Function. Oxford: Oxford University Press. Sakurai Y (1994) Involvement of auditory cortical and hippocampal neurons in auditory working memory and reference memory in the rat. J Neurosci 14:2606– 2623. Sargolini F, Fyhn M, Hafting T, McNaughton BL, Witter MP, Moser MB, Moser EI (2006) Conjunctive representation of position, direction, and velocity in entorhinal cortex. Science 312:758–762. Save E, Buhot MC, Foreman N, Thinus-Blanc C (1992a) Exploratory activity and response to a spatial change in rats with hippocampal or posterior parietal cortical lesions. Behav Brain Res 47:113–127. Save E, Nerad L, Poucet B (2000) Contribution of multiple sensory information to place field stability in hippocampal place cells. Hippocampus 10:64–76. Save E, Poucet B, Foreman N, Buhot MC (1992b) Object exploration and reactions to spatial and nonspatial changes in hooded rats following damage to parietal cortex or hippocampal formation. Behav Neurosci 106:447–456. Schendan HE, Searl MM, Melrose RJ, Stern CE (2003) An FMRI study of the role of the medial temporal
lobe in implicit and explicit sequence learning. Neuron 37:1013–1025. Seamans JK, Floresco SB, Phillips AG (1995) Functional differences between the prelimbic and anterior cingulate regions of the rat prefrontal cortex. Behav Neurosci 109:1063–1073. Segal M (1979) A potent inhibitory monosynaptic hypothalamo-hippocampal connection. Brain Res 162: 137–141. Shapiro ML, Eichenbaum H (1999) Hippocampus as a memory map: synaptic plasticity and memory encoding by hippocampal neurons. Hippocampus 9:365– 384. Shapiro ML, Ferbinteanu J (2006) Relative spike timing in pairs of hippocampal neurons distinguishes the beginning and end of journeys. Proc Natl Acad Sci USA 103:4287–4292. Shapiro ML, Tanila H, Eichenbaum H (1997) Cues that hippocampal place cells encode: dynamic and hierarchical representation of local and distal stimuli. Hippocampus 7:624–642. Sharp PE (2002) The Neural Basis of Navigation: Evidence from Single Cell Recording. Boston: Kluwer Academic Publishers. Sharp PE, Blair HT, Cho J (2001) The anatomical and computational basis of the rat head-direction cell signal. Trends Neurosci 24:289–294. Shors TJ, Matzel LD (1997) Long-term potentiation: what’s learning got to do with it? Behav Brain Sci 20: 597–614. Siapas AG, Lubenov EV, Wilson MA (2005) Prefrontal phase locking to hippocampal theta oscillations. Neuron 46:141–151. Skaggs WE, McNaughton BL (1998) Spatial firing properties of hippocampal CA1 populations in an environment containing two visually identical regions. J Neurosci 18:8455–8466. Skaggs WE, McNaughton BL, Wilson MA, Barnes CA (1996) Theta phase precession in hippocampal neuronal populations and the compression of temporal sequences. Hippocampus 6:149–172. Smith DM, Mizumori SJY (2006) Learning-related development of context-specific neuronal responses to places and events: the hippocampal role in context processing. J Neurosci 26:3154–3163. Smith DM, Wakeman D, Patel J, Gabriel M (2004) Fornix lesions impair context-related cingulothalamic neuronal patterns and concurrent discrimination learning. Behav Neurosci 118:1225–1239. Song EY, Kim YB, Kim YH, Jung MW (2005) Role of active movement in place-specific firing of hippocampal neurons. Hippocampus 15:8–17. Swanson LW, Cowan WM (1979) The connections of the septal region in the rat. J Comp Neurol 186:621–656.
HIPPOCAMPAL PLACE CELLS DURING LEARNING
Tanila H (1998) Unique features in the processing of spatial information by the aged hippocampus as shown by place cell recording studies: a response to Peter Rapp’s commentary ‘‘Representational organization in the aged hippocampus.’’ Hippocampus 8: 436–437. Tanila H, Shapiro ML, Eichenbaum H (1997) Discordance of spatial representation in ensembles of hippocampal place cells. Hippocampus 7:613–623, Taube JS, Muller RU (1998) Comparisons of head direction cell activity in the postsubiculum and anterior thalamus of freely moving rats. Hippocampus 8:87– 108. Treves A (2004) Computational constraints between retrieving the past and predicting the future, and the CA3-CA1 differentiation. Hippocampus 14:539–556. Tulving E (2002) Episodic memory: from mind to brain. Annu Rev Psychol 53:1–25. Vanderwolf C (1969) Hippocampal electrical activity and voluntary movement in the rat. Electroencephalogr Clin Neurophysiol 26:407–418. Vertes RP, Hoover WB, Viana Di Prisco G (2004) Theta rhythm of the hippocampus: subcortical control and functional significance. Behav Cogn Neurosci Rev 3: 173–200. Vinogradova OS (1995) Expression, control, and probably functional significance of the neuronal thetarhythm. Prog Neurobiol 45:523–583. Westmacott R, Black SE, Freedman M, Moscovitch M (2004) The contribution of autobiographical significance to semantic memory: evidence from Alzheimer’s disease, semantic dementia, and amnesia. Neuropsychologia 42:25–48. Whishaw IQ, Gorny B (1999) Path integration absent in scent-tracking fimbria-fornix rats: evidence for hippocampal involvement in ‘‘sense of direction’’ and ‘‘sense of distance’’ using self-movement cues. J Neurosci 19:4662–4673. Wible CG, Findling RL, Shapiro M, Lang EJ, Crane S, Olton DS (1986) Mnemonic correlates of unit activity in the hippocampus. Brain Res 399:97–110. Wiener SI (1996) Spatial, behavioral and sensory correlates of hippocampal CA1 complex spike cell activity: implications for information processing functions. Prog Neurobiol 49:335–361. Wiener SI, Korshunov VA, Garcia R, Berthoz A (1995) Inertial, substratal and landmark cue control of hippocampal CA1 place cell activity. Eur J Neurosci 7: 2206–2219.
43
Wiener SI, Paul CA, Eichenbaum H (1989) Spatial and behavioral correlates of hippocampal neuronal activity. J Neurosci 9:2737–2763. Wilson MA, McNaughton BL (1993) Dynamics of the hippocampal ensemble code for space. Science 261: 1055–1058. Wilson MA, McNaughton BL (1994). Reactivation of hippocampal ensemble memories during sleep. Science 265:676–679. Winson J (1978) Loss of hippocampal theta rhythm results in spatial memory deficit in the rat. Science 201: 160–163. Winson J (1984) Neural transmission through the hippocampus: dependence on behavioral state. In: Cortical Integration (Reinoso-Suarez F, Ajmone-Marsan C, eds.), pp.131–146. New York: Raven Press. Wirth S, Yanike M, Frank LM, Smith AC, Suzuki WA (2003) Single neurons in the monkey hippocampus and learning of new associations. Science 300:1578– 1581. Witter MP, Naber PA, van Haeften T, Machielsen WC, Rombouts SA, Barkhof F, Scheltens P, Lopes da Silva FH (2000) Cortico-hippocampal communication by way of parallel parahippocampal-subicular pathways. Hippocampus 10:398–410. Wood ER, Dudchenko PA, Eichenbaum H (1999) The global record of memory in hippocampal neuronal activity. Nature 397:613–616. Wood ER, Dudchenko PA, Tobitsek RJ, Eichenbaum H (2000) Hippocampal neurons encode information about different types of memory episodes occurring in the same location. Neuron 27:623–633. Yeshenko O, Guazzelli A, Mizumori SJY (2004) Context-dependent reorganization of spatial and movement representations by simultaneously recorded hippocampal and striatal neurons during performance of allocentric and egocentric tasks. Behav Neurosci 118:751–769. Young BJ, Fox GD, Eichenbaum H (1994) Correlates of hippocampal complex-spike cell activity in rats performing a nonspatial radial maze task. J Neurosci 14: 6553–6563. Zhou TL, Tamura R, Kuriwaki J, Ono T (1999) Comparison of medial and lateral septal neuron activity during performance of spatial tasks in rats. Hippocampus 9:220–234. Zugaro MB, Monconduit L, Buzsaki G (2005) Spike phase precession persists after transient intrahippocampal perturbation. Nat Neurosci 8:67–71.
3 Context-Dependent Firing of Hippocampal Place Cells: Does It Underlie Memory? JAMES A. AINGE, PAUL A. DUDCHENKO, AND EMMA R. WOOD
hippocampal activity and tasks that are hippocampus dependent is not clear-cut (see Table 3–1). Although in some hippocampus-dependent tasks context-dependent place field firing is seen, in others it is not. Furthermore, in some tasks that don’t require a hippocampus, context-dependent activity is observed. In this chapter we review these three patterns of results and try to identify the task characteristics that yield context-dependent place cell firing, as well as the characteristics of tasks that require the hippocampus.
One of the basic properties of hippocampal place cells is that an individual cell fires whenever a rat passes through the cell’s receptive field—its place field (O’Keefe, 1979; Muller, 1996; but see Fenton and Muller, 1998). However, when rats are trained on tasks in which the animal passes through the same location repeatedly but on its way to or from different locations, some place cells fire strongly on one type of journey and hardly at all on other journeys (Frank et al., 2000; Wood et al., 2000; Ferbinteanu and Shapiro, 2003; Bower et al., 2005). This conditional or context-dependent firing can be related to where the animal has come from (retrospective coding) or to where the animal is about to go (prospective coding), as well as to other aspects of the situation, such as the expected location of a goal (Smith and Mizumori, 2006b). Therefore, this activity is of interest because it suggests that the hippocampus processes information about both the animal’s current location and its purposive behavior. Smith and Mizumori (2006a) have argued that context-dependent hippocampal activity occurs whenever subjects must distinguish one situation from another to retrieve the correct behavioral responses or memories. They have further suggested that the neural representation of context generated in the hippocampus could be transmitted to extrahippocampal brain regions to facilitate contextual modulation of behavioral responses and memories. Each of these statements implies that the context-dependent activity in the hippocampus is necessary for guiding behavior. However, the relationship between context-dependent
CONTEXT-DEPENDENT HIPPOCAMPAL ACTIVITY IS OBSERVED IN SOME TASKS FOR WHICH THE HIPPOCAMPUS IS NOT REQUIRED One of the initial reports of hippocampal activity that differentiated between behavioral contexts in the absence of changes in the physical environment or the overt behavior of the animal was that of Wood et al. (2000). In this study, rats were trained to enter the left and right arms on alternate trials on a modified T-maze to obtain sweetened water rewards at the ends of the arms. After obtaining a reward at the end of a choice arm, the rats were required to return to the base of the stem of the T-maze via a connecting arm (see Fig. 3–1a) and then run up the central stem and enter the arm not entered on the previous trial. Animals repeated this continuous alternation behavior for 40 trials per session. Importantly, the central stem of the T-maze was traversed both on left-turn trials (in which 44
CONTEXT-DEPENDENT FIRING OF HIPPOCAMPAL PLACE CELLS
45
Table 3–1. Studies on Context-Dependent Place Cell Activity Study Wood et al. (2000); Lee et al. (2006); Ainge et al. (2007a) Frank et al. (2000) Bower et al. (2005)
Ainge et al. (2007a) Lenck-Santini et al. (2001) Holscher et al. (2004) Ferbinteanu and Shapiro (2003) Smith and Mizumori (2006b) Ainge et al. (2007b)
Context-Dependent Hippocampal Activity?
Hippocampus Dependent?
Yes
No
Yes 1. No (complex sequence) 2. Yes (barrier trained) 3. Yes (skipped reward) Very little No Very little Yes Yes Yes
? Probably ? ? Yes ? Probably not Yes Yes 1. No (initial trials) 2. Yes (reversals)
the animals entered the left reward arm) and on rightturn trials (in which they entered the right reward arm). Moreover, the motivational state of the animal as well as its speed, direction of travel, and locations occupied along the stem were the same on both types of trial. Under these circumstances, we found that the majority of CA1 neurons with place fields in the central stem of the T-maze fired differentially depending on whether the rat was performing a left-turn trial or a right-turn trial (Wood et al., 2000). Some cells fired almost exclusively on one trial type but not on the other (Fig. 3–1b), whereas others showed significant changes in firing rate (Fig. 3–1c) or in the location in which they fired on the stem (Fig. 3–1d). These data provide clear evidence that hippocampal neuronal activity in a particular location (in this case the stem of the T-maze) differs, depending on the behavioral context (i.e., whether the rat is traversing the stem during a right-turn or a left-turn trial). Similar context-dependent activity has since been reported in other studies using the continuous alternation T-maze task (Robitsek et al. 2005; Lee et al., 2006; Lipton et al., 2007). This context-dependent activity appears to provide a potential mechanism by which the animal could solve the continuous alternation task. That is, the differential activity could be transmitted to downstream structures enabling the appropriate behavioral response (e.g., if the hippocampal neurons are signaling that it is a left turn trial, then make a left turn at the top of the stem). To test whether context-dependent hippocampal activity is required for alternation behavior on the continuous T-maze alternation task, we assessed the effects of complete hippocampal lesions on performance of the task (Ainge et al., 2007a). A group of 26 rats was trained on the alternation task for 40 trials per day
until they reached criterion performance (90% correct for 8/10 consecutive days). This mirrors the level of performance reached before recordings were carried out in the original recording study (Wood et al., 2000). After reaching criterion, 13 rats received bilateral ibotenic acid lesions of the whole hippocampus, while the remaining 13 rats served as sham-operated, matched-performance controls. After a 2-week recovery period, the rats were tested again on the continuous alternation task for 8 days. There were no differences in the performance of the lesion and control groups during post-surgery testing, and both groups performed at a high level (Fig. 3–1e). These data indicate that performance of the continuous alternation task is not dependent on the hippocampus. Indeed, taken together, this body of new lesion data and the previous recording data (Wood et al., 2000) provide striking evidence for a mismatch between context-dependent hippocampal activity in a particular task and the role of the hippocampus in that task. The observation that context-dependent activity is seen in a task for which the hippocampus is not required is not novel; many other examples of taskrelated hippocampal activity in the absence of a necessary role for the hippocampus have been reported in a variety of studies. Unlike the example above, these typically reflect changes in some aspect of the external environment, such as the specific stimuli that are being presented, or changes in geometric or background contextual features of the environment. For example, in rats and rabbits performing classical eyeblink conditioning tasks, a large fraction of hippocampal neurons have activity that depends on the learned significance of particular stimuli and/or the learned responses (Berger et al., 1983; McEchron and Disterhoft, 1997). This context-dependent activity is seen not only in trace-
46 PLACE CELLS AND SPATIAL CONTEXT
Figure 3–1. Context-dependent activity in CA1 on the continuous spatial-alternation task with no delay. a. Schematic representation of the modified T-maze. Left-turn trials are illustrated in blue and right-turn trials are illustrated in red. b. Example of cell that fired almost exclusively on right-turn trials. Paths taken on left-turn trials (light gray) and rightturn trials (dark gray) are illustrated in the left and right plots, with spikes fired on left-turn trials shown as blue dots and spikes fired in right-turn trials shown as red dots. The central stem is divided into four sectors and the mean firing rate of the cell in each sector for each type of trial is plotted. c. Example of a cell that had a higher firing rate on right- than on left-turn trials. d. Example of a cell that fired in different places on the central stem during left- and right-turn trials. e. Performance (% correct responses) of rats with hippocampal (H) lesions (light gray) and sham lesions (black) on the continuous alternation task during the 8 days of criterion performance. Panels b, c, and d are reprinted from Wood et al. (2000), with permission from Elsevier.
conditioning tasks for which the hippocampus is required but also in delay-conditioning tasks, which do not require the hippocampus (Berger et al., 1983; McEchron and Disterhoft, 1997). Similarly, hippocampal neuron show context-dependent activity as animals perform various recognition memory tasks, such as delayed nonmatching to olfactory stimuli (e.g., Wood et al., 1999), despite no impairment in the performance of these types of recognition memory tasks following hippocampal lesions (Dudchenko et al., 2000). In a later section we will return to the questions
of how and why context-dependent activity might occur in tasks for which the hippocampus is not required.
CONTEXT-DEPENDENT HIPPOCAMPAL ACTIVITY IS NOT ALWAYS SEEN IN HIPPOCAMPUS-DEPENDENT TASKS The data described above show that context-dependent place cell activity can be found in tasks that do not require the hippocampus. But what about tasks that
CONTEXT-DEPENDENT FIRING OF HIPPOCAMPAL PLACE CELLS
require the hippocampus? Clearly, context-dependent activity would be expected in tasks that require the hippocampus and involve discrimination between different behavioral contexts. Although such activity might be expected, Bower et al. (2005) found that hippocampal place cells did not exhibit context-dependent firing on all context-discrimination tasks. They trained rats to run a complex sequence of paths on an open circular platform. The sequence consisted of a series of eight path segments between locations on the periphery of the open field; two segments, which were contiguous, were repeated during the sequence, such that the sequence of paths followed the pattern ABCDEBCF. The whole sequence was repeated several times within a session. Therefore, the animal would need to differentiate between running segments B and C in two different contexts (when preceded by A and followed by D, and also when preceded by E and followed by f) in order to make a choice between running segment D or segment F. Bower et al. (2005) tested the hypothesis that the hippocampus would differentiate between these two contexts by recording unit activity as rats performed this task. However, they found no evidence that hippocampal neurons showed any context-dependent activity during segments B and C, despite good performance by the rats. Thus, context-dependent hippocampal activity was not required for differentiating the overlapping sequences in this task. Is the task dependent on the hippocampus? Bower et al. (2005) argued that it is, as rotation of the sequence with respect to intra- and extramaze cues disrupted performance to the same extent as presentation of a novel sequence. This finding suggests that the animals used a hippocampus-dependent spatial strategy, based on the arrangement of distal and/or proximal cues, as opposed to a purely egocentric or motor strategy (Bower et al., 2005). A second example of this type of mismatch comes from our own studies (Ainge et al., 2007a). We found that although rats with hippocampal lesions are able to perform the continuous T-maze alternation task (as described in the previous section), if the same rats are confined at the base of the stem of the T-maze for a period of 10 s on each trial before being allowed to traverse the stem and enter the left or right arm, rats with hippocampal lesions are significantly impaired (see Fig. 3–2f). Thus, the introduction of a delay changes the task to one that requires the hippocampus. We hypothesized that context-dependent activity, similar to that seen in the no-delay version of the task, would mediate alternation behavior in the delay version of the task. To address this possibility, we trained rats on the continuous alternation task with no delays, and then, after rats were performing well on the task, recorded from CA1 neurons both during the no-delay
47
task and during a session that involved a block of 20 no-delay trials followed by a block of 20 10-s delay trials. Replicating our previous data, 44% of the cells showed context-dependent activity on the stem of the maze during the no-delay task, firing differently on left-turn and right-turn trials (Ainge et al., 2007a). However, our data from a small number of cells recorded during the session in which delays were introduced indicate that 80% of the cells with contextdependent activity during the no-delay block of trials did not show differential activity during the subsequent block of delay trials (see Fig. 3–2). Of these, some cells continued to fire on the stem of the maze during delay trials, but had similar activity on both trial types. Others stopped firing on the stem altogether during the delay trials. Of the cells that had nondifferential place fields on the stem during the nodelay trials, none started to show differential activity during the block of delay trials. Thus, in the delay trials, fewer that 10% of neurons with place fields on the stem showed context-dependent activity on the stem. The most salient aspect of these findings is that when a delay was introduced, the majority of hippocampal neurons no longer discriminated between left-turn and right-turn trials on the stem of the Tmaze. Therefore, in the hippocampal-dependent delay version of the continuous alternation task, very little context-dependent hippocampal activity on the stem of the T-maze was observed. This mismatch mirrors that found in the hippocampus-dependent complexsequence task described above (Bower et al., 2005). Taken together, these data indicate that context-dependent hippocampal activity is not necessary for hippocampus-dependent spatial alternation behavior in these tasks. Although there was little evidence for differential firing during right-turn and left-turn trials in the delay version of our task, the change in activity that resulted from interposing a delay could be considered a form of context-dependent activity. That is, the delay and nodelay trials represent two different contexts, and different patterns of firing were seen in these two contexts. In a similar experiment conducted by Robitsek et al. (2005), CA1 cells were recorded as rats either performed the continuous T-maze alternation task with no delay or were held at the base of the stem for 30 s between trials. They found that many cells with place fields on the maze fired differently in the delay versus no-delay conditions. Furthermore, of those that fired in the delay condition, some fired only when the subsequent arm choice was correct, but not when it was incorrect. Therefore, there is ample evidence of contextdependent hippocampal activity that reflects aspects of the ongoing task and ongoing behavior, but it is not
Figure 3–2. CA1 activity in rats performing the continuous alternation task with and without delays. a. Schematic representation of the continuous alternation task on the modified T-maze with no delay. Left-turn trials are illustrated in blue and right-run trials are illustrated in red. b. Schematic representation of the continuous alternation task on the modified T-maze with 10-s delays between trials. Rats are rewarded for making alternate left (blue) and right (red) turns with a delay of 10 s in the start box between each trial. c. Examples of cells that fire more on right-turn trials than left-turn trials when there is no delay between trials. Paths taken on left-turn trials (dark gray) and right-turn trials (light gray) are illustrated in the left and right plots, with spikes fired on left-turn trials shown as blue dots and spikes fired in right-turn trials shown as red dots. The central stem is divided into six sectors and the firing rate of the cell in each sector for each type of trial is shown. d. The firing patterns of the same cells as in panel c when a 10-s delay is introduced between trials. There is no difference in firing on right- and leftturn trials. e. Performance (% correct) of rats with lesions of the hippocampus (gray) and sham lesions (black) on the continuous alternation task with no delay between trials. f. Performance (% correct) of the same rats on the continuous alternation task with a 10-s delay between trials. Panels e and f depict data previously published in Ainge et al (2007a).
48
CONTEXT-DEPENDENT FIRING OF HIPPOCAMPAL PLACE CELLS
obvious how these types of context-dependent activity directly support alternation behavior in this task.
SOME MEMORY TASKS REQUIRE THE HIPPOCAMPUS AND YIELD CONTEXT-DEPENDENT PLACE CELL ACTIVITY The experiments summarized above provide evidence that context-dependent hippocampal activity is not always associated with a necessary role for the hippocampus in discriminating between behavioral contexts (and vice versa). In this section, we will summarize the studies in which context-dependent hippocampal activity has been observed in tasks that require the integrity of the hippocampus. In the first of these, Ferbinteanu and Shapiro (2003) recorded CA1 activity as rats performed a plus-maze task in which the north and south arms were used as start arms, and the east and west arms as goal arms. On each trial only one goal arm was baited, and the rat was placed in one of the two start arms (chosen according to a pseudorandom schedule). The rat was required to traverse the start arm and enter the baited goal arm to retrieve the food. After retrieving the food, the rat was removed from the maze and held for 10–15 s on a waiting platform before starting the next trial. The same goal arm was baited for a block of trials until the rat reliably visited that arm on 9/10 consecutive trials. At this point, the opposite goal arm was baited for a block of trials. This process was repeated for several blocks of trials (up to 60 trials) within a session. Thus, in order to solve the task, the animals were required to remember the location of the current baited arm in the context of the ongoing block of trials. Hippocampal activity on each arm was analyzed as a function of which journey the animal was making. For example, activity on the north arm was compared between journeys from north to east and from north to west. Similarly, activity on the east arm was compared between journeys starting in the north arm and those starting in the south. The activity of the majority of recorded neurons was journey dependent, and there was evidence for both prospective coding (cells that fired in a start arm, with activity dependent on the subsequent goal arm choice) and retrospective coding (cells that fired in a goal arm, with activity dependent on which start arm the animal had come from). Ferbinteanu et al. (2006) have suggested that the prospective activity could ‘‘guide behavior by anticipating pending events,’’ and that ‘‘retrospective signals [could] inform behavior by retrieving past events.’’ To determine if performance of this spatial serial reversal task required the hippocampus, rats were
49
trained on the task, and pairs of performance-matched animals received fornix lesions or sham surgery. While the sham group maintained their good performance during post-surgery testing, the fornix lesion group’s performance was significantly impaired, and not significantly better than chance. Although the hippocampus was not damaged directly, the fornix lesion would have disrupted normal hippocampal function, leading to the conclusion that accurate performance requires a fully functional hippocampal system. Smith and Mizumori (2006b) have recently obtained similar results. Like Ferbinteanu and Shapiro (2003) they used a plus-maze task, but tested the rats with two consecutive blocks of 15 trials each day. In the first block of trials the west arm was rewarded, and in the second block the east arm was rewarded. Another difference was that all three of the unbaited arms in a block of trials were used as start arms. In addition to journey-dependent prospective and retrospective coding in CA1, similar to that found by Ferbinteanu and Shapiro (2003), Smith and Mizumori (2006b) observed cells whose activity depended on the reward location (i.e., fired differently on the first and second blocks), but fired similarly on a given arm within that block, regardless of which arm the animal had come from or was going to. Thus, different levels of contextdependent activity were observed in this task. The context-dependent activity related to reward location could be used in discriminating between the blocks of trials, which, in this task, would be sufficient for mediating accurate performance. To determine whether acquisition of this task was dependent on the hippocampus, the dorsal hippocampus was temporarily inactivated (by infusion of muscimol) prior to the first two training sessions (Smith and Mizumori, 2006a). Animals were impaired by this inactivation, with many adopting inflexible behavioral strategies (e.g., always turn left). Thus, a functional hippocampus is required for task acquisition, consistent with a role for the context-dependent hippocampal activity in this task. Finally, we have recently found context-dependent hippocampal activity in a spatial, serial reversal task conducted on a concatenated Y-maze (see Fig. 3–3a; Ainge et al., 2007b). At the end of each of the four arms of the maze was a small enclosure (the reward box) containing either an empty or a full food dish. On any given trial, only two of the four reward boxes contained food. After choosing a reward box, rat was replaced in the start box area for approximately 10s while the maze was cleaned and then the next trial would start. The reward locations were kept constant for blocks of 10 trials. After each block of 10 trials the reward locations were changed to another two of the four possible reward boxes (Fig. 3–3a). Rats were
50 PLACE CELLS AND SPATIAL CONTEXT initially given two blocks of 10 trials each day, but as performance improved the rats were given more blocks with reward location changing on every block. On average, the rats completed 40 trials (four blocks) each day. Again, to perform the task correctly, rats had to remember the spatial locations of the rewards in the context of the ongoing block of trials. Hippocampal activity in the start box, maze arms, and choice boxes was analyzed in terms of the destination of the journey. Firing rates in each section of the maze were calculated for each trial, and for cells with place fields in the start box, choice boxes, or maze arms prior to the final choice point, the mean firing rates for the journeys to the four different goal boxes were compared. Consistent with the studies described in this section, 45% of CA1 place cells with place fields in the start box or in the initial maze arm (before the first choice point) differentiated between journeys to different goal boxes. For example, the cell illustrated in Figure 3–3b fired robustly in the initial maze arm when the rat was going to box 2, but much less on journeys to any of the other three boxes. There were also other cells that had place fields between the first and second choice points. When the firing rates of these cells were analyzed, 43% showed contextdependent firing. The example shown in Figure 3–3c depicts a cell that fired specifically when the rat went to box 2 but not to box 1. This pattern of activity is very similar to the prospective coding seen by Ferbinteanu and Shapiro (2003) and is a potential mechanism that could guide behavior on the concatenated Y-maze by anticipating pending events. To assess whether performance of the concatenated Y-maze task was hippocampus dependent, a separate group of rats was trained on the task. Once the rats performed the task readily they were given either hippocampal lesions or sham lesions. The results revealed that rats with lesions of the hippocampus performed the first block of trials each day as accurately as animals with an intact hippocampus. However, when the location of the reward was moved to different reward boxes, animals with hippocampal damage were significantly impaired relative to control animals. Thus, this is another example of a task in which contextdependent place cell activity is found, and which, at least in the reversal phase, requires the hippocampus.
WHAT IS THE RELATIONSHIP BETWEEN CONTEXT-DEPENDENT PLACE CELL ACTIVITY AND BEHAVIOR? Context-dependent place cell activity may be one way in which the mammalian brain distinguishes between different events or intended trajectories occurring in the
same location. The significance of this activity can in part be determined by assessing its relationship to behavior. Thus, if context-dependent activity within the hippocampus is essential for distinguishing contexts and guiding behavior, we would expect the following statements to hold: 1. Removing context-dependent place cell activity (via hippocampectomy) should produce behavioral impairments in the tasks in which contextdependent hippocampal activity is observed. 2. If a task requires context differentiation, then context-dependent activity should be observed in the hippocampus. In addressing the first statement, the data that we have reviewed above indicate that removal of the hippocampus does not produce behavioral impairments in all tasks in which context-dependent activity is observed. Thus, the context-dependent encoding evident in hippocampal place cells is not necessary for all behavioral tasks in which it is observed. However, this does not preclude the possibility, which we will consider below, that such activity is essential for some subset of the tasks in which it is found. But first, what are the characteristics of behavioral tasks in which context-dependent hippocampal activity is observed but is not necessary? In the continuous T-maze alternation task that we have employed (Wood et al., 2000; Ainge et al., 2007a), after receiving a reward on the goal arm, the animals return to the start arm via a connecting return arm. This allows them to run a continuous path from one reward site to the next, which in the no-delay task is uninterrupted. Therefore, one possibility is that the animals could learn and execute a stereotyped motor program, rather than having to maintain in memory some representation of the particular arm it visited on the previous trial (Holscher et al., 2004). According to this scenario, the animals would simply initiate one of two different motor sequences, one going from the left water port to the right water port (involving a left turn at the base of the stem of the T-maze, followed by a right turn at the top of the stem), and another from the right water port to the left water port (involving a right turn followed by a left turn). This type of motor learning would likely be mediated by the basal ganglia (striatum) rather than the hippocampus (McDonald and White, 1994; Packard and Knowlton, 2002). In contrast, the more traditional hippocampus-dependent T-maze alternation tasks cannot be a solved using this type of motor strategy for two reasons. First, in most forcedchoice or spontaneous-alternation T-maze tasks, animals are removed from the goal arm after making a choice (or after a sample phase) and placed on the base of the stem to initiate the next choice trial (Olton et al.,
CONTEXT-DEPENDENT FIRING OF HIPPOCAMPAL PLACE CELLS
51
Figure 3–3. Prospective firing on a concatenated Y-maze. a. Schematic representation of the concatenated Ymaze. Two of the four reward boxes contain reward on each trial. The location of the rewards are changed every 10 trials. b. Example of a cell that fires in the initial maze arm (gray shaded area), but predominantly on trials in which the rat goes to box 3. The left panel shows the paths of all of the trials completed in a session with the spikes fired by a CA1 neuron, shown as red dots. In the middle panel the paths to the different boxes are shown separately. It is clear that this cell fires on the initial arm predominantly when the rat goes to box 3. The right panel shows the mean firing rate of the cell in the initial arm (depicted by the gray shaded area in the left and middle panels) during journeys to each of the four boxes. c. Example of a cell that fires on the maze arm between the two choice points on the left half of the maze (gray shaded area in left and middle panels). This cell has a higher firing rate on trials when the rat goes to box 2. Panel b depicts data previously published in Ainge et al (2007b).
1979). This would interrupt any ongoing learned motor program. Second, the introduction of a delay (both in the traditional tasks and in the delay version of the continuous T-maze task described earlier) would also interrupt the motor sequence. Thus, while the continuous no-delay task may be solved using a striatum-
dependent motor learning strategy, other versions of the task are more likely to depend on hippocampusdependent spatial memory strategies. Could context-dependent hippocampal activity be essential for a subset of the tasks in which it is observed? From an intuitive perspective, one might ex-
52 PLACE CELLS AND SPATIAL CONTEXT pect that the robust differentiation between contexts exhibited by place cells (e.g., Fig. 3–3) contributes to the animal’s ability to separate overlapping trajectories. To see if this view has merit, we may look for common features of tasks that require the hippocampus and in which context-dependent place cell activity is observed. The clearest examples of this are the bottom three studies in Table 3–1: Ferbinteanu and Shapiro (2003), Smith and Mizumori (2006b), and Ainge et al. (2007b). In each of these studies, serial reversal tasks were used. In both the Ferbinteanu and Shapiro (2003) and Smith and Mizumori (2006b) studies, rats were reinforced on a plus maze for first going to one location for a series of trials and then going to an alternative location for a different series of trials. In the Ainge et al. (2007b) concatenated Y-maze study, rats were reinforced for repeatedly visiting one subset of goal boxes and then for switching from these to alternative goal boxes. Thus, in all three studies, rats were repeatedly reinforced for going to one location and then (in the reversal) required to inhibit this response and choose an alternative. A hippocampal representation that distinguishes these alternatives may therefore be essential for these tasks, by virtue of the fact that the animals are required to inhibit a previously rewarded response. Additional tasks that may require the hippocampus and in which context-dependent activity is seen are the barrier-trained and skipped-reward tasks of Bower et al. (2005). After training their animals on the complex-sequence task in which context-dependent activity was not seen (described earlier), the rats were trained on two further tasks in the same open-field environment. In the skipped-reward task, the rats ran a figure-eight sequence with one overlapping segment (analogous to the T-maze), but, unlike the complex sequence task, they were not rewarded at the entrance and exit of the overlapping path segment. In the barriertrained version of the task, barriers were placed on the platform to help guide the rats during their acquisition of the task. The barriers were removed in the final day on the task. In both these versions of the task a significant proportion of cells showed context-dependent activity on the overlapping segment. On the surface, there is little to distinguish the skipped-reward version of the Bower et al. (2005) task from the no-delay task of Wood et al. (2000) and Ainge et al. (2007a). Thus, it is puzzling that the skippedreward task appears to require the hippocampus whereas the no-delay T-maze task does not. One possibility is that the use of an open-field environment by Bower et al. (2005) emphasizes the use of distal spatial landmarks (and thus, presumably, a hippocampal representation), whereas navigation in the structured
T-maze environment used by Wood et al. (2000) may not require the use of these types of cues. However, in both cases it is possible that the differentiation between the two behavioral contexts occurs outside the hippocampus. In addressing the second statement above—that tasks requiring context differentiation should also be tasks in which context-dependent hippocampal activity is observed—we likewise find difficulties. Two possible conclusions can be offered. First, simply, this proposition may not be true. The evidence from the Ainge et al. (2007a) delay task and the Bower et al. (2005) complex sequence task suggests that in some tasks for which the hippocampus is required, contextdependent hippocampal activity is not observed. However, a second alternative is that context-dependent activity may be present in these tasks, but only at the starting points of the journeys. The first conclusion appears the most likely. To the extent that a given task requires that an animal differentiate between contexts, if context-dependent activity does not occur in the hippocampus it must occur elsewhere. This parallels the findings of Jeffery et al. (2003) with place cell remapping: performance of a spatial task was unaffected by a change in the environment that produced place field remapping, suggesting that the information necessary to perform the task was located outside the hippocampus. We have described two hippocampus-dependent tasks that require differentiation between behavioral contexts but in which context-dependent hippocampal activity was not observed (Bower et al., 2005; Ainge et al., 2007a). A possible mechanistic explanation for context-dependent activity not being seen in these tasks is that common aspects of the two contexts dominate in driving hippocampal neuronal activity. But why would this occur in these tasks, and not in the no-delay version of the T-maze task or the skippedreward version of the Bower et al. (2005) task? As described earlier, in both the delayed T-maze task (Ainge et al., 2007a) and the complex sequence task (Bower et al., 2005), the animals’ trajectory was interrupted at the beginning of the overlapping segment— for a 10-s delay on the T-maze, or for a short period during which the animals received rewarding intracranial stimulation on the complex sequence task. This interruption could serve to accentuate the similarities between the two trial types, rather than the differences between them (Bower et al., 2005). Specifically, the trajectory of the rat along the stem of the T-maze, or the repeated segment of the complex sequence task, would be initiated in a common location on both trial types (at the base of the stem of the T-maze, or the entrance to the repeated segment), and in the complex
CONTEXT-DEPENDENT FIRING OF HIPPOCAMPAL PLACE CELLS
sequence it would also end in a common location. In contrast, in the no-delay T-maze and in the skippedreward task, the trajectory begins in different locations for the two different trial types, which may emphasize the differences between them. Bower et al. (2005) have suggested that the lack of hippocampal contextdependent activity on the complex sequence task could result from common inputs in the two contexts that mask or override context-dependent activity transmitted from other structures. The same could be true of the delay task on the T-maze. Although compelling, this explanation cannot fully explain all the instances of context-dependent firing (or lack thereof). For example, Frank et al. (2000) trained rats on a W-shaped maze to alternate responses to the left and right arms. The rats were rewarded at the end of the left and right arms but also at the end of the central arm, which they were required to visit between each outside arm visit. The authors showed that on the inbound journeys there was differential firing in CA1 on the central arm, depending on which outer arm the rat had come from (retrospective coding). This finding fits with the suggestion by Bower et al. (2005) that the difference in start location emphasizes the difference between the journeys, and differential activity would be expected. However, Frank et al. (2000) showed that on outbound journeys along the central arm, when the rat starts in the same place at the base of the central arm for right- and leftturn journeys, cells in CA1 also show differential firing on the central stem, depending on which arm the rat is about to visit (prospective coding). This part of the task is more like the complex sequence task or delay version of the T-maze, as the rat has stopped before it traverses the central arm on outbound journeys, yet context-dependent firing is still seen. In a different study, Lenck-Santini et al. (2001) recorded from the CA1 of rats solving a spatial alternation task on a Y-maze. This task resembles that of Frank et al. (2000), as rats alternated visits to the left and right arms with visits to the center arm. Only the center (goal) arm was rewarded. In this task, no contextdependent activity was seen on the center arm on either outbound or inbound journeys. On outbound journeys, this result could be due to each journey having a common start point. However, on inbound journeys, the trajectory started in different locations, yet no context-dependent activity was seen. One would have to speculate that, despite different start points on the inbound journeys, other features of the task result in the similarities between the two trial types driving CA1 activity. It is worth pointing out that it is not known whether either the W-maze task or the Y-maze task is hippocampus dependent.
53
An alternative and more speculative account of the data is that context-dependent hippocampal activity does occur during hippocampus-dependent tasks that involve context discrimination, but it has not been observed in some studies because it is very restricted in either time or space. Neither Ainge et al. (2007a) nor Bower et al. (2005) found evidence of contextdependent firing in the hippocampus under conditions in which the rat was interrupted at the start of the common segment of two journeys. However, for the rat to perform the task correctly, it must have retained information about the current behavioral context to make the correct choice at the end of the common segment. One possibility, therefore, is that contextdependent hippocampal activity may occur, but only in the area in which the rat is delayed and has to remember which type of trial it is currently running. To address this possibility in our own studies, we analyzed the activity of cells with place fields in the start box area (below the central stem) in a different group of rats performing the T-maze task with delays between each trial. This analysis revealed that a significant proportion of cells with place fields in the start box differentiated between left- and right-turn trials (Ainge et al., 2007a). Thus, despite no evidence of context-dependent activity as rats traversed the central stem in the delay task, context-dependent activity was observed in a more spatially restricted fashion, during the 10 s that rats spent in the start box during the intertrial interval. Could unobserved context-dependent activity also have occurred during the complex sequence task of Bower et al. (2005)? Although it is technically possible that such activity might have occurred following the intracranial stimulation at the end of each segment, such activity would have to have been restricted to the 1-sec artifact-rejection period after the stimulation. We feel that this is a somewhat unlikely possibility. Therefore, we must conclude that, at least for some hippocampus-dependent tasks that involve context discrimination, context-dependent hippocampal activity does not occur. We began this section by considering two implications of the assumption that context-dependent place cell activity in the hippocampus is necessary for behavior. Neither of these implications has proven unassailable. We must conclude, therefore, that contextspecific activity in the hippocampus is not necessary for all tasks in which it is observed. Second, context differentiation must also be supported by structures other than, or in addition to, the hippocampus. But these conclusions raise other issues: If context-specific activity in the hippocampus isn’t necessary for a given behavior, why is it there? Where outside the hippocampus might such context differentiation occur? Does such
54 PLACE CELLS AND SPATIAL CONTEXT extrahippocampal context encoding support behavior? It is to these questions we turn in the next section.
WHAT IS THE NATURE OF CONTEXTDEPENDENT HIPPOCAMPAL PLACE CELL ACTIVITY? Context-dependent hippocampal activity may arise in two ways. First, in both hippocampus-dependent and hippocampus-independent tasks, it may be generated in the intrahippocampal network. Alternatively, it may be that hippocampal place cells simply reflect contextdependent information passed to them from other structures. We will discuss each of these possibilities in turn, and also address a third scenario: in some circumstances (for example, tasks for which the hippocampus is required) context-dependent activity is generated in the hippocampal circuitry, whereas in others (for example, in tasks for which the hippocampus is not required) hippocampal context-dependent activity reflects context-dependent activity in upstream structures. Perhaps the most intuitive possibility is that contextdependent activity is inherent to the hippocampus. If this is so, why is context-dependent activity observed in tasks for which the hippocampus is not required? One explanation is that the hippocampus automatically encodes differences in the behavioral context, regardless of whether it is required for ongoing task performance. This suggestion is consistent with the perspective of Morris and Frey (1997), who have proposed that the hippocampus automatically encodes attended experiences, and that this hippocampal representation can subsequently be used to support episodic memory for those experiences. A consequence of the automatic nature of hippocampal encoding would be that the hippocampus encodes ongoing events during tasks for which it is not required. Therefore, the fact that the hippocampus differentiates between behavioral contexts in a particular situation would not necessarily imply that the context-dependent hippocampal representation is required in that situation. However, if context differentiation only occurred in the hippocampus, then removing the hippocampus should produce impairments in tasks requiring this ability. As we have argued above, this doesn’t appear to be the case. As an alternative, context-dependent hippocampal place cell activity may reflect context encoding generated in input structures to the hippocampus. One candidate structure is the striatum, particularly as performance in some of the tasks described above may involve the execution of learned motor programs that could be mediated by the striatum. Striatal neurons in rats have been shown to have location-related activity that is influenced by the external (spatial) context. For
example, the place fields of striatal neurons reorganize following manipulation of the external cues as rats are performing both place and response tasks on a plus maze (Yeshenko et al., 2004). Similar reorganization of striatal place fields was seen when rats performed a radial-arm maze task in the light and in the dark (Gill and Mizumori, 2006). Thus, striatal neurons, like hippocampal place cells, are influenced by the external context. But what about the behavioral context? Two sets of studies speak to this issue. In the first, Jog et al. (1999) and Barnes et al. (2005) showed that as rats learned a conditional T-maze task in which different tones predict whether the right or left arm will be rewarded, striatal neurons change their firing properties during the course of learning. Specifically, early in training, different neurons fired in all parts of the maze. However, as rats learned the task, more neurons had activity near the start or the goal areas of the maze than in other areas. As training progressed the activity was further concentrated to the start area, but specifically during the period before locomotion commenced, and to the turning behavior. Although very different from that described in hippocampus, this change in striatal activity is dependent on the behavioral context in that it reflects experience with and learning of the behavioral task within a given environment. In the second kind of study, SchmitzerTorbert and Redish (2004) examined the activity of dorsal striatal neurons as rats performed a sequential navigation task on a series of multiple T-mazes joined end-to-end. On each daily session, the animal had to follow a different sequence of turns to negotiate the T-mazes and reach the goal. The task shared some common features with the concatenated Y-maze serial reversal task of Ainge et al. (2007b) in that in each block of trials, a different path had to be taken to reach the goal. In this sequential navigation task, a subset of striatal neurons showed clear location-related firing. Over multiple sessions, this location-related firing was most similar when the rats ran the same sequence of turns through a specific location, but tended to be less similar if a different sequence of turns was made (Schmitzer-Torbert and Redish, 2004). Thus, the behavioral context (the preceding or following sequence of turns) influenced the activity in a specific location. This context-dependent striatal activity shares features with that seen in the hippocampus during many of the tasks described earlier. To our knowledge, there is no direct evidence that activity in the striatum directly mediates or contributes to context-dependent activity of cells in CA1. However, a recent study demonstrated that in rats running a continuous alternation task on a figure-eight-shaped maze, temporary inactivation of the dorsal striatum (by injections of sulpiride) caused a shift in the power
CONTEXT-DEPENDENT FIRING OF HIPPOCAMPAL PLACE CELLS
spectrum of EEG recorded in CA1 (Gengler et al., 2005). It is not clear whether this manipulation influenced the performance of rats on the alternation task, although the authors report that similar sulpiride injections disrupted the performance of a different striatum-dependent motor learning task. However, the disruption of hippocampal EEG by this manipulation indicates that striatal activity is able to influence hippocampal activity. As there are no direct projections from the striatum of the hippocampus, it was suggested that this interaction may occur through indirect pathways via the globus pallidus and temporal cortical structures (Gengler et al., 2005). A second candidate structure is the prefrontal cortex (PFC), which is generally accepted to have a role in short-term, working memory. Many studies have shown that disruption of PFC function causes deficits in tasks that require short-term memory for spatial information (Seamans et al., 1995; Kesner et al., 1996; Floresco et al., 1997; Jones, 2002). All of the tasks reviewed in this chapter involve the use of short-term spatial memory strategies. The PFC projects to the entorhinal cortex (EC), which could influence CA1 either directly or via the dentate gyrus–CA3 intrahippocampal network (Amaral and Witter, 2004). Consequently, one possibility is that the context-dependent firing in CA1 is a result of context-dependent activity generated in and projected from the PFC. There is accumulating evidence for contextdependent activity in the PFC in tasks similar to those in which context-dependent activity is seen in the hippocampus. For example, Jung et al. (1998) recorded from medial prefrontal cortex (mPFC) as rats performed a delayed spatial-alternation task on a figure-eight maze, similar to the delay version of the continuous T-maze task of Ainge et al. (2007a). One difference between the tasks is that the animals were delayed in the middle of the central stem of the figure-eight maze, as opposed to the base of the stem of the T-maze. Some mPFC neurons showed differential firing during the delay period that depended on the previously visited goal arm, analogous to the context-dependent start box cells found in the hippocampus. Another group of neurons showed context-dependent activity on the central stem, either before the animal reached the delay area or after the delay at the choice area at the top of the stem. Thus, the context-dependent activity observed in mPFC during the delayed spatial-alternation tasks shows features of that seen in CA1 during the continuous T-maze spatial-alternation task. In a subsequent study, Baeg et al. (2003) showed that neuronal ensemble activity in prelimbic and infralimbic regions of mPFC during the delay period in the same figure-eight task reflected increased context-dependent activity as training on the task progressed, parallel to the improvement in accu-
55
rate performance by the rats. Additional evidence for context-dependent activity in the mPFC comes from a study by Jones and Wilson (2005), who recorded from CA1 and deep layers of prelimbic and infralimbic mPFC in rats running an egocentric spatial workingmemory task on a double-ended T-maze. Each trial was divided into two phases. In the sample phase the rat was forced to turn either left or right at one end of the maze. In the choice phase the rat was given a free choice of which way to turn at the other end of the maze and was rewarded for making a turn in the direction opposite that made in the sample phase. Using this procedure, it was possible to compare trials on the central stem in which the rat had to use working memory information with trials in which it did not (sample versus choice trials), as well as turn-dependent activity. The authors showed that while cells in CA1 were more spatially selective than those in mPFC, both populations included cells that responded differentially on the central stem of the maze, depending on the type of journey the rat was making. Specifically, in the choice phase of each trial, cells in both areas showed context-dependent activity, differentiating between left- and right-turn trials. These studies provide evidence that neurons in the mPFC, and perhaps the striatum, have contextdependent activity. However, they do not speak to the issue of whether (1) context-dependent activity is generated in these structures and fed forward to the hippocampus, (2) context-dependent activity is generated independently in hippocampus and other structures, or (3) context-dependent activity is generated in the intrahippocampal circuitry and projected to striatum and/or mPFC. Consistent with the possibility that context-dependent activity in CA1 is a result of context-dependent afferent input, Lipton et al. (2007) have recently recorded from grid cells in dorsomedial EC in rats performing the same continuous T-maze alternation task as that used by Wood et al. (2000) and others, described in this chapter. They showed that the EC cells show context-specific firing as the rat traverses the central stem of T, similar to that seen in CA1. In fact, the proportion of cells in this study showing differential activity on the two types of trial was greater in EC (56%) than in CA1 (33%). Since these cells were recorded from the layers of EC that project to the hippocampus, it is possible that, at least in this task, the context-dependent firing is not generated within the hippocampus but rather in structures further upstream. This activity could be generated within the EC. An alternative possible source of this information is the mPFC, which has a direct projection to EC. However, it is important to acknowledge that the converse of this could also be true. As CA1 projects back to the PFC, it is possible that the context-dependent activity in PFC is a result of its inputs from CA1. Indeed, the
56 PLACE CELLS AND SPATIAL CONTEXT mPFC recordings in the studies of Jones and Wilson (2005) and Jung et al. (1998) described above were taken from layers IV and V of prelimbic and infralimbic cortex, which are targets of hippocampal afferents. Moreover, context-dependent activity has also been reported in deep layers of EC (Frank et al., 2000), which also receive projections from CA1. To determine whether context-dependent activity is generated in the hippocampus and projected to other structures or whether it is generated in EC, PFC (or other structures) and projected to the hippocampus, a systematic investigation of the effect of disrupting activity in CA1, EC, PFC, striatum, and other candidate structures is required, as well as an examination of the effects on behavior and neuronal activity in the tasks we have reviewed. While it is too early to suggest that no context-dependent firing is generated in the hippocampus, the fact that context-dependent hippocampal activity is seen in some tasks that can be performed by rats with hippocampal lesions suggests that context-specific activity must be generated in another area of the brain under some conditions and that this is sufficient to solve these tasks in lesioned animals. In other conditions, given that the nature of the context-dependent hippocampal activity tends to differ between hippocampus-dependent tasks and hippocampus-independent tasks, it is possible that the intrahippocampal circuitry contributes to the generation of some forms of context-dependent activity. The recent findings of Leutgeb et al. (2005) may be consistent with the view that, at least under some conditions, context-dependent activity is generated within the intrahippocampal circuitry. They demonstrated a potential mechanism within CA3 that could be used to encode contextual as well as spatial information. When rats were exposed to different enclosures in the same room, place cells in CA3 fired in the same spatial location in each enclosure but the rate of firing changed dramatically, often by more than an order of magnitude. This rate remapping may be the same as the context-dependent activity in the studies reviewed here. That is, place cells fire much more in some conditions than in others and thereby encode both contextual and spatial information. We started this section by asking how contextdependent activity within the hippocampus is generated. Studies in PFC and striatum provide potential sites in the brain that could generate context-dependent firing and pass this information forward to the hippocampus. However, the discovery of rate remapping in CA3 provides a potential intrahippocampal mechanism that could generate context-dependent firing in CA1. One possible way of reconciling these two sets of findings would be if CA1 (and consequently the output from the hippocampus) is influenced by different input
pathways in a dynamic fashion depending on the memory load needed to solve a particular task (Guzowski et al., 2004). Under conditions in which there is low memory load, CA1 may be influenced predominantly by its EC inputs. This would include tasks such as the no-delay continuous T-maze task, for which a striatumdependent motor strategy may be used to solve the task. The context-dependent activity would be generated in the striatum and passed onto the hippocampus; consequently, lesions of the hippocampus would not eliminate the mechanism used to solve the task. As memory load increases, the influence of the CA3 network would increase. For example, the serial reversal tasks require representations of both the current behavioral context (block of trials and current spatial location of the reward) and the correct response needed for that particular context. These functions may be carried out by rate remapping in CA3, thus disruption of activity within the hippocampus would disrupt the mechanism being used by the rat to solve the task. This would account in part for the mismatches described in this chapter between context-dependent firing in the hippocampus and the need for normal hippocampal functioning for correct task performance.
CONCLUSIONS The discovery that hippocampal place cell activity is influenced by aspects of the behavioral context, including recent behavioral history, prospective events, and the location of goals, has provided a potential mechanism by which memory tasks that involve disambiguation of different events or sequences of behavior with common elements may be solved. Does context-dependent activity in the hippocampus underlie context discrimination in these types of memory tasks? The data we have reviewed above suggest that although this could be the case for some tasks (such as the serial reversal tasks), it is clearly not always the case. Therefore, at the very least, structures outside the hippocampus are capable of mediating context discrimination sufficient to support some kinds of context-dependent behaviors. This leads us to suggest that the hippocampus is one component of a network of structures that together allow context discrimination and mediate the performance of context-dependent memory tasks.
acknowledgments We would like to thank Steven Huang for assistance collecting data in the delay version of the continuous T-maze task, and Jane Tulloch for assisting with histology.
CONTEXT-DEPENDENT FIRING OF HIPPOCAMPAL PLACE CELLS
References Ainge JA, van der Meer MAA, Langston RF, Wood ER (2007a) Exploring the role of context-dependent hippocampal activity in spatial alternation behavior. Hippocampus. Published Online: 6 Jun 2007 DOI: 10.1002/hipo.20301 Ainge JA, Woergotter F, Tamosiunaite M, Dudchenko PA (2007b) Hippocampal CA1 place cells encode intended destination on a maze with multiple choice points. Journal of Neuroscience (in press) Amaral DG, Witter MP (2004) Hippocampal formation. In: The Rat Nervous System, 3rd edition (Paxinos G, ed.), pp 637–704. New York: Elsevier. Baeg EH, Kim YB, Huh K, Mook-Jung I, Kim HT, Jung MW (2003) Dynamics of population code for working memory in the prefrontal cortex. Neuron 40:177–188. Barnes TD, Kubota Y, Hu D, Jin DZ, Graybiel AM (2005) Activity of striatal neurons reflects dynamic encoding and recoding of procedural memories. Nature 437: 1158–1161. Berger TW, Rinaldi PC, Weisz DJ, Thompson RF (1983) Single-unit analysis of different hippocampal cell types during classical conditioning of rabbit nictitating membrane response. J Neurophysiol 50:1197–1219. Bower MR, Euston DR, McNaughton BL (2005) Sequential-context-dependent hippocampal activity is not necessary to learn sequences with repeated elements. J Neurosci 25:1313–1323. Dudchenko PA, Wood ER, Eichenbaum H (2000) Neurotoxic hippocampal lesions have no effect on odor span and little effect on odor recognition memory but produce significant impairments on spatial span, recognition, and alternation. J Neurosci 20:2964–2977. Fenton AA, Muller RU (1998) Place cell discharge is extremely variable during individual passes of the rat through the firing field. Proc Natl Acad Sci USA 95(6): 3182–3187. Ferbinteanu J, Kennedy PJ, Shapiro ML (2006) Episodic memory—from brain to mind. Hippocampus 16:691– 703. Ferbinteanu J, Shapiro ML (2003) Prospective and retrospective memory coding in the hippocampus. Neuron 40:1227–1239. Floresco SB, Seamans JK, Phillips AG (1997) Selective roles for hippocampal, prefrontal cortical, and ventral striatal circuits in radial-arm maze tasks with or without a delay. J Neurosci 17:1880–1890. Frank LM, Brown EN, Wilson MA (2000) Trajectory encoding in the hippocampus and entorhinal cortex. Neuron 27:169–178. Gengler S, Mallot HA, Holscher C (2005) Inactivation of the rat dorsal striatum impairs performance in spatial tasks and alters hippocampal theta in the freely moving rat. Behav Brain Res 164:73–82.
57
Gill KM, Mizumori SJ (2006) Context-dependent modulation by D(1) receptors: differential effects in hippocampus and striatum. Behav Neurosci 120:377– 392. Guzowski JF, Knierim JJ, Moser EI (2004) Ensemble dynamics of hippocampal regions CA3 and CA1. Neuron 44:581–584. Holscher C, Jacob W, Mallot HA (2004) Learned association of allocentric and egocentric information in the hippocampus. Exp Brain Res 158:233–240. Jeffery KJ, Gilbert A, Burton S, Strudwick A (2003) Preserved performance in a hippocampal-dependent spatial task despite complete place cell remapping. Hippocampus 13:175–189. Jog MS, Kubota Y, Connolly CI, Hillegaart V, Graybiel AM (1999) Building neural representations of habits. Science 286:1745–1749. Jones MW (2002) A comparative review of rodent prefrontal cortex and working memory. Curr Mol Med 2: 639–647. Jones MW, Wilson MA (2005) Theta rhythms coordinate hippocampal–prefrontal interactions in a spatial memory task. PLoS Biol 3:e402. Jung MW, Qin Y, McNaughton BL, Barnes CA (1998) Firing characteristics of deep layer neurons in prefrontal cortex in rats performing spatial working memory tasks. Cereb Cortex 8:437–450. Kesner RP, Hunt ME, Williams JM, Long JM (1996) Prefrontal cortex and working memory for spatial response, spatial location, and visual object information in the rat. Cereb Cortex 6:311–318. Lee I, Griffin AL, Zilli EA, Eichenbaum H, Hasselmo ME (2006) Gradual translocation of spatial correlates of neuronal firing in the hippocampus toward prospective reward locations. Neuron 51:639–650. Lenck-Santini PP, Save E, Poucet B (2001) Place-cell firing does not depend on the direction of turn in a Ymaze alternation task. Eur J Neurosci 13: 1055–1058. Leutgeb S, Leutgeb JK, Barnes CA, Moser EI, McNaughton BL, Moser MB (2005) Independent codes for spatial and episodic memory in hippocampal neuronal ensembles. Science 309:619–623. Lipton PA, White JA, Eichenbaum H (2007) Disambiguation of overlapping experiences by neurons in the medial entorhinal. J Neurosci 27:5787–5795. McDonald RJ, White NM (1994) Parallel information processing in the water maze: evidence for independent memory systems involving dorsal striatum and hippocampus. Behav Neural Biol 61:260–270. McEchron MD, Disterhoft JF (1997) Sequence of single neuron changes in CA1 hippocampus of rabbits during acquisition of trace eyeblink conditioned responses. J Neurophysiol 78:1030–1044. Morris RG, Frey U (1997) Hippocampal synaptic plasticity: role in spatial learning or the automatic recording
58 PLACE CELLS AND SPATIAL CONTEXT of attended experience? Philos Trans R Soc Lond B Biol Sci 352:1489–1503. Muller RU (1996) A quarter of a century of place cells. Neuron 17:979–990. O’Keefe J (1979) A review of hippocampal place cells. Prog Neurobiol 13:419–439. Olton DS, Becker JT, Handelmann GE (1979) Hippocampus, space and memory. Behav Brain Sci 2:313– 365. Packard MG, Knowlton BJ (2002) Learning and memory functions of the basal ganglia. Annu Rev Neurosci 25: 563–593. Robitsek RJ, Fortin N, Eichenbaum H (2005) Hippocampal unit activity during continuous and delayed T-maze spatial alternation. Program No. 776.777. 2005 Abstract Viewer/Itinerary Planner. Washington, DC: Society for Neuroscience. Online:http://www.sfn .org/index.cfm?pagename¼abstracts_ampublications §ion¼publications. Schmitzer-Torbert N, Redish AD (2004) Neuronal activity in the rodent dorsal striatum in sequential navigation: separation of spatial and reward responses on the multiple T task. J Neurophysiol 91:2259– 2272.
Seamans JK, Floresco SB, Phillips AG (1995) Functional differences between the prelimbic and anterior cingulate regions of the rat prefrontal cortex. Behav Neurosci 109:1063–1073. Smith DM, Mizumori SJ (2006a) Hippocampal place cells, context, and episodic memory. Hippocampus 16: 716–729. Smith DM, Mizumori SJ (2006b) Learning-related development of context-specific neuronal responses to places and events: the hippocampal role in context processing. J Neurosci 26:3154–3163. Wood ER, Dudchenko PA, Eichenbaum H (1999) The global record of memory in hippocampal neuronal activity. Nature 397:613–616. Wood ER, Dudchenko PA, Robitsek RJ, Eichenbaum H (2000) Hippocampal neurons encode information about different types of memory episodes occurring in the same location. Neuron 27:623–633. Yeshenko O, Guazzelli A, Mizumori SJ (2004) Contextdependent reorganization of spatial and movement representations by simultaneously recorded hippocampal and striatal neurons during performance of allocentric and egocentric tasks. Behav Neurosci 118: 751–769.
4 The Place Cells—Cognitive Map or Memory System? KATHRYN J. JEFFERY
Their suggestion provided a great impetus to both the cognitive map hypothesis and cognitivism more generally. For the first time, it seemed possible to look inside the head of a rat and see what it was thinking (or, at least, where it was thinking of and to see evidence of processes that were not always immediately obvious by observing behavior alone. Indeed, it began to seem feasible that one might reconstruct what another sentient being was thinking of simply by observing the activity of its neurons—the complete antithesis of the behaviorist doctrine. Early commentators understandably found this hard to swallow, and thus were skeptical of the existence of place cells and of the notion of a hippocampal ‘‘black box.’’ However, in the decades since place cells were discovered, it has become apparent that the relationship between place cell activity and the rat’s behavior is indisputable. Exactly what this relationship is, however, remains somewhat obscure and a matter of continued debate. The hippocampus is deep in the brain and several synapses removed from either sensory or motor organs, so that numerous transformations occur in the intervening stages between inputs and outputs: it is thus hard to find a clear relationship between the covert activity of place cells and the overt behavior of their owner. The picture has been greatly complicated by the finding that although humans and animals with hippocampal lesions are impaired in navigation (Morris et al., 1982; Maguire et al., 1996), humans with such lesions are also profoundly impaired in episodic memory (Scoville and Milner, 1957; Aggleton and Brown, 1999; Cipolotti and Bird, 2006).
Place cells have provided a platform for one of the most important lines of enquiry in modern psychology, which is the way in which neural activity might, or does, underlie complex behaviors like thinking and planning. In some senses, the discovery of place cells has helped to bring about a revolution in psychology: the demise, more or less (though not completely), of the discipline of behaviorism and the concomitant rise of cognitivism. Behaviorists, beginning with Skinner and his followers, assumed that one did not need to understand the brain in order to understand behavior, and eschewed attempts to open the ‘‘black box’’ inside the head, believing that all the necessary information could be extracted simply by observing behavior under the right conditions. Cognitivism started from the contrasting premise that since all behavior is the end product of complex processes occurring inside the brain, then to fully understand behavior we need to know what these processes are. According to the cognitivist view, although the study of overt behavior is necessary to understand how that behavior came about, and also to understand the ‘‘mind,’’ it is not sufficient—for a complete understanding, both are needed. The first and still most controversial cognitivist hypothesis, put forward by Tolman, is that there exists, in the brain, a neural representation of place akin to a survey map of the environment, which he called a ‘‘cognitive map.’’ O’Keefe’s discovery of place cells in the rat hippocampus, in the early 1970s, prompted O’Keefe and Nadel (1978) to suggest that the hippocampus might be the site of Tolman’s cognitive map. 59
60 PLACE CELLS AND SPATIAL CONTEXT Nevertheless, if we are to make the argument that place cells represent ‘‘knowledge’’ or ‘‘memory,’’ it is necessary to find some principled relationship between the activity of the neurons and the overt behavior of the whole animal. The present chapter will review attempts to determine how activity of neurons in the place system—place cells and the more recently discovered head direction and entorhinal grid cells— relates to what the animal ‘‘knows,’’ as manifest by how it behaves. Beginning with O’Keefe and Nadel’s cognitive map hypothesis, the chapter will explore to what extent behavioral experiments have supported this idea, before turning to the question of how, if at all, these neurons contribute to episodic memory.
IS THE HIPPOCAMPUS A COGNITIVE MAP? The proposal by O’Keefe and Nadel that the hippocampus is the site of an allocentric (‘‘world-centered’’) spatial representation, or ‘‘cognitive map,’’ attracted an enormous amount of attention, both positive and negative, and continues to do so today. The theory was inspired by observations of the remarkable spatial localization of place cell firing fields, and it proposed that the place cells use constellations of landmarks to enable them to form a spatial representation for guiding navigational behavior. The authors made the point that not all kinds of spatial behavior would need such a map, highlighting that some kinds of navigation, which they called ‘‘taxon,’’ can be executed without reference to multiple distant landmarks, if proximal landmarks or body turns could be used instead. It has come to be widely recognized that such ‘‘response’’ behavior, driven by local stimuli and/or motor sequences, may rely on different neural systems from true, allocentric navigation (Packard and McGaugh, 1996; Mizumori et al., 2004), while yielding almost indistinguishable overt behaviors. The distinction between response-based navigation and mapbased navigation is important when linking neural processes with behavior. And, of course, it cannot be ruled out that there are yet more subtypes of navigation, not yet recognized, which complicate attempts to link place cell activity to behavior even further.
What Is a ‘‘Cognitive Map’’? The question of whether the hippocampus is (or at least has) a cognitive map has been bedeviled by the inability of experimenters to agree exactly on what a cognitive map is, which is a necessary prerequisite to finding out whether the hippocampus is or has one. In Tolman’s original (1948) formulation: ‘‘in the course
of learning, something like a field map of the environment gets established in the rat’s brain . . . and it is this tentative map, indicating routes and paths and environmental relationships, which finally determines what responses, if any, the animal will finally release’’ (p. 192). In O’Keefe and Nadel’s (1978) view, a cognitive map is a representation that ‘‘permits an animal to locate itself in a familiar environment without reference to any specific sensory input, to go from one place to another independent of particular inputs (cues) or outputs (responses), and to link together conceptually parts of an environment which have never been experienced at the same time’’ (p. 2). In the words of Downs and Stea (1973), it is ‘‘a process composed of a series of psychological transformations by which an individual acquires, codes, stores, recalls, and decodes information about the relative locations and attributes of phenomena in their everyday spatial environment’’ (p. 9). Many other similar definitions abound (see Bennett, 1996, and Healy et al., 2004, for reviews), and most have in common the idea that a map encodes relationships between objects or places in the environment, independent of the current viewpoint of the animal (that is, allocentrically). While this seems intuitively plausible, the difficulty is that relationship is a vague concept and the definitions, in themselves, rarely specify strict criteria by which the existence of such a map may be proved or disproved. For the purposes of the present discussion, it will be assumed that a cognitive map is a neural representation that allows an animal to localize its position with respect to the environment, by using information combined from two or more spatially discontiguous environmental features, usually in tandem with information from its own movements.
Are the Place Cells a Cognitive Map? Finding out whether the place cells form, or participate in, a cognitive map has been a fraught business. Many studies have looked for correlations between the hypothetical cognitive map (the place cells) and behavior by using tasks in which the animal is navigating to a goal or is otherwise actively engaging (we presume) its cognitive map. Such goal-directed studies have their complications, however. Alterations in behavior may have several sources: arising, for example, from an alteration in the underlying map, or in where the goal is placed on this hypothetical map, or from a disconnection between the goal representation and the actual behavior. Thus, if an animal makes a mistake in its navigation this may be because it is disoriented, or because it is perfectly well oriented but thinks the goal is somewhere else, or because it is
PLACE CELLS—COGNITIVE MAP OR MEMORY SYSTEM?
oriented and knows where the goal is but chooses not to go there on a particular trial. At present, we have no good way of dissociating these possibilities experimentally. This ambiguity is potentially problematic in interpreting the results of behavioral physiology experiments and should be kept in mind when interpreting the results of the studies described below. The first published study of a correlation between place cell activity and navigational behavior was a seminal experiment by O’Keefe and Speakman (1987), who trained animals to perform a spatial reference memory task on a four-armed radial maze (a plusmaze) in a cue-controlled environment. They found that both place fields and the choices of the rat corresponded, in most cases, to the orientation of the cues on a particular trial. Sometimes, the cues were removed after the rat had seen them but before it had made its choice: in such cases, choice accuracy remained high and place fields remained well localized and congruent with the rat’s choice. On trials where the cues were removed before the rat had seen them, performance fell, as expected, to chance: nevertheless, place fields continued to be aligned with the rat’s (faulty) choices. In fact, the experimenters were often able to predict the rat’s choice on the basis of observing its place fields. One implication is that perhaps the place cells ‘‘guessed’’ at the orientation of the environment, and their activity informed the actions of the rat. A conceptually similar experiment was undertaken by Dudchenko and Taube (1997), who recorded head direction cells as rats performed a reference memory task on a radial maze. Head direction cells, first reported in an abstract by Ranck (1984) and later in a more detailed analysis by Taube (Taube et al., 1990), fire when the animal’s head is pointing in a particular direction, irrespective of position. As with O’Keefe and Speakman’s place cells, alterations of head direction cell activity by cue rotation correlated with alterations in the rats’ choice of arms. Similarly, LenckSantini et al. (2001) trained rats to perform a continuous Y-maze alternation task, in which a constant goal arm was only rewarded if the animal visited each of the other two arms in turn. The environment possessed three-fold symmetry and the only disambiguating (‘‘polarizing’’) cue was a cue card hung at a constant spatial relationship to the goal. The cue card was rotated or removed, or the goal was rotated in the absence of the cue card. It was found that in sessions where the place fields assumed an inconsistent location with respect to the cue card and/or goal, the percentage of correct choices also fell, and errors were frequently consistent with the orientation of the fields. A number of other studies have also shown correlations between place field activity and behavior. Studies in which spatial attention was manipulated
61
have found that when animals are required to attend to spatial cues to succeed at a task, place fields are more compact and well formed. For example, Zinyuk et al. (2000) recorded place cells as rats locomoted over the surface of a rotating arena. In those rats that were merely foraging, and hence not needing to attend to distal cues, the fields tended to be poorly localized and messy. By contrast, in rats that performed a navigation task requiring attention to distal cues, the fields were localized and well formed, with a higher proportion of fields being influenced either by distal cues or a combination of distal and local cues, and a much smaller proportion being disrupted. Similarly, Kentros et al. (2004) found that place fields in mice were much more stable when the animals were required to attend to spatial cues in order to navigate. How can we interpret place cell–behavior correlations, like the ones above, with respect to the cognitive map theory? To make the case that a correlation represents a true causal link, it must first be established that the navigational behavior indeed requires a map. In cases where the correct goal locations could only be adduced with reference to the spatial layout of the cues, as in the studies described above, this seems likely according to the heuristic definition of cognitive map outlined earlier (the requirement for use of two or more distal cues in order to navigate). Thus, the observation of concordance between what the place cells did and how the animals behaved is reassuring. However, such correlations do not constitute proof that the cells drove the behavior—it might be that the cue array aligned both the place fields and the rat’s choice, but this does not mean that the place field array caused the rat’s choice. This problem is unavoidable, and so merely correlational experiments can never prove the cognitive map theory. We will return later to the question of whether a lack of correlation can falsify it.
The Problem of Discordant Representations The simplest form of the place cell–cognitive map theory has been challenged by findings in recent years that the so-called map has a distressing tendency toward ambiguity—a hard finding to explain if one believes the place cells represent where the animal ‘‘thinks’’ it is. The nature of this ambiguity and the challenge it poses are as follows. For a number of years after the discovery of place cells it was widely believed that the place cell representation is in fact tightly cohesive, because changes in the animal’s environment were observed to cause, almost invariably, a change of all observed cells simultaneously (Muller and Kubie, 1987). The strong impression this dramatic effect leaves on an observer is
62 PLACE CELLS AND SPATIAL CONTEXT that a new representation (map) has been recruited, leading to widespread adoption of Muller and Kubie’s term ‘‘remapping’’ to describe this phenomenon. As Nadel discusses in Chapter 1 (this volume), small environmental changes tend not to induce remapping whereas large changes induce ‘‘complete’’ remapping, apparently of the whole cell population. This flip-flop, all-or-nothing behavior has suggested to theoreticians (Marr, 1971; O’Reilly and McClelland, 1994; Guzowski et al., 2004) that the hippocampus is constantly balanced between pattern completion (when the changed environment is sufficiently similar to the original to require the same encoding) and pattern separation (when it is sufficiently different to require a completely new map). By this view, the decision about whether to complete or to separate is made by socalled attractor processes implemented in CA3 by the recurrent (i.e., interconnected) network there. This view is intuitively appealing because it tallies with our subjective sense that we either know we are in the same place, despite changes there, or we know we are in a new place, despite its resemblance to a familiar place. Complete remapping thus was considered to support the notion that place cell activity underlies the subjective sense of place (and novelty thereof). This view has been challenged, however, by increasingly frequent reports in recent years of a breakdown in the cohesiveness of the spatial representation following dissociation of environmental cues. Such ‘‘discordance’’—when some cells do one thing and some another—is often known as partial remapping. It was in fact first reported by Muller and Kubie (1987), although the observation attracted little attention at the time because it did not fit with the prevailing all-or-nothing view of the place cell representation. However, partial remapping has now been observed many times in a variety of settings (Bostock et al., 1991; Young et al., 1994; Shapiro et al., 1997; Tanila et al., 1997; Zinyuk et al., 2000; Anderson and Jeffery, 2003; Anderson et al., 2006)—see Figure 4–1. Although it is necessary to be careful that an apparent discordance is not just a statistical effect due to the random nature of remapping (see Brown and Skaggs, 2002, for a discussion of this), observations of partial remapping are nevertheless now too widespread to be ignored. Such ‘‘discordance’’ is troubling, because it does not tally with our intuitive sense that a spatial representation should be unique and unambiguous, as our own subjective sense of place generally is. The occurrence of discordant changes in place cell activity thus raises questions about how the representation relates to the animal’s knowledge. If only some cells change their activity in response to a change in the environment, does this mean that the animal is of two minds,
Figure 4–1. Discordance of the place cell representation induced by ‘‘contextual’’ changes to the visual appearance (black or white) or odor (lemon or vanilla) of an environment. The contour plots show the activity of four simultaneously recorded place cells. Cell 1 responded to a change in visual appearance by shifting its field, and responded to a change in odor by switching its field on or off. Cells 2, 3, and 4, by contrast, responded only when a particular odor and color were paired: cell 2 switched its field off in whitevanilla, cell 3 switched its field on in white-lemon, and cell 4 shifted its field to a different location in whitevanilla. Adapted from Anderson and Jeffery (2003). as it were, about which environment it is in? Or are these simultaneous partial representations merely steps along the road to a decision, somewhere else in the brain, about where the animal thinks it really is? We have suggested the latter possibility, following the results of an open-field place preference task in which rats were required to run to an unmarked goal on the surface of an arena for food reward (Anderson et al., 2006). Because the goal area was undefined by any local cues, we reasoned that only a true cognitive map (relating distant cues to each other) could enable navigation. After the task had been learned, the arena surface was changed from black to white or vice versa. Recording of place cells from two of these animals revealed that some cells changed their fields in response to the arena change while others did not. Thus, the arena change apparently induced a partial remapping. Observation of the rats’ behavior revealed an interesting finding: the behavior too, could be said to have ‘‘partially remapped.’’ When the arena was changed the animals reverted to thigmotaxis, which is the tendency of anxious rats to remain near the edge of the environment and to avoid the open area in the
PLACE CELLS—COGNITIVE MAP OR MEMORY SYSTEM?
middle. Thus, they had clearly noticed the change in the environment: the rats’ place representation was able to represent that something was different. Despite this, the change to the arena produced minimal disruption to the animals’ goal-localization abilities; the representation of the goal location had thus not changed. The correlation between partial remapping of place cells and ‘‘partial remapping’’ of behavior is interesting, and suggests (albeit does not prove) that the dual place cell representation enabled both kinds of behavior to co-occur. Where does this leave us with regard to the unity, or otherwise, of the cognitive map? Partial remapping suggests that the hippocampus does not always choose between pattern completion and pattern separation; in some cases it does both (or neither, depending on one’s point of view), and in such cases, behavior can also be ambiguous. This implies that either the cognitive map is not as cohesive and unitary as we thought, and can represent ambiguity, or, alternatively, that the real cognitive map—the hypothetical representation of unambiguous belief about current location—is somewhere else in the brain.
Might the Real Cognitive Map Be Somewhere Else? A number of lines of evidence point to the possibility that the place cells are only one of several place representations in the brain; many of these lines derive from the sometimes rather dramatic dissociations that have been observed between place cell activity and navigational behavior. Golob et al. (2001) and Muir and Taube (2004) found that head direction cell orientations showed a relatively low correlation with choices that rats made on two cognitive mapping tasks, suggesting that either the head direction cells were not orienting the cognitive map, or the cognitive map was not the sole determinant—or even any determinant at all—of the animals’ choices. We found a similar dissociation in a spatial task in which rats had, when hearing a tone, to proceed to one of the four corners of a square box to retrieve food (Jeffery et al., 2003). Hippocampal lesions impaired the ability of the rats to learn the task, and we presumed, therefore, that the place cell representation in the hippocampus must be needed for the rats to identify which corner of the box was the always-rewarded one. If this were the case, then disrupting this representation ought to have disrupted the rats’ choices. Accordingly, the environment was changed in appearance from black to white, or vice versa, to induce complete remapping of the place cell map. We predicted that if the place cell map is consulted in navigation, then induction of remapping ought to disrupt navigation. We found, in fact,
63
that navigation was little affected by this manipulation. Despite the fact that place cells completely reorganized their representation, the rats continued to navigate at levels considerably above chance. Such failures to find strict neuron–behavior correlates have two broad interpretations: one is that perhaps the rats were not using their cognitive map at the time the recordings were made, and the other is that while they may well be using it, the place or head direction cells are not in fact the basis of this map—it is somewhere else. Disentangling these two possibilities is a difficult task. (There is a third possibility— that the scrambled map somehow conveyed, in its altered form, the same spatial information as the prescrambled map, even though it had never been expressed before. This seems implausible and will not be considered further). Might the animals have been solving tasks such as these without using a cognitive map? In theory this is possible, and perhaps even likely. In the Jeffery et al. box-corner experiment, learning of the task took several hundred trials, which is more typical for a habit-based response task than for the fast-learned navigation underlying presumptive cognitive mapping (by contrast, water maze training may only take 8–10 trials). To form such a habit, the animals may have learned to distinguish the corners and identify the goal corner on the basis of single cues, rather than on the constellation of cues that rats normally use to execute a map-based navigational strategy. Indeed, the rats in that experiment quickly learned to run to the nearest corner of the box on hearing the tone, and only gradually learned to suppress this automatic cue-driven response and proceed to the correct—rather than the nearest—corner. With regard to the apparent hippocampal dependence of the task, it may be that the task required a hippocampus in order to be learned, but not once it had been overlearned, at the time of neuronal recording. It seems, then, that a more rigorous test of the cognitive map hypothesis would require a task that was unequivocally being solved using a cognitive map at the time of recording. This was the basis of the open-arena experiment of Anderson et al. (2006), described above—and as we saw, the partial remapping seen in that experiment continues to leave this question unanswered. The other explanation for place cell–behavior dissociations is that the ‘‘real’’ cognitive map—the hypothetical unambiguous representation of current location—is somewhere else in the brain. This proposition has been advocated by Sharp (1999), who proposed the existence of a ‘‘universal map’’ formed from co-operation between subiculum and entorhinal cortex, with the hippocampal representation being used to contextualize the map according to circum-
64 PLACE CELLS AND SPATIAL CONTEXT stances pertaining at a given time. The proposition that a stable map exists elsewhere in the brain is appealing, given the lability of the hippocampal map and the problems this poses for the idea that our sense of place resides here. However, an alternative reconciliation of representational lability and behavioral stability is outlined below.
Do Place Cells Encode Goals? One feature of a cognitive map is that it might allow the tagging of special places such as goals, and so an ongoing debate in the place cell literature concerns whether the activity of the cells encodes goal information. This deceptively simple question is fraught with methodological difficulties that make a definitive answer still elusive. The principal difficulty is that when a rat navigates to a goal, a vast array of neural systems comes into play, any one of which might plausibly influence place cells in a way not necessarily related to the ‘‘goalness’’ of the goal per se. Some of these systems are related to the motor output of a goal-seeking rat. For example, a rat that has learned about the location of a goal in an environment will tend to spend most of its time either at the goal or making its way toward it, thus biasing the spatial sampling of the place cell in favor of the goal location. In this case, cells that have place fields at the goal may appear to be encoding the goal when in fact they are encoding a place. Additionally, a rat that has stopped locomoting (such as when it has reached a goal) will enter an electrophysiologic state called large-amplitude irregular activity (LIA), the function of which is unknown but which is characterized by synchronous activity of many hippocampal cells, pyramidal cells included (Buzsaki et al., 1983). Thus, cells may seem to have goal-related place fields when actually they have stopping-related place fields. Other less well–characterized confounding phenomena may also come into play when an animal approaches a goal. Possible examples include arousal, expectation of reward, increased sensory processing (e.g., sniffing) associated with goal seeking, and initiation of consummatory behaviors such as salivation and licking. In addition, it is now fairly well established that highly salient features in an environment tend to be associated with an increased density of place fields in the vicinity (Hetherington and Shapiro, 1997)—it is thus necessary to distinguish encoding of the goal because it is a goal from encoding of it as merely a salient landmark in the environment. Dissociating the goalness of a goal from its other aspects is challenging. Despite this, a number of investigators have attempted to determine whether place cells are particularly interested in goals, with mixed
results. Speakman and O’Keefe (1990) tested goal encoding explicitly by moving the goal in their plusmaze task, and failed to see any corresponding shift in place cell activity. Similarly, Jeffery et al. (2003) did not observe preferential encoding of the goal corner in their box-corner task; nor did they see evidence of responding when the tone signaling the goal was sounded. However, in an earlier study Eichenbaum et al. (1987) recorded hippocampal neurons as rats performed a go/no-go olfactory discrimination task, and found that unit activity was frequently synchronized to various phases of the task, including odor sampling and goal approach. Others have also reported enhancement of place cell activity at goal locations (Breese et al., 1989; Wiener et al., 1989). An interesting but puzzling observation that has been made several times now is of place fields apparently moving towards a goal location over repeated trials (Kobayashi et al., 1997; Hollup et al., 2001; Lee et al., 2006). It is a little unclear what to make of these studies. The absence of obvious goal-related activity in many studies of place cells in navigating rats suggests that a straightforward encoding of the goal by place cells, in the way that other places are encoded, is unlikely, a conclusion that is plausible in light of the theoretical difficulties of using a goal representation active at the goal (where it is not needed) rather than away from the goal (where it is needed). However, observations of differential place cell activity at goal locations, insofar as these are not behavioral artifacts, may imply some kind of tagging of the place as significant— perhaps to be used in modulating activity at some other region that is used in goal encoding (such as prefrontal cortex; Hok et al., 2005). An intriguing possibility is that information about the goal is encoded via modulation of place responding, a phenomenon known as gain-field modulation, found prominently in parietal cortex (Andersen, 1997). An alternative is that the synchronous multi-unit activity comprising LIA might incorporate goal information and perhaps be used in tagging that place for future use.
IS SPACE SPECIAL TO THE HIPPOCAMPUS? The cognitive map theory of hippocampus function has engendered a lively debate over the years. Views as to whether the hippocampus is a cognitive map can be aligned along a gradient, roughly as follows: 1. The hippocampus is a purely cognitive map, and processes only spatial information. 2. The hippocampus is a cognitive map, but this map is richly embellished by nonspatial infor-
PLACE CELLS—COGNITIVE MAP OR MEMORY SYSTEM?
mation to form a representation of spatial context, used to (among other things) organize memories. 3. The hippocampus is purely a memory system and space has no privileged role in this system— it just seems that way in some experiments. The evidence is now overwhelming that place cells respond to nonspatial information (the manipulations in Fig. 4–1 being a case in point) and so point 1 above can safely be dismissed. The principal question now, therefore, is whether the primary function of the hippocampus is to represent place, with memory being secondary, or whether its primary role is memory, with space being secondary. An evaluation of the study of hippocampal neuronal responses to nonspatial stimuli is relevant here. Nonspatial stimuli may be classed as focal or as extended along some dimension, and as there may be an important difference between these two kinds of stimuli, they will be discussed separately.
The Hippocampus and Focal Stimuli Stimuli that occupy a narrow extent in space can be thought of as focal or discrete, and those that are narrow in time, as it were, can be thought of as transient. There is quite a lot of evidence that hippocampal neurons are not particularly interested in either discrete or transient stimuli. This is true even within the spatial domain: Poucet and colleagues have observed that place cells do not respond to point-like objects in an arena (Cressant et al., 1997, 1999), unless these are arranged into an extended array and placed at the edge of the apparatus, where they can exert directional control. In contrast to spatially discrete stimuli, many more studies have looked at responsiveness to transient stimuli, and again the picture is mixed: effects are often weak and tend to be statistical rather than overtly apparent. In an early series of studies, Berger and Thompson showed that hippocampal neurons in rabbits developed conditioned responding to paired presentations of a tone with a corneal air puff (Berger et al., 1976). Eichenbaum and colleagues subsequently reported that hippocampal neurons fired in response to behavioral events such as cue sampling and approach to a goal (Eichenbaum et al., 1987; Wiener et al., 1989; Wood et al., 1999), and Deadwyler and colleagues have also reported event-related activity on a number of occasions (Hampson et al., 1999; Deadwyler and Hampson, 2004). These studies generally find that responsiveness to transient stimuli, when it occurs, is overlaid on a fundamentally spatial response. For example, Moita et al. (2003) found that
65
following conditioning, activity within a place field was enhanced on presentation of the conditioned stimulus. It should be noted that in all these studies, the place cell responses were not evident initially but developed over prolonged exposure to the cues. This pattern tallies with the commonplace finding that transient and unique stimuli in the environment (such as experimenters speaking, a door slamming, or other ordinary laboratory sounds) do not generally evoke sporadic out-of-field activity in hippocampal neurons. What conclusions can be drawn from studies of place cells responding to discrete cues? The abundance of studies reporting correlations between presentation of conditioned cues and activity of the place cells suggests that the cells can respond to transient stimuli. Equally clearly, however, is the finding that the responsiveness is not as dramatic as it is in the case of place responding. This suggests that responding to transient cues is not like responding to spatial cues, only in a different stimulus domain. Rather, it seems that these processes may be qualitatively different things. It will be argued later that focal (in space or time) stimuli only come to drive hippocampal neurons if these are repeated sufficiently regularly and often as to become part of the spatial context.
The Hippocampus and Extended Stimuli Contrasting with the discrete and/or transient stimuli discussed above are those stimuli that are extended in space and/or in time. Extended spatial stimuli were discussed above in the section on cognitive mapping. A large class of stimuli, however, lack any metric properties at all, and yet are able to modulate place cell firing quite dramatically; these stimuli tend to have in common the property of persistence. The other trait they share is that they seem to act differently from spatial stimuli, which suggests, as will be argued below, that they are functionally different from spatial stimuli. This is an important conclusion, because it supports the hypothesis that ‘‘space is special’’ in driving place cells. One of the extended nonspatial stimulus dimensions to be implicated in hippocampal function is time. A number of studies have looked at the temporal correlations of unit activity in nonspatial tasks, and a picture of sorts is emerging but it is not straightforward. If time were treated by place cells as simply another dimension of space–time, like the spatial dimensions, then encoding in the time dimension should have similar properties to those of encoding of space: different neurons should encode different intervals, encoding should not be clustered around salient times any more than they are clustered around salient places (which is to say, very slightly but not much), and
66 PLACE CELLS AND SPATIAL CONTEXT subtle manipulations of the temporal interval should cause expansion of the temporal domain of responding in the same way as manipulations of the spatial interval cause scaling of place fields (O’Keefe and Burgess, 1996). In fact, studies of the temporal encoding of hippocampal neurons find that while neurons in conditioning tasks frequently respond to salient times such as trace interval (e.g., McEchron et al., 2003), there is no evidence that other intervals within the same time span are also represented—that is, neural responses do not ‘‘tile’’ the temporal interval in the same way that place fields tile a spatial area. Thus, if place cells encode time, they do not do so in the same way that they encode place. What about other stimulus dimensions? Data on this issue are limited, but preliminary indications are that these are not treated like space either. Studies of continuous deformation of the environmental shape (where ‘‘shape’’ is a stimulus dimension) from a square to a circle, using so-called morph boxes, find that when place cells transition from one state to another, either abruptly (Wills et al., 2005) or gradually (Leutgeb et al., 2005), they do so only once, from on to off or vice versa—that is, they do not exhibit ‘‘fields’’ in the morphing dimension. We have made a similar observation in pilot experiments in which an environment was varied from dark to light (Fig. 4–2). When transitions did occur they tended to be abrupt, and cells that altered their activity all did so together, transitioned only once, and did not exhibit ‘‘fields’’ in the light–dark dimension. This contrasts with the finding that place cells recorded on a three-dimensional apparatus do appear to exhibit fields in the vertical dimension, having similar properties (tiling,
etc.) as they do in the horizontal plane (Verriotis et al., in preparation). Thus, initial indications are that place cells seem prone to forming fields in all three spatial dimensions but not in the other stimulus domains. Another asymmetry between spatial and nonspatial stimuli comes from studies of contextual remapping, where nonspatial stimuli are found to induce remapping. A wide variety of remapping stimuli have now been identified. Many of these are cues that are clearly extended in space and/or time: they include odor and/ or visual appearance of the environmental boundaries (Anderson and Jeffery, 2003), shape of the enclosure (Lever et al., 2002; Leutgeb et al., 2005), ambient lighting (Quirk et al., 1990; Fig. 4–2), and texture. Place cells clearly receive inputs from these sensory modalities because remapping is often robust. However, cells that respond to these nonspatial stimuli also invariably have place fields, whereas the converse is not true: for example, no hippocampal neuron has been reported that switches on and off to a nonspatial stimulus such as odor but fires everywhere. Indeed, the best illustration of the pre-eminence of spatial information in driving place cells comes from a study by Dragoi et al. (2003), in which tetanization of the Schaffer collaterals to induce long-term potentiation of CA1 afferents caused remapping of place fields and thus new fields to be expressed. The implication of this remarkable finding is that the new fields themselves were hard-wired, and synaptic plasticity altered only their tendency to be expressed. In recent years, another interesting class of stimuli have emerged as modulators of place cell activity: these are factors internal to the rat, such as task type (Markus et al., 1995) and route (Frank et al., 2000;
Figure 4–2. Remapping of place cells induced by varying ambient lighting from dark to light. Ambient lighting was increased from dark to light in 10% steps every 120 s. The far left panel shows the data pooled for 0–120 s, and the far right panel shows the data pooled from 240 to 1320 s. The four intermediate panels show a breakdown of the transitional period, between 60 and 300 s— note that the cell shifted its firing field abruptly (arrow) at around 240 s, when the lighting was increased from 10% to 20%, but did not alter again thereafter. This single transition is characteristic of remapping phenomena and contrasts with the on-and-off behavior of spatially localized place fields.
PLACE CELLS—COGNITIVE MAP OR MEMORY SYSTEM?
Wood et al., 2000; Ferbinteanu and Shapiro, 2003). As with the other nonspatial stimuli discussed above, these factors act by modulating space activity (switching place fields on and off) rather than driving the cells independently. A number of investigators have interpreted these studies as evidence of episodic encoding. This seems unlikely, however, as these stimuli, like the physical stimuli discussed above, need to be repeated many times before they come to modulate place cell firing. A more plausible explanation is that by dint of repeated presentation, these ‘‘virtual’’ stimuli become temporally extended (a state of the environment) and are thus incorporated into the place cell representation (Jeffery et al., 2004; Smith and Mizumori, 2006). From the above observations, we can tentatively draw two conclusions. First, hippocampal neurons are not free associators (or relational detectors) in all stimulus domains; they only respond to some kinds of stimuli (those extended in space and/or time). Second, regardless of the stimuli they respond to, they only seem to do so if the animal is also in a place field. Thus, it seems that a stimulus will only acquire the property of driving a place cell if it is extended in space and/or time and if it co-occurs with a constant subset of spatial stimuli—that is, the modulating stimulus always occurs in the same place. These two conditions endow place cells with the property of being able to associate persistent and/or extensive stimuli with a particular part of space, allowing them to encode not space alone but ‘‘spatial context’’ (Mizumori et al., 1999; Jeffery et al., 2004).
67
so that the surface of the navigable terrain is completely tiled (Fig. 4–3). A mosaic map differs from the more traditional continuous, Euclidean survey map (or what Worden [1992] calls ‘‘one big map’’) in that the metric relationships between fragments in the map need not be exact (though they might be): so long as the animal knows how to get from one fragment to the next then it does not matter if its representations of exact distances and directions are somewhat awry. At the level of the whole mosaic it is thus a topological representation. Within a fragment, however, metric relationships are likely to be veridical because the animal can access the relevant stabilizing landmarks more or less simultaneously. A useful discussion of how such a metric/topological hierarchy could function in cognitive mapping is found in Poucet (1993). This idea is developed, below, in light of more recent single neuron data. The properties of place cells argue in favor of a mosaic representation over larger scales, because the cells themselves (at least in rodents) are best influenced by local information, such as would be found in an environmental fragment. Place cells are very strongly influenced by boundaries and express their fields strongly when the rat is near a wall or an edge (O’Keefe and Burgess, 1996) and most strongly when enclosed on all sides (Barry et al., 2006). Studies in which environments have been divided up by barriers find that place fields occurring on one side of a boundary do not continue on the other side (Muller and Kubie, 1987; Hartley et al., 2000; Rivard et al., 2004; Barry et al., 2006, Figure 4–4). This suggests
THE PLACE CELLS AS A MOSAIC MAP? As cognitive maps go, it might seem from the discussion above that the place cell map is not a very good one. Although place cells are inordinately and uniquely responsive to spatial cues, their representation has a fragility and lability (manifest as remapping) that the internal subjective representation of space, certainly in humans, seems not to have. These observations have led some to speculate that if there is a real, unitary cognitive map, the place cells are not it. Might there be a way of reconciling the lability of the place cell maps with the stability of spatial behavior? In this section, it will be argued that the place cells construct a mosaic map of space, to be used as a foundation for navigation and episodic memory. While each piece of the mosaic may be labile, the map as a whole is not. A mosaic map is defined here as one in which fragments of the environment are represented individually, together with their adjacency relationships,
A
B D E
G
F A
D C B
E G F
C B
A
Figure 4–3. A. A mosaic map divides the environment into a set of ‘‘fragments’’ (Worden, 1992). In the hypothetical place-cell mosaic map, each fragment has an intrinsic orientation and is related to neighboring fragments by their allocentric direction, or by landmarks. B. Resistance of a mosaic map to distortion. So long as the fragments are approximately the right size and approximately the right allocentric directions from each other, the map can be used in navigation, with local cues used to hone navigation in the final goal approach stage.
68 PLACE CELLS AND SPATIAL CONTEXT that they are encoding local features rather than more global positions in a larger-scale map. Similarly, head direction cells maintain a consistent orientation within a bounded chamber but do not necessarily carry this orientation across connected chambers (Taube and Burton, 1995), and changes made to place fields in one of a pair of connected chambers do not propagate to the other (Paz-Villagran et al., 2004). These findings in place cells have been supported by recent findings in grid cells, a newly discovered class of neuron found in dorsomedial entorhinal cortex, whose firing fields are laid out in a neat hexagonal array across the surface of the environment (Hafting et al., 2005). It has recently been found that grids are discontinuous at boundary points in the environment (Derdikman et al., 2006). Collectively, these studies suggest that the interior of an enclosed space is uniquely represented and not a fragment of a larger continuous map. Discontinuities occurring at boundaries imply that the map could not be used for vector calculations across boundaries but only within the confines of a particular boundary set. How, then, could an animal, using a mosaic map, navigate from one piece of the mosaic to the next? This could be done easily if the brain represents, as well as the pieces of the mosaic, some kind of information about how these pieces fit together. For example, if the animal has stored information that the fragment containing its burrow is ‘‘north of’’ or ‘‘on the other side of the hedge from’’ the fragment containing the pond, it can move between these areas fairly easily just by knowing roughly where north is or by progressing toward the hedge. Such navigation superficially resembles route navigation but differs from it in that, in principle, a novel shortcut could be taken if the animal knew the approximate relationships of the intervening fragments. For example, in the mosaic illustrated in Figure 4–3, the animal could
Figure 4–4. Three place cells recorded in a square box from which the south wall was slightly separated to allow the rat into the surrounding space. Note that fields lie on one side of a wall or the other, but do not cross it, indicating that the larger space is not continuously represented but that the representation is divided up by the walls. Data courtesy of R. Hayman.
make its way along a novel route from A to E if it knew the relationship between A and G and G and E. That is, it could use its internal representation of the relationships of spatially discontiguous landmarks to construct a navigational plan—the definition of cognitive map argued for earlier. One feature of a mosaic map of potential interest is that it need not possess an accurate metric to function effectively. As long as each fragment of the mosaic is linked to its neighbors by approximately correct compass directions, navigation from one fragment to the next could take place by relatively coarse metric operations, the main necessity being that within a fragment the animal remain relatively well oriented. Inhabitants of large cities know this well: it is possible to find one’s way perfectly well around Central London, for example, while having a poor or nonexistent appreciation of the precise metric relations involved. Mosaicism of the spatial map could also explain why head direction cells in neighboring regions of space need do not always have the same orientation (Taube and Burton, 1995; Dudchenko and Zinyuk, 2005). If the adjacent regions (the ‘‘fragments’’) can be related in some other way, such as by local cues, then a coherent orientation across fragments is not necessary.
CONCLUDING REMARKS—ARE THE PLACE CELLS A MAP, OR A MEMORY SYSTEM? The human hippocampus is widely accepted to be crucially important for episodic memory (Aggleton and Brown, 1999). What does the place cell map, mosaic or otherwise, have to do with episodic memory? The relationship is still obscure and can only be speculated on at present. It was argued earlier that whatever other stimuli might drive place cells, metric stimuli remain funda-
PLACE CELLS—COGNITIVE MAP OR MEMORY SYSTEM?
mentally privileged in this regard. Place cells might be interested in more than just space, but evidence favors the notion that they are primarily interested in space, with other stimuli serving only to modulate the place representation—and even then, only if they possess certain properties (are extended and/or persistent). Data on primates are more limited, but such as they are, they also support the notion of space as a fundamental correlate (Rolls, 1999; Ekstrom et al., 2003; Hori et al., 2003). If one assumes that there is a functional homology between rodents and humans with regard to the anatomy and mechanisms of episodic memory, then spatial and episodic functions need somehow to be unified. Attempts to unify the spatial and memory theories of hippocampal function fall, as discussed earlier, into two broad classes: those in which space is considered pre-eminent, with memory being secondarily associated, and those in which memory is considered preeminent, with space as one form of memory (but not the only one). The strongly spatial character of placecell firing correlates argues for the former: that space is primary, and episodic memory builds upon a spatial framework. An argument put forward in favor of this view is that although an episode contains many kinds of stimuli, only stimuli with certain properties—which tend in fact to be persistent rather than episodic ones— seem able to modulate place cells. It has been suggested here that place cells form a mosaic map of space in which the environment is parceled up by boundaries, with each fragment having its own independent representation. Each independent representation seems to be rather labile, insofar as seemingly minor changes to the environment (such as changes in odor) can induce dramatic reorganization of the place cell representation (remapping). On the face of it, this lability argues against the notion of the place cells as a cognitive map—what use is a map that changes so drastically with changes to the environment? It is suggested here that in fact this need not matter, so long as the alternative (‘‘contextualized’’) representations of a given fragment remain embedded in the same larger mosaic—the animal would still have its ‘‘sense of place’’ and could still navigate. It is just that it is able to simultaneously represent several things about a particular piece of this space. In the same way, one can represent (without becoming disoriented) the notion that a large room in a building is sometimes a concert theater and sometimes a sports hall, depending on the context and without becoming disoriented Could a mosaic map be useful in episodic memory? It has been argued many times, from O’Keefe and Nadel on, that a cognitive map could indeed be useful in episodic memory if such memories need to be ‘‘attached’’ to a map, for retrieval purposes. The argu-
69
ments outlined here support this view. While data suggesting a role for place cells in encoding transient events are scarce, data suggesting that cells may encode the spatial–contextual scaffolding for the attachment of episodic memory are plentiful and plausible. It remains to be seen if this is the real, fundamental role of place cells.
acknowledgments The work described here received support from the Wellcome Trust, the Biotechnology and Biological Sciences Research Council, and the Medical Research Council. The author would like to thank Chris Bird, Paul Dudchenko, and Lynn Nadel for comments on the manuscript.
References Aggleton JP, Brown MW (1999) Episodic memory, amnesia, and the hippocampal-anterior thalamic axis. Behav Brain Sci 22:425–444. Andersen RA (1997) Multimodal integration for the representation of space in the posterior parietal cortex. Philos Trans R Soc Lond B Biol Sci 352:1421–1428. Anderson MI, Jeffery KJ (2003) Heterogeneous modulation of place cell firing by changes in context. J Neurosci 23:8827–8835. Anderson M, Killing S, Morris C, O’Donoghue A, Onyiagha D, Stevenson R, Verriotis M, Jeffery K (2006) Behavioral correlates of the population coding of spatial context. Hippocampus 16:730–742. Barry C, Lever C, Hayman R, Hartley T, Burton S, O’Keefe J, Jeffery KJ, Burgess N (2006) The boundary vector cell model of place cell firing and spatial memory. Rev Neurosci 17:71–97. Bennett AT (1996) Do animals have cognitive maps? J Exp Biol 199 ( Pt 1):219–224. Berger TW, Alger B, Thompson RF (1976) Neuronal substrate of classical conditioning in the hippocampus. Science 192:483–485. Bostock E, Muller RU, Kubie JL (1991) Experience-dependent modifications of hippocampal place cell firing. Hippocampus 1:193–205. Breese CR, Hampson RE, Deadwyler SA (1989) Hippocampal place cells: stereotypy and plasticity. J Neurosci 9:1097–1111. Brown JE, Skaggs WE (2002) Concordant and discordant coding of spatial location in populations of hippocampal CA1 pyramidal cells. J Neurophysiol 88: 1605–1613. Buzsaki G, Leung LW, Vanderwolf CH (1983) Cellular bases of hippocampal EEG in the behaving rat. Brain Res 287:139–171. Cipolotti L, Bird CM (2006) Amnesia and the hippocampus. Curr Opin Neurol 19:593–598.
70 PLACE CELLS AND SPATIAL CONTEXT Cressant A, Muller RU, Poucet B (1997) Failure of centrally placed objects to control the firing fields of hippocampal place cells. J Neurosci 17:2531– 2542. Cressant A, Muller RU, Poucet B (1999) Further study of the control of place cell firing by intra-apparatus objects. Hippocampus 9:423–431. Deadwyler SA, Hampson RE (2004) Differential but complementary mnemonic functions of the hippocampus and subiculum. Neuron 42:465–476. Derdikman D, Fyhn M, Hafting T, Moser MB, Moser E (2006) Breaking up the entorhinal grid in a hairpin maze. Soc Neurosci Abstr 68.10/BB18. Downs RM, Stea D (1973) Cognitive Maps and Spatial Behavior: Process and Products. In: Image and Environments (Downs RM, Stea D, eds.), pp 8–26. Chicago: Aldine Publishing Company. Dragoi G, Harris KD, Buzsaki G (2003) Place representation within hippocampal networks is modified by long-term potentiation. Neuron 39:843–853. Dudchenko PA, Taube JS (1997) Correlation between head direction cell activity and spatial behavior on a radial arm maze. Behav Neurosci 111:3–19. Dudchenko PA, Zinyuk LE (2005) The formation of cognitive maps of adjacent environments: evidence from the head direction cell system. Behav Neurosci 119:1511–1523. Eichenbaum H, Kuperstein M, Fagan A, Nagode J (1987) Cue-sampling and goal-approach correlates of hippocampal unit activity in rats performing an odordiscrimination task. J Neurosci 7:716–732. Ekstrom AD, Kahana MJ, Caplan JB, Fields TA, Isham EA, Newman EL, Fried I (2003) Cellular networks underlying human spatial navigation. Nature 425: 184–188. Ferbinteanu J, Shapiro ML (2003) Prospective and retrospective memory coding in the hippocampus. Neuron 40:1227–1239. Frank LM, Brown EN, Wilson M (2000) Trajectory encoding in the hippocampus and entorhinal cortex. Neuron 27:169–178. Golob EJ, Stackman RW, Wong AC, Taube JS (2001) On the behavioral significance of head direction cells: neural and behavioral dynamics during spatial memory tasks. Behav Neurosci 115:285–304. Guzowski JF, Knierim JJ, Moser EI (2004) Ensemble dynamics of hippocampal regions CA3 and CA1. Neuron 44:581–584. Hafting T, Fyhn M, Molden S, Moser MB, Moser EI (2005) Microstructure of a spatial map in the entorhinal cortex. Nature 801–806. Hampson RE, Simeral JD, Deadwyler SA (1999) Distribution of spatial and non-spatial information in dorsal hippocampus. Nature 402:610–614.
Hartley T, Burgess N, Lever C, Cacucci F, O’Keefe J (2000) Modeling place fields in terms of the cortical inputs to the hippocampus. Hippocampus 10:369–379. Healy SD, Hodgson Z, Braithwaite V (2004) Do animals use maps? In: The Neurobiology of Spatial Behaviour (Jeffrey JK, ed.), pp 104–118. Oxford: Oxford University Press. Hetherington PA, Shapiro ML (1997) Hippocampal place fields are altered by the removal of single visual cues in a distance-dependent manner. Behav Neurosci 111: 20–34. Hok V, Save E, Lenck-Santini PP, Poucet B (2005) Coding for spatial goals in the prelimbic/infralimbic area of the rat frontal cortex. Proc Natl Acad Sci USA 102:4602–4607. Hollup SA, Molden S, Donnett JG, Moser MB, Moser EI (2001) Accumulation of hippocampal place fields at the goal location in an annular watermaze task. J Neurosci 21:1635–1644. Hori E, Tabuchi E, Matsumura N, Tamura R, Eifuku S, Endo S, Nishijo H, Ono T (2003) Representation of place by monkey hippocampal neurons in real and virtual translocation. Hippocampus 13:190–196. Jeffery KJ, Anderson MI, Hayman R, Chakraborty S (2004) A proposed architecture for the neural representation of spatial context. Neurosci Biobehav Rev 28:201–218. Jeffery KJ, Gilbert A, Burton S, Strudwick A (2003) Preserved performance in a hippocampal dependent spatial task despite complete place cell remapping. Hippocampus 13:175–189. Kentros CG, Agnihotri NT, Streater S, Hawkins RD, Kandel ER (2004) Increased attention to spatial context increases both place field stability and spatial memory. Neuron 42:283–295. Kobayashi T, Nishijo H, Fukuda M, Bures J, Ono T (1997) Task-dependent representations in rat hippocampal place neurons. J Neurophysiol 78:597–613. Lee I, Griffin AL, Zilli EA, Eichenbaum H, Hasselmo ME (2006) Gradual translocation of spatial correlates of neuronal firing in the hippocampus toward prospective reward locations. Neuron 51:639–650. Lenck-Santini PP, Save E, Poucet B (2001) Evidence for a relationship between place-cell spatial firing and spatial memory performance. Hippocampus 11:377–390. Leutgeb JK, Leutgeb S, Treves A, Meyer R, Barnes CA, McNaughton BL, Moser MB, Moser EI (2005) Progressive transformation of hippocampal neuronal representations in ‘‘morphed’’ environments. Neuron 48: 345–358. Lever C, Wills T, Cacucci F, Burgess N, O’Keefe J (2002) Long-term plasticity in hippocampal placecell representation of environmental geometry. Nature 416:90–94.
PLACE CELLS—COGNITIVE MAP OR MEMORY SYSTEM?
Maguire EA, Burke T, Phillips J, Staunton H (1996) Topographical disorientation following unilateral temporal lobe lesions in humans. Neuropsychologia 34: 993–1001. Markus EJ, Qin YL, Leonard B, Skaggs WE, McNaughton BL, Barnes CA (1995) Interactions between location and task affect the spatial and directional firing of hippocampal neurons. J Neurosci 15:7079–7094. Marr D (1971) Simple memory: a theory for archicortex. Philos Trans R Soc Lond B Biol Sci 262:23–81. McEchron MD, Tseng W, Disterhoft JF (2003) Single neurons in CA1 hippocampus encode trace interval duration during trace heart rate (fear) conditioning in rabbit. J Neurosci 23:1535–1547. Mizumori SJ, Ragozzino KE, Cooper BG, Leutgeb S (1999) Hippocampal representational organization and spatial context. Hippocampus 9:444–451. Mizumori SJ, Yeshenko O, Gill KM, Davis DM (2004) Parallel processing across neural systems: implications for a multiple memory system hypothesis. Neurobiol Learn Mem 82:278–298. Moita MA, Rosis S, Zhou Y, LeDoux JE, Blair HT (2003) Hippocampal place cells acquire location-specific responses to the conditioned stimulus during auditory fear conditioning. Neuron 37:485–497. Morris RG, Garrud P, Rawlins JN, O’Keefe J (1982) Place navigation impaired in rats with hippocampal lesions. Nature 297:681–683. Muir GM, Taube JS (2004) Head direction cell activity and behavior in a navigation task requiring a cognitive mapping strategy. Behav Brain Res 153:249– 253. Muller RU, Kubie JL (1987) The effects of changes in the environment on the spatial firing of hippocampal complex-spike cells. J Neurosci 7:1951–1968. O’Keefe J, Burgess N (1996) Geometric determinants of the place fields of hippocampal neurons. Nature 381:425–428. O’Keefe J, Nadel L (1978) The Hippocampus as a Cognitive Map. Oxford: Clarendon Press. O’Keefe J, Speakman A (1987) Single unit activity in the rat hippocampus during a spatial memory task. Exp Brain Res 68:1–27. O’Reilly RC, McClelland JL (1994) Hippocampal conjunctive encoding, storage, and recall: avoiding a trade-off. Hippocampus 4:661–682. Packard MG, McGaugh JL (1996) Inactivation of hippocampus or caudate nucleus with lidocaine differentially affects expression of place and response learning. Neurobiol Learn Mem 65:65–72. Paz-Villagran V, Save E, Poucet B (2004) Independent coding of connected environments by place cells. Eur J Neurosci 20:1379–1390.
71
Poucet B (1993) Spatial cognitive maps in animals: new hypotheses on their structure and neural mechanisms. Psychol Rev 100:163–182. Quirk GJ, Muller RU, Kubie JL (1990) The firing of hippocampal place cells in the dark depends on the rat’s recent experience. J Neurosci 10:2008–2017. Ranck J (1984) Head-direction cells in the deep cell layers of dorsal presubiculum in freely moving rats. Soc Neurosci Abstr 10:599. Rivard B, Li Y, Lenck-Santini PP, Poucet B, Muller RU (2004) Representation of objects in space by two classes of hippocampal pyramidal cells. J Gen Physiol 124:9–25. Rolls ET (1999) Spatial view cells and the representation of place in the primate hippocampus. Hippocampus 9:467–480. Scoville WB, Milner B (1957) Loss of recent memory after bilateral hippocampal lesions. J Neuropsychiatry Clin Neurosci 12:103–113. Shapiro ML, Tanila H, Eichenbaum H (1997) Cues that hippocampal place cells encode: dynamic and hierarchical representation of local and distal stimuli. Hippocampus 7:624–642. Sharp PE (1999) Complimentary roles for hippocampal versus subicular/entorhinal place cells in coding place, context, and events. Hippocampus 9:432–443. Smith DM, Mizumori SJ (2006) Hippocampal place cells, context, and episodic memory. Hippocampus 16:716– 729. Speakman A, O’Keefe J (1990) Hippocampal complex spike cells do not change their place fields if the goal is moved within a cue controlled environment. Eur J Neurosci 2:544–555. Tanila H, Shapiro ML, Eichenbaum H (1997) Discordance of spatial representation in ensembles of hippocampal place cells. Hippocampus 7:613–623. Taube JS, Burton HL (1995) Head direction cell activity monitored in a novel environment and during a cue conflict situation. J Neurophysiol 74:1953–1971. Taube JS, Muller RU, Ranck-JB J (1990) Head-direction cells recorded from the postsubiculum in freely moving rats. I. Description and quantitative analysis. J Neurosci 10:420–435. Tolman EC (1948) Cognitive maps in rats and men. Psychol Rev 40:40–60. Wiener SI, Paul CA, Eichenbaum H (1989) Spatial and behavioral correlates of hippocampal neuronal activity. J Neurosci 9:2737–2763. Wills TJ, Lever C, Cacucci F, Burgess N, O’Keefe J (2005) Attractor dynamics in the hippocampal representation of the local environment. Science 308:873–876. Wood ER, Dudchenko PA, Eichenbaum H (1999) The global record of memory in hippocampal neuronal activity. Nature 397:613–616.
72 PLACE CELLS AND SPATIAL CONTEXT Wood ER, Dudchenko PA, Robitsek RJ, Eichenbaum H (2000) Hippocampal neurons encode information about different types of memory episodes occurring in the same location. Neuron 27:623–633. Worden R (1992) Navigation by fragment fitting: a theory of hippocampal function. Hippocampus 2:165–187. Young BJ, Fox GD, Eichenbaum H (1994) Correlates of hippocampal complex-spike cell activity in rats per-
forming a nonspatial radial maze task. J Neurosci 14: 6553–6563. Zinyuk L, Kubik S, Kaminsky Y, Fenton AA, Bures J (2000) Understanding hippocampal activity by using purposeful behavior: place navigation induces place cell discharge in both task-relevant and task-irrelevant spatial reference frames. Proc Natl Acad Sci USA 97:3771–3776.
5 Context-Specific Versus Context-Invariant Spatial Coding in the Hippocampal Formation: Implications for Episodic Memory PATRICIA E. SHARP
HIPPOCAMPAL PLACE CELLS SHOW EVIDENCE OF PATH INTEGRATION CAPABILITIES BUT MAY NOT CONSTITUTE THE INTEGRATION CIRCUITRY
The investigation of navigation-related neuronal firing patterns was initiated by the work of John O’Keefe and colleagues (O’Keefe and Dostrovsky, 1971; O’Keefe, 1976), who discovered the hippocampal place cell phenomenon. These hippocampal place cell signals were so salient, robust, and plentiful that their discovery ignited a new subfield within neuroscience dedicated to the further investigation of spatial signaling in freely behaving animals. The fact that these spatial signals were initially discovered in the hippocampus proper also led to a decidedly ‘‘hippocampocentric’’ approach to this investigation of the neural basis of navigation. Indeed, the vast majority of work done on spatial signaling in moving animals has been conducted on cells in the hippocampus proper. This substantial literature has, at least until recently, implicitly supported the idea that the hippocampus proper is responsible for all or most of the neural computation necessary for navigational cognition. In addition, this work has influenced our thinking about the overall role of the hippocampus in learning and memory. Specifically, it has supported the idea that the critical role played by the hippocampus is to provide a (fundamentally spatial) cognitive map that in turn provides the substrate for many mnemonic and behavioral abilities (O’Keefe and Nadel, 1978).
Enthusiasm for the idea that the hippocampus is a cognitive map was fueled by the observation that, at least sometimes, the hippocampal place cells show an uncanny ability to keep track of the animal’s spatial location, even after the initial orienting cues have been removed or severely diminished (Muller and Kubie, 1987; O’Keefe and Speakman, 1987; Quirk et al., 1990). As initially outlined by O’Keefe (1976), this ability is thought to be due to a process known as path integration, or dead reckoning. According to this idea, the place cell firing patterns are constantly updated according to the animal’s own movement trajectory. To get a sense of how this path integration works, imagine you are in a familiar environment, such as your office, and suddenly the lights go out and it is completely dark. In this case you would probably be able to get up from your desk and make your way toward the door, even in the absence of the visual cues you would normally rely on. As you did so, your mental image of your location would move along with
73
74 PLACE CELLS AND SPATIAL CONTEXT you, so you would have a sense of the area near the door getting closer. Data from hippocampal place cells suggest that these cells would also follow along, with the appropriate place cells for each part of the trajectory toward the door being called up, each in the appropriate spot. Models have been developed to explain how the hippocampal cells could perform this path integration (e.g., McNaughton et al., 1996). In these models, it is assumed that the hippocampal cells are all rigidly linked together, along with cells that signal translational motion and directional heading (see below). Networks like this can be constructed so that any one movement from a given start location (as indicated by the place cell population activity) will necessarily ‘‘call up’’ the next appropriate place cells, based on the direction and speed of movement. As is well known, the idea that the hippocampus functions primarily as a cognitive map has received substantial challenges through the years since the publication of O’Keefe and Nadel’s (1978) original treatise. Two of these challenges are relevant here. First, numerous studies have revealed that hippocampal pyramidal cells will readily show firing correlates other than those that could be considered place fields. Hippocampal recordings during various learning tasks demonstrate that the cells will respond selectively to conditioned stimuli or to various combinations of sensory stimuli, context, and place (Thompson et al., 1980; Eichenbaum et al., 1986; Wiener et al., 1989; Weiss et al., 1996; Wood et al., 1999). This finding suggests that although the hippocampus does reflect information about the animal’s current location within an environment, it is not strictly dedicated to this task. Related to this finding is the fact that hippocampal place cells are remarkably sensitive to environmental context. Early work demonstrated that when a hippocampal cell is recorded in more than one environment, it will likely have entirely different characteristics in each one. For example, a cell with a robust place field in one recording chamber may be entirely silent in a different arena, or it may have a field that is entirely different in size and shape (O’Keefe and Conway, 1978; Kubie and Ranck, 1982; Muller and Kubie, 1987; Thompson and Best, 1989). This finding helped support the idea that the hippocampus plays a role in coding environmental context. Also, as reviewed in many of the chapters in this volume and elsewhere (e.g., Hippocampus, Vol. 16, No. 9, 2006, Special Issue: Place Cells and Episodic Memory) the hippocampal place cells are sensitive to additional environmental variables, other than just the recording chamber itself. These include task demands, trajectory taken through the environment, the environment and/ or entry point experienced immediately prior to arrival
in the environment, and presentation of conditioned stimuli. Vicissitudes in any of these aspects of a situation can cause the hippocampal cells to entirely rearrange their spatial firing characteristics. This extreme sensitivity to contextual variables is reminiscent of the data collected by Eichenbaum and others showing that hippocampal cells are best characterized as responding to various combinations of environmental stimuli, contexts, and place. That is, as hippocampal place cells rearrange themselves multiple times in a given environment, it is difficult to conceive of them as forming a coherent map. Rather, it appears that they respond to various combinations of environmental variables, including the variable consisting of current spatial location. It has been argued that this highly changeable aspect of the hippocampal place fields means that the hippocampal cells are not good candidates to constitute the path integration component of the cognitive map (Sharp, 1999a). Although a wealth of data have demonstrated beyond doubt that the hippocampal cells show evidence of path integration, it now seems that they are likely just reflecting the influence of a path integration system located elsewhere in the brain, rather than performing the integration calculations themselves. This point will be elaborated further below. The second challenge to the idea of the hippocampus as a cognitive map comes from work in which researchers have looked in other parts of the hippocampal formation and beyond, and have found additional robust, intriguing spatial signals. Indeed, as described below, much of the rat limbic system has been shown to contain robust spatial cells. Thus, in retrospect, it was just an historical accident that the hippocampus proper was so strongly identified as the cognitive map. The place-specific firing fields and path integration abilities discovered by O’Keefe and colleagues were truly remarkable, and these basic observations and accompanying theoretical ideas have stood the test of time. However, it now seems that the hippocampal place cells were only a hint of what was to come and that the idea of a cognitive map must be at least expanded, if not entirely moved, to other, nearby, limbic system areas.
THE LIMBIC SYSTEM CONTAINS SEVERAL DISTINCT CELL TYPES THAT SHOW ENVIRONMENTALLY INVARIANT SPATIAL FIRING PROPERTIES Head Direction Cells Head direction cells were initially discovered in the postsubicular region of the hippocampal formation (Ranck, 1984; Taube et al., 1990b) and have since been
SPATIAL CODING AND EPISODIC MEMORY
recorded in numerous additional limbic system regions, including the mammillary bodies, anterior thalamic nucleus, dorsolateral thalamic nucleus, retrosplenial cortex, anterior hippocampus, and entorhinal cortex (Sharp et al., 2001; Sargolini et al., 2006). These cells have a spatial firing pattern complimentary to that of the place cells in that they fire throughout all regions of the recording chamber (i.e., they are not place specific), but do so only when the rat is facing one particular direction. Each of these head direction cells fires in one particular directional heading (Fig. 5–1A), usually covering an approximately 908 range centered on that direction. Within
75
each cell’s 908 preferred range, there is a peak firing rate in the middle of this range and a symmetrical falloff on either side of this peak, so that the cell makes a Gaussian, or triangular, function of firing rate versus direction. These cells are context invariant in two ways. First, any cell that acts as a head direction cell in one environment will also do so in any other environment in which the animal is tested (Taube et al., 1990a). It will also have approximately the same peak firing rate and range in each different environment. Second, as illustrated in Figure 5–2A, if two or more head direction cells are recorded at the same time, they will maintain
Figure 5–1. Cartooned examples illustrating each of the four cell types discussed in the text. Each circle represents an overhead view of the type of cylindrical chamber often used during recordings of these cells. Within each circle is a representation of the spatial firing properties of one particular cell. During recording sessions the rats continuously travel through the chamber, constantly searching for tiny food pellets that are dropped down throughout the session. A. Representation of three different head direction cells. The set of three, parallel arrows for each cell represent the preferred direction for that cell. They are meant to show that each head direction cell fires throughout the area of the cylinder, but only when the animal faces in the indicated direction. B. Examples of three different entorhinal grid cells. Within any one rat, the grid cells vary in terms of their phase, angular orientation, and grid spacing. C. Examples of three representative subicular place cells. Darker shading indicates higher relative firing rates. D. Examples of three representative hippocampal place cells.
76 PLACE CELLS AND SPATIAL CONTEXT the same directional relationship to each other in each different environment, although they may all shift in relation to earth-centered coordinates (Yoganarasimha et al., 2006). Thus, it appears that these head direction cells are rigidly linked together to form a network in which the momentary directional heading of the animal is indicated by the momentary population firing rate vector within the network.
Figure 5–2. Cartooned representations for each of the cell types discussed in the text. Each circle represents the spatial firing patterns of two or more cells recorded at the same time in a small cylinder, with each cell shown using a different color. Each square shows the firing patterns when the same set of cells is recorded in an environment (large square) that differs from the cylinder in both size and shape. A. Each arrow represents the preferred direction for a different head direction cell. These cells maintain the same directional headings, relative to one another, in the two chambers. (In this particular illustration, the cells also maintain the same directional heading relative to the larger world, but, as described in the text, this is not always the case.) B. Examples of three representative entorhinal grid cells. Each cell forms a hexagonal grid-like framework of locational hot spots. In the example shown here, the three cells share a similar grid spacing and angular orientation, as is the case for cells recorded within close proximity to one another. C. Illustration of two representative subicular place cells. Both cells expand and reshape to fit the larger square, and they maintain a similar spatial relationship to one another. D. Representation of several hippocampal place cells. There is a different subset of cells active in each of the two chambers. Only the cell shown in blue has a place field in both chambers, and the size, shape, and relative location of the field for this cell is different for the cylinder than for the square.
This linking together is thought to be accomplished through a process of path integration similar to that described above for the place cells (Skaggs et al., 1995; Redish et al., 1996; Song and Wang, 2005). In this case, however, the relevant movement information consists of that related to angular movements of the rat’s head, such as might be provided by the vestibular system or by motor command signals. Neural network models have been developed that link the
SPATIAL CODING AND EPISODIC MEMORY
populations of head direction and angular head velocity cells together so that any given combination of current heading and angular motion will lead to the activation of the head direction cells that signal the resulting, appropriate heading. For example, activity in head direction cells that are active when the rat faces ‘‘north,’’ when paired with activity in angular velocity cells that signal a 908 clockwise head movement, will automatically turn off those same ‘‘north’’ head direction cells and turn on the ‘‘east’’ head direction cells.
Grid Cells A recent finding that has generated tremendous excitement in the field has been the discovery of grid cells in the medial entorhinal cortex (Fyhn et al., 2004; Hafting et al., 2005; Sargolini et al., 2006). These cells are similar to the hippocampal place cells in that they fire in relation to the animal’s current location within the recording environment. However, these cells have multiple regions (spots) of high firing and, remarkably, these hot spots form a hexagonal grid-like structure (Fig. 5–1B). Grid cells recorded within any one, local region of the entorhinal cortex share a common angular orientation and spacing between the hot spots, as illustrated by the set of three grid cells depicted in Figure 5–2B. This means that sets of local cells form an interleaved hexagonal lattice, like that shown in Figure 5–2B. However, as illustrated in Figure 5–1B, cells from different medial entorhinal regions have different orientations and grid spacing. These cells are also environmentally invariant. Any cell that has been identified as a grid cell in one environment will also behave as a grid cell in any other environment, and will show a similar grid spacing (McNaughton et al., 2006). Also, any two or more grid cells recorded simultaneously will show the same phase and angular relationships relative to each other in any environment (McNaughton et al., 2006). As illustrated in Figure 5–2B, this means that when a cell goes from a smaller to a larger environment it will simply use a larger set of spots to cover the additional area. Thus, similar to the head direction cells discussed above, the grid cells seem to form a network in which the relations between cells are rigidly maintained. This context-independent, lattice-like structure suggests that these cells are also linked together through a path integration process. Models for how this could work have already been developed (Fuhs and Touretzky, 2006; McNaughton et al., 2006). These models incorporate the head direction cells described above, along with cells that simply indicate translational mo-
77
tion. In these models, the activity of the head direction and movement cells serves to shift the activity pattern through the lattice formed by the interleaved grid-cell firing fields. For example, if the animal starts out on one of the blue spots in Figure 5–2B, a translational movement will serve to turn off the ‘‘blue’’ cell, and activate either a ‘‘pink’’ or a ‘‘yellow’’ cell. The choice between these latter two cells is determined by which head direction cells are currently active. Note that these models postulate the existence of cells with combined grid and directional properties, so that the cell firing rate in each hot spot is also influenced by the animal’s current directional heading. Cells of this type have recently been reported (Sargolini et al., 2006).
Subicular Place Cells The subiculum also has cells that show place (location)-related firing patterns and often have multiple hot spots (Sharp and Green, 1994). Cells of this type, illustrated in Figure 5–1C, are often tonically active, thus they fire throughout the recording chamber. Nonetheless, they often have one or more regions of consistent, relatively high firing. These cells are similar to the hippocampal place cells in that they have idiosyncratic place fields that vary in size and shape across the cell population, and in some cases they show only a single firing field within a given environment. Interestingly, however, recent analyses have shown that in the relatively rare cases in which subicular cells have at least three hot spots, the arrangement of these fields shows a clear grid pattern, similar to that shown by the entorhinal grid cells (Wilent et al., unpublished observations). Thus, the overall subicular cell population exhibits properties similar to both the hippocampal and entorhinal spatial cells. Of relevance here, however, is that subicular cells show context invariance, unlike hippocampal cells (Sharp, 1997, 1999b). In addition, surprisingly, the manner in which they exhibit this context invariance is strikingly different from that shown by the entorhinal grid cells, as illustrated in Figure 5–2C. When subicular cells are recorded in two environments that differ in visual stimulus properties, size, and shape, each subicular cell shows a spatial firing pattern with the same relative spatial firing fields in each chamber. That is, the subicular cells reshape and resize to fit the same general pattern onto each environment. This is quite different from the manner in which grid cells expand to fill a larger environment (Fig. 5–2B). Specifically, the grid cells ‘‘add on’’ extra firing fields, continuing on with the same hexagonal pattern, to fill up the added space. In contrast, the subicular cells stretch the same basic pattern, rather than add on extra spots.
78 PLACE CELLS AND SPATIAL CONTEXT It should also be noted that an early study of entorhinal cells showed a similar phenomenon (Quirk et al., 1992). The cells in this study did not show an apparent grid-like pattern but, rather, showed idiosyncratic, distributed fields like those shown here for the subiculum. When the cells were recorded first in a cylinder and then in a square of the same area, they showed a similar pattern in each. However, these cells were not tested in environments of different size. Thus, it is not clear whether these cells would expand, as for the subicular cells described here, or whether they would add on additional fields, perhaps ultimately revealing a grid-like pattern, given a large enough environment. Because subicular cells show context-invariant spatial firing properties and appear to be linked together, in that the spatial firing fields of different cells retain the same relative positions, it has been suggested that the subiculum is also a candidate region for translational path integration (Sharp, 1999a). In fact, it appears that both the subiculum and the entorhinal cortex could provide ideal circuitry for translational path integration. One possibility is that the two regions could work together to perform the neural calculations necessary for the integration process, since they are strongly interconnected anatomically. The postulated integration mechanism would have to be quite different for the two regions, however. Specifically, the ‘‘gain’’ on the integration mechanism (that is, the rate at which distance traveled cause shifting of the place cell population firing vector) would have to be constant for the entorhinal grid cells, but context dependent for the subicular cells. To see this, recall that the subicular patterns shrink and grow to fit different-sized environments. If these spatial firing patterns are based on path integration, this means that in a relatively large environment, each step the animal takes would need to update the spatial pattern more slowly than when the animal is in a smaller environment. This ‘‘lowering of the gain’’ would permit the pattern to be stretched all the way across the larger environment. In contrast, path integration for the entorhinal grid cells would require a constant gain for the integration process, since the distance between hot spots for any one cell does not change along with the size of the environment. These considerations suggest that the subiculum and entorhinal cortex may contain at least two different translational path integration mechanisms: one to stretch the same pattern to fit each environment, and another to keep a fixed grid pattern throughout each environment. To resolve these issues it will be necessary to examine more closely the path integration properties of both grid and non-grid cells in the entorhinal and subicular regions.
Cells with Place and/or Directional Properties Are Located throughout the Hippocampal Formation It should be noted that many of the locational cell types described above also often have a directional component, so the firing rate within any one spot is influenced by direction as well as by location. This directional bias has been observed for both the entorhinal grid cells (Sargolini et al., 2006) and the subicular place cells (Sharp and Green, 1994). It is often assumed that this directional bias is superimposed on the locationspecific patterns by the influence of local head direction cells. Cells indicating place by direction have also been documented in the pre-, post-, and parasubiculum (Taube, 1995; Sharp, 1996; Cacucci et al., 2004). In most cases, cells in these regions have not been as thoroughly investigated as the four cell types discussed above. A full description of these cells is beyond the scope of this review.
THE HIPPOCAMPUS CONTAINS CELLS THAT ARE CONTEXT SPECIFIC Cartooned examples of hippocampal place cells are shown in Figures 5–1D and 5–2D for comparison. As can be seen, the subset of cells that exhibit a place field changes from one environment to another. In addition, any two cells that do happen to fire in both environments are likely to show a different shape, size, and relative location in the two environments (Muller and Kubie, 1987).
THEORETICAL IMPLEMENTATION OF PATH INTEGRATION HAS SHIFTED FROM THE HIPPOCAMPUS PROPER TO OTHER LIMBIC SYSTEM REGIONS As described in the introduction, the first demonstration of path integration at the level of the single cell was provided during early work on hippocampal place cells (O’Keefe, 1976; O’Keefe and Conway, 1978). This work provided the initial demonstrations that cells could maintain their location-specific firing patterns on the basis of information about the animal’s own direction and speed of movement through the environment. Because this path integration ability was first documented in the hippocampus, many of the early models of path integration were developed with the idea that the hippocampal place cells themselves were part
SPATIAL CODING AND EPISODIC MEMORY
of the fundamental path integration network (e.g., McNaughton et al., 1996). However, with the discovery of the subicular place cells and entorhinal grid cells described above, this idea has shifted. Indeed, as described above, the most recent models of translational path integration have used grid cells (along with head direction cells) as the basis for the path integration network. The main motivation for this shift from hippocampal cells to entorhinal grid cells is the context invariance of the entorhinal cells. As already described, any model of path integration must be put together using a set of movement, direction, and locational cells that are rigidly and permanently hooked up together in such a way that any particular combination of movement, direction, and starting location leads to a consistent prediction of the next location. It does not make sense to think that hippocampal place cells could play the role of location-specific cells in these models because of their context specificity. Any one hippocampal cell may or may not fire in any one environment, and when it does, it will likely not retain a constant spatial relationship to other hippocampal cells. Thus, it is difficult to conceive of how these cells could provide the basis for a path integration network. In contrast, the grid cells are ideally suited to this task. The repeating pattern shown by the interwoven grid cells in Figure 5–2B implies that the path integration task could be accomplished by a relatively small set of local cells that simply repeat the same integration process to lay out the grid pattern as many times as necessary to fill in the environment. As an example, for the cells in Figure 5–2B, the animal can travel along any straight line of cells by simply calling up an ordered sequence of cell activity, consisting of either yellow, blue, red, yellow, blue, red, etc., or yellow, red, blue, yellow, red, blue, etc. The choice of sequence depends on current directional heading. This idea implies that any behavioral system needing to integrate position over long distances would need to somehow count up the number of these grid-cell sequence iterations through which the animal has passed.
IF THE HIPPOCAMPUS DOES NOT CONSTITUTE THE COGNITIVE MAP, WHAT IS THE BEHAVIORAL SIGNIFICANCE OF THE HIPPOCAMPAL ‘‘PLACE CELL’’ FIRING FIELDS? The entorhinal grid cells have been found to exist in layers II and III of the entorhinal cortex (Fyhn et al., 2004), thus these cells likely provide at least part of the entorhinal projection onto the hippocampus. Ac-
79
cordingly, it seems likely that the hippocampal place fields, along with their observed path integration capabilities, are in part a result of convergent inputs from grid cells. Indeed, recent work has demonstrated that realistic hippocampal place fields could be generated on the basis of Moire interference patterns produced by multiple entorhinal grid cell firing patterns (Blair et al., 2007). However, grid cell input cannot be the sole determinant of the hippocampal cell firing patterns, since the entorhinal grid cells are context invariant, while hippocampal cells are easily influenced by contextual variables. Thus to explain the context-specific place fields of hippocampal cells, we must consider what would happen for a set of cells that receive both convergent input from a set of entorhinal grid cells and input from cells coding for additional aspects of the situation, such as particular sensory stimuli, particular trajectories or journeys through the environment, and task demands. Imagine a hippocampal cell that receives input from a set of N grid cells. Blair et al. (2007) have explained that for any two of these cells that differ from each other in angular orientation and/or grid spacing (see examples in Fig. 5–1B), there will be periodically spaced regions in which the spots of the two grids overlap (the Moire interference pattern). It follows, then, that regions within the environment where subsets of the grid cells overlap would provide relatively high levels of excitatory input to the recipient hippocampal cell, thus biasing that cell toward having a place field at that location. However, to explain the context specificity of the hippocampal cells, it must be assumed that these regions of overlapping grid cell activity are not alone sufficient to drive the cell to fire. Rather, it must be assumed that hippocampal cells also require input from the postulated nonspatial, context-specific cells. Thus, any place field that arises in that particular environment would be an idiosyncratic combination of the pattern of grid cell inputs that are a constant for that cell, along with environment-specific aspects of the current context. In environments that are fixed (have no alterations of stimuli, task, trajectory, etc.) this will result in stable hippocampal place fields. However, any change in context could result in a remapping of the hippocampal cells, as is often observed. In the extreme case, we could imagine an event (episode) that happens just once within a given environment. Following the above reasoning, this could result in an instantaneous firing pattern within the hippocampus that would be completely unique to that particular environment, the location (place) of the animal within the environment during the event, and
80 PLACE CELLS AND SPATIAL CONTEXT stimulus aspects of that event. Thus, any one episode would induce an entirely novel, instantaneous pattern within the hippocampal population. This novel pattern would constitute the hippocampal representation of that episode. According to this idea, the place fields of hippocampal place cells are simply a reflection of the postulated grid cell path integration system, and this tendency toward place specificity within a given environment would contribute part of the location representation of the hippocampal snapshot taken during formation of an episodic memory.
References Blair HT, Welday A, Zhang K (2007) Scale-invariant memory representations emerge from Moire interference between grid fields that produce theta oscillations: a computational model. J Neurosci 27:3211–3229. Cacucci F, Lever C, Wills TJ, Burgess N, O’Keefe J (2004) Theta-modulated place-by-direction cells in the hippocampal formation in the rat. J Neurosci 24: 8265–8277. Eichenbaum H, Kuperstein M, Fagan A, Nagode J (1986) Cue-sampling and goal approach correlates of hippocampal unit activity in rats performing an odor discrimination task. J Neurosci 7:716–732. Fuhs MC, Touretzky DS (2006) A spin glass model of path integration in rat medial entorhinal cortex. J Neurosci 26:4266–4276. Fyhn M, Molden S, Witter MP, Moser EI, Moser M-B (2004) Spatial representation in the entorhinal cortex. Science 305:1258–1264. Hafting T, Fyhn M, Molden S, Moser M-B, Moser EI (2005) Microstructure of a spatial map in the entorhinal cortex. Nature 436:801–806. Kubie JL, Ranck Jr JB (1982) Tonic and phasic firing of rat hippocampal complex spike cells in three different situations: context and place. Adv Behav Biol 26:89– 98. McNaughton BL, Barnes CA, Gerrard J, Gothard K, Jung MW, Knierim JJ, Kudrimoti H, Qin Y, Skaggs WE, Suster M, Weaver KL (1996) Deciphering the hippocampal polyglot: the hippocampus as a path integration system. J Exp Biol 199:173–185. McNaughton BL, Battaglia FP, Jensen O, Moser EI, Moser M-B (2006) Path integration and the neural basis of the cognitive map. Nat Rev Neurosci 7:663–678. Muller RU, Kubie JL (1987) The effects of changes in the environment on the spatial firing of hippocampal complex spike cells. J Neurosci 7:1951–1986. O’Keefe J (1976) Place units in the hippocampus of the freely moving rat. Exp Neurol 51:78–109. O’Keefe J, Conway DH (1978) Hippocampal place cells in the freely moving rat: why they fire where they fire. Exp Brain Res 31:573–590.
O’Keefe J, Dostrovsky J (1971) The hippocampus as a spatial map. Preliminary evidence from unit activity in the freely-moving rat. Brain Res 34:171–175. O’Keefe J, Nadel L (1978) The Hipppocampus as a Cognitive Map. New York: Oxford University Press. O’Keefe J, Speakman A (1887) Single unit activity in the rat hippocampus during a spatial memory task. Exp Brain Res 68:1–27. Quirk GJ, Muller RU, Kubie JL (1990) The firing of hippocampal place cells in the dark depends on the rat’s recent experience. J Neurosci 10:2008–2017. Quirk GJ, Muller RU, Kubie JL, Ranck Jr JB (1992) The positional firing properties of medial entorhinal neurons: description and comparison with hippocampal place cells. J Neurosci 12:1945–1963. Ranck Jr JB (1984) Head-direction cells in the deep layers of the dorsal presubiculum in freely moving rats. Abstr Soc Neurosci 10:599. Redish AD, Elga AN, Touretzky DS (1996) A coupled attractor model of the rodent head direction system. Network 7:671–685. Sargolini F, Fyhn M, Hafting T, McNaughton BL, Witter MP, Moser M-B, Moser EI (2006) Conjunctive representation of position, direction, and velocity in entorhinal cortex. Science 312:758–762. Sharp PE (1996) Multiple spatial/behavioral correlates for cells in the rat postsubiculum: Multiple regression analysis and comparison to other hippocampal areas. Cereb Cortex 6:238–259. Sharp PE (1997) Subicular cells generate similar spatial firing patterns in two geometrically and visually distinctive environments; comparison with hippocampal place cells. Behav Brain Res 85:71–92. Sharp PE (1999a) Complimentary roles for hippocampal versus subicular/entorhinal place cells in coding place, context, and events. Hippocampus 9:432– 443. Sharp PE (1999b) Subicular place cells expand/contract their spatial firing patterns to fit the size of the environment in an open field, but not in the presence of barriers: Comparison with hippocampal place cells. Behav Neurosci 113:643–662. Sharp PE, Blair HT, Cho J (2001) The anatomical and computational basis of the rat head-direction cell signal. Trends Neurosci 24:289–294. Sharp PE, Green C (1994) Spatial correlates of firing patterns of single cells in the subiculum of the freely moving rat. J Neurosci 14:2339–2356. Skaggs WE, Knierim JJ, Kudrimoti H, McNaughton BL (1995) A model of the neural basis of the rat’s sense of direction. In: Advances in Neural Information Processing Systems (Tesauro G, Touretzky D, Leen T, eds.), pp 173–180. Cambridge, MA: MIT Press. Song PC, Wang XJ (2005) Angular path integration by moving ‘‘hill of activity’’: a spiking neuron model
SPATIAL CODING AND EPISODIC MEMORY
without recurrent excitation of the head-direction system. J Neurosci 25:1002–1014. Taube JS (1995) Place cells recorded in the parasubiculum of freely moving rats. Hippocampus 5: 569–583. Taube JS, Muller RU, Ranck Jr JB (1990a) Head direction cells recorded from the postsubiculum in freely moving rats. II. Effects of environmental manipulations. J Neurosci 10:436–447. Taube JS, Muller RU, Ranck Jr JB (1990b) Head-direction cells recorded from the postsubiculum in freely moving rats. I. Description and quantitative analysis. J Neurosci 10:420–435. Thompson RF, Berger TW, Berry SD, Hoehler FK, Kettner RE, Weisz DJ (1980) Hippocampal substrate of classical conditioning. Physiol Psychol 8: 262–279.
81
Thompson LT, Best PJ (1989) Place cells and silent cells in the hippocampus of freely behaving rats. J Neurosci 9:2382–2390. Weiss C, Kronforst-Collins MA, Disterhoft JF (1996) Activity of hippocampal pyramidal neurons during trace eyeblink conditioning. Hippocampus 6:198–209. Wiener SI, Paul CA, Eichenbaum H (1989) Spatial and behavioral correlates of hippocampal neuronal activity. J Neurosci 9:2737–2763. Wood ER, Dudchenko PA, Eichenbaum H (1999) The global record of memory in hippocampal neuronal activity. Nature 397:613–616. Yoganarasimha D, Yu X, Knierim JJ (2006) Head direction cell representations maintain internal coherence during conflicting proximal and distal cue rotations: comparison with hippocampal place cells. J Neurosci 26:622–631.
6 The Roles of Hippocampal Subfields in Processing Spatial Contexts of Events: Neurophysiological and Behavioral Analyses INAH LEE, RAYMOND P. KESNER, AND JAMES J. KNIERIM
2001; Kesner et al., 2004; Lee et al., 2005ab; Ferbinteanu et al., 2006; Kesner and Hopkins, 2006). Why is the hippocampus necessary for processing events occurring in space? Several theories (for example, relational theory, contextual theory, configural theory, and cognitive map theory) have commonly emphasized the role of the hippocampus in associating or relating individual cues in an environment for constructing a unitary, ‘‘contextual’’ representation of the spatial environment (O’Keefe and Nadel, 1978; Nadel et al., 1985; Sutherland et al., 1989; Good and Honey, 1991; O’Reilly and McClelland, 1994; McClelland and Goddard, 1996; Cohen et al., 1997; Shapiro et al., 1997; Fanselow, 2000; Rudy and O’Reilly, 2001; Burgess et al., 2002). The link between the hippocampal role in memory and contextual information processing was originally proposed by Hirsh (1974), who suggested that the hippocampus was important in ‘‘contextual retrieval’’ of memory. He defined contextual retrieval as ‘‘retrieval of an item of stored information initiated by a cue which refers to but is not necessarily described within the information that is retrieved’’ (p. 422). In addition to physical and external stimuli, Hirsh also included motivational state as an internal contextual cue. Other contextual memory theories for the hippocampus (Fanselow, 2000; Rudy and O’Reilly, 2001) have argued that a significant event (e.g., foot shock) is associated with a contextual representation of the environment in which the event takes place. Despite the frequent usage in the literature, however, the term context has been loosely defined. In this chapter, discussion
Since the historic case report of the patient H.M. (Scoville and Milner, 1957), neuropsychological and animal studies have focused on revealing the function of the hippocampus and adjacent cortical regions in learning and memory (Hirsh, 1974; Mishkin, 1978; O’Keefe and Nadel, 1978; Olton et al., 1978; Morris et al., 1982; Kesner, 1985; Zola-Morgan et al., 1986; McNaughton and Morris, 1987; Kesner and Beers, 1988; Squire, 1992; Alvarez et al., 1995; VarghaKhadem et al., 1997; Tulving and Markowitsch, 1998; Eichenbaum, 2000). Although the debate is still ongoing, most researchers now agree that the hippocampus is essential for the formation and retrieval of memories for discrete events occurring at certain places and times. This type of memory has been called ‘‘episodic memory’’ (Tulving, 1972, 2002), and hippocampal amnesic patients, when compared to normal subjects, are significantly impaired in remembering the details of episodic events (Zola-Morgan et al., 1986; Rempel-Clower et al., 1996; Vargha-Khadem et al., 1997; Mishkin et al., 1998; Spiers et al., 2001). Although it has been debated whether episodic memory is unique to humans (Morris, 2001; Tulving, 2002; Clayton et al., 2003), nonhuman animals have also been tested for their abilities to remember events associated with particular places and/or times and it has been recognized that the hippocampus plays a key role in those tasks (Kim and Fanselow, 1992; Dusek and Eichenbaum, 1997; Clayton and Dickinson, 1998; Griffiths et al., 1999; Suzuki and Clayton, 2000; Aggleton and Pearce, 2001; Gilbert et al., 2001; Morris, 82
HIPPOCAMPAL SUBFIELDS AND SPATIAL CONTEXT OF EVENTS
will be restricted to ‘‘spatial context’’ (Tolman, 1948; Fanselow, 2000; Maren and Holt, 2000; Burgess et al., 2002; Nadel and Hardt, 2004; Rudy et al., 2004; Anderson et al., 2006), that is, the configuration of background landmarks that define a particular environment and allow an animal to recognize that environment. It is assumed in this chapter that the contextual representation of the environment is associated in the hippocampus with the events that occur in that environment, and this representation may play an important role in retrieving the memories of events later. In this regard, one of the important characteristics of spatial context is that it serves as a static background, or ‘‘state’’ (Burgess et al., 2001), that does not change as various events occur in it. In humans, an episodic event tends to be remembered with its spatial context (as well as its temporal context). For example, when a human subject navigates a virtual town in a three-dimensional, virtualreality environment and receives an object from a virtual character, it can be considered as an episodic event. By asking the subject to remember the spatial context associated with the event later, Burgess et al. (2001b) showed that the activity in the hippocampus was increased when retrieving the event-related information with the background spatial context as a cue. Human subjects involved in other types of event memory tasks exhibit a significantly higher level of activity in the hippocampus as well as its associated structures in the medial temporal lobe (Lepage et al., 1998; Eldridge et al., 2000, 2005; Small et al., 2001; Davachi et al., 2003; Dobbins et al., 2003; Zeineh et al., 2003; Addis et al., 2004). Furthermore, when patients who had suffered bilateral damage to the hippocampus at early ages were examined, they were mostly impaired on tasks requiring the processing of spatial and episodic information (Vargha-Khadem et al., 1997); the patients were unable to process their surrounding environments to figure out their locations during navigation and were unable to remember the personal events in their daily lives. These patients showed abnormally small hippocampi in both hemispheres, based on volumetric measurements of imaging data. These results demonstrate a fairly selective role of the hippocampus when the information processing involves forming the association between a certain event and the spatial context in which the event occurred, as well as retrieving the event memory when cued by the context. Animal models of hippocampal amnesia provide more direct evidence for the involvement of the hippocampus in spatial contextual information processing. A contextual fear-conditioning paradigm in rodents, for example, has served as an important paradigm (Kim and Fanselow, 1992; Phillips and
83
LeDoux, 1992; Rudy, 1993; Corcoran and Maren, 2001; Lee and Kesner, 2004a). The paradigm involves placing animals in a cue-controlled environment to condition a certain behavioral response, such as freezing behavior (Blanchard and Blanchard, 1969), to the contextual representation of the background environment. After the brief exploration period, a tone stimulus is presented briefly and an aversive event such as a mild foot shock is experienced at the termination of the tone. The animal associates the noxious event not only with the conditioned stimulus (i.e., tone) but also with the spatial context in which the aversive event has occurred. This is clearly demonstrated when the animal is placed again in the environment at a later time (e.g., 24 hr after the aversive event), as the animal shows freezing behavior even though the conditioned stimulus and the foot shock are not presented. Importantly, the animal’s freezing behavior is only expressed in the original environment in which the aversive event had taken place, but not in a different spatial environment (Kim and Fanselow, 1992). Freezing behavior is a highly adaptive form of behavior in species such as rodents to avoid being detected by the movement-sensitive visual systems of their predators. Therefore, rapid retrieval of an event memory (e.g., foot shock) that has been selectively associated with the contextual representation of a particular environment will be beneficial to the animal to quickly activate appropriate defensive systems (selectively in a threatening environment only) and thus increase the chance of survival. Ablating or inactivating the hippocampus eliminates the capability of the animals to discriminate different spatial contexts (Kim and Fanselow, 1992; Phillips and LeDoux, 1992; Maren and Fanselow, 1997; Maren and Holt, 2004), minimizing the difference between the freezing responses when tested in different contexts. Although the phenomenological evidence described above suggests that the hippocampus is indeed essential for forming and retrieving contextual representations (and associated events), the exact mechanisms through which individual neural circuits (i.e., subfields) in the hippocampus perform this task are largely unknown. Because most theories have focused on the role of the hippocampus as a unitary structure, the majority of behavioral experiments have manipulated the hippocampus as a whole (e.g., complete hippocampal lesions) (Morris et al., 1982; Parkinson et al., 1988; Kim and Fanselow, 1992; Angeli et al., 1993; Moser et al., 1993; Fortin et al., 2002; Gilbert and Kesner, 2004; Kennedy and Shapiro, 2004; Maren and Holt, 2004; Forwood et al., 2005; Kirwan et al., 2005; Banta Lavenex et al., 2006). The paucity of experimental evidence testing the functions of detailed hippocampal circuits stems from the technical difficulties in
84 PLACE CELLS AND SPATIAL CONTEXT manipulating or physiologically recording from anatomically distinct regions in the hippocampus. In contrast, computational models of the detailed functions of the hippocampal subfields in memory have been available for more than 30 years (Marr, 1971; McNaughton and Morris, 1987; Treves and Rolls, 1992; O’Reilly and McClelland, 1994; Hasselmo et al., 1995; Levy, 1996; Samsonovich and McNaughton, 1997; Wallenstein and Hasselmo, 1997; Lisman et al., 2005; Rolls and Kesner, 2006). As will be reviewed in this chapter, improvements in lesion and inactivation techniques and ensemble recording techniques have begun to remedy this situation. Inspired by a recent surge of experimental evidence demonstrating the differential contributions of the hippocampal subfields in memory (Kesner et al., 2000, 2004, 2005; Gilbert et al., 2001; Lee and Kesner, 2002, 2003a, 2004a,b; Gilbert and Kesner, 2003, 2006; Lee et al., 2004a,b, 2005a,b; Leutgeb et al., 2004, 2005, 2006; Vazdarjanova and Guzowski, 2004; Jerman et al., 2006), this chapter will provide a selective review of recent experimental evidence from animal studies that test some of the critical hypotheses proposed by the theoretical models with respect to the hippocampal mechanisms of contextual information processing. The chapter is organized as follows. We will first selectively review some of the influential computational models and the contextual information processes that have been considered critical by those models when processing event memory in the hippocampus. Experimental evidence from both recent perturbation studies and physiological studies that have tested the predictions from the computational models will then be reviewed. The goal of this chapter is to establish the link, based on experimental evidence, between some of the key computational principles of the hippocampal subfields and their cognitive functions for episodic event memory.
COMPUTATIONAL HYPOTHESES Although there is a consensus that the hippocampus is critical for contextual information processing for event memory (Sutherland et al., 1989; O’Reilly and McClelland, 1994; McClelland and Goddard, 1996; Cohen et al., 1997; Shapiro et al., 1997; Fanselow, 2000; Rudy and O’Reilly, 2001; Burgess et al., 2002), the underlying neural mechanisms are not fully understood. What processes in the hippocampus are necessary to form spatial representations of external cues in the environment? In which hippocampal networks are each of these processes implemented? What are the network dynamics between the hippocampal subfields responsible for forming and retrieving con-
textual information as the animal performs a memory task? With our current knowledge of the hippocampus, it is difficult to provide substantial answers to these critical questions, mainly because the differential roles of the hippocampal subfields remain largely untested. This is in contrast to the detailed anatomical, cellular, and physiological knowledge that has been available for many decades (Ramon Y Cahal, 1911; Lorente de No, 1934; Vanderwolf, 1969; Andersen et al., 1971; Marr, 1971; O’Keefe and Dostrovsky, 1971; Bliss and Lomo, 1973). Therefore, we argue that the focus of experimental efforts needs to be shifted to empirically testing the functional significance of the different anatomical circuits in the hippocampus in memory (Kesner and Hopkins, 2006; Rolls and Kesner, 2006). Specific cognitive processes necessary for embodying episodic memory need to be identified and the principles by which those processes are implemented in different circuits in the hippocampus need to be tested. The hippocampus is composed of different subfields (e.g., CA1, CA3, and dentate gyrus) that are distinguished from each other on the basis of anatomical and physiological characteristics (Amaral and Witter, 1989, 1995; Li et al., 1994). Furthermore, the subfields are interconnected with each other and with other cortical and subcortical structures in a systematic fashion. Briefly, the hippocampal subfields receive their major sensory inputs from the entorhinal cortex in parallel via the perforant path. Within the hippocampus, the dentate gyrus sends its information to the CA3 subfield and CA3 sends its outputs to the CA1 subfield. Therefore, in addition to the parallel input streams from the entorhinal cortex to different subfields, a serial information-processing stream exists in the hippocampus from the dentate gyrus to CA1 via CA3. Understanding the dynamic information processing occurring in these subfields in support of contextual information processing can potentially provide great insights into the cognitive function of the hippocampus. Testing of the differential roles of the hippocampal subfields requires the construction of detailed hypotheses for the subfield functions relevant to the cognitive processes underlying hippocampus-dependent memory. One of the investigative approaches in line with such a perspective is computational modeling that involves biologically plausible, mathematical descriptions of the inputs and outputs of the hippocampal network, as well as the information processing that occurs in the different circuits within the hippocampus. Here, we will identify two cognitive processes (i.e., pattern completion and pattern separation) critically involved in processing contextual information when event memories are formed and retrieved. We
HIPPOCAMPAL SUBFIELDS AND SPATIAL CONTEXT OF EVENTS
will also selectively review some of the computational models that have influenced the recent experimental investigations on the functional dissociation of the hippocampal subfields in contextual information processing. Since Hirsh’s (1974) proposition that event information can be retrieved on the basis of the contextual information originally associated with the event, the formation and retrieval of contextual memory have been tested for many years in laboratory settings (Kim and Fanselow, 1992; Mizumori et al., 1999; Rudy and O’Reilly, 1999; Burgess et al., 2002; Davachi et al., 2003; Jeffery et al., 2004; Lee and Kesner, 2004a; Lee et al., 2004b; Leutgeb et al., 2004; Maren and Holt, 2004; Vazdarjanova and Guzowski, 2004). It is important to note, however, that differences among spatial contexts in real life often are not as evident as in the laboratory settings; some minor and major changes can occur in the same environment over time or different environments may look very similar to each other. Thus, the cognitive task of retrieving event memory based on contextual information may critically depend on a ‘‘match–mismatch’’ process (Honey et al., 1998; Vinogradova, 2001) between a certain context (given as a retrieval cue) and the original context associated with the event. This match–mismatch, or noveltydetection, process has been proposed as a critical function of the hippocampus by many researchers (Myhrer, 1988; Knight, 1996; Honey et al., 1998; Vinogradova, 2001; Fyhn et al., 2002; Mumby et al., 2002; Jenkins et al., 2004; Kesner et al., 2004; Mizumori et al., 2004; Lee et al., 2005a). This function is not only central to the contextual information processing for episodic memory in the hippocampus, it also connects to one of the key computational problems (i.e., pattern completion versus pattern separation) in computational modeling whenever a certain network needs to retrieve a stored pattern of information in response to variable input patterns that may or may not be identical to the stored information (Hopfield, 1982; McNaughton and Morris, 1987; McNaughton and Nadel, 1990; Hertz et al., 1991; Churchland and Sejnowski, 1992; Rolls and Treves, 1998). A pioneering and most influential computational model that brought these issues to the hippocampal field is Marr’s (1971) theory for archicortex. Marr hypothesized that the archicortex in general is specially structured for ‘‘simple representations’’ of discrete memories, whereas the neocortex is developed for a more sophisticated organization of memory. Marr took the hippocampus as a model system in the archicortex for constructing his theory. With respect to processing event memory, Marr assumed that the representation of a ‘‘whole event’’ is composed of the representations of ‘‘subevents.’’ Importantly, his
85
theory notes that the simple representation in the hippocampus is formed only of parts of the event and the hippocampal network puts together those parts of information, using modifiable synapses. For retrieval of the event memory, based on the collateral synapses found among pyramidal cells in the hippocampus (e.g., in CA3, although Marr did not specify the subfield), Marr proposed ‘‘collateral effects’’ as a mechanism to recognize individual subevents represented in the hippocampal network. The collateral effects refer to the process of retrieving an original pattern of information stored in the hippocampal network, based on modifiable synapses of the recurrent collaterals (in CA3), when only a partial information pattern is given to the network. Recognition of the individual subevents, according to Marr, then leads to recovery of the whole event. Although Marr did not construct his theory using the term context, his theory provides insights into how hippocampal circuits may process contextual information. His model for the retrieval of event memory (based on associations between partial patterns of information stored in the hippocampus) bears resemblance to theories of the role of the hippocampus in contextual information processing, especially the contextual retrieval theory originally proposed by Hirsh (1974). Collateral effects are better known as ‘‘pattern completion’’ in modern computational modeling (McNaughton and Morris, 1987; McNaughton and Nadel, 1990; Hertz et al., 1991; Churchland and Sejnowski, 1992; Rolls and Treves, 1998). Pattern completion is a critical process for recovering an original pattern of representation when corrupted input patterns are given to the network (because the inputs are either partial or altered, compared to the inputs used at the time of constructing the original representations). Marr also paid attention to the problem of storage capacity, as he modeled the amount of information that can be stored in the hippocampal network when separate representations of discrete subevents need to be maintained in the network without interference (Marr, 1971). Although Marr did not ascribe the process of storing separate events (i.e., pattern separation) solely to any particular hippocampal network, he modeled the pattern separation process in the cerebellum between the mossy fiber system and granule cells (Marr, 1969). Succeeding models (McNaughton and Morris, 1987; McNaughton and Nadel, 1990; O’Reilly and McClelland, 1994; Rolls, 1996; Rolls and Kesner, 2006) have implemented similar pattern separation mechanisms in the hippocampus between the granule cells in the dentate gyrus and the pyramidal cells in CA3 via the mossy fiber system. These models suggest that the cells in the dentate gyrus and their connections to the pyramidal cells in CA3 make
86 PLACE CELLS AND SPATIAL CONTEXT it possible to represent events sparsely in the hippocampal network so that different events can be separately coded with minimal interference. Pattern separation is vital for discriminating similar environments that contain overlapping contextual patterns and for storing those similar contextual patterns as orthogonal representations in the hippocampus. Inspired by Marr, other researchers have made pattern completion and pattern separation key computational principles of their hippocampal models elaborating the functions of different anatomical structures in the hippocampus (McNaughton and Morris, 1987; O’Reilly and McClelland, 1994; Rolls, 1996; O’Reilly and Rudy, 2000; Rolls and Kesner, 2006). For example, McNaughton and Nadel (1990) proposed that the dentate gyrus performed pattern separation by a ‘‘codon expansion’’ strategy that was directly analogous to that proposed by Marr (1969) in the cerebellum. Treves and Rolls (Treves and Rolls, 1992, 1994; Rolls and Treves, 1998) implemented pattern completion and pattern separation processes in discrete circuits in the hippocampus. According to their model, pattern separation is achieved mainly through the dentate gyrus, which functions as a competitive network that reduces redundancy in the perforant path inputs from the entorhinal cortex. The entorhinal cortical information orthogonally represented in the dentate gyrus is then transferred to the CA3 network via the sparse, but powerful, mossy fibers from the granule cells in the dentate gyrus. Given the competitive informationcoding scheme of the dentate gyrus and its sparse connectivity to pyramidal neurons in CA3, it is likely that similar patterns of entorhinal inputs to the dentate gyrus may activate orthogonal populations of CA3 cells. With respect to pattern completion, Rolls and Treves (Treves and Rolls, 1992, 1994; Rolls and Treves, 1998) and McNaughton and Morris (1987) emphasized the role of the recurrent collateral fibers in CA3 and the direct projections from the entorhinal cortex to CA3. That is, during retrieval, incomplete or altered input patterns are provided to CA3 via the perforant path from the entorhinal cortex. These input patterns initially activate an incomplete version of the original representation in CA3. However, the network restores the original (thus complete) representation as the recurrent collaterals between pyramidal cells in CA3 reinstate the rest of the information that was originally associated with the given, partial input patterns. Computational models have recognized that the capability of the hippocampus for performing the dual function of pattern completion and pattern separation may create a potential competition or ‘‘trade-off’’ (O’Reilly and McClelland, 1994; McClelland and Goddard, 1996) between the two processes. When does the CA3 network complete altered input patterns to
retrieve a single, original representation? When does it orthogonalize the modified input patterns to retrieve multiple, distinct representations? Computational models (Treves and Rolls, 1992, 1994; O’Reilly and McClelland, 1994; Hasselmo et al., 1995; Levy, 1996; McClelland and Goddard, 1996; Rolls and Treves, 1998; Lisman and Otmakhova, 2001) have attempted to simulate the dynamic response of the hippocampal network as changes occur in the direct and indirect inputs from the entorhinal cortex and the dentate gyrus, respectively, to the network. These simulations led to the conclusion that, depending on the input streams (e.g., entorhinal cortical inputs versus mossy fiber inputs from the dentate gyrus) that are active at the time of memory formation (or retrieval) and the proportion of overlap in the input patterns, the CA3 network can undergo qualitatively different retrieval modes (i.e., pattern completion or pattern separation).
EVIDENCE FROM PERTURBATION STUDIES The roles of the hippocampal subfields suggested by the computational models have been tested recently by perturbation studies. Perturbation studies in this chapter refer to the experiments that investigate the function of a selected neural structure by examining behavioral deficits associated with the damage in the structure (caused by lesions or pharmacological inactivation, for example). Case studies of human amnesics are one type of perturbation study. In amnesic patients, because of the variability across subjects in the extent of damage to different subfields of the hippocampus, it is often difficult to differentiate the roles of the subfields. Nonetheless, it appears that the extent of damage to the hippocampal subfields affects the severity of memory impairment. For example, patients with damage limited to CA1 exhibit memory deficits when episodic information (e.g., phone conversations or stories from TV) needs to be remembered (ZolaMorgan et al., 1986; Rempel-Clower et al., 1996). When patients had more extended damage (involving CA3 and the dentate gyrus in addition to CA1) in the hippocampus and other medial temporal lobe structures (i.e., subiculum and entorhinal cortex), they exhibited more severe deficits in retrieving their episodic memories (especially those memories that had occurred more recently) than did normal subjects or patients with damage limited to CA1. These studies further support the idea that monitoring the outputs of CA1 (which may be justified if and only if the information is transmitted solely through the serially connected circuits in the hippocampus) is not sufficient to understand the whole function of the hippocampus.
HIPPOCAMPAL SUBFIELDS AND SPATIAL CONTEXT OF EVENTS
Rather, it seems that the dentate gyrus and CA3 subfields perform additional operations that might not be accurately assessed by monitoring just CA1. Anatomically, although the hippocampus sends its outputs to cortical regions through CA1, it also sends major outputs to subcortical regions such as the mammillary bodies, anterior thalamic nuclei, and lateral/medial septal nuclei (Amaral and Witter, 1995; Jakab and Leranth, 1995). Among these subcortical connections, the projections from CA3 to the septal regions may explain the discrepancy in the severity of amnesic symptoms between the patients with damage only in CA1 and those with more extended damage including CA3. Because the septal projections reach the subiculum and entorhinal cortex (Amaral and Witter, 1995; Jakab and Leranth, 1995), CA3 damage may thus indirectly affect the inputs and outputs between the hippocampus and its associated cortical areas. Supporting this possibility, it was observed that rats whose CA3 efferents to the septal regions were disconnected showed deficits in acquiring a delayed nonmatch-to-sample task in a radial eight-arm maze (Hunsaker and Kesner, 2007). These animals were also impaired in performing the learned task when they were transferred to a novel environment. Although further investigations are necessary, the direct connections between CA3 and subcortical structures may play an important role in controlling the dynamics for information processing in the hippocampus, possibly through cholinergic modulation (Hasselmo et al., 1996; Hasselmo, 1999, 2006). A specific function of a given subfield can be better studied in animal models, since more localized lesions can be produced than in humans. With the availability of specific neurotoxins such as colchicine, for example, researchers can study an animal model with selective damage to the dentate gyrus and hilar region of the hippocampus, leaving other subfields (e.g., CA1 and CA3) mostly intact (Walsh et al., 1986; McLamb et al., 1988; Tilson et al., 1988; Emerich and Walsh, 1989; McNaughton et al., 1989; Xavier et al., 1999; Lee and Kesner, 2004a,b). In addition, knock-out techniques can target the genes responsible for the expression of specific receptors (e.g., NMDA receptors) located in a particular subfield of the hippocampus (Rondi-Reig et al., 2001; Nakazawa et al., 2002, 2003). Using these selective perturbation techniques, researchers have attempted to functionally map the individual circuits in the hippocampus to specific cognitive processes. For example, forming a novel contextual representation in event memory requires rapidly associating the spatial relationships among myriad cues in the environment. According to computational models, the CA3 subfield has an ideal structure to quickly represent a novel spatial context
87
(McNaughton and Morris, 1987; Treves and Rolls, 1992, 1994; O’Reilly and McClelland, 1994; Rolls, 1996; Rolls and Treves, 1998). Rolls (1996) referred to such a representational characteristic of CA3 as a ‘‘snapshot’’ quality. In CA3, the mossy fibers from the dentate gyrus to CA3 pyramidal neurons are essential (particularly during memory acquisition) to provide powerful ‘‘teaching signals’’ to activate specific neuronal populations in CA3 (McNaughton and Morris, 1987; McNaughton and Nadel, 1990; Treves and Rolls, 1992, 1994; Rolls, 1996; Rolls and Treves, 1998). The recurrent collaterals connecting the individual neurons in CA3 then may quickly form arbitrary associations among the neurons representing the environmental cues, thus forming a contextual representation of the novel spatial environment. This ‘‘autoassociative’’ process, forming extensive associations among the items within the same network (Marr, 1971; Hopfield, 1982; Kohonen, 1997), requires NMDA receptor–based, long-term potentiation of individual synapses in the hippocampus (Bains et al., 1999; Kesner and Rolls, 2001; Nakazawa et al., 2002). For example, selective blockade of NMDA receptors in CA3 in rats produced marked impairment in performing a spatial working-memory task in a novel environment (Lee and Kesner, 2002). Specifically, rats were trained to remember on each trial one of two arms adjacently located in a radial, eight-arm maze (Fig. 6–1a). The two arms used in a trial were randomly chosen. When a trial began, the rat visited one of the arms in the maze and retrieved food reward at the end of the arm (sample phase). In the following test phase, after either a short (10 s) or intermediate (5 min) delay, two adjacent arms were made available (including the arm that had been visited during the sample phase and a novel arm adjacent to it) and the animal needed to visit a novel arm that had not been visited during the sample phase. Importantly, the animal was confined in a bucket in the center platform of the maze, thus being unable to view the cues in the testing room during the delay period. Rats usually make rotational movements in random directions in the bucket and lose the directional information associated with the visited arm in the previous sample phase. This was confirmed when it was observed that they often headed toward a wrong side of the maze when the bucket was raised after the delay. After being trained in the task, a group of rats were implanted bilaterally with cannulae in one of the subfields (CA1, CA3, or dentate gyrus) of the hippocampus. After recovery from surgery, the animals showed normal performance in the familiar spatial context. Furthermore, blocking NMDA receptors (by infusing the NMDA-receptor antagonist, APV) in any of the hippocampal subfields did not disrupt the performance in
88 PLACE CELLS AND SPATIAL CONTEXT the familiar environment (Fig. 6–1b). However, when the animals were tested in the same task, yet in a completely novel room with the blockade of NMDA receptors, different results were obtained depending on the site of the NMDA receptor blockade in the hippocampus. Specifically, when the delay was short
(10 s), the animals with the blockade of CA3 NMDA receptors were severely impaired in performing the spatial working-memory task in the novel room, whereas the blockade of NMDA receptors in other subfields did not significantly affect the performance (Fig. 6–1c).
Figure 6–1. Performance in a delayed nonmatch-to-sample task in a familiar spatial environment versus a novel spatial environment. a. Delayed nonmatch-to-sample task. In a sample phase, the rat visited a randomly chosen arm (drawn with solid lines, in contrast to the dotted-lined arms, indicating inaccessible arms) in the eight-arm maze surrounded by distinct visual cues (not shown in the illustration). The rat retrieved a cereal reward (invisible from the center platform) at the end of the arm and returned to the center platform. In a delay period (10 s in this figure), the animal was confined in a bucket in the center platform, during which another arm adjacent to the visited arm (either left or right side of the sampled arm) was made accessible. During a choice phase, the rat needed to choose the arm that had not been visited during the sample phase to obtain reward. b. Comparison of within-subjects performance of different groups implanted with cannulae in different hippocampal subfields in the familiar room under the influence of vehicle solution (phosphate-buffered saline [PBS]; 2 days of 16 trials) followed by APV (2 days of 16 trials). None of the groups exhibited a significant difference in performance between the two conditions (PBS versus APV). c. Performance of the same task in a novel spatial environment for different cannulae groups. The PBS-injected groups (CA1-PBS and CA3-PBS) showed no difference, and were hence regrouped as one control group (CT-PBS). The CA3-APV group exhibited marked impairment of performance for the first two blocks compared to other groups (CA1-APV and CT-PBS), and its performance improved in the remaining two blocks (blocks 3 and 4). Adapted from Lee and Kesner (2002, Fig. 2) with permission of Nature Neuroscience.
HIPPOCAMPAL SUBFIELDS AND SPATIAL CONTEXT OF EVENTS
The Lee and Kesner study (2002) above was originally inspired by a theoretical model by Kesner and Rolls (2001), who provided a computational hypothesis to describe the hippocampal network functions that potentially underlie the successful performance of the animals in this task. They hypothesized that remembering a novel spatial location (i.e., an arm in the maze) requires the formation of a new spatial representation of the set of cues associated with that location. The CA3 subfield has been hypothesized to store the representation of such a local, spatial context that defines the arm location in the room. Their model suggests that NMDA receptor–dependent, synaptic plasticity mechanisms in CA3 are critical for setting up the new spatial representations associated with different arms of the maze, especially when the animal first explores the environment. Kesner and Rolls also emphasized the capability of the CA3 network for selectively activating and maintaining (even in the absence of cues during delay) the local contextual representation associated with the arm visited during the sample phase. In computational modeling, when the network maintains a stable activity pattern (or ‘‘state’’) in the absence of the inputs that triggered the activity state, such a network is called an ‘‘attractor network’’ and the state maintained in the network is called an ‘‘attractor state’’ (Hopfield, 1982; Hertz et al., 1991; Wang, 2001). Therefore, the CA3 network may form and maintain separate attractor states for different spatial contexts associated with the different arms of the maze (Kesner and Rolls, 2001; Rolls and Kesner, 2006). Their model thus predicts that, when the animals need to perform the same task in a completely novel room, CA3 would require the NMDA receptor– dependent, synaptic plasticity again to set up new spatial representations in the network. The results from the Lee and Kesner study (2002) can be explained by this model because it hypothesizes that activating a discrete attractor state for the familiar context (that has already been learned during training) does not require the NMDA receptors in CA3 (rather, it may require AMPA receptors). The results thus suggest that formation of representations of novel contextual information requires NMDA receptor–dependent, synaptic plasticity in CA3. The contribution of NMDA receptors in CA3 to representing novel contextual information was also confirmed by a study that tested knock-out mice with deficient NMDA receptors in CA3 in a behavioral task (Nakazawa et al., 2003). The knock-out mice were normal when they were required to find well-learned platform locations in a water maze, whereas they exhibited significant impairment when they were required to rapidly learn new platform locations. Supporting evidence has also been presented in a study in
89
which the rats with localized lesions in CA3 demonstrated deficits in rapidly associating a spatial context with foot shock in a contextual fear-conditioning paradigm (Lee and Kesner, 2004a). That is, when freezing behavior was measured during the acquisition of the contextual fear conditioning, the CA3-lesioned animals were deficient in exhibiting the freezing response only at the earlier stages of conditioning, whereas the animals with lesions in CA1 and dentate gyrus showed more sustained deficits during the acquisition than CA3-lesioned animals (Fig. 6–2). This result further demonstrates that the rapid formation of contextual information is dependent on the CA3 network. A few memory tasks that tested the role of CA3 in specific computational processes such as pattern completion have recently shown that the CA3 network is indeed necessary to perform memory tasks in spatial-cue-reduced environments (Nakazawa et al., 2002; Gold and Kesner, 2005). These experiments disturbed the normal operations of CA3 either through localized lesions in CA3 (Gold and Kesner, 2005) or by knocking out the genes for the NMDA receptors in CA3 pyramidal cells (Nakazawa et al., 2002). When Nakazawa et al. tested the CA3 NMDA-receptor knock-out mice in a water maze task in which the mice were required to swim to a previously learned location, they were significantly impaired when only a subset of the spatial cues was presented around the maze (but not when the complete set of cues was available). Gold and Kesner (2005) tested the same hypothesis, using a delayed nonmatch-to-place task on a dry-land version of the water maze (Fig. 6–3). In this task, the rats visited a certain location in a circular platform during a sample phase. When the rats were required to revisit the same location after a brief delay (10 s), the animals with CA3 lesions were significantly impaired in finding the same location when only a subset of cues was presented in the room. These results support the computational models that have ascribed the role of pattern completion to CA3 and provide converging evidence that CA3 is indeed necessary for retrieving the original representation of the environment when only partial cues are provided. With respect to pattern separation, computational models have emphasized the role of the dentate gyrus (McNaughton and Morris, 1987; O’Reilly and McClelland, 1994; Rolls, 1996; Rolls and Kesner, 2006). Using a spatial pattern separation paradigm, Gilbert and colleagues (2001) tested the rats with selective lesions in the dentate gyrus (made neurotoxically by colchine). In this task, the rats sampled one of the locations marked by a block object in an open arena to receive food reward (Fig. 6–4). After a brief delay period, the animals were allowed to revisit the same location, but this time with an identical foil-object
90 PLACE CELLS AND SPATIAL CONTEXT
Figure 6–2. Freezing behavior in subfield-lesion groups during the acquisition of contextual fear after the pre-acquisition period on day 1. Left: Representative photomicrographs for subfield-specific lesions in the dorsal hippocampus. Control lesion (CONT), CA1 lesion, dentate gyrus (DG) lesion, and CA3 lesion (from top to bottom). Right: Percent freezing exhibited during the intertrial intervals (i.e., averaged for three consecutive intertrial intervals ¼ 1 block) in an acquisition period commenced immediately after the pre-acquisition period. Note the significant impairment of freezing in all lesion groups in block 1 as well as in block 2 except for the CA3 lesion group. All groups eventually reached the control group’s level of freezing shown in block 3. Adapted from Lee and Kesner (2004a) with permission of Hippocampus.
placed at varying distances (i.e., 15, 37.5, 60, 82.5, or 105 cm) from the original object. Similar to the computational hypothesis proposed by Kesner and Rolls (2001) for the spatial working-memory task (see above), successful performance in this task may require the rats to differentially activate and maintain (during the delay) the spatial representations associated with the two locations marked by the objects. When the two locations are close to each other, it is likely that the spatial contexts associated with the two locations overlap greatly. This may necessitate the pattern separation process in the hippocampal network as input patterns reach the dentate gyrus and CA3 from the entorhinal cortex. Compared to sham-lesioned animals, the animals with lesions in the dentate gyrus were markedly impaired in revisiting the original spatial locations, especially when the two locations were close to each other (i.e., 15, 37.5, or 60 cm). The
behavioral deficits in the dentate gyrus–lesioned animals were likely due to the inability of the rats to form and retrieve the orthogonal spatial representations separately associated with the target and foil locations. When the two locations were far apart (i.e., 105 cm), the spatial contexts associated with those two locations minimally overlapped and pattern separation probably became less important. Supporting such a possibility, the performance between the control group and the lesion group was not significantly different in that condition, demonstrating that involvement of the spatial pattern separation process by the dentate gyrus is a function of the amount of overlap between input patterns. The results support the computational models that propose a critical role of the dentate gyrus for implementing the pattern separation mechanisms in the hippocampus when similar locations or events need to be disambiguated (see also McHugh et al., 2007).
HIPPOCAMPAL SUBFIELDS AND SPATIAL CONTEXT OF EVENTS
91
Figure 6–3. CA3 lesions and performance deficits in a delayed nonmatch-to-place task. Left and middle: Schematic representations of the apparatus on the sample phase and choice phase. There were 177 food wells on the surface of the apparatus and the center-most five rows had 15 food wells across. On the maze, there were Plexiglas partitions positioned on the tenth row (for encouraging discrete choices) and a start box (block rectangle). In the testing room, around the maze, four distinct cue cards served as distal cues. During the sample phase, the food well that contained a cereal reward was marked by a block object (denoted by a black dot) that was not presented during the choice phase. Note that some of the distal cues (two cues in the middle in this example) were eliminated during choice phases. Right: Performance by the control- and CA3-lesioned rats, measured by the mean degree of error (i.e., the absolute value of the angle of the choice location subtracted from the angle of the target location) as a function of cues available during the choice phase. Note the impairment of performance of CA3-lesioned animals in relatively cue-deprived conditions. Adapted from Gold and Kesner (2005) with permission of Hippocampus.
EVIDENCE FROM PHYSIOLOGICAL STUDIES The function of the hippocampus most certainly relies on dynamic interactions among its discrete subfields and associated regions. Although the perturbation studies are helpful in diagnosing the behavioral effects of the absence of a certain structure in the hippocampus, they are inadequate to reveal the normal mode of operation of that structure. The dynamic information processing at the level of individual circuits may be monitored best by recording neural activity as the animal processes the contextual information. In 1971, O’Keefe and Dostrovsky discovered that individual pyramidal neurons in the hippocampus fired when the rat positioned itself in particular locations in the environment. These neurons were called ‘‘place cells’’ and the spatial receptive fields of the place cells were named ‘‘place fields’’ (O’Keefe and Dostrovsky, 1971; O’Keefe and Nadel, 1978). The discovery of the place cell led to the idea that an internal, ‘‘cognitive map’’ of the environment resides in the hippocampus (O’Keefe and Nadel, 1978). The mechanisms by which the cells in the hippocampus represent a particular location in space, how-
ever, have been unclear. It appears that the firing of place cells is influenced by cues that are both distally and locally placed in the environment in reference to the animal’s position (O’Keefe and Nadel, 1978; Muller and Kubie, 1987; Young et al., 1994; Gothard et al., 1996b; Shapiro et al., 1997; Knierim, 2002). Idiothetic cues, such as vestibular information, can also have profound, sometimes dominant, influences on the spatial firing pattern of place cells in rats (McNaughton et al., 1983, 1996; Wiener et al., 1995; Knierim et al., 1998), as well as on hippocampal neurons in primates (Feigenbaum and Rolls, 1991). Furthermore, behavioral tasks appear to exert influences on the firing patterns of place cells (Breese et al., 1989; Markus et al., 1995; Kobayashi et al., 1997; Frank et al., 2000; Fyhn et al., 2002; Ferbinteanu and Shapiro, 2003; Lee et al., 2006; Shapiro et al., 2006). With respect to the influence from environmental cues (which is the focus of this chapter), place cells are more affected when the topological relationships among cues are manipulated, either by changing the locations of multiple stimuli in the room at the same time or by removing a significant number of cues from the environment, than by manipulations that leave the topological relationships among the cues intact (O’Keefe and Nadel, 1978;
92 PLACE CELLS AND SPATIAL CONTEXT
Figure 6–4. Performance of the control group and the dentate gyrus (DG) lesion group in a spatial pattern separation paradigm using a delayed match-to-sample task. Top: Schematic of spatial separation task apparatus and an example of a sample phase and a choice phase (objects are 15 cm apart in this example). Bottom: Mean percent correct performance as a function of spatial separation in the control group and DGlesion group. Each graph shows the performance averaged over 80 trials before surgery (pre-surgical, left), the first 160 trials after surgery (post-surgical 1/2; middle), and the last 160 trials after surgery (postsurgical 2/2; right). Note the inter-cue distance-dependent impairment in performance in the DG-lesioned group, demonstrating the role of DG in spatial pattern separation. Adapted from Gilbert et al. (2001) with permission of Hippocampus. O’Keefe, 1979; Muller and Kubie, 1987; O’Keefe and Speakman, 1987; Hetherington and Shapiro, 1997; Shapiro et al., 1997; Tanila et al., 1997). By contrast, when a single cue is removed from the environment, place cells often exhibit robust, unaltered firing (O’Keefe and Conway, 1978; Muller and Kubie, 1987; Shapiro et al., 1997). As suggested by computational models, the removal of an insignificant number of cues does not affect the spatial representation formed in the hippocampus, presumably because the network can retrieve the original spatial representation through pattern completion. The alterations observed in the firing patterns of place cells when the relationships between environmental cues are significantly manipulated (e.g., novel spatial context) suggest that the hippocampal network undergoes pattern separation when independent spatial representations need to be formed. Shapiro et al. (1997) directly tested the hypothesis that hippocampal neurons represent the spatial relationships between external cues in the environment by systematically manipulating the local cues (i.e., cues
that are contacted by the animal as it performs a working memory task on an elevated plus maze) and distal cues (i.e., cues that are remotely placed and cannot be contacted by the animal) in an environment and examining how hippocampal neurons respond to different spatial contexts (constructed by altering the relationships among the cues in the environment). The cue manipulations involved double rotation, distal scrambling, local scrambling, distal deletion, and local deletion of cues. The double-rotation manipulation involved rotating a set of distal cues in the room and a set of local cues along the surface of the maze in opposite directions to alter the familiar spatial relationships between distal and local cues. The scrambling manipulations involved the random scrambling of the distal or local cues, and the cue-deletion conditions involved the removal of a single distal or local cue from the environment. As a result, when distal and local cues were simultaneously manipulated (double rotation), most neurons in the hippocampus altered their preferred firing locations, ceased to fire, or star-
HIPPOCAMPAL SUBFIELDS AND SPATIAL CONTEXT OF EVENTS
ted to fire in the altered spatial context. Scrambling the cues also produced similar changes in firing patterns of the hippocampal neurons, whereas removing a single cue (either distal or local) minimally affected the preferred firing locations of the neurons. These results demonstrate that the overall spatial relationships among multiple cues in the environment are more important than individual single cues in controlling the firing properties of the hippocampal neurons (O’Keefe and Conway, 1978; Muller and Kubie, 1987; Hetherington and Shapiro, 1997). It has been reported that hippocampal neurons recorded from primates exhibit firing properties similar to those of place cells in that they respond maximally when monkeys look at a particular part of a spatial environment, that is, a local visual context (Rolls et al., 1997; Robertson et al., 1998; Georges-Francois et al., 1999; Rolls, 1999). Specifically, when the monkeys changed their positions in a room (either passively in a mobile chair or actively by walking), instead of firing maximally when they were positioned at certain locations of the room as in rodents, the hippocampal neurons selectively responded to specific ‘‘views’’ of the environment, irrespective of the locations of the animals (but see Ono et al., 1991, 1993). These ‘‘spatial view’’ cells also showed mnemonic properties because the cells maintained their firing patterns when the visual details of the scene were obscured by black curtains or by turning off the illumination of the room, as long as the monkeys looked toward the scene. Rolls interpreted this phenomenon as pattern completion using vestibular and proprioceptive inputs in the hippocampus (Rolls and Treves, 1998). Although the exact firing properties of the hippocampal neurons are different between species (i.e., location-bound firing versus spatial view–dependent firing in rodents and primates, respectively), presumably due to the substantial differences in their sensory systems, a common theme that arises from these physiological recording studies is that neuronal activity in the hippocampus is tuned to a contextual stimulus defined by multiple elemental cues in the environment and the spatial relationships among them.
DIFFERENCES BETWEEN HIPPOCAMPAL SUBFIELDS Given that the hippocampal neurons play an essential role in processing the spatial relationships among stimuli (see above), the next logical step is to provide a mechanistic explanation of how different hippocampal circuits work together to process the information. Computational models have emphasized that different hippocampal subfields perform different functions
93
when animals form and retrieve memory of a certain environment—in particular, the role of the CA3 subfield has been greatly emphasized (McNaughton and Morris, 1987; Treves and Rolls, 1992, 1994; O’Reilly and McClelland, 1994; McClelland and Goddard, 1996; Rolls and Treves, 1998; Wallenstein et al., 1998). Most physiological studies that recorded singleunit activity in the hippocampus, however, targeted CA1 alone or they recorded CA3 neurons only after multiple sessions of CA1 recording (Barnes et al., 1990; Wilson and McNaughton, 1993; Markus et al., 1994; Wiener et al., 1995; Gothard et al., 1996b; Mizumori and Kalyani, 1997; Shapiro et al., 1997; Mizumori et al., 1999). To investigate how the contextual information is processed at different circuit levels in the hippocampus, neuronal activity needs to be monitored simultaneously from different hippocampal subfields as the animal processes contextual information. Such experiments provide the best opportunity to test the detailed hypotheses from the computational models. In this section we will selectively review recent physiological evidence testing critical predictions made by computational models with respect to the differential functions of hippocampal subfields in processing contextual information. Lee et al. (2004b) tested the hypothesis that CA3, compared to CA1, is specialized for pattern completion when contextual cues are altered. This was examined by monitoring simultaneously the neural representations of CA1 and CA3 in a situation in which the spatial relationships among different types of cues in a familiar context were manipulated. The experiment was similar to a previous study (Shapiro et al., 1997) that manipulated the local and distal cues simultaneously in a familiar environment (see above). Specifically, rats were trained in a circular, curtained environment in which distinctive distal cues were available along the curtains and the floor (Fig. 6–5). In the center of the curtained environment was an elevated ring track. The surface of the ring track was divided into four sections, each of which contained a unique local cue (brown sandpaper, gray rubber mat, duct tape with white cross-stripes, and beige rubber shelf liner). Rats were trained to circle the ring track clockwise while consuming chocolate rewards placed at random locations on the ring track. During this pretraining, the distal and local cues maintained their locations in the environment as well as their spatial relationships with each other (i.e., standard condition). After pre-training, the rats were implanted with multi-electrode drives that targeted the CA1 and CA3 subfields to record a large number of neurons simultaneously as the animal performed the same foraging task in the environment. However, when the animal was reintroduced to the environment after the drive
Figure 6–5. Coherent representation of modified spatial contexts in CA3. a. Experimental design. The ring track (center) with distinctive local cues on its surface was positioned in a curtained environment (black outer circle). Distal cues were positioned along the curtained wall. Each day, the standard session (STD) was repeated three times, interleaved with cue-mismatch sessions (MIS) of different mismatch amounts (908 and 1358 in this example). b. Representative locations of CA1 (blue arrowhead) and CA3 (red arrowhead) electrodes within a subject. c. Correlation matrices between population firing-rate vectors. Matrices shown are between population firing-rate vectors of the first and second standard sessions (STD-1 versus STD-2 correlation matrix) or between those of standard and mismatch sessions (STD versus MIS correlation matrix) for each mismatch condition. The abscissa and the ordinate represent the location (in degrees) on the linearized track in the standard and/or mismatch sessions. The CA3 representations remained coherent at all mismatch angles, as indicated by the high correlation values along the diagonal of the matrix, whereas the CA1 representation abruptly lost coherence when mismatch angles exceeded 458. Adapted from Lee et al. (2004b) with permission of Nature.
94
HIPPOCAMPAL SUBFIELDS AND SPATIAL CONTEXT OF EVENTS
implantation, the following contextual manipulations were made. Three standard conditions were given, interleaved with two mismatch conditions, in which the topological relationships between distal and local cues were systematically manipulated by rotating the distal cues in a clockwise direction and the local cues in a counterclockwise direction. For example, a 908 mismatch condition involved rotating the whole set of distal cues 458 clockwise and rotating the whole set of local cues (i.e., ring track) 458 counterclockwise. Different mismatch conditions (458, 908, 1358, and 1808) were randomly placed between standard conditions in the experiment. When the alteration in the spatial relationships among cues was minimal (i.e., the 458 mismatch condition), similar firing patterns were observed between CA1 and CA3; that is, similar proportions of neurons in CA1 and CA3 followed either subset of cues (distal or local cues) or their firing fields appeared or disappeared in the mismatch environment. It may be considered that the CA3 subfield was in a pattern-completion mode in this minimal cue-mismatch condition. However, as the topological relationships between cues deviated further from the standard configuration, the two networks (CA1 and CA3) exhibited drastically different dynamics in response to the altered environments. The ensemble behavior of the neuronal firing fields in CA3 was more coherent than that of CA1, in that the CA3 neurons tended to follow local cues (with a few place fields following the distal cues), whereas CA1 place fields followed the distal and local cues equally. Although it is unclear why CA3 followed local cues more than distal cues, the net result of this network behavior was that the new representation of the altered environment was more correlated with the representation of the original environment, as demonstrated by a population vector correlation analysis between the firing fields recorded in the original and altered environments. This result can be considered an instantiation of pattern completion (or pattern generalization more accurately, as the same spatial cues were used to produce altered environments instead of removing the existing cues). However, as the altered spatial context deviated to a greater extent from the original spatial context (e.g., 1808 mismatch condition), the correlation between the population representations of the standard environment and mismatch environment became weaker. The gradual dissipation of the similar representational patterns of the environments (between the standard and mismatch conditions) in CA3 suggests that the CA3 network might undergo the transition from pattern completion and generalization to pattern separation modes as the degree of mismatch becomes larger. The reason for the CA3 network showing mostly pattern completion and generalization in the Lee et al.
95
study, even in the most extreme 1808 mismatch sessions, might be that the experiments were performed in a familiar environment with familiar cue sets, despite the alterations in the spatial relationships among cues. It is conceivable that certain environmental stimuli (e.g., familiar odors in the environment, the shape of the recording room, the overhead lighting, etc.) not directly manipulated in the experiment might still give rats a sense of familiarity in the experiment. That is, manipulating familiar cue sets in the familiar room might have restricted the expression of the full range of dynamics from pattern completion to pattern separation in the CA3 network. It appears that introducing novel environmental components is necessary to test the pattern separation function of CA3. Such an experiment was conducted by Leutgeb et al. (2004), who simultaneously monitored neural firing patterns in CA1 and CA3 when rats explored environments in two rooms with distinct visual cues. When the animals explored an environment (i.e., a cylindrical or box enclosure) placed in the same room repeatedly between the exploration sessions, neurons in CA1 and CA3 maintained similar firing rates between the two environments. However, when the rats explored the cylindrical or box environments placed in two different rooms, the changes in firing rate were differentially affected in CA1 and CA3; neurons in CA3 changed their firing rates more significantly than those in CA1 between the two different environments, a result suggesting that CA3 representations underwent a pattern separation process to a larger extent than CA1. The evidence for both pattern completion (Lee et al., 2004b) and pattern separation (Leutgeb et al., 2004) processes in CA3 provides convincing evidence that the CA3 network is capable of both computations, as predicted by theoretical models. However, these two computational processes were independently shown in two different experiments. More convincing experimental evidence would show both pattern completion and pattern separation in CA3 (including the transition between the processes), using a single experimental paradigm. This condition was realized by an experiment performed by Vazdarjanova and Guzowski (2004), who allowed the rats to explore a pair of environments composed of a familiar environment and an altered version (to varying degrees) of it. Minimally altered environments were made by rearranging the familiar objects in the environment or by replacing the familiar objects with novel ones, without altering the positions of the objects (thus maintaining their spatial relationships). A maximally altered environment was made by changing the locations of novel objects in a novel environment (e.g., novel enclosure in a novel room). The overlap in hippocampal
96 PLACE CELLS AND SPATIAL CONTEXT neuronal populations that represent the familiar and altered environments was measured by quantifying the expression of two immediate early genes (IEG), Homer1a and Arc, which reflect neural activity levels with different time courses. HomerIa is expressed later than Arc; when the time for the exploration of an environment is appropriately timed to the expression time course of each IEG, it is possible to identify different populations of neurons that participate in processing information in different environments. For example, if a neuron participates in processing contextual information from two environments explored in sequence, both Homer1a and Arc will be identified in the neuron. Otherwise, if a neuron is involved in processing only one of the environments, only one of the IEGs will be detected (Homer1a for exploring the earlier environment and Arc for the later environment). The results from the Vazdarjanova and Guzowski study (2004) indicate that the neuronal populations in CA3 that were involved in exploring the familiar environment and the altered environment had greater overlap than those in CA1 when the modified environment consisted of only minor changes in the original environment. This finding demonstrates the pattern completion process in CA3 and can be interpreted as a similar phenomenon to that shown in the Lee et al. (2004b) study. When the modification of the original environment involved more drastic changes, however, the neurons activated between the original and altered environments were different, showing the pattern separation process in CA3 as demonstrated in the Leutgeb et al. (2004) study. By contrast, the neurons in CA1 reflected the changes in the environment more linearly without the nonlinear network dynamics observed in CA3 for categorizing the incoming contextual information. Computational models (see the model by Kesner and Rolls above, for example) have emphasized the role of the CA3 subfield in rapidly forming a novel contextual representation as the animal encounters a new spatial environment. Because of the powerful connections between the dentate gyrus and CA3 via the mossy fiber system, it has been suggested that the ‘‘snapshot’’ quality of memory formation in the hippocampus is provided mainly by CA3 (Rolls, 1996). For example, Rolls has proposed that the perforant path synapses are too weak to sufficiently activate the pyramidal neurons with only one-time experience of events. According to this computational hypothesis, the novel context-related changes in neural activity should be observed first in CA3 in the hippocampus. The perturbation study by Lee and Kesner (2002) supports this prediction: the animals with the NMDA receptor antagonist in CA3, but not in CA1 or the dentate gyrus, were selectively impaired when they performed the
spatial working-memory task in a novel room. Physiological evidence for the possibility of differential time courses of encoding environmental changes between subfields was recently provided by Lee et al. (2004a), using the same cue-rotation paradigm as described above. Specifically, place cells change their preferred firing locations as the animal runs a stereotyped route (Mehta et al., 1997, 2000; Ekstrom et al., 2001); the field moves in a direction opposite to the animal’s movement direction (thus called ‘‘backward shifting’’ of place fields) and the field’s shape is negatively skewed. These plastic changes in the firing properties may be a reflection of synaptic changes that occur as the hippocampal neurons create representations of learned sequences of locations or events (Blum and Abbott, 1996; Knierim, 2000). Since the shift in place field location occurs in an experience-dependent manner, by monitoring which subfield (between CA1 and CA3) exhibits such changes earlier or later as rats experience novel spatial relationships among familiar cues, Lee et al. (2004a) attempted to dissociate the time course of contextual information processing in CA1 and CA3 (Fig. 6–6). Their results show that the neurons in CA3 shifted their firing locations backward across laps when the animals encountered the altered environment for the first time, compared to when they experienced the environmental changes repeatedly over several days. CA1 neurons, however, did not shift their firing fields when the rats experienced the modified environment for the first time, but exhibited experience-dependent, plastic changes in their firing locations after the animals experienced the environments repeatedly across days (see also Yu et al., 2006). Moreover, the results suggest that long-term storage of these spatiotemporal sequences may be localized to the CA3 network, with the CA1 network showing only temporary storage of the sequences (consistent with the original CA1 findings by Mehta et al., 1997, 2000). The double dissociations observed between CA1 and CA3 point to promising lines of research, by showing that the differential, intrahippocampal dynamics for constructing and retrieving the spatial contextual information between different subfields can be simultaneously monitored in the hippocampus experimentally, thus bringing closer the possibility of testing the predictions from computational models in the future.
CONCLUDING REMARKS Memory of episodic events that include rich contextual details in the environment is significantly affected by hippocampal damage in human patients and animal models (Scoville and Milner, 1957; Hirsh, 1974; Mishkin, 1978; O’Keefe and Nadel, 1978; Olton et al.,
HIPPOCAMPAL SUBFIELDS AND SPATIAL CONTEXT OF EVENTS
97
Figure 6–6. Double dissociation between the time courses of CA3 and CA1 place cells for processing altered spatial contexts. Top: Illustration of the calculation procedure for the difference (DCOM in degrees) between a lap-based COM (COMt or COMtþ1) and the COM of the session-based, average place field (COMAVG). When the lap-based COM (COMt) is past the COMAVG, the DCOM is positive (DCOMt), whereas the DCOM is negative when the lap-based COM (DCOMtþ1) is ahead of the COM relative to the direction of movement of the rat. Bottom: Lap-based DCOM of the CA1 and CA3 place fields in the mismatch sessions across 4 days (D1-MIS to D4-MIS). The ordinate scales for D3MIS and D4-MIS are different from those of D1-MIS and D2-MIS for better presentation of the trends in the data. Note the backward shifting of CA3 place fields, but not CA1 fields, on day 1 and the reversed pattern on day 4, when the rats encountered the altered spatial contexts across days. Adapted from Lee et al. (2004a) with permission of Neuron.
1978; Morris et al., 1982; Kesner, 1985; Zola-Morgan et al., 1986; Kesner and Beers, 1988; Kim and Fanselow, 1992; Squire, 1992; Alvarez et al., 1995; Rempel-Clower et al., 1996; Dusek and Eichenbaum, 1997; Vargha-Khadem et al., 1997; Mishkin et al., 1998; Tulving and Markowitsch, 1998; Griffiths et al., 1999; Eichenbaum, 2000; Suzuki and Clayton, 2000; Aggleton and Pearce, 2001; Morris, 2001; Spiers et al., 2001; Ferbinteanu et al., 2006). On the basis of theoretical models for the hippocampus (Sutherland et al., 1989; Treves and Rolls, 1992, 1994; O’Reilly
and McClelland, 1994; McClelland and Goddard, 1996; Rolls, 1996; Cohen et al., 1997; Shapiro et al., 1997; Rolls and Treves, 1998; Fanselow, 2000; Kesner and Rolls, 2001; Rudy and O’Reilly, 2001; Burgess et al., 2002; Rolls and Kesner, 2006), this is mostly likely due to its central role in representing the relationships among numerous sensory stimuli in an environment and integrating those representations with a path integration–based spatial map probably conveyed by the medial entorhinal cortex (Hafting et al, 2005; McNaughton et al., 2006), as well as with other essen-
98 PLACE CELLS AND SPATIAL CONTEXT tial components of an episodic event such as goals and motivations (Breese et al., 1989; Eichenbaum et al., 1990; Markus et al., 1995; Wood et al., 2000; Moita et al., 2003; Lee et al., 2006; Shapiro et al., 2006; Smith and Mizumori, 2006a,b). The direction of future research thus needs to be shifted from establishing the relationship between episodic memory and the hippocampus as a whole to identifying the discrete functions of individual subfields of the hippocampus (and the connected structures outside the hippocampus; Knierim, 2006) in specific cognitive processes essential for event memory. The success of this endeavor requires detailed, process-oriented hypotheses (e.g., pattern completion and separation) for the formation and retrieval of event memory and the assignment of those cognitive processes to discrete anatomical circuits in the hippocampus and related structures. Computational models have contributed significantly to generating mechanistic explanations of how different subfields function for encoding, storage, and retrieval of event information. These models still await rigorous experimental validation, however, and they almost certainly will require modification as experimental results accumulate. Such a collaborative, investigative strategy for producing biologically plausible models, as selectively reviewed in this chapter, has produced some fruitful results in physiological studies as well as in the perturbation paradigms using localized lesions or knock-out techniques. Although there is experimental evidence to support some of the key computational principles in the hippocampus, this evidence is still insufficient to explain the hippocampal contributions to complex behavior. When an animal is engaged in a memory task, there are interactions among multiple systems responsible for processing different internal variables such as goals, strategies, attention, motivations, and selfmotion cues (e.g., vestibular and proprioceptive senses), in addition to the contextual information on the external environment. It is highly likely that hippocampal network activity is dynamically influenced by those other systems as the animal performs a hippocampus-dependent, event-memory task. Supporting experimental evidence indicates that the firing properties of the hippocampal neurons cannot be exclusively explained by external spatial cues in the environment in some situations (Breese et al., 1989; Markus et al., 1995; Wiener and Korshunov, 1995; Kobayashi et al., 1997; Wood et al., 2000; Fyhn et al., 2002; Rolls and Xiang, 2005; Lee et al., 2006; Shapiro et al., 2006; Smith and Mizumori, 2006a). That is, the hippocampal network dynamics are modulated by the nature of the task at hand and by other internal-state variables. For example, place cells are significantly modulated by idiothetic (or internally generated)
cues regarding directional and distance information (O’Keefe and Nadel, 1978; Knierim et al., 1995, 1998; Wiener et al., 1995; Gothard et al., 1996a, 2001; McNaughton et al., 1996; Leutgeb et al., 2000). Such modulations become particularly salient when the animal is engaged in path integration–based navigation in which the current location is estimated according to the movements made since the last known position. In one study, the rat ran back and forth between a fixed reward location at the end of a linear track and a start box that shifted its location along the track (Gothard et al., 1996a). The majority of CA1 cells that were active during the initial outbound journey from the start box fired at fixed distances from the start box, whereas other cells fired at fixed locations (including the reward location) on the track irrespective of the distance from the shifting start-box location. In another study, when the rats alternated the two reward locations (via the center stem of the maze, following a ‘‘figure-8’’ trajectory on the maze) positioned on the opposite sides of a continuous T-maze in a familiar environment, the majority of CA1 neurons that exhibited their place fields in the stem fired conditionally, depending on the reward locations visited immediately before and after traversal of the stem (Wood et al., 2000; Lee et al., 2006). Those neurons, therefore, seem to represent more than the physical location in the stem, because there was no change in the position information associated with the stem (as well as no changes in environmental cues around the maze) between the two types of journeys (i.e., left-toright reward locations versus right-to-left reward locations via the stem). Moreover, the firing fields of those neurons shifted toward a certain reward location within a behavioral session (Fig. 6–7), whereas other classic place cells with stationary firing fields (mostly near the reward or goal locations) were simultaneously observed in the same task (Lee et al., 2006). These results collectively suggest that the firing patterns of hippocampal neurons are influenced not only by the external stimuli composed of spatial cues and their topological relationships but also by internal variables (‘‘internal context’’; Kennedy and Shapiro, 2004) that influence the animal’s behavior in space. In a goal-oriented, complex memory task, the internal context may encompass other variables such as emotional variables and task demands (Hirsh, 1974; Davidson and Jarrard, 1993; Hock and Bunsey, 1998; Kennedy and Shapiro, 2004; Rolls and Xiang, 2005) than simply idiothetic sensory cues (e.g., vestibular and proprioceptive inputs). Although it is beyond the scope of this chapter, it is important to investigate how these variables interact in the hippocampal circuits when events need to be remembered. For example, investigation of the discrepancy in the activity of hip-
Figure 6–7. Trial type–dependent, conditional firing and forward shifting of the spatial firing fields of CA1 neurons in a continuous T-maze alternation task. a. Left: Illustration of the animal’s trajectory associated with left-to-right (L-R; red) and right-to-left (R-L; blue) trials. The two trajectories are separately shown (arranged vertically) only for illustrative purposes. Right: Representative examples of CA1 spatial firing fields selectively active on the stem for either L-R or R-L trial type of the alternation task. Firing fields for a single CA1 neuron are shown following the illustrative scheme shown on the left. Gray, position data for each trial type; red and blue, position data associated with spikes for L-R and R-L trial type, respectively. Numbers associated with the stem denote average firing rate (Hz) on the stem associated with each trial type. b. Representative examples of CA1 neurons that shifted their spatial representations forward via the stem of the maze across trials, plotted horizontally in sequential five-trial blocks (L-R [red] and R-L [blue] fields for each neurons shown separately as in panel a). c. Forward shifting of preferred firing locations. The DCOM represents the location of an individual neuron’s firing field relative to its average field location (DCOM ¼ 0), calculated from all the trials combined. Inset: Speed profile after the first 10 trials. Error bars ¼ SEM. Adapted from Lee et al. (2006) with permission of Neuron.
99
100
PLACE CELLS AND SPATIAL CONTEXT
pocampal neurons between different experimental paradigms, especially between the memory tasks that require animals to differentially respond to discrete events and non-memory tasks in which no such conditional responses are required (e.g., foraging freely in an environment), may provide great insights into how the hippocampal network dynamics change in response to different cognitive components required to process various events. If that is the case, the theoretical or computational models for the hippocampal circuits need to be verified in various behavioral paradigms. It would also suggest the importance of interpreting the network dynamics in the hippocampus in relation to other associated networks such as the entorhinal cortex, amygdala, and prefrontal cortex (Kesner, 1985; Chiba et al., 1994; Burgess et al., 2001b; Petrovich et al., 2001; Brun et al., 2002; Lee and Kesner, 2003b; Kesner and Rogers, 2004; Mizumori et al., 2004; Ferbinteanu et al., 2006; Isomura et al., 2006; Knierim et al., 2006; Parron et al., 2006; Sargolini et al., 2006; Knierim, 2006). Such an approach will allow a better diagnosis of the unique role of the hippocampal network by knowing what proportion of the network behavior observed in the hippocampus needs to be interpreted in relation to the functions of the other structures and will lead to a better functional delineation of different brain structures in the remembering of episodic events.
References Addis DR, Moscovitch M, Crawley AP, McAndrews MP (2004) Recollective qualities modulate hippocampal activation during autobiographical memory retrieval. Hippocampus 14:752–762. Aggleton JP, Pearce JM (2001) Neural systems underlying episodic memory: insights from animal research. Philos Trans R Soc Lond B Biol Sci 356:1467–1482. Alvarez P, Zola-Morgan S, Squire LR (1995) Damage limited to the hippocampal region produces longlasting memory impairment in monkeys. J Neurosci 15:3796–3807. Amaral DG, Witter MP (1989) The three-dimensional organization of the hippocampal formation: a review of anatomical data. Neuroscience 31:571–591. Amaral DG, Witter MP (1995) The hippocampal formation. In: The Rat Nervous System (Paxinos G, ed.), pp 443–493. San Diego: Academic Press. Andersen P, Bliss TV, Skrede KK (1971) Unit analysis of hippocampal population spikes. Exp Brain Res 13: 208–221. Anderson MI, Killing S, Morris C, O’Donoghue A, Onyiagha D, Stevenson R, Verriotis M, Jeffery KJ (2006) Behavioral correlates of the distributed coding of spatial context. Hippocampus 16:730–742.
Angeli SJ, Murray EA, Mishkin M (1993) Hippocampectomized monkeys can remember one place but not two. Neuropsychologia 31:1021–1030. Bains JS, Longacher JM, Staley KJ (1999) Reciprocal interactions between CA3 network activity and strength of recurrent collateral synapses. Nat Neurosci 2:720– 726. Banta Lavenex P, Amaral DG, Lavenex P (2006) Hippocampal lesion prevents spatial relational learning in adult macaque monkeys. J Neurosci 26:4546– 4558. Barnes CA, McNaughton BL, Mizumori SJ, Leonard BW, Lin LH (1990) Comparison of spatial and temporal characteristics of neuronal activity in sequential stages of hippocampal processing. Prog Brain Res 83: 287–300. Blanchard RJ, Blanchard DC (1969) Crouching as an index of fear. J Comp Physiol Psychol 67:370–375. Bliss TV, Lomo T (1973) Long-lasting potentiation of synaptic transmission in the dentate area of the anaesthetized rabbit following stimulation of the perforant path. J Physiol 232:331–356. Blum KI, Abbott LF (1996) A model of spatial map formation in the hippocampus of the rat. Neural Comput 8:85–93. Breese CR, Hampson RE, Deadwyler SA (1989) Hippocampal place cells: stereotypy and plasticity. J Neurosci 9:1097–1111. Brun VH, Otnass MK, Molden S, Steffenach HA, Witter MP, Moser MB, Moser EI (2002) Place cells and place recognition maintained by direct entorhinalhippocampal circuitry. Science 296:2243–2246. Burgess N, Becker S, King JA, O’Keefe J (2001a) Memory for events and their spatial context: models and experiments. Philos Trans R Soc Lond B Biol Sci 356:1493–1503. Burgess N, Maguire EA, O’Keefe J (2002) The human hippocampus and spatial and episodic memory. Neuron 35:625–641. Burgess N, Maguire EA, Spiers HJ, O’Keefe J (2001b) A temporoparietal and prefrontal network for retrieving the spatial context of lifelike events. Neuroimage 14: 439–453. Chiba AA, Kesner RP, Reynolds AM (1994) Memory for spatial location as a function of temporal lag in rats: role of hippocampus and medial prefrontal cortex. Behav Neural Biol 61:123–131. Churchland PS, Sejnowski TJ (1992) The Computational Brain. Cambridge, MA: MIT Press. Clayton NS, Bussey TJ, Dickinson A (2003) Can animals recall the past and plan for the future? Nat Rev Neurosci 4:685–691. Clayton NS, Dickinson A (1998) Episodic-like memory during cache recovery by scrub jays. Nature 395:272– 274.
HIPPOCAMPAL SUBFIELDS AND SPATIAL CONTEXT OF EVENTS
Cohen NJ, Poldrack RA, Eichenbaum H (1997) Memory for items and memory for relations in the procedural/ declarative memory framework. Memory 5:131– 178. Corcoran KA, Maren S (2001) Hippocampal inactivation disrupts contextual retrieval of fear memory after extinction. J Neurosci 21:1720–1726. Davachi L, Mitchell JP, Wagner AD (2003) Multiple routes to memory: distinct medial temporal lobe processes build item and source memories. Proc Natl Acad Sci USA 100:2157–2162. Davidson TL, Jarrard LE (1993) A role for hippocampus in the utilization of hunger signals. Behav Neural Biol 59:167–171. Dobbins IG, Rice HJ, Wagner AD, Schacter DL (2003) Memory orientation and success: separable neurocognitive components underlying episodic recognition. Neuropsychologia 41:318–333. Dusek JA, Eichenbaum H (1997) The hippocampus and memory for orderly stimulus relations. Proc Natl Acad Sci USA 94:7109–7114. Eichenbaum H (2000) A cortical-hippocampal system for declarative memory. Nat Rev Neurosci 1:41–50. Eichenbaum H, Stewart C, Morris RG (1990) Hippocampal representation in place learning. J Neurosci 10: 3531–3542. Ekstrom AD, Meltzer J, McNaughton BL, Barnes CA (2001) NMDA receptor antagonism blocks experiencedependent expansion of hippocampal ‘‘place fields.’’ Neuron 31:631–638. Eldridge LL, Engel SA, Zeineh MM, Bookheimer SY, Knowlton BJ (2005) A dissociation of encoding and retrieval processes in the human hippocampus. J Neurosci 25:3280–3286. Eldridge LL, Knowlton BJ, Furmanski CS, Bookheimer SY, Engel SA (2000) Remembering episodes: a selective role for the hippocampus during retrieval. Nat Neurosci 3:1149–1152. Emerich DF, Walsh TJ (1989) Selective working memory impairments following intradentate injection of colchicine: attenuation of the behavioral but not the neuropathological effects by gangliosides GM1 and AGF2. Physiol Behav 45:93–101. Fanselow MS (2000) Contextual fear, gestalt memories, and the hippocampus. Behav Brain Res 110:73–81. Feigenbaum JD, Rolls ET (1991) Allocentric and egocentric spatial information processing in the hippocampal formation of the behaving primates. Psychobiology 19:21–40. Ferbinteanu J, Kennedy PJ, Shapiro M (2006) Episodic memory—from brain to mind. Hippocampus 16:691– 703. Ferbinteanu J, Shapiro ML (2003) Prospective and retrospective memory coding in the hippocampus. Neuron 40:1227–1239.
101
Fortin NJ, Agster KL, Eichenbaum HB (2002) Critical role of the hippocampus in memory for sequences of events. Nat Neurosci 5:458–462. Forwood SE, Winters BD, Bussey TJ (2005) Hippocampal lesions that abolish spatial maze performance spare object recognition memory at delays of up to 48 hours. Hippocampus 15:347–355. Frank LM, Brown EN, Wilson M (2000) Trajectory encoding in the hippocampus and entorhinal cortex. Neuron 27:169–178. Fyhn M, Molden S, Hollup S, Moser MB, Moser E (2002) Hippocampal neurons responding to first-time dislocation of a target object. Neuron 35:555–566. Georges-Francois P, Rolls ET, Robertson RG (1999) Spatial view cells in the primate hippocampus: allocentric view not head direction or eye position or place. Cereb Cortex 9:197–212. Gilbert PE, Kesner RP (2003) Localization of function within the dorsal hippocampus: the role of the CA3 subregion in paired-associate learning. Behav Neurosci 117:1385–1394. Gilbert PE, Kesner RP (2004) Memory for objects and their locations: the role of the hippocampus in retention of object-place associations. Neurobiol Learn Mem 81:39–45. Gilbert PE, Kesner RP (2006) The role of the dorsal CA3 hippocampal subregion in spatial working memory and pattern separation. Behav Brain Res 169:142– 149. Gilbert PE, Kesner RP, Lee I (2001) Dissociating hippocampal subregions: double dissociation between dentate gyrus and CA1. Hippocampus 11:626–636. Gold AE, Kesner RP (2005) The role of the CA3 subregion of the dorsal hippocampus in spatial pattern completion in the rat. Hippocampus 15:808–814. Good M, Honey RC (1991) Conditioning and contextual retrieval in hippocampal rats. Behav Neurosci 105: 499–509. Gothard KM, Hoffman KL, Battaglia FP, McNaughton BL (2001) Dentate gyrus and CA1 ensemble activity during spatial reference frame shifts in the presence and absence of visual input. J Neurosci 21:7284– 7292. Gothard KM, Skaggs WE, McNaughton BL (1996a) Dynamics of mismatch correction in the hippocampal ensemble code for space: interaction between path integration and environmental cues. J Neurosci 16: 8027–8040. Gothard KM, Skaggs WE, Moore KM, McNaughton BL (1996b) Binding of hippocampal CA1 neural activity to multiple reference frames in a landmark-based navigation task. J Neurosci 16:823–835. Griffiths D, Dickinson A, Clayton N (1999) Episodic memory: what can animals remember about their past? Trends Cogn Sci 3:74–80.
102
PLACE CELLS AND SPATIAL CONTEXT
Hafting T, Fhyn M, Molden S, Moser MB, Moser EI (2005) Microstructure of a spatial map in the entorhinal cortex. Nature 436:801–806. Hasselmo ME (1999) Neuromodulation: acetylcholine and memory consolidation. Trends Cogn Sci 3:351–359. Hasselmo ME (2006) The role of acetylcholine in learning and memory. Curr Opin Neurobiol 16:710–715. Hasselmo ME, Schnell E, Barkai E (1995) Dynamics of learning and recall at excitatory recurrent synapses and cholinergic modulation in rat hippocampal region CA3. J Neurosci 15:5249–5262. Hasselmo ME, Wyble BP, Wallenstein GV (1996) Encoding and retrieval of episodic memories: role of cholinergic and GABAergic modulation in the hippocampus. Hippocampus 6:693–708. Hertz J, Palmer RG, Krogh AS (1991) Introduction to the Theory of Neural Computation. Redwood City, CA: Addison-Wesley. Hetherington PA, Shapiro ML (1997) Hippocampal place fields are altered by the removal of single visual cues in a distance-dependent manner. Behav Neurosci 111: 20–34. Hirsh R (1974) The hippocampus and contextual retrieval of information from memory: a theory. Behav Biol 12:421–444. Hock BJ, Jr., Bunsey MD (1998) Differential effects of dorsal and ventral hippocampal lesions. J Neurosci 18:7027–7032. Honey RC, Watt A, Good M (1998) Hippocampal lesions disrupt an associative mismatch process. J Neurosci 18:2226–2230. Hopfield JJ (1982) Neural networks and physical systems with emergent collective computational abilities. Proc Natl Acad Sci USA 79:2554–2558. Hunsaker MR, Allan KD, Kesner RP (2007) Role of dCA3 efferents via the fimbria in the acquisition of a delay nonmatch to place task. Hippocampus 17:494–502. Isomura Y, Sirota A, Ozen S, Montgomery S, Mizuseki K, Henze DA, Buzsaki G (2006) Integration and segregation of activity in entorhinal-hippocampal subregions by neocortical slow oscillations. Neuron 52:871–882. Jakab RL, Leranth C (1995) Septum. In: The Rat Nervous System (Paxinos G, ed.). San Diego: Academic Press. Jeffery KJ, Anderson MI, Hayman R, Chakraborty S (2004) A proposed architecture for the neural representation of spatial context. Neurosci Biobehav Rev 28:201–218. Jenkins TA, Amin E, Pearce JM, Brown MW, Aggleton JP (2004) Novel spatial arrangements of familiar visual stimuli promote activity in the rat hippocampal formation but not the parahippocampal cortices: a cfos expression study. Neuroscience 124:43–52. Jerman T, Kesner RP, Hunsaker MR (2006) Disconnection analysis of CA3 and DG in mediating encoding
but not retrieval in a spatial maze learning task. Learn Mem 13:458–464. Kennedy PJ, Shapiro ML (2004) Retrieving memories via internal context requires the hippocampus. J Neurosci 24:6979–6985. Kesner RP (1985) Correspondence between humans and animals in coding of temporal attributes: role of hippocampus and prefrontal cortex. Ann NY Acad Sci 444:122–136. Kesner RP, Beers DR (1988) Dissociation of data-based and expectancy-based memory following hippocampal lesions in rats. Behav Neural Biol 50:46–60. Kesner RP, Gilbert PE, Wallenstein GV (2000) Testing neural network models of memory with behavioral experiments. Curr Opin Neurobiol 10:260–265. Kesner RP, Hopkins RO (2006) Mnemonic functions of the hippocampus: a comparison between animals and humans. Biol Psychol 73:3–18. Kesner RP, Hunsaker MR, Gilbert PE (2005) The role of CA1 in the acquisition of an object-trace-odor paired associate task. Behav Neurosci 119:781–786. Kesner RP, Lee I, Gilbert P (2004) A behavioral assessment of hippocampal function based on a subregional analysis. Rev Neurosci 15:333–351. Kesner RP, Rogers J (2004) An analysis of independence and interactions of brain substrates that subserve multiple attributes, memory systems, and underlying processes. Neurobiol Learn Mem 82:199–215. Kesner RP, Rolls ET (2001) Role of long-term synaptic modification in short-term memory. Hippocampus 11: 240–250. Kim JJ, Fanselow MS (1992) Modality-specific retrograde amnesia of fear. Science 256:675–677. Kirwan CB, Gilbert PE, Kesner RP (2005) The role of the hippocampus in the retrieval of a spatial location. Neurobiol Learn Mem 83:65–71. Knierim JJ (2000) LTP takes route in the hippocampus. Neuron 25:504–506. Knierim JJ (2002) Dynamic interactions between local surface cues, distal landmarks, and intrinsic circuitry in hippocampal place cells. J Neurosci 22: 6254–6264. Knierim JJ (2006) Neural representations of location outside the hippocampus. Learn Mem 13:405–415. Knierim JJ, Kudrimoti HS, McNaughton BL (1995) Place cells, head direction cells, and the learning of landmark stability. J Neurosci 15:1648–1659. Knierim JJ, Kudrimoti HS, McNaughton BL (1998) Interactions between idiothetic cues and external landmarks in the control of place cells and head direction cells. J Neurophysiol 80:425–446. Knierim JJ, Lee I, Hargreaves EL (2006) Hippocampal place cells: parallel input streams, subregional processing, and implications for episodic memory. Hippocampus 16:755–764.
HIPPOCAMPAL SUBFIELDS AND SPATIAL CONTEXT OF EVENTS
Knight R (1996) Contribution of human hippocampal region to novelty detection. Nature 383:256–259. Kobayashi T, Nishijo H, Fukuda M, Bures J, Ono T (1997) Task-dependent representations in rat hippocampal place neurons. J Neurophysiol 78:597–613. Kohonen T (1997) Associative Memory: A System Theoretical Approach. New York: Springer-Verlag. Lee I, Kesner RP (2002) Differential contribution of NMDA receptors in hippocampal subregions to spatial working memory. Nat Neurosci 5:162–168. Lee I, Kesner RP (2003a) Differential roles of dorsal hippocampal subregions in spatial working memory with short versus intermediate delay. Behav Neurosci 117:1044–1053. Lee I, Kesner RP (2003b) Time-dependent relationship between the dorsal hippocampus and the prefrontal cortex in spatial memory. J Neurosci 23:1517– 1523. Lee I, Kesner RP (2004a) Differential contributions of dorsal hippocampal subregions to memory acquisition and retrieval in contextual fear-conditioning. Hippocampus 14:301–310. Lee I, Kesner RP (2004b) Encoding versus retrieval of spatial memory: double dissociation between the dentate gyrus and the perforant path inputs into CA3 in the dorsal hippocampus. Hippocampus 14:66–76. Lee I, Rao G, Knierim JJ (2004a) A double dissociation between hippocampal subfields: differential time course of CA3 and CA1 place cells for processing changed environments. Neuron 42:803–815. Lee I, Hunsaker MR, Kesner RP (2005a) The role of hippocampal subregions in detecting spatial novelty. Behav Neurosci 119:145–153. Lee I, Jerman TS, Kesner RP (2005b) Disruption of delayed memory for a sequence of spatial locations following CA1- or CA3-lesions of the dorsal hippocampus. Neurobiol Learn Mem 84:138–147. Lee I, Yoganarasimha D, Rao G, Knierim JJ (2004b) Comparison of population coherence of place cells in hippocampal subfields CA1 and CA3. Nature 430: 456–459. Lee I, Griffin AL, Zilli EA, Eichenbaum H, Hasselmo ME (2006) Gradual translocation of spatial correlates of neuronal firing in the hippocampus toward prospective reward locations. Neuron 51:639–650. Lepage M, Habib R, Tulving E (1998) Hippocampal PET activations of memory encoding and retrieval: the HIPER model. Hippocampus 8:313–322. Leutgeb S, Leutgeb JK, Barnes CA, Moser EI, McNaughton BL, Moser MB (2005) Independent codes for spatial and episodic memory in hippocampal neuronal ensembles. Science 309:619–623. Leutgeb S, Leutgeb JK, Moser EI, Moser MB (2006) Fast rate coding in hippocampal CA3 cell ensembles. Hippocampus 16:765–774.
103
Leutgeb S, Leutgeb JK, Treves A, Moser M-B, Moser EI (2004) Distinct ensemble codes in hippocampal areas CA3 and CA1. Science 305:1295–1298. Leutgeb S, Ragozzino KE, Mizumori SJ (2000) Convergence of head direction and place information in the CA1 region of hippocampus. Neuroscience 100: 11–19. Levy WB (1996) A sequence predicting CA3 is a flexible associator that learns and uses context to solve hippocampal-like tasks. Hippocampus 6:579–590. Li XG, Somogyi P, Ylinen A, Buzsaki G (1994) The hippocampal CA3 network: an in vivo intracellular labeling study. J Comp Neurol 339:181–208. Lisman JE, Otmakhova NA (2001) Storage, recall, and novelty detection of sequences by the hippocampus: elaborating on the SOCRATIC model to account for normal and aberrant effects of dopamine. Hippocampus 11:551–568. Lisman JE, Talamini LM, Raffone A (2005) Recall of memory sequences by interaction of the dentate and CA3: a revised model of the phase precession. Neural Netw 18:1191–1201. Lorente de No R (1934) Studies in the structure of the cerebral cortex II. Continuation of the study of the ammonic system. J Psychol Neurol 46:113–177. Maren S, Fanselow MS (1997) Electrolytic lesions of the fimbria/fornix, dorsal hippocampus, or entorhinal cortex produce anterograde deficits in contextual fear conditioning in rats. Neurobiol Learn Mem 67:142– 149. Maren S, Holt W (2000) The hippocampus and contextual memory retrieval in Pavlovian conditioning. Behav Brain Res 110:97–108. Maren S, Holt WG (2004) Hippocampus and Pavlovian fear conditioning in rats: muscimol infusions into the ventral, but not dorsal, hippocampus impair the acquisition of conditional freezing to an auditory conditional stimulus. Behav Neurosci 118:97–110. Markus EJ, Barnes CA, McNaughton BL, Gladden VL, Skaggs WE (1994) Spatial information content and reliability of hippocampal CA1 neurons: effects of visual input. Hippocampus 4:410–421. Markus EJ, Qin YL, Leonard B, Skaggs WE, McNaughton BL, Barnes CA (1995) Interactions between location and task affect the spatial and directional firing of hippocampal neurons. J Neurosci 15:7079–7094. Marr D (1969) A theory of cerebellar cortex. J Physiol 202:437–470. Marr D (1971) Simple memory: a theory for archicortex. Philos Trans R Soc Lond B Biol Sci 262:23–81. McClelland JL, Goddard NH (1996) Considerations arising from a complementary learning systems perspective on hippocampus and neocortex. Hippocampus 6:654–665.
104
PLACE CELLS AND SPATIAL CONTEXT
McHugh TJ, Jones MW, Quinn JJ, Balthasar N, Coppari R, Elmquist JK, Lowell BB, Fanselow MS, Wilson MA, Tonegawa S (2007) Dentate gyrus NMDA receptors mediate rapid pattern separation in the hippocampal network. Science 317:94–99. McLamb RL, Mundy WR, Tilson HA (1988) Intradentate colchicine impairs acquisition of a two-way active avoidance response in a Y-maze. Neurosci Lett 94: 338–342. McNaughton BL, Barnes CA, Gerrard JL, Gothard K, Jung MW, Knierim JJ, Kudrimoti H, Qin Y, Skaggs WE, Suster M, Weaver KL (1996) Deciphering the hippocampal polyglot: the hippocampus as a path integration system. J Exp Biol 199:173–185. McNaughton BL, Barnes CA, Meltzer J, Sutherland RJ (1989) Hippocampal granule cells are necessary for normal spatial learning but not for spatially selective pyramidal cell discharge. Exp Brain Res 76:485– 496. McNaughton BL, Barnes CA, O’Keefe J (1983) The contributions of position, direction, and velocity to single unit activity in the hippocampus of freely moving rats. Exp Brain Res 52:41–49. McNaughton BL, Morris RGM (1987) Hippocampal synaptic enhancement and information storage within a distributed memory system. Trends Neurosci 10: 408–415. McNaughton BL, Nadel L (1990) Hebb-Marr networks and the neurobiological representation of action in space. In: Neuroscience and Connectionist Theory (Gluck MA, Rumelhart DE, eds.), pp 1–63. Hillsdale, NJ: Lawrence Erlbaum Associates. Mehta MR, Barnes CA, McNaughton BL (1997) Experience-dependent, asymmetric expansion of hippocampal place fields. Proc Natl Acad Sci USA 94: 8918–8921. Mehta MR, Quirk MC, Wilson MA (2000) Experiencedependent asymmetric shape of hippocampal receptive fields. Neuron 25:707–715. Mishkin M (1978) Memory in monkeys severely impaired by combined but not by separate removal of amygdala and hippocampus. Nature 273:297–298. Mishkin M, Vargha-Khadem F, Gadian DG (1998) Amnesia and the organization of the hippocampal system. Hippocampus 8:212–216. Mizumori SJ, Kalyani A (1997) Age and experiencedependent representational reorganization during spatial learning. Neurobiol Aging 18:651–659. Mizumori SJ, Ragozzino KE, Cooper BG, Leutgeb S (1999) Hippocampal representational organization and spatial context. Hippocampus 9:444–451. Mizumori SJY, Yeshenko O, Gill KM, Davis DM (2004) Parallel processing across neural systems: implications for a multiple memory system hypothesis. Neurobiol Learn Mem 82:278–298.
Moita MA, Rosis S, Zhou Y, LeDoux JE, Blair HT (2003) Hippocampal place cells acquire location-specific responses to the conditioned stimulus during auditory fear conditioning. Neuron 37:485–497. Morris RG (2001) Episodic-like memory in animals: psychological criteria, neural mechanisms and the value of episodic-like tasks to investigate animal models of neurodegenerative disease. Philos Trans R Soc Lond B Biol Sci 356:1453–1465. Morris RG, Garrud P, Rawlins JN, O’Keefe J (1982) Place navigation impaired in rats with hippocampal lesions. Nature 297:681–683. Moser E, Moser MB, Andersen P (1993) Spatial learning impairment parallels the magnitude of dorsal hippocampal lesions, but is hardly present following ventral lesions. J Neurosci 13:3916–3925. Muller RU, Kubie JL (1987) The effects of changes in the environment on the spatial firing of hippocampal complex-spike cells. J Neurosci 7:1951– 1968. Mumby DG, Gaskin S, Glenn MJ, Schramek TE, Lehmann H (2002) Hippocampal damage and exploratory preferences in rats: memory for objects, places, and contexts. Learn Mem 9:49–57. Myhrer T (1988) Exploratory behavior and reaction to novelty in rats with hippocampal perforant path systems disrupted. Behav Neurosci 102:356–362. Nadel L, Hardt O (2004) The spatial brain. Neuropsychology 18:473–476. Nadel L, Wilner J, Kurtz EM (1985) Cognitive maps and environmental context. In: Context and Learning (Blalsam PD, Tomie A, eds.), pp 385–406. Hillsdale, NJ: Lawrence Erlbaum Associates. Nakazawa K, Quirk MC, Chitwood RA, Watanabe M, Yeckel MF, Sun LD, Kato A, Carr CA, Johnston D, Wilson MA, Tonegawa S (2002) Requirement for hippocampal CA3 NMDA receptors in associative memory recall. Science 297:211–218. Nakazawa K, Sun LD, Quirk MC, Rondi-Reig L, Wilson MA, Tonegawa S (2003) Hippocampal CA3 NMDA receptors are crucial for memory acquisition of onetime experience. Neuron 38:305–315. O’Keefe J (1979) A review of the hippocampal place cells. Prog Neurobiol 13:419–439. O’Keefe J, Conway DH (1978) Hippocampal place units in the freely moving rat: why they fire where they fire. Exp Brain Res 31:573–590. O’Keefe J, Dostrovsky J (1971) The hippocampus as a spatial map: preliminary evidence from unit activity in the freely moving rat. Brain Res 34:171–175. O’Keefe J, Nadel L (1978) The Hippocampus as a Cognitive Map. Oxford: Clarendon Press. O’Keefe J, Speakman A (1987) Single unit activity in the rat hippocampus during a spatial memory task. Exp Brain Res 68:1–27.
HIPPOCAMPAL SUBFIELDS AND SPATIAL CONTEXT OF EVENTS
Olton DS, Branch M, Best PJ (1978) Spatial correlates of hippocampal unit activity. Exp Neurol 58:387–409. Ono T, Nakamura K, Nishijo H, Eifuku S (1993) Monkey hippocampal neurons related to spatial and nonspatial functions. J Neurophysiol 70:1516–1529. Ono T, Tamura R, Nakamura K (1991) The hippocampus and space: are there ‘‘place neurons’’ in the monkey hippocampus? Hippocampus 1:253–257. O’Reilly RC, McClelland JL (1994) Hippocampal conjunctive encoding, storage, and recall: avoiding a trade-off. Hippocampus 4:661–682. O’Reilly RC, Rudy JW (2000) Computational principles of learning in the neocortex and hippocampus. Hippocampus 10:389–397. Parkinson JK, Murray EA, Mishkin M (1988) A selective mnemonic role for the hippocampus in monkeys: memory for the location of objects. J Neurosci 8: 4159–4167. Parron C, Poucet B, Save E (2006) Cooperation between the hippocampus and the entorhinal cortex in spatial memory: a disconnection study. Behav Brain Res 170: 99–109. Petrovich GD, Canteras NS, Swanson LW (2001) Combinatorial amygdalar inputs to hippocampal domains and hypothalamic behavior systems. Brain Res Brain Res Rev 38:247–289. Phillips RG, LeDoux JE (1992) Differential contribution of amygdala and hippocampus to cued and contextual fear conditioning. Behav Neurosci 106:274–285. Ramon Y Cahal S (1911) Histology of the Nervous System of Man and Vertebrates, Vol. II. Oxford: Oxford University Press. Rempel-Clower NL, Zola SM, Squire LR, Amaral DG (1996) Three cases of enduring memory impairment after bilateral damage limited to the hippocampal formation. J Neurosci 16:5233–5255. Robertson RG, Rolls ET, Georges-Franc¸ois P (1998) Spatial view cells in the primate hippocampus: effects of removal of view details. J Neurophysiol 79:1145–1156. Rolls ET (1996) A theory of hippocampal function in memory. Hippocampus 6:601–620. Rolls ET (1999) Spatial view cells and the representation of place in the primate hippocampus. Hippocampus 9:467–480. Rolls ET, Kesner RP (2006) A computational theory of hippocampal function, and empirical tests of the theory. Prog Neurobiol 79:1–48. Rolls ET, Robertson RG, Georges-Francois P (1997) Spatial view cells in the primate hippocampus. Eur J Neurosci 9:1789–1794. Rolls ET, Treves A (1998) Neural Networks and Brain Function. Oxford: Oxford University Press. Rolls ET, Xiang JZ (2005) Reward-spatial view representations and learning in the primate hippocampus. J Neurosci 25:6167–6174.
105
Rondi-Reig L, Libbey M, Eichenbaum H, Tonegawa S (2001) CA1-specific N-methyl-D-aspartate receptor knockout mice are deficient in solving a nonspatial transverse patterning task. Proc Natl Acad Sci USA 98:3543–3548. Rudy JW (1993) Contextual conditioning and auditory cue conditioning dissociate during development. Behav Neurosci 107:887–891. Rudy JW, Huff NC, Matus-Amat P (2004) Understanding contextual fear conditioning: insights from a two-process model. Neurosci Biobehav Rev 28:675– 685. Rudy JW, O’Reilly RC (1999) Contextual fear conditioning, conjunctive representations, pattern completion, and the hippocampus. Behav Neurosci 113: 867–880. Rudy JW, O’Reilly RC (2001) Conjunctive representations, the hippocampus, and contextual fear conditioning. Cogn Affect Behav Neurosci 1:66–82. Samsonovich A, McNaughton BL (1997) Path integration and cognitive mapping in a continuous attractor neural network model. J Neurosci 17:5900–5920. Sargolini F, Fyhn M, Hafting T, McNaughton BL, Witter MP, Moser M-B, Moser EI (2006) Conjunctive representation of position, direction, and velocity in entorhinal cortex. Science 312:758–762. Scoville WB, Milner B (1957) Loss of recent memory after bilateral hippocampal lesions. J Neurol Neurosurg Psychiatry 20:11–21. Shapiro ML, Kennedy PJ, Ferbinteanu J (2006) Representing episodes in the mammalian brain. Curr Opin Neurobiol 16:701–709. Shapiro ML, Tanila H, Eichenbaum H (1997) Cues that hippocampal place cells encode: dynamic and hierarchical representation of local and distal stimuli. Hippocampus 7:624–642. Small SA, Nava AS, Perera GM, DeLaPaz R, Mayeux R, Stern Y (2001) Circuit mechanisms underlying memory encoding and retrieval in the long axis of the hippocampal formation. Nat Neurosci 4:442–449. Smith DM, Mizumori SJ (2006a) Learning-related development of context-specific neuronal responses to places and events: the hippocampal role in context processing. J Neurosci 26:3154–3163. Smith DM, Mizumori SJY (2006b) Hippocampal place cells, context, and episodic memory. Hippocampus 16:716–729. Spiers HJ, Burgess N, Maguire EA, Baxendale SA, Hartley T, Thompson PJ, O’Keefe J (2001) Unilateral temporal lobectomy patients show lateralized topographical and episodic memory deficits in a virtual town. Brain 124:2476–2489. Squire LR (1992) Memory and the hippocampus: a synthesis from findings with rats, monkeys, and humans. Psychol Rev 99:195–231.
106
PLACE CELLS AND SPATIAL CONTEXT
Sutherland RJ, McDonald RJ, Hill CR, Rudy JW (1989) Damage to the hippocampal formation in rats selectively impairs the ability to learn cue relationships. Behav Neural Biol 52:331–356. Suzuki WA, Clayton NS (2000) The hippocampus and memory: a comparative and ethological perspective. Curr Opin Neurobiol 10:768–773. Tanila H, Shapiro ML, Eichenbaum H (1997) Discordance of spatial representation in ensembles of hippocampal place cells. Hippocampus 7:613–623. Tilson HA, McLamb RL, Shaw S, Rogers BC, Pediaditakis P, Cook L (1988) Radial-arm maze deficits produced by colchicine administered into the area of the nucleus basalis are ameliorated by cholinergic agents. Brain Res 438:83–94. Tolman EC (1948) Cognitive maps in rats and men. Psychol Rev 55:189–208. Treves A, Rolls ET (1992) Computational constraints suggest the need for two distinct input systems to the hippocampal CA3 network. Hippocampus 2:189– 199. Treves A, Rolls ET (1994) Computational analysis of the role of the hippocampus in memory. Hippocampus 4:374–391. Tulving E (1972) Episodic and semantic memory. In: Organization of Memory (Tulving E, Donaldson W, eds.), pp 381–403. New York: Academic Press. Tulving E (2002) Episodic memory: from mind to brain. Annu Rev Psychol 53:1–25. Tulving E, Markowitsch HJ (1998) Episodic and declarative memory: role of the hippocampus. Hippocampus 8:198–204. Vanderwolf CH (1969) Hippocampal electrical activity and voluntary movement in the rat. Electroencephalogr Clin Neurophysiol 26:407–418. Vargha-Khadem F, Gadian DG, Watkins KE, Connelly A, Van Paesschen W, Mishkin M (1997) Differential effects of early hippocampal pathology on episodic and semantic memory. Science 277:376– 380. Vazdarjanova A, Guzowski JF (2004) Differences in hippocampal neuronal population responses to modifications of an environmental context: evidence for distinct, yet complementary, functions of CA3 and CA1 ensembles. J Neurosci 24:6489–6496. Vinogradova OS (2001) Hippocampus as comparator: role of the two input and two output systems of the hippocampus in selection and registration of information. Hippocampus 11:578–598.
Wallenstein GV, Eichenbaum H, Hasselmo ME (1998) The hippocampus as an associator of discontiguous events. Trends Neurosci 21:317–323. Wallenstein GV, Hasselmo ME (1997) GABAergic modulation of hippocampal population activity: sequence learning, place field development, and the phase precession effect. J Neurophysiol 78:393–408. Walsh TJ, Schulz DW, Tilson HA, Schmechel DE (1986) Colchicine-induced granule cell loss in rat hippocampus: selective behavioral and histological alterations. Brain Res 398:23–36. Wang XJ (2001) Synaptic reverberation underlying mnemonic persistent activity. Trends Neurosci 24:455– 463. Wiener SI, Korshunov VA (1995) Place-independent behavioural correlates of hippocampal neurones in rats. Neuroreport 7:183–188. Wiener SI, Korshunov VA, Garcia R, Berthoz A (1995) Inertial, substratal and landmark cue control of hippocampal CA1 place cell activity. Eur J Neurosci 7: 2206–2219. Wilson MA, McNaughton BL (1993) Dynamics of the hippocampal ensemble code for space. Science 261: 1055–1058. Wood ER, Dudchenko PA, Robitsek RJ, Eichenbaum H (2000) Hippocampal neurons encode information about different types of memory episodes occurring in the same location. Neuron 27:623–633. Xavier GF, Oliveira-Filho FJ, Santos AM (1999) Dentate gyrus–selective colchicine lesion and disruption of performance in spatial tasks: difficulties in ‘‘place strategy’’ because of a lack of flexibility in the use of environmental cues? Hippocampus 9:668–681. Young BJ, Fox GD, Eichenbaum H (1994) Correlates of hippocampal complex-spike cell activity in rats performing a nonspatial radial maze task. J Neurosci 14: 6553–6563. Yu X, Yoganarasimha D, Knierim JJ (2006) Backward shift of head direction tuning curves of the anterior thalamus: comparison with CA1 place fields. Neuron 52:717–729. Zeineh MM, Engel SA, Thompson PM, Bookheimer SY (2003) Dynamics of the hippocampus during encoding and retrieval of face-name pairs. Science 299:577–580. Zola-Morgan S, Squire LR, Amaral DG (1986) Human amnesia and the medial temporal region: enduring memory impairment following a bilateral lesion limited to field CA1 of the hippocampus. J Neurosci 6: 2950–2967.
7 Plasticity, Attention, and the Stabilization of Hippocampal Representations DAVID C. ROWLAND AND CLIFFORD G. KENTROS
The true art of memory is the art of attention. (Samuel Johnson, 1759) The consensus view among neuropsychologists is that the hippocampal formation (hippocampus proper and its interconnected regions in the medial temporal lobe) is largely responsible for the acquisition of what most people think of as memories: the conscious recollection of experiences, including facts, events, places, etc. On a more cellular level, when one records from hippocampal neurons of behaving animals, the most obvious firing correlate by far is the ‘‘place cell’’ phenomenon, reviewed elsewhere in this book. At the subcellular level, memory is generally thought to involve plastic changes in synaptic weights, most notably in the hippocampus. Even though the experiments done at each level are quite distinct, they all purport to study the same thing, and should therefore be mutually reinforcing with regard to shared attributes such as timescale. We therefore present a speculative hypothesis regarding hippocampal function that attempts to synthesize results from various levels of analysis, drawing on growing electrophysiological evidence that distinct firing correlates of hippocampal neurons can be differentially stored and expressed. One of the key challenges in the study of episodic memory is determining how the brain decides which features of an episode should be fated for long-term storage. We simply cannot remember everything. Indeed, it would be maladaptive to not be able to preferentially store biologically important information over the irrelevant ‘‘background noise’’ of the sen-
sorium. Recall, for example, the many woes of Dr. Luria’s patient with remarkably augmented memory (Luria, 1987). The brain performs what has been called ‘‘mnemonic selection’’ (Summerfield et al., 2006), preferentially selecting behaviorally relevant information for long-term storage. What neural processes could underlie such phenomena? This kind of selective attribution of neural resources has been called ‘‘attention’’ at least since the days of William James (1890): ‘‘Everyone knows what attention is. It is the taking possession by the mind, in clear and vivid form, of one out of what seem several simultaneously possible objects or trains of thought.’’ (pp 403–404) When one puts all of these pieces together one comes up with the hypothesis that attention (or some other executive process resembling it) may select particular stimuli for long-term memory storage by affecting the consolidation of plastic changes in synaptic weights in the hippocampal formation. This commonsense hypothesis is nevertheless very difficult to prove one way or another. First and foremost, the neural mechanisms of attention remain obscure. Like memory itself, attention cannot be definitively and directly measured electrophysiologically. Its presence or absence must be inferred from the relation of the firing patterns to the structure of a behavioral task; and it should also be confirmed by alterations in sensorimotor correlates such as a reduction in perceptual threshold and reaction time (Kastner and Ungerleider, 2000; Goldberg et al., 2006). Nevertheless, investigators working primarily in the primate visual system have made great strides in characterizing attentional effects
107
108
PLACE CELLS AND SPATIAL CONTEXT
on single neurons. A typical experiment involves comparing firing rates of visual neurons when their receptive fields are or are not relevant to a behavioral task, and cross-checking attention by measuring taskassociated decreases in perceptual threshold for that field (e.g., Cook and Maunsell, 2002). The problem is that such techniques do not translate well to recordings of hippocampal neurons in rodents for a variety of reasons: most rodents are small nocturnal animals in which vision is not a primary sensory modality. They are afoveate, which makes classic attention paradigms that require the animal to fixate while targets and distracters presented in the visual field impossible. They do not take well to head fixation— restraint is one of the more effective means to generate a large stress response in rodents (e.g., Pham et al., 2003). Moreover, although hippocampal neurons do sometimes fire in correlation to sensory input or motor output, the hippocampus is neither a sensory nor a motor structure. It is a supramodal association area with ‘‘receptive fields’’ that do not necessarily depend upon any single sensory modality (e.g., Quirk et al., 1990), and require the animal’s active participation to be revealed (Foster et al., 1989). It is worth recalling that much of the groundbreaking work in visual cortex was performed in anesthetized animals (e.g., Hubel and Wiesel, 1959). Many of the properties of the receptive fields of sensorimotor neurons are maintained in unconscious animals, which is certainly not the case for hippocampal neurons: place fields only occur during active locomotion (Foster et al., 1989) This means that it is exceedingly difficult to perform meaningful cross-checks of attention using sensorimotor correlates such as perceptual threshold and reaction time. This leaves just task-relevance as a yardstick, but this has a troubling lack of specificity, as performance of even the simplest behavioral tasks requires engaging a variety of cognitive processes. However, it must be stressed that these are more practical problems than evidence against attentional effects on hippocampal processing. It is doubtful that such a fundamental and highly adaptive neural process as the ability to select between concomitant neural processing streams evolved so recently as to follow the divergence of the primate from the rodent line. Indeed, the septohippocampal theta rhythm, discovered and studied primarily in rodents, is arguably one of the best electrophysiologically measurable candidates of attention. It correlates with behavioral states of ‘‘heightened attentiveness’’ such as active navigation and freezing in response to a predator. The problem is that it has many other justifiable behavioral correlates as well, including motivation and arousal (see Buzsaki, 2005, for a review that describes both sides of the coin). Specificity is the crux of the problem with attention—it
is a word that can sometimes mean so much that it means in fact very little. We therefore attempt to critically evaluate the following questions, each of which is necessary for our deceptively simple hypothesis to remain a possibility. (1) Are the firing patterns of hippocampal neurons learned? (2) Do they require plasticity like that seen in reduced hippocampal preparations? (3) Are the firing properties of hippocampal neurons attentionally modulated? Each of these questions gets increasingly difficult to address empirically, but an emerging body of work seems to point toward executive selection of firing patterns in hippocampal neurons.
ARE THE FIRING PATTERNS OF HIPPOCAMPAL NEURONS LEARNED? The spatially restricted firing of hippocampal neurons has led to the term place field (O’Keefe and Dostrovsky, 1971; O’Keefe, 1976), by analogy to the receptive fields of sensory structures. The similarities between receptive fields of sensory cortices and place fields are obvious, but the differences are perhaps more important. The receptive fields of sensory structures are largely thought to be hardwired, meaning that they stably respond to the same sensory inputs from early in life to death. Learning, experience, and attention can modify their receptive fields, but these modifications are often seen as changes in firing rate or small shifts in the cell’s preferred stimulus (e.g., Li et al., 2004; Fu et al., 2002). By contrast, changes in place fields are frequently much more pronounced; place fields ‘‘remap’’ (Muller and Kubie, 1987) to create a unique representation of a novel environment (e.g., Leutgeb et al., 2005), and place fields develop over the course of minutes in response to novel environments (Frank et al., 2004), suggesting a ‘‘softwired’’ system. Such a soft-wired system allows for tremendous memory space and engenders flexibility to the system, but also requires accessory plastic mechanisms to store the representation over the long term. The malleable and delayed nature suggests that the hippocampus actually constructs a behaviorally relevant representation from the entorhinal cortex and presumably stores this representation in sets of synaptic weights. The recent landmark studies demonstrating that the entorhinal cortex provides a representation of position, direction, and speed has challenged our understanding of spatial representations in the hippocampus (Hafting et al., 2005; McNaughton et al., 2006; Sargolini et al., 2006). If the entorhinal cortex provides such a clear spatial representation, what is the role of the hippocampus? Even though both the hippocampus and the
PLASTICITY OF HIPPOCAMPAL REPRESENTATIONS
entorhinal cortex represent space, the differences between grid cells and place fields of hippocampal neurons are striking. Entorhinal grid cells are topographically organized, show a regular arrangement over the surface of the environment, are active in all environments, and are almost immediately stable in novel environments (Hafting et al., 2005). This suggests that many of the properties of these neurons, such as the spacing, may be hardwired during development (McNaughton et al., 2006). By contrast, hippocampal place cells show little topographic organization (but see Hampson et al., 1999), most cells are silent in a given environment, and the cells take minutes to form stable fields. This suggests a more labile, soft-wired system. The ‘‘complete’’ or ‘‘global’’ remapping of hippocampal place cells is unique for receptive fields in the nervous system. By comparison, the striate and extrastriate visual cortex, the somatosensory cortex, the motor cortex, and the auditory cortex all exhibit a ‘‘topographic’’ representation along some dimension of the cortex (Mountcastle, 1997). Even higher order association areas like the prefrontal cortex exhibit some topography (Hagler and Sereno, 2006). From an anatomical perspective, the complete remapping of place cells is supported by the essentially random connectivity within hippocampal neurons. Neurons are no more likely to contact neighbors than more distant peers and the probability of contact is roughly 2%–5% (Buzsaki, 2006). This random connectivity sparks an analogy to random access memory (RAM) found in computers, where arrays of transistors allow memory storage without much regard to neighbor effects. By contrast, the entorhinal cortex has the sort of columnar structure found in most of the cortex where neurons are densely interconnected with their immediate neighbors (Witter and Moser, 2006). This dense local connectivity forms microcircuits within the structure, supporting a topographic organization (Yoshimura et al., 2005). The expansive memory space in the hippocampus could allow for environmental specificity (Jeffery and Burgess, 2006). These unique structural features of the hippocampus provided a foundation for early speculation that the hippocampus formed rapid associative memories without ‘‘organizing information in a complicated sense’’ (Marr, 1971). However, the hippocampus does seem to be reorganizing information over the course of minutes to even days, suggesting an added layer of complexity to the system. The wide difference in time course for development between the hippocampus and entorhinal cortex suggests that the hippocampus does not simply inherit a representation from the entorhinal cortex but reorganizes these inputs into a behaviorally useful representation. The grid cells of the entorhinal cortex form instantaneously within novel environments (Hafting
109
et al., 2005), whereas CA1 place fields development in about 4 min (Frank et al., 2004) and CA3 place fields take roughly 10 min (Leutgeb et al., 2004). CA1 place fields have been observed to change over the course of days, sometimes called ‘‘slow remapping’’ (Lever et al., 2002). This time difference far exceeds the time expected if the singular role of the hippocampus were to disambiguate the multipeaked firing of entorhinal cortex into a single-peaked place field; if this were true then the time course would be essentially indistinguishable from that of the grid cells (i.e., it would be almost instantaneous). The delayed time-course of development suggests that the hippocampus constructs a representation with experience in an environment. It is worth noting that the plasticity in the hippocampus is still largely referred to as ‘‘rapid’’ by some researchers, in part because of the traditional view that the hippocampus supports rapid associations (Marr, 1971), and because the systems-level consolidation in the neocortex presumably still takes even longer (Frank et al., 2006). However, it is still unclear how a system that takes minutes to form a stable representation supports memories that appear to be formed instantaneously.
DO THE FIRING PATTERNS OF HIPPOCAMPAL NEURONS REQUIRE PLASTICITY? One consequence of having a system capable of complete remapping is that there must be a physiological mechanism in place to cause the new representation to stick. The hippocampus is believed to support associative memories through plastic processes, such as long-term potentiation (LTP). If this hypothesis is true, what is the relationship between the formation of spatial representations and these plastic processes? Xu et al. (1998) showed that exposure to novel environments affects plastic processes. The experimenters first induced LTP and then placed the animal in either a novel or familiar environment. Placing the animal in a novel environment resulted in a complete abolition of LTP in the potentiated pathway, suggesting that novelty-induced changes in synaptic strength competed with the potentiated pathway for the plastic resources (Xu et al., 1998). What is particularly striking here is that the high-frequency stimulation used to induce LTP is frequently thought to be excessive and unnatural, yet this induced LTP was no match for the behaviorally induced changes. Similar results have been achieved in place cell studies, where LTP induction forced a remapping, but this remapping lasted only briefly before the representation switched to the previously stored, and behaviorally formed,
110
PLACE CELLS AND SPATIAL CONTEXT
representation (Dragoi et al., 2003). Recordings from neurons in behaving animals in conjunction with genetic or pharmacological manipulations have provided interesting insights into the interactions between plasticity and spatial representations. The role of NMDA receptor–mediated plasticity in the formation of spatial representations has been particularly well studied. Four studies have been conducted on place cells while blocking NMDA receptors through systemic injections (Kentros et al., 1998) or targeted genetic deletions of the NR1 subunit in either the CA1 region (McHugh et al., 1996) or the CA3 region (Nakazawa et al., 2002, 2003). In all of these cases, hippocampal neurons still showed spatially specific firing in novel environments (although McHugh et al. and Nakazawa et al. noted an increase in the size of the fields), yet these same manipulations create deficits in spatial tasks and an abolition of LTP. The newly established fields were stable over the short term for both the pharmacological and genetic studies, but after 24 hr, the fields in the pharmacological study were unstable (Fig. 7–1). The representation of space in the familiar environment, however, was unaffected over both the short and long term. This result marshals against the hypothesis that NMDA receptor activation, a process critical for LTP induction and normal learning in spatial tasks, is required for the formation of spatially specific firing of hippocampal neurons. The fact that place fields were formed and briefly main-
tained suggests that one form of plasticity is responsible for the formation of fields, while a second, NMDA receptor–mediated plasticity is responsible for creating representations that stick. In contrast to NMDA receptors, mice lacking particular AMPA receptor subunits display a near-complete lack of spatial specificity and coherence. Thus, AMPA receptors are probably involved in the formation of normal place fields (Yan et al., 2002). One intriguing hypothesis is that NMDAmediated plasticity encodes the distance between fields (Muller et al., 1996) and as such binds together the disparate locations that together constitute a space. Along the same lines, Dragoi and Buzsaki (2006) implicated NMDA receptors as a potential mechanism for bringing together members of a cell assembly. Consistent with the pharmacological place-cell result, rats given direct chlorophenylpiperazine (CPP) injections into the hippocampus perform the same as control animals after short delays, but are impaired in a 24-hr retention trial of a newly learned platform location in the water maze (McDonald et al., 2005). It has also been proposed that NMDA receptors predominantly mediate the temporary storage of novel events within pre-existing representations (Morris and Frey, 1997). The hippocampus is well suited for the role of novelty detection, because current sensory information arriving via the entorhinal cortex can be directly compared with stored representations in the CA3 collateral network. When a mismatch is detected
Figure 7–1. The NMDA receptor antagonist chlorophenylpiperazine (CPP) prevents the long-term maintenance of the place fields in a novel environment. Yellow pixels indicate areas of low firing and dark pixels indicate areas of high firing. The recording of cell 4 was lost on day 2. Note that the fields in the novel environment (B) develop normal fields and are maintained over the short term even after drug injection (compare D1W1 with D2W2) but remap on the following day (compare D1W2 with D2W1). Unstable fields yield low correlation scores while stable fields yield high correlation scores. Adapted with permission from Kentros et al. (1998).
PLASTICITY OF HIPPOCAMPAL REPRESENTATIONS
the representation can be updated or, if the change is large enough, it can force an entirely new output pattern, a ‘‘remapping’’ (Guzowski et al., 2004). Consistent with this hypothesis, the hippocampus is required for detecting novel stimuli (Knight, 1996). Both functional magnetic resonance imaging (fMRI) (Kumaran and Maguire, 2006) and electrophysiological recordings (Fyhn et al., 2002) suggest that the hippocampus responds to unexpected events. However, one would expect this representation to be fleeting, owing to rapid habituation (Vinogradova, 2001) and the kinetics of NMDA receptors, unless the event is reinforced with some neuromodulatory signal (Lisman and Grace, 2005; Fig. 7– 2). At the level of single-unit recordings, two questions remain. First, do salient novel events lead to long-lasting changes in the firing patterns of hippocampal neurons? And second, does the long-term maintenance of this updated representation require NMDA receptor–mediated plasticity? Such work highlights the importance of having multiple forms of plasticity in the hippocampus. Furthermore, parallel studies like these can allow for meaningful dialogue between disparate methodological approaches. In this case it appears that NMDA receptor–mediated (associative) plasticity is not necessary for acquisition of hippocampal representations but is required for either the consolidation of the representation (Kentros et al., 1998; McDonald et al., 2005), the acquisition of nonspatial information (Morris and Frey, 1997), or, potentially, both.
111
Figure 7–2. Hypothetical action of dopamine (DA). The arrow represents either artificial stimulation or a novel event and the height of the lines is the synaptic strength. A. Without dopamine or some other reinforcing neuromodulator, the novel event is briefly registered, but not stored long term. B. Gray box gives a hypothetical dopamine signal that overlaps and slightly precedes stimulation. When dopamine is available before stimulation in vitro (Huang and Kandel, 1995) or when dopamine agonists are applied before exposure to a novel environment (Li et al., 2003), dopamine prevents depotentiation. In humans, performance in declarative memory tasks is improved when midbrain dopamine regions are active prior to hippocampal regions (Adcock et al., 2006).
IS THERE ATTENTIONAL MODULATION OF HIPPOCAMPAL FIRING PROPERTIES? If hippocampal neurons are modulated by attention to salient or task-relevant stimuli, then one might expect the hippocampal representation to change as the task changes. When Markus and colleagues (1995) recorded hippocampal neurons during goal-directed behavior and random foraging in the same space, they found that the cells frequently remapped between the two conditions. Thus, goal-directed behaviors change the hippocampal representation of space. Place fields certainly do not suddenly all jump to the goal area as the animal learns a spatial task; rather, they tend to simply maintain a coherent structure, although a tendency of place fields to cluster around goal locations has been reported (Hollup et al., 2001). We will forego the more complicated issue of whether place cells ‘‘encode’’ behavioral goals and focus instead on the ways in which goal-directed behaviors change the firing of hippocampal neurons. In the Markus study it was not clear if the representation changed in response to task-relevant stimuli
and, more importantly, if the changes were correlated with behavioral performance. Thus, a critical second step is to draw meaningful correlations between the changes in place fields and behavioral performance. Perhaps the most fruitful paradigm for studying the behavioral effects of goals on place fields has been in using the forced-choice T-maze. When the animal is placed in the T-maze, it must choose the correct turn in a sequence in order to get a reward (usually a simple left–right alternation). Under these conditions, the hippocampal neurons not only fire directionally, as on the linear track (McNaughton et al., 1983), but many of the cells are also strongly modulated by which turn the animal is about to make, often called ‘‘prospective coding’’ (Frank et al., 2000; Wood et al., 2000; Ferbinteanu and Shapiro, 2003). In other words, a cell that fires along the stem portion of the maze before a leftward movement will frequently fire less or not at all when the animal traverses the same region prior to a rightward movement. Thus, place fields in the Tmaze are altered by the behavioral context.
112
PLACE CELLS AND SPATIAL CONTEXT
A recent study found that the fields in the T-maze actually shifted forward toward the goal locations as the animal acquired the task, but importantly did not shift past the goal location (Lee et al., 2006b). Previous reports have shown that place fields shift backwards along a linear track, perhaps owing to asymmetric nature of spike timing–dependent plasticity (Mehta et al., 2002). Apparently, the mnemonic demands of this task modify the representation to a point that it overrides the inherent plasticity observed during non-demanding tasks. Because there are mnemonic demands of the Tmaze task not present in simple track running, it can be concluded that the hippocampal neurons are not responding merely to the physical context but to the task demands as well. The above data demonstrate that hippocampal neurons respond to the behavioral demands of the task, but do the neurons preferentially represent task-relevant stimuli? An intriguing series of experiments progressively shifting the start points of the rat’s trajectory in a directed task found that place fields are frequently aligned relative to the start point, but then rapidly flip to the room-based frame of reference as the animal approaches the goal (Gothard et al., 1996). This finding shows that both reference frames can be expressed in the hippocampus; however, in this case, neither reference frame was made more relevant to the animal, and the location of the reward was a visible food well at the end of the track, perhaps contributing to the individual variation observed in this study (Gothard et al., 1996). The researchers then looked to see if the representation could be biased by having the roomaligned reference frame uniquely give the location of the goal (Redish et al., 2000; Rosenzweig et al., 2003). Under these conditions, the rats that accurately performed the task were able to switch to a room-aligned map as the animal approached the goal. Aged animals were impaired in the task; this result correlated with an inability to switch their representations between the two reference frames. Thus, hippocampal representations not only reflect multiple reference frames but can also be selectively employed for a given reference frame if it is more behaviorally relevant. The choice of reference frame is not strictly limited to cases in which the animal must choose one frame or another but also includes instances when one frame is made more stable than another (Knierim et al., 1995). In a related study, Zinyuk et al. (2000) trained rats to find a small sector of a rotating arena within a stable room for a food reward. The sector in this case was dictated by the stable room frame. When untrained animals were placed on a rotating arena, the fields were not aligned to either reference frame (when this is plotted it looks as if the place fields show no spatial preference at all). However, when the animal was
trained to find the unmarked sector in the task, the fields were frequently stable with respect to the room frame, although some cells were stable with respect to the arena frame only or both frames. The authors concluded that cells show stable fields only after some behavioral significance is assigned to the environment and this representation is biased toward, although not exclusive to, the relevant reference frame (Zinyuk et al., 2000). We have shown similar findings with mice (Kentros et al., 2004). Rats typically form stable spatial stable fields even if the available cues are inconsequential to the animal. Under essentially the same conditions, wild-type C57B16 mice form consistently stable place fields far less often than their larger relatives. However, when we trained mice to solve a spatial task, the place fields of the animals that learned the task stabilized. In the task, the animals are put into an environment for 5 min, after which negative stimuli (bright lights and car alarms, leading some to call it the ‘‘New York maze’’) are turned on (Fig. 7–3). The animal must learn to turn off the negative stimuli by going to a completely unmarked goal area, defined entirely by tracker output (analogous to the hidden platform in the Morris water maze). Whenever the animal enters this goal area, the negative stimuli turn off. After the animal stays in the goal area for a number of seconds (which is actually quite difficult, given the elevated locomotor drive of this strain) the negative stimuli stay off for a safe period of 2–3 min. After the safe period, during which the animal moves around the environment normally, the negative stimuli come back on again, requiring the animal to go back to the goal area again. Thirty minutes of such training was given to the animals daily for 5 days, after which we performed conjoint and disjoint rotations of the arena cues relative to the goal, to establish cue control and the spatial nature of the task, respectively. Mice can readily learn this task within days, enabling the recording of hippocampal units throughout the duration of the task. Remarkably, when we compared the fields of animals that were put into an environment simply for recording sessions without any task contingencies (‘‘no task’’) with those of animals that had to do this spatial task in the same environment, we found that the animals in the spatial-task group had on average significantly more stable place fields than animals that were put into the same environment without any task contingencies (Fig. 7–4). When we separated animals out by performance, the effect was even more marked: animals that performed well had place fields that were just as stable as those we and others see in rats, while animals that performed poorly had fields as unstable as those in the no-task group.
PLASTICITY OF HIPPOCAMPAL REPRESENTATIONS
113
Figure 7–3. Two of the behavioral groups from the Kentros et al. (2004) study. A. Notask animals were repeatedly placed back into the same environment. B. Spatial-task animals were required to find an unmarked goal location to turn of the bright lights and car alarms. Adapted from Kentros et al. (2004) with permission from Elsevier.
The only explanation for this result is that attention to the spatial cues forced a stable spatial map in animals performing the spatial task. Place field stability seems to require not only the same plastic processes that underlie in vitro LTP but also increased attention to the spatial cues. It would follow that attention to spatial cues might impinge upon the plastic mechanisms within the hippocampus in order to store the spatial layout of the environment over the long term. Attention appears to be mediated by modulatory influences outside of the hippocampus, as briefly outlined in the following section.
PLACE FIELDS ARE STABILIZED BY ATTENTION TO SALIENT STIMULI, BUT HOW? The results of our work with mice demonstrate that performance of animals covaries with the stability of their hippocampal representation of the task environment. This is exactly what one would expect from part of a spatial memory trace. The fact that no-task mice showed more stable fields over the short term bears resemblance to findings in aged rats (Barnes et al., 1997) and rats given systemic injections of NMDA receptor antagonists, but in this case the mice were essentially genetically identical, presumably perfectly capable learners with all the necessary plastic neural machinery. It might simply be that the animal has no reason to notice or care which shapes we paint on the
arena walls, so while it saw them and apparently encoded them (the place fields of no-task animals typically could follow arena rotations), it did not assign enough significance to them to consistently stabilize their hippocampal representation. Likewise, in the Zinyuk et al. and Rosenzweig et al. studies, the representation changed in favor of the attended reference frame. When the cues become task-relevant, the animal pays more attention to them, with attention in turn recruiting neuromodulatory input that facilitates the consolidation of hippocampal plasticity and, thereby, memory. Thus, we favor a model of long-term stability as facilitated by neuromodulation arising from subcortical structures. To this end, we found that injection of mice with a dopamine agonist improved their place field stability. Similarly, rats given D1 receptor antagonists showed difficulty in adjusting to context changes, as revealed by the lack of reliability and specificity in their place fields (Gill and Mizumori, 2006). Investigators studying human attentional networks have postulated three networks related to different aspects of attention: alerting, orienting, and executive attention. Alerting is the induction and maintenance of a high level of sensitivity to incoming stimuli; orienting is the selection of information from incoming inputs. Executive attention involves mechanisms for monitoring and resolving conflicts. Recent work has suggested that acetylcholine predominately mediates orienting, norepinephrine mediates alerting, and dopamine mediates executive attention (Posner and
114
PLACE CELLS AND SPATIAL CONTEXT
neuromodulators involved in attention or arousal and highlight in vitro and in vivo evidence suggesting that these neuromodulators aid in the consolidation of memories (relevant data are summarized in Table 7–1, and anatomical connections in Fig. 7–5). To help keep the two constructs of attention and memory delineated we also review evidence that attention updates the moment-by-moment activity of hippocampal neurons.
Neuromodulators Implicated in Attention, Plasticity, and Memory Consolidation Dopamine
Figure 7–4. A. Short-term stability (30 min) of wildtype mouse place cells is better than long-term stability (6þ hr), but neither is as high as that of animals performing a spatial task. B. Long-term stability is significantly improved when the animals are performing a spatial task. Heights of the bars give mean correlation scores. Adapted from Kentros et al. (2004) with permission form Elsevier.
Rothbart, 2007). The hippocampus is not typically considered a part of these attentional networks, but most studies of attention use perceptual cues that are briefly held in working memory and therefore would not necessarily engage the hippocampus (e.g., Posner, 1980). Indeed, recent evidence suggests that not only is the encoding of long-term memories influenced by attentional networks (Uncapher et al., 2006), but longterm memory also aids in directing attention through familiar scenes. For example, long-term associative links between objects and targets improve detection of objects while weakening detection of distracters (Moores et al., 2003), and the hippocampus is required for cueing salient information within complex visual scenes, or ‘‘contextual cueing’’ (Chun, 2000). This finding suggests a deep connection between attention and memory. In this section, we outline some of the
Many studies implicate dopamine in plasticity and memory. In the slice preparation dopamine inhibits depotentiation (prolongs the late phase of LTP) (Huang and Kandel, 1995; Otmakhova and Lisman, 1998; Fig. 7–2). Dopamine-mediated late-phase LTP requires the CREB/ATF transcription pathway; knocking out this pathway prevents acquisition of hippocampal-dependent tasks (Pittenger et al., 2002). Consistent with its role in LTP, dopamine agonists delivered either directly into the hippocampus of behaving rats or systemically improve performance in hippocampal-dependent tasks (Bernabeu et al., 1997; Bach et al., 1999). Activity of dopamine neurons in the midbrain is often coupled with the activity of hippocampal formation in human fMRI tasks and the level of activity of dopamine neurons during encoding predicted subsequent recall (Wittmann et al., 2005; Adcock et al., 2006). Taken together, the effects of dopamine can be viewed as facilitating long-term memory consolidation. Largely through the work of Schultz and colleagues, the firing patterns of the dopaminergic neurons of the ventral tegmental area (VTA) relative to reward expectancy and uncertainty have been characterized in great detail, incorporating physiological insight into reinforcement learning models (Schultz et al., 1997; Schultz and Dickinson, 2000). In reinforcement learning models, dopamine is used to reinforce actions that most often lead to rewards. Reinforcement learning may also be an appropriate model for place field stability and place field modulation in the hippocampus (Johnson and Redish, 2005). One problem with reinforcement learning is that it takes many trials (or, alternatively, either awake or sleeping reply of experiences) for the learning to take hold. One way dopamine reinforces attended experience is by responding to novel events. Li et al. (2003) found that LTP is induced in the hippocampus when animals are briefly exposed to novel environments, and this plasticity requires activation of D1/D5 dopamine receptors. Even though Li et al. used an entirely new
PLASTICITY OF HIPPOCAMPAL REPRESENTATIONS
115
Table 7–1. Neuromodulators and Their Functions at Various Levels of Analysis* Neuromodulator/ Study Type LTP
Place cell (single session)
Place cell (longitudinal studies) Hippocampaldependent memory
Attention
Dopamine
Norepinephrine
Acetylcholine
Promotes late-phase LTP (Huang and Kandel, 1995; Otmakhova and Lisman, 1996) D1/D5 antagonists reduce spatial specificity following context change (turning off lights) (Gill and Mizumori, 2006)
Promotes late-phase LTP (Katsuki et al., 1997)
Enhances LTP (Shinoe et al., 2005)
Alpha2 receptor agonists increase firing rates in a spatially nonspecific way (Tanila, 2001)
Is involved in theta oscillations (Buzsaki, 2002) Excessively rigid representations in response to novel environment (Ikonen et al., 2002) Sopolamine reduces the reproducibility of fields in familiar environments (Brazhnik et al., 2003) Consolidation of hippocampal-dependent memories (Power et al., 2003) Working/reference memory in radial maze (Everitt and Robbins, 1997) Loss of cholinergic neurons in Alzheimer patients
D1/D5 agonists improve stability while antagonists decrease stability (Kentros et al., 2004) D1/D5 agonists improve memory in aged animals (Bach et al., 1999) D1 antagonists impair performance in spatial tasks (Morris et al., 2003) Dopamine activity predicts declarative memory performance (Wittman et al., 2005) Reinforcement learning (Schultz and Dickinson, 2000) Executive attention (Posner and Rothbart, 2007)
Alpha2 receptor agonists and antagonists decrease stability (Tanila, 2001) Consolidation of memories (McGaugh, 2000) Retrieval of memories (Murchison et al., 2004)
Attention (Aston-Jones and Cohen, 2005) Arousal and collection of salient information (Berridge and Waterhouse, 2003) General arousal (Lee et al., 2006a) Alerting (Posner and Rothbart, 2007)
Optimizes processing of signals in attentionally demanding contexts (Sarter et al., 2005) Reflexive and voluntary attention (Beane and Marrocco, 2004) Orienting (Posner and Rothbart, 2007)
*This is not an exhaustive list, but rather gives an idea of the convergence and divergence of function. The role of neuromodulators in LTP depends heavily on the stimulation protocol (see Li et al., 2003, for an illustration).
environment, they propose that this could be a mechanism for incorporating novel experiences into preexisting representations or for detecting mismatches in stored versus current environments. The mice in our study, however, formed consistently stable fields in response to the behavioral conditions of the task, a result suggesting that pure novelty is not sufficient. The information about goals and saliency could come via the prefrontal cortex (PFC) (Hasselmo, 2005). The activity of prefrontal neurons is known to phase lock to hippocampal theta during choice points in mazes, presumably when working memory is needed (Jones and Wilson, 2005), and the PFC is required for acquisition of the hippocampal-dependent trace eyeblink conditioning task (Kronforst-Collins and Disterhoft, 1998; Weible et al., 2000). The PFC sends projections downward to dopaminergic neurons, which in turn send
projections to the hippocampus (Lisman and Grace, 2005). This sufficiently complex anatomy suggests that dopamine can signal information about novelty and behavioral relevance, which can allow for hippocampal representations to be stabilized in a behavioral conditional fashion (Fig. 7–5). Such goal-directed information is consistent with models of selective attention (Desimone and Duncan, 1995), but the fact that mouse place fields are stable following training in the spatial task is consistent with either simple increased vigilance or selective attention (Moser, 2004).
Adrenergic System Another potential candidate mechanism for the stability of place fields that has been implicated in the consolidation of memories is the adrenergic system
116
PLACE CELLS AND SPATIAL CONTEXT
Figure 7–5. Schematic representation of the major excitatory and neuromodulatory connections within the hippocampal formation. The view is dramatically simplified, as it leaves out several important connections, including serotonergic inputs from the raphe nuclei and the neuromodulatory projections to the entorhinal cortex. The amygdala provides direct glutamatergic drive to areas CA3 and CA1, but can also facilitate modulatory activity by recruiting subcortical structures. The connections between the subiculum and cortical structures are not assigned either ‘‘spatial’’ or ‘‘nonspatial,’’ because it is possible that the subiculum parses out spatial and nonspatial information (McNaughton et al., 2006). Ach, acetylcholine; DA, dopamine; LEC, lateral entorhinal cortex; MEC, medial entorhinal cortex; MS/DBB, medial septum and diagonal band of Broca; NE, norepinephrine; PFC, prefrontal cortex; VTA, ventral tegmental area. All red and blue arrows are excitatory connections.
(McGaugh, 2000, but for a potential role of NE in retrieval, see Murchison et al., 2004). Traditionally, the locus coeruleus–norepinephrine (LC-NE) system has been implicated in arousal or alerting (Lee et al., 2006a; Posner and Rothbart, 2007), but work with monkeys suggests a more complex role in control over behavior. This complexity might arise from its connections with the anterior cingulate and orbitofrontal cortices (AstonJones and Cohen, 2005). A related idea is that NE enhances or modulates the collection of salient sensory information (Berridge and Waterhouse, 2003). The LC accounts for the entire input of NE to the hippocampus and, while the pathways seem to reach all points in the hippocampal formation, the greatest innervation is to the dentate gyrus (Loy et al., 1980). Stimulation of the LC has been shown to facilitate LTP induced in the dentate gyrus in a protein synthesis–dependent fashion (Walling and Harley, 2004). Neither agonists nor antagonists of alpha-2 receptors improve place field stability, and agonists increase the firing rate of place cells in a spatially nonspecific fashion, especially in response to novel situations (Tanila, 2001). This result suggests that adrenergic modulation is required for appropriately responding to novel situations, and disrup-
tion of the balance of adrenergic tone can impair spatial representations.
Amygdala Emotionally laden episodic memories tend to be vivid and enduring. It is clear that the amygdala plays a role in facilitating consolidation of these episodes, although the mechanisms are unknown (Pelletier and Pare, 2004). Glutamatergic contacts from the amygdala to the hippocampus provide a direct excitatory route; however, because the amygdala is interconnected with subcortical structures (McGaugh, 2004), it is believed to facilitate the formation of emotionally charged memories in a relatively nonspecific way. To this end, stimulation of the basolateral amygdala within 30 min of tetanization induces a long-lasting potentiation in the perforant pathway to the dentate gyrus in vivo, and this LTP requires muscarinic and beta-adrenergic, but not dopaminergic, receptor activation (Frey et al., 2001). In behaving rats, direct infusion of amphetamines into the amygdala post-training enhances memory in the water maze, but lesioning the amygdala after inducing the enhancement does not block retention (Packard et al.,
PLASTICITY OF HIPPOCAMPAL REPRESENTATIONS
1994). This result suggests that stimulation of the amygdala facilitates learning in other structures, most likely the hippocampus in this case, and does not store the memory itself. In summary, the amygdala modulates memory consolidation in the hippocampus by promoting plastic mechanisms (McGaugh, 2000). The amygdala also seems to interact with attention through facilitating detection of emotionally salient stimuli (Phelps and LeDoux, 2005). For example, patients with amygdala damage can comprehend the meaning of both neutral and aversive words. However, these patients display an inability to rapidly pick out aversive words under conditions of reduced perceptual awareness (‘‘attentional blink’). This ability of the amygdala to rapidly detect aversive stimuli could indicate a role in facilitating attention (Anderson and Phelps, 2001). In rats, the amygdala and the rhinal cortices can become transiently synchronized following unexpected, behaviorally significant events (Paz et al., 2006).
Acetylcholine Finally, cholinergic neurons emanating from the basal forebrain could also be involved in formation and maintenance of hippocampal representations. In vitro evidence shows that acetylcholine can facilitate LTP in area CA1 (Huerta and Lisman, 1995; Shinoe et al., 2005). The role of cholinergic inputs in learning standard spatial tasks is disputed and depends on the method of lesioning. Complete lesions of the medial septum, which provides cholinergic inputs and direct inhibition to the hippocampus, produce deficits in spatial working memory and the water maze (Bannerman et al., 2004), but the more selective 192 IgGsaporin lesion of cholinergic neurons often fails to produce the same deficits (Baxter et al., 1995). Cholinergic modulation of the hippocampus is deeply intertwined with cholinergic (atropine-sensitive or type II) theta. The medial septum is required for hippocampal theta, and both cholinergic and GABAergic cells in the medial septum fire at theta frequencies (Buzsaki, 2002). Adding carbachol (an acetylcholine receptor agonist) to the slice preparation can create endogenous rhythms at the theta frequency in the hippocampus, so-called in vitro theta (Whittington and Traub, 2003; Traub et al., 2004). Blocking of cholinergic receptors does not prevent walking (atropine-insensitive or type I) theta, but abolishes type II theta (Buzsaki, 2002, 2006). The involvement of cholinergic modulation in type II theta is particularly interesting from an attention standpoint, because type II theta has been associated with salient sensory information (Bland, 1986). The cholinergic system has been generally implicated in optimizing the processing of signals in at-
117
tentionally demanding contexts (Sarter et al., 2005). The integrity of the cholinergic system and cholinergic modulation of certain cortical regions are necessary for normal attentional performance in rodents (McGaughy et al., 1999, 2000; Dalley et al., 2004) and monkeys (Witte et al., 1997). Microdialysis experiments have shown an increase in cortical acetylcholine during sustained visual-attention paradigms in rodents (Passetti et al., 2000; Dalley et al., 2001). Thus, in contrast to memory, the role of acetylcholine in attention is not disputed. This discrepancy suggests a potential dissociation between attention and memory, but may simply reflect the low attentional load of standard learning tasks (Sarter et al., 2003). The cholinergic system contributes to attention by optimizing the processing of signals in attentionally demanding contexts, or ‘‘orienting’’ (Sarter et al., 2005; Hasselmo, 2006). The cortical-projecting cholinergic neurons increase the efficacy of thalamic inputs while suppressing intracortical or feedback inputs, which suggests a role for acetylcholine in detecting current sensory information at the expense of stored representations (Gil et al., 1997). The long-lasting influence of cortical cholinergic inputs is to induce task-related plasticity favoring detection of the learned or attended stimuli (Weinberger, 2004). For example, lesions of cholinergic neurons prevent cortical map reorganization and impair learning in motor tasks (Conner et al., 2003). Likewise, acetylcholine enhances inputs from the entorhinal cortex to the CA3 region of the hippocampus, perhaps enhancing the contribution of incoming sensory information compared to the stored representation (Giocomo and Hasselmo, 2005). The integrity of the hippocampal–cholinergic pathway does not seem to be required for a normal spatial map per se. Lesioning of the medial septum has no significant effect on the space specificity of CA1 place fields, while slightly reducing the specificity of CA3 place fields (Mizumori et al., 1989). Instead, cholinergic modulation seems to be required for detecting changes in context (Leutgeb and Mizumori, 1999). Damage to the medial septum or change in the cholinergic tone largely prohibits remapping, resulting in an overly stable representation (Leutgeb and Mizumori, 1999; Ikonen et al., 2002). Scopolamine, a competitive antagonist of muscarinic acetylcholine receptors, reduces the reproducibility of fields in familiar environments (Brazhnik et al., 2003). The contribution of acetylcholine in modifying the place field activity in response to novel conditions is consistent with its role in detecting salient sensory information. Thus, each of the neuromodulatory systems implicated in attention are also implicated in plasticity and memory, providing a potential mechanism for the long-term storage of attended stimuli. The action of
118
PLACE CELLS AND SPATIAL CONTEXT
the various neuromodulators provides for storage of arousing stimuli and the selection of salient information at the expense of irrelevant information.
Attention Dynamically Alters Hippocampal Representations on a Time Scale Too Short for Long-Term Synaptic Plasticity Thus far, we have described the effects of attention on the long-term stability of hippocampal representations and would suggest that this long-term stability requires recruitment of neuromodulators from subcortical structures and long-lasting changes in synaptic strength. Why call this attention and not just learning? The hippocampus is capable of responding to all forms of sensory input, yet not all of these are stored long term. Thus, the hippocampus (and the structures associated with it) selects relevant from irrelevant information through a process like attention (Vinogradova, 2001). It is difficult to argue this point, however, without slipping into semantics, because attention and memory are so deeply interrelated that they are barely dissociable (Desimone, 1996). One way researchers studying sensory cortex have dissociated attention from learning is to place trained animals under different attentional demands while recording neurons. For example, monkeys trained to perform both a tactile and visual task show increased firing and synchrony among somatosensory neurons as the animal is performing the tactile task, compared with the same stimuli as the animal is performing the visual task (Steinmetz et al., 2000). These changes occur too quickly to be readily explained by longlasting changes in synaptic strength (although these processes probably occurred during the time it took to train the animal). Instead, it seems clear that some topdown modulation is facilitating these rapid changes (Fritz et al., 2003). Our claim that attention modulates hippocampal neurons might similarly be supported if hippocampal neurons display distinct and task-related firing as the animal is ‘‘on-task.’’ To discuss the moment-by-moment dynamics of hippocampal neurons in demanding contexts, it is first necessary to consider the firing of neurons in fixed environments. The phenomenon of the place field has led to the strong belief that hippocampal neurons are responding to the animal’s location in space in fixed environments. It has been estimated that of the minority of cells that show robust firing in an environment 98% respond to the animal’s position in space and are therefore classified as ‘‘place cells’’ (Harris et al., 2003). Furthermore, ensembles of place cells can be used to reconstruct the position of the animal. The reconstruction error frequently approaches the error of the tracking
system, suggesting a tight correlation between the ensemble of active neurons and the animal’s spatial position (Wilson and McNaughton, 1993). If we invert the question and ask whether location in space can predict the activity of hippocampal neurons, the answer is not definitive. Even in fixed environments, nearidentical passes through the firing field produce widely different firing patterns, sometimes called ‘‘overdispersion’’ (Fenton and Muller, 1998). On the linear track, where it is easier to control for the path the animal takes, the prediction based on position alone does a poor job of explaining the variance. Adding in theta modulation of place cells improves the prediction, but it is still not perfect (Harris et al., 2003). Olypher et al. (2002) suggest that rodents are switching between reference frames under simple foraging conditions. Simply put, they hypothesize that the animal uses primarily idiothetic cues to orient itself in space and sporadically updates this estimate using extramaze cues. This hypothesis is grounded in three studies: place fields can be aligned to different reference frames (Touretzky and Redish, 1996); both CA1 and dentate granule cells switch from idiothetic to allothetic cues 1 to 2 s after leaving a start box; and the place representation degrades over time in the absence of visual cues (Gothard et al., 1996, 2001). Olypher et al. suggest that this reference-frame switching can account for the overdispersion of place fields. However, spike trains of nearly all cortical areas can be described as random, and it would be nearly impossible to determine experimentally how frequently the animal shifts reference frames during random foraging. To substantiate their claim, Olypher et al. predicted that the overdispersion of place fields would be reduced when the animals are explicitly required to solve a spatial task in only one reference frame. Taking data from the Zinyuk et al. study, they showed a reduction in overdispersion for trained animals versus untrained animals as well as for trained animals navigating toward a goal versus trained animals that are not navigating towards the goal. This incremental improvement is reminiscent of studies on the somatosensory (Steinmetz et al., 2000) and visual systems (Reynolds et al., 2000), where the degree of synchrony and firing rate is dependent on the attentional state. Another potentially important factor is the relationship between active cells and their simultaneously active peer cells. Harris et al. (2003) made a similar discovery that the spikes of the neuron are only weakly predicted by space. The addition of theta modulation improved the prediction, but adding simultaneously recorded peer cells improved the prediction even more. The peer cells that positively predicted the firing of the target cell were considered members of the same assembly. The lifetime of these assemblies was esti-
PLASTICITY OF HIPPOCAMPAL REPRESENTATIONS
mated to be 25ms. This exceptionally brief lifetime exceeds simple sensory control and suggests top-down processes, which Harris et al. term ‘‘internal cognitive processes.’’ It is unclear whether these cell assemblies develop with time or how they are modulated with task demands, but they nonetheless present an intriguing idea that internal cognitive processes might reshape sensory inputs into behaviorally useful representations. In a related study, injection of delta 9tetrahydrocannabinol (THC) into rats spared the firing rate and place specificity of hippocampal neurons but abolished the cell assembly organization and impaired performance in a hippocampal-dependent task (Robbe et al., 2006). This result suggests that such temporal coordination, potentially mediated by attention, is an important property of hippocampal function.
SELECTIVE ATTENTION OR GENERAL AROUSAL? As mentioned above, our results with mice would be consistent with either the animal being in an aroused state or the animal selectively attending to the visual cues. Even though the currently available data are inconclusive, we speculate that the stabilization occurred through selective attention to the visual cues. In theories of selective attention, learning about a stimulus depends on attending to that stimulus. Attention is directed to the relevant stimuli, which become encoded into memory and thus diverted away from irrelevant stimuli, which are not encoded into memory (Makintosh, 1975; Dayan et al., 2000). General arousal, by contrast, facilitates consolidation in an entirely nonspecific way. Thus, the primary difference between the two competing hypotheses is that selective attention enables the preferential encoding of attending stimuli at the expense of unattended stimuli, while general arousal does not. Neuromodulation is thought to play a critical role in the process of selection by reinforcing behaviorally relevant perceived stimuli through heterosynaptic plasticity. At the level of single-unit recordings we would expect to see more representational bandwidth dedicated toward the attended stimuli at the expense of the unattended stimuli. The critical test, then, is to train the animals to associate the reward with one set of stimuli and ignore a second, while recording hippocampal units. The Zinyuk et al. study addressed this question by placing the arena frame in conflict with the room frame and, indeed, the authors saw an increase in the number of fields stable with respect to the room frame. However, it was unclear if this came at the expense of the arena frame; in fact, some fields were stable with respect to the arena frame after training. Furthermore, the two frames of
119
reference in the Zinyuk et al. study were both spatial and therefore interacting. A much better test would be to place spatial and nonspatial stimuli in opposition, as an intriguing ongoing study has done (Muzzio et al., 2005).
What Are the Options for Selection? Selective attention can broadly be defined as selection between ‘‘processing streams.’’ Thus, it is necessary to define the features that the hippocampus can select. The hippocampus is a supramodal structure capable of responding to all modalities in isolation as well as cross-modal stimuli. The rabbit hippocampus seems to respond robustly to presentation of stimuli such as tone clicks. By contrast, the proportion of rat hippocampal neurons responding to simple tones, odors, tastes, etc. is quite small, but can become much larger as these simple stimuli are set in relation to other cues (Eichenbaum, 1997). At the level of single-unit recordings from behaving animals, place fields also respond to nongeometric features of the environment. Hippocampal neurons can: maintain place-specific firing when the lights are turned off (Quirk et al., 1990), remap when the color of a cue card is changed (Kentros et al., 1998), remap when the color of the floor paper is changed (Jeffery and Anderson, 2003), fire directionally on a linear track (McNaughton et al., 1983), remap if the animal must perform a behavioral task in the environment (Markus et al., 1995), and so on. In short, it appears that the hippocampus can encode multiple ‘‘maps’’ for the same environment according to the spatial context (Touretsky and Redish, 1996; Redish and Tourestky, 1997). The problem with defining the hippocampus as a ‘‘supramodal’’ structure (true as that description may be) is that it is a rather vague description of a structure that has just one major cortical input: the entorhinal cortex. Indeed, a large proportion of the selection may well occur in the rhinal cortices, before the information even reaches the hippocampus (de Curtis and Pare, 2004). Thus, the critical test is to determine the type of information relayed to the hippocampus from the entorhinal cortex. Recent experimental efforts have shown that neurons in the medial entorhinal cortex are strongly modulated by position in space whereas neurons in the lateral entorhinal cortex are not (Hargreaves et al., 2005). It is presumed that the lateral enthorhinal cortex provides nonspatial information (e.g., object representations) to the hippocampus (Knierim et al., 2006). The entorhinal cortex, therefore, provides two input streams: one spatial and one nonspatial. The spatial and nonspatial information are then mixed to form object–place associations within the hippocampus
120
PLACE CELLS AND SPATIAL CONTEXT
(Knierim et al., 2006). The selection might also follow this spatial and nonspatial divide. If spatial stimuli are more behaviorally relevant to the animal, as they were for our mice, then we would expect the representation to heavily favor this spatial information. If nonspatial information is more behaviorally relevant, then we would expect the representation to be biased toward the relevant nonspatial stimuli and, at least in theory, come at the expense of the spatial information (Makintosh, 1975; Dayan et al., 2000).
Are Selected Nonspatial Stimuli Mapped? A plausible hypothesis is that the hippocampus uses the spatial signals from the entorhinal cortex to construct a context-specific scaffolding onto which information about items, objects, or context are overlaid. Under this view, space holds a distinct but not exclusive role in hippocampal processing. An obvious prediction from this hypothesis is that the gain of the signal will be altered as the context is changed slightly, but not enough to force a complete or global remapping. Such modulations of the firing rates have been observed following a contextual change (Leutgeb et al., 2005), after the animal is trained in a fear conditioning paradigm (Moita et al., 2003), or when an odor is added to the behavioral testing room (Anderson and Jeffery, 2003). The working model goes as follows: the hippocampus takes the multipeaked firing of the grid cells and collapses it into a single peak (or at least fewer peaks) and adds the nonspatial information from the lateral entorhinal cortex (Knierim et al., 2006). Small changes in context are registered as changes in firing rates while large changes in context are registered as a shift in the grid orientation and a delayed, but concomitant, complete remapping of hippocampal place fields occurs (McNaughton et al., 2006).
CONCLUSIONS At the beginning of the chapter we put forward a simple speculative hypothesis attempting to link memory processes operating at the cellular, systems, and cognitive levels. Simply put, we posit that attention aids memory formation by providing neuromodulatory input that turns transient, homosynaptic plasticity to long-lasting heterosynaptic plasticity. We then attempted to evaluate the feasibility of the model by using hippocampal place fields as a ‘‘model memory trace’’ and asking the following questions: (1) Are they learned? (2) Are they formed via plastic processes resembling those seen in vitro? (3) Are their firing patterns modulated by attention? Place cells do seem to be a good cellular model of hippocampal memory, as
having a normal, stable place field representation correlates with spatial memory performance. Transgenic mice and aged rats that have poor spatial memory and impaired plasticity also have abnormal place fields, typically being more unstable and/or less specific. The balance of evidence suggests that place fields require experience to be constructed, and that plasticity is involved in this process, so the first two questions can be answered positively even though the mechanistic details remain elusive. The last part of the hypothesis, the role of attention in the firing patterns of hippocampal neurons, remains more problematic. This really is not surprising because this part deals with cognitive rather than cellular mechanisms, and our mechanistic understanding of cognitive neuroscience is in its infancy relative to that of cellular neuroscience. Even though the various neurotransmitter systems implicated in attention can also be shown to have effects on plasticity, this is not proof that attention modulates plasticity, let alone memory. Nevertheless, while there is surely no ‘‘smoking gun’’ directly and specifically implicating attention in the firing of hippocampal neurons, one can still find a variety of indirect experimental data to support this premise. Indeed, the hypothesis arose from an attempt to explain our experimental data indicating that optimal stabilization of mouse place fields only happens in animals successfully performing a spatial task, i.e., when the animal is paying attention to the available spatial information. While this may remain the most convincing result, there are indirectly supportive data from many different groups. First of all, place-specific firing requires active navigation, thus place fields are not an automatic calculation of the sensory information available in the field. Place fields can remap as the animal switches between different tasks in the same constant environment, which implies that the map shifts as the animal shifts its attention to perform the second task. When animals are presented with distinct reference frames in opposition to each other (e.g., room versus arena), both reference frames have their own place cell map. Significantly, the cells appear to shift as one or the other reference frames is made behaviorally relevant. It is difficult to avoid the conclusion that the place fields shift as the animal shifts its attention from the room cues to the arena cues. Moreover, hippocampal neurons appear to encode more than just place. Not only do hippocampal neurons fire in response to a variety of nonspatial stimuli, there is also a large amount of unexplained variance in a cell’s firing patterns even when the animal is in the place field. Unexplained variance remains to be elucidated, but a potential explanation for such data is that the more the animal is currently attending to its position in space, the more spatial a particular pyramidal neuron is.
PLASTICITY OF HIPPOCAMPAL REPRESENTATIONS
Thus, we are left with a tantalizing picture, a veritable compendium of correlations that support the idea of some form of executive selection of the firing patterns of hippocampal neurons without actually proving it. It is admittedly not a hypothesis overly encumbered with direct electrophysiological evidence, and there are serious structural limitations to our ability to rectify this. What we call attention is so vaguely defined that it can be stretched to fit most behavioral results, and it is very challenging to specify it further. Nevertheless, the balance of evidence suggests that ignoring such executive influences in hippocampal recordings from awake, behaving animals may be in the end just as problematic as studying them overtly: just because it is very difficult to isolate such effects does not mean they are not there.
References Adcock RA, Thangavel A, Whitfield-Gabrieli S, Knutson B, Gabrieli JDE (2006) Reward-motivated learning: mesolimbic activation precedes memory formation. Neuron 50:507–517. Anderson AK, Phelps EA (2001) Lesions of the human amygdala impair enhanced perception of emotionally salient events. Nature 411:305–309. Anderson MI, Jeffery KJ (2003) Heterogeneous modulation of place cell firing by changes in context. J Neurosci 23:8827–8835. Aston-Jones G, Cohen JD (2005) Adaptive gain and the role of the locus coeruleus–norepinephrine system in optimal performance. J Comp Neurol 493:99–110. Bach ME, Barad M, Son H, Zhuo M, Lu YF, Shih R, Mansuy I, Hawkins RD, Kandel ER (1999) Agerelated defects in spatial memory are correlated with defects in the late phase of hippocampal long-term potentiation in vitro and are attenuated by drugs that enhance the cAMP signaling pathway. Proc Natl Acad Sci USA 96:5280–5285. Bannerman DM, Matthews P, Deacon RMJ, Rawlins JNP (2004) Medial septal lesions mimic effects of both selective dorsal and ventral hippocampal lesions. Behav Neurosci 118:1033–1041. Barnes CA, Suster MS, Shen J, McNaughton BL (1997) Multistability of cognitive maps in the hippocampus of old rats. Nature 388:272–275. Baxter MG, Bucci DJ, Gorman LK, Wiley RG, Gallagher M (1995) Selective immunotoxic lesions of basal forebrain cholinergic cells: effects on learning and memory in rats. Behav Neurosci 109:714–722. Bernabeu R, Bevilaqua L, Ardenghi P, Bromberg E, Schmitz P, Bianchin M, Izquierdo I, Medina JH (1997) Involvement of hippocampal cAMP/cAMPdependent protein kinase signaling pathways in a late memory consolidation phase of aversively motivated
121
learning in rats. Proc Natl Acad Sci USA 94:7041– 7046. Beane M, Marrocco RT (2004) Norepinephrine and acetylcholine mediation of the components of reflexive attention: implications for attention deficit disorders. Prog Neurobiol 74:167–181. Berridge CW, Waterhouse BD (2003) The locus coeruleus–noradrenergic system: modulation of behavioral state and state-dependent cognitive processes. Brain Res Rev 42:33–84. Bland BH (1986) The physiology and pharmacology of hippocampal formation theta rhythms. Prog Neurobiol 26:1–54. Brazhnik ES, Muller RU, Fox SE (2003) Muscarinic blockade slows and degrades the location-specific firing of hippocampal pyramidal cells. J Neurosci 23: 611–621. Buzsaki G (2002) Theta oscillations in the hippocampus. Neuron 33:325–340. Buzsaki G (2005) Theta rhythm of navigation: link between path integration and landmark navigation, episodic and semantic memory. Hippocampus 15:827– 840. Buzsaki G (2006) Rhythms of the Brain. Oxford: Oxford University Press. Chun MM (2000) Contextual cueing of visual attention. Trends Cogn Sci 4:170–178. Conner JM, Culberson A, Packowski C, Chiba AA, Tuszynski MH (2003) Lesions of the basal forebrain cholinergic system impair task acquisition and abolish cortical plasticity associated with motor skill learning. Neuron 38:819–829. Cook EP, Maunsell JHR (2002) Attentional modulation of behavioral performance and neuronal responses in middle temporal and ventral intraparietal areas of macaque monkey. J Neurosci 22:1994–2004. Dalley JW, McGaughy J, O’Connell MT, Cardinal RN, Levita L, Robbins TW (2001) Distinct changes in cortical acetylcholine and noradrenaline efflux during contingent and noncontingent performance of a visual attentional task. J Neurosci 21:4908–4914. Dalley JW, Theobald DE, Bouger P, Chudasama Y, Cardinal RN, Robbins TW (2004) Cortical cholinergic function and deficits in visual attentional performance in rats following 192 IgG-saporin-induced lesions of the medial prefrontal cortex. Cereb Cortex 14:922–932. Dayan P, Kakade S, Montague PR (2000) Learning and selective attention. Nat Neurosci 3:1218–1223. de Curtis M, Pare D (2004) The rhinal cortices: a wall of inhibition between the neocortex and the hippocampus. Prog Neurobiol 74:101–110. Desimone R (1996) Neural mechanisms for visual memory and their role in attention. Proc Natl Acad Sci 93:13494–13499.
122
PLACE CELLS AND SPATIAL CONTEXT
Desimone R, Duncan J (1995) Neural Mechanisms of Selective Visual Attention. Annual Review of Neuroscience 18:193–222. Dragoi G, Buzsaki G (2006) Temporal Encoding of Place Sequences by Hippocampal Cell Assemblies. Neuron 50:145–157. Dragoi G, Harris KD, Buzsaki G (2003) Place representation within hippocampal networks is modified by long-term potentiation. Neuron 39:843–853. Eichenbaum H (1997) To cortex: thanks for the memories. Neuron 19:481–484. Everitt BJ, Robbins TW (1997) Central cholinergic systems and cognition. Annu Rev Neurosci 48:649–684. Fenton AA, Muller RU (1998) Place cell discharge is extremely variable during individual passes of the rat through the firing field. Proc Natl Acad Sci USA 95: 3182–3187. Ferbinteanu J, Shapiro ML (2003) Prospective and retrospective memory coding in the hippocampus. Neuron 40:1227–1239. Foster T, Castro C, McNaughton B (1989) Spatial selectivity of rat hippocampal neurons: dependence on preparedness for movement. Science 244:1580–1582. Frank LM, Brown EN, Wilson M (2000) Trajectory encoding in the hippocampus and entorhinal cortex. Neuron 27:169–178. Frank LM, Brown EN, Stanley GB (2006) Hippocampal and cortical place cell plasticity: Implications for episodic memory. Hippocampus 16: 775–784. Frank LM, Stanley GB, Brown EN (2004) Hippocampal plasticity across multiple days of exposure to novel environments. J Neurosci 24:7681–7689. Frey S, Bergado-Rosado J, Seidenbecher T, Pape H-C, Frey JU (2001) Reinforcement of early long-term potentiation (early-LTP) in dentate gyrus by stimulation of the basolateral amygdala: heterosynaptic induction mechanisms of late-LTP. J Neurosci 21: 3697–3703. Fritz J, Shamma S, Elhilali M, Klein D (2003) Rapid taskrelated plasticity of spectrotemporal receptive fields in primary auditory cortex. Nat Neurosci 6:1216– 1223. Fu, Y-X, Djupsund K, Gao H, Hayden B, Shen K, Dan Y. (2002) Temporal Specificity in the Cortical Plasticity of Visual Space Representation. Science 296: 1999– 2003. Fyhn M, Molden S, Hollup S, Moser M-B, Moser EI (2002) Hippocampal neurons responding to first-time dislocation of a target object. Neuron 35:555–566. Gil Z, Connors BW, Amitai Y (1997) Differential regulation of neocortical synapses by neuromodulators and activity. Neuron 19:679–686. Gill K, Mizumori S (2006) Context-dependent modulation by D-1 receptors: differential effects in hippocampus and striatum. Behav Neurosci 120:377–392.
Giocomo L, Hasselmo ME (2005) Nicotinic modulation of glutamatergic synaptic transmission in region CA3 of the hippocampus. Eur J Neurosci 22:1349–1356. Goldberg ME, Bisley JW, Powell KD, Gottlieb J (2006) Saccades, salience and attention: the role of the lateral intraparietal area in visual behavior. Prog Brain Res 155:157–175. Gothard KM, Hoffman KL, Battaglia FP, McNaughton BL (2001) Dentate gyrus and CA1 ensemble activity during spatial reference frame shifts in the presence and absence of visual input. J Neurosci 21:7284–7292. Gothard KM, Skaggs WE, Moore KM, McNaughton BL (1996) Binding of hippocampal CA1 neural activity to multiple reference frames in a landmark-based navigation task. J Neurosci 16:823–835. Guzowski JF, Knierim JJ, Moser EI (2004) Ensemble dynamics of hippocampal regions CA3 and CA1. Neuron 44:581–584. Hafting T, Fyhn M, Molden S, Moser M-B, Moser EI (2005) Microstructure of a spatial map in the entorhinal cortex. Nature 436:801–806. Hagler JDJ, Sereno MI (2006) Spatial maps in frontal and prefrontal cortex. NeuroImage 29:567–577. Hampson RE, Simeral JD, Deadwyler SA (1999) Distribution of spatial and nonspatial information in dorsal hippocampus. Nature 402:610–614. Hargreaves EL, Rao G, Lee I, Knierim JJ (2005) Major dissociation between medial and lateral entorhinal input to dorsal hippocampus. Science 308:1792–1794. Harris KD, Csicsvari J, Hirase H, Dragoi G, Buzsaki G (2003) Organization of cell assemblies in the hippocampus. Nature 424:552–556. Hasselmo ME (2005) A model of prefrontal cortical mechanisms for goal-directed behavior. J Cogn Neurosi 17:1115–1129. Hasselmo ME (2006) The role of acetylcholine in learning and memory. Curr Opin Neurobiol 16:710– 715. Hollup SA, Molden S, Donnett JG, Moser M-B, Moser EI (2001) Accumulation of hippocampal place fields at the goal location in an annular watermaze task. J Neurosci 21:1635–1644. Huang Y, Kandel ER (1995) D1/D5 receptor agonists induce a protein synthesis-dependent late potentiation in the CA1 region of the hippocampus. Proc Natl Acad Sci USA 92:2446–2450. Hubel D, Wiesel T (1959) Receptive fields of single neurones in the cat’s striate cortex. J Physiol 148: 574–591. Huerta PT, Lisman JE (1995) Bidirectional synaptic plasticity induced by a single burst during cholinergic theta oscillation in CA1 in vitro. Neuron 15:1053– 1063. Ikonen S, McMahan R, Gallagher M, Eichenbaum H, Tanila H (2002) Cholinergic system regulation of
PLASTICITY OF HIPPOCAMPAL REPRESENTATIONS
spatial representation by the hippocampus. Hippocampus 12:386–397. James W (1890) Principles of Psychology. Macmillan and Co, London 1901. Orignially published by Henry Holt Co 1890. Jeffery KJ, Anderson MI (2003) Dissociation of the geometric and contextual influences on place cells. Hippocampus 13:868–872. Jeffery KJ, Burgess N (2006) A metric for the cognitive map: found at last? Trends Cogn Sci 10:1–3. Johnson A, Redish AD (2005) Hippocampal replay contributes to within session learning in a temporal difference reinforcement learning model. Neural Networks 18:1163–1171. Jones MW, Wilson MA (2005) Theta rhythms coordinate hippocampal–prefrontal interactions in a spatial memory task. PLoS Biol 3:e402. Kastner S, Ungerleider LG (2000) Mechanisms of visual attention in the human cortex. Annu Rev Neurosci 23: 315–341. Katsuki H, Izumi Y, Zorumski CF (1997) Noradrenergic regulation of synaptic plasticity in the hippocampal CA1 region. J Neurophys 77:3013–3020. Kentros C, Hargreaves E, Hawkins RD, Kandel ER, Shapiro M, Muller RV (1998) Abolition of long-term stability of new hippocampal place cell maps by NMDA receptor blockade. Science 280:2121–2126. Kentros CG, Agnihotri NT, Streater S, Hawkins RD, Kandel ER (2004) Increased attention to spatial context increases both place field stability and spatial memory. Neuron 42:283–295. Knierim JJ, Kudrimoti HS, McNaughton BL (1995) Place cells, head direction cells, and the learning of landmark stability. J Neurosci 15:1648–1659. Knierim JJ, Lee I, Hargreaves EL (2006) Hippocampal place cells: parallel input streams, subregional processing, and implications for episodic memory. Hippocampus 16:755–764. Knight RT (1996) Contribution of human hippocampal region to novelty detection. Nature 383:256–259. Kronforst-Collins MA, Disterhoft JF (1998) Lesions of the caudal area of rabbit medial prefrontal cortex impair trace eyeblink conditioning. Neurobiol Learn Mem 69:147–162. Kumaran D, Maguire EA (2006) An unexpected sequence of events: mismatch detection in the human hippocampus. PLoS Biol 4:e424. Lee AW, Devidze N, Pfaff DW, Zhou J (2006a) Functional genomics of sex hormone–dependent neuroendocrine systems: specific and generalized actions in the CNS. Prog Brain Res 158:243–272. Lee I, Griffin AL, Zilli EA, Eichenbaum H, Hasselmo ME (2006b) Gradual translocation of spatial correlates of neuronal firing in the hippocampus toward prospective reward locations. Neuron 51:639–650.
123
Leutgeb S, Leutgeb JK, Barnes CA, Moser EI, McNaughton BL, Moser M-B (2005) Independent codes for spatial and episodic memory in hippocampal neuronal ensembles. Science 309:619–623. Leutgeb S, Leutgeb JK, Treves A, Moser M-B, Moser EI (2004) Distinct ensemble codes in hippocampal areas CA3 and CA1. Science 305:1295–1298. Leutgeb S, Mizumori SJY (1999) Excitotoxic septal lesions result in spatial memory deficits and altered flexibility of hippocampal single-unit representations. J Neurosci 19:6661–6672. Lever C, Wills T, Cacucci F, Burgess N, O’Keefe J (2002) Long-term plasticity in hippocampal placecell representation of environmental geometry. Nature 416:90–94. Li W, Pie¨ch V, Gilbert CD (2004) Perceptual learning and top-down influences in primary visual cortex. Nat Neurosci 7: 651–657 Li S, Cullen WK, Anwyl R, Rowan MJ (2003) Dopaminedependent facilitation of LTP induction in hippocampal CA1 by exposure to spatial novelty. Nat Neurosci 6:526–531. Lisman JE, Grace AA (2005) The hippocampal-VTA loop: controlling the entry of information into longterm memory. Neuron 46:703–713. Loy R, Koziell D, Lindsey J, Moore R (1980) Noradrenergic innervation of the adult rat hippocampal formation. J Comp Neurol 189:699–710. Luria A (1987) The Mind of the Mnemonist: A Little Book about a Vast Memory. Cambridge, MA: Harvard University Press. Makintosh N (1975) A theory of attention: variations in the associability of stimuli with reinforcement. Psychol Rev 82:276–298. Markus EJ, Qin YL, Leonard B, Skaggs WE, McNaughton BL, Barnes CA (1995) Interactions between location and task affect the spatial and directional firing of hippocampal neurons. J Neurosci 15:7079–7094. Marr D (1971) Simple memory: a theory for archicortex. Philos Trans R Soc B 262: 23–81. McDonald RJ, Hong NS, Craig LA, Holahan MR, Louis M, Muller RU (2005) NMDA-receptor blockade by CPP impairs post-training consolidation of a rapidly acquired spatial representation in rat hippocampus. Eur J Neurosci 22:1201–1213. McGaugh JL (2000) Memory—a century of consolidation. Science 287:248–251. McGaugh JL (2004) The amygdala modulates the consolidation of memories of emotionally arousing experiences. Annu Rev Neurosci 27:1–28. McGaughy J, Decker MW, Sarter M (1999) Enhancement of sustained attention performance by the nicotinic acetylcholine receptor agonist ABT-418 in intact but not basal forebrain-lesioned rats. Psychopharmacology 144:175–182.
124
PLACE CELLS AND SPATIAL CONTEXT
McGaughy J, Everitt BJ, Robbins TW, Sarter M (2000) The role of cortical cholinergic afferent projections in cognition: impact of new selective immunotoxins. Behav Brain Res 115:251–263. McHugh TJ, Blum KI, Tsien JZ, Tonegawa S, Wilson MA (1996) Impaired Hippocampal Representation of Space in CA1-Specific NMDAR1 Knockout Mice. Cell 87:1339–1349. McNaughton BL, Barnes CA, O’Keefe J (1983) The contributions of position, direction, and velocity to single unit activity in the hippocampus of freely moving rats. Exp Brain Res 52:41–49. McNaughton BL, Battaglia FP, Jensen O, Moser EI, Moser M-B (2006) Path integration and the neural basis of the ‘cognitive map.’ Nat Rev Neurosci 7:663–678. Mehta MR, Lee AK, Wilson MA (2002) Role of experience and oscillations in transforming a rate code into a temporal code. Nature 417:741–746. Mizumori SJ, McNaughton BL, Barnes CA, Fox KB (1989) Preserved spatial coding in hippocampal CA1 pyramidal cells during reversible suppression of CA3c output: evidence for pattern completion in hippocampus. J Neurosci 9:3915–3928. Moita MAP, Rosis S, Zhou Y, LeDoux JE, Blair HT (2003) Hippocampal place cells acquire locationspecific responses to the conditioned stimulus during auditory fear conditioning. Neuron 37:485–497. Moores E, Laiti L, Chelazzi L (2003) Associative knowledge controls deployment of visual selective attention. Nat Neurosci 6:182–189. Morris RGM, Frey U (1997) Hippocampal synaptic plasticity: role in spatial learning or the automatic recording of attended experience? Philos Trans R Soc B 352:1489–1503. Morris RGM, Moser EI, Riedel G, Martin SJ, Sandin J, Day M, Carroll CO (2003) Elements of a neurobiological theory of the hippocampus: the role of activitydependent synaptic plasticity in memory. Philosophical Transactions of the Royal Society B: Biological Sciences 358:773–786. Moser EI (2004) Hippocampal Place Cells Demand Attention. Neuron 42:183–185. Mountcastle VB (1997) The columnar organization of the neocortex. Brain 120:701–722. Muller RU, Kubie JL (1987) The effects of changes in the environment on the spatial firing of hippocampal complex-spike cells. J Neurosci 7:1951–1968. Muller RU, Stead M, Pach J (1996) The hippocampus as a cognitive graph. J Gen Physiol 107:663–694. Murchison CF, Zhang X-Y, Zhang W-P, Ouyang M, Lee A, Thomas SA (2004) A distinct role for norepinephrine in memory retrieval. Cell 117:131–143. Muzzio IA, Levita L, Kentros C, de Boisblanc M, Kandel ER (2005) Attention-dependent learning of modality-
specific cues differentially modulates the firing pattern of CA1 neurons. In: Society for Neuroscience. Washington DC. Program No. 68.13. Nakazawa K, Sun LD, Quirk MC, Rondi-Reig L, Wilson MA, Tonegawa S (2003) Hippocampal CA3 NMDA receptors are crucial for memory acquisition of onetime experience. Neuron 38:305–315. Nakazawa K, Quirk MC, Chitwood RA, Watanabe M, Yeckel MF, Sun LD, Kato A, Carr CA, Johnston D, Wilson MA, Tonegawa S (2002) Requirement for hippocampal CA3 NMDA receptors in associative memory recall. Science 297:211–218. O’Keefe J (1976) Place units in the hippocampus of the freely moving rat. Exp Neurol 51:78–109. O’Keefe J, Dostrovsky J (1971) The hippocampus as a spatial map. Preliminary evidence from unit activity in the freely moving rat. Brain Res 34:171–175. Olypher AV, Lansky P, Fenton AA (2002) Properties of the extra-positional signal in hippocampal place cell discharge derived from the overdispersion in locationspecific firing. Neuroscience 111:553–566. Otmakhova NA, Lisman JE (1998) D1/D5 Dopamine Receptors Inhibit Depotentiation at CA1 Synapses via cAMP-Dependent Mechanism. J Neurosci 18:1270– 1279. Otmakhova NA, Lisman JE (1998) D1/D5 dopamine receptors inhibit depotentiation at CA1 synapses via cAMP-dependent mechanism. J Neurosci 18:1270– 1279. Packard MG, Cahill L, McGaugh JL (1994) Amygdala modulation of hippocampal-dependent and caudate nucleus–dependent memory processes. Proc Natl Acad Sci USA 91:8477–8481. Passetti F, Dalley JW, O’Connell MT, Everitt BJ, Robbins TW (2000) Increased acetylcholine release in the rat medial prefrontal cortex during performance of a visual attentional task. Eur J Neurosci 12:3051– 3058. Paz R, Pelletier JG, Bauer EP, Pare D (2006) Emotional enhancement of memory via amygdala-driven facilitation of rhinal interactions. Nat Neurosci 9:1321–1329. Pelletier JG, Pare D (2004) Role of amygdala oscillations in the consolidation of emotional memories. Biol Psychiatry 55:559–562. Pham K, Nacher J, Hof PR, McEwen BS (2003) Repeated restraint stress suppresses neurogenesis and induces biphasic PSA-NCAM expression in the adult rat dentate gyrus. Eur J Neurosci 17:879–886. Phelps EA, LeDoux JE (2005) Contributions of the amygdala to emotion processing: from animal models to human behavior. Neuron 48:175–187. Pittenger C, Huang YY, Paletzki RF, Bourtchouladze R, Scanlin H, Vronskaya S, Kandel ER (2002) Reversible inhibition of CREB/ATF transcription factors in region
PLASTICITY OF HIPPOCAMPAL REPRESENTATIONS
CA1 of the dorsal hippocampus disrupts hippocampusdependent spatial memory. Neuron 34:447–462. Posner MI (1980) Orienting of attention. Q J Exp Psychol 32:3–25. Posner MI, Rothbart MK (2007) Research on attention networks as a model for the integration of psychological science. Annu Rev Psychol 58:1–23. Power AE, Vazdarjanova A, McGaugh JL (2003) Muscarinic cholinergic influences in memory consolidation. Neurobiol Learn Mem 80:178–193. Quirk GJ, Muller RU, Kubie JL (1990) The firing of hippocampal place cells in the dark depends on the rat’s recent experience. J Neurosci 10:2008–2017. Redish AD, Rosenzweig ES, Bohanick JD, McNaughton BL, Barnes CA (2000) Dynamics of hippocampal ensemble activity realignment: time versus space. J Neurosci 20:9298–9309. Redish AD, Touretzky DS (1997) Cognitive maps beyond the hippocampus. Hippocampus 7:15–35. Reynolds JH, Pasternak T, Desimone R (2000) Attention increases sensitivity of V4 neurons. Neuron 26:703– 714. Robbe D, Montgomery SM, Thome A, Rueda-Orozco PE, McNaughton BL, Buzsaki G (2006) Cannabinoids reveal importance of spike timing coordination in hippocampal function. Nat Neurosci 9:1526– 1533. Rosenzweig ES, Redish AD, McNaughton BL, Barnes CA (2003) Hippocampal map realignment and spatial learning. Nat Neurosci 6:609–615. Sargolini F, Fyhn M, Hafting T, McNaughton BL, Witter MP, Moser M-B, Moser EI (2006) Conjunctive representation of position, direction, and velocity in entorhinal cortex. Science 312:758–762. Sarter M, Bruno JP, Givens B (2003) Attentional functions of cortical cholinergic inputs: what does it mean for learning and memory? Neurobiol Learn Mem 80:245–256. Sarter M, Hasselmo ME, Bruno JP, Givens B (2005) Unraveling the attentional functions of cortical cholinergic inputs: Interactions between signal-driven and cognitive modulation of signal detection. Brain Res Rev 48:98–111. Schultz W, Dayan P, Montague PR (1997) A neural substrate of prediction and reward. Science 275:1593– 1599. Schultz W, Dickinson A (2000) Neuronal coding of prediction errors. Annu Rev Neurosci 23:473–500. Shinoe T, Matsui M, Taketo MM, Manabe T (2005) Modulation of synaptic plasticity by physiological activation of M1 muscarinic acetylcholine receptors in the mouse hippocampus. J Neurosci 25:11194–11200. Steinmetz PN, Roy A, Fitzgerald PJ, Hsiao SS, Johnson KO, Niebur E (2000) Attention modulates synchro-
125
nized neuronal firing in primate somatosensory cortex. Nature 404:187–190. Summerfield C, Greene M, Wager T, Egner T, Hirsch J, Mangels J (2006) Neocortical connectivity during episodic memory formation. PLoS Biol 4:e128. Tanila H (2001) Noradrenergic regulation of hippocampal place cells. Hippocampus 11:793–808. Touretzky DS, Redish AD (1996) Theory of rodent navigation based on interacting representations of space. Hippocampus 6:247–270. Traub RD, Bibbig A, LeBeau FEN, Buhl EH, Whittington MA (2004) Cellular mechanisims of neuronal population oscillations in the hippocampus in vitro Annu Rev Neurosci 27:247–278. Uncapher MR, Otten LJ, Rugg MD (2006) Episodic encoding is more than the sum of its parts: an fMRI investigation of multifeatural contextual encoding. Neuron 52:547–556. Vinogradova OS (2001) Hippocampus as comparator: role of the two input and two output systems of the hippocampus in selection and registration of information. Hippocampus 11:578–598. Walling SG, Harley CW (2004) Locus coeruleus activation initiates delayed synaptic potentiation of perforant path input to the dentate gyrus in awake rats: a novel betaadrenergic- and protein synthesis–dependent mammalian plasticity mechanism. J Neurosci 24:598–604. Weible AP, McEchron MD, Disterhoft JF (2000) Cortical involvement in acquisition and extinction of trace eyeblink conditioning. Behav Neurosci 114:1058– 1067. Weinberger NM (2004) Specific long-term memory traces in primary auditory cortex. Nat Rev Neurosci 5: 279–290. Whittington MA, Traub RD (2003) Interneuron diversity series: inhibitory interneurons and network oscillations in vitro. Trends Neurosci 26:676–682. Wilson M, McNaughton BL (1993) Dynamics of the hippocampal ensemble code for space. Science 261: 1055–1058. Witte EA, Davidson MC, Marrocco RT (1997) Effects of altering brain cholinergic activity on covert orienting of attention: comparison of monkey and human performance. Psychopharmacology 132:324–334. Witter MP, Moser EI (2006) Spatial representation and the architecture of the entorhinal cortex. Trends Neurosci 29:671–678. Wittmann BC, Schott BH, Guderian S, Frey JU, Heinze H-J, Duzel E (2005) Reward-related fMRI activation of dopaminergic midbrain is associated with enhanced hippocampus-dependent long-term memory formation. Neuron 45:459–467. Wood ER, Dudchenko PA, Robitsek RJ, Eichenbaum H (2000) Hippocampal neurons encode information
126
PLACE CELLS AND SPATIAL CONTEXT
about different types of memory episodes occurring in the same location. Neuron 27:623–633. Xu L, Anwyl R, Rowan MJ (1998) Spatial exploration induces a persistent reversal of long-term potentiation in rat hippocampus. Nature 394:891–894. Yan J, Zhang Y, Jia Z, Taverna FA, McDonald RJ, Muller RU, Roder JC (2002) Place-cell impairment in glutamate receptor 2 mutant mice. J Neurosci 22(3): 204RC.
Yoshimura Y, Dantzker JLM, Callaway EM (2005) Excitatory cortical neurons form fine-scale functional networks. Nature 433:868–873. Zinyuk L, Kubik S, Kaminsky Y, Fenton AA, Bures J (2000) Understanding hippocampal activity by using purposeful behavior: place navigation induces place cell discharge in both task-relevant and task-irrelevant spatial reference frames. Proc Natl Acad Sci USA 97:3771–3776.
8 What Do Place Cells Tell Us about Learning to Associate and Learning to Segregate? EDUARD KELEMEN AND ANDRE´ A. FENTON
Substantial effort has gone into our current understanding of a mechanism for memory storage and its maintenance, with the molecular details of these processes having been largely worked out (Bliss and Collingridge, 1993; Pastalkova et al., 2006). These mechanisms lead to storing multiple representations (memories) during a lifetime. Such memories are latently present in the synapses of the brain and constitute a pool of experiences and information acquired by an animal. Most of the memories are irrelevant most of the time and remain inactive. At any given moment only a small subset from this pool of stored memories is activated and used. Normally, the active subset of memories is relevant for a particular context or ethological situation. The questions of how the appropriate memory is selected for activation and how in turn activation switches from one memory to the next as the context changes are organizing themes of our laboratory’s research efforts. In many of life’s situations more than one memory is relevant; therefore, the nervous system is challenged to activate several memories concurrently without confusing them, and the task of segregating simultaneously relevant but distinct memories arises. In this chapter we will concentrate on exactly this problem of segregating different, concurrently relevant representations. We will study this question in the context of hippocampal spatial representations, first reviewing behavioral evidence that rats use two concurrent spatial reference frames. We then review evidence that the hippocampus is important for the ability to segregate locations in the different spatial frames. Finally, we present preliminary
evidence for a functional grouping of place cell responses, as one of the mechanisms to achieve this segregation of distinct memories. There are two main reasons for hippocampal place cells being used to study the segregation of memories. The first is the well-established role of the hippocampus in spatial cognition and the thoroughly studied relationship between hippocampal place-cell firing and the rat’s apparent spatial knowledge. The other reason is that the hippocampus is indeed involved in the process of segregation of memories, as demonstrated by evidence from behavioral and pharmacological studies.
PLACE CELLS AS PART OF A SPATIAL KNOWLEDGE SYSTEM In formulating the cognitive-map theory of hippocampal function, O’Keefe and Nadel (1978) stressed that the cells were components of a knowledge system capable of providing a basis for spatial perception and spatial judgments. They were careful to point out that place cells were not just sensory elements that reflect stimuli in the environment. Instead, place cell firing reflects the rat’s spatial knowledge. A good deal of work supports this core idea that place cells do not merely reflect what can be currently sensed. Place cells respond to the absence of expected stimuli (O’Keefe, 1976) and the presence of unexpected task-relevant stimuli (Fyhn et al., 2002). Place cell responding is maintained after removing demonstrably salient
127
128
PLACE CELLS AND SPATIAL CONTEXT
stimuli (Quirk et al., 1990), and yet firing can change radically when the rat’s foraging behavior changes in a physically unaltered space (Markus et al., 1995). There are many other examples of the complexity of place cell behavior; the following discussion, however, dramatically illustrates the point that place cell firing is controlled by knowledge about the environment rather than the environment itself. Rotenberg and Muller (1997) demonstrated that place cell responses in rats in a cylinder were controlled by the location of a card on the wall. When they moved the card in 458 increments in front of the rat, place-cell firing fields shifted an equal amount. In the key experiment they shifted the card 1808 in front of the rat. Firing fields failed to shift, as if the card were being ignored. However, firing fields followed subsequent 458 card shifts. After four of these shifts, the card returned to its original location, but the cells were firing opposite their original firing field locations. Although the environment was identical at the beginning and at the end of the experiment, the place cells had quite different responses—their firing fields at the end of the experiment were rotated 1808 from the locations at the start of it. This evidence that place cell firing reflects the animal’s spatial knowledge, not just the current environment, has several important implications. We will focus here on just one, the process of segregation. Although the process of segregating is arguably as important to cognition as the process of associating, much of the research on place cells has focused on understanding how they make the associations expressed in their discharge. Accordingly, experimental paradigms have been designed to attenuate the salience of the irrelevant spatial information to minimize the need for segregation of relevant and irrelevant information. In the Morris (1981) water maze, this was achieved by adding water to theenvironment, making the behavioral substrate homogeneous and unstable so that the animal must use distal visual stimuli to find the hidden platform. In the pellet-chasing task introduced by Muller et al. (1987), the aim was to control place cell firing with a single stimulus—a salient card on the enclosure wall. The potential influence on place cell firing from other environmental features was effectively diminished through the use of nonpolarized, azimuthal geometry of a cylinder for the recording chamber and by hiding other stimuli as much as possible by enclosing the cylinder with a curtain, replacing the floor between sessions, and delivering scattered food automatically from overhead. The effort to exclude the cognitive process of segregation from consideration yielded a substantial body of knowledge about navigation and explicit memory. Building on this success, in recent years we have sought an
understanding of how these associations are segregated by developing a task that requires segregation of different sets of associations.
HIPPOCAMPUS IS INVOLVED IN SEGREGATION We tested whether the hippocampus is involved in segregating spatial information by training rats on a series of active place-avoidance tasks (Fig. 8–1). The rat was placed on a circular disk in an experimental room and was required to avoid a 608 sector of the environment, called a ‘‘shock zone.’’ Different variants of the task were used, all of which required the rat to associate shock with the same shock zone that could be defined by distal visual cues. However, the tasks varied systematically in the demand to segregate relevant and irrelevant information. In the task variant with low segregation demand, the shock zone could be identified by its position relative to the distal cues in the room, as well as by scent marks on the arena surface and by knowledge about rat’s self-motion. We called this variant the (Room&Arena)þ task variant because an arbitrary combination of room and arena frame cues could form the conditioned stimulus (CS) for conditioning the avoidance (Fig. 8–1C). The arena rotated at 1 rpm in the task variant with the highest segregation demand; the rotation made locations on the disk and selfmotion irrelevant and even misleading for identifying the room-frame shock zone. We called this version the RoomþArena—task variant because avoidance of room frame locations was reinforced but that of arena frame locations was not (Fig. 8–1E). This encouraged the rat to ignore arena frame locations when remembering where it was shocked. The experimental conditions were identical in the task variant with intermediate segregation demand, except that we covered the arena with shallow water to hide the irrelevant olfactory cues and attenuate the salience of arena frame locations. We called this the Roomþ variant (Fig. 8–1D); it was presumably easier to ignore arena frame information in this task. We used functional lesions to test the role of the hippocampus in segregating room and arena frame information in these tasks. Tetrodotoxin (TTX) was injected into one hippocampus to induce partial hippocampal dysfunction (Klement et al., 2005; Olypher et al., 2006). If the hippocampus is important for segregating room and arena frame locations, then the TTX injection should have the greatest effect on the RoomþArena—task and the least effect on the (Room&Arena)þ task. This is exactly what was observed. The TTX injection did not impair the
PLACE CELLS IN LEARNING TO SEGREGATE AND ASSOCIATE
129
Figure 8–1. Photograph (A) and schematic (B) of the place-avoidance behavioral setup. C–F. Cartoons depicting environmental stimuli that are available in four task variants. Three cues off the disk (large circle) and the large square represent room frame stimuli. A and B represent arena frame stimuli that are fixed on the disk surface. The edges of the shock zone are outlined around the shock symbol. The curved arrow just off the disk indicates disk rotation in the room. The arena frame is unstable relative to the room shock zone in the RoomþArena—and RoomþArenaþ task variants. This is indicated by dispersed arena cues around the room frame depiction of the shock zone. The room frame is unstable relative to the arena shock zone in the RoomþArenaþ task variant. This is indicated in the arena frame depiction by the dispersed room cues around the rotating arena shock zone.
(Room&Arena)þ task with low segregation demand but it devastated performance of the RoomþArena— task with high segregation demand (Wesierska et al., 2005). In fact, the TTX injection prevented learning, recall, and even the consolidation of the RoomþArena—task (Cimadevilla et al., 2001). In the Roomþ task variant with intermediate segregation demand, the TTX injection did not impair a familiar Roomþ avoidance, but it did impair learning of a new Roomþ avoidance (Kubik and Fenton, 2005). The differential effect of the TTX injection on learning and recall of Roomþ avoidance can be predicted by reasoning that begins from common knowledge. Good communicators and the best teachers have a knack for focusing a discussion on the essential subset of elements that need to be associated to understand a concept. They often deliberately ignore or leave out nonessential details. Once the idea has been properly acquired, a student’s awareness of the details and even of contradictions is much less likely to be misleading or confusing. By analogy, the data suggest that the proper segregation of stimuli into relevant and irrelevant cue subsets is especially important during the initial learning of spatial associations and that once these associ-
ations are learned, a failure to segregate relevant from irrelevant information is less likely to be disturbing. This series of experiments demonstrates that hippocampal function becomes more important in conditions of increasing segregation demand, even though the spatial associative demand remained essentially constant.
COORDINATED USE OF TWO SPATIAL REPRESENTATIONS We naturally wanted to study how place cell activity segregates the representations of room locations and arena locations. We chose a variant of the place avoidance task in which both the independently manipulated room and arena spatial frames are relevant. The rat must avoid two shock zones in the Roomþ Arenaþ task variant (Fig. 8–1F; Fenton et al., 1998). One shock zone is defined by distal room cues and is stationary. The other shock zone is defined by arena frame cues and rotates with the disk. Because avoidance of separate room and arena locations is reinforced in the RoomþArenaþ variant, neither spatial frame
130
PLACE CELLS AND SPATIAL CONTEXT
can be entirely ignored if the rat is to avoid shock optimally. Before describing the place cell activity, we will first show that during RoomþArenaþ avoidance, rats indeed used information from both the room and arena frames. Rats learn to avoid both shock zones in about four sessions (Fenton et al., 1998). Once well trained, they enter each shock zone about three times in 15 min. Figure 8–2 illustrates the typical trajectory of a welltrained rat during the task. The rat must organize its behavior in two spatial reference frames, so the rat’s trajectory in the two reference frames is plotted separately. When the rat is shocked for being in the room shock zone, these shocks occur in a cluster within the room frame but at dispersed locations in the arena frame. Similarly, when the rat is shocked for being in the arena shock zone, these shocks are clustered in the arena frame but dispersed in the room frame. When the rat is shocked during the RoomþArenaþ task the animal should segregate its notion of where it is into two distinct places: its room frame location and its arena frame location. The location associated with shock depends on where the rat remembers it was
shocked previously. It would be best for the rat not to associate shock with the other location. We quantified place avoidance in various ways. In counting the number of times the rat entered a shock zone in each session, we found that avoidance decreased with training (Fig. 8–2D). Measurement of the time between putting the rat on the arena and the time it was first shocked also gave us a measure of avoidance. Naı¨ve rats entered a shock zone within about 15 s. Once well trained, a rat would delay entering the shock zone for almost 6 min, an interval during which the arena would have rotated six times (Fig. 8–2E). The paths rats take during the task and these measures indicate that rats can avoid the two shock zones, but what does it tell us about the ways in which the rat’s room and arena spatial representations are organized? Several lines of evidence suggest that avoidance of the room shock zone expresses a separate memory from avoidance of the arena shock zone. An initial clue came from training rats to do (Room&Arena)þ on the stationary arena. Once they were well trained, the rotation was switched on and the shock was turned off. The rats avoided both the appropriate part of the
Figure 8–2. Behavior illustrating RoomþArenaþ place avoidance. A–C. The trajectories of the rat (gray) in the separate room and arena spatial frames are shown in the left and right circles, respectively. The room shock zone is outlined in red and the arena shock zone in blue in the corresponding frames. Red dots indicate locations where the rat was shocked after entering the room shock zone, and blue dots indicate locations where the rat was shocked after entering the arena shock zone. A shows typical behavior early in training, and B shows behavior after asymptotic performance was reached. C. Sometimes the rats avoided shock for the whole session. Learning of the task is depicted in D and E. Each session was performed on a separate day. Avoidance is scored by number of entrances to both shock zones (D) and the time to first enter either shock zone (E). Typically, avoidance is equally good in both spatial frames. The rats entered the shock zone unusually soon on day 6 for unknown reasons. Avoidance on this day was otherwise very good.
PLACE CELLS IN LEARNING TO SEGREGATE AND ASSOCIATE
room and the appropriate part of the arena. This result suggests that they formed a room-frame avoidance memory and a separate arena-frame avoidance memory. We have also demonstrated that room-frame and arena-frame avoidance memories can be acquired and extinguished independently (Bures et al., 1998; Fenton and Bures, 2003). Two ways of using segregated room and arena representations of space are compatible with the data reviewed thus far. One is that the representations are used in parallel, with each representation being independent of the other and each leading to separate avoidance behaviors. The room representation leads to avoidance of the room shock zone and the arena representation leads to avoidance of the arena shock zone. The two behaviors can be combined and as long as they are not in conflict, both can be expressed without interference. This possibility reflects the notion that there are multiple parallel memory systems (White and McDonald, 2002). Alternatively, separate room and arena representations can be used in a synergistic way by integrating them so that the behavioral output is not merely the sum of the two elemental behaviors. To determine whether the room and arena avoidance behaviors are used independently in parallel or synergistically, we analyzed the behavior of the rats during two probe conditions. The rats were trained to do RoomþArenaþ avoidance on the rotating arena, but the arena did not rotate during these probes. First we put the rats on the stable arena arranged in such a way that the two shock zones overlapped. Then we tested the rats on the stable arena when the two shock zones were opposite each other. Figure 8–3A1 illustrates the trajectories of the rats during these probe trials. The rats visited all parts of the disk except for the shock zone when the two shock zones were overlapping (in the north). This result demonstrated that when the arena was no longer rotating, the rats continued to express the conditioned place avoidance. The rats preferred to stay on the east side of the room when the two shock zones were opposite each other (the room shock zone in the north and the arena shock zone in the south). In fact, the rats never entered the west side (Fig. 8–3A2) on this probe. All the rats visited the west side of the room when the two shock zones were overlapping in the north, but not one of the six rats visited the same place when the arena shock zone was 1808 away in the south. These data indicate that the room- and arena-frame avoidance behaviors were not simply combined as the parallel model predicts. We conclude that although the rats form separate room and arena place-avoidance memories, they do not use these as two independent sources of information. It is more likely that the rats integrated the separate room and arena representations to guide synergistic behavior.
131
Why did the rats also avoid the west side of the disk when the two shock zones were opposite each other in the north and south? During RoomþArenaþ training, the disk was always rotated clockwise. Whenever the two shock zones were 1808 apart during training, the arena shock zone was moving toward the west. Thus, in this case, the east side of the room was better for avoiding shock. If this explanation is valid, then rotating the arena in the opposite direction should change how the rats react to the probe with the two shock zones opposite each other. We tested this prediction by rotating the arena in a counterclockwise direction during RoomþArenaþ place avoidance training. The arena rotation was stopped with the two shock zones opposite each other at the end of each 16-min training session. Within a few sessions, all of the rats walked to the west side of the arena, and half of the rats stopped visiting the east side altogether (Fig. 8–3B). This pattern confirmed our prediction that the direction of arena rotation during training influenced place avoidance behavior on the stationary probe sessions. This was also confirmed in a separate group of four rats that were exposed to both directions of rotation from the very beginning of RoomþArenaþ training. These rats visited both the east and west sides of the disk during the probe session with the two shock zones opposite each other (Fig. 8–3C). Our interpretation of the place avoidance data on the rotating arena is that the rats formed independent room-frame and arena-frame avoidance memories and used them in an integrated manner to display synergistic behavior. There is another possibility, however: the rats may have learned to identify room and arena positions in a single, integrated three-dimensional space. In this representation, the x-y dimensions would correspond to positions in one frame (e.g., the room) and the z dimension would correspond to a relation between the room and arena frames, such as their displacement. Although this seems unlikely, and it is difficult to conceive of the task in such an abstract manner, it is nonetheless a possible alternative. We turned to place cell recordings during RoomþArenaþ avoidance to better understand the organization of spatial information on the rotating arena.
PLACE CELL DISCHARGE DURING ROOMþARENAþ AVOIDANCE We have demonstrated that on the rotating arena, the hippocampus is important for segregating spatial information into the room and arena frames (Kubik and Fenton, 2005). We then reviewed evidence that spatial information from the two frames is represented as
132
PLACE CELLS AND SPATIAL CONTEXT
Figure 8–3. Stable arena probe sessions revealed that room- and arena-frame avoidance behavior is synergistic. The room shock zone is outlined in red and the arena shock zone in blue in the corresponding frames. Red and blue dots correspond to the locations of shocks when the rat entered the room-frame shock zone and the arenaframe shock zone, respectively. A1. Overlapping shock zones: the rats walked all over the arena, except for the shock zone in the north. A2. Opposite shock zones: the rats avoided both shock zones but also avoided the west side of the disk. Prior to these probe sessions the arena rotated clockwise during RoomþArenaþ training. Training resumed after these probes. This time the arena rotated counterclockwise. After each daily 16-min session, the rotation was stopped with the shock zones opposite each other. Behavior after the first (B1), third (B2), and fifth (B3) retraining sessions is shown. The rats learned to visit the west side of the disk. Four other rats were trained with the arena rotation alternating between clockwise and counterclockwise from the outset. These rats visited both the east and west sides of the disk on the probe sessions with the shock zones overlapping (C1) and opposite (C2) each other.
separate memories, and showed that the room- and arena-frame avoidance memories are integrated to direct synergistic avoidance behavior. The rat’s behavior during RoomþArenaþ avoidance demonstrates unequivocally that it uses spatial information from both the room and the arena frames. This spatial information should be reflected in place cell firing according to the cognitive-map theory of hippocampal function (O’Keefe and Nadel, 1978). We recorded ensembles of dorsal CA1 place cells from one or two tetrodes while rats performed Roomþ
Arenaþ avoidance on the rotating arena. Our first question was whether place cell activity would signal locations in the distinct room and arena frames. A sample of 152 cells was recorded from eight rats while the rats performed place avoidance in the stable–rotating– stable sequence of the (Room&Arena)þ then Roomþ Arenaþ then (Room&Arena)þ place-avoidance task variants. The two shock zones were overlapping when the arena was stable. We estimated the firing field quality using coherence (COH; Muller and Kubie, 1989) and information content (IC; Skaggs et al.,
PLACE CELLS IN LEARNING TO SEGREGATE AND ASSOCIATE
133
Figure 8–4. Frame-specific, angular-displacement spike maps during rotation illustrate the three main classes of place cell firing that can be observed during RoomþArenaþ avoidance on the rotating disk. The room shock zone is outlined in red and the arena shock zone in blue in the corresponding frames. Red and blue dots mark where the rat was in shocked in the room and arena shock zones, respectively. A room-coding cell (A), an arena-coding cell (B), and a conjunctive place cell (C) are shown. Spiking is summed across the whole session (column 1) or across episodes when the arena was displaced within 608. The arc indicates the range of the angular displacement that the shock zone moved during each class of episode. The upper row illustrates data from the viewpoint of the room frame (the room shock zone is stationary). The lower row illustrates data from the arena frame viewpoint (the arena shock zone is stationary).
1993). There were equal proportions of cells with firing fields (COH > 0.4 and IC > 0.3 bits/spike) when the arena was stable (80/152) and rotating (77/152). This finding suggests that place cell activity was normal during rotation; we will now focus on the discharge properties during rotation. We found examples of cells that had well-defined firing fields in the room frame. In the arena frame the firing of these cells was dispersed. We call such cells ‘‘room-coding’’ cells; an example is shown in Figure 8–4A. We stress that the firing of room-coding cells could also signal information about arena frame locations. Since we restricted our analyses to standard time-averaged measurements of firing field quality, it remains possible that a minority of the time a room-
coding cell’s discharge signaled arena locations. We also found examples of what we call ‘‘arena-coding’’ cells. These are cells with well-defined firing fields in the arena frame and dispersed firing in the room frame (Fig. 8–4B). Most (94%) cells with firing fields during rotation were room coding or arena coding. Some cells had firing fields in both frames. Often these cells had firing fields near the center of the disk. In any case, the number of cells with firing fields in the room frame was similar to the number of cells with fields in the arena frame. We also saw five cells that only fired in well-defined firing fields when the arena was at a certain displacement within the room (Fig. 8–4C). Neurons such as these are conjunctive cells (Gothard et al., 1996) and have firing fields in both frames.
134
PLACE CELLS AND SPATIAL CONTEXT
Consistent with the cognitive-map theory (O’Keefe and Nadel, 1978), we found room-coding and arenacoding place cells in conditions where rats were using spatial information about locations in the room and arena frames. This observation poses an interesting problem when we consider the dominant hypotheses on place cell coding of information about place. How can the same set of cells signal two conflicting types of information at the same time? This task is not especially difficult if individual place cells signal where the rat is in the labeled-line style of coding that can be seen in sensory systems. However, the current view is that location is signaled in a distributed place-cell population code that can remap. This requires some mechanism to subgroup activity representing room frame locations from activity representing arena frame locations. We next consider how the observation that place cells can represent room and arena locations impacts ideas about distributed coding. In population codes, each cell’s response is individually tuned to an external stimulus or mental item. In a place cell we assume it is a location. Cells apparently respond as independent entities in the most commonly used population codes. It is only necessary to know each cell’s activity to decode the signal at a moment because the independence assumption ignores the possibility of activity being correlated between cells. The vector sum of the activity of all cells in the population can provide an accurate signal (Georgopoulos et al., 1986, 1988). Wilson and McNaughton (1993) assumed a population code for location when they decoded the rat’s location from place cell activity. A second population-coding viewpoint asserts that place cell firing might signal the rat’s location in a population code that carries information in the neuronal activity correlations within the population. This means that although we can gain information about what is being represented by the sum of information from each cell, we can get more information by studying the joint activity of the cells (Gawne and Richmond, 1993; see review by Latham and Nirenberg, 2005). Recent place-cell work has used evidence of correlations in spike times representing information to argue that place cell activity is organized into cell assemblies as postulated by Hebb (1949). Harris et al. (2003) assumed cell assembly coding to optimize decoding of the rat’s location from ensemble place cell activity. The discharge of a set of cells together signals a stimulus or mental item in such ‘‘cell assembly’’ population codes. The activity in the population of cells is segregated into different cell assemblies by temporal grouping; a cell assembly is defined by the subset of cells that discharge together, and the activity correlations of cells within and between cell assemblies carry information about the rat’s location.
Regardless of the details of the particular populationcoding hypotheses, the cognitive-map theory predicts that one should find room-coding place cells and arenacoding place cells during the RoomþArenaþ task, because these cells should signal the spatial information the rat seems to be using to avoid shock on the rotating arena. However, simply summing the activity of the place cell population at any moment will yield an average somewhere between the correct room location and the correct arena location. This location will not correspond to where the rat is in either frame. There must be some mechanism to segregate information about location in the room frame from information in the arena frame. This requirement for segregation is inherent in a cell assembly type of code. Von der Malsburg (1981) called failure of this segregation the ‘‘superposition catastrophe.’’ A cell assembly is defined by discharge in the set of coactive cells that signals a mental item. Once it is necessary to signal two items at once with two cell assemblies that have cells in common, there must be some means to segregate the activity of the two assemblies, or they will have merged into a single, third assembly, with the consequence being catastrophic information loss. It has been suggested that superposition catastrophe is avoided by temporal grouping, which entails the appropriate segregation of activity from the two assemblies in time (see review by Singer, 1999). Although there is evidence for temporal grouping, it is not yet widely accepted (see reviews by Averbeck and Lee, 1994; Shadlen and Movshon, 1999). It may be that the analytical tools required to detect temporal grouping are substantially more complex than calculating individual firing rates (and firing rate maps that are traditional in place cell work). Proposed mechanisms to achieve appropriate temporal grouping are based on oscillations. The oscillations themselves have been well characterized but the means by which they lead to appropriate temporal grouping for coding multifactorial information are largely not understood. An alternative mechanism to segregate place cell activity is functional grouping. Cells that signal a common category of information, in this case room frame or arena frame locations, will be separated from cells that signal the other category of information. Functional grouping is common in sensory and motor systems and has even been reported for hippocampal activity during nonmatch-to-sample lever-pressing (Hampson et al., 1999, 2002). Functional grouping has not been observed, however, in place cell activity (Redish et al., 2001). We analyzed the ensemble organization of place cell activity during the rotation. We considered only cells that were recorded simultaneously and compared the coherence of each cell in the room frame with its coherence in the arena
PLACE CELLS IN LEARNING TO SEGREGATE AND ASSOCIATE
135
Figure 8–5. Functional grouping of ensemble frame-preference, with six simultaneously recorded cells during rotation. These are a subset of the cells recorded as an ensemble. A1. Discharge is represented in frame-specific (left) spike and (right) color-coded firing rate maps. The ensemble was judged to be room frame preferring because firing fields tended to be better defined in the room frame than in the arena frame. A2. Firing-rate map coherence in the two frames is represented as a scatter plot. The points deviate above the diagonal, indicating that the ensemble preferred the room frame. B1. An arena frame–preferring ensemble. B2. The coherence values of these cells tend to fall below the diagonal, indicating a general preference for the arena frame. The color of the cell numbers corresponds to the color of the points in the scatter plot. Notice that the coherence of cells 5 and 6 in A and cell 5 in B fall on the diagonal. These cells had poor spatial firing in both spatial frames. frame by plotting the coherence in one frame against that in the other. There was a surprising pattern: most cells in an ensemble tended to have better coherence in one frame. Figure 8–5A shows a room-preferring ensemble and Figure 8–5B shows an arena-preferring ensemble. The ensemble-level frame preference was quantified by calculating the difference between the coherence of the room-frame and arena-frame rate maps. The distribution of the differences tended to shift to positive values for room-preferring ensembles and to negative values for arena-preferring ensembles. Of seven ensembles recorded with 10 or more place cells, three were room-preferring and three were arena-preferring. A frame preference in 6/7 ensembles
(z ¼ 1.9, p < 0.03) is unlikely if the presence or absence of a frame preference is equally likely.
SUMMARY Our aim in this chapter was to make the case that, in addition to forming associations, the hippocampus must also segregate information into appropriate categories of associations. At the very least, cognitive segregation results in reducing the likelihood of making inappropriate associations while providing for the flexible and synergistic use of stored information (Cohen and Eichenbaum, 1993).
136
PLACE CELLS AND SPATIAL CONTEXT
We have summarized work from our laboratory that provides substantial evidence that rat behavior reflects their ability to easily use at least two spatial frames. Rats use spatial frame–specific information in ways suggesting that they store the information as separate associations. Hippocampal dysfunction impairs the coordinated use of this information. Place cell recordings have confirmed that place cell discharge was also frame specific in ways that matched framespecific behavior. The firing of single place cells seemed to preferentially signal the rat’s position in one spatial frame over that in the other. Across a group of rats, room-coding and arena-coding place cells were equally represented, but both classes of cell were rarely observed in a single session. This is clear evidence of functional grouping of place cell activity that has not been described previously. In fact, Redish et al. (2001) failed to observe functional grouping of place cell firing in linear environments. Nearby place cells did not have neighboring firing fields. This finding contrasts with the functional grouping of hippocampal activity that was described during a delayed nonmatching-to-sample lever-pressing task (Hampson et al., 1999; 2002) in which place cell responses are difficult to characterize. Cells within 600-mm anatomical domains had similar left–right lever response preferences. Cells in different adjacent domains had different response preferences. Evidence consistent with such anatomical clustering of place cells was reported by Brown and Skaggs (2002), but this idea is inconsistent with the lack of preservation of remapped firing field distances in open fields (Muller and Kubie, 1987; Lever et al., 2002). In contrast to other place cell data, we observed evidence of functional grouping of place cell discharge according to spatial frame in small ensembles recorded from the dorsal CA1 region. How can we reconcile our observations with those from other place cell studies? We suggest that functional grouping only occurs across categories of information that require segregation. In this case, functional grouping of place cell responses would only be detected when it is possible to distinguish between qualitatively distinct categories of location. Although locations change during standard maze running, there are no obvious criteria for distinguishing qualitatively distinct categories of location. The functional grouping of spatial frame preference that we observed was only detected during arena rotation that dissociated locations into the categories of the room and arena frames. The functional grouping of place cells into framespecific categories matches rat spatial behavior during arena rotation, with one salient exception. Each rat demonstrated that it uses spatial information from both the room and the arena frames. In contrast, for single
sessions, we were only able to identify place cell firing that reflected location in one of the two spatial frames. There are at least two ways to account for this. We recorded from only one or two locations in dorsal CA1; it is very likely that these locations were within a single 600-mm region. This would correspond to a single functional anatomical domain as reported by Hampson et al. (1999). Alternatively, it could be that all place cells in a rat preferentially respond to location in one spatial frame. In this case, information about locations in the other spatial frame is represented elsewhere. Either version of this segregation would avoid representation of conflicting information in the hippocampus and thus reduce the likelihood of the superposition catastrophe. Recordings from distributed hippocampal sites could distinguish between these two possibilities and reveal how functional grouping of place cell responses is achieved. Regardless of the mechanism, the observation of functional grouping in place cell frame-specific responses demonstrates that in response to a demand to segregate information into spatial frames, rat behavior and place cell firing becomes segregated into the corresponding spatial frames. This is just what a knowledge system should do.
References Averbeck BB, Lee D (2004) Coding and transmission of information by neural ensembles. Trends Neurosci 27:225–230. Bliss TVP, Collingridge GL (1993) A synaptic model of memory: long-term potentiation in the hippocampus. Nature 361:31–39. Brown JE, Skaggs WE (2002) Concordant and discordant coding of spatial location in populations of hippocampal CA1 pyramidal cells. J Neurophysiol 88: 1605–1613. Bures J, Fenton AA, Kaminsky Y, Wesierska M, Zahalka A (1998) Rodent navigation after dissociation of the allocentric and idiothetic representations of space. Neuropharmacology 37:689–699. Cimadevilla JM, Wesierska M, Fenton AA, Bures J (2001) Inactivating one hippocampus impairs avoidance of a stable room-defined place during dissociation of arena cues from room cues by rotation. Proc Natl Acad Sci USA 98:3531–3536. Cohen NJ, Eichenbaum H (1993) Memory, Amnesia, and the Hippocampal System. Cambridge, MA: MIT Press. Fenton AA, Bures J (2003) Navigation in the moving world. In The Neurobiology of Spatial Behaviour (Jeffery JK, ed.). Oxford: Oxford University Press. Fenton AA, Wesierska M, Kaminsky Y, Bures J (1998) Both here and there: simultaneous expression of au-
PLACE CELLS IN LEARNING TO SEGREGATE AND ASSOCIATE
tonomous spatial memories. Proc Natl Acad Sci USA 95:11493–11498. Fyhn M, Molden S, Hollup S, Moser MB, Moser E (2002) Hippocampal neurons responding to first-time dislocation of a target object. Neuron 35:555–566. Gawne T, Richmond B (1993) How independent are the messages carried by adjacent inferior temporal cortical neurons? J Neurosci 13:2758–2771. Georgopoulos AP, Kettner RE, Schwartz AB (1988) Primate motor cortex and free arm movements to visual targets in three-dimensional space. II. Coding of the direction of movement by a neuronal population. J Neurosci 8:2928–2937. Georgopoulos AP, Schwartz AB, Kettner RE (1986) Neuronal population coding of movement direction. Science 233:1416–1419. Gothard KM, Skaggs WE, Moore KM, McNaughton BL (1996) Binding of hippocampal CA1 neural activity to multiple reference frames in a landmark-based navigation task. J Neurosci 16:823–835. Hampson RE, Simeral JD, Deadwyler SA (1999) Distribution of spatial and nonspatial information in dorsal hippocampus. Nature 402:610–614. Hampson RE, Simeral JD, Deadwyler SA (2002) ‘‘Keeping on track’’: firing of hippocampal neurons during delayed-nonmatch-to-sample performance. J Neurosci 22:RC198. Harris KD, Csicsvari J, Hirase H, Dragoi G, Buzsaki G (2003) Organization of cell assemblies in the hippocampus. Nature 424:552–556. Hebb DO (1949) The Organization of Behavior. New York: Wiley. Klement D, Pastalkova E, Fenton AA (2005) Tetrodotoxin infusions into the dorsal hippocampus block non-locomotor place recognition. Hippocampus 15:460–471. Kubik S, Fenton AA (2005) Behavioral evidence that segregation and representation are dissociable hippocampal functions. J Neurosci 25:9205–9212. Latham PE, Nirenberg S (2005) Synergy, redundancy, and independence in population codes, revisited. Neuroscience 25:5195–506. Lever C, Wills T, Cacucci F, Burgess N, O’Keefe J (2002) Long-term plasticity in hippocampal placecell representation of environmental geometry. Nature 416:90–94. Markus EJ, Qin Y, Leonard B, Skaggs WE, McNaughton BL, Barnes CA (1995) Interactions between location and task affect the spatial and directional firing of hippocampal neurons. J Neurosci 15:7079–7094. Morris RGM (1981) Spatial localisation does not depend on the presence of local cues. Learn Motiv 12:239–261. Muller RU, Kubie JL (1989) The firing of hippocampal place cells predicts the future position of freely moving rats. J Neurosci 9:4101–4110.
137
Muller RU, Kubie JL, Ranck JBJr (1987) Spatial firing patterns of hippocampal complex-spike cells in a fixed environment. J Neurosci 7:1935–1950. Muller RU, Kubie JL (1987) The effects of changes in the environment on the spatial firing of hippocampal complex-spike cells. J Neurosci 7:1951–1968. O’Keefe J (1976) Place units in the hippocampus of the freely moving rat. Exp Neurol 51:78–109. O’Keefe J, Nadel L (1978) The Hippocampus as a Cognitive Map. Oxford: Claredon Press. Olypher A, Klement VD, Fenton AA (2006) Cognitive disorganization in hippocampus: a physiological model of the disorganization in psychosis. J Neurosci 26:158–168. Pastalkova E, Serrano P, Pinkhasova D, Wallace E, Fenton AA, Sacktor TC (2006) Storage of spatial information by the maintenance mechanism of LTP. Science 313:1141–1144. Quirk GJ, Muller RU, Kubie JL (1990) The hippocampal place cells in the dark reflect the rat’s recent experience. J Neurosci 10:2008–2017. Redish AD, Battaglia FP, Chawla MK, Ekstrom AD, Gerrard JL, Lipa P, Rosenzweig ES, Worley PF, Guzowski JF, McNaughton BL, Barnes CA (2001) Independence of firing correlates of anatomically proximate hippocampal pyramidal cells. J Neurosci 21:RC134. Rotenberg A, Muller RU (1997) Variable place-cell coupling to a continuously viewed stimulus: evidence that the hippocampus acts as a perceptual system. Philos Trans R Soc Lond B Biol Sci 352:1505–1513. Shadlen MN, Movshon JA (1999) Synchrony unbound: a critical evaluation of the temporal binding hypothesis. Neuron 24:67–77. Singer W (1999) Time as coding space? Curr Opin Neurobiol 9:189–194. Skaggs WE, McNaughton BL, Gothard KM, Markus EJ (1993) An information theoretic approach to deciphering the hippocampal code. In Advances in Neural Information Processing, Vol. 5 (Hanson SJ, Cowan JD, Giles CL, ed.). San Mateo, CA: Morgan Kaufmann. von der Malsburg C (1981) The Correlation Theory of Brain Function. Technical Report 81–2. Biophysical Chemistry, Max Planck Institute. Wesierska M, Dockery C, Fenton AA (2005) Beyond memory, navigation and inhibition: behavioural evidence for hippocampus-dependent cognitive coordination in the rat. J Neurosci 25:2413–2419. White NM, McDonald RJ (2002) Multiple parallel memory systems in the brain of the rat. Neurobiol Learn Mem 77:125–184. Wilson MA, McNaughton BL (1993) Dynamics of the hippocampal ensemble code for space. Science 261: 1055–1058.
9 Do Place Cells Guide Spatial Behaviors? ETIENNE SAVE AND BRUNO POUCET
Spatial behaviors are essential to the survival of most animal species. Very frequently, animals leave their nest, explore the surroundings, travel in their home range to fulfill fundamental needs, find food and water, interact with mates, and then return to their nest. However, because of the inherent complexity of the environment and its likeliness to abruptly change, goal-directed behaviors, i.e., navigation, may sometimes be difficult to perform. Animals have developed a variety of navigational strategies that have been shown to be efficient to maintain and/or restore their navigation performance when the environment is modified. Following Tolman’s work, certain strategies have been hypothesized to depend on the animal’s ability to generate a spatial representation in their brain (Tolman, 1946). Such a representation is thought to encode the spatial relationships between landmarks, irrespective of the animal’s location or point of view (allocentric coding), thus forming a ‘‘cognitive map’’ of the environment (O’Keefe and Nadel, 1978). Important places such as goal or home locations are defined relative to each other and relative to configurations of landmarks. The cognitive map thus provides the animal with the ability to flexibly navigate in its environment, for example, by using different trajectories to reach a goal (shortcuts, detours), and to memorize locations and their spatial relationships a (Save, et al., 1998). The question of the neuroanatomical and neurobiological foundations of spatial behaviors has generated a huge amount of work in the last 40 years. In rodents, lesion and electrophysiological studies have identified
the hippocampus as a key structure in the processing of spatial information. There is overwhelming evidence that the hippocampus is crucial for accurate performance in spatial tasks. Lesion studies have shown that the hippocampus does not mediate all strategies. Hippocampus-dependent strategies are thought to involve an allocentric spatial map (cognitive map), as in the Morris water-maze place navigation task, and are altered following hippocampal damage (Morris et al., 1982). In contrast, visually guided or response-based strategies do not depend on the hippocampus (Morris et al., 1982; Packard and McGaugh, 1996; Save and Poucet, 2000). In the early 1970s, the seminal discovery of spatially selective place cells in the CA3 and CA1 hippocampal fields in freely moving rats led to the proposal that the hippocampus is the neural substrate of an allocentric spatial map (O’Keefe and Dostrovsky, 1971; O’Keefe and Nadel, 1978). To understand the role of place cells, at least two fundamental questions need to be addressed. First, what are the determinants of location-specific firing, and second, how do place cells contribute to spatial behaviors? The issue of sensory and cognitive (mnemonic) determinants of spatial firing has been tackled in numerous studies. In particular, it has been shown that place cell activity is controlled by a variety of information including external (allothetic: visual, olfactory, etc.) and internal (idiothetic: vestibular, proprioceptive, etc.) information (O’Keefe and Conway, 1978; Muller and Kubie, 1987; Sharp et al., 1995; Gothard et al., 1996a; Save et al., 2000). Other work has proposed that the function of place cells is to
138
DO PLACE CELLS GUIDE SPATIAL BEHAVIORS?
encode sequences of multidimensional events that form episodic memory (Eichenbaum et al., 1999). In contrast, the second question has received little attention, and it remains unclear whether place cells control the animal’s spatial behavior. Thus, our aim in the present chapter is to review relevant data on the relationships between place cell firing and animals’ spatial behavior. Overall, although relatively few experiments have addressed this issue, most support the idea that hippocampal place cell activity is tightly coupled to spatial behavior and may encode various aspects of goal-directed behavior.
PLACE CELL FIRING DURING RANDOM FORAGING VERSUS GOAL-DIRECTED BEHAVIOR Investigation of the relationship between place cell firing and an animal’s spatial behavior is not straightforward because of constraints inherent to the recording technique. In a number of classical studies, place cell activity could be correctly sampled as the rats searched for small pellets dropping from the ceiling in the recording cylinder (e.g., Muller and Kubie, 1987). However, this situation is not the most appropriate one for studying the coupling between place cell firing and spatial behavior because it does not impose much constraint on the animal’s behavior, which, for the most part, appears to be erratic. For that matter, it could instead be considered a control situation in which the animal has no explicit task to perform, although it must be clear that the rat keeps track of its position in the environment. This task, hereafter called random foraging task, has been used in numerous studies as a ‘‘notask’’ situation (although it is undoubtedly a task). To investigate the effect of behavioral tasks on place cell activity, a number of studies have explored the possibility that place cell firing is influenced by the task demand. Markus and colleagues (1995) recorded place cells as rats performed successively a random foraging task and a goal-directed task both in the same apparatus (a circular platform or a plus maze). In the goal-directed task, rats were trained to visit sequentially four baited spots located at the periphery of the circular platform. The authors found that a change in behavior resulted in modifications in place cell firing. A proportion of the fields changed their location in the apparatus and the new fields displayed increased selectivity to head direction. This finding suggests that the goal-directed task involved the recruitment of a different population of neurons with properties (directionality) in relation to the specific behavior. In the same vein, Kobayashi and colleagues (2003) trained rats first in a random foraging task and then in
139
a goal-directed task to receive intracranial self-stimulation rewards. In the goal-directed task, animals were rewarded to shuttle between two diametrically opposite goal zones in a circular arena. The results indicated that for a number of cells, the spatial firing pattern changed as the animal learned the goal-directed task and performed optimal trajectories. It is remarkable that changes resulted in new fields to appear tightly associated to the two goal zones, suggesting some goal-encoding process by place cells. This aspect will be developed in the goal-encoding section below. Differences in place fields between a random foraging and a goal-directed task were also found in mice (Kentros et al., 2004). The goal-directed task was fairly different from that used by Markus et al. (1995). In a circular arena, mice were trained to locate an unmarked area and stay in it for 1s to turn off unpleasant auditory and visual stimuli for a while. The area could be located only by using environmental cues. Independent groups of animals were trained in the random foraging and goal-directed tasks. The results show that place fields were unstable between sessions in the foraging task but were more stable in the goal-directed task, suggesting that the representation of space was more reliable in the goal-directed task than in the foraging task. Note however, that unstable place fields in constant conditions are seen in mice but usually not in rats (e.g., Save et al., 2005). In the latter, unstable fields generally result from environmental manipulations. One view of these results is that a stable hippocampal map is necessary for accurate place learning. Another view is that behavior may influence place cell activity, possibly via the animal’s attentional, motivational, or emotional states (Kentros et al., 2004). Relevant to this view, Zinyuk et al. (2000) have reported that place firing was more affected in animals initially trained in a random foraging task (‘‘foragers’’) than in those in a goal-directed task (‘‘navigators’’) when the arena rotated continuously. The goaldirected task was a place preference task in which the animals had to locate an unmarked area by using room cues, to release a food pellet. On the rotating arena, the animals were required to dissociate a room reference frame (relevant to the task) from an arena reference frame (irrelevant to the task). Rotation was found to affect more markedly the firing characteristics of place cells (such as rate of discharge, spatial coherence of firing, and spatial information content) in foragers than in navigators. More specifically, clear-cut place fields were maintained in predictable locations (mostly relative to the room reference frame) in navigators during arena rotation. In contrast, rotation disrupted the spatial selectivity of place fields in foragers. Thus, the choice of whether rats were trained in a random
140
PLACE CELLS AND SPATIAL CONTEXT
foraging task or a goal-directed task had an effect on how they dealt with the rotating arena to maintain coherent place cell firing. As proposed by the authors, this finding suggests that the use of environmental cues that allow place field anchoring depends on the animal’s behavioral state, consistent with the hypothesis that task-related cognitive processes (e.g., attention) modulate place cell firing (Kentros et al., 2004). Other firing properties such as the variability of discharge have also been shown to be different in foragers and navigators. Place cell firing displays high variability as the animal performs multiple passes through the field, a phenomenon called ‘‘overdispersion.’’ Because such variability cannot be explained only by the positional discharge of the cell, it was hypothesized that it reflects extrapositional signals. Olypher et al. (2002) examined whether overdispersion was different in foragers and navigators rats. As in Zinyuk et al.’s study (2000), foragers had to collect randomly scattered pellets and navigators were trained to use the room frame on the rotating platform. They found that overdispersion was decreased in navigators relative to that of foragers, thus suggesting that it is modulated by the task demand. The authors proposed that overdispersion corresponds to switches of attention between different spatial reference frames, thus explaining the lower overdispersion in rats that used to rely on one reference frame only. Overall, the data comparing place cell activity in goal-directed and random foraging tasks are remarkably convergent. They show that place cell firing is different in the two behavioral contexts. More specifically, place cell firing in goal-directed behavior exhibits properties consistent with the task demand (directionality, goal firing, stability, resistance to change, etc.). In contrast, spatial firing in random foraging lacks these properties. These results suggest that place cells can display distinct functional modes and can switch from one mode to another as a function of the behavioral requirements. The nature of the mechanisms underlying such change is still undetermined, but the data suggest that it involves recruitment of cognitive resources. Random foraging and goal-directed behavior are two behavioral contexts that differ on many aspects. Changes in firing may thus be explained by a variety of factors such as locomotor, motivational, or emotional factors. In a recent study, Smith and Mizumori (2006) controlled these factors to restrict the behavioral context to the rewarded location. Rats were trained to receive some food at one location on a plus maze during the first half of a recording session and at another location during the second half of the session. These two halves constituted two separate contexts. Comparing place cell activity in the two halves, the authors de-
scribed clear context-specific firing patterns, suggesting that moderate behavioral differences can influence place cell activity and yield distinct representations in the hippocampus.
IS PLACE CELL FIRING COUPLED WITH PERFORMANCE? A further step in investigating the relationships between place cell activity and the behavior is to examine whether place cell firing is correlated to the animal’s performance in a spatial task. The earliest study reporting such a relationship is that of O’Keefe and Speakman (1987). Place cell activity was recorded while rats performed a spatial memory task in a cross maze surrounded by controlled cues. On spatial working memory trials, rats were placed on the start arm for a ‘‘perceptual’’ period. The cues were then removed and, after a delay (‘‘memory’’ period), the rats were allowed to reach the goal arm. Place cell firing patterns were maintained in the absence of the controlling cues and were similar to the firing patterns in the perceptual period. In addition, the rats were given a few control trials in which they were not exposed to the perceptual period. Rather, they were placed on the start arm after removal of environmental cues. The animals performed at chance levels but the firing fields corresponded to the rat’s choice, even if this choice led to an incorrect location. Thus, O’Keefe and Speakman’s results show that place cells still fire in the absence of landmarks and that the rat’s behavior is coherent with the spatial map. One can speculate that, because there were no anchoring cues, the animals might have arbitrarily oriented their spatial representation relative to the maze and therefore produced a choice consistent with this map but not in register with the external layout. In line with O’Keefe and Speakman’s study, an appropriate way to investigate the relationships between spatial firing and performance is to disrupt the alignment of the place cell map and examine its consequence on the animal’s performance in a spatial task. If firing and behavior are coupled, then incoherent place cell activity should yield incorrect spatial performance. To test this idea, Lenck-Santini et al. (2001a) trained rats in a continuous alternation spatial task. Animals had to alternate between the two arms of a Y-maze to receive a food reward at the end of the third arm. The apparatus was located in a cue-controlled environment containing a conspicuous cue card. Some manipulations such as removing the cue card (in the presence of the rat) and rotating the goal to a different arm (in the absence of the rat) resulted in incoherent placement of the fields relative to the learned goal location on some sessions. Incoherent firing was accompa-
DO PLACE CELLS GUIDE SPATIAL BEHAVIORS?
nied by a deterioration of the rats’ performance. Interestingly, the most frequent errors were not alternation errors (memory of the previous choice) but orientation errors (failures to return to the goal arm, perseveration), suggesting that diminished performance was not due to a memory deficit but to a spatial deficit. Basically, these results are consistent with those of O’Keefe and Speakman in that they suggest a coupling between place cell activity and behavior. Furthermore, they indicate that performance in more complex spatial tasks also depends on coherent place cell firing. This impairment, however, may arise from a basic orientation deficit that would result from the conflict between the hippocampal map and the environment. In a follow-up study, Lenck-Santini et al. (2002) also used a conflict procedure to investigate involvement of the place cell system in the use of three different spatial strategies: place, guidance, and cue strategies. Rats were trained to perform a continuous place-preference task (see Rossier et al., 2000; Zinyuk et al., 2000) in which they had to locate a 20-cmdiameter goal zone in the recording cylinder in order for a small pellet to be dropped to the floor. The only available landmark was a salient cue card attached to the cylinder wall. The goal zone was located in the middle of a quadrant at a distance from the cue card in the place task or at the bottom of the cue card in the guidance task, or it was marked by a black disc in the cue task (Fig. 9–1). Manipulations of the cue card allowed one to produce coherent or incoherent placement of the place fields relative to the goal zone. Ninetydegree cue card rotation in the absence of the animal yielded coherent field placement (908 rotation) whereas rotation in the presence of the animal very often yielded incoherent field placement (the goal zone was rotated by the same amount of 908 in the two manipulations). The performance of the rats (number of entries in the goal zone) decreased when field placement was incoherent but was maintained when it was coherent. This effect was much greater in the place task than in the guidance task. In the cue task, the performance of the rat was maintained despite incoherent field placement. Thus, these results are consistent with the hypothesis of a coupling between place cell firing and the animal’s performance. Most importantly, they suggest that this coupling is dependent on the strategy used by the animal. Strategies based on the use of a spatial representation involve a coupling, whereas other strategies (guidance, cue learning) are more or less decoupled from place cell activity. Interestingly, each session involving cue manipulation was started with an extinction period, i.e., the animal was not rewarded. This enabled the authors to examine the animal’s expectation about the location of the goal zone, in particular when field placement was inco-
141
herent. Was the animal’s search consistent with the place field or with the cue card? The results clearly indicate that in the place task, search was consistent with the field location, whereas in the guidance and cue tasks it was consistent with the card location or the disc, respectively. Taken together with previous results (O’Keefe and Speakman, 1987, Lenck-Santini et al., 2001a), this indicates that behavior is coherent with the hippocampal representation, even if environmental information is conflicting. Instead of generating a conflict, Fenton and colleagues examined the effects of slightly manipulating the distance between two cues, a white card and a black card, attached to the wall arena, on place cell activity (Fenton et al., 2000a,b) and place-learning abilities (Kubie et al., 2004). In the place-cell recording study, there were two cue-manipulation conditions: the angular distance between the two cue cards was increased or decreased by 258. Fields tended to move in the direction corresponding to the direction of the cue displacement. In addition, they found that fields located in the sector of the apparatus delimited by a particular cue card, black or white, tended to move in association with this cue card. Such topological changes were accompanied by a decrease in the field rate and coherence. In a subsequent study, Kubie et al. trained rats in a spatial accuracy task and asked whether the behavioral response to shifting cues apart or together paralleled the place cell response reported by Fenton et al. (2000). The spatial accuracy task is akin to the place preference task described above. The rat was rewarded for stopping for more than 1 s in a small (3 cm diameter), unmarked goal location. Performance was measured during an extinction period. Cue card manipulations induced behavioral shifts in a number of rats, and these shifts were found to parallel very closely the shifts observed in place cells. These studies support the hypothesis that place cells control spatial behaviors. However, they involved mainly place learning, thus leaving the possibility that control is task specific and is achieved in certain circumstances (e.g., overtraining). Accordingly, it is necessary to look at different kinds of behaviors and examine their relationship with place cell firing. In recent work, we recorded place cells as the animals were exposed to a shortcut situation (Alvernhe et al., unpublished results). Rats were first trained to shuttle between the extremities of an M-shaped maze to receive a food reward (Fig. 9–2). Each inner wall of the maze was made of three parts that could be individually removed, thus allowing the animal to take a shortcut to reach the goal. Nine different shortcut problems could be proposed to the animal but only six were used. The aim was to simultaneously examine the dynamics of
142
PLACE CELLS AND SPATIAL CONTEXT
Figure 9–1. A. The continuous place preference task (Rossier et al., 2000) used in the Lenck-Santini et al.’s study (2002). Once trained, the rat must enter an unmarked goal zone and remain still for 2 s to receive a small pellet dropped in the cylinder. The rat then leaves the goal zone and forages in the cylinder to find the pellet. B. Manipulation of the cue card (represented as a thick, curved line at the periphery of the cylinder) in the place task allows one to produce coherent or incoherent placement of the place fields relative to the goal zone (represented as a dotted-line circle). Ninety-degree cue card rotation (as indicated by the arrow) in the presence of the animal (visible rotation) very often yields inconsistent field placement. In contrast, rotation in the absence of the animal (hidden rotation) yields consistent field placement (908 rotation). C. Left: Examples of rate maps of consistent and inconsistent session pairs. Yellow pixels indicate no firing and purple indicates maximum firing. Orange, red, green, and blue indicate intermediate firing rates from low to high. The cross represents the goal zone. In consistent session pairs, rotation of the cue card (Test) was followed by an equivalent rotation of the fields. In inconsistent session pairs, the fields were not controlled by the cue card. In the present example, the field remained stable relative to background cues. Right: Performance of the animals (number of responses/min) in consistent and inconsistent session pairs. Asterisks indicate a significant difference between the Reference (Ref) session and the Test session in inconsistent session pairs ( p < 0.0.01).
shortcut selection (relative to the familiar path) and hippocampal neural representation to see whether place cell activity would change when the animal took a shortcut and whether these changes paralleled behavioral changes. Three successive recording sessions were made, including a standard session (no shortcut), a test session (with the shortcut), and a second standard session (no shortcut). Throughout the study, the animals experienced the six possible shortcut problems in pseudo-random order. Behaviorally, there was abrupt learning of the shortcut: all rats chose the shortcut the first time they faced it. This behavioral change was accompanied by immediate place field
change. The fields located nearby or far from the shortcut region were differentially affected: a majority of far fields remained unaffected, whereas most near fields were modified, mainly in terms of shape and/or firing rate (remapping). These preliminary results, though underlying only a correlation between shortcut behavior and place cell firing, are consistent with the hypothesis that spatial behaviors involving the use of a spatial map are coupled with place cell activity. Place field reorganization (remapping) does not appear to be the only firing property correlated to a change in performance. In a recent study, it was found that low performance in a radial arm-maze memory
DO PLACE CELLS GUIDE SPATIAL BEHAVIORS?
Figure 9–2. A. Apparatus used in the shortcut study (Alverhne et al., unpublished results). The rats were trained to shuttle between the two extremities of an M-shaped maze to receive a pellet from two feeders (light circles). The inner walls of the maze were made of three parts (indicated by vertical lines) that could be removed individually, therefore allowing the rat to take a shortcut. Nine possible shortcut problems could be proposed to the animal but only six were actually used. The procedure involved three successive recording sessions, including a first standard session, a shortcut session, and a second standard session. B. Examples of rate maps in three different cells. Yellow pixels indicate no firing and purple indicates maximum firing. Orange, red, green, and blue indicate intermediate firing rates from low to high. The shortcut problem used while recording each cell is shown on the right. Fields that were near the removed wall (cell 1 and cell 2) were often affected by the shortcut. In contrast, fields that were far from the removed wall (cell 3) were usually not affected and remained stable relative to the standard session.
task performed in darkness was better correlated with a reduced specificity of place fields than with the spatial reorganization of these fields (Puryear et al., 2006). In contrast to the results presented thus far, there are a few studies that report a lack of coupling between behavior and place cell activity. Jeffery and coworkers (2003) recorded place cells as the rats per-
143
formed a spatial task in which they had to locate food in one corner of a square arena. One wall of the square contrasted with the other walls, therefore providing a spatial cue to the animal. Additional distal cues were located on the room wall. Changing the color of the arena produced a remapping in place cells but the animals still performed well. Although these results appear incompatible with the hypothesis of place cells controlling behavior, other possible accounts may resolve the discrepancy. For example, because the rats were largely overtrained at the time of recordings, it is possible that they had switched to a different strategy, such as a beacon strategy based on the use of an extramaze cue. This strategy would not depend on the hippocampus (Save and Poucet, 2000) but on some extrahippocampal region. Thus, lesioning the hippocampus after overtraining may not produce dramatic deficits. This hypothesis is consistent with previous results suggesting that place cell firing is not coupled with behavior when rats use beacon strategies (Trullier et al., 1999; Lenck-Santini et al., 2002). Further support for this conclusion is provided by the finding that place cells strongly react to spatial changes in the environment, but much less so to nonspatial change (Lenck-Santini et al., 2005), a pattern of results in line with the well-documented effects of hippocampal lesions (Save et al., 1992). Overall, the results largely support the notion that there is a relationship between place cell firing and spatial behavior, shown in various spatial tasks that are all assumed to depend on the use of an allocentric spatial representation, such as place learning, shortcut behavior, and spatial memory tasks. In contrast, strategies that do not involve a representation are not linked to place cell firing.
IS PLACE CELL FIRING CONTROLLED BY THE MEMORY OF BEHAVIORAL EPISODES? That the hippocampus is involved in encoding episodic memory is supported by studies showing that place cell activity can fire in relation to previous or intended behaviors. Wood et al. (2000) trained rats in a continuous delayed spatial-alternation task in a modified T-maze. Each goal arm of the T-maze was connected to the stem by a return alleyway so that after its choice, the rat could go back to the stem and perform the next trial. To exhibit correct performance in such a task, an animal had to remember its previous choice, i.e., a right or left turn, when running in the stem of the Tmaze. The authors examined whether firing in the stem depended on the last behavioral episode and found that most cells exhibited different discharge (field
144
PLACE CELLS AND SPATIAL CONTEXT
robustness, rate or location) on left-turn and right-turn trials. Because behavior was constant in the stem, explanations appealing to behavioral differences while traversing the field could be ruled out. The results suggest that place cells are influenced by the memory of behavioral events. Converging results were found by Franck et al. (2000) in hippocampal and entorhinal cortex neurons, but inconsistent results were found by Lenck-Santini et al. (2001b). In this latter study, place cells were recorded as the animals performed a continuous alternation task in a Y-maze (see previous section). However, unlike in Wood et al.’s study, place cell firing did not display any behavioral correlate with the previous or future turn of the rat. One possible explanation relates to procedural differences between the two tasks. In the Wood et al. study, due to a return procedure, the rats had to perform three successive turns on the same side, right or left, between two trials whereas in the Lenck-Santini et al. experiment, the rats permanently alternated between right and left turns. This may have strengthened the encoding of turns in Wood et al. study. Interestingly, in a recent study, Bower et al. (2005) showed that similar learning sequences in different contexts do not systematically underlie different firing in hippocampal cells. A recent study that used the same continuousalternation task in a T-maze (Lee et al., 2006) confirmed Wood et al.’s results and showed that location of the fields gradually shifted forward, toward the goal within a session, thus reflecting prospective memory coding. Continuous tasks, however, do not allow clear dissociation of retrospective and prospective memory control of place cell firing because the goal serves as the start point for the next trial. This dissociation was achieved by Ferbinteanu and Shapiro (2003), who trained rats to perform a place-learning task in a plus maze. In distinct trials (‘‘journeys’’), the animals had to run from two opposite start arms to one or the other of the two remaining arms. Analysis of firing in goal arms and start arms revealed that some fields were journey dependent and others journey independent. Journey-dependent fields were assumed to reflect retrospective coding when located in the goal arm (and thus were affected by the immediately preceding turn) and prospective coding when located in the start arm (and thus were affected by the intended turn). Journeyindependent fields showed spatial correlates only. Interestingly, the proportion of journey-dependent fields decreased in trials in which the rat made an error (entered an incorrect goal arm), thus suggesting a relationship between spatial performance and place cell firing. In a following study, Shapiro and Ferbinteanu showed that the temporal firing pattern of pairs of journey-independent cells, i.e., cells that fired for the same location, was different across different journeys.
In addition, a shift in spike timing between the two cells occurred more often in goal arms than in start arms, suggesting that the hippocampal cells could distinguish the beginning and the end of the journeys (Shapiro and Ferbinteanu, 2006). Overall, these results show that, in certain conditions, place cell activity can be modulated by the rat’s experience and behavioral context—that is, by the memory of past and future events.
DO PLACE CELLS ENCODE GOALS? One essential component of a navigational system would be the identification and specification of meaningful places such as goals, thus allowing the animal to compute accurate trajectories. The notion that the place cell system encodes goals or displays goalrelated firing in spatial tasks is appealing and has been addressed in a number of studies. These studies investigated several possibilities of goal–place cell firing relationships. Are place fields controlled by the goal? Speakman and O’Keefe (1990) examined the effect of rotating the goal in a plus maze located in a cue-controlled environment. Place fields did not rotate accordingly but remained stable relative to the external cues, indicating that the goal did not control place-cell firing (see also Lenck-Santini et al., 2001a). In contrast, Breese and colleagues (1989) found that manipulation of food location did influence field position. Rats were first trained to collect food from five randomly baited cups located in the corners and center of a square arena. When baiting was restricted to only one cup, the field shifted to the corresponding cup. Are place fields concentrated at the goal location? Gothard et al. (1996b) recorded place cells as the rats performed a place navigation task to get a food reward in a circular arena containing proximal and distal cues. They found that a proportion of the cells had fields bound to the goal location. Consistent with these results, Hollup and co-workers (2001) found an accumulation of fields at the platform location in an annular water maze. Accumulation of fields has also been described in nonspatial tasks (Eichenbaum et al., 1987). In contrast, Lenck-Santini et al. (2002) failed to observe such accumulation. Are place cells influenced by the behavioral significance of stimuli? A negative response to this question was given by Tabuchi and colleagues (2003), who suggested that place cells did not encode the reward value. In contrast, place cell firing was found to increase when the animal entered a baited arm in the radial arm maze relative to a non-baited arm (Ho¨lscher et al., 2003).
DO PLACE CELLS GUIDE SPATIAL BEHAVIORS?
Overall, these data are quite inconsistent, thus making any conclusion difficult. Goal-related firing in place cells must be interpreted with care on account of some confounding factors. First, it remains unclear whether place cells are controlled by the cue or the rewardrelated properties of the goal, or by some other more abstract property of goals. One possibility would be to diminish the cue component by providing a hidden goal as in Hollup et al.’s water-maze study (2001). Second, place cell firing may be influenced by particular behaviors exhibited at the goal by the animal (stillness, food consumption, movements, etc.), an issue that may be controlled by dissociating the goal from the reward site. The place preference task (Rossier et al., 2000; see above) allows one to dissociate these aspects. In this task the rats are trained to stay in a 20-cm-diameter goal zone for 2s to release a pellet from a feeder above. The goal zone is an unmarked area that the animal has to locate using a cue card on the wall and is spatially dissociated from the location where the animal is rewarded, since pellets drop and roll randomly in the arena. We recently reanalyzed the data of the Lenck-Santini et al. study (2002) in which rats were trained in the place preference task (Hok et al., 2007). In addition to having widely distributed firing fields, place cells also displayed weak but reliable discharge while the rat was in the goal zone (Fig. 9–3). To examine whether goal-related activity was specific to the goal status of the zone and did not result from the requirements of the task (staying still for 2 s), we conducted a number of analyses. Goal-related activity occurred when the animal navigated to the goal zone but not when it foraged erratically to collect pellets. Firing at the goal location did not simply result from low walking speed, since equivalent behavior occurring elsewhere in the apparatus was not associated with increased firing. It was not correlated with sharp waves and ripples associated with large irregular activity during quiet alertness (Buzsa`ki, 1986). Noticeably, the theta rhythm was not abolished while the rats paused at the goal location during correct responses but displayed a frequency shift. Finally, goal-related activity could not be explained by variations in interneuron activity. With these arguments, one can rule out the possibility that this activity results from nonspecific behavior and thus support the hypothesis that it is related to some specific processing at the goal. Furthermore, comparison of the place and cue tasks (in the cue task the goal zone was closely associated with a visible cue on the apparatus floor to the cue card) revealed that goal-related firing was found in the two tasks. However, they were characterized by different temporal profiles of discharge in the goal zone. In the place task, a firing increase occurred 1 s after the rat had entered the goal zone, whereas in the cue task it occurred before
145
entry into the goal zone. This pattern suggests that goalrelated firing is modulated by the nature of the task and thus by the processes underlying the task. It seems to be in operation in tasks requiring the use of a spatial map. Thus, we propose that this discharge is a feedback signal that would reflect computation of the animal’s location to ascertain that it is in the correct place. These results complement quite nicely those obtained in a previous study aimed at recording prefrontal cortex neurons during performance of the same task (Hok et al., 2005). The authors found that cells from the infralimbic and prelimbic areas displayed a spatial selective firing in relation to specific places such as the goal zone or the area where the pellets released from the overhead feeder landed. These fields were very different from those observed in the hippocampus by Hok et al.’s (2007) hippocampus study. They were much larger and noisier in the prefrontal cortex, which suggests that they did not mediate accurate spatial localization of the goal but instead provided gradientlike information allowing the animal to plan its paths. Taken together, these results indicate that goal encoding in the brain activates a functional network that at least involves the hippocampus and the prefrontal cortex. How these two structures interact remains to be determined, however. Overall, there is a growing amount of evidence suggesting that place cells contribute to goal encoding. This is likely a very schematic view, however. In particular, the notion of goal is somewhat abstract; one must take into account that there are multiple kinds of goals. For example, there are goals defined by the experimenter (e.g., the goal zone in the place preference task) or by the animal (e.g., the pellet drop-zone), explicit goals (associated with reward), implicit goals (home base), appetitive goals (rewarded), and aversive goals (places to avoid). Thus, these different goals may not involve similar encoding processes and may activate different brain structures. These differences may explain some of the inconstancies in the available data.
CONCLUSION Only recently have experiments been specifically designed to study the relationships between hippocampal neurons and spatial behavior. In the last few years, data have accumulated that point toward the idea that there is an interaction between place cells and behavior. Two complementary aspects of this interaction have emerged from these studies—namely, that place cells guide spatial behavior and, conversely, that behavior influences place cell firing. Place-cell guidance of spatial behavior is supported by two types of results. First, place cell firing is
146
PLACE CELLS AND SPATIAL CONTEXT
Figure 9–3. Examples of raster plots, peri-event time histograms (PETH), and time/rate maps for fourplace cells showing goal-related firing (Hok et al., 2007). Such firing was clearly visible in cells that had their main place field away from the goal zone. In the raster plots, each row shows the firing of a place cell in one trial with the small ticks representing one spike. The left vertical line indicates the animal’s entry into the goal zone and the right line indicates the end of the 2-s goal period. In the PETH, activity is accumulated in 200-ms bins, starting 12 s before the rat entered the goal and ending 4 s after a pellet was released (R). The 2-s goal period is bracketed by vertical lines. Raster plots and PETH show that in the place task, a firing increase occurred about 1s after the rat had entered the goal zone. In contrast, in the cue task, an increase occurred before entry into the goal zone (not shown). Variations of activity before and after the goal period resulted from crossings of the main field. The color-coded maps show the time spent by the animal in each pixel cumulated for the entire session. Yellow pixels indicate minimum time, purple indicates maximum time, and orange, red, green, and blue indicate intermediate times from low to high. The firing rate for the entire session is also shown. Yellow pixels indicate no firing and purple indicates maximum firing. Orange, red, green, and blue indicate intermediate firing rates from low to high. The firing rate corresponding to 4-sec navigation episodes, i.e., before pellet release (correct), and the firing rate corresponding to foraging episodes (other) are also indicated. Goal-related firing is seen in the ‘‘correct’’ and ‘‘rate’’ maps (indicated by the red arrow) but not in the ‘‘other’’ map. coupled with performance in tasks involving use of a spatial map. Alteration of firing is correlated with a drop in performance on various spatial tasks, including place learning in a plus maze or a y-maze, and place navigation in a circular arena. In addition,
modification of behavior correlated with modification of firing in a shortcut study. Noticeably, strategies that do not involve a spatial map are not controlled by place cells. Second, place cells may contribute to goal encoding, but such encoding appears to be modulated
DO PLACE CELLS GUIDE SPATIAL BEHAVIORS?
by factors that are not yet understood. The significance of goal-related firing remains to be clarified. Consistent with the basic properties of place cells, goalrelated firing occurs when the animal is at the goal. Thus it can hardly be seen as a prospective signal anticipating the arrival at the goal, nor as a strictly retrospective signal conveying the memory of the goal or some other event. In addition, the possibility that goal-related firing reflects only motivational or attentional states is unlikely because its temporal dynamics contrast markedly in two tasks that involve similar emotional and attentional demand but different spatial demands. Rather, these brief episodes of firing seem to indicate that the rat is at the correct location. Together with a number of results (Kobayashi et al., 1997), our findings suggest that individual place cells can be involved in both the coding of spatial goals (by virtue of goal-related firing) and the more general coding of space (by virtue of their firing at the main cell’s place field). Would this functional organization allow faster path computation? How flexible is this dual activity in response to environmental changes? Can multiple goals be represented in one cell? Is this signal processed by hippocampal circuits or is it conveyed to other brain structures? Recent neural models of navigation have delineated the contours of large networks that include numerous subcortical and cortical structures (Mizumori et al., 2000; Poucet et al., 2004). In this network, the prefrontal cortex likely contributes to goal-based action selection (Matsumoto et al., 2003). Thus, one hypothesis is that goal-related firing, in addition to reflecting computation of the animal’s location, serves as a feedback signal conveyed to the prefrontal cortex to inform the animal that the behavior was successful. Other studies indicate that behavior influences place cell firing. First, place-cell firing changes when the spatial demand of the task changes. Although place cells display positional firing in both the random foraging and goal-directed tasks, the firing properties are clearly different in the two tasks. One implication is that the place cell system is able to sustain different experience-dependent functional modes and readily alternates between these modes. Consequently, there would be a mode corresponding to a random foraging task and a mode corresponding to the goal-directed task. It turns out that reliability and robustness of spatial firing cells are decreased in random foraging tasks relative to goal-directed tasks. Successive training in the two tasks results in modifications of the firing properties of place cells, thus indicating that the place cell system is able to readily switch from one mode to the other. Although the mechanisms underlying these changes are unknown, most studies agree that cognitive processes such as attention, mo-
147
tivation, and emotion play a major role in activating and switching between different functional modes. Such aspects could be mediated by subcortical–cortical circuits involving, in particular, the hypothalamus, the septum, the basolateral amygdala, the parietal cortex, and the prefrontal cortex. Second, place cell activity is influenced by the memory of past and future events. Evidence that cells discharge differentially according to the last or next behavioral episode has accumulated, although the conditions under which it occurs are somewhat unclear. Time evolution of memory-dependent coding has been suggested in overtrained rats. However, it is not known how the differential discharge is established in relation to the dynamics of learning. The relationships between place cell activity and behavior are multifaceted and complex. Hippocampal place cells seem to play a pivotal role in spatial behavior, but they may represent only one element in a widely extended functional and anatomical network. This network includes other cell types exhibiting spatial signals, such as the head direction and grid cells. Head direction cells fire according to the rat’s head orientation and have been found in numerous brain areas of the limbic system (see Taube, 1998, for a review). Grid cells show a regular geometric pattern of spatially selective discharge and have been found in the medial entorhinal cortex (Hafting et al., 2005). Because head direction and grid cells are undoubtedly functionally related to hippocampal place cells, one important issue to examine is whether they also contribute to spatial behaviors. So far, this issue has been scarcely addressed (see Golob et al., 2001, for head direction cells), although it will probably generate numerous studies in the next decade. A peculiar property of hippocampal neurons is that their activity seems to be related to many aspects of spatial processing, ranging from sensory processing to behavior activation, via memory encoding. This variety of function rests on the intrinsic functional organization of the hippocampal place cell system and on its interactions with the rest of the brain. Any conceptual separation between sensory processing, memory encoding, and behavioral activation is probably artificial. Rather, these functions may be considered different facets of a functional continuum that involves many neural structures.
References Breese CR, Hampson RE, Deadwyler SA (1989) Hippocampal place cells: stereotypy and plasticity. J Neurosci 9:1097–1111. Bower MR, Euston DR, McNaughton BL (2005) Sequential-context-dependent hippocampal activity is
148
PLACE CELLS AND SPATIAL CONTEXT
not necessary to learn sequences with repeated elements. J Neurosci 25:1313–1323. Buzsa`ki G (1986) Hippocampal sharp waves: their origin and significance. Brain Res 398:242–252. Eichenbaum H, Dudchenko P, Wood E, Shapiro M, Tanila H (1999) The hippocampus, memory and place cells: is it spatial memory or a memory space? Neuron 23:209–226. Eichenbaum H, Kuperstein M, Fagan A, Nagode J (1987) Cue-sampling and goal-approached correlates of hippocampal unit activity in rats performing an odordiscrimination task. J Neurosci 7:716–732. Fenton AA, Csizmadia G, Muller RU (2000) Conjoint control of hippocampal place cell firing by two visual stimuli. I. The effects of moving the stimuli on firing field positions. J Gen Physiol 116:191–209. Ferbinteanu J, Shapiro ML (2003) Prospective and retrospective memory coding in the hippocampus. Neuron 40:1227–1239. Franck LM, Brown EN, Wilson M (2000) Trajectory encoding in the hippocampus and entorhinal cortex. Neuron 27:169–178. Golob EJ, Stackman RW, Wong AC, Taube JS (2001) On the behavioral significance of head direction cells: neural and behavioral dynamics during spatial memory tasks. Behav Neurosci 115:285–304. Gothard KM, Skaggs WE, McNaughton, BL (1996a) Dynamics of mismatch correction in the hippocampal ensemble code for space: interaction between path integration and environmental cues. J Neurosci 16:8027– 8040. Gothard KM, Skaggs WE, Moore KM, McNaughton BL (1996b) Binding of hippocampal CA1 neural activity to multiple reference frames in a landmark-based navigation task. J Neurosci 16:823–835. Hafting T, Fyhn M, Molden S, Moser M-B, Moser EI (2005) Microstructure of a spatial map in the entorhinal cortex. Nature 436:801–806. Hok V, Lenck-Santini PP, Roux S, Save E, Muller RU, Poucet B (2007) Goal-related activity in hippocampal place cells. J Neurosci 17:27:472–482. Hok V, Save E, Lenck-Santini P-P, Poucet B (2005) Coding for spatial goals in the prelimbic/infralimbic area of the rat frontal cortex. Proc Natl Acad Sci USA 102:4602–4607. Hollup SA, Molden S, Donnett JG, Moser M-B, Moser EI (2001) Accumulation of hippocmapl place fields at the goal location in an annular watermaze task. J Neurosci 21:1635–1644. Holscher C, Jacob W, Mallot HA (2003) Reward modulates neuronal activity in the hippocampus of the rat. Behav Brain Res 142:181–191. Jeffery KJ, Gilbert A, Burton S, Strudwick A (2003) Preserved performance in a hippocampal-dependent
spatial task despite complete place cell remapping. Hippocampus 13:175–189. Kentros CG, Agnihotri NT, Streater S, Hawkins RD, Kandel ER (2004) Increased attention to spatial context increases both place field stability and spatial memory. Neuron 42:283–295. Kobayashi T, Noshijo H, Fukuda M, Bures J, Ono T (1997) Task-dependent representations in rat hippocampal place neurons. J Neurophysiol 78:597–613. Kobayashi T, Tran AH, Nishijo H, Ono T, Matsumoto G (2003) Contribution of hippocampal place cell activity to learning and formation of goal-directed navigation in rats. Neuroscience 117:1025–1035. Kubie JL, Fenton AA, Muller R, Novikov N (2004) Behavioral responses parallel place cell responses to environmental manipulations. Soc Neurosci Abst 434.7 Lee I, Griffin AL, Zilli EA, Eichenbaum H, Hasselmo ME (2006) Gradual translocation of spatial correlates of neuronal firing in the hippocampus toward propective reward locations. Neuron 51:639–650. Lenck-Santini PP, Muller RU, Save E, Poucet B (2002) Relationships between place cell firing fields and navigational decisions by rats. J Neurosci 22:9035– 9047. Lenck-Santini PP, Rivard B, Muller RU, Poucet B (2005) Study of CA1 place cell activity and exploratory behavior following spatial and nonspatial changes in the environment. Hippocampus 15:356–369. Lenck-Santini P-P, Save E, Poucet B (2001a) Evidence for a relationship between place-cell spatial firing and spatial memory performance. Hippocampus 11:377– 390. Lenck-Santini PP, Save E, Poucet B (2001b) Place cells firing does not depend on the direction of turn in a Ymaze alternation task. Eur J Neurosci 13:1055–1058. Markus EJ, Qin Y, Leonard B, Skaggs WE, McNaughton BL, Barnes CA (1995) Interactions between location and task affect the spatial and directional firing of hippocampal neurons. J Neurosci 15:7079–7094. Matsumoto K, Suzuki W, Tanaka K (2003) Neuronal correlates of goal-based motor selection in the prefrontal cortex. Science 301:229–232. Mizumori SJ, Cooper BG, Leutgeg S, Pratt WE (2000) A neural systems analysis of adaptive navigation. Mol Neurobiol 21:57–82. Morris RGM, Garrud P, Rawlins JNP, O’Keefe J (1982) Place navigation impaired in rats with hippocampal lesions. Nature 297:681–683. Muller RU, Kubie JL (1987) The effects of changes in the environment on the spatial firing of hippocampal complex-spike cells. J Neurosci 7:1951–1968. O’Keefe J, Conway DH (1978) Hippocampal place cells in the freely moving rat: why they fire where they fire. Exp Brain Res 31:573–590.
DO PLACE CELLS GUIDE SPATIAL BEHAVIORS?
O’Keefe J, Dostrovsky J (1971) The hippocampus as a spatial map. Preliminary evidence from unit activity in the freely moving rat. Brain Res 34:171–175. O’Keefe J, Nadel L (1978) The Hippocampus as a Cognitive Map. Oxford: Clarendon Press. O’Keefe J, Speakman A (1987) Single unit activity in the rat hippocampus during a spatial memory task. Exp Brain Res 68:1–27. Olypher AV, Lansky P, Fenton AA (2002) Properties of the extra-positional signal in hippocampal place cell discharge derived from the overdispersion in locationspecific firing. Neuroscience 111:553–66. Packard MG, McGaugh JL (1996) Inactivation of hippocampus or caudate nucleus with lidocaine differentially affects expression of place and response learning. Neurobiol Learn Mem 65:65–72. Poucet B, Lenck-Santini PP, Hok V, Save E, Banquet JP, Gaussier P, Muller RU (2004) Spatial navigation and hippocampal place cell firing: the problem of goal encoding. Rev Neurosci 15:87–109. Puryear CB, King M, Mizumori SJY (2006) Specific changes in hippocampal spatial codes predict spatial working memory performance. Behav Brain Res 169: 168–175. Rossier J, Schenk F, Kaminsky Y, Bures J (2000) The place preference task: a new tool for studying the relation between behavior and place cell activity in rats. Behav Neurosci 114:373–284. Save E, Nerad L, Poucet B (2000) Contribution of multiple sensory information to place field stability in hippocampal place cells. Hippocampus 10:64–76. Save E, Paz-Villagra`n V, Alexinsky T, Poucet B (2005) Functional interaction between the parietal cortex and hippocampal place cell firing in the rat. Eur J Neurosci 21:522–530. Save E, Poucet B (2000) Involvement of the hippocampus and associative parietal cortex in the use of proximal and distal landmarks for navigation. Behav Brain Res 109:195–206. Save E, Poucet B, Foreman N, Buhot MC (1992) Object exploration and reactions to spatial and nonspatial changes in hooded rats following damage to parietal cortex or hippocampal formation. Behav Neurosci 106:447–456.
149
Save E, Poucet B, Thinus-Blanc C (1998) Landmark use and the cognitive map in the rat. In: Spatial Representation in Animals (Healy SD, ed.), pp 119–132. Oxford: Oxford University Press. Shapiro ML, Ferbinteanu J (2006) Relative spike timing in parts of hippocampal neurons distinguishes the beginning and the end of journeys. Proc Natl Acad Sci USA 103:4287–4292. Sharp PE, Blair HT, Etkin D, Tzanetos DB (1995) Influences of vestibular and visual motion information on the spatial firing pattern of hippocampal place cells. J Neurosci 15:173–189. Smith DM, Mizumori SJY (2006) Learning-related development of context-specific neuronal responses to places and events: the hippocampal role in context processing. J Neurosci 26:3154–3163. Speakman A, O’Keefe J (1990) Hippocampal complex spike cells do not change their place fields if the goal is moved within a cue-controlled environment. Eur J Neurosci 2:544–555. Tabuchi ET, Mulder AB, Wiener SI (2003) Reward value invariant place responses and reward site associated activity in hippocampal neurons of behaving rats. Hippocampus 13:117–132. Taube JS (1998) Head direction cells and the neurophysiological basis for a sense of direction. Prog Neurobiol 55:225–256. Tolman EC (1946) Cognitive maps in rats and men. Psychol Rev 55:189–208. Trullier O, Shibata R, Mulder AB, Wiener SI (1999) Hippocampal neuronal position selectively remains fixed to room cues only in rats alternating between place navigation and beacon approach tasks. Eur J Neurosci 11:4381–4388. Wood ER, Dudchenko PA, Robitsek RJ, Eichenbaum H (2000) Hippocampal neurons encode information about different types of memory episodes occurring in the same location. Neuron 27:623–633. Zinyuk L, Kubik S, Kaminsky Yu, Fenton, AA, Bures J (2000) Understanding hippocampal activity by using purposeful behavior: place navigation induce vs place cell discharge in both task-relevant and task-irrelevant spatial reference frames. Proc Natl Acad Sci USA 97:3771–3776.
10 Place Cells Identify Hippocampus with Location-Specific Construction of Mental Images NEIL BURGESS AND CHRIS M. BIRD
Place cells signal the current location of the animal within its environment, an obviously useful online function for navigation, and probably also a useful representation to be stored in spatial memory so that an unmarked location can be revisited. However, human neuropsychology points to a crucial role for the hippocampus more generally in context-dependent, or ‘‘episodic,’’ memory. Can the detailed characterization of place cell firing, built up over more than three decades, inform our understanding of the mechanisms underlying human memory? We argue that it can. Thus the long-held idea that the right hippocampus provides the ‘‘spatial context’’ for events within episodic memory (O’Keefe and Nadel, 1978, Chapter 14) can be made much more precise (see also Chapter 1, this volume). Specifically, three functional constraints on the hippocampal contribution to memory and imagery are implied by the functional characteristics of place cell firing. First, consistent with the perceptual influence on place cell firing, the hippocampus aids the retrieval or construction of visuospatial scenes by enabling efficient search through the mass of abstract, viewpointindependent information in the medial temporal lobes by constraining retrieved information to be consistent with perception from one single location. Second, consistent with the path-integrative influence on place cell firing, retrieval or imagery is facilitated by the ability to move the locus of the retrieval location within the retrieved or constructed scene. In addition, this ability results in the hippocampus supporting ‘‘allocentric’’ spatial memory over both short and long durations.
Third, consistent with remapping of place cell representations in different environments, and closest to the original idea of spatial context, retrieved information from a familiar environment is additionally constrained to be consistent with perception from a location within that specific environment. Overall, we use place cell data to outline the way in which the hippocampus provides the spatial context for retrieval or imagery that would otherwise remain acontextual or semantic. In this chapter we briefly review the three aspects of place cell firing referred to above, and explain how each leads to the corresponding functional constraint. We then discuss the ramifications of these constraints for the role of the hippocampus in memory and imagery. We outline a specific (re)constructive role for the hippocampus in these processes. With reference to other theories relating the hippocampus specifically to episodic memory, we note that, although the proposed construction of rich and dynamic first-person perspective imagery likely also creates a subjective feeling of ‘‘re-experience,’’ the same proposed constructive function will also occur in novel imagery and incorrect retrieval. Similarly, the same proposed function could underlie the distinction made between rich contextual recollection and feelings of familiarity. It is important to note, however, that while the hippocampus is necessary for the above processes, it is not sufficient. As well as imposing a single consistent location for retrieval, imagining the products of retrieval requires imposition of a specific viewing direction so that the retrieved (allocentric) information
150
LOCATION-SPECIFIC CONSTRUCTION OF MENTAL IMAGES
can be translated into an egocentric (head-centered) reference frame. This requires the contribution of multiple brain regions including the ‘‘head direction system’’ (Papez’s circuit), the retrosplenial cortex, medial parieto-occipital sulcus, and the posterior parietal lobe. In addition, mental movements of viewpoint during retrieval or planning are driven by mock motor-efference signals generated in the prefrontal cortex. Detailed discussion of the specific roles of these additional regions is beyond the scope of this chapter, but will be summarized in the Discussion.
PERCEPTUAL INFLUENCE ON PLACE CELL FIRING Place cells recorded in freely moving rats each fire whenever the animal enters a specific portion of its environment (the ‘‘place field’’; O’Keefe, 1976), thus signaling the rat’s location to the rest of the brain (O’Keefe and Nadel, 1978; Wilson and McNaughton, 1993; Muller, 1996). The firing of these cells is independent of the orientation of the rat during free exploration of open environments (Muller et al., 1994) and correlates with behavioral responses in some spatial memory tasks (O’Keefe and Speakman, 1987; Lenck-
151
Santini et al., 2001; see also Chapters 4 and 9, this volume). The orientation of the set of place fields within a symmetrical arena is controlled by distal cues (i.e., cues at or beyond the arena boundary; O’Keefe and Conway, 1978; Muller and Kubie, 1987; Cressant et al., 1997), most likely mediated by the head-direction system that they also control (Taube, 1998). In asymmetric environments, both distal cues and environmental geometry contribute to orientation (Jeffery et al., 1997), as well as ‘‘path integration’’ (Knierim et al., 1995; Jeffery and O’Keefe, 1999; see also below). Aside from overall orientation, the locations and shapes of place fields in open environments reflect conjunctions of the distances to the nearest boundaries in the various (allocentric) directions around the rat (O’Keefe and Burgess, 1996; Hartley et al., 2000; Barry et al., 2006). We refer to the cortical input to place cells carrying the sensory information concerning boundary distances and directions as ‘‘boundary vector cells’’ (BVCs, see Fig. 10–1). Whether these cells actually exist, or whether they are just functionally equivalent to the perceptual inputs driving place cell firing is not yet clear. Firing patterns consistent with BVCs have been seen in the subiculum (Sharp, 1999; Barry et al., 2006), but it is also possible that BVCs correspond to linear combinations of ‘‘grid cells’’ (Burgess et al., 2007).
Figure 10–1. Boundary vector cell (BVC) model of the geometrical influence on place fields, assuming a stable directional reference frame (adapted from Hartley et al., 2000; Barry et al., 2006). Place fields are composed from thresholded linear sums of the firing rates of BVCs. a. Top: Each BVC has a Gaussian tuned response to the presence of a boundary at a given distance and bearing from the rat (independent of its orientation). Its firing rate (bars on the left) decreases as a boundary’s distance and direction differ from the preferred values. Bottom: The tuning is sharper for shorter distances. The only free parameters of a BVC are the distance and direction of peak response. b. Place fields recorded from the same cell in four differently shaped environments. c. Simulation of the place fields in b using 4 BVC inputs (BVCs shown on the left, simulated fields on the right). d. Predicted firing of the modeled place cell in three novel-shaped environments. e. Observed firing of the actual cell in the three novel-shaped environments.
152
PLACE CELLS AND SPATIAL CONTEXT
IMPLICATIONS FOR LOCATION-SPECIFIC RETRIEVAL Many researchers, going back to Marr (1971), have argued that the CA3 field of the hippocampus, with its dense recurrent collaterals and (indirect) bidirectional connectivity with neocortical association areas, performs pattern completion to retrieve specific memories from partial input cues. According to this model, neural codes in hippocampal region CA3 can be reactivated by partial cues and in turn reactivate stored information in the neocortex via their back-projections (see also Teyler and DiScenna, 1986; see Burgess et al., 2001, for a review, and McClelland et al., 1995, for further details). In the case of spatial memory, there is some evidence that these index-like codes could be the place cell representations of location. Nakazawa and colleagues (2002) demonstrated that CA3-NR1 knockout mice were able to perform a spatial memory task in an environment containing many familiar distal landmarks (where no pattern completion was required), but that the removal of a subset of the landmarks impaired performance in the knock-out but not the healthy mice. Furthermore, the place cell representations of the environment were fully reactivated in the knock-out mice when all the orientation cues were present in a familiar environment, but not when the mice were reexposed to a familiar environment with some of the cues removed. These data are consistent with the ability of place cells to perform pattern completion from partial input cues to reinstate a full representation of a familiar environment. This view of the hippocampus performing pattern completion to retrieve memories necessarily implies that our memories are ‘‘reconstructive’’ in nature (see below for further discussion). However, the BVC model allows us to make more specific predictions about the type of memories that the hippocampus will retrieve. Our characterization of the BVC inputs to place cells implies a specific contribution to spatial memory within a familiar environment. That is, place cells are likely to form ‘‘attractor’’ representations in which the place cells that fire at a given location in the environment are associated both with each other and with the BVCs whose activity is consistent with the distances of environmental boundaries around that location (Burgess et al., 2001; Byrne et al., 2007; see also Fig. 10–2). In the context of robotics, a similar proposal was made: place cells represent the specific sonar pattern generated at a location within the environment, and fire as a function of the similarity of current sonar readings to this pattern (Recce and Harris, 1996). According to this model, if the input to the system is sufficient to retrieve the index code (or ‘‘chart,’’ see
Samsonovich and McNaughton, 1997) for a specific familiar environment, the learned associations between place cells and BVCs constrain the process of pattern completion to retrieve patterns of activation consistent with perception from a single location. For example, if the familiar environment has east–west dimensions 30 m, then the distances represented by active eastward BVCs and westward BVCs must sum to 30 m (orientations relative to distal cues rather than compasses), with similar constrains for other dimensions.
IMPLICATIONS FOR THE HIPPOCAMPAL ROLE IN EPISODIC MEMORY We store a vast amount of information in long-term memory and this potentially leads to a significant problem: when retrieving details about a single episode, how do we constrain the retrieval process? The critical role of the hippocampus, as suggested by our characterization of place cell firing, is to restrict the set of retrieved information to be consistent with perception from a single specific location. This automatically places a powerful constraint on the amount of information that needs be searched through when trying to remember a specific event. Of course, information about the locations of environmental features perceivable from a specific location is not the only information necessary to form an image of a scene. Imagery is invariably egocentric in nature, most probably taking a head-centered frame of reference. This is the case even if the images are from a third-person (or even a bird’s-eye-view) perspective. So to generate an internal image it is necessary to also impose a specific direction of view on the retrieved information. Consequently, the information retrieved by active place cells must be transformed from an allocentric representation into an egocentric representation. This translation is thought to be performed by the retrosplenial and parietal cortical regions, via the parahippocampus, with imagery occurring in the precuneus (see Burgess et al., 2001; Byrne et al., 2007, and Discussion for further details). More generally, constraining retrieved information to be consistent with a specific viewpoint is a prerequisite, not just for the reconstruction of personally experienced events but also for imagery more generally. It follows that the hippocampus is necessary to construct a spatially coherent imagined scene. In support of this idea, it was recently demonstrated that amnesic patients, with damage limited largely to the hippocampus, were impaired in their ability to imagine spatially coherent scenarios (e.g., being in a mar-
LOCATION-SPECIFIC CONSTRUCTION OF MENTAL IMAGES
153
Figure 10–2. Consistent place and boundary vector cell (BVC) activation. a. The current location of a person in an arena surrounded by buildings. b. Place cell activation consistent with being in this location. c. The BVCs associated with this place cell in this environment. Since the place cells and BVCs are reciprocally connected, activation of this place cell will reinstate activity in the BVCs, signaling the distances and directions of the surrounding buildings associated with being in that location.
ket or on a beach) (Hassabis et al., 2007). By contrast with healthy adults, the patients’ constructed scenes were fragmentary and spatially incoherent in nature. Functional neuroimaging has also demonstrated remarkable concordance with the brain regions recruited while recalling personally experienced episodes and those recruited when imaging future events (Addis et al., 2007). Importantly, the hippocampus was activated in both tasks.
PATH-INTEGRATIVE INFLUENCE ON PLACE CELL FIRING Path integration, sometimes known as dead reckoning, is the ability to keep track of a start location on the basis of subsequent self-motion information (Mittelstaedt and Mittelstaedt, 1982; see Etienne and Jeffery, 2004, for a recent review). It is thought that the shortterm coherence of place cell representations is auto-
matically maintained via path integration (O’Keefe and Nadel, 1978; McNaughton et al., 1996). This process may reflect an interaction between the parietal and hippocampal systems (Alyan and McNaughton, 1999; Save et al., 2001). For example, lesions to the associative parietal cortex of rats result in altered place cell firing, suggesting that egocentric sensory information must travel through parietal cortex to elicit appropriate place cell firing in the hippocampus (Save et al., 2005). The interaction between parietal and medial temporal areas likely involves retrosplenial cortex, lesions of which selectively disrupt path integration (Cooper et al., 2001), and which contains cells coding for locations in space (Galletti et al., 1995). Where path integration, arena shape, and distal cues to orientation are all pitted against one another, distal cues seem the most potent (Jeffery et al., 1997) unless they are seen to be unstable (Jeffery and O’Keefe, 1999) or the rat is systematically disoriented before each trial (Knierim et al., 1995).
154
PLACE CELLS AND SPATIAL CONTEXT
There have been various models of how path integration could influence place cell firing (see, e.g., McNaughton et al., 2006; Burgess, 2007, for reviews). Recent models are influenced by the discovery of grid cells in the entorhinal cortex—cells that fire when a rat occupies multiple locations laid out on a hexagonal grid (Hafting et al., 2005). Local sets of grid cells could be tuned to perform path integration irrespective of the location of the animal. Thus, grid cells may operate as a continuous attractor for path integration, with the stable association of each grid to the environment provided by place cells associated with distal cues observable at a particular location (O’Keefe and Burgess, 2005; Burgess et al., 2007; see also Fuhs and Touretzky, 2006; McNaughton et al., 2006).
IMPLICATIONS FOR THE HIPPOCAMPAL ROLE IN EPISODIC MEMORY Path integration is a subset of the more general process of motion-related spatial updating in which position of multiple locations around one can be updated using knowledge of one’s own motion. In humans, there is evidence that actual spatial updating and imagined movements of viewpoint share common cognitive processes (Wang and Simons, 1999; Burgess et al., 2004; see Burgess, 2006, for a review). Both processes may be driven by real motor-efference signals or by prefrontally generated mock motor-efference using the same temporoparietal circuitry (see Discussion). We argue that the combination of mental spatial updating and the memory codes stored in the hippocampus facilitates a key component of episodic memory—the ability to move one’s viewpoint within a remembered scene. Without this ability, memories would be static ‘‘snapshots.’’ Just as path integration maintains the short-term coherence of the place cell representation, the same mechanism could be used for smoothly linking events within episodic memory. Thus, memories (and novel, imagined scenarios) can be dynamic in nature, with the viewpoint changing as the observer moves around. For example, in order to fully recollect aspects of an episode, it may be useful to retrieve details from a particular viewpoint (e.g., standing outside a friend’s house) and then ‘‘playing’’ what happened next (e.g., seeing who was there as you move into the house). There is evidence that the hippocampus is necessary to shift viewpoints within an imagined scene, consistent with the model outlined above. King and colleagues designed a task that required the remembering of object locations within a virtual town square, with memory tested from the same view as or a different view from that seen at study. A shift in the
remembered viewpoint was only required in the latter condition. In healthy subjects, reaction times were proportional to the angular shift in viewpoint between study and test, suggesting that a rotation from the remembered to the currently observed viewpoint was carried out. Importantly, a patient with bilateral hippocampal damage was selectively impaired in the different view conditions (King et al., 2002, 2004). If participants are allowed to watch the locations rotate in front of them, no such mental shift of viewpoint is required and performance is unaffected by hippocampal damage (Shrager et al., 2007). If the hippocampus has a key role in moving viewpoints within a remembered (or imagined) scene, it necessarily follows that the hippocampus is involved in short-term as well as long-term memory if a shift in viewpoint is required. This association was investigated by Hartley et al. (2007), who showed that hippocampal patients were unable to match shifted views of mountain scenes across a delay of only 2s (see also Lee et al., 2005a,b; Hannula et al., 2006; Olson et al., 2006). It is even plausible to expect that the hippocampus should play a role in the perception and online comparison of complex spatial scenes. However, little evidence was found for this by Hartley et al. (2007), and the authors concluded that the parahippocampus is probably sufficient to support this ability using a point-by-point comparison strategy, provided that the stimuli remain visible (see also Shrager et al., 2006).
REMAPPING OF PLACE CELL REPRESENTATIONS: WHAT DEFINES A ‘‘CONTEXT’’? Although terms such as context and contextual cues are commonly used in memory research, they remain ill defined. Place cell recording has offered a clear rationale for conceptualizing context. As reviewed above, place cells perform pattern completion to retrieve details about a familiar environment, given partial input about features in that environment. However, they also appear to play another role, signaling the environment in which the animal is at the moment (e.g., Kubie and Muller, 1991). Thus, the place cell representations of two very different environments are radically different, or ‘‘remapped’’ (e.g., Bostock et al., 1991). Wills et al. (2005) demonstrated that two differently shaped environments acted as ‘‘attractors’’ for place cell representations of those environments. In this experiment, rats were exposed to circular and square environments of different color, and place cells showed distinct patterns of firing dependent upon the shape of the environment. Next the rats were trans-
LOCATION-SPECIFIC CONSTRUCTION OF MENTAL IMAGES
ferred to a deformable box in which the circular or square pattern of firing was found according to the shape of the box. Then the shape of the box was configured to be more or less ‘‘square-like’’ or ‘‘circularlike’’ intermediate shapes. Place cells abruptly and simultaneously changed their patterns of firing following incremental changes in the shapes of the environments–returning to either the square-like or the circle-like representation in the intermediate environments. The Wills et al. study demonstrated the place cell representation’s ability to generalize across similar environments to produce stable representations of those places. Remapping occurs when two environments are sufficiently different to signal a change in ‘‘context’’ (see also Jeffery et al., 2004). Direct predictions can be made from this about an animal’s behavior in paradigms such as fear conditioning. If a tone is paired with a shock in one context, then the rat will exhibit context-dependent freezing to the tone—i.e., it will again freeze when hearing the tone in that context, but will freeze much less when in another context (Kim and Fanselow, 1992; Phillips and LeDoux, 1994). It is not clear which aspects of a new environment define whether it acts as the same or a different context. However, a simple prediction is that an environment that elicits a remapped place cell representation compared to the original environment will act as a new context. Nadel and colleagues (O’Keefe and Nadel, 1978, see also Chapter 1, this volume) have particularly stressed the role of the hippocampus in ‘‘spatial context,’’ which they separate from other types of context such as ‘‘temporal context.’’ Similarly, Maguire and colleagues have discussed the hippocampal role in providing a ‘‘spatial framework’’ (Hassabis et al., 2007). We can formalize these concepts more precisely as the ability to construct the surface geometry of an environment from a given viewpoint (see above). Furthermore, place cell remapping implies that viewpoints can be constrained to be located in a given environment (i.e., all of the locations within an environment form an attractor or ‘‘chart’’). This also allows pattern completion appropriate to that environment: a spatial-contextual interpretation of Marr’s 1971 model of the hippocampus as an associative memory system. This discussion has focused on ‘‘fast remapping,’’ where place cells rapidly change their firing patterns following a change in context and respond coherently to intermediate contexts (Wills et al., 2005). It is possible that an ‘‘event’’ is necessary to trigger fast remapping. However, there is another form of remapping that takes place incrementally over several days (Lever et al., 2002). This may be likened to models of long-term memory (Howard and Kahana, 2002) or memory for
155
temporal order (Burgess and Hitch, 1999), which depend on a slowly evolving representation of context. One possibility is that the hippocampus provides this (temporal) context signal (see Burgess and Hitch, 2005; Wallenstein et al., 1998, for further discussion).
DISCUSSION Relationship of Construction to Subjective ‘‘Re-experience’’ We have proposed that the hippocampus plays a specific role in the reconstruction of a coherent spatial context. In some circumstances these contexts are constrained to be from a previously encountered environment (i.e., reconstructed from memory), in others they may be completely novel. Several lines of evidence suggest that in humans this (overtly spatial) role is likely to be underpinned by the right hippocampus (see Burgess et al., 2002). However, spatial information is not the only type of contextual information that can be retrieved within an ‘‘episode.’’ It may be that other information is supported by the left hippocampus, such as narrative descriptions, temporal ordering, odors, etc. (e.g., O’Keefe and Nadel, 1978; Burgess et al., 2002). Our characterization of the hippocampus performing pattern completion to recreate a familiar context might be misinterpreted to imply that we view the retrieval of episodes as ‘‘holistic’’ in nature; that is, the retrieval of some details of an episode will cue the retrieval of all details. Such a process would allow one to ‘‘re-experience’’ all aspects of an event—a phenomenon that has been argued to epitomize episodic memory (Wheeler et al., 1997; Tulving and Markowitsch, 1998; Tulving, 2001). On the contrary, in our model of how the hippocampus reconstructs the spatial context of an event, or indeed other aspects of the event, retrieved information is constrained to be consistent with perception from a single viewpoint. It is not constrained to be veridical, or for the various elements of the actual event to be necessarily bound together in any holistic way. In fact, there is little evidence that they are. For example, Trinkler et al. (2006) described three experiments in which virtual reality was used to present a ‘‘walk’’ thorough a virtual town, during which different characters were encountered in different places and gave objects to the testee (and in one experiment, distinctive odors were sniffed during each encounter). Despite demonstration of the hippocampal dependence of this task (King et al., 2004), there was no evidence that the ‘‘episodes’’ were encoded in a holistic manner, the data being well fitted by assuming independent pair-wise associations between the various aspects of the events.
156
PLACE CELLS AND SPATIAL CONTEXT
The view that memories characterized as episodic are reconstructed is not new (Bartlett, 1932) and can result in striking distortions in many situations (e.g., Schacter and Dodson, 2001; Loftus, 2005). In fact, if retrieval is partial and constructive in nature, rather than all-or-none, the subjective experience of ‘‘reexperience’’ or ‘‘recollection’’ may simply reflect the construction of a sufficient amount of plausible contextual information into an ‘‘episode’’ irrespective of its relationship to the past (or future).
Relationship of Hippocampus to Other Areas Involved in Memory: Translation between Allocentric and Egocentric Representations Most models of memory propose that consciously accessible information or knowledge is stored in the neocortex. We would argue that this information is retained in a relatively ‘‘piecemeal’’ or context-less state. This is similar to the conception of how semantic knowledge is stored (Marr, 1970). In fact, the brain may not distinguish between knowledge that has traditionally been termed ‘‘semantic’’ and that termed ‘‘episodic.’’ The difference may be merely that retrieval of episodes requires the elements to be (re)constructed to form a coherent scene or episode, while the retrieval of semantic information does not—or it is reconstructed in a qualitatively different way that is independent of the hippocampus. We propose that the codes, indices, or charts necessary for reconstructing stored information into a coherent episode are stored within the hippocampus in allocentric coordinates, providing an efficient method of storage and retrieval. Furthermore, long-term memory for locations is best served by allocentric representations (i.e., relative to stable landmarks) because the location and configuration of the body at retrieval typically will be unrelated to that at encoding (see Burgess et al., 2001, for further discussion). Equally, the short-term retention of perceptual information for the purpose of immediate action will be best served by egocentric representations appropriate to the corresponding sensory and motor systems (Milner et al., 1999). Thus, when we call to mind an episode, we are in effect creating a working memory representation that is egocentric in nature. Spatial location is encoded by place cells, and the combination of the subsets of spatial information necessary to recreate a scene is likely to take place in the parahippocampal gyrus—specifically, its boundary geometry ordered by allocentric direction (Byrne et al., 2007). However, the imposition of an egocentric viewpoint on this allocentric information requires complementary spatial information—the encoding of orientation independent of location. Such ‘‘head
direction cells’’ (Taube, 1998) are found along an anatomical circuit largely homologous to Papez’s circuit (Papez, 1937) leading from the mammillary bodies to the presubiculum via the anterior thalamus. Furthermore, head direction–selective neurons that exhibit responses tuned to various different reference frames have been found in the retrosplenial cortex of the rat (Chen et al., 1994). The parietal lobe is implicated in spatial working memory and sensorimotor integration and probably plays a role in the translation from allocentric to egocentric coordinate frames. The posterior parietal cortex contains neurons that exhibit egocentrically tuned responses to visual locations that are also modulated by variables such as eye position and head and body orientation (e.g., Andersen et al., 1985; Snyder et al., 1998). Such coding can allow transformation of locations between reference frames (Zipser and Andersen, 1988; Pouget and Sejnowski, 1997). In humans, transcranial magnetic stimulation and fMRI studies also indicate that areas surrounding the right intraparietal sulcus are essential in the generation and manipulation of egocentric mental imagery (Formisano et al., 2002; Sack et al., 2002). Damage to this area frequently results in hemispatial neglect—a severe disorder of spatial cognition that operates in multiple reference frames (Mort et al., 2003). Medial parietal cortex may play a particularly important role in imagery for the products of retrieval. Galletti and colleagues (1995) documented neurons in the anterior bank of the medial parieto-occipital sulcus that represent the positions of visual stimuli in a craniotopic reference frame. Human studies have implicated regions of the medial parietal cortex (such as the precuneus) in mental imagery (Fletcher et al., 1996) and visuospatial working memory (Wallentin et al., 2006). Damage to the medial parieto-occipital sulcus has been associated with topographical disorientation in humans (Maguire, 2001). Such patients, while able to recognize familiar landmarks, are unable to derive directional information from the landmarks necessary for successful navigation (Aguirre and D’Esposito, 1999). Prefrontal regions are also implicated in spatial working memory, with parietal areas predominantly associated with storage and prefrontal areas with the application of control processes such as active maintenance or planning (Stuss and Knight, 2002), using the posterior spatial representations. Thus, fMRI studies have shown activation in both of these areas when subjects were required to remember the locations of various objects for short periods of time (Smith and Jonides, 1999). Manipulations of working memory may also involve making or planning eye movements to direct attention to spatial locations in imagery. In
LOCATION-SPECIFIC CONSTRUCTION OF MENTAL IMAGES
support of this notion, voluntary eye movements disrupt spatial working memory (Postle et al., 2006), while patients with left hemispatial neglect show abnormal eye movements that deviate rightward during visual search (Behrmann et al., 1997) as well as during rest (Fruhmann-Berger and Karnath, 2005). Studies involving mental navigation and route planning consistently find elevated activation in frontal regions, especially on the left side (Ghaem et al., 1997; Maguire et al., 1998; Ino et al., 2002). For example, Maguire et al. (1998) found additional activation in left prefrontal cortex associated with the planning of detours when subjects were navigating in a familiar virtual town in which the most obvious route had unexpectedly been blocked. This suggests that left prefrontal areas contribute to route planning, perhaps guiding egocentric mental imagery within the temporoparietal systems activated by the basic navigation condition. A model for the interaction of these various brain regions with the hippocampus has recently been proposed by Byrne and colleagues (2007). According to their framework, and as we have argued above, the hippocampus indexes the retrieved information to be consistent with an imagined location, allowing reconstruction of the set of visual textures, distances, and allocentric directions of surrounding landmarks in the parahippocampal gyrus. Head direction cells, in combination with processing in posterior parietal and retrosplenial cortices, allow translation of this allocentric information into egocentric coordinates for imagery according to the direction of view. Within this model, modulation of the allocentric to egocentric translation by motor efference allows ‘‘spatial updating’’ of egocentric parietal representations, and the generation of mock motor efference in prefrontal cortex allows the simulation of planned movements. This enables mental exploration in both retrieval and imagery, making a potential contribution to spatial planning (e.g., to see whether the goal location would be visible after going left or right at the next corner).
CONCLUSIONS Single-cell recording from hippocampal place cells has enabled detailed models of spatial memory to be developed. We have focused on one, the BVC model. This model predicts a role for the hippocampus in the construction of rich and dynamic first-person perspective imagery. Episodic memory is simply one form of imagery—the reconstruction of previously experienced events. The ability to construct such images gives rise to feelings of recollection or reexperiencing the events. The model also predicts a role for the hippocampus in some forms of spatial
157
working memory and the ability to construct novel images, consistent with recent experimental data.
References Addis DR, Wong AT, Schacter DL (2007) Remembering the past and imagining the future: common and distinct neural substrates during event construction and elaboration. Neuropsychologia 45:1363–1377. Aguirre GK, D’Esposito M (1999) Topographical disorientation: a synthesis and taxonomy. Brain 122: 1613–1628. Alyan S, McNaughton BL (1999) Hippocampectomized rats are capable of homing by path integration. Behav Neurosci 113:19–31. Andersen RA, Essick GK, Siegel RM (1985) Encoding of spatial location by posterior parietal neurons. Science 230:456–458. Barry C, Lever C, Hayman R, Hartley T, Burton S, O’Keefe J, Jeffery KJ, Burgess N (2006) The boundary vector cell model of place cell firing and spatial memory. Rev Neurosci 17:71–97. Bartlett FC (1932) Remembering: A Study in Experimental and Social Psychology. Cambridge, UK: Cambridge University Press. Behrmann M, Watt S, Black SE, Barton JJ (1997) Impaired visual search in patients with unilateral neglect: an oculographic analysis. Neuropsychologia 35:1445– 1458. Bostock E, Muller RU, Kubie JL (1991) Experience-dependent modifications of hippocampal place cell firing. Hippocampus 1:193–205. Burgess N (2006) Spatial memory: how egocentric and allocentric combine. Trends Cogn Sci 10:551–557. Burgess N (2007) Computational models of the spatial and mnemonic functions of the hippocampus. In: The Hippocampus Book (Andersen P, Morris RG, Amaral DG, Bliss TV, O’Keefe J, eds.), pp 715–750. New York: Oxford University Press. Burgess N, Barry C, O’Keefe J (2007) An oscillatory interference model of grid cell firing. Hippocampus, in press (DOI 10.1002/hipo.20327) Burgess N, Becker S, King JA, O’Keefe J (2001) Memory for events and their spatial context: models and experiments. Philos Trans R Soc Lond B Biol Sci 356: 1493–1503. Burgess N, Hitch GJ (1999) Memory for serial order: a network model of the phonological loop and its timing. Psychol Rev 106:551–581. Burgess N, Hitch GJ (2005) Computational models of working memory: putting long-term memory into context. Trends Cogn Sci 9:535–541. Burgess N, Maguire EA, O’Keefe J (2002) The human hippocampus and spatial and episodic memory. Neuron 35:625–641.
158
PLACE CELLS AND SPATIAL CONTEXT
Burgess N, Spiers HJ, Paleologou E (2004) Orientational manoeuvres in the dark: dissociating allocentric and egocentric influences on spatial memory. Cognition 94:149–166. Byrne P, Becker S, Burgess N (2007) Remembering the past and imagining the future: a neural model of spatial memory and imagery. Psychol Rev 114:340–375. Chen LL, Lin LH, Barnes CA, McNaughton BL (1994) Head-direction cells in the rat posterior cortex. II. Contributions of visual and ideothetic information to the directional firing. Exp Brain Res 101:24–34. Cooper BG, Manka TF, Mizumori SJY (2001) Finding your way in the dark: the retrosplenial cortex contributes to spatial memory and navigation without visual cues. Behav Neurosci 115:1012–1028. Cressant A, Muller RU, Poucet B (1997) Failure of centrally placed objects to control the firing fields of hippocampal place cells. J Neurosci 17:2531–2542. Dodson CS, Schacter DL (2002) The cognitive neuropsychology of false memories: theory and data. In: The Handbook of Memory Disorders (Baddeley AD, Kopelman MD, Wilson BA, eds.), pp 343–362. New York: John Wiley & Sons. Etienne AS, Jeffery KJ (2004) Path integration in mammals. Hippocampus 14:180–192. Fletcher PC, Shallice T, Frith CD, Frackowiak RS, Dolan RJ (1996) Brain activity during memory retrieval. The influence of imagery and semantic cueing. Brain 119: 1587–1596. Formisano E, Linden DE, Di SF, Trojano L, Esposito F, Sack AT, Grossi D, Zanella FE, Goebel R (2002) Tracking the mind’s image in the brain I: timeresolved fMRI during visuospatial mental imagery. Neuron 35:185–194. Fruhmann-Berger M, Karnath HO (2005) Spontaneous eye and head position in patients with spatial neglect. J Neurol 252:1194–1200. Fuhs MC, Touretzky DS (2006) A spin glass model of path integration in rat medial entorhinal cortex. J Neurosci 26:4266–4276. Galletti C, Battaglini PP, Fattori P (1995) Eye position influence on the parieto-occipital area PO (V6) of the macaque monkey. Eur J Neurosci 7:2486–2501. Ghaem O, Mellet E, Crivello F, Tzourio N, Mazoyer B, Berthoz A, Denis M (1997) Mental navigation along memorized routes activates the hippocampus, precuneus, and insula. Neuroreport 8:739–744. Hafting T, Fyhn M, Molden S, Moser MB, Moser EI (2005) Microstructure of a spatial map in the entorhinal cortex. Nature 436:801–806. Hannula DE, Tranel D, Cohen NJ (2006) The long and the short of it: relational memory impairments in amnesia, even at short lags. J Neurosci 26:8352–8359. Hartley T, Bird CM, Chan D, Cipolotti L, Husain M, Vargha-Khadem F, Burgess N (2007) The hippocam-
pus is required for short-term topographical memory in humans. Hippocampus 17:34–48. Hartley T, Burgess N, Lever C, Cacucci F, O’Keefe J (2000) Modeling place fields in terms of the cortical inputs to the hippocampus. Hippocampus 10:369–379. Hassabis D, Kumaran D, Vann SD, Maguire EA (2007) Patients with hippocampal amnesia cannot imagine new experiences. Proc Natl Acad Sci USA 104:1726–1731. Howard M, Kahana MJ (2002) A distributed representation of temporal context. J Math Psychol 46:269–299. Ino T, Inoue Y, Kage M, Hirose S, Kimura T, Fukuyama H (2002) Mental navigation in humans is processed in the anterior bank of the parieto-occipital sulcus. Neurosci Lett 322:182–186. Jeffery KJ, Anderson MI, Hayman R, Chakraborty S (2004) A proposed architecture for the neural representation of spatial context. Neurosci Biobehav Rev 28:201–218. Jeffery KJ, Donnett JG, Burgess N, O’Keefe JM (1997) Directional control of hippocampal place fields. Exp Brain Res 117:131–142. Jeffery KJ, O’Keefe JM (1999) Learned interaction of visual and idiothetic cues in the control of place field orientation. Exp Brain Res 127:151–161. Kim JJ, Fanselow MS (1992) Modality-specific retrograde amnesia of fear. Science 256:675–677. King JA, Burgess N, Hartley T, Vargha-Khadem F, O’Keefe J (2002) Human hippocampus and viewpoint dependence in spatial memory. Hippocampus 12: 811–820. King JA, Trinkler I, Hartley T, Vargha-Khadem F, Burgess N (2004) The hippocampal role in spatial memory and the familiarity-recollection distinction: a single case study. Neuropsychology 18:405–417. Knierim JJ, Kudrimoti HS, McNaughton BL (1995) Place cells, head direction cells, and the learning of landmark stability. J Neurosci 15:1648–1659. Kubie JL, Muller RU (1991) Multiple representations in the hippocampus. Hippocampus 1:240–242. Lee AC, Buckley MJ, Pegman SJ, Spiers HJ, Scahill VR, Gaffan D, Bussey TJ, Davies RR, Kapur N, Hodges JR, Graham KS (2005a) Specialization in the medial temporal lobe for processing of objects and scenes. Hippocampus 15:782–797. Lee AC, Bussey TJ, Murray EA, Saksida LM, Epstein RA, Kapur N, Hodges JR, Graham KS (2005b) Perceptual deficits in amnesia: challenging the medial temporal lobe ‘mnemonic’ view. Neuropsychologia 43:1–11. Lenck-Santini PP, Save E, Poucet B (2001) Evidence for a relationship between place-cell spatial firing and spatial memory performance. Hippocampus 11:377–390. Lever C, Burgess N, Cacucci F, Hartley T, O’Keefe J (2002) What can the hippocampal representation of environmental geometry tell us about Hebbian learning? Biol Cybern 87:356–372.
LOCATION-SPECIFIC CONSTRUCTION OF MENTAL IMAGES
Loftus EF (2005) Planting misinformation in the human mind: a 30-year investigation of the malleability of memory. Learn Mem 12:361–366. Maguire EA (2001) The retrosplenial contribution to human navigation: a review of lesion and neuroimaging findings. Scand J Psychol 42:225–238. Maguire EA, Burgess N, Donnett JG, Frackowiak RS, Frith CD, O’Keefe J (1998) Knowing where and getting there: a human navigation network. Science 280: 921–924. Marr D (1970) A theory for cerebral cortex. Proc R Soc Lond B Biol Sci 176:161–234. Marr D (1971) Simple memory: a theory for archicortex. Philos Trans R Soc Lond B Biol Sci 262:23–81. McClelland JL, McNaughton BL, O’Reilly RC (1995) Why there are complementary learning systems in the hippocampus and neocortex: insights from the successes and failures of connectionist models of learning and memory. Psychol Rev 102:419–457. McNaughton BL, Barnes CA, Gerrard JL, Gothard K, Jung MW, Knierim JJ, Kudrimoti H, Qin Y, Skaggs WE, Suster M, Weaver KL (1996) Deciphering the hippocampal polyglot: the hippocampus as a path integration system. J Exp Biol 199:173–185. McNaughton BL, Battaglia FP, Jensen O, Moser EI, Moser MB (2006) Path integration and the neural basis of the ‘cognitive map.’ Nat Rev Neurosci 7:663– 678. Milner AD, Dijkerman HC, Carey DP (1999) Visuospatial processing in a case of visual form agnosia. In: The Hippocampal and Parietal Foundations of Spatial Cognition (Burgess N, Jeffery KJ, O’Keefe J, eds.), pp 443–466. Oxford: Oxford University Press. Mittelstaedt H, Mittelstaedt M-L (1982) Homing by path integration. In: Avian Navigation (Papi F, Wallraff HG, eds.), pp 290–297. New York: SpringerVerlag. Mort DJ, Malhotra P, Mannan SK, Rorden C, Pambakian A, Kennard C, Husain M (2003) The anatomy of visual neglect. Brain 126:1986–1997. Muller R (1996) A quarter of a century of place cells. Neuron 17:813–822. Muller RU, Bostock E, Taube JS, Kubie JL (1994) On the directional firing properties of hippocampal place cells. J Neurosci 14:7235–7251. Muller RU, Kubie JL (1987) The effects of changes in the environment on the spatial firing of hippocampal complex-spike cells. J Neurosci 7:1951–1968. Nakazawa K, Quirk MC, Chitwood RA, Watanabe M, Yeckel MF, Sun LD, Kato A, Carr CA, Johnston D, Wilson MA, Tonegawa S (2002) Requirement for hippocampal CA3 NMDA receptors in associative memory recall. Science 297:211–218. O’Keefe J (1976) Place units in the hippocampus of the freely moving rat. Exp Neurol 51:78–109.
159
O’Keefe J, Burgess N (1996) Geometric determinants of the place fields of hippocampal neurons. Nature 381: 425–428. O’Keefe J, Burgess N (2005) Dual phase and rate coding in hippocampal place cells: theoretical significance and relationship to entorhinal grid cells. Hippocampus 15:853–866. O’Keefe J, Conway DH (1978) Hippocampal place units in the freely moving rat: why they fire where they fire. Exp Brain Res 31:573–590. O’Keefe J, Nadel L (1978) The Hippocampus as a Cognitive Map. Oxford: Clarendon Press. O’Keefe J, Speakman A (1987) Single unit activity in the rat hippocampus during a spatial memory task. Exp Brain Res 68:1–27. Olson IR, Page K, Moore KS, Chatterjee A, Verfaellie M (2006) Working memory for conjunctions relies on the medial temporal lobe. J Neurosci 26:4596– 4601. Papez JW (1937) A proposed mechanism of emotion. Arch Neurol Psychiatry 38:724–744. Phillips RG, LeDoux JE (1994) Lesions of the dorsal hippocampal-formation interfere with background but not foreground contextual fear conditioning. Learn Mem 1:34–44. Postle BR, Idzikowski C, Sala SD, Logie RH, Baddeley AD (2006) The selective disruption of spatial working memory by eye movements. Q J Exp Psychol (Colchester) 59:100–120. Pouget A, Sejnowski TJ (1997) A new view of hemineglect based on the response properties of parietal neurones. Philos Trans R Soc Lond B Biol Sci 352:1449–1459. Recce M, Harris KD (1996) Memory for places: a navigational model in support of Marr’s theory of hippocampal function. Hippocampus 6:735–748. Sack AT, Sperling JM, Prvulovic D, Formisano E, Goebel R, Di SF, Dierks T, Linden DE (2002) Tracking the mind’s image in the brain II: transcranial magnetic stimulation reveals parietal asymmetry in visuospatial imagery. Neuron 35:195–204. Samsonovich A, McNaughton BL (1997) Path integration and cognitive mapping in a continuous attractor neural network model. J Neurosci 17:5900–5920. Save E, Guazzelli A, Poucet B (2001) Dissociation of the effects of bilateral lesions of the dorsal hippocampus and parietal cortex on path integration in the rat. Behav Neurosci 115:1212–1223. Save E, Paz-Villagran V, Alexinsky T, Poucet B (2005) Functional interaction between the associative parietal cortex and hippocampal place cell firing in the rat. Eur J Neurosci 21:522–530. Schacter DL, Dodson CS (2001) Misattribution, false recognition and the sins of memory. Philos Trans R Soc Lond B Biol Sci 356:1385–1393.
160
PLACE CELLS AND SPATIAL CONTEXT
Sharp PE (1999) Subicular place cells expand or contract their spatial firing pattern to fit the size of the environment in an open field but not in the presence of barriers: comparison with hippocampal place cells. Behav Neurosci 113:643–662. Shrager Y, Bayley PJ, Bontempi B, Hopkins RO, Squire LR (2007) Spatial memory and the human hippocampus. Proc Natl Acad Sci USA 104:2961–2966. Shrager Y, Gold JJ, Hopkins RO, Squire LR (2006) Intact visual perception in memory-impaired patients with medial temporal lobe lesions. J Neurosci 26:2235–2240. Smith EE, Jonides J (1999) Storage and executive processes in the frontal lobes. Science 283:1657–1661. Snyder LH, Grieve KL, Brotchie P, Andersen RA (1998) Separate body- and world-referenced representations of visual space in parietal cortex. Nature 394:887–891. Stuss DT, Knight RT (2002) Principles of Frontal Lobe Function. New York: Oxford University Press. Taube JS (1998) Head direction cells and the neuropsychological basis for a sense of direction. Prog Neurobiol 55:225–256. Teyler TJ, DiScenna P (1986) The hippocampal memory indexing theory. Behav Neurosci 100:147–154. Trinkler I, King JA, Spiers HJ, Burgess N. (2006) Part or parcel? Contextual binding of events in episodic memory. In: Handbook of Binding and Memory: Perspectives from Cognitive Neuroscience (Zimmer HD, Mecklinger A, Lindenberger U, eds.), pp 53–83. Oxford: Oxford University Press.
Tulving E (2001) Episodic memory and common sense: how far apart? Philos Trans R Soc Lond B Biol Sci 356:1505–1515. Tulving E, Markowitsch HJ (1998) Episodic and declarative memory: role of the hippocampus. Hippocampus 8:198–204. Wallenstein GV, Eichenbaum H, Hasselmo ME (1998) The hippocampus as an associator of discontiguous events. Trends Neurosci 21:317–323. Wallentin M, Roepstorff A, Glover R, Burgess N (2006) Parallel memory systems for talking about location and age in precuneus, caudate and Broca’s region. Neuroimage 32:1850–1864. Wang RF, Simons DJ (1999) Active and passive scene recognition across views. Cognition 70:191–210. Wheeler MA, Stuss DT, Tulving E (1997) Toward a theory of episodic memory: the frontal lobes and autonoetic consciousness. Psychol Bull 121:331–354. Wills TJ, Lever C, Cacucci F, Burgess N, O’Keefe J (2005) Attractor dynamics in the hippocampal representation of the local environment. Science 308:873– 876. Wilson MA, McNaughton BL (1993) Dynamics of the hippocampal ensemble code for space. Science 261: 1055–1058. Zipser D, Andersen RA (1988) A back-propagation programmed network that simulates response properties of a subset of posterior parietal neurons. Nature 331: 679–684.
11 Hippocampal Neuronal Activity and Memory: Should We Still Be Talking about Place Cells? HOWARD EICHENBAUM
The hippocampus is critical for memory, so it seems almost obvious that its neuronal activity should be examined from the perspective of firing properties related to memory performance. Yet, historically, most studies of hippocampal neuronal activity have not focused on this aspect, but instead have been directed at questions about spatial information processing by hippocampal neurons called ‘‘place cells.’’ How did this situation come to be and why is there now such a strong interest in the relationship between place cells and memory? In this chapter I will summarize and extend arguments made in recent reviews that challenge the notion that hippocampal cells are ‘‘place cells,’’ and suggest instead that hippocampal neuronal activity should be examined with regard to memory processing functions of the hippocampus (Eichenbaum et al., 1999; Eichenbaum, 2004). I will begin by reconsidering the case for a selective role of the hippocampus in spatial processing and the phenomenon of place cells, and then outline an alternative hypothesis about hippocampal function in memory and offer a reconciliation of the spatial and memory processing views of hippocampal neuronal activity.
COGNITIVE MAPPING AND PLACE CELLS: THE NOTION OF A DEDICATED SPATIAL PROCESSING FUNCTION OF THE HIPPOCAMPUS One of the most influential hypotheses about hippocampal function is the notion that the hippocampus is
specialized for the representation and use of spatial memories (O’Keefe and Nadel, 1978). According to this view, the hippocampus mediates the development of cognitive maps, or representations of the environment that are allocentric, that is, referenced to the global layout of features in the environment and not dependent on egocentric reference from a particular viewpoint. The main lines of evidence supporting this hypothesis are two-fold: (1) animals and humans with damage to the hippocampus may be severely and selectively impaired in certain spatial-memory tasks; and (2) place cells were discovered—principal neurons in the hippocampus that fire when an animal is in a particular location in its environment, as defined by the spatial configuration of any subset of the salient cues, independent of its direction and ongoing behavior (O’Keefe and Dostrovsky, 1971; O’Keefe, 1979). The allocentric perspective of hippocampal neuronal firing patterns is contrasted with the egocentric reference of other spatial processing systems such as motor areas and the parietal cortex, where cells fire in association with the animal’s egocentric direction or particular movements in space (Burgess et al., 2002). While the cognitive mapping hypothesis was proposed almost 30 years ago and remains largely unchanged in its basic tenet that the hippocampus formulates maps of space (Muller, 1996; Best et al., 2001; McNaughton et al., 2006), several lines of evidence call into question its ability to account for the full pattern of findings on spatial-memory performance in animals and humans with hippocampal damage and on the firing properties of place cells.
161
162
PLACE CELLS AND SPATIAL CONTEXT
Does Hippocampal Damage Result in a Selective Deficit in Allocentric Spatial Representation? Concerning role of the hippocampus in producing spatial memory, there is a large body of evidence indicating that the hippocampus is essential for some forms of spatial-memory performance (see reviews by Corkin, 1984; Nadel, 1994; Eichenbaum and Cohen, 2001; Burgess et al., 2002). However, a requirement for cognitive mapping fails to account for observations on impaired and preserved performance of animals and humans with hippocampal damage, and fails to account for patterns of impaired and preserved nonspatial memory. In spatial tasks, among the common observations of impaired spatial memory are flawed performance on spatial alternation, on tasks of nonmatching to sample, and of ‘‘working memory.’’ However, the same subjects may perform fully normally within the identical spatial environment on spatial discrimination and ‘‘reference memory’’ (Olton et al., 1979). Among these tasks there are clear-cut differences in the demand for remembering recent episodes in the tasks where performance is impaired following hippocampal damage and preserved performance when this demand is absent. But there are no differences in the spatial cues or the demand for processing or remembering the allocentric layout within the same environment where impairment or spared performance is observed. Conversely, animals with hippocampal damage are impaired in memory for recent episodes in a nonspatial version of these tasks (Olton and Feustle, 1981). Strengthening the interpretation that it is not spatial representation per se that is lost following hippocampal damage, a study by Alyan et al. (2000) showed that rats with hippocampal lesions can form representations of a constellation of distant landmarks that compose a spatial layout of the environment and guide spatial memory. In this study, rats were rewarded for shuttling between two food cups, one in the center of a round table and the other at a point in the periphery defined by the arrangement of a set of distant cues placed on the walls of a circular curtain that surrounded the table. In a series of probe tests, rats were first fed at the central food cup, then watched to determine where in the periphery they subsequently ran. When all the distal cues were present, both intact rats and rats with hippocampal lesions appropriately headed toward the peripheral food cup. When the distal cues were removed, both groups headed in random directions. By contrast, when all the cues were rotated in concert, both intact and hippocampusdamaged rats ran to the locus predicted by the distal
cues. These findings indicate that the distal cues were used by both groups to identify the peripheral reward location. When a subset of cues proximal to the peripheral reward locus was removed, both groups continued to chose correctly, indicating that both groups could solve the problem with a reduced set of cues. However, when all but one cue located opposite the peripheral reward locus was removed, neither group could solve the problem, indicating that both groups depended on multiple cues. Finally, when the arrangement of the cues was scrambled, both groups chose randomly, indicating that they depended on the spatial relations among the cues. This pattern of findings shows that, like normal rats, rats with hippocampal damage construct and use a representation of distal spatial cues to guide movements to a site of reward. Consistent with these findings, rats with hippocampal damage can successfully solve the water maze problem under conditions where the heading can be guided by a single trajectory toward a constellation of distal cues (Eichenbaum et al., 1990). In the conventional version of the water maze task, rats are trained to locate an escape platform hidden just under the surface of cloudy water in a large swimming pool, starting each swim from one of four randomly selected cardinal loci in the periphery (Morris et al., 1982). Under these training conditions, rats with hippocampal damage or disconnection of the hippocampus by fornix lesions are severely impaired in learning. However, in this study, rats were gradually trained to locate the escape platform from a single start locus, and under this condition, rats with fornix lesions succeeded. Furthermore, in probe trials where the platform was removed, those rats swam in the vicinity of the former platform, just as intact rats do, showing that the hippocampus is not required to identify the platform locus on the basis of distal spatial cues. However, whereas normal rats could subsequently efficiently locate the platform from any of the novel starting points, rats with fornix lesions could not. Whishaw and colleagues extended these findings, showing that rats with hippocampal lesions can gradually learn to find the platform from all four cardinal starting points (Whitshaw et al., 1995; Whitshaw and Tomie, 1997). However, unlike normal rats, rats with hippocampal lesions learn a new escape locus in the same pool. These studies confirm that rats with hippocampal lesions can acquire a spatial representation of the environment, and can use it to guide a welllearned swim path. However, they fail when required to integrate multiple swim trajectories concurrently, and cannot use the knowledge gained in learning one path to generate novel headings or learn new escape loci.
HIPPOCAMPAL NEURONAL ACTIVITY AND MEMORY
Spatial Memory in Human Amnesic Patients Studies on human amnesic subjects also offer insights into the critical role of the hippocampus in spatial memory. Notably, amnesic patients are impaired in spatial learning, just as they are impaired in a broad range of learning and memory capacities (Corkin, 1984; Cohen and Eichenbaum, 1993). Recent studies have focused on whether amnesic patient are impaired in spatial cognition not secondary to their memory impairment, and whether the use of cognitive maps learned well before hippocampal damage always depends on hippocampal function. Two recent studies strongly indicate that the hippocampus is not required for spatial cognition per se, nor is it required for the storage or use of cognitive maps acquired remotely prior to hippocampal damage. One study focused on a profoundly amnesic subject (known by his initials H.M.) with damage to the hippocampus and surrounding cortex who had moved away from the town in which he had grown up over 60 years prior to becoming amnesic (Teng and Squire, 1999). The patient had virtually no knowledge of his current neighborhood, confirming the inclusion of a spatial-memory impairment in his profile of anterograde amnesia. However, he remembered and could navigate within his childhood neighborhood at least as well as a group of control subjects who had also moved away from the same area many years before. He could describe how to navigate from his home or other places to several locations in the area, could describe alternative routes when the primary route was imagined as blocked, and could imagine himself in several locations and point successfully in the direction of major landmarks. All of these capacities would be deficient if the hippocampus were critical to spatial cognition or permanently stored cognitive maps. Similarly, another patient (K.C.) with more diffuse damage including the hippocampus demonstrated intact spatial memory and navigational capacities for the neighborhood in which he lived for 40 years prior to onset of amnesia (Rosenbaum et al., 2000). Also notable is that, as introduced above, H.M. gradually acquired semantic memory for the layout of a house in which he moved after onset of amnesia. This pattern of results on intact spatial-memory performances in human amnesia conforms to the pattern that characterizes preserved nonspatial memory in amnesia. Thus, for both spatial and nonspatial materials, information acquired remotely prior to the onset of amnesia is intact, and there is a spared capacity for gradual acquisition of new semantic memories. These findings confirm that spatial memory is indeed part of the domain of memory mediated by the hippocampus.
163
But the hippocampus is not critical to spatial cognition per se, and the spatial-memory performance follows the same general pattern of impaired and preserved abilities in nonspatial memory. Parallels with findings on animals are drawn closer by recent studies on the performance of amnesic humans on allocentric and egocentric spatial tasks (Burgess et al., 2002; King et al., 2002). In these studies, normal subjects and patients with limited hippocampal damage were shown to remember from one perspective. Then they were tested on their memory for the loci as seen from the same or a different perspective in the identical environment. The investigators reasoned that recognition from the same, ‘‘egocentric’’ perspective could be supported by visual pattern matching. However, correct identification from a different perspective relies on memory for the location relative to other cues, using an ‘‘allocentric’’ memory. The amnesic subjects were deficient in both tests, but the impairment was more severe on ‘‘allocentric’’ memory. Notably, the severity of the deficit was increased by greater memory demands, produced either by elongating the delay between initial presentation and subsequent test or by increasing the number of loci to be remembered. The authors interpreted the greater deficits in different-view conditions as indicating a selective role for the hippocampus in allocentric memory. However, there is an alternative explanation that emphasizes differences in the memory demands of the two memory tests. The ‘‘egocentric’’ tests can be solved by familiarity with the identical view seen initially. By contrast, the ‘‘allocentric’’ tests require recollection of the previous view and subsequent reconciliation with the test view and generation of a flexible response; this integration could not be accomplished by a simple visual pattern-matching. Thus, the distinction between performances on ‘‘egocentric’’ and ‘‘allocentric’’ memory is confounded by a differential demand for recollective memory in the allocentric condition as compared to the absence of this demand in the egocentric condition.
Do Hippocampal Neurons Compose an Allocentric Spatial Representation? With regard to the phenomenon of place cells, there is only one behavioral protocol in which the firing patterns of hippocampal neurons meet the fundamental criteria for place cells as supporting an allocentric representation independent of viewpoint and ongoing behavior (O’Keefe, 1979). In the protocol of Muller et al. (1987), rats forage in an open field marked by a salient orienting cue for randomly placed food pellets continuously for several minutes. In this situation,
164
PLACE CELLS AND SPATIAL CONTEXT
most hippocampal neurons show spatially specific activity with no other remarkable behavioral correlates. Notably, this situation is unique in that it provides a set of conditions where spatial cues provide virtually the only regularities of the experimental protocol; the spatial cue offers a continuous orientation signal at all times. Conversely, the delivery of rewards and the onset, direction, speed, and punctuation of movements and other behaviors are intentionally randomized in time and space in an effort to distribute their influence among all locations equally. It is only under these specific conditions that true place cells, neurons whose activity reflects the location of the animal in relation to the layout of all the salient cues regardless of the animal’s orientation and behavior, are observed. Notably, even in this situation, most or all spatial firing patterns would fall short of meeting O’Keefe’s (1979) original criterion of place cells being equally under the control of any subset of prominent spatial cues. In one recent study using the open-field protocol, most cells were controlled only by two of the four prominent cues and no cells were influenced by all the cues (O’Keefe and Burgess, 1996). Also, a recent analysis suggests that the capacity of firing rates to predict location varies with the animal’s position with relation to proximal cues; even in this situation, individual stimuli can influence hippocampal firing patterns (Olypher et al., 2002). In addition, whereas one previous study indicated a quantitative control over the position and shape of spatial representations by specific combinations of spatial cues (O’Keefe and Burgess, 1996), another study showed that altering of the overall shape of the environment, usually the major determinant of hippocampal spatial representation, was not reflected immediately in any change in place cell firing patterns (Lever et al., 2002). The combination of these findings indicates that, even within the random foraging protocol, something more than spatial dimensions is incorporated into hippocampal representations. There is also substantial evidence that hippocampal neurons encode a broad variety of information in addition to spatial cues. Even the earliest descriptions of place cells acknowledged that the activity of some hippocampal cells reflects nonspatial information processing. For example, O’Keefe’s (1976) detailed analysis described cells that fired only during particular behaviors including eating, grooming, and exploratory sniffing. These findings were consistent with Ranck’s (1973) description of hippocampal cells that fired during orientating, approach, and consummatory behaviors. Even the identical cells that seem to represent location independent of behavioral patterns in the openfield task fail to do so when the animal performs spatially directed behaviors in the radial-maze task (Muller
et al., 1994). Moreover, there have subsequently been several reports of hippocampal neural activity associated with nonspatial stimuli and behaviors in animals performing various memory tasks (for reviews see Deadwyler and Hampson, 1997; Eichenbaum et al., 1999). Other studies have employed stimuli that appear in multiple locations and have observed hippocampal neural activity associated with these stimuli regardless of their location (e.g., Young et al., 1994; Wood et al., 1999—see above). A compelling body of data is consistent with the view that neural ensembles in the hippocampus encode a broad range of behavioral events as well as the places where they occur.
THE RELATIONAL MEMORY HYPOTHESIS: AN ALTERNATIVE ACCOUNTING OF HIPPOCAMPAL NEURONAL ACTIVITY The results described above and many other findings have led to an alternative view that hippocampal networks form representations of relations among stimuli across the full range of stimuli and actions that constitute everyday experiences (Cohen and Eichenbaum, 1993; Eichenbaum et al., 1994). In recent years, the nature of the critical relations and how they might be combined to serve memory has become considerably clearer. To explain these features of relational memory organization that have emerged from these considerations, I will begin with a story based on one of my own memories. At a scientific meeting a few years ago, I met a student in the hallway and discussed her poster presentation. At the outset of this discussion, I recalled her standing next to her poster in the large convention center floor. My recollection of her presentation unfolded as a ‘‘mental replay’’ of her study extended over time (Tulving, 2002). And that memory did not occur in isolation, but instead also brought to mind memories of other studies that shared related information. This memory provides examples of common properties of declarative memory dependent on the hippocampus. Declarative memories involve memories of items in the context in which they were experienced wherein each discrete event includes the relevant people, their actions, and the place where that event occurred (e.g., the student next to her poster); a flow of events that composes that unique experience (the sequence of components of the student’s study); and an organization of memories into a network that links related memories by common features (e.g., the other experiments that shared items with the student’s study). I have suggested that the combined representation of events as items in their context, episodes as the sequences of events that constitute the flow of informa-
HIPPOCAMPAL NEURONAL ACTIVITY AND MEMORY
tion in memories, and networks generated by the linking of memories into an organized structure constitute the essential features of relational memory (Eichenbaum, 2004). In the following sections, I will consider how well these features of declarative memory are captured by the activity of hippocampal neurons.
Events Are Represented as Items in Context Considerable evidence indicates that the hippocampus encodes associations among stimuli, actions, and places that compose discrete events, that is, items in the context of other stimuli. In studies on humans, several functional imaging studies have shown that the hippocampus is more activated during the encoding or retrieval of associations among many elements of a memory than during that for single or unrelated elements (for review see Cohen et al., 1999). For example, Eldridge et al. (2000) reported that the hippocampus was activated during correct recollections that included perceptual details or associations made with a word during a previous study phase as compared to a sense of familiarity with individual words. Henke et al. (1997) observed greater hippocampal activation when subjects associated a person with a house than when making independent judgments about the person and house. Giovanello et al. (2003) reported greater hippocampal activation when subjects recognized previously presented word pairings than when recognizing rearranged word pairs or words studied separately. Davachi and Wagner (2002) reported that the hippocampus is activated during encoding of multiple items and more activated when subjects are required to link the items to one another by systematic comparisons, as compared to rote rehearsal of individual items. Other recent studies have reported activation within subfields of the hippocampus during the encoding of face–name associations (Zeineh et al., 2003) and along the entire longitudinal extent of the hippocampus when subjects studied name–face pairs that they later remembered with high confidence (Sperling et al., 2003). Also, Small et al. (2001) reported that, whereas viewing faces and names independently activated separate areas within the hippocampus, viewing of faces and names in combination activated a distinct area bridging the separate face and name activations; the latter area was also activated during recall of names when cued by the face. Yet other studies have described activation of the hippocampus during the retrieval of autobiographical experiences. Maguire (2001) reported selective activation in the medial temporal region during retrieval of multiple aspects of autobiographical events but not retrieval of public events. Addis et al. (2004) also
165
found that the level of detail, personal significance, and emotionality each contributed to hippocampal activation of autobiographical memories. The involvement of the hippocampus in processing complex material is not limited to autobiographical details but extends broadly, for example, to recollection of the context of learning in formal tests of memory (e.g., Davachi et al., 2003; Ranganath et al., 2003). Studies on single-neuron activity in animals have also provided considerable evidence that hippocampal neurons represent associations among stimuli. As described above, many studies have shown that hippocampal neurons fire in association with ongoing behavior and the context of events as well as the animal’s location (Eichenbaum et al., 1999). The combination of spatial and nonspatial features of events captured by hippocampal neuronal activity is consistent with the view that the hippocampus encodes many features of events and the places where they occur. In addition, two recent studies highlight the associative coding of events and places by hippocampal neurons. In one study rats were trained on an auditory fear-conditioning task (Moita et al., 2003). Prior to fear conditioning, few hippocampal cells were activated by an auditory stimulus. Following pairings of tone presentations and shocks, many cells fired briskly to the tone when the animal was in a particular place where the cell fired above baseline. Another recent study examined the firing properties of hippocampal neurons in monkeys performing a task in which they rapidly learned new scene–location associations (Wirth et al., 2003). Just as the monkeys acquired a new response to a location in the scene, neurons in the hippocampus changed their firing patterns to become selective to particular scenes. These scene–location associations persisted even long after learning was completed (Yanike et al., 2004). In a study explicitly designed to examine the representation of nonspatial cues presented in multiple places, Wood et al. (1999) trained rats to perform the same memory judgments at many locations in the environment. Animals performed a task in which they had to recognize any of nine olfactory cues placed in any of nine locations. Because the location of the discriminative stimuli was varied systematically, neuronal activity related to the stimuli and behavior could be dissociated from that related to the animal’s location. A large subset of hippocampal neurons fired only in association with the odors, the place where odors were sampled, or the match–nonmatch status of the odors, and the largest fraction of these cells encoded combinations of these cues. In a similar study on humans, Ekstrom et al. (2003) recorded the activity of hippocampal neurons as subjects played a taxi driver game, searching for passengers picked up and dropped off at various locations in a virtual-reality town. Many
166
PLACE CELLS AND SPATIAL CONTEXT
of these cells fired selectively associated with a view, a place, or the current goal, and the largest fraction fired in association with specific combinations of a place and the view of a particular scene or under a particular goal. In addition, other investigators have reported that hippocampal cells represent specific salient objects in the context of a particular environment in studies of rats engaged in foraging (Gothard et al., 1996; Rivard et al., 2004) and escape behavior (Hollup et al., 2001) in open fields. Thus, in rats, monkey, and humans, a prevalent property of hippocampal firing patterns involves the representation of unique associations of stimuli, their significance, specific behaviors, and the places where these events occur.
Episodes Are Represented as Sequences of Events Hippocampal neurons become activated during virtually every moment of task performance, including during simple behaviors such as foraging for food (e.g., Muller, 1996) as well as learning related behaviors directed at relevant stimuli that have to be remembered (e.g., Hampson et al., 1993). This general pattern is also observed in a broad range of learning protocols, from studies that involve classical conditioning, discrimination learning, and nonmatching or matching to sample tasks to tests and a variety of maze tasks (for reviews see Eichenbaum et al., 1999, Eichenbuam, 2004). In each of these paradigms, animals are repeatedly presented with specific stimuli and reinforcers and execute appropriate cognitive judgments and conditioned behaviors. Corresponding to each of these regular events, many hippocampal cells show time-locked activations associated with each sequential event. Also, as described above, many of these cells show striking specificities corresponding to particular combinations of stimuli, behaviors, and the spatial location of the event. Thus, within the overall network, cellular activity can be characterized as a sequence of firings representing the step-by-step events in each behavioral episode. Additionally, several recent studies have shown that the processing of previous spatial experiences as the sequential activation of places may continue off-line for a substantial period (Nadasdy et al., 1999; Louie and Wilson, 2001; Lee and Wilson, 2002). Other recent studies have shown that these sequential codings can be envisioned to represent a series of events and their places that compose a meaningful episode, and the information contained in these representations both distinguishes and links related episodes. These studies focus on animals traversing routes through a maze that overlap in multiple places. An examination of firing patterns as animals run
through the overlapping segments of these routes provides a direct comparison of whether neural activity is governed solely by spatial cues and movement trajectory, or instead is influenced by recent past and near future events that constitute the ongoing episode. Recent studies on the spatial firing patterns of hippocampal neurons provide compelling data consistent with the characterization of hippocampal neurons as representing spatially extended episodes. In one study, rats were trained on a spatial alternation task in a modified T-maze (Wood et al., 2000; see also Frank et al., 2000; Bower et al., 2005). Performance on this task requires that the animal distinguish left-turn and right-turn episodes and that it remember the immediately preceding episode to guide the choice on the current trial; in that way, the task is similar in demands to those of episodic memory. If hippocampal neurons encode each sequential behavioral event and its locus within one type of episode, then most cells should fire only when the rat is performing within either the left-turn or the right-turn type of episode. This should be particularly evident when the rat is on the ‘‘stem’’ of the maze, when the rat traverses the same locations on both types of trials. Indeed, a large proportion of cells that fired when the rat was on the maze stem fired differentially on left-turn versus right-turn trials. The majority of cells showed strong selectivity, some firing at over 10 times the rate on one trial type, suggesting they were part of the representations of only one type of episode. Other cells fired substantially on both trial types, potentially providing a link between left-turn and right-turn representations by the common places traversed on both trial types. Ferbinteanu and Shapiro (2003) also reported that many hippocampal neurons fire in association with serial locations occupied as rats traverse different routes within the same environment. Furthermore, they modified the task to distinguish whether the activity of these cells reflected the immediate past experience of the animal (retrospective coding) or predicted its future path (prospective coding). They found that these cells encode both past events and future goals of each route, with some cells encoding both kinds of information on the same trials. Furthermore, retrospective and prospective coding diminished on error trials, and some cells fired in association with the intention to proceed to a particular location even when a detour was required. These findings indicate that the overall hippocampal representations are neither retrospective or prospective per se, and do not necessarily capture the precise details of behaviors or places that distinguish qualitatively similar episodes. Rather, the hippocampal network encodes routes through space as a meaningful sequence of events that characterize a particular spatially extended experience.
HIPPOCAMPAL NEURONAL ACTIVITY AND MEMORY
Another recent study (Lee et al., 2006) provided even more striking evidence that the representation of sequential events predominates hippocampal neuronal activity in animals performing the T-maze alternation task. In this study, animals performed a large number of repetitive trials on the task, raising the demand to disambiguate the most recent episode from many preceding experiences. Under this condition, the spatial correlates of CA1 neurons that distinguished leftand right-turn trials gradually changed from their original firing locations, shifting toward prospective goal locations as animals repetitively performed many trials. The relative locations of simultaneously recorded firing fields, however, were preserved within the ensemble spatial representation during this shifting, indicating that the sequential ordering of events for each trial type was consistent. The striking withinsession shifts in preferred firing locations in the absence of any changes in the environment suggested a mechanism by which hippocampal neuronal ensembles distinguish individual episodes in this task.
Networks Are Composed by Linking Related Memories The networking of memories involves a linking of related experiences and the consequent ability to make generalize across items or events that are only indirectly related. A network of memories has two potentially quite useful properties. First, memories for the common features become ‘‘timeless’’ semantic elements, not bound to any particular episode in which they were acquired. Second, the elements that encode common features link memories to one another, allowing one to compare and contrast memories and to make inferences among indirectly related events. These properties of relational memory representation underlie the hallmark ‘‘flexibility’’ of declarative memory expression (Cohen, 1984). Also, the continued abstraction of common features and linking of memories may underlie a more extensive association and interleaving of memories during the prolonged period of consolidation (McClelland et al., 1995). Experimental analyses on the role of the hippocampus in memory networking in animals and humans have focused directly on the learning of multiple related problems and their integration into networks of memory that support flexible, inferential judgments. One analysis of this capacity in animals compared the ability of normal rats and rats with selective damage to the hippocampus on their ability to learn a set of odor problems and to link the representations of these problems in support of novel inferential judgments (Bunsey and Eichenbaum, 1996). Animals were initially trained on two sets of overlapping odor paired
167
associates (e.g., A goes with B, B goes with C). Then the rats were given probe tests to determine if they could infer the relationships between items that were only indirectly associated through the common elements (A goes with C?). Normal rats learned the paired associates and showed strong transitivity in the probe tests. Rats with selective hippocampal lesions also learned the pairs over several trials but were severely impaired in the probes, showing no evidence of transitivity. In another experiment, rats learned a hierarchical series of overlapping odor choice judgments (e.g., A > B, B > C, C > D, D > E), then were probed on the relationship between indirectly related items (B > D?). Normal rats learned the series and showed robust transitive inference on the probe tests. Rats with hippocampal damage also learned each of the initial premises but failed to show transitivity (Dusek and Eichenbaum, 1997). The combined findings from these studies show that rats with hippocampal damage can learn even complex associations, such as those embodied in the odor paired-associates and conditional discriminations. But, without a hippocampus, they do not interleave the distinct experiences according to their overlapping elements to form a relational network that supports inferential and flexible expression of their memories. Complementary evidence on hippocampal neuronal activation associated with the networking of memories comes from two recent studies indicating that the hippocampus is involved when humans make inferential memory judgments. In one study, subjects initially learned to associate each of two faces with a house and, separately, learned to associate pairs of faces (Preston et al., 2004). Then, during brain scanning, the subjects were tested on their ability to judge whether two faces who were each associated with the same house were therefore indirectly associated with each other, and on whether they could remember trained face pairs. The hippocampus was selectively activated during performance of the inferential judgment about indirectly related faces compared to that during memory for trained face–house or face–face pairings. In the other study, subjects learned a series of choice judgments between pairs of visual patterns that contained overlapping elements, just as in the studies on rats and monkeys, and as a control they also learned a set of non-overlapping choice judgments (Heckers et al., 2004). The hippocampus was selectively activated during transitive judgments as compared to novel nontransitive judgments. Although under some circumstances it may be possible to associate indirectly related items without a relational network (O’Reilly and Rudy, 2001; Van Elzakker et al., 2003), the findings on hippocampal involvement in transitive
168
PLACE CELLS AND SPATIAL CONTEXT
inference indicate that the hippocampus plays a critical role in linking related memories according to their common features, and this linkage results in a network that can support inferences between items in memory that are only indirectly related. Such performance suggests how a general networking of memories can underlie the flexibility of declarative memory expression. By what mechanism does the hippocampus link memories? The simple key to such a linkage is to devote some of the neuronal population to representing common elements across related memories. In virtually all the studies described above, some hippocampal neurons indeed encoded features that are common among different experiences. For example, in the Wood et al. (1999) study on odor recognition memory, some cells showed striking associative coding of odors, their match–nonmatch status, and places, whereas other cells fired in association with one of those features across different trials. Some cells fired during a particular phase of the approach toward any stimulus cup. Others fired differentially as the rat sampled a particular odor, regardless of its location or match– nonmatch status. Other cells fired only when the rat sampled the odor at a particular place, regardless of the odor or its status. Yet other cells fired differentially associated with the match and nonmatch status of the odor, regardless of the odor or where it was sampled. Similarly, in Ekstrom and colleagues’ (2003) study on humans performing a virtual navigation task, some hippocampal neurons fired in associaton with combinations of views, goals, and places, whereas other cells fired when subjects viewed particular scenes, occupied particular locations, or had particular goals in finding passengers or locations for drop off. Also, in Moita and colleagues’ (2003) study of auditory fear conditioning, whereas some cells only fired to a tone when the animal was in a particular place, others fired in association with the tone wherever it was presented across trials. And in Rivard and colleagues ‘(2004) study of rats exploring objects in open fields, some cells fired selectively in association with an object in one environment, whereas others fired in association with the same object across environments. The notion that these cells might reflect the linking of important features across experiences and the abstraction of common information has also been highlighted in recent studies on monkeys and humans. Hampson et al. (2004) trained monkeys on matching to sample problems, then probed the nature of the representation of stimuli by recording from hippocampal cells when the animals were shown novel stimuli that shared features with the trained cues. They found many hippocampal neurons that encoded meaningful categories of stimulus features and appeared to employ these representations to recognize the same features
across many situations. Similarly, Kreiman et al. (2000a) characterized hippocampal firing patterns in humans during presentations of a variety of visual stimuli. They reported a substantial number of hippocampal neurons that fired when the subject viewed specific categories of material, e.g., faces, famous people, animals, scenes, and houses, across many exemplars of each. A subsequent study showed that these neurons are activated when a subject simply imagines its optimal stimulus, supporting a role for hippocampal networks in recollection of specific memories (Krieman et al., 2000b). Qurioga et al. (2005) described hippocampal neurons in humans that fired selectively when subjects viewed a particular person or familiar object in a large number of perspectives and variations. These neurons seemed to capture the ‘‘invariant’’ features of a particular object or person across a broad variety of experiences. The combination of findings across species indicating that hippocampal neurons capture common elements across related memories provides compelling evidence for the notion that some hippocampal cells represent common features among the various episodes that could serve to link memories obtained in separate experiences.
An Alternative Characterization of the Spatial Firing Patterns of Hippocampal Neurons The characterization of hippocampal activity in relational representation can be extended to account for the findings on place cells. Consider, for example, observations on the ‘‘directionality’’ of place cells in rats. When rats forage randomly for food in an open field, place cells are characterized as firing similarly when rats move in any direction. However, in the radial-maze task, animals regularly run outward on each maze arm to obtain a reward, and then return to the central platform to initiate another arm choice (Olton et al., 1979). Here outward and inward arm movements are associated with meaningfully distinct behavioral episodes. Correspondingly, hippocampal neuronal activity reflects the relevant ‘‘directional structure’’ imposed by this protocol, and almost all place cells fire primarily during outward or inward journeys (McNaughton et al., 1983; Muller et al., 1994). Similarly, place cells are activated selectively during distinct approach or return episodes and from variable goal and start locations in both open fields and on linear tracks (Wiener et al., 1989; Gothard et al., 1996). Markus et al. (1995) directly compared the directionality of place cells under different task demands within the identical open-field environment, and found that place cells that were nondirectional when rats foraged randomly were strongly directional when they
HIPPOCAMPAL NEURONAL ACTIVITY AND MEMORY
systematically visited specific reward locations. Taken together, these findings emphasize that place cells exhibit movement-related firing patterns whenever the animal’s movements are associated with meaningfully different events. Often the ‘‘directional’’ selectivity of place cells is interpreted as representation of a second-order, spatial-vector parameter. However, a more detailed analysis of firing patterns of hippocampal neurons during spatial-memory performance suggests an alternative account. One study involved a variant of the radial-maze task where rats began each trial at the center of the open field, and then approached any of the four corners to obtain a reward (Wiener et al., 1989). Subsequently, they had to return to the center and then approach a different corner to obtain another reward, and so on, remembering where they had previously received rewards and not repeating visits to those locations. A large number of hippocampal neurons had location-specific firing patterns, like place cells. However, the majority of these cells fired differentially depending on whether the animal was running outward toward a corner or inward returning to the center. In addition, most cells were also ‘‘tuned’’ for a particular speed of movement, and fired differentially depending on the angle of turn the rat made on the approach or return path. Thus, one could describe each neuron as a ‘‘place cell,’’ where spatial firing was conditional on the animal moving at a particular speed, direction, and turning angle. Alternatively, the firing patterns of these cells could be characterized as a sequence of activations representing discrete sequential behavioral events, and the location where they occur, within each type of episode as defined by the task. The recent studies on the spatial firing patterns of hippocampal neurons in rats performing the T-maze alternation task provide compelling data consistent with this characterization (Wood et al., 2000). Performance on this task requires that the animal distinguish left-turn and right-turn episodes and that it remember the immediately preceding episode to guide the choice on the current trial. Thus, Olton (1984, 1986) considered this class of task similar in demands to that of episodic memory. Unlike radial-maze tasks, in the T-maze, rats run in the same direction on different types of trial episodes, yet many of the cells demonstrated a strong selectivity that distinguished the different types of episodes. Thus, hippocampal neurons can disambiguate journeys on a maze even when the animal is running in the same direction, when the episodes are distinctly meaningful. Additional evidence that the hippocampus forms distinct and linked representations for related spatial experiences comes from studies of rats exploring openfield environments. Rats perform spatial-memory tasks
169
or forage randomly for rewards, and the spatial firing patterns of hippocampal cells are compared between variations of the same environmental cues (for a review see Eichenbaum et al., 1999). In one study, rats performed a working memory task on a radial maze that contained a combination of distal cues off the maze arms and local cues on the surfaces of the maze arms (Shapiro et al., 1997; Tanila et al., 1997). The rats were repetitively exposed to a standard configuration of distal and local cues, and a variant where the distal and local cues were rotated 908 in opposite directions. Over many presentations of both conditions, the spatial firing patterns of some of the cells were consistent and those of others changed. Some cells acquired distinct spatial representations across the two conditions, whereas the other cells maintained the same spatial firing pattern associated with the distal cues, with the local cues, or with the fixed cues in the environment that were not rotated. Thus, part of the hippocampal network captured the unique conjunctions of cues within each distinct condition, whereas others captured regularities in the stimuli that were common across both conditions. In another study rats initially foraged randomly in a rectangular open field, then were allowed to walk through a door into another identical environment (Skaggs and McNaughton, 1998). Upon crossing into the second environment, the rats were presented with a conflict between the identical spatial cues in the two environments and the presumed self-knowledge that they walked between two different environments. Consistent with the view that the environments were represented as distinct and linked experiences, approximately half of the hippocampal cells had the same firing patterns in the two environments and half had distinct firing patterns. These findings support the notion that one subset of hippocampal neurons encodes conjunctions of common and distinguishing features that disambiguate the two spatial experiences, and another subset encodes the common features that link the two experiences.
Reconciling Cognitive Mapping and Relational Memory Views of Hippocampal Function A major dilemma for the cognitive-mapping view is that cognitive maps are a particularly good example of semantic memory structures, in that they involve logically organized representations of spatial relations, independent of particular experiences of travel through the environment. Yet it is generally agreed that the hippocampus is critical for episodic memory. Burgess et al. (2002) attempted to reconcile the roles of the hippocampus in cognitive mapping and episodic
170
PLACE CELLS AND SPATIAL CONTEXT
memory, suggesting that temporal ‘‘stamps’’ are added to spatial representations in order to date spatially organized memories. The findings described above on the pattern of impaired and spared spatial-memory performances and on the nature of hippocampal neuronal firing patterns in rats performing spatial tasks suggest an alternative conception of how the hippocampus mediates cognitive mapping. Studies of human spatial cognition have revealed that there are multiple forms of representation for large-scale spatial environments, and these are differentially emphasized in numerous theories about the nature of cognitive maps (for review see Kitchin and Blades, 2002). Among these forms of spatial representation, the most prominent ones are ‘‘route’’ and ‘‘survey’’ knowledge (McNamara, 2002). Survey knowledge refers to the construction and use of a maplike representation of the overall layout of the environment. This strategy predominates when people have the aid of a map or if the goal is to create a maplike representation. Alternatively, when a map is not available and the goal is to learn a particular path, the initial knowledge is usually constituted as a representation of a route. With sufficient experience on routes within the same environment, people may learn global spatial dimensions, including distances and directions, and may subsequently construct a map-like representation. Regardless of whether a survey representation exists, it is easier for people to retrieve information from some perspectives than from others, which suggests that memory of space from egocentric views is the primary means of representation, and the ability to make judgments from novel views involves manipulations and integration of those representations (Wang and Spelke, 2000; McNamara, 2002). These considerations suggest a framework for reconciliation of the cognitive-mapping and relationalmemory hypotheses. Imagine that you have arrived at a hotel in a city not previously visited and for which you have not obtained a map. As you go out for dinner the first evening, you travel along a route, taking in several views and making several turns along the way to the restaurant. At this point, it is very unlikely that you have a map-like representation of the environment, but it is likely that you can recall much of the route you had followed. The next day you go out again, to a meeting in a different direction. A similar routerepresentation is formed for that journey. However, imagine that today’s journey intersects with the route followed yesterday. You may recognize the scene at the locus of intersection of the two journeys and recall the routes back to the hotel and to the restaurant along the previous journey. As you make additional journeys through the city, these also intersect, allowing
you to construct a large-scale representation in which route-representations are linked by their intersections. Eventually, as more and more of the journeys share common locations and the particular routes are less important to remember, you develop a representation of the common loci and their spatial relations independent of particular routes—in other words, a maplike, ‘‘semantic’’ representation of the environment organized by the spatial regularities. At this point, you can probably draw a reasonable map of those parts of the city you have visited, and can navigate along new routes deduced from the map-like representation. This scheme suggests a common role for the hippocampus in spatial and nonspatial memory, consistent with the relational-memory hypothesis. Just as in nonspatial learning, in spatial learning the hippocampus initially forms a representation of each spatial episode. In this case, a spatial episode consists of a series of views from places and your own movements through space, as well as any events witnessed along the way, that constitute a route for each journey. As multiple routes intersect, the common places are represented by some of the same hippocampal neurons, such that the route-representations become linked, just as in the examples of learning overlapping nonspatial problems described above. Eventually, as more intersections are shared among journeys, some of those neurons represent only the common places, and those cells constitute true ‘‘place cells.’’ In addition, when one arrives at a common location, the hippocampus may be able to generate many of the routes that included that location, mediating a capacity for navigation in novel directions. The evidence on hippocampal neuronal firing patterns described above supports this accounting of spatial representations. Thus, in the Wood et al. (2000) study, separate subpopulations of hippocampal neurons can be described as representing the series of successive places that compose each type of episode, and some cells of each population represent the common places that link the different types of episodes. In addition, recent evidence from functional brain imaging suggests that the human hippocampus is also specifically involved in the representation of routes. Several functional imaging studies have now been performed on humans performing spatial tasks, and these reveal a common network of cortical areas activated during different aspects of spatial performance (Flitman et al., 1997; Mellet et al., 2000). Furthermore, when humans simply view complex spatial scenes, both the parahippocampal region and parts of the hippocampus can be activated (Epstein and Kanwisher, 1998; Kohler et al., 2002). Other studies have directly investigated hippocampal activation during
HIPPOCAMPAL NEURONAL ACTIVITY AND MEMORY
recall or encoding of large-scale environments. In one study humans initially explored an imaginary town using virtual-reality technology (Maguire et al., 1997). Strong activation was observed in the right hippocampus during successful recall of routes as compared to following arrows through a route or unsuccessful navigational attempts. Similarly, in London taxi drivers, the hippocampus is strongly activated when they recall specific routes through the town (Maguire et al., 1998). These studies demonstrate hippocampal activation during route retrieval, but they do not distinguish whether the activations reflect recollection of specific routes or generation of a route from a map-like representation. Indeed, after initially studying either routes or maps, the hippocampus is activated during subsequent navigation of an environment (Mellet et al., 2000). A recent study that examined hippocampal activation specifically during the encoding of a novel environment speaks more directly to the nature of initial spatial representation (Shelton and Gabrieli, 2002). In this study, subjects were scanned as they explored a large-scale virtual-reality environment in one of two ways. Under one condition, they explored the environment from a ‘‘bird’s eye’’ view, encouraging a survey representation. Under the other condition, they explored the environment from a ‘‘ground’’ perspective by entering through a doorway and traversing corridors and rooms, encouraging the representation of routes. As in previous imaging studies, a common network of cortical areas was activated in both conditions, but there were differences as well. In particular, survey encoding recruited greater activation in the inferior temporal and posterior parietal cortex, suggesting that survey representations are mainly encoded as complex visual scenes. By contrast, route encoding recruited regions not activated by survey encoding, including other cortical areas, the parahippocampal region, and the hippocampus. These findings suggest a conception that contrasts with the view of Burgess et al. (2002), who have proposed that the hippocampus encodes episodic memory by adding a temporal dimension to a fundamentally map-like survey representation. Instead, the Shelton and Gabrieli findings support the view that the hippocampus initially encodes a large-scale space as representations of routes. These episodic representations can then be recalled and used to develop survey representations and in the navigation of well-known environments. The strong consistency between data on route encoding from studies on hippocampal neuronal firing patterns in rats and hippocampal activation in human functional imaging supports the notion that hippocampal representations of space, like those for non-
171
spatial memory, are fundamentally organized as sequences of events and places where they occur.
CONCLUSIONS: THE ROLE OF ‘‘PLACE CELLS’’ IN MEMORY This review of the place cell phenomenon confirms the representation of location by hippocampal neurons, but at the same time suggests that the role of hippocampal neuronal activity in memory is found by also considering the broad variety of nonspatial features of events and behavioral actions encoded by these neurons. Furthermore, the present analysis suggests that a reconciliation of the cognitive-mapping and relational-memory hypotheses can be reached. By this view, the three central features of relational-memory representation incorporate aspects of spatial processing highlighted in the cognitive-mapping account. First, events are encoded as items in the context in which they occur. Thus, the prominent representation of location by hippocampal place cells reflects the encoding of places where important events occur. Second, episodes are encoded as sequences of events. Thus, the representation of many types of memories, including spatially extended routes, by hippocampal place cells reflects the sequential series of events that compose a unique, spatially extended experience. Third, relational networks involve the linking of multiple episodes and the capacity to make inferential judgments across related experiences. Thus, the representation of many features of episodes, including locations common across multiple journeys, by hippocampal place cells reflects the linking of routes into a network (a map) that can guide navigation across novel routes. According to this view, the cognitivemapping and relational-memory accounts converge and the study of ‘‘place cells’’ will continue to be enormously useful in characterizing a more truly general role of hippocampal neuronal activity in relational memory.
References Addis DR, Moscovitch M, Crawley AP, McAndrews MP (2004) Recollective qualities modulate hippocampal activation during autobiographical memory retrieval. Hippocampus 146:752–762. Alyan SH, Jander R, Best PJ (2000) Hippocampectomized rats can use a constellation of landmarks to recognize a place. Brain Res 876:225–237. Best PJ, White AW, Minai A (2001) Spatial processing in the brain: the activity of hippocampal place cells. Annu Rev Neurosci 24:459–486.
172
PLACE CELLS AND SPATIAL CONTEXT
Bower MR, Euston DR, McNaughton BL (2005) Sequential-context dependent hippocampal activity is not necessary to learn sequences with repeated elements. J Neurosci 15:1313–1323. Bunsey M, Eichenbaum H (1996) Conservation of hippocampal memory function in rats and humans. Nature 379:255–257. Burgess N, Maguire EA, O’Keefe J (2002) The human hippocampus and spatial and episodic memory. Neuron 35:625–641. Cohen NJ (1984) Preserved learning capacity in amnesia: evidence for multiple memory systems. In: The Neuropsychology of Memory (Butters N, Squire LR, eds.), pp 83–103. New York: Guilford Press. Cohen NJ, Eichenbaum H (1993) Memory, Amnesia, and the Hippocampal System. Cambridge, MA: MIT Press. Cohen NJ, Ryan J, Hunt C, Romine L, Wszalek T, Nash C (1999) Hippocampal system and declarative relational memory: summarizing the data from functional neuroimaging studies. Hippocampus 9:83–98. Corkin S (1984) Lasting consequences of bilateral medial temporal lobectomy: clinical course and experimental findings in H.M. Semin Neurol 4:249–259. Davachi L, Mitchell JP, Wagner AD (2003) Multiple routes to memory: distinct medial temporal lobe processes build item and source memories. Proc Natl Acad Sci USA 100:2157–2162. Davachi L, Wagner AG (2002) Hippocampal contributions to episodic encoding: insights from relational and item-based learning. J Neurophysiol 88:982– 990. Deadwyler SA, Hampson RE (1997) The significance of neural ensemble codes during behavior and cognition. Annu Rev Neurosci 20:217–244. Dusek J, Eichenbaum H (1997) The hippocampus and memory for orderly stimulus relations. Proc Natl Acad Sci USA 94:7109–14. Eichenbaum H (2004) Hippocampus: cognitive processes and neural representations that underlie declarative memory. Neuron 44:109–20. Eichenbaum H, Cohen NJ (2001) From Conditioning to Conscious Recollection: Memory Systems of the Brain. New York: Oxford University Press. Eichenbaum H, Dudchencko P, Wood E, Shapiro M, Tanila H (1999) The hippocampus, memory, and place cells: is it spatial memory or a memory space? Neuron 23:209–226. Eichenbaum H, Otto T, Cohen NJ (1994) Two functional components of the hippocampal memory system. Brain Behav Sci 17:449–518. Eichenbaum H, Stewart C, Morris RGM (1990) Hippocampal representation in spatial learning. J Neurosci 10:331–339. Ekstrom AD, Kahana MJ, Caplan JB, Fields TA, Isham EA, Newman EL, Fried I (2003) Cellular networks
underlying human spatial navigation. Nature 425:184– 187. Eldridge LL, Knowlton BJ, Furmanski CS, Brookheimer SY, Engel SA (2000) Remembering episodes: a selective role for the hippocampus during retrieval. Nat Neurosci 3:1149–1152. Epstein R, Kanwisher N (1998) A cortical representation of the local visual environment. Nature 392:598– 601. Ferbinteanu J, and Shapiro ML (2003) Prospective and retrospective memory coding in the hippocampus. Neuron 40:1227–1239. Flitman S, O’Grady J, Cooper V, Grafman J (1997) PET imaging of maze processing. Neuropsychologia 35: 409–420. Frank LM, Brown EN, Wilson M (2000) Trajectory encoding in the hippocampus and entorhinal cortex. Neuron 27:169–178. Giovanello KS, Verfaellie M, Keane MM (2003) Disproportionate deficit in associative recognition relative to item recognition in global amnesia. Cogn Affect Behav Neurosci 3:186–194. Gothard KM, Skaggs WE, Moore KM, McNaughton BL (1996) Binding of hippocampal CA1 neural activity to multiple reference frames in a landmark-based navigation task. J Neurosci 16:823–835. Hampson RE, Heyser CJ, Deadwyler SA (1993) Hippocampal cell firing correlates of delayed-match-to-sample performance in the rat. Behav Neurosci 107:715–739. Hampson RE, Pons TP, Stanford TR, Deadwyler SA (2004) Categorization in the monkey hippocampus: a possible mechanism for encoding information into memory. Proc Natl Acad Sci USA 101:3184– 3189. Heckers S, Zalezak M, Weiss AP, Ditman T, Titone D (2004) Hippocampal activation during transitive inference in humans. Hippocampus 14:153–162. Henke K, Buck A, Weber B, Wieser HG (1997) Human hippocampus establishes associations in memory. Hippocampus 7:249–256. Hollup SA, Molden S, Donnett JG, Moser MB, Moser EI (2001) Accumulation of hippocampal place fields at the goal location in an annular watermaze task. J Neurosci 21:1635–1644. King JA, Burgess N, Hartley T, Vharga-Khadem F, O’Keefe J (2002) Human hippocampus and viewpoint dependence in spatial memory. Hippocampus 12:811– 820. Kitchin R, Blades M (2002) The Cognition of Geographic Space. London: I.B. Tauris. Kohler S, Crane J, Milner B (2002) Differential contributions of the parahippocampal place area and anterior hippocampus to human memory for scenes. Hippocampus 12:718–723.
HIPPOCAMPAL NEURONAL ACTIVITY AND MEMORY
Kreiman K, Kock C, Fried I (2000a) Category-specific visual responses of single neurons in the human medial temporal lobe. Nat Neurosci 3:946–953. Kreiman K, Kock C, Fried I (2000b) Imagery neurons in the human brain. Nature 408:357–361. Lee AK, Wilson MA (2002) Memory of sequential experience in the hippocampus during slow wave sleep. Neuron 36:1183–1194. Lee I, Griffin AL, Zilli EA, Eichenbaum H, Hasselmo M (2006) Gradual translocation of spatial correlates of neuronal firing in the hippocampus toward prospective reward locations. Neuron 51:539–550. Lever C, Wills T, Cacucci F, Burgess N, O’Keefe J (2002) Long-term plasticity in hippocampal placecell representation of environmental geometry. Nature 416:90–94. Louie K, Wilson MA (2001) Temporally structured replay of awake hippocampal ensemble activity during rapid eye movement sleep. Neuron 29:145–156. Maguire EA (2001) Neuroimaging studies of autobiographical events memory. Philos Trans R Soc Lond B 356:1441–1452. Maguire EA, Burgess N, Donnett JG, Frackowiak RSJ, Frith CD, O’Keefe J (1998) Knowing where and getting there: a human navigational network. Science 280: 921–924. Maguire EA, Frackowiak RSJ, Frith CD (1997) Recalling routes around London: activation of the right hippocampus in taxi drivers. J Neurosci 17:7103–7110. Markus EJ, Qin Y-L, Leonard B, Skaggs WE, McNaughton BL, Barnes CA (1995) Interactions between location and task affect the spatial and directional firing of hippocampal neurons. J Neurosci 15: 7079–7094. McClelland JL, McNaughton BL, O’Reilly RC (1995) Why are there complementary learning systems in the hippocampus and neocortex: insights from the successes and failures of connectionist models of learning and memory. Psychol Rev 102:419–457. McNamara TP (2002) Spatial memory. In: Learning and Memory, second edition (Byrne JR, ed.) pp 633–636. New York: Macmillan. McNaughton BL, Barnes CA, O’Keefe J (1983) The contributions of position, direction, and velocity to single unit activity in the hippocampus of freely moving rats. Exp Brain Res 52:41–49. McNaughton BL, Battaglia FP, Jensen O, Moser EI, Moser MB (2006) Path integration and the neural basis of the ‘cognitive map.’ Nat Rev Neurosci 7:663– 678. Mellet E, Briscogne S, Tzourio-Mazoyer N, Ghae¨m O, Petit L, Zago L, Etard O, Berthoz A, Mazoyer B, Denis M (2000) Neural correlates of topographic mental exploration: the impact of route versus survey perspective learning. Neuroimage 12:588–600.
173
Moita MAP, Moisis S, Zhou Y, LeDoux JE, Blair HT (2003) Hippocampal place cells acquire location specific location specific responses to the conditioned stimulus during auditory fear conditioning. Neuron 37:485–497. Morris RGM, Garrud P, Rawlins JP, O’Keefe J (1982) Place navigation impaired in rats with hippocampal lesions. Nature 297:681–683. Muller RU (1996) A quarter of a century of place cells. Neuron 17:813–822. Muller RU, Kubie JL, Ranck JB Jr (1987) Spatial firing patterns of hippocampal complex spike cells in a fixed environment. J Neurosci 7:1935–1950. Muller RU, Bostock E, Taube JS, Kubie JL (1994) On the directional firing properties of hippocampal place cells. J Neurosci 14:7235–7251. Nadel L (1994) Multiple memory systems: what and why, an update. In: Memory Systems (Schacter DL, Tulving E, eds.) pp 39–63. Cambridge, MA: MIT Press. Nadasdy Z, Hirase H, Czurko A, Csicsvari J, Buzsaki G (1999) Replay and time compression of recurring spike sequences in the hippocampus. J Neurosci 19: 9497–9507. O’Keefe JA (1976) Place units in the hippocampus of the freely moving rat. Exp Neurol 51:78–109. O’Keefe JA (1979) A review of hippocampal place cells. Prog Neurobiol 13:419–439. O’Keefe J, Burgess N (1996) Geometric determinants of the place fields of hippocampal neurons. Nature 381: 425–28. O’Keefe J, Dostrovsky J (1971) The hippocampus as a spatial map. Preliminary evidence from unit activity in the freely moving rat. Brain Res 34:171–175. O’Keefe JA, Nadel L (1978) The Hippocampus as a Cognitive Map. New York: Oxford University Press. O’Reilly RC, Rudy JW (2001) Conjunctive representations in learning and memory: principles of cortical and hippocampal function. Psychol Rev 108:311–345. Olton DS (1984) Comparative analyses of episodic memory. Behav Brain Sci 7:250–251. Olton DS (1986) Hippocampal function and memory for temporal context. In: The Hippocampus, Vol. 4 (Isaacson RL, Pribram KK, eds.). New York: Plenum Press. Olton DS, Becker JT, Handlemann GE (1979) Hippocampus, space, and memory. Behav Brain Sci 2:313–365. Olton DS, Feustle WA (1981) Hippocampal function required for nonspatial working memory. Exp Brain Res 68:1–27. Olypher AV, Lansky P, Fenton AA (2002) On the location-specific positional and extra-positional information in the discharge of rat hippocampal cells. Biosystems 67:167–175. Quiroga RQ, Reddy L, Kreiman G, Koch C, Fried I (2005) Invariant visual representation by single neurons in the human brain. Nature 23(7045):1102–1107.
174
PLACE CELLS AND SPATIAL CONTEXT
Preston A, Shrager Y, Dudukovic NM, Gabrieli JDE (2004) Hippocampal contribution to the novel use of relational information in declarative memory. Hippocampus 14:148–152. Ranck JB Jr (1973) Studies on single neurons in dorsal hippocampal formation and septum in unrestrained rats. Part I. Behavioral correlates and firing repertoires. Exp Neurol 41:461–531. Ranganath C, Yonelinas AP, Cohen, MX, Dy CJ, Tom SM, D’Esposito MD (2003) Dissociable correlates of recollection and familiarity with the medial temporal lobes. Neuropsychologia 42:2–13. Rivard B, Li Y, Lenck-Santini P-P, Poucet B, Muller RU (2004) Representation of objects in space by two classes of hippocampal pyramidal cells. J Gen Physiol 124:9–25. Rosenbaum RS, Priselac S, Kohler S, Black SE, Gao F, Nadel N, Moscovitch M (2000) Remote spatial memory in an amnesic person with extensive bilateral hippocampal lesions. Nat Neurosci 3:1044–1048. Shapiro ML, Tanila H, Eichenbaum H (1997) Cues that hippocampal place cells encode: dynamic and hierarchical representation of local and distal stimuli. Hippocampus 7:624–642. Shelton AL, Gabrieli JDE (2002) Neural correlates of encoding space from route and survey perspectives. J Neurosci 22:2711–2717. Skaggs WE, McNaughton BL (1998) Spatial firing properties of hippocampal CA1 populations in an environment containing two visually identical regions. J Neurosci 18:8455–8466. Small SA, Nava AS, Perera GM, DeLaPaz R, Mayeux R, Stern Y (2001) Circuit mechanisms underlying memory encoding and retrieval in the long axis of the hippocampal formation. Nat Neurosci 4:442–49. Sperling R, Chua E, Cocchiarella A, Rand-Giovannetti E, Poldrack R, Schacter DL, Albert M (2003) Putting names to faces: successful encoding of associative memories activates the anterior hippocampal formation. NeuroImage 20:1400–1410. Tanila H, Shapiro M, Gallagher M, Eichenbaum H (1997) Brain aging: impaired coding of novel environmental cues. J Neurosci 17:5167–5174.
Teng E, Squire LR (1999) Memory for places learned long ago is intact after hippocampal damage. Nature 400:675–677. Tulving E (2002) Episodic memory: from mind to brain. Annu Rev Psychol 53:1–25. Van Elzakker M, O’Reilly RC, Rudy JW (2003) Transitivity, flexibility, conjunctive representations, and the hippocampus. I. An empirical analysis. Hippocampus 13:334–340. Wang RF, Spelke ES (2000) Updating egocentric representations in human navigation. Cognition 77:215–250. Whishaw IQ, Cassel J-C, Jarrard LE (1995) Rats with fimbria-fornix lesions display a place response in a swimming pool: a dissociation between getting there and knowing where. J Neurosci 15:5779–88. Whishaw IQ, Tomie J (1997) Perseveration on place reversals in spatial swimming pool tasks: further evidence for place learning in hippocampal rats. Hippocampus 7(3):361–370. Wiener SI, Paul CA, Eichenbaum H (1989) Spatial and behavioral correlates of hippocampal neuronal activity. J Neurosci 9:237–263. Wirth S, Yanike M, Frank LM, Smith AC, Brown EN, Suzuki WA (2003) Single neurons in the monkey hippocampus and learning of new associations. Science 300:1578–1581. Wood ER, Dudchenko PA, Eichenbaum H (1999) The global record of memory in hippocampal neuronal activity. Nature 397:613–616. Wood E, Dudchenko P, Robitsek JR, Eichenbaum H (2000) Hippocampal neurons encode information about different types of memory episodes occurring in the same location. Neuron 27:623–633. Yanike M, Wirth S, Suzuki WA (2004) Representations of well-learned information in the monkey hippocampus. Neuron 42:477–487. Young, B.J., Fox, G.D. & Eichenbaum, H. (1994) Correlates of hippocampal complex spike cell activity in rats performing a non-spatial radial arm maze task. Journal of Neuroscience 14:6553–6563. Zeineh MM, Engel SA, Thompson PM, Brookheimer SY (2003) Dynamics of the hippocampus during encoding and retrieval of face-name pairs. Science 299:577–580.
II PRIMATE HIPPOCAMPUS AND PLACE REPRESENTATION
This page intentionally left blank
12 Place-Differential Neural Responses in Monkey Hippocampal Formation during Real and Virtual Navigation HISAO NISHIJO, ETSURO HORI, AND TAKETOSHI ONO
Recent neuropsychological data from humans have demonstrated a pivotal role of the medial temporal lobe, including the hippocampal formation (HF), in allocentric (environment-centered) spatial learning and memory. Patients with temporal lobe damage extending to the HF could not remember locations of landmarks in space or specific spatial relationships among the landmarks, and had deficits in a radial-arm maze task (Maguire et al., 1996a; Abrahams et al., 1997). Consistent with these findings, studies using positron emission tomography (PET) and functional magnetic resonance imaging (fMRI) of normal humans demonstrated that blood flow in the HF increased during various types of spatial tasks in a real or realistic virtual environment (Aguirre et al., 1996; Maguire et al., 1996b, 1997; Aguirre and D’Esposito, 1997; Ghaem et al., 1997). Neurophysiological studies in rats support the human data. Activity of some HF neurons is localized to a specific location in the environment. These neurons are called ‘‘place cells’’ (O’Keefe and Dostrovsky, 1971; Olton et al., 1978; McNaughton et al., 1983; Eichenbaum et al., 1987; Muller et al., 1987). Unitrecording studies in monkeys also indicated that some primate HF neurons responded to spatial cues (Watanabe and Niki, 1985; Cahusac et al., 1989; Miyashita et al., 1989; Rolls et al., 1989; Feigenbaum and Rolls, 1991; Eifuku et al., 1995; Suzuki et al., 1997). Using a paradigm in which a monkey could change its location while in a motorized movable device (spatial moving task), we have reported place correlates of monkey HF
neurons (Ono et al., 1993a,b; Nishijo et al., 1993, 1997b). The HF neurons in the monkey have also been reported to respond to spatial view (Rolls and O’Mara, 1995) and whole-body motion (O’Mara et al., 1994) during passive translocation. The relationship between spatial functions and episodic memory in the HF is one of the main issues. Changes in place-related activity (remapping) is suggested to be a neural basis of episodic memory (Guzowski et al., 2004; Jeffery et al., 2004; McNaughton et al., 2006). Basic spatial tuning of place cells is essentially formed by local geometrical intramaze and idiothetic cues (Jeffery et al., 2003, 2004; Knierim and Rao, 2003). Other factors modulate this spatial tuning, including extramaze distal cues, which might set the orientation of the place fields of place cells or work as contexts to gate local geometrical and idiothetic cues (Jeffery et al., 2003, 2004). Another factor is cognitive requirement or task paradigms, which are reported to influence activity of place cells even in the same spatial environment (Markus et al., 1995; Smith and Mizumori, 2006). Thus, rodent studies suggest that remapping of place fields might be induced by both cognitive (or nonspatial) factors such as task requirements and changes in extramaze distal cues. In this chapter, we review our studies in which monkey HF neurons were analyzed during performance of either different tasks in the same spatial environment (Matsumura et al., 1999; Hori et al., 2003) or essentially the same task in different spatial environments (i.e., different extramaze distal cues; Hori et al., 2005).
177
178 PRIMATE HIPPOCAMPUS AND PLACE REPRESENTATION
EFFECTS OF TASK PARADIGMS ON HIPPOCAMPAL NEURONAL RESPONSES We previously reported that activity of rodent and monkey HF place cells is context or task dependent (Nishijo et al., 1993; Kobayashi et al., 1997; Furusawa et al., 2006). This suggests that the HF could encode different reference frames (or charts), which might be the neural basis of episodic memory (Nishijo et al., 1993; Kobayashi et al., 1997; Redish and Touretzky, 1997; Samsonovich and McNaughton, 1997). In this section, to analyze task-dependent HF neuronal responses in monkeys, HF neurons were tested with four spatial tasks (two tasks and two variants) to observe whether the HF neurons showed different activity in the various tasks in the same spatial environment (Matsumura et al., 1999).
Behavioral Paradigms The monkey sat in a chair within a cab and could see several landmarks in the experimental room that were available to identify its position in the experimental field. The monkey always faced toward the front of the experimental field, since the monkey’s head was connected to a stereotaxic apparatus in the cab (Fig. 12–1Aa). The front wall of the cab contained a color liquid crystal display (LCD) monitor (Fig. 12–1Ab) and a joystick for moving the cab in the experimental field and/or a pointer on the LCD monitor. The cab and/ or the pointer could be moved at constant velocity in all directions during continuous manipulation of the joystick by the monkey. The movement direction of the cab and/or the pointer was linked to the direction to which the joystick was brought down by the monkey. During recording, eye movements were also monitored by an eye monitor system with an infrared CCD camera. The basic behavior required in the tasks was to either move the cab to one of four reward areas in the experimental field or move the pointer on the LCD monitor to one of four reward areas in the four corners of the LCD monitor by manipulating the joystick (Fig. 12–1Ab). Reward areas on the LCD monitors (target circle; TC) corresponded to rewarded areas in the experimental field (target area; TA). In the real translocation task with no pointer during continuous presentation of a target circle on the LCD monitor (RT/ TC) (Fig. 12–1B), no pointer was visible while the target circle was continuously presented on the LCD monitor throughout the discrimination and manipulatingresponse phases. After the cab was placed at a starting point, the task was initiated by a 1-s warning tone (1300 Hz), followed by presentation of a 20 20 cm blue square frame on the LCD monitor (warning phase). After a 2-s presentation of the target circle (discrimi-
nation phase), the monkey could move the cab toward the target area by manipulating the joystick (manipulating/response phase). When the monkey arrived at the target area indicated by the target circle on the LCD monitor, a reward of 6 ml of orange juice was given for 3 s (reward phase). In the real translocation task with a pointer during continuous presentation of a target circle on the LCD monitor (RT/P-TC) (Fig. 12–1B), the task sequence and behavioral requirements were similar to those in the RT/TC task, except that the pointer, indicating the location of the cab, was also shown on the LCD monitor. That is, the monkey had a map on a scale ratio of 1:12.61 and car navigation in the RT/P-TC task. In the virtual translocation task with the pointer (VT/P) (Fig. 12–1B), the cab was stationary at a starting point throughout the trial. In this task, a target circle was transiently presented on the LCD monitor only during the discrimination phase. Therefore, the monkey was required to memorize the location of the target circle during the discrimination phase and to move the pointer to the location based on memory. In the virtual translocation task with a pointer during continuous presentation of a target circle on the LCD monitor (VT/P-TC) (Fig. 12–1B), the task sequence and behavioral requirements were similar to those in the VT/P task, except that a target circle was presented not only during the discrimination phase but also during the manipulating phase.
Place-Differential Responses The activity of 389 neurons was recorded from the monkey HF. Of these, 166 (42.7%) neurons had place fields in the experimental field and/or on the LCD monitor. Figure 12–2 shows an example of an HF neuron with place-differential responses. The trail of the cab and the corresponding place-differential responses in the RT/TC task are illustrated in Fig. 12– 2A. The trail of the cab is shown as dotted lines in which each dot corresponds to a position of a center of the cab at each moment (Fig. 12–2Aa, left panel). Although the monkey moved and visited various sites of the experimental field and received juice rewards in the four corners of the experimental field, the activity increased in the right back corner of the experimental field (Fig. 12–2Aa, right panel). Superimposed spike waves of the HF neuron and an autocorrelogram of the neuronal spikes are shown in Figures 12–2Ab and 12–2Ac, respectively. The autocorrelogram showed that a refractory period of the neuron was 2–3 ms, indicating that these spikes were recorded from a single neuron. Figure 12–2Ba shows the relation between eye positions and neuronal activity of the same neuron
Figure 12–1. Experimental setup (A) and task paradigms (B). A. Freely movable monkey cab (a) and front panel of the cab (b). A monkey sat in a cab, which was freely moved in an experimental field in a room. On the upper part of the front wall there was a color LCD monitor. The lower part of the front wall was equipped with a joystick to freely move a cab in the experimental field and/or a pointer presented on the LCD monitor. Juice reward, controlled by an electromagnetic valve, was delivered from a tube that projected through the rear wall. In the real translocation (RT) tasks (a), the monkey moved the cab to one of four reward areas in the four corners of the experimental field by manipulating the joystick. In the virtual translocation (VT) tasks (b) the monkey moved a pointer on the LCD monitor to one of four reward areas in the four corners of the LCD monitor by manipulating the joystick. TC, target circle; TA, target area. B. Time chart of the four behavioral tasks. Each square indicates the frame that appeared on the LCD monitor. All four tasks consisted of four phases: (1) warning phase: the warning tone and the frame were presented for 1 s; (2) discrimination phase: a red target circle appeared on the LCD monitor for 2 s; (3) manipulating/response phase: a monkey manipulated a joystick to drive a cab or move a pointer on the LCD monitor; and (4) reward phase: approximately 6 ml of orange juice was delivered for 3 s. In the RT tasks, one of four target circles, corresponding to one of four target areas in the experimental field, was presented on the LCD monitor. Then the monkey manipulated the joystick to drive the cab toward the indicated target area with (RT/P-TC task) or without (RT/TC task) guidance of a pointer (P), indicating location of the cab on the monitor. In the VT tasks, the cab was stationary throughout the trial. After one of the target circles was presented on the LCD monitor, the monkey manipulated the joystick to move a pointer toward the indicated location of the target circle without (VT/P task) or with (VT/P-TC task) presentation of the target circle during the manipulating/response phase. 179
180 PRIMATE HIPPOCAMPUS AND PLACE REPRESENTATION shown in Figure 12–2A, when the monkey drove the cab from the outside of the place field to the place field during the RT/TC task. The trails of the cab are shown in Figure 12–2Bb. Neuronal activity was observed only when the monkey moved within or near the place field. Active eye movements were observed regardless of neuronal activity, indicating no correlation between eye movements and neuronal activity. No neurons
recorded in this study showed any correlation to eye movements, as shown in Figure 12–2.
Responsiveness of Hippocampal Formation Neurons across Tasks Of the 166 place-differential HF neurons, 150 (90.4%, 150/166) had place fields in the RT (RT/TC and RT/P-
Figure 12–2. An example of a hippocampal formation (HF) neuron with place-differential responses. A. Aa. The trail of the cab (left panel) and the corresponding place-differential responses (right panel) in the translocation task with no pointer during presentation of a target circle on the LCD monitor (RT/TC). A place field is surrounded by thick lines. Calibration is shown on the right; a mean firing rate for each pixel was expressed as a relative firing rate (R) in which the mean firing rate in each pixel was divided by the grand mean firing rate (M) in each task, and shown in five steps (R 2 M; 2.0 M > R 1.5 M; 1.5 M > R 1.0 M; 1.0 M > R 0.5 M; R < 0.5 M). Three values in the calibration indicate those of 2 M, 1 M, and 0, respectively. Regions not visited by the monkey during the session(s) are shown by blank pixels. Note that the activity increased in the right back corner of the experimental field, although the monkey moved and visited various sites of the experimental field and received juice reward at the four corners of the experimental field. Ab. Superimposed spike waves of the HF neuron shown in Aa. Ac. Autocorrelogram of the HF neuron shown in Aa. Ordinate, number of spikes; abscissa, time in ms; bin size, 1 ms. Note that a refractory period of the neuron was 2–3 ms, which indicates that these spikes were recorded from a single neuron. Ba,b. Relations between eye movements and activity of the neuron shown in A during the RT/TC task. Eye trace, left eye positions (a); cab location, location indicated by coordinates on X and Y axes (a); X, coordinates on X-axis; Y, coordinates on Y-axis (a); Raster, raster display of neuronal activity (a). R, right; L, left; U, upper; D, down (a). Hatched bars indicate duration of cab movements (a), light stippled areas indicate duration during which a cab was located within a place field of the neuron (b). Bb indicates actual movements (arrow) of the cab and the place field (srtiped area) of the neuron.
PLACE-DIFFERENTIAL NEURAL RESPONSES IN NAVIGATION
TC) tasks while 100 (60.2%, 100/166) had place fields in the VT (VT/P and VT/P-TC) tasks. The difference in ratios was statistically significant (Fisher’s exact probability test, p < 0.01). Furthermore, when responsiveness of the HF neurons to two RT tasks was compared, the ratio of the HF neurons that had place fields in the RT/TC task (71.7%, 119/166) was significantly larger than that of HF neurons that had place fields in the RT/P-TC task (44.0%, 73/166) (Fisher’s exact probability test, p < 0.01). In the RT/TC task, no pointer was presented on the LCD monitor. Furthermore, the monkey always faced in a fixed direction; consequently, the same landmarks were seen in each trial. Therefore, the monkey had to judge its position according to the survey knowledge of the landmarks in the experimental room (i.e., relative spatial knowledge of place; Thorndyke and Hayes-Roth, 1982; Aguirre and D’Esposito, 1997). Furthermore, the monkey could flexibly change its course during translocation when movement direction deviated from the destination. This evidence strongly suggests that the monkey’s behavior was based on a cognitive map (locale system) in which the spatial relationships of various landmarks are represented, rather than on taxon systems, where a set of stimulus (single landmark)–response (action or movement) associations are represented (O’Keefe and Nadel, 1978). However, the monkey did not necessarily judge its position based on the cognitive maps in the RT/P-TC task, since its position was indicated by the pointer on the LCD monitor, although behavioral requirements in the RT/P-TC task were similar to those in the RT/TC tasks. These results indicate that more HF neurons responded in the RT/TC than in RT/P-TC tasks, and strongly suggest that the HF is more important for information processing in the locale system than in other systems. This role is consistent with the cognitive-map theory advanced by O’Keefe and Nadel (1978), in which the HF is suggested to be a neural substrate of the cognitive map. The VT tasks have similar characteristics to tabletop visuospatial tasks used with humans (e.g., Corsi Tapping task, Rey-Osterrieth figure, stylus-maze learning). Patients with topographical disorientation seldom show deficits in such tabletop tests (Habib and Sirigu, 1987; Maguire et al., 1996a). These clinical studies, along with the PET and fMRI studies demonstrating an increase in blood flow in the HF in a realistic virtual environment, suggest that it is important to test topographical disorientation in a real environment or a large-scale realistic virtual environment. In the RT-VT study, more HF neurons responded in the RT than in the VT tasks, a finding supporting this idea. Furthermore, previous unitrecording studies of the monkey HF reported that approximately or less than 10% of the HF neurons showed
181
spatial responses when the monkey remained at a fixed location or was moved passively by the experimenter (Cahusac et al., 1989; Miyashita et al., 1989; Rolls et al., 1989; Feigenbaum and Rolls, 1991). In the present study, about 40 % of the HF neurons showed place-differential responses. These results indicate that population activity of HF neurons is dependent on cognitive demands in allocentric spatial processing and are consistent with an fMRI study in humans in which the amount of neuronal activity was dependent on the computational demand that a given task imposed (Just et al., 1996). Taken together, this evidence strongly suggests a pivotal role of the HF in allocentric spatial information processing in primates as well as in rats.
Correlation of Place Fields across Tasks Of the 166 place-differential neurons, 68 had place field(s) in only one task. Of the remaining 98 neurons that had place fields in more than two tasks, only 17 neurons had overlapped place fields across the tasks. Thus, most HF place-differential neurons (149/ 166, 89.8%) had non-overlapped place fields rather than overlapped place fields. Of the 17 neurons that had overlapped place fields, 9 had overlapped place fields in only some of the tasks but not all the tasks in which given neurons had place fields (partial overlap). Figure 12–3 shows an example of RT/TC-responsive HF neurons that had place fields in not only the RT/TC task but also the VT/P and VT/P-TC tasks, and showed partially overlapped place fields. The place fields of the HF neuron in the RT/TC and VT/P task were overlapped (Fig. 12–3A,C), but not with the place field in the VT/P-TC task (Fig. 12–3D). The remaining six neurons had place fields in more than two tasks and displayed overlapped place fields in those tasks. Although these place-differential HF neurons had overlapping place fields across some tasks, these neurons did not have common place fields across all four tasks, since place-differential activity was observed only in some tasks but not across the four tasks. That is, no place-differential neurons displayed overlapped place fields across the four tasks. In this study, most HF place-differential neurons (89.8%) had non-overlapped place fields rather than overlapped place fields. In another study, activity of rodent place cells in the CA1 and CA3 subfields was highly sensitive to environmental changes and showed different representation in each different environment (see review by Redish and Touretzky, 1997). These results are consistent with our results, in which different neuronal representations were established in different tasks. Recent theoretical studies have proposed that the HF represents the external world by a reference frame
182 PRIMATE HIPPOCAMPUS AND PLACE REPRESENTATION
Figure 12–3. An example of trail and firing rate maps of the RT/TC-responsive neurons that had place fields in the translocation task with no pointer during presentation of a target circle on the LCD monitor (RT/TC), the virtual translocation task with the pointer (VT/P), and the virtual translocation task with a pointer during presentation of a target circle on the LCD monitor (VT/P-TC). Note that the hippocampal formation neuron had partially overlapped place fields. In the RT/TC and VT/P tasks the place fields in the left front area on the frame overlapped (Ab, Cb), but the place field was in the right lower area on the frame in the VT/ P-TC task (Db). Other descriptions are given in Figure 12–2.
(Redish and Touretzky, 1997) or chart (Samsonovich and McNaughton, 1997) system in which different assemblies of different HF neurons or HF neurons with different place fields are created in different environments or behavioral contexts. The four different maps of the place fields in the four different tasks might correspond to four different reference frames or charts. Rodent and human neurophysiological and behavioral studies support this idea that the HF is important in the creation of reference frames and consequently in reducing interference among different environments and contexts. Place fields of the place cells of aged rats with spatial memory deficits were shown to be unstable when tested repeatedly in the same environment, thus these animals likely have deficits in selecting a correct reference frame in a familiar environment (Barnes et al., 1997). Rats with HF lesions exhibited the same behavioral responses to contextual cues, regardless of the degree of conditioning (Wincour et al., 1987), and humans and animals with HF damage are impaired in acquiring conditional relation between stimuli in a variety of situations (Hirsh, 1980; Ross
et al., 1984). All of this evidence suggests that an assembly of reference frames is a neural basis of episodic memory for various events occurring in different environmental and contextual situations. Nonetheless, in our study 10.2% of the placedifferential neurons had overlapped place fields across the tasks in which place-differential activity was observed (partial overlap). We speculate that the monkey might have accepted the LCD monitor as a map of a real space, since the four tasks used in this study had similar characteristics. These HF neurons with overlapped place fields might connect the four different reference frames, among which were significant pointto-point relationships. Future computational studies may clarify and test this hypothesis.
SPATIAL REPRESENTATION IN MONKEY HIPPOCAMPAL FORMATION Previous neurophysiological studies as well as that described above have demonstrated that the activity of
PLACE-DIFFERENTIAL NEURAL RESPONSES IN NAVIGATION
complex-spike cells in the HF increases when a rat or monkey is in a specific location in the environment (O’Keefe and Dostrovsky, 1971; Olton et al., 1978; McNaughton et al., 1983; Eichenbaum et al., 1987; Muller and Kubie, 1987; Wiener et al., 1989; Ono et al., 1993a,b; Wilson and McNaughton, 1993, 1994; Kobayashi et al., 1997; Nishijo et al., 1997b). O’Keefe and Nadel (1978) proposed a cognitive-map theory in which the HF was the neural substrate of the cognitive map. This theory and the relational-association theory (Eichenbaum et al., 1990; Young et al., 1994) suggest that spatial relationships among various environmental stimuli are crucial to neural representation of space, and that the spatial relationships of environmental cues surrounding the animal (i.e., cognitive map) are represented by the activity pattern of an ensemble of HF place cells. The relationship between HF place cells and the cognitive map has not been clarified in monkeys, however. In this section, to investigate how HF placedifferential neurons encode spatial relationships of the various environmental stimuli and represent space across activity patterns, we reanalyzed the same data set of the monkey HF neuronal responses described in the previous section (Matsumura et al., 1999) by means of multivariate statistical analyses. Multidimensional scaling (MDS) is one method of simplifying the relationships within a complex array of data; through MDS one can construct a geometric representation of data to show the degree of relation between stimuli represented by the data matrix (see Young, 1987, for more details). We hypothesized that environmental stimuli (i.e., different locations) work to elicit neural activity of each HF neuron. We then used MDS to analyze the relations among various locations in the experimental field where the monkey moved to see if they might correspond to the cognitive map (Hori et al., 2003).
Representation of Space by Place-Differential Hippocampal Formation Neurons Data from the same 166 place-differential responses that had place field(s) in at least one of four tasks (Matsumura et al., 1999) were analyzed by MDS. Each data set in each of 64 small pixels (an 8 8 array; 25 25 cm pixels in the experimental field and 1.98 1.98 cm pixels on the LCD monitor) was converted into 16 large pixels (places A–P in Fig. 12–4A) by averaging neuronal activity of 4 small pixels. In all of the four tasks, the pixels in the 4 4 array were represented as places A to P (Fig. 12–4A). Thus, a data matrix of neural activity in a 4 4 array derived from all place-differential neurons in each task was generated. Pearson’s correlation coef-
183
ficients between all possible pairs of two places were calculated by using the place-differential responses. If Pearson’s correlation coefficient between a given pair of two places was 1.0, response patterns of the placedifferential neurons to the two places were identical. A Pearson’s correlation coefficient of 0 indicated that the response patterns to the two places were completely different. In MDS the stimuli (i.e., places A–P in the present study) are positioned in a geometric configuration such that the closeness between the stimuli corresponds as much as possible to original proximities (i.e., Pearson’s correlation coefficients in this study; see Shepard, 1962; Kruskal, 1964). The MDS program computed the Pearson’s correlation coefficients between all possible pairs of two places to give interplace relationships by plotting relative positions of places A–P in a two-dimensional space. In the twodimensional space derived from MDS, the distance between a given pair of two places is negatively correlated to the Pearson’s correlation coefficient between the two places. Figure 12–4B illustrates results by MDS analysis for two-dimensional space. To facilitate a comparison between real positions of places A–P in the experimental room and ideal relative positions of places A–P in the two-dimensional space derived from MDS, the MDS spaces were superimposed on the spatial arrangement of the experimental room in Figure 12–4A. The position of each place in the MDS space was normalized so that X–Y coordinates of place A were the number pair (0, 3) and those of place M were the number pair (0, 0) (i.e., distance between places A and M was 3.0). In RT/TC and RT/P-TC tasks, the 16 places spread over two-dimensional space and their relative positions in two-dimensional space were similar to those in the experimental room (Figs. 12–4Ba and 12–4Bb). However, the relative positions of the 16 places in two-dimensional space in VT/P and VT/ P-TC tasks were rather distorted (Figs. 12–4Bc and 12–4Bd). HF place cells in rats increase firing rates in specific spatial locations during free movement (Wilson and McNaughton, 1993;O’Keefe andBurgess,1996;Barnes et al., 1997). Activity of monkey HF neurons is also place related (Ono et al., 1993a,b; Eifuku et al., 1995; Nishijo et al., 1997b; Matsumura et al., 1999). Although the emergent dimensions in MDS may not be inherently meaningful, this does not preclude identification of meaningful dimensions. We interpreted the two dimensions in MDS to reflect the two-dimensional space in the experimental field (or the LCD monitor), since the relative positions of 16 places in the MDS space were very similar to those in the experimental field (or on the monitor) (Fig. 12–4). These results and those from the neurophysiological studies suggest that
184 PRIMATE HIPPOCAMPUS AND PLACE REPRESENTATION
Figure 12–4. Spatial representation in the monkey hippocampal formation neuron. A. Places A–P in real space were superimposed on a twodimensional space. Expri, experimenter; Oscillo, oscilloscope; D.W. Amp, data wave amplifier; Contr, controller; PC-98, NEC computer. Ba–d. Multidimensional scaling (MDS) arrangements of places in the RT/TC task (a), RT/P-TC task (b), VT/P task (c), and VT/P-TC task (d). Note that the MDS spatial arrangement of places derived from neuronal activity represented well the real spatial arrangement in the RT/TC task.
spatial information was integrated in the HF or other brain areas such as the medial entorhinal cortex that project to the HF to represent a map-like configuration of environmental cues (Skaggs and McNaughton, 1998; McNaughton et al., 2006), where the geometric relations were well preserved. These results strongly suggest that the configuration of places A–P derived from MDS corresponds to a kind of cognitive map in the RT tasks and that the activity patterns of monkey HF neurons represent spatial arrangements of environmental cues.
Comparison of Multidimensional Scaling Spaces among Tasks To numerically analyze the similarity between relative positions of 16 places in the MDS space and those in the experimental field (or on the monitor), Pearson’s correlation coefficients between these two data sets in the MDS and real space were calculated
using distance between possible pairs of two places in the MDS and real space. Correlation coefficients between the MDS and real space were relatively high in the RT/TC and in RT/P-TC tasks (0.964 and 0.873, respectively). Correlation coefficients between real space and two kinds of the virtual tasks (VT/P and VT/P-TC tasks), by contrast, were relatively low (0.659 and 0.710, respectively). Correlation coefficients between MDS and real space were significantly larger in the RT/TC task than in the other tasks (twotailed t-test after Fisher’s z transformation, p < 0.01). In the RT/P-TC task, correlation coefficients between the MDS and real space were also significantly larger than in the two virtual translocation tasks (VT/P and VT/P-TC tasks) (two-tailed t-test after Fisher’s z transformation, p < 0.01). By contrast, correlation coefficients between the MDS and real space were not significantly different between the VT/P and of VT/PTC tasks (two-tailed t-test after Fisher’s z transformation, p > 0.05).
PLACE-DIFFERENTIAL NEURAL RESPONSES IN NAVIGATION
In the real translocation tasks (RT/TC and RT/PTC), MDS arrangement had a higher correlation to the real-space arrangement than in the virtual translocation (VT/P and VT/P-TC) tasks. Furthermore, this correlation with real space was higher in RT/TC than in RT/PTC tasks. The pointer in the LCD monitor was not presented in the RT/TC task. In addition, the monkey always faced in a fixed direction; consequently, the same landmarks were seen in each trial. The monkey had to judge its position on the basis of survey knowledge of the landmarks in the experimental room in the RT/TC task (i.e., relative spatial knowledge of place; Thorndyke and Hayes-Roth, 1982; Aguirre and D’Esposito, 1997). Furthermore, the monkey could flexibly change its course during translocation when movement direction deviated from the destination. This evidence strongly suggests that the monkey’s behavior was based on a cognitive map (locale system) in which the spatial relationships of various landmarks are represented, rather than on taxon systems in which a set of stimulus–response (e.g., single landmarks to actions or movements) associations are represented (O’Keefe and Nadel, 1978). However, the monkey did not necessarily judge its position according to the cognitive maps in the RT/P-TC task, since its position was indicated by the pointer on the LCD monitor, although the behavioral requirements in the RT/P-TC task were similar to those in the RT/TC tasks. In the VT tasks, cognitive spatial information processing (information processing in the locale system) was not necessary because the pointer and/or a target circle were continuously presented on the monitor and the cab was located in a fixed place. These results suggest that the degree of cognitive spatial information processing is correlated with accuracy of representation of real space in the HF, and that neuronal events in the HF are highly dependent on cognitive demands required for spatial tasks and the animal’s active interaction with the environment. Distortion of spatial representation in the VT tasks might be ascribed to ambiguous place fields in the VT/P-TC task (Matsumura et al., 1999). Furthermore, place-differential neurons were reported in the monkey septal nuclei (Kita et al., 1995), which receive direct inputs from the HF pyramidal cells (Risold and Swanson, 1996). The same MDS analysis indicated that the spatial configuration in MDS space based on septal neuron responses was correlated with behavioral performance of place discrimination (Nishijo et al., 1997a). These results strongly suggest that functions of the septohippocampal system are directly related to spatial navigation ability, consistent with previous behavioral studies demonstrating that HF and septal lesions result in navigation deficits in various spatial tasks (Morris et al., 1982; Eichenbaum et al., 1990; M’Harzi and Jarrard, 1992).
185
MONKEY HIPPOCAMPAL FORMATION NEURONAL RESPONSES DURING NAVIGATION IN DIFFERENT VIRTUAL SPACES The results described in the previous two sections have demonstrated that cognitive factors such as task requirements have a great impact on place cell activity. In our and previous studies on primates, HF neurons were recorded in either a single experimental room or in virtual space (Ono et al., 1993a,b; Nishijo et al., 1997b; Matsumura et al., 1999; Ekstrom et al., 2003; Ma et al., 2004), but they were not tested in multiple spaces (i.e., multiple arrangements of extramaze distal cues). Thus the purpose in this section is to investigate effects of extramaze distal cues on HF place-differential activity (i.e., remapping of placedifferential activity) while the monkey navigated in a same arena with different arrangements of the extramaze distal cues. Monkeys were required to navigate in a virtual water maze (Hori et al., 2005), where human subjects and monkeys with hippocampal damage had been reported to display spatial memory impairments (Astur et al., 2002; Hampton et al., 2004).
Behavioral Paradigms Monkeys were trained to sit on a chair in front of a 1.5 (h) 1.9 (w) screen in a dark room. Fitted with glasses with polarized lenses, the monkeys viewed polarized three-dimensional visual images of virtual spaces projected from a LCD projector onto the screen for a stereopsis (binocular stereopsis). For the virtual navigation task (VNT), a large-scale virtual space (diameter, 100 m) was created using 3D software (Fig. 12–5Ab). A circular open field (diameter, 20 m; movable arena) surrounded by a wall (height, 0.3 m) was located in the center of the virtual space. The movable arena contained five reward areas (diameter, 0.6 m) that were symmetrically placed (north [N], east [E], south [S], west [W], and center [C]), and extramaze distal cues (house, tree, building, rock, posters) were placed outside the movable arena. In this VNT task, the monkey had to navigate (velocity, 1 m/s) by manipulating a joystick among the three reward areas aligned in a line (i.e., N–C–S or E–C–W) to acquire a juice reward at each of these three reward areas. There were three different arrangements of the virtual spaces, based on the placement of the extramaze distal cues (VNT1–3) (Fig. 12–5B). Each trial consisted of 20 rewards; the monkey visited each side of the reward area 5 times and the center reward area 10 times in each trial. Each neuron was tested in at least two trials with the different reward areas (i.e., N–C–S and W–C–E) in each virtual space.
186 PRIMATE HIPPOCAMPUS AND PLACE REPRESENTATION In a pointer translocation task (PTT) (Fig. 12– 5Aa), a pointer (diameter, 11 cm) and two reward circles (diameter, 22 cm; distance between the centers, 110 cm) appeared on the screen. Two reward areas were located diametrically opposite each other on the screen (right/left or upper/lower). The monkeys had to manipulate the joystick to move the pointer to one of the two reward circles (velocity; 0.2 m/s). Following a correct trial, when the pointer entered one of the two reward circles after a visit to the other one, the monkeys received a juice reward.
Place-Differential Neurons in Virtual Spaces A total of 228 single neurons were recorded from the primate HF. Of these 228 neurons, 72 displayed sig-
nificant spatial correlates responses (place-differential neurons). Figure 12–6 illustrates an example of activity of a typical place-differential neuron. The trail (path) of the monkey in the movable arena and firing rate map of the neuron in the VNT1 are shown in Figure 12–6Aa. Although the monkey navigated along a line of N–C–S or W–C–E, activity of the neuron increased around the reward area at the N (Ab). Figure 12–6Ac–j indicates peri-event histograms of the same HF neuron, which were aligned with delivery of juice reward in each reward area. Activity of the neuron increased when the monkey navigated from the center reward area to the north reward area (Fig. 12–6Ad), and before and after the reward delivery at the north reward area (Fig. 12–6Ah). That is, the monkey navigated toward the north before the reward delivery and toward the south after the reward delivery, and the
Figure 12–5. Layout of spatial cues in virtual space in the pointer translocation task (PTT) and virtual navigation task (VNT). A. Arrangement of the reward areas in the PTT (a) and VNT (b). The square (a) indicates a screen where four circles for the reward areas were projected from a projector. The outer large circle (b) indicates a movable arena (diameter, 20 m) that was surrounded by a wall (height, 0.3 m), and five reward areas were located symmetrically (north, east, south, west, center) within the movable arena. The monkey navigated toward the reward areas during the virtual navigation task. Curved lines, indicate trails of the pointer (a) and monkey (b); filled dots indicate juice rewards. B. Arrangements of the distal cues in VNT1 (a), VNT2 (b), and VNT3 (c). Outer and inner circles indicate the entire virtual-reality space and the movable arena, respectively. The entire virtual-reality space was 100 m in diameter and there were some landmarks outside the circular movable arena, i.e., a house, a tree, a building, a rock, and three posters.
PLACE-DIFFERENTIAL NEURAL RESPONSES IN NAVIGATION
activity increased in both directions. These results indicate that activity of this neuron increased regardless of movement directions, suggesting that this activity reflected place around the north reward area, rather than a specific visual scene. Figure 12–6Ba shows a trail (path) of the pointer and a firing rate map of the same neuron in the PTT. In this task, activity of the neuron increased only around the right reward area. Of 144 neurons tested with both the VNT and PTT, 59 displayed place-differential responses in the VNT and/or PTT tasks. Of these 59 place-differential neurons, 21 displayed place-differential activity in both the VNT and PTT, and 29 and 9 neurons displayed activity only in the VNT and PTT, respectively. These results indicate that monkey HF neurons display place-differential activity in virtual spaces,
187
even though no specific landmarks except the symmetrically placed reward areas were located within the movable arena. In previous studies of rats, specific characteristics of place cells were formed by exploratory navigation, and idiothetic cues such as proprioceptive and vestibular information were essential for place cell activity in real environments (Muller and Kubie, 1987; O’Keefe and Speakman, 1987; Breese et al., 1989; Foster et al., 1989; Barnes et al., 1997; Kobayashi et al., 1997; Stackman et al., 2002). In our study, the virtual images provided optic flow (one of idiothetic information) to the monkey, even though the monkey was physically fixed in the same position in the experimental room. Thus place-differential activity could be formed through an interaction between the intramaze local cues and optic flow without
Figure 12–6. Example of a place-differential neuron in virtual space.A. Trail (a) and average firing rate map (b) of a place-differential neuron in VNT1. Activity of the neuron increased around the reward area in the north, even though the monkey visited all reward areas. Histograms in c–j indicate those aligned with reward delivery in each reward area. In the center reward area, peri-event histograms were created according to movement directions of the monkey. Large and small circles in Aa indicate the movable arena and reward areas, respectively. B. Trail (a) of the pointer and average firing rate map (b) in the PTT of the same neuron shown in A. In this task, activity of the neuron increased only around the right reward area. Both firing rate maps are displayed in a three-dimensional manner with a color scale.
188 PRIMATE HIPPOCAMPUS AND PLACE REPRESENTATION proprioceptive and vestibular information during virtual navigation. Consistent with this suggestion, recent noninvasive studies have shown that mental imagery of spatial navigation in a map and navigation in a virtual town increased HF blood flow (Maguire et al., 1998; Mellet et al., 2000). Interestingly, human patients with chronic bilateral vestibular failure show deficits in a virtual water-maze task, which suggests that chronic elimination of vestibular inputs might alter activity of HF circuits involved in ‘‘top-down’’ spatial information processing that responds to both real and virtual navigation (Schautzer et al., 2003). These results are consistent with the idea that the primate HF is involved in higher ‘‘top-down’’ spatial information–processing functions such as cognitive maps that are commonly activated during both real and virtual navigation, as well as in mental imagery.
Remapping of Hippocampal Formation Place-Differential Activity In our study remapping (changes in place-differential activity) was defined as follows: (1) the neurons displayed place-differential activity (place field) in at least one VNT task, but no place-differential activity in at least one of the remaining VNT tasks (type 1 remapping); and (2) when the neurons displayed placedifferential activity in the two or three VNT tasks, pixel-to-pixel correlation coefficients were less than 0.3 in at least one pair of the firing-rate maps in the VNT tasks (type 2 remapping). The effects of the extramaze distal cues on the place-related responses of the monkey HF neuron in the three VNT tasks are shown in Figure 12–7. Activity of the neuron increased only in the VNT1, and not in VNT2 (Fig. 12–7B) or
Figure 12–7. Example of a place-differential neuron that displayed remapping of place-differential activity among the three VNT tasks. A–C. Trails of the monkey (a) and average firing rate maps (b) of the placedifferential neuron in VNT1 (A), VNT2 (B), and VNT3 (C). Activity of the neuron increased only in VNT1. D. Firing-rate histogram according to movement directions of the monkey in the VNT1. Other descriptions are as for Figure 12–6.
PLACE-DIFFERENTIAL NEURAL RESPONSES IN NAVIGATION
VNT3 (Fig. 12–7C). The activity of the neuron in the VNT1 was not correlated to movement direction (Fig. 12–7D). This response is typical of type 1 remapping, in which the neuron displays differential responses to the specific space. Of the 127 neurons tested with more than two VNT tasks, 40 displayed place-related activity in at least one VNT task. Of these 40 placerelated neurons, 21 and 6 neurons showed type 1 and 2 remapping, respectively. These results indicate that the response of most of the HF place-related neurons depended on the arrangement of the extramaze distal cues, consistent with previous studies in rodents in which place cells were highly sensitive to distal cues (Tanila et al., 1997; Brown and Skaggs, 2002; Hayman et al., 2003; Jeffery et al., 2003, 2004). Thus primate HF neurons appear to share similar characteristics to those of rodents. These results also suggest that distal cues work as a context to gate local and idiothetic cues that are essential for spatial tuning within a maze (Hayman et al., 2003). Furthermore, even in the same experimental room where monkeys navigated by driving a cab, most HF neurons had non-overlapped place fields across the different tasks (Matsumura et al., 1999). In conclusion, the findings presented above suggest that the HF represents allocentic space with contexts (Jeffery et al., 2004), reference frames (Redish and Touretzky, 1997), or a chart system (Samsonovich and McNaughton, 1997), in which different assemblies of different HF neurons with different spatial tuning are created for each different environment or behavioral context. The HF appears to be crucial for the processing of both allocentric information and encoding different reference frames (contexts), which may be the neural basis of episodic memory (Nadel and Willner, 1980).
References Abrahams S, Pickering A, Polkey CE, Morris RG (1997) Spatial memory deficits in patients with unilateral damage to the right hippocampal formation. Neuropsychologia 35:11–24. Aguirre GK, D’Esposito M (1997) Environmental knowledge is subserved by separable dorsal/ventral neural areas. J Neurosci 17:2512–2518. Aguirre GK, Detre JA, Alsop DC, D’Esposito M (1996) The parahippocampus subserves topographical learning in man. Cereb Cortex 6:823–829. Astur RS, Taylor LB, Mamelak AN, Philpott L, Sutherland RJ (2002) Humans with hippocampus damage display severe spatial memory impairments in a virtual Morris water task. Behav Brain Res 132:77–84. Barnes CA, Suster MS, Shen J, McNaughton BL (1997) Multistability of cognitive maps in the hippocampus of old rats. Nature 388:272–275.
189
Breese CR, Hampson RE, Deadwyler SA (1989) Hippocampal place cells: stereotypy and plasticity. J Neurosci 9:1097–1111. Brown JE, Skaggs WE (2002) Concordant and discordant coding of spatial location in populations of hippocampal CA1 pyramidal cells. J Neurophysiol 88: 1605–1613. Cahusac PMB, Miyashita Y, Rolls ET (1989) Responses of hippocampal formation neurons in the monkey related to delayed spatial response and object-place memory tasks. Behav Brain Res 33:229–240. Eichenbaum H, Kuperstein M, Fagan A, Nagode J (1987) Cue-sampling and goal-approach correlates of hippocampal unit activity in rats performing an odordiscrimination task. J Neurosci 7:716–732. Eichenbaum H, Stewart C, Morris RGM (1990) Hippocampal representation in place learning. J Neurosci 17:3531–3542. Eifuku S, Nishijo H, Kita T, Ono T (1995) Neuronal activity in the primate hippocampal formation during a conditional association task based on the subject’s location. J Neurosci 15:4952–4969. Ekstrom AD, Kahana MJ, Caplan JB, Fields TA, Isham EA, Newman EL, Fried I (2003) Cellular networks underlying human spatial navigation. Nature 425: 184–188. Feigenbaum JD, Rolls ET (1991) Allocentric and egocentric spatial information processing in the hippocampal formation of the behaving primate. Psychobiology 19:21–41. Foster TC, Castro CA, McNaughton BL (1989) Spatial selectivity of rat hippocampal neurons: dependence on preparedness for movement. Science 244:1580–1582. Furusawa AA, Hori E, Umeno K, Tabuchi E, Ono T, Nishijo H (2006) Unambiguous representation of overlapping serial events in the rat hippocampal formation. Neuroscience 137:685–698. Ghaem O, Mellet E, Crivello F, Tzourio N, Mazoyer B, Berthoz A, Denis M (1997) Mental navigation along memorized routes activates the hippocampus, precuneus, and insula. Neuroreport 8:739–744. Guzowski JF, Knierim JJ, Moser EI (2004) Ensemble dynamics of hippocampal regions CA3 and CA1. Neuron 44:581–584. Habib M, Sirigu A (1987) Pure topographical disorientation: a definition and anatomical basis. Cortex 23:73– 85. Hampton RR, Hampstead BM, Murray EA (2004) Selective hippocampal damage in rhesus monkeys impairs spatial memory in an open-field test. Hippocampus 14:808–818. Hayman RM, Chakraborty S, Anderson MI, Jeffery KJ (2003) Context-specific acquisition of location discrimination by hippocampal place cells. Eur J Neurosci 18:2825–2834.
190 PRIMATE HIPPOCAMPUS AND PLACE REPRESENTATION Hirsh R (1980) The hippocampus, conditional operations, and cognition. Physiol Psychol 8:175–182. Hori E, Nishio Y, Kazui K, Umeno K, Tabuchi E, Sasaki K, Endo S, Ono T, Nishijo H (2005) Place-related neural responses in the monkey hippocampal formation in a virtual space. Hippocampus 15:991–996. Hori E, Tabuchi E, Matsumura N, Tamura R, Eifuku S, Endo S, Nishijo H, Ono T (2003) Representation of place by monkey hippocampal neurons in real and virtual translocation. Hippocampus 13:190–196. Jeffery KJ, Anderson MI, Hayman R, Chakraborty S (2004) A proposed architecture for the neural representation of spatial context. Neurosci Biobehav Rev 28:201–218. Jeffery KJ, Gilbert A, Burton S, Strudwick A (2003) Preserved performance in a hippocampal-dependent spatial task despite complete place cell remapping. Hippocampus 13:175–189. Just MA, Carpenter PA, Keller TA, Eddy WF, Thulborn KR (1996) Brain activation modulated by sentence comprehension. Science 274:114–116. Kita T, Nishijo H, Eifuku S, Terasawa K, Ono T (1995) Place and contingency differential responses of monkey septal neurons during conditional place-object discrimination. J Neurosci 15:1683–1703. Knierim JJ, Rao G (2003) Distal landmarks and hippocampal place cells: effects of relative translation versus rotation. Hippocampus 13:604–617. Kobayashi T, Nishijo H, Fukuda M, Bures J, Ono T (1997) Task-dependent representations in rat hippocampal place neurons. J Neurophysiol 78:597–613. Kruskal JB (1964) Multidimensional scaling by optimizing goodness of fit to a nonmetric hypothesis. Psychometrika 29:1–27. Ma YY, Ryou JW, Kim BH, Wilson FA (2004) Spatially directed movement and neuronal activity in freely moving monkey. Prog Brain Res 143:513–520. Maguire EA, Burgess N, Donnet JG, Frackowiak RSJ, Frith CD, O’Keefe J (1998) Knowing where and getting there: a human navigation network. Science 280:921–924. Maguire EA, Burke T, Phillips J, Staunton H (1996a) Topographical disorientation following unilateral temporal lobe lesions in humans. Neuropsychologia 34: 993–1001. Maguire EA, Frackowiak RSJ, Frith CD (1996b) Learning to find your way: a role for the human hippocampal formation. Proc R Soc Lond B 263:1745–1750. Maguire EA, Frackowiak RSJ, Frith CD (1997) Recalling routes around London: activation of the right hippocampus in taxi drivers. J Neurosci 15:7103–7110. Markus EJ, Qin YL, Leonard B, Skaggs WE, McNaughton BL, Barnes CA (1995) Interactions between location and task affect the spatial and directional
firing of hippocampal neurons. J Neurosci 15:7079– 7094. Matsumura N, Nishijo H, Tamura R, Eifuku S, Endo S, Ono T (1999) Spatial- and task-dependent neuronal responses during real and virtual translocation in the monkey hippocampal formation. J Neurosci 19:2381– 2393. McNaughton BL, Barnes CA, O’Keefe J (1983) The contribution of position, direction, and velocity to single unit activity in the hippocampus of freely moving rats. Exp Brain Res 52:41–49. McNaughton BL, Battaglia FP, Jensen O, Moser EI, Moser MB (2006) Path integration and the neural basis of the ‘cognitive map.’ Nat Rev Neurosci 7:663–678. Mellet E, Briscogne S, Tzourio-Mazoyer N, Ghaem O, Petit L, Zago L, Etard O, Berthoz A, Mazoyer B, Denis M (2000) Neural correlates of topographic mental exploration: the impact of route versus survey perspective learning. Neuroimage 12:588–600. M’Harzi M, Jarrard LE (1992) Effects of medial and lateral septal lesions on acquisition of a place and cue radial maze task. Behav Brain Res 49:159–165. Miyashita Y, Rolls ET, Cahusac PMB, Niki H, Feigenbaum JD (1989) Activity of hippocampal formation neurons in the monkey related to a conditional spatial response task. J Neurophysiol 61:669–678. Morris RG, Garrud P, Rawlins JN, O’Keefe J (1982) Place navigation impaired in rat with hippocampal lesions. Nature 297:681–683. Muller RU, Kubie JL (1987) The effects of changes in the environment on the spatial firing of hippocampal complex-spike cells. J Neurosci 7:1951–1968. Muller RU, Kubie JL, Ranck JB Jr (1987) Spatial firing patterns of hippocampal complex-spike cells in a fixed environment. J Neurosci 7:1935–1950. Nadel L, Willner J (1980) Context and conditioning. Physiol Psychol 8:218–228. Nishijo H, Kita T, Tamura R, Uwano T, Terasawa K, Ono T (1997a) Septal neuronal responses related to spatial representation in monkeys. Hippocampus 7:460–464. Nishijo H, Ono T, Eifuku S, Tamura R (1997b) The relationship between monkey hippocampus placerelated neural activity and action in space. Neursci Lett 226:57–60. Nishijo H, Ono T, Tamura R, Nakamura K (1993) Amygdalar and hippocampal neuron responses related to recognition and memory in monkey. Prog Brain Res 95: 339–357. O’Keefe J, Burgess N (1996) Geometric determinants of the place fields of hippocampal neurons. Nature 381: 425–428. O’Keefe J, Dostrovsky J (1971) The hippocampus as a spatial map. Preliminary evidence from unit activity in the freely moving rat. Brain Res 34:171–175.
PLACE-DIFFERENTIAL NEURAL RESPONSES IN NAVIGATION
O’Keefe J, Nadel L (1978) The Hippocampus as a Cognitive Map. Oxford: Clarendon Press. O’Keefe J, Speakman A (1987) Single unit activity in the rat hippocampus during a spatial memory task. Exp Brain Res 68:1–27. Olton DS, Branch M, Best PJ (1978) Spatial correlates of hippocampal unit activity. Exp Neurol 58:387–409. O’Mara SM, Rolls ET, Berthoz A, Kesner RP (1994) Neurons responding to whole-body motion in the primate hippocampus. J Neurosci 14:6511–6523. Ono T, Eifuku S, Nakamura K, Nishijo H (1993a) Monkey hippocampal neuron responses related to spatial and non-spatial influence. Neurosci Lett 159:75–78. Ono T, Nakamura K, Nishijo H, Eifuku S (1993b) Monkey hippocampal neuron related to spatial and non-spatial functions. J Neurophysiol 70:1516–1529. Redish AD, Touretzky DS (1997) Cognitive maps beyond the hippocampus. Hippocampus 7:15–35. Risold PY, Swanson LW (1996) Structural evidence for functional domains in the rat hippocampus. Science 272:1484–1486. Rolls ET, Miyashita Y, Cahusac PMB, Kesner RP, Niki H, Feigenbaum JD, Bach L (1989) Hippocampal neurons in the monkey with activity related to the place in which a stimulus is shown. J Neurosci 9:1835–1845. Rolls ET, O’Mara SM (1995) View-responsive neurons in the primate hippocampal complex. Hippocampus 5:409–424. Ross RT, Orr WB, Holland PC, Berger TW (1984) Hippocampectomy disrupts acquisition and retention of learned conditioned responding. Behav Neurosci 98: 211–225. Samsonovich A, McNaughton BL (1997) Path integration and cognitive mapping in a continuous attractor neural network model. J Neurosci 17:5900–5920. Schautzer F, Hamilton D, Kalla R, Strupp M, Brandt T (2003) Spatial memory deficits in patients with chronic bilateral vestibular failure. Ann NY Acad Sci 1004:316–324. Shepard RN (1962) The analysis of proximities: multidimensional scaling with an unknown distance function. Psychometrika 27:125–140.
191
Skaggs WE, McNaughton BL (1998) Spatial firing properties of hippocampal CA1 populations in an environment containing two visually identical regions. J Neurosci 18:8455–8466. Smith DM, Mizumori SJ (2006) Learning-related development of context-specific neuronal responses to places and events: the hippocampal role in context processing. J Neurosci 26:3154–3163. Stackman RW, Clark AS, Taube JS (2002) Hippocampal spatial representations require vestibular input. Hippocampus 12:291–303. Suzuki WA, Miller EK, Desimone R (1997) Object and place memory in the macaque entorhinal cortex. J Neurophysiol 78:1062–1081. Tanila H, Shapiro ML, Eichenbaum H (1997) Discordance of spatial representation in ensembles of hippocampal place cells. Hippocampus 7:613–623. Thorndyke PW, Hayes-Roth B (1982) Difference in spatial knowledge acquired from maps and navigation. Cogn Psychol 14:560–589. Watanabe T, Niki H (1985) Hippocampal unit activity and delayed response in the monkey. Brain Res 325:241–254. Wiener SI, Paul CA, Eichenbaum H (1989) Spatial and behavioral correlates of hippocampal neuronal activity. J Neurosci 9:2737–2763. Wilson MA, McNaughton BL (1993) Dynamics of the hippocampal ensemble code for space. Science 261: 1055–1058. Wilson MA, McNaughton BL (1994) Reactivation of hippocampal ensemble memories during sleep. Science 265:676–679. Winocur G, Rawlins JNP, Gray JA (1987) The hippocampus and conditioning to contextual cues. Behav Neurosci 101:617–625. Young BJ, Fox GD, Eichenbaum H (1994) Correlates of hippocampal complex-spike cell activity in rats performing a non-spatial radial maze task. J Neurosci 14: 6553–6563. Young FW (1987) Multidimentional scaling. History, Theory, and Applications. Hillsdale, NJ:Lawrence Erlbaum Associates.
13 Spatial View Cells in the Primate Hippocampus, and Memory EDMUND T. ROLLS
Hippocampal spatial view neurons in primates provide allocentric representations of a view of space ‘‘out there.’’ The responses depend on where the monkey is looking, and can be updated by idiothetic (self-motion) inputs provided by eye movements when the view is hidden. In a room-based object–place memory task, some hippocampal neurons respond to the objects shown, some to the places viewed, and some to combinations of the places viewed and the objects present in those locations. In an object–place recall task, when the location in space at which an object has been seen is recalled by the presentation of the object, some primate hippocampal neurons maintain their responding to the object recall cue in a delay period without the object being visible while the place is recalled; other neurons respond to the place being recalled. Other spatial view neurons form associations with the rewards present at particular locations in space. These findings and computational models of the hippocampus help to show how the primate, including human, hippocampus provides spatial representations of places being looked at that are involved in memory functions such as the memory of where an object or reward has been seen and of recalling this information. The aims of this chapter are to consider how space is represented in the primate hippocampus, how this is related to the memory and spatial functions performed by the hippocampus, and how the hippocampus performs these functions. The neurophysiological studies described were performed (unless stated otherwise) with macaque monkeys to provide information as relevant as possible to understanding memory and spatial
systems in humans. Given the great development of vision in primates relative to that of rodents, and with it the temporal cortical visual areas concerned with vision (which provide many inputs to the hippocampus via, for example, the perirhinal cortex), it is important to investigate whether primate hippocampal processing of space is identical to that of rats, in which place cells are found (McNaughton et al., 1983; O’Keefe, 1984; Muller et al., 1991; Markus et al., 1995). Because of the developments of the primate brain, some of the connections received by the primate hippocampus are reviewed here, as they are relevant to understanding the types of neuron found in the primate hippocampus. The primate hippocampus receives inputs via the entorhinal cortex (area 28) and the highly developed parahippocampal gyrus (areas TF and TH) as well as the perirhinal cortex from the ends of many processing streams of the cerebral association cortex, including the visual and auditory temporal-lobe association cortical areas, the prefrontal cortex, and the parietal cortex (Van Hoesen, 1982; Amaral, 1987; Amaral et al., 1992; Suzuki and Amaral, 1994b) (see Fig. 13–1). The hippocampus is thus through its connections potentially able to associate together object and spatial representations. In addition, the entorhinal cortex receives inputs from the amygdala and the orbitofrontal cortex, which could provide reward-related information to the hippocampus (Suzuki and Amaral, 1994a; Carmichael and Price, 1995; Stefanacci et al., 1996; Pitkanen et al., 2002). There are also direct subcortical inputs from the amygdala and septum (Amaral, 1986). The hippocampus in turn projects
192
SPATIAL VIEW CELLS AND MEMORY
193
Figure 13–1. Forward connections (solid lines) from areas of cerebral association neocortex via the parahippocampal gyrus and perirhinal cortex and then entorhinal cortex to the hippocampus, and backprojections (dashed lines) via the hippocampal CA1 pyramidal cells, subiculum, and parahippocampal gyrus to the neocortex. There is great convergence in the forward connections down to the single network implemented in the CA3 pyramidal cells, and great divergence again in the backprojections. Left: Block diagram. Right: More detailed representation of some of the principal excitatory neurons in the pathways. Abbreviations: D, deep pyramidal cells; DG, dentate granule cells; F, forward inputs to areas of the association cortex from preceding cortical areas in the hierarchy; mf, mossy fibers; PHG, parahippocampal gyrus; pp, perforant path; rc, recurrent collateral of the CA3 hippocampal pyramidal cells; S, superficial pyramidal cells; 2, pyramidal cells in layer 2 of the entorhinal cortex; 3, pyramidal cells in layer 3 of the entorhinal cortex; 4, pyramidal cells in layer 4 of the entorhinal cortex. The thick lines above the cell bodies represent the dendrites.
back via the subiculum, entorhinal cortex, and parahippocampal gyrus (area TF-TH) to the cerebral cortical areas from which it receives inputs (Van Hoesen, 1982), as well as to subcortical areas such as the mammillary bodies (see Fig. 13–1). In the studies of neuronal responses in the primate hippocampus described in this chapter, the recordings of neuronal activity were generally made while the hippocampus was performing the functions for which lesion studies have shown it is needed. Lesion studies have shown that damage to the hippocampus or to some of its connections such as the fornix in monkeys produces deficits in learning about the places of objects and about the places where responses should be made. For example, macaques and humans with damage to the hippocampal system or fornix are impaired
in object–place memory tasks in which not only the objects seen but also their location must be remembered (Smith and Milner, 1981; Gaffan and Saunders, 1985; Parkinson et al., 1988; Gaffan, 1994). Posterior parahippocampal lesions in macaques impair even a simple type of object–place learning in which the memory load is just one pair of trial-unique stimuli (Malkova and Mishkin, 2003). Further, neurotoxic lesions that selectively damage the primate hippocampus impair spatial scene memory (Murray et al., 1998). Also, fornix lesions impair conditional left–right discrimination learning in which the visual appearance of an object specifies whether a response is to be made to the left or the right (Rupniak and Gaffan, 1987). A comparable deficit is found in humans (Petrides, 1985). Fornix-sectioned monkeys are also
194 PRIMATE HIPPOCAMPUS AND PLACE REPRESENTATION impaired in learning on the basis of a spatial cue which object to choose (e.g., if two objects are on the left, choose object A, but if the two objects are on the right, choose object B) (Gaffan and Harrison, 1989b). Monkeys with fornix damage are also impaired in using information about their place in an environment. For example, Gaffan and Harrison (1989a) found learning impairments when the monkey’s choice among two or more objects depended on the position of the monkey in the room. In recordings made in the primate hippocampus under conditions similar to those in which place cells would be found in rats, we have not found neurons that respond in relation to the place where the monkey is. Instead, we have found spatial view cells, which may be thought of as responding to the place at which the monkey is looking. Because these neurons are in some sense concerned with place, their properties are described in this chapter. The way in which these cells were discovered, and some of the tasks in which they respond, are described below.
MEMORY FOR THE POSITIONS OF RESPONSES AND FOR PLACES OF STIMULI IN MEMORY TASKS Watanabe and Niki (1985) analyzed hippocampal neuronal activity while monkeys performed a delayed spatial-response task. In a delayed spatial-response task, a stimulus is shown on, for example, the left; there is then a delay period, and after this the monkey can respond, by touching the left stimulus position. The authors reported that 6.4% of hippocampal neurons responded differently while the monkey was remembering left as compared to right. The responses of these neurons could reflect preparation for the spatial response to be made, or they could reflect memory of the spatial position in which the stimulus was shown. To provide evidence on which element was important, Cahusac, Miyashita, and Rolls (1989) analyzed hippocampal activity in this task and in an object–place memory task. In the object–place memory task, the monkey was shown a sample stimulus in one position on a video screen, there was a delay of 2 s, and then the same or a different stimulus was shown in the same or in a different position. The monkey remembered the sample and its position, and if both matched the delayed stimulus, he licked to obtain fruit juice. Of the 600 neurons analyzed in this task, 3.8% responded differently for the different spatial positions, with some of these responding differentially during the sample presentation, some in the delay period, and some in the match period. Thus some hippocampal neurons (those differentially active in the sample or
match periods) respond differently for stimuli shown in different positions in space, and some (those differentially active in the delay period) respond differently when the monkey is remembering different positions in space. In addition, some of the neurons responded to a combination of object and place information, in that they responded only to a novel object in a particular place. These neuronal responses were not due to any response being made or prepared by the monkey, because information about the behavioral response required was not available until the match stimulus was shown. Cahusac et al. (1989) also found that most of the neurons that responded in the object– place memory task did not respond in the delayed spatial-response task. Instead, a different population of neurons (5.7% of the total) responded in the delayed spatial-response task, with differential left–right responses in the sample, delay, or match periods. Thus, this latter population of hippocampal neurons had activity that was related to the preparation for or initiation of a spatial response, which in the delayedresponse task could be encoded as soon as the sample stimulus was seen. These recordings show that there are some neurons in the primate hippocampus with activity related to the spatial position of stimuli or to the memory of the spatial position of stimuli (as shown in the object– place memory task). The recordings also showed that information about which visual stimulus was shown and where it was shown was combined onto some neurons in the primate hippocampus.
OBJECT–PLACE MEMORY TASKS The response of hippocampal neurons in primates with activity related to the place in which a stimulus is shown was further investigated using a serial multiple object–place memory task. The task required a memory for the position on a video monitor in which a given object had appeared previously (Rolls et al., 1989). This task was designed to allow a wider area of space to be tested than in the previous study, and was chosen because memory of where objects had been seen previously in space was known to be disrupted by hippocampal damage (Gaffan, 1987, 1994). In the task, a visual image appeared in one of four or nine positions on a screen. If the stimulus had been seen in that position before, the monkey could lick to obtain fruit juice, but if the image had not appeared in that position before, the monkey had to learn not to lick, to avoid the taste of saline. Each image appeared in each position on the screen only twice, once as novel and once as familiar. The task thus required memory not only of which visual stimuli had been seen before but
SPATIAL VIEW CELLS AND MEMORY
also of the positions in which they had been seen—an object–place memory task. It was found that 9% of neurons recorded in the hippocampus and parahippocampal gyrus had spatial fields on this and related tasks, in that they responded whenever there was a stimulus in some but not in other positions on the screen. Also, 2.4% of the neurons responded to a combination of spatial information and information about the object seen, in that they responded more the first time a particular image was seen in any position. Six of these neurons were found to show this combination even more clearly—they responded only to some positions, and only provided that it was the first time that a particular stimulus had appeared there. Thus, not only is spatial information processed by the primate hippocampus, it can also be combined, as shown by the responses of single neurons with information about which stimuli had been seen before (Rolls et al., 1989). The ability of the hippocampus to form such arbitrary associations of information probably originating from the parietal cortex about position in space with information originating from the temporal lobe about objects may be important for its role in memory. Moreover, these findings provide neurophysiological support for the computational theory according to which arbitrary associations are formed onto single neurons in the hippocampus between signals originating in different parts of the cerebral cortex, e.g., about objects and about position in space (Treves and Rolls, 1994; Rolls, 1996c; Rolls and Treves, 1998).
AN ALLOCENTRIC REPRESENTATION OF SPACE IN THE PRIMATE HIPPOCAMPUS These studies show that some hippocampal neurons in primates have spatial fields. To study how space is represented in the hippocampus, Feigenbaum and Rolls (1991) investigated whether the spatial fields use egocentric or some form of allocentric coordinates, by finding a neuron with a space field, and then moving the monitor screen and the monkey relative to each other and to different positions in the laboratory. For 10% of the spatial neurons, the responses remained in the same position relative to the monkey’s body axis when the screen was moved or the monkey was rotated or moved to a different position in the laboratory. These neurons thus represented space in egocentric coordinates. For 46% of the spatial neurons analyzed, the responses remained in the same position on the screen or in the room when the monkey was rotated or moved to a different position in the laboratory. These neurons thus represented space in allocentric coordinates. Evidence for two types of allocentric encoding
195
was found. In the first type, the field was defined by its position on the monitor screen independent of the position of the monitor relative to the monkey’s body axis and independent of the position of the monkey and the screen in the laboratory. These neurons were called ‘‘frame of reference’’ allocentric, in that their fields were defined by the local frame provided by the monitor screen. The majority of the allocentric neurons responded in this way. In the second type of allocentric encoding, the field was defined by its position in the room at which they monkey was looking, and was relatively independent of position relative to the monkey’s body axis or to position on the monitor screen face. These neurons were called ‘‘absolute’’ allocentric, in that their fields were defined by position in the room; they are what we have gone on to show subsequently are spatial view neurons. These results show that in addition to neurons with egocentric spatial fields, which have also been found in other parts of the brain such as the parietal cortex (Andersen, 1995), there are neurons in the primate hippocampal formation that encode space in allocentric coordinates.
SPATIAL VIEW NEURONS IN THE PRIMATE HIPPOCAMPUS In rats, place cells are found, which respond according to the place where the rat is in a spatial environment (see McNaughton et al., 1983; O’Keefe, 1984; Muller et al., 1991). In a first investigation to analyze whether such cells might be present in the primate hippocampus, Rolls and O’Mara (1993, 1995) recorded the responses of hippocampal cells when macaques were moved in a small chair or robot on wheels in a cue-controlled testing environment (a 2 2 2 m chamber with matte black internal walls and floors). Tests were performed to determine whether cells might be found that could be described as ‘‘place-related,’’ i.e., firing differently when macaques were moved to different places in this environment, or according to the position in space at which the monkey was looking, or according to his ‘‘head direction.’’ The most common type of cell responded to the part of space at which the monkey was looking, independent of the place where the monkey was. These neurons were called ‘‘view’’ neurons, and are similar to the space neurons described by Rolls et al. (1989) and Feigenbaum and Rolls (1991) (the main difference being that, in the study of Rolls et al. [1989] and Feigenbaum and Rolls [1991], the allocentric representation was defined by the place on a video monitor where a stimulus was shown, whereas spatial view cells respond when the monkey looks at a particular part of a spatial environment). Some of these view neurons had responses
196 PRIMATE HIPPOCAMPUS AND PLACE REPRESENTATION that depended on the proximity of the monkey to what was being viewed. Thus in this study, the neuronal representation of space found in the primate hippocampus was shown to be defined primarily by the view of the environment, and not by the place where the monkey was (Rolls and O’Mara, 1993, 1995). Ono et al. (1993) performed studies on the representation of space in the primate hippocampus while the monkey was moved in a cab to different places in a room. They found that 13.4% of hippocampal formation neurons fired more when the monkey was at some places than when at other places in the test area, and although some neurons responded more when the monkey was at some places than at others, it was not clear whether the responses of these neurons responded to the place where the monkey was independent of spatial view, or whether the responses of place-like cells were view dependent. This critical issue is discussed below, after the properties of spatial view cells have been described further, when tests that can distinguish spatial view cells from place cells will become more clear. In rats, place cells fire best during active locomotion by the rat (Foster et al., 1989). To investigate whether place cells might be present in monkeys while active locomotion is being performed, we (Rolls et al., 1997a, 1998; Robertson et al., 1998; Georges-Franc¸ois et al., 1999) recorded from single hippocampal neurons while monkeys moved themselves around the test environment by walking (or running) on all fours. Also, to bring out the responses of spatial cells in the primate hippocampus, instead of the cue-controlled environment of Rolls and O’Mara (1995), which was matte black, apart from four spatial cues, we used the much richer environment of the open laboratory, within which the monkey had a 2.5 2.5 m area to walk. The position and head direction of the monkey were tracked continuously, and the eye position (i.e., the horizontal and vertical directions of the eyes with respect to the head) was recorded continuously with the scleral search coil technique so that we could measure exactly where the monkey was looking in the environment at all times. An example of a hippocampal pyramidal cell recorded during active locomotion in this environment is shown in Figure 13–2; all the firing that occurred during a period of 6 min when the monkey was walking around the laboratory is shown in the outer rectangles, representing the walls (Fig. 13–2a). The icons of the cart position printed every 250 ms show that a wide area of the laboratory was explored during the period. The cell fired mainly when the monkey was looking at a part of wall 3, as shown in Figures 13–2b and 13–2c; a spot on the walls indicates where the monkey was looking only when the firing rate was above 12 spikes/s, the half-maximal firing rate. The fact that the cell responded when the
monkey was looking at the spatial view field on wall 3 from a large number of different places in the room is shown in Figure 13–2b, in which every tenth cart position and horizontal gaze direction when the cell fired at greater than 12 spikes/s are shown. The range of different cart positions and head directions (which were aligned with the cart direction) over which the cell fired when the cell responded at more than 12 spikes/s are indicated in Figure 13–2c, in which every cart position and head direction for this response rate are shown. Further analyses of the response properties of this cell, including evidence that it responded for a whole set of different head positions, head directions, and eye positions, and that it had similar spatial view fields when the monkey was actively walking and when he was stationary but exploring the environment with eye movements, are provided by Georges-Franc¸ois et al (1999). The responses of another cell in relation to spatial view and not to place, head direction, or eye position per se are shown in Figure 13–3. The firing of the cell as a function of horizontal and vertical eye position is shown (Fig. 13–3, left), as well as the monkey’s stationary position and head direction (Fig. 13–3, right). (The firing rate of the neuron was measured whenever the eyes were stationary to within 18 for more than 250 ms, and data for several minutes of recording are shown.) The highest firing rate of the cell was found when the monkey was looking approximately 108 left and level in the vertical plane. The response field of the cell is plotted against wall 1 in Figure 13–3a (right). The recording time for the data shown here was approximately 4 min. The monkey was then moved to a different place with a different head direction (Fig. 13–3b). Here the highest firing rate was when the monkey was looking approximately 308 right. The response field of the cell is again plotted against wall 1 in Figure 13–3b (right). Data for the monkey at a different place (but the same head direction as in Fig. 13–3b) are shown in Figure 13–3c. Here the cell fired most when the monkey looked approximately 308 left. The response field, however, was at the same place on wall 1 as in Figures 13–3a and 13–3b. It is clear from this type of experiment that it was where the monkey was looking that determined whether the neuron responded, and not a particular head direction, eye position, or place where the monkey was located. This finding was confirmed in one-way analyses of variance (ANOVA), in which the several hundred firing rate and eye position data pairs used to construct Figures 13–3a–c were sorted according to different hypotheses. When the data for level eye position plus and minus 78 (the level of gaze where the cell fired if it was going to) were sorted according to where the monkey was looking on the wall (binned
Figure 13–2. Examples of the firing of a hippocampal cell (az033) when the monkey was walking around the laboratory. a. The firing of the cell is indicated by the spots in the outer set of four rectangles, each of which represents one of the walls of the room. There is one spot on the outer rectangle for each action potential. The base of the wall is toward the center of each rectangle. The positions on the walls fixed during the recording sessions are indicated by points in the inner set of four rectangles, each of which also represents a wall of the room. The central square is a plan view of the room, with a triangle printed every 250 ms to indicate the position of the cart, thus showing that many different places were visited during the recording sessions. b. A similar representation of the same three recording sessions as in a, but modified to indicate some of the range of cart positions and horizontal gaze directions when the cell fired. To enable individual cart/eye gaze direction, icons to be distinguished only every tenth icon were plotted when the cell fired faster than 12 spikes/ sec. A spot was placed in the rectangles whenever the cell fired at greater than 12 spikes/s. c. A similar representation of the same three recording sessions as in b, but modified to indicate more fully the range of cart positions when the cell fired. To enable individual cart/eye gaze direction, icons to be distinguished only every tenth icon were plotted when the cell fired faster than 12 spikes/sec. (A rate of 12 spikes/s was selected, as it was half the peak firing rate of the cell and thus helped to reveal the conditions under which the cell was strongly activated.) The triangles indicate the current position of the monkey, and the lines projecting from them show which part of the wall was being viewed at any one time while the monkey was walking. One spot is shown for each action potential. Adapted from Georges-Franc¸ois, Rolls, and Robertson (1999, Fig. 1).
197
198
SPATIAL VIEW CELLS AND MEMORY
into six wall positions visible in Figs. 13–3a–c), the one-way ANOVA was significant at p < 0.001, and the cell provided an average information (about spatial view) of 0.217 bits in a 500-ms epoch. When the same data were sorted according to eye position (binned into six bins), the one-way ANOVA was not significant ( p < 0.8), and the cell provided an average information (about eye position) of 0.006 bits in a 500-ms epoch. When the same data were sorted according to head direction (binned into two bins), the one-way ANOVA was not significant ( p < 0.5), and the cell provided an average information (about head direction) of 0.0 bits in a 500-ms epoch. When the same data were sorted according to the place where the monkey was located (binned into two bins), the oneway ANOVA was not significant ( p < 0.9), and the cell provided an average information (about place) of 0.001 bits in a 500-ms epoch. This analysis leads to the conclusion that the cell responded significantly differently for different allocentric spatial views and had information about spatial view in its firing rate, but did not respond differently just on the basis of eye position, head direction, or place. Across the population of cells analyzed, it was possible to confirm that it was where the monkey was looking, and not the eye position, head direction, or head position in the room per se, that accounted for the firing of these neurons and about which they conveyed the most information (Georges-Franc¸ois et al., 1999). This series of experiments proved that the representation was not in egocentric spatial coordinates (with respect to the head or body) but was instead allocentric, representing positions in space in world-based coordinates. In further experiments on these spatial view neurons, it was shown that the responses of some neurons reflected quite an abstract representation of space, in that if the visual details of the view were completely obscured by floor-to-ceiling black curtains, many of the neurons could still respond when the monkey looked toward the area where the view had been (Robertson et al., 1998). There was sometimes a slight drift of the spatial view field when the curtains were closed, consistent with the hypothesis that a remembered spatial view is not as accurately located as a seen one and with the observation that the actual view
199
of the scene was the normal determinant of the spatial response field of the cell. The slight drift of the spatial field of the cell is also consistent with evidence from the study by Georges-Franc¸ois et al. (1999) that the coordinate system used by these cells is not in eye position coordinates, nor in a combination of eye position and head direction coordinates, but in allocentric, i.e. world, coordinates. The experiment by Robertson et al. (1998; see also Rolls et al., 1997a) shows that primate hippocampal spatial view neurons can be updated for at least short periods by idiothetic cues including eye position and head direction signals, and that the drift produced by the necessary temporal integration of these signals can then be corrected by showing the visual scene again. The cells that responded with only a small decrease in their response when the view details were obscured (or the room was placed into darkness) were found in the CA1 region, the parahippocampal gyrus and the presubiculum. Other cells had a large decrease (to on average 23%of their normal response) when the monkey looked toward the normally effective location in the environment when the view details were obscured. These cells were in the CA3 region of the hippocampus. This finding provides additional evidence that visual inputs are important in defining the response properties of spatial view neurons (Robertson et al., 1998). This reduction in the firing of the CA3 cells reflects the reduction in the visual sensory drive or recall cue to a CA3 memory system. The results indicate that, for CA3 cells, the visual input is necessary for the normal spatial response of the neurons, and for other cells in the primate hippocampal formation, the response still depends on the monkey gazing toward that location in space when the view details are obscured (Robertson et al., 1998). These latter cells could therefore reflect the operation of a memory system in which the neuronal activity can be triggered by factors that probably include not only (idiothetic) eye position command/feedback signals but also vestibular and/or proprioceptive inputs. The fact that the CA3 neurons continue to fire in the dark and with the view obscured is evidence that there is an attractor (autoassociation) network implemented in the CA3 recurrent collateral system that can be triggered into an
3 Figure 13–3. Examples of the firing of another hippocampal cell (av216) when the monkey was at different positions in the room, with different head directions, looking at wall 1 of the room. The details of the spatial view field are shown by the different firing rates indicated with grayscale levels. The firing rate of the cell in spikes/s as a function of horizontal and vertical eye position is indicated by the blackness of the diagram on the left (with the calibration bar in spikes/s shown below). (Positive values of eye position represent right in the horizontal plane and up in the vertical plane.) The hatched box in the diagram on the right represents the approximate position of the spatial view field. Reprinted with permission from Georges-Franc¸ois, Rolls, and Robertson (1999, Fig. 5.).
200 PRIMATE HIPPOCAMPUS AND PLACE REPRESENTATION attractor state by the appropriate idiothetic signals (Rolls, 1989a,b, 1996c; Treves and Rolls, 1994; Rolls and Treves, 1998). These findings are consistent with partial recovery of information in the CA3 network, which may operate by autoassociation, and further recovery of information in the CA3 to CA1 associatively modifiable synapses as shown by Rolls (1995), who demonstrated this retrieval effect of the Schaffer collaterals in simulations of the hippocampus. Schultz and Rolls (1999) also produced a quantitative analysis of this effect. Another factor that could contribute to the better responses of CA1 cells when the spatial view is obscured is the direct perforant path input to the CA1 cells, which may provide additional input to the CA1 cells (see Rolls and Treves, 1998). The spatial view field of these cells typically occupies a part of space about as large as 1/16 of all the four walls of the testing room. Each cell has a different view to which it responds. Thus over a population of many such neurons, these partly overlapping view fields represent rather precise information about the part of space being viewed. This has been quantified using information theory; indeed, it has been shown that the amount of information about which part of space the monkey is viewing increases approximately linearly with the number of neurons in the sample. Thus an independent contribution is made by each of the cells in an ensemble to represent allocentric space (Rolls et al., 1998). Because information is a logarithmic measure, this means that the number of spatial views (or the accuracy of the representation) increases exponentially with the number of neurons in the ensemble, a powerful result. Moreover, it appears that most of this information is contained in the number of spikes that each neuron produces within a short time window, and not in the relative time of firing of the spikes of different neurons (Panzeri et al., 1999). Many spatial view (or ‘‘space’’ or ‘‘view’’) cells have been found in this series of experiments (Rolls et al., 1997a, 1998; Robertson et al., 1998; GeorgesFranc¸ois et al., 1999). (The number of spatial view cells in the initial sample of 352 cells recorded under these conditions is 40; see Rolls et al. [1997a]. It is simply noted here that their average spontaneous rate is low, with a mean 0.5 spikes/s, and that their average peak firing rate is 17 spikes/s, with an interquartile range of 11–20 spikes/s. This low spontaneous rate and low peak response rate is similar to that of place cells in rats.) No place cells have been found that responded on the basis of where the monkey was, and not where he was looking in the environment. Although Ono et al. (1993; see also Matsumura et al., 1999) have described macaque cells whose firing rate depended on the location of the macaque, I note that very extensive testing with formal contrasts of different hypotheses
performed along the lines described by GeorgesFranc¸ois et al. (1999) is in general needed to show whether a cell in the primate hippocampus responds to the place where the monkey is rather than to spatial view. For example, given that a region of allocentric space in a room that defines the spatial view field of a spatial view cell will not be visible from all places in a room, it is not sufficient to show that the firing rate depends on the place where the monkey is, because the spatial view does as well. Another example is shown in Figure 2 of Rolls (1996a,b), showing a cell (av057) that might have been interpreted as a place cell if testing with different head directions of the monkey allowing different spatial views had not been performed. It is essential to measure the firing rate of a primate hippocampal cell with different head directions so that different spatial views can be compared, as testing with just one head direction (Matsumura et al., 1999) cannot provide evidence that will distinguish a place cell from a spatial view cell. These points will need to be borne in mind in future studies of hippocampal neuronal activity in primates including humans (cf. Fried et al., 1997; Kreiman et al., 2000; Ekstrom et al., 2003), and simultaneous recording of head position, head direction, and eye position as described in this chapter will be needed. To distinguish spatial view from place cells it will be important to test neurons while the primate or human is in one place with all the different spatial views visible from that place. The same neuron should also be tested when the organism is in a different place but at least some of the same spatial views are visible, as has been done in our primate recording. It is also necessary to test primate hippocampal cells during active locomotion in case this is an important factor as in the rat. Having said this, we have found that spatial view cells in the primate hippocampus have similar responses during active locomotion to those when the monkey is stationary but is allowed to look around and actively explore the environment with eye movements. Indeed, it is possible that this active exploration of an environment by eye movements is somewhat analogous to the active exploration that a rat does by running around. The actual recording system we use allows the monkey to be very active when he is moving on all four legs, in that the chair on wheels is attached only to his head and allows head angular velocities as large as 1008/s and linear motion of 0.6 m/s, so it is unlikely that this setup would not find place cells in the primate hippocampus, if they are present. Having said this, we do not have a strong position on this issue, but simply note that thus far we have not observed place cells in the primate hippocampus and great care is needed to show the presence of place cells. We instead draw attention to a remarkable new
SPATIAL VIEW CELLS AND MEMORY
type of cell, spatial view cells, which in primates respond to places ‘‘out there’’ and are well suited to participating in the memory for where objects have been seen in an environment. Hippocampal spatial view cells are very different from inferior temporal cortex neurons that respond to objects or faces wherever they are moved to in an environment (Rolls and Deco, 2002; Rolls et al., 2003), in that it is normally the combination of a set of features in a fixed position in the world that activates spatial view neurons. In a sense, an object can be defined as a set of co-occurring features that can be easily moved to different locations in an environment, whereas a place is defined by a set of co-occurring features that remain together in the same location in the world and are not normally seen to move independently of the rest of the world. Spatial view cells are also very different from head direction cells, which are found in the primate presubiculum and parahippocampal gyrus (Robertson et al., 1999). For example, for a given head direction, if the monkey is moved to different places in the environment where the spatial view is different, spatial view cells give different responses. In contrast, the response of head direction cells remains constant for a given head direction, even when the spatial view is very different (Robertson et al., 1999). To provide a simple concept to emphasize the difference, one can think of head direction cells as responding like a compass attached to the top of the head, which will signal head direction even when the compass is in different locations, including in a totally different, and even novel, spatial environment.
OBJECT–PLACE NEURONS IN THE PRIMATE HIPPOCAMPUS A fundamental question about the function of the primate, including human, hippocampus is whether object and allocentric spatial information is represented. To investigate this, Rolls, Xiang, and Franco (2005) made recordings from single hippocampal formation neurons while macaques performed an object–place memory task that required the monkeys to learn associations between objects and where they were shown in a room. Some neurons (10%) responded differently to different objects independent of location; other neurons (13%) responded to the spatial view independent of which object was present at the location; and some neurons (12%) responded to a combination of a particular object and the place where it was shown in the room. These results show that there are separate and combined representations of objects and their locations in space in the primate hippocampus. This property is necessary in an episodic memory system for which associations
201
between objects and the places where they are seen is prototypical. Thus a requirement for a human episodic memory system, separate and combined neuronal representations of objects and of where they are seen ‘‘out there’’ in the environment, is present in the primate hippocampus (Rolls et al., 2005).
RECALL-RELATED NEURONS IN THE PRIMATE HIPPOCAMPUS Having described the discovery of spatial view neurons in the primate hippocampus and neurons involved in associations between spatial view and the objects present at the places viewed (Rolls et al., 2005) or the rewards present at the places viewed (Rolls and Xiang, 2005), a new investigation of how hippocampal neuronal activity is related to the recall of memories is now described (Rolls and Xiang, 2006). Although we have a full and quantitative theory of how the hippocampus and its backprojection pathways to the neocortex are involved in the recall of previously stored episodic memories from just a fragment of the original memory (Rolls, 1989b, 1996c, 2008; Treves and Rolls, 1994; Rolls and Treves, 1998; Rolls and Kesner, 2006), this is the first neurophysiological investigation of the hippocampal recall process in primates (Rolls and Xiang, 2006). We used an object–place memory task because this is prototypical of episodic memory, and there is evidence that the primate hippocampal system is required for this type of memory. (Posterior parahippocampal lesions in macaques impair even a simple type of object–place learning in which the memory load is just one pair of trial-unique stimuli [Malkova and Mishkin, 2003]; further, it has been shown that a onetrial odor–place recall memory task is hippocampal dependent in rodents [Day et al., 2003].) We designed a one-trial object–place recall task in which the whole memory was recalled from a part of it (Fig. 13–4). Images of new objects were used each day, and within a day the same objects were used, so that with nontrial-unique objects within a day, the recall task is quite difficult. Recordings were made from 347 neurons in the hippocampus of a macaque performing the object– place recall task. The following types of neurons were found in the task (Rolls and Xiang, 2006). One type of neuron had responses that occurred to one of the objects used in the task. A number of these neurons had activity that was related to the recall process. An example of one of these neurons is shown in Figure 13–5. The neuron had activity that was greater to object 1 (O1) not only when it was shown in stages 1, 2, and 3 of the task but also in the delay period
202 PRIMATE HIPPOCAMPUS AND PLACE REPRESENTATION
Figure 13–4. The object–place recall task. One trial is shown. After a 0.5-s tone to indicate the start of a trial, in stage 1, one of two objects (O1) is shown at one of the places (P1). (The object and the place are chosen randomly on each trial.) To ensure that the monkey sees the stimulus, the monkey can touch the screen at the place to obtain one drop of juice reward by licking. After a 0.5-s delay, in stage 2, the other of the two objects (O2) is shown at the other place (P2). (One drop of fruit juice was available as in stage 1.) After a 0.5-s delay, in stage 3, the recall cue, one of the objects chosen at random, is shown at the top of the screen in the middle. (One drop of fruit juice was available as in stage 1.) After a 0.5-s delay, in stage 4, the macaque must then recall the place where the object shown as the recall cue in stage 3 was presented, and must then touch that place on the screen to obtain four licks of fruit juice, thus indicating that he has recalled the location correctly. In stage 4 of the trials, the left and right positions (P1 and P2) have no image present, with the two possible locations for a response indicated by identical circles. The task requires the monkey to perform recall of the place from the object, within the period beginning at the presentation of the recall cue at the start of stage 3 and ending when the response is made in stage 4. following stage 3 when the object was no longer visible and in stage 4, when the object was no longer visible and the macaque was touching the remembered location of that object. Thus while the location was being recalled from the object, this type of neuron continued to respond as if the object were present— that is, it kept the representation of the object active after the object was no longer visible, and the place to touch was being recalled. Sixteen of the neurons responded in this way, and an additional six had objectrelated firing that did not continue following stage 3 of the task in the recall period. The difference in firing rates of these 22 neurons to the different objects was in many cases highly statistically significant (e.g., p < 10–6). We performed a Fisher’s exact probability test to confirm that the set of statistically significant results in the 22 neurons could not have arise by chance within the 347 tests performed and were able to reject this with p < 5.4 10–8. Thus the population of 22 neurons had statistically very highly significance in its objectrelated responses. None of these neurons had differential responses for the different places used in the object–place recall task.
A second type of neuron had responses related to the place (left or right) in which an object was shown in stages 1 or 2 of each trial. An example of one of these neurons is shown in Figure 13–6. The neuron responded more when an object was shown in the left position (P1) than in the right position (P2) on the screen. Interestingly, when the recall object was shown in stage 3 of the trial in the top center of the screen, the neuron also responded as if the left position (P1) were being processed on trials in which the left position had to be recalled. This firing continued in the delay period after the recall cue had been removed at the end of stage 3 and into stage 4. Thus this type of neuron appeared to reflect the recall of the position on the screen at which the object had been represented. Analysis of trials on which errors were made indicated that the responses were not just motor response related, for if due to some response bias the monkey touched the incorrect side, the neuron could still respond according to the correct recalled location. Thirteen neurons had differential responses to the different places P1 and P2 and continued to show place-related activity in the recall part of the task,
SPATIAL VIEW CELLS AND MEMORY
203
Figure 13–5. Activity of a neuron with responses related to one of the objects used in the object–place recall task. The firing rates to object 1 (O1) and object 2 (O2) are shown (mean firing rate in spikes/s across trials ± SEM). The first histogram pair (on the left) shows the responses to the two objects measured throughout the trial whenever object 1 or object 2 was on the screen. The second histogram pair shows the neuronal responses when the objects were being shown in stage 3 as the recall cue. The third histogram pair shows the neuronal responses in the 0.5-s delay period after one of the objects had been shown in stage 3 as the recall cue. The neuron continued to respond more after object 1 than after object 2 had been seen, in this recall period in which the place was being recalled from the object. The fourth histogram pair shows the neuronal responses in stage 4 when the macaque was recalling and touching the place at which the cue recall object had been shown. The responses of the neuron were object related even when the object was not being seen, but was being used as a recall cue, in the delay after stage 3 of the task and in stage 4. ** p < 0.01; * p < 0.05.
stage 3. Five other neurons had left–right place-related responses without a memory recall component, in that they did not respond in stage 3 of the task, when a nonspatial recall stimulus was being shown and a place should be recalled (see Table 13–1). We performed a Fisher’s exact probability test to confirm that the set of statistically significant results in the 18 neurons could not have arisen by chance within the 347 tests performed and were able to reject this with p < 0.05. Thus the population of 18 neurons as a population had statistically significant place-related responses. The new finding is that 13 of the neurons had place-related responses when a place was being recalled by an object cue. The responses of the population of neurons recorded in one macaque are shown in Table 13–1. In addition to the neurons described above, three further neurons responded to particular combinations of ob-
jects and places, e.g., to object 1 when it was shown in place 1, but not to other combinations. The recording sites of the object and of the place neurons are shown in Figure 13–7. All the neurons were within the hippocampus proper. The mean firing rate of the population of responsive neurons (see Table 13–1) to the most effective object or place was 7.2 ± 0.6 spikes/s (±SEM), and their mean spontaneous rate was 3.2 ± 0.6 spikes/s. These findings (Rolls and Xiang, 2006) are the first we know in the primate hippocampus of neuronal activity related to recall. It is particularly interesting that the neurons with continuing activity to the object after it had disappeared in the recall phase of the task could reflect the operation of the object–place recall process hypothesized to take place in the CA3 cells. By continuing to respond to the object while the place is being recalled in the task, the object-related neurons
204 PRIMATE HIPPOCAMPUS AND PLACE REPRESENTATION
Figure 13–6. Activity of a neuron with responses related to the left place (P1) in the object–place recall task. The firing rate to place 1 (P1) and place 2 (P2) are shown (mean firing rate in spikes/s across trials ± SEM). The first histogram pair (on the left) shows the responses to the two places measured when a stimulus was on the screen in stage 1 or stage 2. The second histogram pair shows the neuronal responses when the objects were being shown in stage 3 as the recall cue, and depending on whether the place to be recalled was place 1 or place 2. The third histogram pair shows the neuronal responses in the 0.5-s delay period after one of the objects had been shown in stage 3 as the recall cue. The neuron responded more when place 1 was the correct place to be recalled on a trial. The fourth histogram pair shows the neuronal responses in stage 4 when the macaque was recalling and touching the place at which the cue recall object had been shown. The responses of the neuron were place related even in stage 3 when the object being shown as a place recall cue was at the top of the screen, in the delay after stage 3 of the task and in stage 4. ** p < 0.01; * p < 0.05.
could be part of the completion of the whole object– place combination memory from an autoassociation or attractor process in CA3 (Rolls and Kesner, 2006). The neurons with recall-related activity in the object–place recall task also provide neurophysiological evidence on the speed of association learning in the hippocampal formation. Given that this is a one-trial object–place recall task, with the association between the object and its place being made in stages 1 and 2 of each trial (see Fig. 13–4), it is clear that it takes just one trial for the object–place associations to be formed that are relevant to the later recall on that trial. This is the speed of learning required for episodic memory, and this neurophysiological evidence shows that this type of rapid, one-trial, object–place learning is represented in the primate hippocampus.
REWARD–PLACE NEURONS IN THE PRIMATE HIPPOCAMPUS It is suggested that whenever memories are stored, part of the context is stored with the memory. This process very likely happens in associative neuronal networks such as those in the hippocampus (Rolls, 1989b, 1990, 1996c, 2000, 2004; Treves and Rolls, 1994; Rolls and Treves, 1998; Rolls et al., 2002). The CA3 part of the hippocampus may operate as a single autoassociative memory capable of linking together almost arbitrary co-occurrences of inputs, including inputs about emotional state that reach the entorhinal cortex from, for example, the amygdala and orbitofrontal cortex (Rolls, 2005). Recall of a memory occurs best in such networks when the input key to the memory is nearest to
SPATIAL VIEW CELLS AND MEMORY
Table 13–1. Numbers of Neurons in the Hippocampus with Different Types of Response during the Object–Place Recall Task Response Object with activity continuing afterthe recall cue Object with activity not continuing after the recall cue Place with activity during and after the recall cue Place with activity duringthe recall cue Object–place Total
No. 16 6 13 5 3 347
205
the original input pattern of activity that was stored (Rolls and Treves, 1990, 1998; Treves and Rolls, 1991, 1992, 1994; Rolls et al., 1997b; Rolls and Deco, 2002). It thus follows that a memory of, for example, a happy episode is recalled best when one is in a happy mood state. This is a special case of a general theory of how context is stored with a memory and of how context influences recall (Treves and Rolls, 1994; Rolls, 1996c; Rolls and Stringer, 2001). The recall itself from the hippocampus is likely to use the highly developed backprojections from the hippocampus to the neocortex shown in Figure 1 of Treves and Rolls (1994). The effect of emotional state on cognitive processing and memory is thus suggested to be a particular case of a more general way in which context can affect the storage and retrieval of memories or can affect cognitive processing (Rolls, 2005).
Figure 13–7. The recording sites of the different neuron types in the object–place recall task are shown. Amyg, amygdala; Hipp, hippocampus; opt, optic tract; Prh, perirhinal cortex; rhs, rhinal sulcus; sts, superior temporal sulcus; TE, inferior temporal visual cortex; Tf, TH, parahippocampal gyrus.
206 PRIMATE HIPPOCAMPUS AND PLACE REPRESENTATION There is now direct evidence that the hippocampus, which is implicated in the memory for past episodes (Rolls and Treves, 1998; Rolls, 1999; Rolls et al., 2002), contains neurons in primates that respond to combinations of spatial information and reward information (Rolls and Xiang, 2005; Rolls et al., 2005), described below. The ability to form associations between events including where they occur and what is present is a fundamental property of episodic memory (Treves and Rolls, 1994; Rolls, 1996c), and this new neurophysiological evidence shows that rewardrelated information, relevant to affect and mood, is associated with other events in representations in the primate hippocampus. The primate anterior hippocampus (which corresponds to the rodent ventral hippocampus) receives inputs from brain regions involved in reward processing such as the amygdala and orbitofrontal cortex (Suzuki and Amaral, 1994a; Carmichael and Price, 1995; Stefanacci et al., 1996; Pitkanen et al., 2002). To investigate how this affective input may be incorporated into primate hippocampal function, Rolls and Xiang (2005) recorded neuronal activity while macaques performed a reward–place association task in which each spatial scene shown on a video monitor had one location that, if touched, yielded a preferred fruit juice reward, and a second location that yielded a less-preferred juice reward. Each scene had different locations for the different rewards. Of 312 hippocampal neurons analyzed, 18% responded more to the location of the preferred reward in different scenes, and 5% to the location of the less-preferred reward (Rolls and Xiang, 2005). When the locations of the preferred rewards in the scenes were reversed, 60% of 44 neurons tested reversed the location to which they responded, showing that the reward–place associations could be altered by new learning in a few trials. The majority (82%) of these 44 hippocampal reward–place neurons tested did not respond to object–reward associations in a visual discrimination object–reward association task. Thus the primate hippocampus contains a representation of the reward associations of places ‘‘out there’’ being viewed; this is a way in which affective information can be stored as part of an episodic memory and how the current mood state may influence the retrieval of episodic memories. There is consistent evidence that rewards available in a spatial environment can influence the responsiveness of rodent place neurons (Ho¨lscher et al., 2003; Tabuchi et al., 2003) that respond to the place where the animal is located, not to the view of a place ‘‘out there’’ (Rolls, 1999; de Araujo et al., 2001). Thus the primate hippocampus can combine by associative learning a representation of places ‘‘out there’’ not only with which object is present at the
viewed location (Rolls et al., 2005) but also with which reward is present at the viewed location (Rolls and Xiang, 2005). The general principle here, then, is that the hippocampus may store information about where emotion-related (e.g., rewarding) events happened, may take part in the recall of emotions when particular places are seen again, and may provide a system in which the current mood can influence which memories are recalled. Before discussing the possible functions of primate spatial view cells and their relation to rat place cells, it is useful to summarize the properties of other cells in the primate hippocampus that are relevant to understanding the representation of space by the primate hippocampus.
NEURONS RELATED TO LEARNING ASSOCIATIONS BETWEEN VISUAL STIMULI AND SPATIAL RESPONSES In another type of task for which the primate hippocampus is needed, conditional spatial response learning, in which the monkeys have to learn which spatial response to make to different stimuli, to acquire associations between visual stimuli and spatial responses, 14% of hippocampal neurons responded to particular combinations of visual stimuli and spatial responses (Miyashita et al., 1989). The firing of these neurons could not be accounted for by the motor requirements of the task, nor wholly by the stimulus aspects of the task, as demonstrated by testing their firing in related visual discrimination tasks. These results show that single hippocampal neurons respond to combinations of visual stimuli and the spatial responses with which they must become associated in conditional response tasks and are consistent with the computational theory described above in which part of the mechanism of this learning involves associations between visual stimuli and spatial responses learned by single hippocampal neurons. In a subsequent study by Cahusac et al. (1993), during such conditional spatial response learning 22% of this type of neuron analyzed in the hippocampus and parahippocampal gyrus altered their responses so that their activity, which was initially equal to the two new stimuli, became progressively differential to the two stimuli when the monkey learned to make different responses to the two stimuli (Cahusac et al., 1993). These changes occurred for different neurons just before, at, or just after the time when the monkey learned the correct response to make to the stimuli, and are consistent with the hypothesis that when new associations between objects and places (in this case the places for responses) are learned, some hippo-
SPATIAL VIEW CELLS AND MEMORY
campal neurons learn to respond to the new associations required to solve the task. Similar findings have been described by Wirth et al. (2003).
RESPONSES OF NEURONS IN THE PRIMATE HIPPOCAMPUS TO WHOLE-BODY MOTION Another type of cell found in the primate hippocampus responds to whole-body motion (O’Mara et al., 1994), an idiothetic cue. For example, such cells respond when the monkey is rotated about the vertical axis, with a much larger response for clockwise than for counterclockwise rotation. By occluding the visual field, it was possible to show that in some cases the response of these cells required visual input. For other cells, visual input was not required, and it is likely that such cells responded on the basis of vestibular inputs. Some cells were found that responded to a combination of body motion and view or place. In some cases these neurons responded to linear motion, in others to axial rotation (n ¼ 43). In some cases these neurons required visual input for their responses, in other cases the neurons appeared to be driven by vestibular inputs. Some cells responded to a combination of movement together with either a particular local view seen by the monkey (n ¼ 2) or a particular place toward which the monkey was moving (n ¼ 1). These (idiothetic) whole-body motion cells may be useful in a memory system for remembering trajectories through environments, for example, in short-range spatial navigation and path integration (O’Mara et al., 1994).
207
PRIMATE PRESUBICULAR HEAD DIRECTION CELLS Rat head direction cells have a firing rate that is a simple function of head direction in the horizontal plane (see Muller et al., 1996; Taube et al., 1996). The firing does not depend on the place where the rat is located. The cells in the rat are found in the dorsal presubiculum (also referred to as the postsubiculum), as well as in some other brain structures including the anterior thalamic nuclei (Taube et al., 1996). We (Robertson et al., 1999) have analyzed a similar population of head direction cells in primates and shown that they are place independent, can be idiothetically updated in the dark, and encode information about head direction that is independent for different neurons (up to several neurons).
CONTINUOUS AND DISCRETE ATTRACTOR NETWORKS AND EPISODIC MEMORY Space is continuous, and object representations are discrete. If these representations are to be combined in an object–place memory, then we need to understand the operation of networks that combine these representations. A class of network that can maintain the firing of its neurons to represent any location along a continuous physical dimension such as spatial position, head direction, etc., is a ‘‘continuous attractor neural network’’ (CANN) (see references provided below and Chapter 7 in Rolls and Deco, 2002). It uses
Figure 13–8. Architecture of a continuous attractor neural network (CANN). Recurrent collateral axons with associatively modifiable synaptic connections make contact with the excitatory pyramidal cells in the network. The vertical lines are the dendrites, the cell bodies are triangles, and the axons extend out of the bottom of each cell body. The synaptic weight or strength for axon j to the dendrite of neuron i is wij. The external firing rate input to the network is conveyed by axons ei. Feedback inhibitory interneurons are not shown. For details see Rolls et al. (2002) and Rolls and Deco (2002).
208 PRIMATE HIPPOCAMPUS AND PLACE REPRESENTATION excitatory recurrent collateral connections with associative modifiability between the neurons to reflect the distance between the neurons in the state-space of the animal (e.g., head direction space). These networks can maintain the packet or bubble of neural activity constant for long periods, wherever it is started, to represent the current state (head direction, position, etc.) of the animal, and are likely to be involved in many aspects of spatial processing and memory, including spatial vision. Global inhibition (implemented by feedback inhibitory interneurons) is used to keep the number of neurons in a bubble or packet of actively firing neurons relatively constant, and to help to ensure that there is only one activity packet. Continuous attractor networks can be thought of as being very similar to autoassociation or discrete attractor networks (Rolls and Treves, 1998; Rolls and Deco, 2002; Rolls, 2008) and have the same architecture, as illustrated in Figure 13–8. The main difference is that the patterns stored in a CANN are continuous patterns, with each neuron having broadly tuned firing that decreases with, for example, a Gaussian function as the distance from the optimal firing location of the cell is varied, and with different neurons having tuning that overlaps throughout the space. Such tuning is illustrated in Figure 13–9, together with examples of discrete (separate) patterns (each pattern implemented by the firing of a particular subset of the neurons), with no continuous distribution of the patterns throughout the space, that are useful for storing arbitrary events or objects. A consequent difference is that the CANN can maintain its firing at any location in the trained continuous space, whereas a discrete attractor or autoassociation network moves its population of active neurons toward one of the previously learned attractor states and thus implements the recall of a particular previously learned pattern from an incomplete or noisy (distorted) version of one of the previously learned patterns. It has been shown that attractor networks can store both continuous patterns and discrete patterns and can thus be used to store the location in (continuous, physical) space (e.g., the place ‘‘out there’’ in a room represented by spatial view cells) where an object (a discrete item) is present (Rolls et al., 2002; cf. Rolls, 1989b, 1996c). Such associations between an object and the place where it is located are prototypical of episodic or event memory and may be implemented in the primate hippocampus (Rolls et al., 2005). In this network, when events are stored that have both discrete (object) and continuous (spatial) aspects, the whole place can be retrieved later by the object, and the object can be retrieved by using the place as a retrieval cue. Such networks are likely to be present in parts of the brain such as the hippocampus, which receive and combine inputs both from systems that
contain representations of continuous (physical) space, and from brain systems that contain representations of discrete objects, such as the inferior temporal visual cortex. The combined continuous and discrete attractor network described by Rolls et al. (2002) shows that in brain regions where the spatial and object processing streams are brought together, a single network can represent and learn associations between both types of input. Indeed, in brain regions such as the hippocampal system, it is essential that the spatial and object processing streams are brought together in a single network, for it is only when both types of information are in the same network that spatial information can be retrieved from object information and vice versa, a fundamental property of episodic memory (Rolls and Treves, 1998; Rolls and Deco, 2002; Rolls, 2008).
CONTINUOUS ATTRACTOR NETWORKS AND PATH INTEGRATION We have considered how spatial representations could be stored in continuous attractor networks and how the activity can be maintained at any location in the statespace in a form of short-term memory when the external (e.g., visual) input is removed (Rolls and Deco, 2002). However, many networks with spatial representations in the brain can be updated by internal, selfmotion (i.e., idiothetic), cues even when there is no external (e.g., visual) input. Examples are head direction cells in the presubiculum of rats and macaques, place cells in the rat hippocampus, and spatial view cells in the primate hippocampus. A major question that arises is how such idiothetic inputs could drive the activity packet in a continuous attractor network and, in particular, how such a system could be set up biologically by self-organizing learning. One approach to simulating the movement of an activity packet produced by idiothetic cues (which is a form of path integration whereby the current location is calculated from recent movements) is to employ a look-up table that stores (taking head direction cells as an example) for every possible head direction and head rotational velocity input generated by the vestibular system the corresponding new head direction (Samsonovich and McNaughton, 1997). Another approach involves modulating the strengths of the recurrent synaptic weights in the continuous attractor on one but not the other side of a currently represented position, so that the stable position of the packet of activity, which requires symmetric connections in different directions from each node, is lost, and the packet moves in the direction of the temporarily increased weights. No possible biological process has been proposed, however, for achieving the appropriate dynamic synaptic weight
SPATIAL VIEW CELLS AND MEMORY
209
Figure 13–9. The types of firing patterns stored in continuous attractor networks for the patterns present on neurons 1–1000 for memory 1 (A, when the firing is that produced when the spatial state represented is that for location 300) and for memory 2 (B, when the firing is that produced when the spatial state represented is that for location 500). The continuous nature of the spatial representation results from the fact that each neuron has a Gaussian firing rate that peaks at its optimal location. This particular mixed network also contains discrete representations that consist of discrete subsets of active binary firing-rate neurons in the range 1001–1500. The firing of these latter neurons can be thought of as representing the discrete events that occur at the location. Continuous attractor networks by definition contain only continuous representations, but this particular network can store mixed continuous and discrete representations and is illustrated to show the difference of the firing patterns normally stored in separate continuous attractor and discrete attractor networks. For this particular mixed network, during learning, memory 1 is stored in the synaptic weights, then memory 2, etc., and each memory contains part that is continuously distributed to represent physical space, and part that represents a discrete event or object. Adapted from Rolls et al. (2002).
changes (Zhang, 1996). Another mechanism (for head direction cells) (Skaggs et al., 1995) relies on a set of cells, termed (head) rotation cells, which are co-activated by head direction cells and vestibular cells and drive the activity of the attractor network by anatomically distinct connections for clockwise and counterclockwise rotation cells, in what is effectively a lookup table. However, no proposal has been made to explain how this could be achieved through a biologically plausible learning process. This has been the case until recently for most approaches to path integration in continuous attractor networks, which rely heavily on rather artificial preset synaptic connectivities.
Stringer and colleagues (2002a) introduced a proposal with more biological plausibility for explaining how the synaptic connections from idiothetic inputs to a continuous attractor network can be learned by a selforganizing learning process. The mechanism associates a short-term memory trace of the firing of neurons in the attractor network reflecting recent movements in the state-space (e.g., of places) with an idiothetic velocity of movement input. This has been applied to head direction cells (Stringer et al., 2002a), rat place cells (Stringer et al., 2002a,b), and primate spatial view cells (Stringer et al., 2004, 2005; Rolls and Stringer, 2005). These attractor networks provide a basis for
210 PRIMATE HIPPOCAMPUS AND PLACE REPRESENTATION understanding cognitive maps and how they are updated by learning and self-motion.
SPATIAL-VIEW NEURONS IN PRIMATES COMPARED TO PLACE CELLS IN RODENTS Primate spatial view cells are unlike place cells found in the rat (O’Keefe, 1979, 1990, 1991; Kubie and Muller, 1991; Wilson and McNaughton, 1993). Primates, with their highly developed visual and eye movement control systems, can explore and remember information about what is present at places in the environment without having to visit those places. Such spatial view cells in primates would thus be useful as part of a memory system, because they would provide representation of a part of space that would not depend on exactly where the monkey or human was and that could be associated with items that might be present in those spatial locations. An example of the utility of such a representation in humans is remembering where a particular person had been seen. Primate spatial representations would also be useful in remembering trajectories through environments of use, for example, in short-range spatial navigation (O’Mara et al., 1994; Rolls and Deco, 2002). The representation of space in the rat hippocampus, which is of the place where the rat is, may be related to the fact that with a much less developed visual system than that of the primate, the rat’s representation of space may be defined more by the olfactory and tactile as well as distant visual cues present, and may thus tend to reflect the place where the rat is. A hypothesis on how this difference could arise from essentially the same computational process in rats and monkeys is described below (Rolls, 1999; de Araujo et al., 2001). The starting assumption is that in both the rat and primate, the dentate granule cells and the CA3 and CA1 pyramidal cells respond to combinations of the inputs received. In the primate, because of the fovea providing high spatial resolution over a typical viewing angle of perhaps 108–208, a combination of visual features in the environment will result in the formation of a spatial view cell, the effective trigger for which will thus be a combination of visual features within a relatively small part of space. In contrast, in the rat, given the very extensive visual field subtended by the rodent retina, which may extend over 1808–2708, a combination of visual features formed over such a wide visual angle would effectively define a position in space that is a place. The actual processes by which the hippocampal formation cells would come to respond to feature combinations could be similar in rats
and monkeys, involving, for example, competitive learning in the dentate granule cells, autoassociation learning in CA3 pyramidal cells, and competitive learning in CA1 pyramidal cells (Treves and Rolls, 1994; Rolls and Treves, 1998). Thus the selective properties of spatial view cells in primates and place cells in rats might arise by the same computational process but be different by virtue of the fact that primates are foveate and view a small part of the visual field at any one time, whereas the rat has a very wide visual field (de Araujo et al., 2001). Although the representation of space in rats may therefore be in some ways analogous to the representation of space in the primate hippocampus, the difference does have implications for theories, and modeling, of hippocampal function. Self-motion, that is, idiothetic, inputs are represented on the rodent medial entorhinal cortex by grid cells, each of which has peaks of firing in a hexagonal grid. Different entorhinal cells have different grid frequencies and phases (Hafting et al., 2005). The connections from the entorhinal cortex to the dentate granule cells are thought to act as a competitive network to produce the sparse representations useful for enabling the hippocampus to store a large number of memories (Rolls, 1989b, 2008; Rolls and Kesner, 2006). Exactly this type of competitive network can account for the mapping of entorhinal cortex grid cells to place cells in the rodent dentate granule and hippocampal CA3 cells (Rolls et al., 2006) (see Fig. 13–10). It will be of interest in future studies to investigate whether there is a similar grid-like representation of spatial view in the primate entorhinal cortex. The utility of such a grid representation is that it could provide a representation useful for path integration based on eye position and head direction of spatial representations of where the animal is looking in the environment, with a similar conversion to spatial view representations being performed by competitive learning in the dentate granule cells. I suggest that the importance of transforming a grid cell representation in the entorhinal cortex into a place cell representation (or in primates a spatial view cell representation) in the hippocampus is as follows. At least one crucial function this allows is the formation of memories, formed, for example, between an object and a place; this is proposed to be a prototypical function of the hippocampus that is fundamental to episodic memory (Rolls, 1989b, 1996c, 2008; Rolls and Treves, 1998; Rolls and Kesner, 2006). For the formation of such object–place memories in an associative network (in particular, an autoassociative network in the CA3 region of the hippocampus) to occur, the place must be made explicit in the representation. Moreover, for high-capacity memory, that is, the
Figure 13–10. Simulation of competitive learning in the dentate gyrus to produce place cells from the entorhinal cortex grid cell inputs. a,b. Firing rate profiles of two entorhinal cortex (EC) grid cell with frequencies of 4 and 7 cycles. c,d. Firing rate profiles of two dentate gyrus (DG) cells after training. Simulation details: The two-dimensional space was 100 100 training and testing locations. There were 125 EC cells with frequencies of 3, 4, 5, 6, and 7 cycles (to capture the frequencies described neurophysiologically (Hafting et al., 2005), and 25 phases for each frequency. (A phase is defined as an offset in the X and Y directions, and five offset values were used in each direction.) There were 100 DG cells, with the sparseness of the representation set to a ¼ 0.02. A standard competitive net was used (Rolls and Deco 2002). Adapted from Rolls, Stringer, and Elliot (2006).
211
212 PRIMATE HIPPOCAMPUS AND PLACE REPRESENTATION ability to store and retrieve many memories, the representation must be sparse. The sparseness can be defined as a¼
X i ¼ 1;n
!2 ri =n
=
X
(ri2 =n)
(1)
i ¼ 1;n
where ri is the firing rate of the i’th neuron in the set of n neurons to a given stimulus or, in this case, place (Rolls and Treves, 1998; Rolls and Deco, 2002; Rolls, 2008). The sparseness takes a low value if few neurons are firing to a given place. Information can be read off easily from the firing rates of the neurons, with different places producing very different sets of neurons firing, so that different representations are relatively orthogonal. The entorhinal cortex representations are not only not sparse (in our simulations, the sparseness was 0.54), but in addition the typical overlap between the sets of firing of the neuronal populations representing two different places is high (with a mean cosine of the angle or normalized dot product 0.54). In contrast, the sparseness of the representations formed in the dentate gyrus cells with Hebbian rule training was 0.024, and a typical cosine of the angle was 0.000. (Low values of this measure of sparseness, a, indicate sparse representations. The cosine of the angle between two vectors takes the value 1 if they point in the same direction, and the value 0 if they are orthogonal.) These sparse and orthogonal representations are required for high-capacity storage of object and place, object and reward, and, in general, episodic memories. This is the function we believe,of the mapping from entorhinal cortex to hippocampal cells for which Rolls et al. (2006) produced a computational model. This concept maps well onto utility in an environment, for it is the place where we are located or at which we are looking in the world that we wish to associate with objects or rewards, and this association is made explicit in the dentate gyrus–hippocampal representation (Rolls and Xiang, 2005, 2006; Rolls et al., 2005; Rolls and Kesner, 2006). In contrast, the entorhinal cortex may represent self-motion space in a way that is suitable for idiothetic path integration in any environment, that is, independent of the visual cues and landmarks that define the environment, apart from rotation using a head direction signal. In rats, the presence of place cells has led to theories that the rat hippocampus is a spatial cognitive map and can perform spatial computations to implement navigation through spatial environments (O’Keefe and Nadel, 1978; O’Keefe, 1991; Burgess et al., 1994; Burgess and O’Keefe, 1996). The details of such navigational theories could not apply in any direct way to what is found in the primate hippocampus. Instead, what is applicable to both the primate and rat hippo-
campal recordings is that hippocampal neurons contain a representation of space (for the rat, primarily where the rat is, and for the primate primarily of positions ‘‘out there’’ in space) that is a suitable representation for an episodic memory system. In primates, this would enable one to remember, for example, where an object was seen. In rats, it might enable memories to be formed of where particular objects (for example, those defined by olfactory, tactile, and taste inputs) were found. Thus, at least in primates and possibly also in rats, the neuronal representation of space in the hippocampus may be appropriate for forming memories of events (which usually in these animals have a spatial component). Such memories would be useful for spatial navigation. Neuronal recordings have shown that what is represented in the nonhuman primate hippocampal system may also be present in humans, namely that regions of the hippocampal formation can be activated when humans look at spatial views (Epstein and Kanwisher, 1998; O’Keefe et al., 1998).
CONCLUSIONS The spatial view cells described in this and related studies in the primate hippocampus, as well as in some parts of the parahippocampal cortex that send afferents to and receive efferents from the hippocampus (Rolls et al., 1997a, 1998, 2005; Robertson et al., 1998; Georges-Franc¸ois et al., 1999; Rolls and Xiang, 2005), are, in the ways described above, unlike place cells found in the rat (O’Keefe, 1979; Muller et al., 1991). Primates, with their highly developed visual and eye movement control systems, can explore and remember information about what is present at places in the environment without having to visit those places. Such spatial view cells in primates would thus be useful as part of a memory system, becasue they would provide a representation of a part of space that would not depend on exactly where the monkey was and could be associated with items that might be present in those spatial locations. An example of the utility of such a representation in monkeys might be in enabling a monkey to remember where it had seen ripe fruit, or in humans of remembering where they had seen a person or where they had left their keys. The representation of space provided by primate hippocampal spatial view–responsive neurons may thus be useful in forming memories of what has been seen where, an example of an episodic memory. Such memories would be useful for spatial navigation, for which, according to the present hypothesis, the hippocampus would implement the memory component but not the spatial computation component. A detailed
SPATIAL VIEW CELLS AND MEMORY
and quantitative model of how the hippocampus could operate as a memory system and how information stored in the hippocampus could be recalled has been developed elsewhere (Rolls, 1989a,b, 1995, 1996c, 2008; Treves and Rolls, 1994; Rolls et al., 1997b; Treves et al., 1997; Rolls and Treves, 1998; Schultz and Rolls, 1999; Rolls and Stringer, 2005; Rolls and Kesner, 2006). Some of the cells with spatial responses in the primate hippocampus and presubiculum described here could be involved in functions other than purely episodic memory. For example, head direction and wholebody motion neurons could be useful as part of a system for remembering the compass bearing (head direction) and distance traveled to enable one to find one’s way back to the origin, even with a number of sectors of travel and over a number of minutes. This process is referred to as path integration. Spatial memory and navigation can also benefit from visual information about places being looked at, which can be used as landmarks, and spatial view cells added to the head direction cells and whole-body motion cells would provide the basis for a memory system useful in navigation. Another possibility is that primate head direction cells are part of a system for computing during navigation which direction to head toward next. For this process to take place, not only would a memory system of the type elaborated elsewhere be needed (Rolls, 1989a,b, 1996c; Treves and Rolls, 1994; Rolls and Treves, 1998) that can store the spatial information found in the hippocampus, but also an ability to use this information in spatial computation of the appropriate next bearing. Such a system might be implemented using a hippocampal memory system that linked together spatial views, whole-body motion, and head direction information. The findings described here certainly implicate the hippocampus in the update of spatial view cells’ firing produced in the dark by idiothetic cues including eye position and head direction signals. This system would be different from that in the rat (Burgess et al., 1994; McNaughton et al., 1996), in that spatial view is represented in the primate hippocampus. Evidence that primate hippocampal spatial-view cells could be involved in arbitrary associations with the objects and rewards present at particular viewed locations has been described above. These studies provide direct evidence that the primate hippocampus contains the necessary representations for forming such associations, such as representations of objects and of spatial view. Moreover, the new investigation described here of object–place recall memory (Rolls and Xiang, 2006) shows some of the representations that become active within the hippocampus when places are recalled from objects. This recall operation and the learning of the associated events that precede
213
it are described in a model of how the hippocampus is involved in episodic memory and the subsequent retrieval of a whole episodic memory from one part of it in recall. The model describes quantitatively not only the storage and recall operations within the hippocampus but also their recall to the neocortex (Amit, 1989; Treves, 1990; Treves and Rolls, 1991; Rolls and Stringer, 2005; Rolls and Kesner, 2006; Rolls, 2008). In this theory, the CA3 network forms an autoassociative or attractor memory (Amit, 1989) that operates with sparse representations and incomplete connectivity (Rolls, 1989b,1995; Treves, 1990; Treves and Rolls, 1991, 1994; Rolls et al., 1997b; Schultz and Rolls, 1999). Modifiable backprojection synapses to the neocortex implement the recall (Rolls, 1989b, 1995, 1996c; Treves and Rolls, 1994; Rolls and Treves, 1998). The studies described here provide fundamental evidence about the information represented in the primate hippocampus and are of considerable interest for understanding what the primate (including human) hippocampus does and how it works as a memory system (Rolls and Treves, 1998). Indeed, the relevance of this work in primates to humans is attested to by findings from neuroimaging studies in humans showing that the sight of simple spatial views can activate hippocampus-related areas (for example, Epstein and Kanwisher, 1998; and O’Keefe et al., 1998). However, it is only at the neuronal level that one can address issues such as the spatial coordinate frame used (Georges-Franc¸ois et al., 1999), how the information is represented (which has important implications for how it is stored) (Rolls et al., 1998), and how similar the recall state is to the stored memory state when retrieval occurs to a partial cue (Robertson et al., 1998). It would be difficult with neuroimaging studies to show, for example, that there is an allocentric representation of space ‘‘out there’’ accessed by either looking at the particular location in space or rotating the head and moving the eyes to another head direction–eye position (and even head position) combination that would result in looking toward that location in space when the view details are made invisible. Nor can such neuroimaging studies show that other neurons in the hippocampus respond to wholebody motion that for some neurons is based on vestibular signals, for other neurons on optic flow signals, and for yet other neurons on either. Analyses at the neuronal level are thus essential because they provide clear evidence about what is being represented in a brain structure. They are especially relevant to understanding how a part of the brain operates, because they show what information is being exchanged between the computing elements of the brain (Rolls and Treves, 1998; Rolls and Deco, 2002; Rolls, 2008).
214 PRIMATE HIPPOCAMPUS AND PLACE REPRESENTATION acknowledgments The author has worked on some of the experiments described here with A. Berthoz, P.M.B. Cahusac, J.D. Feigenbaum, P. GeorgesFranc¸ois, R.P. Kesner, Y. Miyashita, H. Niki, S. Panzeri, R.G. Robertson, S. Stringer, A.Treves, and J.-Z. Xiang, and their collaboration is sincerely acknowledged. This research was supported by the Medical Research Council and by the Human Frontier Science Program.
References Amaral DG (1986) Amygdalohippocampal and amygdalocortical projections in the primate brain. In: Excitatory Amino Acids and Epilepsy (Schwarz R, Ben-Ari Y, eds.), pp 3–17. New York: Plenum Press. Amaral DG (1987) Memory: anatomical organization of candidate brain regions. In: Handbook of Physiology. Section 1, The Nervous System (Moutcastle VB, ed.), pp 211–294. Washington DC: American Physiological Society. Amaral DG, Price JL, Pitkanen A, Carmichael ST (1992) Anatomical organization of the primate amygdaloid complex. In: The Amygdala (Aggleton JP, ed.), pp 1–66. New York: Wiley-Liss. Amit DJ (1989) Modeling Brain Function. Cambridge, UK: Cambridge University Press. Andersen RA (1995) Coordinate transformations and motor planning in posterior parietal cortex. In: The Cognitive Neurosciences (Gazzaniga MS, ed.), pp 519–532. Cambridge, MA: MIT Press. Burgess N, O’Keefe J (1996) Neuronal computations underlying the firing of place cells and their role in navigation. Hippocampus 6:749–762. Burgess N, Recce M, O’Keefe J (1994) A model of hippocampal function. Neural Netw 7:1065–1081. Cahusac PMB, Miyashita Y, Rolls ET (1989) Responses of hippocampal formation neurons in the monkey related to delayed spatial response and object-place memory tasks. Behav Brain Res 33:229–240. Cahusac PMB, Rolls ET, Miyashita Y, Niki H (1993) Modification of the responses of hippocampal neurons in the monkey during the learning of a conditional spatial response task. Hippocampus 3:29–42. Carmichael ST, Price JL (1995) Limbic connections of the orbital and medial prefrontal cortex in macaque monkeys. J Comp Neurol 346:403–434. Day M, Langston R, Morris RG (2003) Glutamatereceptor-mediated encoding and retrieval of pairedassociate learning. Nature 424:205–209. de Araujo IET, Rolls ET, Stringer SM (2001) A view model which accounts for the spatial fields of hippocampal primate spatial view cells and rat place cells. Hippocampus 11:699–706.
Ekstrom AD, Kahana MJ, Caplan JB, Fields TA, Isham EA, Newman EL, Fried I (2003) Cellular networks underlying human spatial navigation. Nature 425: 184–188. Epstein R, Kanwisher N (1998) A cortical representation of the local visual environment. Nature 392:598–601. Feigenbaum JD, Rolls ET (1991) Allocentric and egocentric spatial information processing in the hippocampal formation of the behaving primate. Psychobiology 19:21–40. Foster TC, Castro CA, McNaughton BL (1989) Spatial selectivity of rat hippocampal neurons: dependence on preparedness for movement. Science 244:1580– 1582. Fried I, MacDonald KA, Wilson CL (1997) Single neuron activity in human hippocampus and amygdala during recognition of faces and objects. Neuron 18:753–765. Gaffan D (1987) Amnesia, personal memory and the hippocampus: experimental neuropsychological studies in monkeys. In: Cognitive Neurochemistry (Stahl SM, Iversen SD, Goodman EC, eds.), pp 46–56. Oxford: Oxford University Press. Gaffan D (1994) Scene-specific memory for objects: a model of episodic memory impairment in monkeys with fornix transection. J Cogn Neurosci 6:305– 320. Gaffan D, Harrison S (1989a) Place memory and scene memory: effects of fornix transection in the monkey. Exp Brain Res 74:202–212. Gaffan D, Harrison S (1989b) A comparison of the effects of fornix section and sulcus principalis ablation upon spatial learning by monkeys. Behav Brain Res 31: 207–220. Gaffan D, Saunders RC (1985) Running recognition of configural stimuli by fornix transected monkeys. Q J Exp Psychol 37B:61–71. Georges-Franc¸ois P, Rolls ET, Robertson RG (1999) Spatial view cells in the primate hippocampus: allocentric view not head direction or eye position or place. Cereb Cortex 9:197–212. Hafting T, Fyhn M, Molden S, Moser MB, Moser EI (2005) Microstructure of a spatial map in the entorhinal cortex. Nature 436:801–806. Ho¨lscher C, Jacob W, Mallot HA (2003) Reward modulates neuronal activity in the hippocampus of the rat. Behav Brain Res 142:181–191. Kreiman G, Koch C, Fried I (2000) Category-specific visual responses of single neurons in the human medial temporal lobe. Nat Neurosci 3:946–953. Kubie JL, Muller RU (1991) Multiple representations in the hippocampus. Hippocampus 1:240–242. Malkova L, Mishkin M (2003) One-trial memory for object-place associations after separate lesions of hippocampus and posterior parahippocampal region in the monkey. J Neurosci 23:1956–1965.
SPATIAL VIEW CELLS AND MEMORY
Markus EJ, Qin YL, Leonard B, Skaggs W, McNaughton BL, Barnes CA (1995) Interactions between location and task affect the spatial and directional firing of hippocampal neurons. J Neurosci 15:7079–7094. Matsumura N, Nishijo H, Tamura R, Eifuku S, Endo S, Ono T (1999) Spatial- and task-dependent neuronal responses during real and virtual translocation in the monkey hippocampal formation. J Neurosci 19:2318– 2393. McNaughton BL, Barnes CA, Gerrard JL, Gothard K, Jung MW, Knierim JJ, Kudrimoti H, Qin Y, Skaggs WE, Suster M, Weaver KL (1996) Deciphering the hippocampal polyglot: the hippocampus as a path integration system. J Exp Biol 199:173–185. McNaughton BL, Barnes CA, O’Keefe J (1983) The contributions of position, direction, and velocity to single unit activity in the hippocampus of freely moving rats. Exp Brain Res 52:41–49. Miyashita Y, Rolls ET, Cahusac PM, Niki H, Feigenbaum JD (1989) Activity of hippocampal formation neurons in the monkey related to a conditional spatial response task. J Neurophysiol 61:669–678. Muller RU, Kubie JL, Bostock EM, Taube JS, Quirk GJ (1991) Spatial firing correlates of neurons in the hippocampal formation of freely moving rats. In: Brain and Space (Paillard J, ed.), pp 296–333. Oxford: Oxford University Press. Muller RU, Ranck JB Jr, Taube JS (1996) Head direction cells: properties and functional significance. Curr Opin Neurobiol 6:196–206. Murray EA, Baxter MG, Gaffan D (1998) Monkeys with rhinal cortex damage or neurotoxic hippocampal lesions are impaired on spatial scene learning and object reversals. Behav Neurosci 112:1291–1303. O’Keefe J (1979) A review of the hippocampal place cells. Prog Neurobiol 13:419–439. O’Keefe J (1984) Spatial memory within and without the hippocampal system. In: Neurobiology of the Hippocampus (Seifert W, ed.), pp 375–403. London: Academic Press. O’Keefe J (1990) A computational theory of the hippocampal cognitive map. Prog Brain Res 83:301–312. O’Keefe J (1991) The hippocampal cognitive map and navigational strategies. In: Brain and Space (Paillard J, ed.), pp 273–295. Oxford: Oxford University Press. O’Keefe J, Burgess N, Donnett JG, Jeffery KJ, Maguire EA (1998) Place cells, navigational accuracy, and the human hippocampus. Philos Trans R Soc B 353: 1333–1340. O’Keefe J, Nadel L (1978) The Hippocampus as a Cognitive Map. Oxford: Clarendon Press. O’Mara SM, Rolls ET, Berthoz A, Kesner RP (1994) Neurons responding to whole body motion in the primate hippocampus. J Neurosci 14:6511–6523.
215
Ono T, Nakamura K, Nishijo H, Eifuku S (1993) Monkey hippocampal neurons related to spatial and nonspatial functions. J Neurophysiol 70:1516–1529. Panzeri S, Schultz SR, Treves A, Rolls ET (1999) Correlations and the encoding of information in the nervous system. Proc R Soc Lond B 266:1001–1012. Parkinson JK, Murray EA, Mishkin M (1988) A selective mnemonic role for the hippocampus in monkeys: memory for the location of objects. J Neurosci 8: 4059–4167. Petrides M (1985) Deficits on conditional associativelearning tasks after frontal- and temporal-lobe lesions in man. Neuropsychologia 23:601–614. Pitkanen A, Kelly JL, Amaral DG (2002) Projections from the lateral, basal, and accessory basal nuclei of the amygdala to the entorhinal cortex in the macaque monkey. Hippocampus 12:186–205. Robertson RG, Rolls ET, Georges-Franc¸ois P (1998) Spatial view cells in the primate hippocampus: effects of removal of view details. J Neurophysiol 79:1145– 1156. Robertson RG, Rolls ET, Georges-Franc¸ois P, Panzeri S (1999) Head direction cells in the primate presubiculum. Hippocampus 9:206–219. Rolls ET (1989a) The representation and storage of information in neuronal networks in the primate cerebral cortex and hippocampus. In: The Computing Neuron (Durbin R, Miall C, Mitchison G, eds.), pp 125–159. Wokingham, England: Addison-Wesley. Rolls ET (1989b) Functions of neuronal networks in the hippocampus and neocortex in memory. In: Neural Models of Plasticity: Experimental and Theoretical Approaches (Byrne JH, Berry WO, eds.), pp 240–265. San Diego: Academic Press. Rolls ET (1990) Theoretical and neurophysiological analysis of the functions of the primate hippocampus in memory. Cold Spring Harb Symp Quant Biol 55: 995–1006. Rolls ET (1995) A model of the operation of the hippocampus and entorhinal cortex in memory. Int J Neural Syst 6:51–70. Rolls ET (1996a) The representation of space in the primate hippocampus, and episodic memory. In: Perception, Memory and Emotion: Frontier in Neuroscience (Ono T, McNaughton BL, Molotchnikoff S, Rolls ET, Nishijo H, eds.), pp 375–400. Amsterdam: Elsevier. Rolls ET (1996b) The representation of space in the primate hippocampus, and its relation to memory. In: Brain Processes and Memory (Ishikawa K, McGaugh JL, Sakata H, eds.), pp 203–227. Amsterdam: Elsevier. Rolls ET (1996c) A theory of hippocampal function in memory. Hippocampus 6:601–620. Rolls ET (1999) Spatial view cells and the representation of place in the primate hippocampus. Hippocampus 9:467–480.
216 PRIMATE HIPPOCAMPUS AND PLACE REPRESENTATION Rolls ET (2000) Memory systems in the brain. Annu Rev Psychol 51:599–630. Rolls ET (2004) The operation of memory systems in the brain. In: Computational Neuroscience: A Comprehensive Approach (Feng J, ed.), pp 291–534. London: CRC Press. Rolls ET (2005) Emotion Explained. Oxford: Oxford University Press. Rolls ET (2008) Memory, Attention, and DecisionMaking: A Unifying Computational Neuroscience Approach. Oxford: Oxford University Press. Rolls ET, Aggelopoulos NC, Zheng F (2003) The receptive fields of inferior temporal cortex neurons in natural scenes. J Neurosci 23:339–348. Rolls ET, Deco G (2002) Computational Neuroscience of Vision. Oxford: Oxford University Press. Rolls ET, Kesner RP (2006) A computational theory of hippocampal function, and empirical tests of the theory. Prog Neurobiol 79:1–48. Rolls ET, Miyashita Y, Cahusac PMB, Kesner RP, Niki H, Feigenbaum J, Bach L (1989) Hippocampal neurons in the monkey with activity related to the place in which a stimulus is shown. J Neurosci 9: 1835–1845. Rolls ET, O’Mara S (1993) Neurophysiological and theoretical analysis of how the primate hippocampus functions in memory. In: Brain Mechanisms of Perception and Memory: From Neuron to Behavior (Ono T, Squire LR, Raichle ME, Perrett DI, Fukuda M, eds.), pp 276–300. New York: Oxford University Press. Rolls ET, O’Mara SM (1995) View-responsive neurons in the primate hippocampal complex. Hippocampus 5:409–424. Rolls ET, Robertson RG, Georges-Franc¸ois P (1997a) Spatial view cells in the primate hippocampus. Eur J Neurosci 9:1789–1794. Rolls ET, Stringer SM (2001) A model of the interaction between mood and memory. Network: Computation in Neural Systems 12:111–129. Rolls ET, Stringer SM (2005) Spatial view cells in the hippocampus, and their idiothetic update based on place and head direction. Neural Netw 18:1229– 1241. Rolls ET, Stringer SM, Elliot T (2006) Entorhinal cortex grid cells can map to hippocampal place cells by competitive learning. Network: Computation in Neural Systems 17:447–465. Rolls ET, Stringer SM, Trappenberg TP (2002) A unified model of spatial and episodic memory. Proc R Soc Lond B 269:1087–1093. Rolls ET, Treves A (1990) The relative advantages of sparse versus distributed encoding for associative neuronal networks in the brain. Network 1:407–421.
Rolls ET, Treves A (1998) Neural Networks and Brain Function. Oxford: Oxford University Press. Rolls ET, Treves A, Foster D, Perez-Vicente C (1997b) Simulation studies of the CA3 hippocampal subfield modelled as an attractor neural network. Neural Netw 10:1559–1569. Rolls ET, Treves A, Robertson RG, Georges-Franc¸ois P, Panzeri S (1998) Information about spatial view in an ensemble of primate hippocampal cells. J Neurophysiol 79:1797–1813. Rolls ET, Xiang J-Z (2005) Reward-spatial view representations and learning in the hippocampus. J Neurosci 25:6167–6174. Rolls ET, Xiang J-Z (2006) Spatial view cells in the primate hippocampus, and memory recall. Rev Neurosci 17:175–200. Rolls ET, Xiang J-Z, Franco L (2005) Object, space and object-space representations in the primate hippocampus. J Neurophysiol 94:833–844. Rupniak NMJ, Gaffan D (1987) Monkey hippocampus and learning about spatially directed movements. J Neurosci 7:2331–2337. Samsonovich A, McNaughton BL (1997) Path integration and cognitive mapping in a continuous attractor neural network model. J Neurosci 17:5900–5920. Schultz S, Rolls ET (1999) Analysis of information transmission in the Schaffer collaterals. Hippocampus 9:582–598. Skaggs WE, Knierim JJ, Kudrimoti HS, McNaughton BL (1995) A model of the neural basis of the rat’s sense of direction. In: Advances in Neural Information Processing Systems (Tesauro G, Touretzky DS, Leen TK, eds.), pp 173–180. Cambridge, MA: MIT Press. Smith ML, Milner B (1981) The role of the right hippocampus in the recall of spatial location. Neuropsychologia 19:781–793. Stefanacci L, Suzuki WA, Amaral DG (1996) Organization of connections between the amygdaloid complex and the perirhinal and parahippocampal cortices in macaque monkeys. J Comp Neurol 375:552–582. Stringer SM, Rolls ET, Trappenberg TP (2004) Selforganising continuous attractor networks with multiple activity packets, and the representation of space. Neural Netw 17:5–27. Stringer SM, Rolls ET, Trappenberg TP (2005) Self-organizing continuous attractor network models of hippocampal spatial view cells. Neurobiol Learn Mem 83:79–92. Stringer SM, Trappenberg TP, Rolls ET, Araujo IET (2002a) Self-organizing continuous attractor networks and path integration: one-dimensional models of head direction cells. Network: Computation in Neural Systems 13:217–242.
SPATIAL VIEW CELLS AND MEMORY
Stringer SM, Rolls ET, Trappenberg TP, Araujo IET (2002b) Self-organizing continuous attractor networks and path integration. Two-dimensional models of place cells. Network: Computation in Neural Systems 13:429–446. Suzuki WA, Amaral DG (1994a) Perirhinal and parahippocampal cortices of the macaque monkey— cortical afferents. J Comp Neurol 350:497–533. Suzuki WA, Amaral DG (1994b) Topographic organization of the reciprocal connections between the monkey entorhinal cortex and the perirhinal and parahippocampal cortices. J Neurosci 14:1856–1877. Tabuchi E, Mulder AB, Wiener SI (2003) Reward value invariant place responses and reward site associated activity in hippocampal neurons of behaving rats. Hippocampus 13:117–132. Taube JS, Goodridge JP, Golob EJ, Dudchenko PA, Stackman RW (1996) Processing the head direction signal: a review and commentary. Brain Res Bull 40:477–486. Treves A (1990) Graded-response neurons and information encodings in autoassociative memories. Phys Rev A 42:2418–2430. Treves A, Rolls ET (1991) What determines the capacity of autoassociative memories in the brain? Network 2:371–397.
217
Treves A, Rolls ET (1992) Computational constraints suggest the need for two distinct input systems to the hippocampal CA3 network. Hippocampus 2: 189–199. Treves A, Rolls ET (1994) A computational analysis of the role of the hippocampus in memory. Hippocampus 4:374–391. Treves A, Rolls ET, Simmen M (1997) Time for retrieval in recurrent associative memories. Phys D 107:392– 400. Van Hoesen GW (1982) The parahippocampal gyrus. New observations regarding its cortical connections in the monkey. Trends Neurosci 5:345–350. Watanabe T, Niki H (1985) Hippocampal unit activity and delayed response in the monkey. Brain Res 325:241–254. Wilson MA, McNaughton BL (1993) Dynamics of the hippocampal ensemble code for space. Science 261: 1055–1058. Wirth S, Yanike M, Frank LM, Smith AC, Brown EN, Suzuki WA (2003) Single neurons in the monkeys hippocampus and learning of new associations. Science 300:1578–1581. Zhang K (1996) Representation of spatial orientation by the intrinsic dynamics of the head-direction cell ensemble: a theory. J Neurosci 16:2112–2126.
14 Learning, Memory, and the Monkey Hippocampus WENDY A. SUZUKI
Ever since the early 1900s, it has been known that brain damage can lead to profound memory impairment (von Bechterew, 1900). However, it was the dramatic description of the severely amnesic patient H.M. by Scoville and Milner (1957) that first focused attention on the medial temporal lobe and, in particular, the role of the hippocampus in normal memory function. Subsequent studies showed that the memory impairment following medial temporal lobe damage in humans was limited to declarative memory for facts, events, and relationships. The memory impairment was seen for declarative information in all sensory modalities tested and spared general cognitive abilities as well as other forms of memory, including perceptual priming, skeletomotor conditioning, and habit learning. While Scoville and Milner (1957) highlighted the possible role of damage to the hippocampus in the severe memory deficit seen in patient H.M., they could not rule out the possibility that combined damage to the hippocampus and amygdala or hippocampus together with the surrounding cortical structures might underlie the impairment. To clarify the role of the various medial temporal lobe structures in memory, experimentalists set out to develop an animal model system. In 1978, the publication of two seminal reports, one by Mishkin in the monkey model system and the other by O’Keefe and Nadel in the rat model system, helped launch the experimental study of the role of the hippocampus in memory. In a classic study, Mishkin (1978) showed that macaque monkeys with large medial temporal lobe lesions that included both the hippocampus and
amygdala made to approximate the lesion sustained by patient H.M. exhibited a severe recognition memory deficit on a delayed nonmatching-to-sample task. Recognition memory is defined as the ability to determine that a given stimulus has been seen before; it is consistently impaired in amnesic patients with medial temporal lobe damage. Because monkeys with smaller lesions limited to either the hippocampus or amygdala were unimpaired on this task, Mishkin suggested that it was combined damage to the hippocampus and amygdala that was responsible for H.M.’s severe memory deficit. The important role of the hippocampus in memory was highlighted in the late 1980s, when histological material from two well-characterized amnesic patients showed that bilateral damage limited to the hippocampus resulted in significant declarative memory impairment (Zola-Morgan et al., 1986; Victor and Agamanolis, 1990). More recently, results from larger groups of patients with MRI-confirmed hippocampal damage have supported these findings (Manns et al., 2003; Bayley et al., 2005). The deficit following selective hippocampal damage in humans was not as severe as the deficit exhibited by patient H.M., thus damage to structures in addition to the hippocampus must be contributing to the more severe memory impairment seen following larger medial temporal lobe lesions. Subsequent studies designed to separate out the relative contributions of the hippocampus and the amygdala in memory in the monkey model system showed that it was not the conjoint damage of the hippocampus and the amygdala that was underlying recognition
218
LEARNING, MEMORY, AND MONKEY HIPPOCAMPUS
memory performance (Zola-Morgan et al., 1989a), but conjoint damage of the hippocampus together with the surrounding and strongly interconnected entorhinal, perirhinal, and parahippocampal cortices (Zola-Morgan et al., 1989b; Meunier et al., 1990, 1993; Suzuki et al., 1993). While large medial temporal lobe lesions including the hippocampus together with those of the surrounding cortical areas produce the most severe memory impairments in monkeys (Mishkin, 1978; Zola-Morgan and Squire, 1985), damage limited to the hippocampus also produced significant memory impairment on a range of memory tasks, including visual recognition memory (Zola et al., 2000), as well as tasks that require memory for sequences of color, spatial, or object information (Beason-Held et al., 1999). Thus, findings from both human and nonhuman primates show that the hippocampus contributes importantly to a range of declarative memory abilities. Paralleling the report by Mishkin (1978) was a second major line of hippocampal research in rodents, started in 1978 with the publication of O’Keefe and Nadel’s highly influential book, The Hippocampus as a Cognitive Map. This book presented the theory that the major role of the hippocampus was to process, represent, and remember spatial information. This theory was based largely on the groundbreaking discovery by O’Keefe and Dostrovsky (1971) of place cells in the rat hippocampus, which increased their firing dramatically whenever the rat moved into a particular location in the environment. These striking physiological findings, together with parallel findings showing that hippocampal lesions in rats produced profound impairments on various tasks of spatial learning and memory, suggested an obvious role of the hippocampus in processing spatial information in memory. The discovery of place cells in the hippocampus and the cognitive map theory was the inspiration for an enormous amount of subsequent research focused on defining the spatial functions of the hippocampus as well as the spatial firing characteristics of hippocampal place cells. However, because a major pillar of the cognitive map theory is that the hippocampus is involved exclusively in processing spatial information, this theory came into conflict with findings from both human neuropsychological studies and experimental studies in monkeys showing that the hippocampus played a wider role in declarative memory for all kinds of information, including spatial information. These differing theoretical perspectives caused a division in the experimental learning and memory literature that lasted for many years. More recently, it has become clear that the rat hippocampus encodes much more than just spatial information. For example, even the earliest studies of hippocampal activity in behaving rats reported activ-
219
ity in response to particular movements or rewards (Ranck, 1973). The use of a wider range of memory tasks to probe rat hippocampal activity including tasks in the visual and olfactory modalities have provided definitive evidence that the rat hippocampus signals a wide range of information beyond space. In an attempt to synthesize the data showing this broader scope of information processed by the rat hippocampus, Eichenbaum and colleagues (1999) proposed an alternative to the cognitive/spatial map theory of the hippocampus, referred to as the ‘‘memory space hypothesis.’’ The core of this theory suggests that hippocampal neurons do not just represent spatial information, but are better characterized as representing the full temporal sequence of events that compose ongoing episodes (Fig. 14–1). This idea is similar to the suggestion of Morris and Frey (1997) that the hippocampus encodes information about all ongoing and attended episodes. This includes not only information about the spatial location of the rats in the environment (i.e., place cell activity) but also other nonspatial elements, including information about the timing of particular events, mnemonic signals used in the episode, as well as information about the regularities that may be common across similar episodes. Neurons that signal the commonalities between related episodes that presumably build up over time were termed ‘‘nodal’’ cells and are thought to play a critical role in linking repeated and related episodes into a ‘‘memory space.’’ According to the memory space view, hippocampal place cells are not part of a cognitive or spatial map of the environment, but instead represent nodal information that can link events that occur in the same location across episodes. The hippocampus plays a critical role in processing episodic information as well as semantic information represented by the commonalities across episodes. Both episodic and semantic information are thought to enter memory through Hebbian synaptic plasticity. These physiological predictions are also consistent with the broader relational theory of hippocampal function put forth by this same group that suggests that the hippocampus is required for the formation and flexible use of representations of all kinds of relations among items, spatial locations, or events (Cohen and Eichenbaum, 1993; Eichenbaum and Cohen, 2001). The memory space hypothesis provided a more comprehensive description of the wide range of hippocampal activity that had been reported in the rodent hippocampus. Its emphasis on the episodic and semantic patterns of neural encoding also provided a strong link to the human and nonhuman primate neuropsychological literature. While the memory space hypothesis was based largely on findings from rat hippocampal physiology and lesion studies, a valuable database with which to test these ideas is the monkey
220 PRIMATE HIPPOCAMPUS AND PLACE REPRESENTATION ings in the monkey hippocampus. This review will show that, consistent with the memory space hypothesis, hippocampal neurons provide a rich array of responses that signal all aspects of the ongoing memory trial. I will also discuss more recent studies showing strong associative learning signals as well as long-term memory signals in the monkey hippocampus that provide further insight into the plastic and mnemonic properties of hippocampal neurons.
MEMORY SIGNALS DURING TASKS WITH SHORT DELAY INTERVALS IN THE MONKEY HIPPOCAMPUS
Figure 14–1. Schematic illustration of the pattern of hippocampal coding seen during two spatial trajectories with a single overlapping point. According to the memory space concept, hippocampal neurons should encode all unique aspects of these episodes, including not only spatial locations but also motor-related activity, reward-related activity, and commonalities or ‘‘nodes’’ between related episodes. From Eichenbaum et al. (1999), with permission.
hippocampal physiological literature. Since the early 1980s, neurophysiologists have been recording activity in the monkey hippocampus during a variety of memory tasks with short delay intervals, as well as in long-term memory tasks. These studies report that hippocampal neurons signal a wide range of taskrelated information. However, the relevance of these signals to memory performance was difficult to infer, because only a small proportion of the hippocampal cells gave responses specific for particular to-beremembered stimuli (i.e., stimulus-selective memory responses). These stimulus-selective memory responses (i.e., strong activity when a cup but not other stimuli are correctly remembered) are considered to be a prototypical memory signal. A re-evaluation of these data in the context of the memory space hypothesis suggests that while relatively few stimulus-selective memory signals were seen in the monkey hippocampus, many other task-related signals, including signals selective for commonalities across trials (i.e., nodal signals), reward, or motor responses, provide information about all aspects of the ongoing trial. Thus the overall pattern of activity seen in monkey hippocampal neurons is consistent with key predictions of the memory space hypothesis. In this chapter, I will review over 30 years of findings from behavioral neurophysiological record-
Starting in the early 1980s, a range of memory tasks with short delay intervals between 500 ms and several seconds in length were used to probe hippocampal activity in awake behaving monkeys. These tasks are typically described as tasks of ‘‘short-term’’ or ‘‘working’’ memory and include variations of the delayed matching-to-sample tasks (Brown, 1982; Wilson et al., 1988; Riches et al., 1991; Colombo and Gross, 1994; Hampson et al., 2004), delayed-response tasks (Watanabe and Niki, 1985), object–place memory tasks (Cahusac et al., 1989; Rolls et al., 1989, 2005), or serial recognition tasks (Rolls et al., 1993). Despite differences in task details, each of these tasks shares several common features (Fig. 14–2, top panel). They typically include a sample stimulus presentation phase, during which a to-be-remembered stimulus is presented. Most include a delayphase, during which time the sample stimulus must be kept in memory, and a test phase, when the information about the to-beremembered sample stimulus can be used to solve the task. All tasks also include a motor response phase as well as a reward phase. To solve these tasks, animals must differentiate between the to-be-remembered sample stimuli, remember a particular sample stimulus over the delay interval, and use that remembered information to respond appropriately during the test phase of the task. Feedback about whether the response was correct or wrong is provided by the reward given at the end of the trial. Hippocampal neurons exhibit a rich range of responses during all phases of these short-term or working memory tasks, including sensory- and memory-related activity specific for particular stimuli shown in the task (stimulus-selective mnemonic activity), category-selective responses, and a surprisingly large proportion of responses selective for task phase (i.e., sample, delay, or test phase; Fig. 14–2A–K). In the following sections, I describe this task-related hippocampal activity during shortterm and working memory tasks, organized by task phase.
Figure 14–2. Top panel illustrates a typical short-term recognition memory task including a sample stimulus presentation, delay interval during which the sample stimulus must be held in memory, a test interval when a comparison must be made between the information held in memory and the presented stimulus, a response phase (typically a motor response), and a reward phase if the trial was correct. A–K. Schematic illustration of the range of different signals that have been seen in the monkey hippocampus during performance of simple short-term or working memory tasks. The different patterns of activity associated with each task period are shown below that task period. A. Schematic illustration of a cell responsive during the sample period of the task (putative nodal signal) but not selective for the individual sample stimuli. B. Schematic illustration of a sample selective response that differentiates between sample 1 (S1) and sample 2 (S2). For panels A and B, the thick black line indicates the time period during which the sample stimulus was presented C. Schematic illustrations of a cell responsive during the delay period of the task. D. Schematic illustration of a cell that exhibits selective delay activity, differentiating between sample 1 held in memory (S1) and sample 2 held in memory (S2). For panels C and D, the thick black line indicates the duration of the delay interval. E. Example of a cell that is responsive during the test period of the task, but that does not differentiate between particular test items (T1 and T2). F. Abstracted match-nonmatch signal that differentiates whether a test stimulus matches a stimulus held in memory or not, but does not provide specific information about the identity of the test stimulus. F. Selective match– nonmatch signal that differentiates between a preferred stimulus (S1) shown as a match compared to the same stimulus shown as a nonmatch. In panels E–G, the thick black line indicates the time period that the test stimulus was presented. H. Activity selective for the response phase of the task. I. Activity that differentiates between different responses made during the response period of the task. For panels H and I, responses are aligned to the motor response. J. Cell that signals a correct trial has been completed. K. Cell that signals the absence of reward on error trials. For panels J and K, responses are aligned to the delivery of reward.
221
222 PRIMATE HIPPOCAMPUS AND PLACE REPRESENTATION
Sample Period Activity: Responses to Visual and Auditory Stimuli To determine if hippocampal neurons participate in the encoding of the sample stimuli used in memory tasks with short-delay intervals, researchers have examined the pattern of neural activity seen during the sample stimulus presentation. In general, a relatively substantial proportion of hippocampal neurons respond during the sample presentation (Fig. 14–2A), though fewer cells respond selectively to particular to-be-remembered stimuli (stimulus-selective response (Fig. 14–2B). Riches et al. (1991) reported that approximately 30%–40% of the recorded hippocampal cells responded significantly to the sample stimulus during the performance of a simple delayed matching-to-size task using geometric objects as stimuli. However, only 10%–15% of the hippocampal cells differentiated between square and non-square stimuli. Colombo and Gross (1994) reported that 56% (17/30) of hippocampal neurons responded significantly during the visual sample stimuli of a visual–visual delayed matching-tosample task, and 33% of the cells (10/30) responded differentially to the two different visual stimuli used (i.e., stimulus-selective response). That is, 33% of hippocampal cells would respond more when one or the other visual stimulus was shown as a sample. That same study reported 25% (16/63) of hippocampal neurons recorded responded significantly during the auditory sample stimulus of an auditory–visual delayed match-to-sample task and 14% of the hippocampal cells (9/63) responded selectively to one of the auditory stimuli used. In a more recent example, Hampson et al. (2004) reported that 21% of hippocampal cells responded significantly and exclusively during the sample period of a delayed match-to-sample task using clip art images as stimuli, but none of these cells differentiated between the various to-be-remembered sample stimuli. While cells that respond selectively during the sample stimulus presentation may participate in the encoding of the to-be-remembered sample stimulus, what is the function of the non-stimulusselective hippocampal cells that respond during the sample presentation? These cells may signal the occurrence of the particular task phase (i.e., the sample phase) or provide timing information about the occurrence of the sample phase of the task. This interpretation is consistent with a prediction of the memory space hypothesis that hippocampal neurons convey nodal or semantic-like information about the commonalities across related trials or episodes. The finding of note here is that an even larger proportion of hippocampal cells provide this nodal, sample period–selective response than those that signal stimulus-selective sample period information.
While Hampson et al. (2004) reported no stimulusselective hippocampal activity during the sample period presentation, the most striking observation from this study was the population of 20% of hippocampal neurons that responded to particular categories of the visual clip art images shown during the sample period of the task. For example, some neurons responded to any of the clip art images that included people. Others responded to categories of images that included a particular color. Consistent with these category-selective responses seen in the monkey hippocampus, Quiroga et al. (2005) described single neurons in the human hippocampus of epileptic patients that were selectively activated by different pictures of particular famous individuals, landmarks, or objects. Some cells were also responsive to the letter strings naming the target stimuli. In one widely described example, a cell from the left posterior hippocampus responded to a variety of pictures of the actress Jennifer Aniston. Like the category-selective cells described in the monkey hippocampus (Hampson et al., 2004), these cells appeared to respond selectively to the category of Jennifer Aniston, irrespective of the details of the visual image representing that category. This is one of the most striking examples of hippocampal cells being able to link common but very different elements together in memory. This ability to tie common elements together in memory has been taken as evidence that the hippocampal neurons are involved in networks that support inferential memory expression (Eichenbaum and Cohen, 2001). This is a function required to link different items or episodes in memory by their common elements and is a critical requirement of recollection. This kind of representation might also be particularly useful during tasks of transitive inference, which refers to the ability to infer that if item A is bigger than item B and item B is bigger than item C, then item A is bigger than item C. Tasks requiring transitive inference are known to be sensitive to damage to the hippocampal system in rats (Bunsey and Eichenbaum, 1996) and monkeys (Buckmaster et al., 2004); these tasks have also been shown to activate the hippocampus in human functional imaging studies (Preston et al., 2004). It will be of great interest to explore the neurophysiological correlates of hippocampal activity during tasks of transitive inference in animal model systems.
Responses to Spatial Stimuli A similar proportion of both stimulus-selective and nonselective hippocampal responses have been seen in response to visual–spatial sample stimuli. For example, using a spatial delayed-response task, Watanabe and Niki (1985) reported that 24% (65/272) of the recorded
LEARNING, MEMORY, AND MONKEY HIPPOCAMPUS
hippocampal cells responded significantly to light cues shown in two different spatial locations on the screen during the sample period of the task. However, only 3% (9/272) of the neurons responded selectively to the different spatial locations. Another study used a similar object–place association task where computer monitors in two different spatial locations in the room served as the spatial cues. While 32% of the hippocampal cells responded to the test stimuli, 12% and 10% exhibited selective place or combined object–place selectivity, respectively (Rolls et al., 2005). Thus, in general, modest numbers of hippocampal cells respond selectively to cues shown in different spatial locations. While the studies described above required animals to make egocentric judgments about the relative spatial positions of stimuli on a fixed computer monitor, both Rolls and Ono have examined allocentric spatial responses of neurons in the monkey hippocampus (see Chapters 12 and 13, this volume). Rolls and colleagues (Rolls and O’Mara, 1995; George-Francois et al., 1999) recorded in the monkey hippocampus as animals were either placed or moved to different spatial locations in the environment. Typically, animals were not required to perform any spatial memory tasks during these experiments. They reported that between 6% and 11% of the hippocampal cells recorded responded selectively to particular views of the spatial environment. In contrast, theses studies failed to find cells that responded according to the spatial location of the subject in the environment similar to rat hippocampal place cells. Ono and colleagues (Matsumura et al., 1999; Hori et al., 2003) recorded hippocampal activity as monkeys navigated through either real or virtual environments while performing simple spatial tasks. Using this paradigm, they reported that 32% to 43% of the recorded hippocampal neurons exhibited activity specific to the actual or virtual location of the monkey in the environment (place cell–like activity). Like rat hippocampal place cells, Ono’s group reported that the monkey hippocampal cells were sensitive to arrangements of distal cues. Both groups suggest that either the view-selective or place-selective information may participate in coding of particular spatial contexts in memory. These studies together provide strong evidence that like rat hippocampal cells, monkey hippocampal cells signal information about allocentric spatial information. It will be important to test the mnemonic capacities of these cells further by recording hippocampal activity as monkeys perform spatial tasks known to be sensitive to hippocampal damage.
Delay Period Activity A major advance in our understanding of the neurophysiology of working memory came in 1971, when
223
Fuster and Alexander described cells in the macaque prefrontal cortex that exhibited robust sustained activity during the delay interval of a spatial working memory task (spatial delayed-response task). Because this delay activity was selective for particular to-beremembered information (i.e., strong delay activity when a north target but not a south, east, or west target is remembered), it has been widely viewed as a neural correlate of spatial working memory. Many previous studies have shown large proportions of neurons in the prefrontal cortex that exhibit stimulus-selective delay activity during various object or spatial working memory tasks (Funahashi et al., 1989, 1990, 1991; Miller et al., 1995). Similarly, many cells in the monkey hippocampus are also active during the delay interval of various memory tasks. However, in contrast to the prefrontal cortex, only a small proportion of these delayactive cells in the monkey hippocampus are stimulus selective. Instead, the majority of the delay-active hippocampal cells are not selective for particular to-beremembered stimuli. Watanabe and Niki (1985) reported that 43% of all the recorded hippocampal cells responded significantly during the delay interval of a spatial delayedresponse task, but that only 8/272 (3%) of the cells responded selectively depending on the location of the sample stimulus (Fig. 14–2D). Similarly, Colombo and Gross (1994) reported that while 33% and 41% of the hippocampal neurons exhibited significant responses during the delay interval of a visual–visual delayed matching-to-sample task or an auditory– visual delayed matching-to-sample task, respectively, only 10% and 18% of the hippocampal cells responded in a stimulus-selective fashion. Cahusac et al. (1989) examined hippocampal activity during either an object– place memory task or a delayed spatial-response task and reported that 4% of the population of 600 cells fired selectively for either right or left positions during the delay intervals of the task. Hampson et al. (2004) reported that 11% of the hippocampal cells exhibited significant activity during the delay period of a delayed matching-to-sample task but none of the cells exhibited stimulus-selective activity (Fig. 14–2C). The relatively small proportion of hippocampal cells exhibiting selective delay activity suggests that the hippocampus does not play a primary role in working memory. This interpretation is consistent with lesion findings indicating that damage to the prefrontal cortex (Pribram et al., 1952) but not the hippocampus (ZolaMorgan and Squire, 1985) yields clear working memory deficits over short delay intervals. Recently, results from fMRI studies showing hippocampal activations during the delay interval of both short-term and long-term memory tasks have been taken as evidence that the medial temporal lobe may
224 PRIMATE HIPPOCAMPUS AND PLACE REPRESENTATION be important for memory over a short delay interval (Ranganath and D’Esposito, 2001; working memory; Ranganath et al., 2005). This contrasts with the widely held belief that the medial temporal lobe is critical for long-term but not short-term memory. However, the neurophysiological evidence presented above suggests that while both the hippocampus and prefrontal cortex are indeed active during the delay interval of simple working memory tasks with a single item held in memory, they appear to be engaged in quite different functions. The prefrontal cortex is involved in conveying selective information about the to-be-remembered stimulus over the delay interval, consistent with a primary role in working memory. In contrast, the hippocampus appears more involved in providing more general or nodal information about the timing or occurrence of a particular task phase. This activity may be better described as nodal, or semantic-like, information about commonalities across trials rather than as a working memory signal. Other recent studies support the idea that the hippocampus is not involved in memory over short delay intervals per se, but instead is required any time there is a requirement for the formation and flexible use of representations of relationships between items (Cohen and Eichenbaum, 1993; i.e., relational memory; Eichenbaum and Cohen, 2001). Its critical role in processing relational information results in impairment on not only long-term memory tasks but also shortterm memory tasks that use relational stimuli as memorandum. Consistent with this idea, several recent studies have shown that short-term memory for the relationship between items in a scene or memory for face–scene relationships (Hannula et al., 2006) and short-term memory for spatial configurations, faces, color–place associations, or object–location conjunctions (Olson et al., 2006a,b) are impaired in amnesic patients with medial temporal lobe damage or more selective hippocampal damage. These findings support the idea that the medial temporal lobe plays a critical role in processing all forms of relational information irrespective of the duration of the delay interval used in the task. These findings also make the prediction that working memory tasks that use relational stimuli as memoranda may elicit a different pattern of activity in the medial temporal lobe from that of the largely non-stimulus-selective responses seen with the simple nonrelational working memory tasks described above. Specifically, these relational stimuli may result in a much more robust and selective pattern of hippocampal activity during both the stimulus presentation period as well as the memory delay period. Future neurophysiology experiments will be needed to test this prediction directly.
Effects of Memory on StimulusResponsive Cells While the previous section summarized hippocampal responses to sample stimuli, here we ask how hippocampal cells respond during the test phase of the task when animals must use memory of the sample stimulus to compare with the test stimulus and respond appropriately. Consistent with findings in the sample period, some studies report that hippocampal neurons respond during the test phase, but do not differentiate between the different stimuli shown nor provide information about match or nonmatch status of the stimulus (Fig. 14–2E; Hampson et al., 2004) Other studies, however, have described clear recognition memory-related signals in the hippocampus. For example, Wilson et al. (1988) recorded in the hippocampus during a size matching-to-sample task using simple geometric shapes in which a sample stimulus was followed after a short delay interval by a test stimulus. If the test stimulus matched the sample stimulus, the animal was required to press the right panel to get reward. If the two stimuli were different, the animal could press the left panel to get reward. Wilson et al. (1988) reported that 102/303 (34%) of the hippocampal units recorded responded differentially during the test stimulus phase depending on whether the trial was a match (go left) or a nonmatch (go right). Thus, irrespective of the particular sample or test stimulus used, these cells differentiated matching from nonmatching trials during the test phase. Unfortunately, control trials examining the motor dependency of this response were not done. Control trials for motor-related activity, however, were done in a study in the rat hippocampus during an olfactory version of a delayed nonmatching-to-sample task (Otto and Eichenbaum, 1992). This study reported that 15 of 120, or 12.5%, of the responsive hippocampal neurons responded differentially to all stimuli used in the task when they were shown as matches compared to when the same stimuli were shown as nonmatches. This pattern of activity has been referred to as an ‘‘abstracted’’ recognition memory signal in the sense that these hippocampal neurons do not convey specific information about particular stimuli, but instead signal the outcome of the match–nonmatch comparison for all stimuli (Otto and Eichenbaum, 1992; Fig. 14–2F). Importantly, in this latter study, control experiments showed that these abstracted match–nonmatch signals could be clearly dissociated from motor responses. Another common recognition memory signal described during performance of various memory tasks are neurons that respond selectively to different visual test items (i.e., visually selective neurons) and whose stimulus-selective response can be either enhanced or suppressed, depending on whether the neuron’s
LEARNING, MEMORY, AND MONKEY HIPPOCAMPUS
preferred stimulus is held in memory or not. Thus, stimulus-selective neural activity can be significantly modulated by a memory requirement. This has been referred to as a stimulus-selective match–nonmatch signal (Fig. 14–2F). While many neurons in the perirhinal and entorhinal cortices exhibit stimulus-selective match–nonmatch signals, only small proportions of monkey hippocampal neurons (Wilson et al., 1988) or rat hippocampal neurons (Hampson et al., 1993; Wiebe and Staubli, 1999; Wood et al., 1999) have been described with selective match–nonmatch responses. Thus, while both the hippocampus and the surrounding entorhinal and perirhinal cortices signal memory for previously seen stimuli, these areas signal recognition memory in different but complementary ways. The cortex provides predominantly stimulus-selective mnemonic signals including stimulus-selective match–nonmatch signals (Fig. 14–2G). In contrast, the hippocampus conveys strong abstracted match–nonmatch signals (Fig. 14–2F), although some selective mnemonic signals have also been reported (Suzuki and Eichenbaum, 2000). These neurophysiological observations are consistent with reports showing that the severity of the recognition memory deficit seen in monkeys is correlated with the precise locus and extent of the medial temporal lobe region damaged (Zola-Morgan et al., 1994). This finding could be explained by the fact that these medial temporal lobe areas all contribute in complementary ways to recognition memory, allowing for partial compensation when part of the system is damaged. However, the most severe recognition deficit is seen only when all of these medial temporal lobe areas have been compromised.
Motor-Related Activity While the place cells in the rat hippocampus were emphasized for many years, it was clear from the earliest recordings that motor movements could strongly influence the activity in the rat hippocampus (Ranck, 1973; O’Keefe and Nadel, 1978). Consistent with this observation in the rat hippocampus, several studies have described clear motor-related activity during the performance of various memory-demanding tasks. Watanabe and Niki (1985), for example, found that 18.8% of the hippocampal neurons either increased or decreased their activity at the time of initiation of the motor response (i.e., release of the hold lever; Fig. 14– 2H). Miyashita et al. (1989) reported that 33% (63 of 192 cells) of hippocampal neurons responded during the right or left arm movement of two versions of a conditional spatial-response task (Fig. 14–2I). In addition to the sample-related and match-nonmatchrelated signals observed in the Kornorski delayed
225
matching task, Wilson et al. (1990) reported that 26% of the recorded hippocampal neurons responded with either increasing or decreasing firing during the motor response period of the task, and 16% of the total number of cells responded differentially to the rightversus-left reaches required by the task. While not directly related to the ability of the animal to solve the memory tasks, these relatively prominent motor signals suggest that as in rodents, hippocampal neurons in monkeys signal a wide range of task-related information, including motor-related signals.
Reward-Related Signals Recent studies have described cells in the monkey perirhinal cortex that signal the association between particular stimuli and reward (Liu and Richmond, 2000; Mogami and Tanaka, 2006). Given that essentially all the memory tasks used to study the neurophysiology of memory in the hippocampus are reward based and the feedback from reward is absolutely essential for continued good performance on the task, it is not surprising that information about rewards are represented in the medial temporal lobe. Watanabe and Niki (1985) described a small population of hippocampal cells that differentiated between correct and error trials during the performance of a spatial working memory task (17 or 272 cells studied, or 6.3%). One subpopulation either increased or decreased firing at the time of the juice delivery during the task but did not respond to juice given outside the context of the task (Fig. 14–2J). A second category of cells increased or decreased their firing rate after the key-press response when the animal either made errors or juice was omitted from a correct trial (Fig. 14–2K). The correctresponding cells and omit-reward cells in monkeys have similarities to the approach-consummate cells and approach-consummate mismatch cells described in early reports of hippocampal activity in rats (Ranck, 1973). Rolls and Xiang (2005) asked monkeys to respond to different cue locations in complex visual scenes in which one of the cues was associated with a preferred reward and the other was associated with a less preferred reward. They showed that 18% of hippocampal neurons responded differentially to the location of a preferred reward compared to the non-preferred reward, and 60% of the neurons tested reversed the location to which they responded when the location of the preferred reward was changed. Thus, hippocampal neurons signal information about reward available at particular locations (reward–location associations). To summarize, this review of the mnemonic signals seen during memory tasks with short delay intervals in the monkey hippocampus shows that essentially all
226 PRIMATE HIPPOCAMPUS AND PLACE REPRESENTATION aspects of the ongoing trial from the sample presentation to the reward is represented by activity of hippocampal neurons (Fig. 14–2A–K). This information includes both to-be-remembered information about the trial (Fig. 14–2D, F, G) and information about commonalities across trials (Fig. 14–2A,C,E,H,J,K), movements being made (Fig. 14–2H,I), and information about reward outcome (Fig. 14–2J,K). Overall, these findings confirm some of the major predictions of the memory space hypothesis and emphasize the similarities between the patterns of task-related activity described in the monkey and rodent hippocampus. While the previous sections focused on activity seen during short-term or working memory tasks, the next section addresses activity seen during tasks of associative learning and long-term memory.
ASSOCIATIVE LEARNING SIGNALS IN THE MONKEY HIPPOCAMPUS A critical function of the medial temporal lobe is the ability to successfully acquire new declarative information about facts and events that includes new associations between initially unrelated items (associative learning). Both Wirth et al. (2003) and Cahusac et al. (1993) examined the patterns of hippocampal activity during the learning of novel conditional motor associations, where behavioral learning could be easily measured and compared with changes in neural activity (Cahusac et al., 1993; Wirth et al., 2003). This category of associative learning task, also known as arbitrary sensory motor mapping or conditional visuomotor learning, requires animals to associate a given sensory stimulus (typically a visual image presented on a computer screen) with a motor response (i.e., look right). Post-training lesions to the medial temporal lobe in monkeys impair their ability to learn novel conditional motor associations while well-learned associations remain unaffected (Rupniak and Gaffan, 1987; Murray and Wise, 1996; Wise and Murray, 1999; Murray et al., 2000; Brasted et al., 2002, 2003). In the study by Wirth et al. (2003), animals were first shown four identical target stimuli superimposed on a complex visual scene that filled the video monitor. Following a delay interval, during which the scene disappeared but the targets remained on the screen, the animal was cued to make a single eye movement to one of the peripheral targets on the screen (Fig. 14– 3A). For each visual scene, only one of the four targets was associated with reward. Each day, the animals learned two to four new scenes by trial and error. These new scenes were also randomly intermixed with well-learned ‘‘reference’’ scenes that the animals had seen for many months before the recording experi-
ments began. Responses to the reference scenes were used to control for motor-related activity in the hippocampal cells. In contrast to the short-delay memory tasks described above, this study found that 61% of the hippocampal cells examined responded selectively to the different scenes shown in the task during the scene period, the delay period, or both periods of the task (i.e., differential responses to different visual stimuli, also termed visually selective responses). The substantially larger number of stimulus-selective responses seen in this task compared to that in the studies reviewed above might be due to a number of factors, including the use of highly complex visual images and the use of a spatial memory task dependent on intact hippocampal function. Moreover, both the spatial task and the complex overlapping visual-scene stimuli used in this task are highly relational, which may also explain why so many hippocampal cells responded selectively. Selectively responding cells with learningrelated activity were identified by correlating a moving average of the raw neural activity with a moving average of the raw behavioral performance during learning. Using this criterion, 28% of the selectively responding cells showed a significant positive or negative correlation with learning. These cells were termed ‘‘changing cells.’’ Two categories of changing cells were described. Sustained changing cells (54% of the population of changing cells) signaled learning with a change in neural activity that was maintained for as long as we were able to hold the cell (Fig. 14– 3B). This category resembles the sustained cells described by Cahusac et al. (1993). Baseline-sustained changing cells making up the remaining 45% of changing cells started out with a scene-selective response during either the scene or delay period of the task, even before the animal learned the association, and signaled learning by returning to baseline activity (Fig. 14–3C). This return to baseline activity was anti-correlated with the animal’s learning curve for that particular scene. While Causac et al. (1993) did not describe baseline-sustained cells, they did describe a third population of hippocampal neurons that only showed differential activity to the two visual stimuli transiently, near the time of learning before returning to baseline levels of response (transient cells). Both transient cells and baseline-sustained cells are similar in that they signal learning with a return to baseline levels of activity. To examine the specific relationship between the changing cell activity and the behavioral correlates of learning, the temporal relationship between behavioral learning and changes in neural activity was examined in detail. Specifically, for each new learning condition for which neural activity changed, comparisons were
Figure 14–3. A. Schematic illustration of the location–scene association task (see text for details). B. Illustration of the trial-by-trail probability correct performance (dotted line reads from the left axis) as a function of the trial-by-trial activity of cells during either the scene or delay period of the task (solid line reads from the right axis) for a sustained changing cell. Filled circles at the top of graphs B and C indicate error trials, while open circles indicate correct trials. r value refers to the correlation between behavioral and neural changes with learning. C. Illustration of a baseline-sustained cell. Note the strong positive or negative correlation between neural activity and learning. D. Scatter plot illustrating the temporal relationship between trial number of behavioral changing (i.e., learning) and trial number of neuronal change. Note that about half the cells change before or at the same time as learning while the remaining half of the cells change before learning. Moreover, there is a strong tendency for cells to change after learning if the scene was learned quickly (i.e., faster than about 15 trials) and to change after learning if the scene was learned more slowly.
227
228 PRIMATE HIPPOCAMPUS AND PLACE REPRESENTATION made between the estimated trial number of neural change and the estimated trial number of learning. This comparison showed that hippocampal cells can signal learning before (n ¼ 18), at the same time (n ¼ 1), and after (n ¼ 18) learning. Hippocampal cells signaled learning starting from as much as 13 trials before learning to 15 trials after learning (Fig. 14–3D). Similar to the Wirth et al.’s (2003) study, Cahusac et al. (1993) reported that that the learning-related signals could occur within a wide range of lag or lead times relative to behavioral learning, mainly from 30 trials before learning to 40 trials after learning. Taken together, the studies of learning-related activity in the hippocampus show that hippocampal neurons appear to participate at all stages of the learning process, from several trials before behavioral learning is expressed, when the observed activity may be involved in driving the learned behavior, to several trials after learning, when the activity may be involved in a strengthening process. These findings not only emphasize the key role of the hippocampus in encoding new regularities or relationships but suggest that the hippocampus participates in the network of structures that drive the behavioral changes that underlie new learning. The memory space hypothesis predicts that when a rat is introduced to a new environment, hippocampal neurons may initially show little activity, but activity may develop with repeated experience as the relationships, commonalities, and other features of the environment are learned (Eichenbaum et al., 1999). These learning-related changes are thought to follow rules of classic Hebbian synaptic plasticity. These predictions are consistent with elegant studies in the rat showing the development of place cell activity as rats are introduced to a novel environment (Wilson and McNaughton, 1993). However, spatial learning was not directly examined in these studies. More recently, Frank et al. (2004) analyzed the development of place cells as animals were exposed to either a familiar set of arms in a modified T-maze or during the first 3 days of exposure to a novel set of arms. They showed that strong and selective place cell activity would develop on the first exposure to a novel arm. They further showed that the largest changes in place field activity in the novel arm on days 1 and 2 corresponded to the most striking increase in running speed, suggesting a correlation with learning-related behavior (increased running speed) and place cell development. This study also suggests changes in hippocampal activity during learning; however, because this study did not directly compare changes in neural activity and behavior, the specific role of these neural changes to learning could not be specified. The studies by Wirth et al. (2003) and Cahusac et al. (1993) provide further evidence of clear, learning-related activity when behavioral learning is
well controlled. The sustained, baseline-sustained, and transient cells described in these studies provide some of the strongest examples of learning-related activity in the monkey hippocampus. Similar findings have also been reported in rabbit hippocampus during trace eyeblink conditioning (McEchron and Disterhoft, 1997). Moreover, these findings suggest that hippocampus neurons change early enough relative to the behavioral change to underlie the learning process. In rats it has been shown that the long-term stability of place cell activity is dependent on NMDA receptors (Kentros et al., 1998). It will be of great interest to examine the role of NMDA receptors in both the initial formation as well as the long-term consolidation of new location– scene associations in the monkey hippocampus.
LONG-TERM MEMORY SIGNALS IN THE MONKEY HIPPOCAMPUS While most early studies in the monkey hippocampus used tasks of short-term or working memory, Miyashita et al. (1989) recorded hippocampal activity during a conditional spatial-response task requiring longterm memory for particular visual–motor associations. In this task, the presentation of one stimulus required three consecutive reach responses to obtain reward while the presentation of a second stimulus required no response (i.e., a no-go response). They reported that 14% of the hippocampal cells recorded responded differentially to the two different visual stimuli used. It was suggested that these signals were consistent with long-term memory signal specific for the learned object–response associations. This interpretation was supported by control trials showing that these neurons did not responding selectively to the particular visual stimuli used, nor to particular arm movements. While this study suggests that information about well-learned associations are signaled by hippocampal neurons, other studies have shown that various cortical areas signal long-term memory by increases in their stimulus-selective responses to particular well-learned stimuli (Logothetis and Pauls, 1995; Kobatake et al., 1998; Baker et al., 2002; Sigala et al., 2002). To determine if hippocampal neurons also convey this form of long-term memory, Yanike et al. (2004) examined responses of hippocampal neurons to very well-learned location–scene associations using the same task described by Wirth et al. (2003; see also Fig. 14–3A). Yanike et al. (2004) focused their analysis on the hippocampal responses to the very well-learned ‘‘reference’’ associations. These reference associations were identical to the new associations except that the particular location–scene combinations were highly familiar to the animals. In the study by Wirth et al.
LEARNING, MEMORY, AND MONKEY HIPPOCAMPUS
Figure 14–4. A. Distribution of selectivity index (SI) for all selectively responding hippocampal cells shown separately for new and reference scenes. The average SI for the reference scene and new scenes was 0.55 and 0.47, respectively. These values are significantly different from each other. B. Response of a single selective hippocampal cell to four reference scenes and three new scenes. This cell responded highly selectively to a particular reference scene. The SI to the references scenes was 0.7 while the corresponding value for the new scenes was 0.3.
(2003), the reference associations served as an important control for motor-related responses. In most cases, the animals had between 6 and 22 months of previous experience with the reference associations before the cells were recorded. Similar to reports in other cortical areas, the selectivity of hippocampal neurons was also modulated by extensive training with the well-learned associations compared to novel associations. Further analysis showed that hippocampal neurons responded more
229
selectively (i.e., differentially) to the well-learned associations than to the new associations during both the scene period of the task and the delay period of the task, as measured by a selectivity index (SI; Moody et al., 1998). Figure 14–4B illustrates the response of an individual hippocampal cell to four references scenes and three new scenes. Note how much more strongly this cell differentiates between the reference scenes than between the new scenes. This strongly differential response (i.e., much stronger response for one of the reference scenes than for other reference scenes) translates into overall higher SI scores. While many previous studies showed that neocortical cells in sensory areas can shift their sensory tuning with practice, these findings show that hippocampal neurons exhibit this same pattern of increasing selectivity to highly familiar stimuli. These results are particularly surprising, given previous reports that damage to the hippocampus does not impair performance of well-learned associations (Murray and Wise, 1996; Wise and Murray, 1999). These physiology studies provide new insight into the role of the hippocampus in representing well-learned information. Specifically, while lesion studies have shown that the hippocampus is not necessary for the successful retrieval of remotely learned information (Bayley et al., 2006), these findings show that that information is still nonetheless represented in the hippocampus at the single-unit level. They are consistent with a proposed role in retrieval of well-learned information, suggested by functional imaging studies in humans (Schacter et al., 1995, 1996; Nyberg et al., 1996a,b; Schacter and Wagner, 1999; Preston et al., 2004). Further studies will be needed to determine the relative roles of the hippocampus and neocortical sites and their interaction during the retrieval process.
CONCLUSIONS This review of the monkey hippocampal literature has shown that the patterns of task-related neural activity seen in the monkey hippocampus parallel those described in the rat and are consistent with major predictions of the memory space hypothesis (Eichenbaum et al., 1999). A major goal of the memory space hypothesis was to take into account the broad range of task-related activity that had been reported in the rat hippocampus, particularly the nonspatial correlates of neural activity. In this review, a broad range of responses have been shown to be present in the monkey hippocampus as well. These parallel findings in monkeys and rats emphasize the value of parallel study of animal model systems in understanding the patterns of
230 PRIMATE HIPPOCAMPUS AND PLACE REPRESENTATION neural activity that underlie memory in the hippocampus. While the review of memory-related neural activity during tasks with short delay intervals suggests that essentially all aspects of task performance are represented in the hippocampus, one of the most striking observations is the large proportion of responses selective for a particular task period (scene, delay, or test phase), but not selective for the particular stimulus being processed during those task periods. In the framework of the memory space hypothesis, the task phase–selective responses signal commonalities or ‘‘nodes’’ across trials. It is thought that this nodal information provides a network of signals that may support the flexible expression of memory. However, more information is needed before the role of these phase-selective cells is confirmed. For example, while the nodal activity appears consistent with a role in signaling the commonalities across tasks, it will be important to examine the responses of these cells on other tasks to determine how general or specific these responses are. Also, if these nodal signals are important for binding related information together in memory, it will be important to test this idea directly with a task that requires the animals to manipulate information about the commonalities across trials to further probe these cells. This review has also shown that hippocampal neurons play a critical role in both the acquisition of new conditional motor associations (Cahusac et al., 1993; Wirth et al., 2003; Frank et al., 2004) and the long-term representation of well-learned associations (Miyashita et al., 1989; Yanike et al., 2004). These findings suggest that, over time and with extensive experience, the striking changing cell activity seen during acquisition of new associations (Fig. 14–2B,C) gradually develops into an overall more highly tuned response for highly familiar stimuli (Fig. 14–4B). It remains unclear how long it takes for these more selective responses to develop or how the hippocampus interacts with the neocortex during this process, but these findings suggest that conditional motor learning tasks like the location–scene association task offer a useful paradigm by which one can start examining the neural correlates of long-term memory consolidation. In conclusion, enormous insight into the functions of the hippocampus has been gained by examining the patterns of hippocampal activity on a range of memory tasks, including not only tasks of spatial representation and memory but also tasks of object memory, olfactory memory, and auditory memory. These studies have revealed the wide range of task-related signals present in the hippocampus in both rats and monkeys. Predictions from the memory space hypothesis, developed primarily from the rat physiology
and lesion literature, also provide a powerful framework for understanding the patterns of neural activity in the monkey hippocampal literature. Taken together, these findings suggest that hippocampal activity in both rat and monkey represent the full temporal sequence of events that compose ongoing episodes, including information specific to an episode and more semantic or ‘‘nodal’’ information that is shared across related episodes. It will be important to continue the development of novel tasks that tap both the episodic and semantic signals that have been described in the hippocampus, and compare and contrast these rich hippocampal signals to the signals seen in the surrounding entorhinal, perirhinal and parahippocampal cortices. This strategy will provide an integrated picture of how these strongly interconnected medial temporal lobe areas participate in both episodic and semantic-like memory processes.
References Baker CL, Behrmann M, Olson CR (2002) Impact of learning on representation of parts and wholes in monkey inferotemporal cortex. Nat Neurosci 5:1210– 1216. Bayley PJ, Gold JJ, Hopkins RO, Squire LR (2005) The neuroanatomy of remote memory. Neuron 46:799– 810. Bayley PJ, Hopkins RO, Squire LR (2006) The fate of old memories after medial temporal lobe damage. J Neurosci 26:13311–13317. Beason-Held L, Rosene DL, Killainy RJ, Moss MB (1999) Hippocampal formation lesions produce memory impairment in the rhesus monkey. Hippocampus 9:562– 574. Brasted PJ, Bussey TJ, Murray EA, Wise SP (2002) Fornix transection impairs conditional visuomotor learning in tasks involving nonspatially differentiated responses. J Neurophysiol 87:631–633. Brasted PJ, Bussey TJ, Murray EA, Wise SP (2003) Role of the hippocampal system in associative learning beyond the spatial domain. Brain 126:1202–1223. Brown MW (1982) Effect of context on the response of single units recorded from the hippocampal region of behaviourally trained monkeys. In: Neuronal Plasticity and Memory Formation (Ajmone Marsan C, Matthies H, eds.), pp 557–573. New York: Raven Press. Buckmaster CA, Eichenbaum H, Amaral DG, Suzuki WA, Rapp PR (2004) Entorhinal cortex lesions disrupt the relational organization of memory in monkeys. J Neurosci 24:9811–9825. Bunsey M, Eichenbaum H (1996) Conservation of hippocampal memory functions in rats and humans. Nature 379:255–257.
LEARNING, MEMORY, AND MONKEY HIPPOCAMPUS
Cahusac PM, Miyashita Y, Rolls ET (1989) Responses of hippocampal formation neurons in the monkey related to delayed spatial response and object-place memory tasks. Behav Brain Res 33:229–240. Cahusac PM, Rolls ET, Miyashita Y, Niki H (1993) Modification of the responses of hippocampal neurons in the monkey during the learning of a conditional spatial response task. Hippocampus 3:29–42. Cohen NJ, Eichenbaum H (1993) Memory, Amnesia, and the Hippocampal System. Camgridge, MA: MIT Press. Colombo M, Gross CG (1994) Responses of inferior temporal cortex and hippocampal neurons during delayed matching to sample in monkeys (Macaca fascicularis). Behav Neurosci 108:443–455. Eichenbaum H, Cohen NJ (2001) From Conditioning to Conscious Recollection. New York: Oxford University Press. Eichenbaum H, Dudchenko P, Wood E, Shapiro M, Tanila H (1999) The hippocampus, memory, and place cells: is it spatial memory or a memory space? Neuron 23:209–226. Frank LM, Stanley GB, Brown EN (2004) Hippocampal plasticity across multiple days of exposure to novel environments. J Neurosci 24(35):7681–7689. Funahashi S, Bruce CJ, Goldman-Rakic PS (1989) Mnemonic coding of visual space in the monkey’s dorsolateral prefrontal cortex. J Neurophysiol 61: 331–349. Funahashi S, Bruce CJ, Goldman-Rakic PS (1990) Visuospatial coding in primate prefrontal neurons revealed by oculomotor paradigms. J Neurophysiol 63: 814–831. Funahashi S, Bruce CJ, Goldman-Rakic PS (1991) Neuronal activity related to saccadic eye movements in the monkey’s dorsolateral prefrontal cortex. J Neurophysiol 65:1464–1483. Fuster JM, Alexander GE (1971) Neuron activity related to short-term memory. Science 173:652–654. George-Francois P, Rolls ET, Robertson RG (1999) Spatial view cells in the primate hippocampus: allocentric view not head direction or eye position. Cereb Cortex 9: 197–212. Hampson RE, Heyser CJ, Deadwyler SA (1993) Hippocampal cell firing correlates of delayed-matching-tosample performance in the rat. Behav Neurosci 107: 715–739. Hampson RE, Pons TP, Stanford TR, Deadwyler SA (2004) Categorization in the monkey hippocampus: a possible mechanism for encoding information into memory. Proc Natl Acad Sci USA 101:3184–3189. Hannula DE, Tranel D, Cohen NJ (2006) The long and the short of it: relational memory impairments in amnesia, even at short lags. J Neurosci 26:8352–8359. Hori E, Tabuchi E, Matsumura N, Tamura R, Eifuku S, Endo S, Nishijo H, Ono T (2003) Representation of
231
place by monkey hippocampal neurons in real and virtual translocation. Hippocampus 13:190–196. Kentros C, Hargreaves E, Hawkins RD, Kandel ER, Shapiro M, Muller RV (1998) Abolition of long-term stability of new hippocampal place cell maps by NMDA receptor blockade. Science 280:2121–2126. Kobatake E, Wang G, Tanaka K (1998) Effects of shapediscrimination training on the selectivity of inferotemporal cells in adult monkeys. J Neurophys 80: 324–330. Liu Z, Richmond BJ (2000) Response differences in monkey TE and perirhinal cortex: stimulus association related to reward schedules. J Neurophysiol 83: 1677–1692. Logothetis NK, Pauls J (1995) Psychophysical and physiological evidence for viewer-centered object representations in the primate. Cereb Cortex 3:270–288. Manns JR, Hopkins RO, Squire LR (2003) Semantic memory and human hippocampus. Neuron 38: 127– 133. Matsumura N, Nishijo H, Tamura R, Eifuku S, Endo S, Ono T (1999) Spatial- and task-dependent neuronal responses during real and virtual translocation in the monkey hippocampal formation. J Neurosci 19:2381– 2393. McEchron MD, Disterhoft JF (1997) Sequence of single neuron changes in CA1 hippocampus of rabbits during acquisition of trace eyeblink conditioned responses. J Neurophysiol 78:1030–1044. Meunier M, Bachevalier J, Mishkin M, Murray EA (1993) Effects on visual recognition of combined and separate ablations of the entorhinal and perirhinal cortex in rhesus monkeys. J Neurosci 13:5418–5432. Meunier M, Murray EA, Bachevalier J, Mishkin M (1990) Effects of perirhinal cortical lesions on visual recognition in monkeys. Soc Neurosci Abstr 16:616. Miller EK, Erickson CA, Desimone R (1996) Neural mechanisms of visual working memory in prefrontal cortex of the macaque. J Neurosci 16: 5154–5167. Mishkin M (1978) Memory in monkeys severely impaired by combined but not by separate removal of amygdala and hippocampus. Nature 273:297–298. Miyashita Y, Rolls ET, Cahusac PM, Niki H, Feigenbaum JD (1989) Activity of hippocampal formation neurons in the monkey related to a conditional spatial response task. J Neurophys 61:669–678. Mogami T, Tanaka K (2006) Reward association affects neuronal responses to visual stimuli in macaque te and perirhinal cortices. J Neurosci 26:6761–6770. Moody SL, Wise SP, di Pellegrino G, Zipser DA (1998) A model that accounts for activity in primate frontal cortex during a delayed matching to sample task. J Neurosci 18:399–410. Morris RGM, Frey U (1997) Hippocampal synaptic plasticity: role in spatial learning or the automatic
232 PRIMATE HIPPOCAMPUS AND PLACE REPRESENTATION recording of attended experience? Philos Trans R Soc Lond B Biol Sci 352:1489–1503. Murray EA, Bussey TJ, Wise SP (2000) Role of prefrontal cortex in a network for arbitrary visuomotor mapping. Exp Brain Res 133:114–129. Murray EA, Wise SP (1996) Role of the hippocampus plus subjacent cortex but not amygdala in visuomotor conditional learning in rhesus monkeys. Behav Neurosci 110:1261–1270. Nyberg L, McIntosh AR, Cabeza R, Habib R, Houle S, Tulving E (1996a) General and specific brain regions involved in encoding and retrieval of events: what, where, and when. Proc Natl Acad Sci USA 93:11280– 11285. Nyberg L, McIntosh AR, Houle S, Nilsson LG, Tulving E (1996b) Activation of medial temporal structures during episodic memory retrieval. Nature 380: 715–717. O’Keefe J, Dostrovsky J (1971) The hippocampus as a spatial map. Preliminary evidence from unit activity in the freely moving rat. Brain Res 34:171–175. O’Keefe J, Nadel L (1978) The Hippocampus as a Cognitive Map. New York: Oxford University Press. Olson IR, Moore KS, Stark M, Chatterjee A (2006a) Visual working memory is impaired when the medial temporal lobe is damaged. J Cogn Neurosci 18:1087– 1097. Olson IR, Page K, Moore KS, Chatterjee A, Verfaellie M (2006b) Working memory for conjunctions relies on the medial temporal lobe. J Neurosci 26:4596–4601. Otto T, Eichenbaum H (1992) Neuronal activity in the hippocampus during delayed nonmatch to sample performance in rats: evidence for hippocampal processing in recognition memory. Hippocampus 2:323– 334. Preston AR, Shrager Y, Dudukovic NM, Gabrieli JD (2004) Hippocampal contribution to the novel use of relational information in declarative memory. Hippocampus 14:148–152. Pribram KH, Mishkin M, Rosvold HE, Kaplan SJ (1952) Effects on delayed-response performance of lesions of dorsolateral and ventromedial frontal cortex of baboons. J Comp Physiol Psychol 45:565–575. Quiroga RQ, Reddy L, Kreiman G, Koch C, Fried I (2005) Invariant visual representation by single neurons in the human brain. Nature 435:1102–1107. Ranganath C, Cohen MX, Brozinsky CJ (2005) Working memory maintenance contributes to long-term memory formation: neural and behavioral evidence. J Cogn Neurosci 17:994–1010. Ranganath C, D’Esposito M (2001) Medial temporal lobe activity associated with active maintenance of novel information. Neuron 31:865–873. Rank JBJr (1973) Studies on single neurons in dorsal hippocampal formation and septum in unrestrained rats. Exp Neurol 41:531.
Riches IP, Wilson FA, Brown MW (1991) The effects of visual stimulation and memory on neurons of the hippocampal formation and the neighboring parahippocampal gyrus and inferior temporal cortex of the primate. J Neurosci 11:1763–1779. Rolls ET, Cahusac PM, Feigenbaum JD, Miyashita Y (1993) Responses of single neurons in the hippocampus of the macaque related to recognition memory. Exp Brain Res 93: 299–306. Rolls ET, Miyashita Y, Cahusac PM, Kesner RP, Niki H, Feigenbaum JD, Bach L (1989) Hippocampal neurons in the monkey with activity related to the place in which a stimulus is shown. J Neurosci 9: 1835–1845. Rolls ET, O’Mara SM (1995) View-responsive neurons in the primate hippocampal complex. Hippocampus 5:409–424. Rolls ET, Xiang JZ (2005) Reward-spatial view representations and learning in the primate hippocampus. J Neurosci 25:6167–6174. Rolls ET, Xiang J, Franco L (2005) Object, space, and object-space representations in the primate hippocampus. J Neurophysiol 94:833–844. Rupniak NM, Gaffan D (1987) Monkey hippocampus and learning about spatially directed movements. J Neurosci 7:2331–2337. Schacter DL, Alpert NM, Savage CR, Rauch SL, Albert MS (1996) Conscious recollection and the human hippocampal formation: evidence from positron emission tomography. Proc Natl Acad Sci USA 93:321–325. Schacter DL, Reiman E, Uecker A, Polster MR, Yun LS, Cooper LA (1995) Brain regions associated with retrieval of structurally cohrent visual infomation. Nature 376:587–590. Schacter DL, Wagner AD (1999) Medial temporal lobe activations in fMRI and PET studies of episodic encoding and retrieval. Hippocampus 9:7–24. Scoville WB, Milner B (1957) Loss of recent memory after bilateral hippocampal lesions. J Neurol Neurosurg Psychiatry 20:11–21. Sigala N, Gabbiani F, Logothetis NK (2002) Visual categorization and object representation in monkeys and humans. J Cogn Neurosci 14:187–198. Suzuki WA, Eichenbaum H (2000) The neurophysiology of memory. Ann NY Acad Sci 911:175–191. Suzuki WA, Zola-Morgan S, Squire LR, Amaral DG (1993) Lesions of the perirhinal and parahippocampal cortices in the monkey produce long-lasting memory impairment in the visual and tactual modalities. J Neurosci 13:2430–2451. Victor M, Agamanolis D (1990) Amnesia due to lesions confined to the hippocampus: a clinical-pathologic study. J Cogn Neurosci 2:246–257. von Bechterew WV (1900) Demonstration eines Gehirns mit Zerstorung der vorderen und inneren Theile der
LEARNING, MEMORY, AND MONKEY HIPPOCAMPUS
Hirnrinde bieder Schlafenlappen. Neurologisches Centralblatt 19: 900–991. Watanabe T, Niki H (1985) Hippocampal unit activity and delayed response in the monkey. Brain Res 325: 241–254. Wiebe S, Staubli U (1999) Dynamic filtering of recognition memory codes in the hippocampus. J Neurosci 19:10562–10574. Wilson FA, Riches IP, Brown MW (1990) Hippocampus and medial temporal cortex: neuronal activity related to behavioural responses during the performance of memory tasks by primates. Behav Brain Res 40:7–28. Wilson FAW, Brown MW, Riches IP (1988) Neuronal activity in the inferomedial temporal cortex compared with that in the hippocampal formation. In: Cellular Mechanisms of Conditioning and Behavioral Plasticity (Woody CD, Alkon DL, McGaugh JL, eds.), pp 313–328. New York: Plenum Press. Wilson MA, Mcnaughton BL (1993) Dynamics of the hippocampal ensemble code for space. Science 261: 1055–1058. Wirth S, Yanike M, Frank LM, Smith AC, Brown EN, Suzuki WA (2003) Single neurons in the monkey hippocampus and learning of new associations. Science 300:1578–1581. Wise SP, Murray EA (1999) Role of the hippocampal system in conditional motor learning: mapping antecedents to action. Hippocampus 9:101–117. Wood ER, Dudchenko P, Eichenbaum H (1999) The global record of memory in hippocampal neuronal activity. Nature 397:613–616.
233
Yanike M, Wirth S, Suzuki WA (2004) Representation of well-learned information in the monkey hippocampus. Neuron 42:477–487. Zola-Morgan S, Squire LR (1985) Medial temporal lesions in monkeys impair memory on a variety of tasks sensitive to human amnesia. Behav Neurosci 99:22– 34. Zola-Morgan S, Squire LR, Amaral DG (1986) Human amnesia and the medial temporal region: enduring memory impairment following a bilateral lesion limited to field CA1 of the hippocampus. J Neurosci 6:2950–2967. Zola-Morgan S, Squire LR, Amaral DG (1989a) Lesions of the amygdala that spare adjacent cortical regions do not impair memory or exacerbate the impairment following lesions of the hippocampal formation. J Neurosci 9:1922–1936. Zola-Morgan S, Squire LR, Amaral DG, Suzuki WA (1989b) Lesions of perirhinal and parahippocampal cortex that spare the amygdala and hippocampal formation produce severe memory impairment. J Neurosci 9:4355–4370. Zola-Morgan S, Squire LR, Ramus SJ (1994) Severity of memory impairment in monkeys as a function of locus and extent of damage within the medial temporal lobe memory system. Hippocampus 4:483– 495. Zola SM, Squire LR, Teng E, Stefanacci L, Buffalo EA, Clark RE (2000) Impaired recognition memory in monkeys after damage limited to the hippocampal region. J Neurosci 20:451–463.
This page intentionally left blank
III NEURAL SYSTEMS PERSPECTIVE ON THE SIGNIFICANCE OF PLACE FIELDS
This page intentionally left blank
15 Entorhinal Grid Cells and the Neural Basis of Navigation MARIANNE FYHN, TRYGVE SOLSTAD, AND TORKEL HAFTING
The seminal discovery of place cells inspired the hypothesis that the hippocampus is the locus of an allocentric spatial map in which episodic events are embedded (O’Keefe and Nadel, 1978). Whereas there is ample evidence for the role of the hippocampus in spatial memory, there has been controversy over the neural basis for navigation (McNaughton et al., 1996; O’Keefe, 1999; Eichenbaum, 2000; Redish, 2001). Accumulating evidence suggests that the functions of place cells extend well beyond a specific role in mapping the physical space (Eichenbaum et al., 1999). Place cells have been shown to respond to salient nonspatial information and to make multiple representations of the same location (Markus et al., 1995; Wood et al., 2000; Fyhn et al., 2002; Leutgeb et al., 2004; 2005a,b; Moita et al., 2004; Wills et al., 2005). These and other observations suggest that the hippocampus associates spatial and nonspatial features of an environment or event into episodic memories, possibly leaving spatial computations to adjacent structures. One synapse upstream of the hippocampus, information about position, direction, speed, and self-motion converges in medial entorhinal cortex (MEC). Accordingly, principal neurons in this area express metric, positional, and directional information, activity suggesting that MEC plays a key role in navigation and spatial processing (Fyhn et al., 2004, 2007; Hafting et al., 2005; Sargolini et al., 2006). In this chapter we will review current knowledge about spatial representations in MEC, its possible role in navigation, and how cell ensembles in MEC might
contribute to the spatial component of hippocampal place cell representations.
SPATIAL REPRESENTATION IN MEDIAL ENTORHINAL CORTEX A dynamic representation of space is dependent on the integration of information about position, direction, and distance. In the mammalian brain this is likely to rely on a distributed brain network including MEC as a central component. The entorhinal cortex can be divided into a lateral (LEC) and a medial (MEC) subdivision based on cytoarchitecture and differences in connectivity to other cortices (Witter and Amaral, 2004). The LEC is strongly innervated by the perirhinal cortex and, directly or indirectly, the frontal, piriform, insular, olfactory, and temporal cortices (Deacon et al., 1983; Burwell and Amaral, 1998). Neurons in this region have modest or no spatial correlates but respond to nonspatial aspects of the environment (Hargreaves et al., 2005), which suggests that LEC processes nonspatial contextual information. In contrast, MEC receives most of the visuospatial information emanating from the occipital, retrosplenial, and parietal cortices directly or through the postrhinal cortex (Burwell, 2000). Furthermore, MEC is the main terminal of dense projections from the head-direction system of the dorsal presubiculum (van Groen and Wyss, 1990), and it receives input from parasubiculum, which communicates information about head
237
238
NEURAL SYSTEMS PERSPECTIVE ON PLACE FIELDS
direction and movement through space from the retrosplenial cortex and anterior thalamus (van Groen and Wyss, 1990; Chen et al., 1994; Taube, 1995; Witter and Amaral, 2004). The MEC is therefore favorably situated to integrate spatial information. Contrary to this proposition, early physiological recordings from MEC revealed neurons with only moderate to weak spatial modulation, pointing to intrahippocampal computation as the origin of spatial representations (Barnes et al., 1990; Mizumori et al., 1992; Quirk et al., 1992; Frank et al., 2000). However, anatomical studies have established that the entorhinal cortex can be divided into three fairly non-overlapping bands, running perpendicular to the mediolateral subdivision: the dorsolateral band, the intermediate band, and the ventromedial band (Witter et al., 1989b; Dolorfo and Amaral, 1998b). The three bands differ in both their input and output projections. The dorsal-to-ventral organization of the bands corresponds to the septal-to-temporal (dorsal-to-ventral) gradient of their terminations along the longitudinal axis of the hippocampus (Dolorfo and Amaral, 1998b). Most of the visuospatial input to MEC arrives in the medial part of the dorsolateral band. This is also the area projecting to the dorsal hippocampus, which shows the most confined place fields and plays a more central role in spatial learning than the ventral hippocampus (Moser et al., 1993; Moser and Moser, 1998). A re-examination of the spatial modulation of neurons in the three bands demonstrated strong spatial modulation of neurons recorded from the most dorsocaudal part of MEC, and gradually weaker spatial specificity in neurons at increasingly ventral positions (Fyhn et al., 2004). The entorhinal cortex consists of four layers of principal neurons, the superficial layers (layers II and III) and deep layers (layers V and VI) (Fig. 15–1). We have identified three cell types in MEC on the basis of their spatial response patterns. The characteristic cell type of the network is the ‘‘grid cell,’’ which has multipeaked firing fields comprising the vertices of equilateral triangles that tessellate the environment (Hafting et al., 2005) (Fig. 15–2). The ensemble activity of only eight simultaneously recorded grid cells is enough to reconstruct the trajectory of a rat exploring a 1 m2 recording arena (Fyhn et al., 2004). All grid cells recorded at the same location in MEC share a common spacing and orientation of the grid, but the place of firing, the spatial phase, is randomly distributed among neurons in such a way that a small ensemble of grid cells covers the entire recording arena. In the dorsoventral axis of MEC, grid cells show a topographic organization of grid size, with spacing increasing from less than 30 cm at the most dorsal recording locations to several meters at more ventral locations (Fyhn et al., 2004; Hafting et al., 2005; Sargolini et al., 2006; K.G.
Kjelstrup, V.H. Brun, T. Solstad, T. Hafting, M. Fyhn, M.-B. Moser, and E.I. Moser, unpublished observations). How grid orientation is distributed in the grid cell network is presently an element of suspense, but there is evidence that at least grid cells from opposite hemispheres can show different grid orientations (Hafting et al., 2005; Fyhn et al., 2006). While grid cells are found in all principal layers of MEC, the proportion of grid cells of the cell population is layer specific (Sargolini et al., 2006). The second cell type in the MEC is the head direction cell (Sargolini et al., 2006) (Fig. 15–3), which is selectively active when the head of the rat is pointing in a specific direction relative to external cues, irrespective of the animal’s position. The MEC head direction cells may inherit their properties from the head direction network in the dorsal presubiculum, which holds a high density of this cell type (Taube, 1998; Witter and Moser, 2006). While no head direction cells were found in layer II, head direction cells intermingle with grid cells in layers III, V, and VI. The third cell phenotype found in layers III, V, and VI of MEC is named ‘‘conjunctive cells’’ because of mixed grid and head-direction properties (Sargolini et al., 2006; Fig. 15–3). Conjunctive cells may not comprise a separate cell type, but are rather thought to reflect the continuum of MEC cells expressing conjoint properties ranging from pure head direction cells with no spatial modulation to grid cells with directional properties. The existence of grid cells in all principal layers of MEC is somewhat surprising, considering the classic view of the entorhinal–hippocampal connectivity, where the main purpose for the superficial and deep layers of entorhinal cortex was thought to be input of neocortical and output of hippocampal information, respectively. The distribution of grid cells in all layers of MEC, however, implies that the MEC operates as an integrated structure, in accordance with studies that point to extensive intrinsic connections within the entorhinal cortex (Kohler, 1986; Dolorfo and Amaral, 1998a; Kloosterman et al., 2003; van Haeften et al., 2003). The anatomical positioning and the neuronal representation of space in MEC point to this structure as the hub in the brain network for navigation, but does MEC encompass the properties of a contextindependent spatial map?
MEDIAL ENTORHINAL CORTEX AS A UNIVERSAL SPATIAL MAP Spatial representation in the hippocampus is sensitive to the internal state of the animal and to nonspatial
ENTORHINAL GRID CELLS AND NAVIGATION
239
Figure 15–1. Overview of the anatomy of the rat entorhinal cortex. a. Sketch of rat brain with entorhinal cortex outlined in green. b. Saggital Nissl-stained section of a rat brain through the hippocampus (CA1 and dentate gyrus [DG] indicated) and medial entorhinal cortex (MEC). The green outline surrounds MEC. The different layers of MEC are indicated between the black lines and numbered II, III, V and VI. c. Schematic presentation of information flow in the hippocampal-MEC network. Projection neurons from superficial layers (layers II and III) of MEC constitute the medial perforant pathways. Layer II projects to the dentate gyrus and CA3, which in turn projects to CA1 by the Schaffer collaterals. Layer III projects to the CA1 and subiculum (Sub). Back-projections from CA1 and subiculum convey information from the hippocampus to the deep layers (layers V and VI) of MEC. Deep layers also project to superficial layers.
properties of external stimuli (Markus et al., 1995; Wood et al., 2000; Fyhn et al., 2002; Leutgeb et al., 2004; Moita et al., 2004; Wills et al., 2005). This suggests that positional information is computed outside of the hippocampus, because this information is likely to rely on algorithms that integrate self-motion information into a metric and directionally oriented representation that is valid in all contexts. The observed properties of spatial representation in MEC suggest that it holds the information of such a universal spatial map. First, the firing locations of grid cells are stable in repeated exposures to the same enclosure across sessions, days, and weeks (Fyhn et al., 2004; Hafting et al., 2005) Retrieval of the same grid vertex positions at repeated trials is not dependent on where the rat is released from in the recording en-
closure, a finding suggesting that the grids are anchored to external cues. This is further illustrated by rotating a polarizing cue-card in the enclosure where all grid, conjunctive, and head direction cells follow the rotation (Hafting et al., 2005; Sargolini et al., 2006). Simultaneously recorded grid, head direction, and conjunctive cells rotate in synchrony, indicating a coherent representation of space. Even though grid cells are strongly influenced by external cues, grids are not dependent on distal (visual) landmarks, neither for maintenance of the firing fields in a familiar environment nor for the establishment of grids in a novel environment. Removal of all visual inputs does not affect the grid-like firing pattern, although the spatial stability of the fields shows a moderate decrease in darkness (Hafting et al., 2005). A decrease in spatial
240
NEURAL SYSTEMS PERSPECTIVE ON PLACE FIELDS
Figure 15–2. The key principal cell type of medial entorhinal cortex (MEC) in the grid cell, which shows a repetitive triangular firing pattern in two-dimensional environments. Left: Saggital parvalbuminstained section of a rat brain through the MEC. Postrhinal cortex (POR) and CA1 are indicated. Grid cells show a topographic organization with increasing spacing and field size from dorsal to ventral in MEC. Right: The firing fields of one representative grid cell are shown from each recording location (arrows). Spikes (red dots) are superimposed on the rat’s trajectory (black lines).
correlation in darkness compared to the light condition is consistent with a role for allocentric cues in aligning the grid with the external reference frame. The persistence of the grid-like firing pattern in darkness suggests that the grid is to a large extent based on self-motion cues. Grid fields seem to appear immediately in a novel environment, and the fields stabilize within minutes. This is also the case when rats are
Figure 15–3. Grid cells exist in all principal cell layers of medial entorhinal cortex and are intermingled with conjunctive and head-direction cells in layers III, V and VI. One example of each cell type is shown. First row (starting form left): Trajectory (black) with superimposed spike locations (red dots); second row: color-coded rate map with peak rate indicated. Red is maximum, dark blue is zero. Pixels not covered are white. Third row: Autocorrelation matrix for the rate map. The color scale is from blue (r¼1) through green (r¼0) to red (r¼1). The distance scale of the autocorrelogram is the same as for the rate map; only the central part of the autocorrelogram is shown. Fourth row: Polar plots indicating strong directional tuning of firing rate of the conjunctive and head-direction cells, and no directional tuning of the grid cell.
introduced to a novel environment in total darkness (Hafting et al., 2005). Grid cells are omni-directional in an open field, indicating that they code for the explicit location in the environment, since the sensory input is dependent on the direction from which the rat enters the firing field (Fyhn et al., 2004). The directional information of MEC is expressed in the headdirection and conjunctive cells.
ENTORHINAL GRID CELLS AND NAVIGATION
The described properties of the spatial representation in MEC suggest that it is generated by self-motion cues, but updated by allocentric cues. This pattern resembles models of path integration (McNaughton et al., 1996; Redish and Touretzky, 1997). Path integration, or dead reckoning, is the representation of location relative to a reference point by which animals continuously add up small increments in distance and direction as they move away from an origin. The resulting self-motion-based vector is derived independently of external landmarks, and errors accumulate with distance from the origin. Landmarks or position fixes may serve as reference points for the path integrator, resetting it as error accumulates. In contrast, geometric navigation refers to a process whereby position is estimated from landmarks in a triangulation process. This kind of mapping of the environment depends heavily on the particular external objects, and the relation between the landmarks must be learned for every environment. Hafting et al. (2005) hypothesized that MEC continuously computes the animal’s position from the self-motion-generated grid cell map while stored (learned) representations in the hippocampus serve as reference points and frequently reset and calibrate the grid cells. Another unresolved question is whether the long-term stability of the grid cell map in a given environment hinges on the hippocampus, entorhinal cortex, or other brain areas. To update its spatial representation with the movement of the animal, the grid cell network must shift from the active grid-cell population sharing a spatial phase to a grid cell population with an adjacent spatial phase in the direction of movement. This shift can be realized through the following information exchange. The omnidirectional grid cells of layer II project to conjunctive cells (unidirectional grid cells) in layers III and V, implicitly sharing the same spatial phase as the active layer II cells. The conjunctive cells, possibly assisted by the head direction system, in turn project asymmetrically back to layer II grid cells with a spatial phase shifted in the preferred direction of the conjunctive cell, effectively shifting the network activity from the current grid cell population to a population with an adjacent spatial phase (Sargolini et al., 2006). Conceptually, this does not assume any specific topography in the neuronal organization; the relative difference in spatial phases between grid cells is determined by the hardwired connections with conjunctive cells and is therefore conserved for all environments explored by the animal. The intrinsic network of MEC points to strong communication within and between superficial and deep layers. The layer II network shows extensive inhibitory recurrent connectivity, but excitatory connections between stellate cells were not detected until
241
recently (Kumar et al., 2007). Neurons in the deep layers have their apical dendrites in layer I and synapse on local axons of cells in the superficial layers, indicating that the deep layers have access to information processed in the superficial layers (Iijima et al., 1996). Thus, it is conceivable that grid cells in the deep layers inherit their spatial pattern from the principal cells of layer II. Conversely, the deep layers also project strongly to the superficial layers. Axons from layer V cells ascend to the pial surface and branch widely when entering the upper three layers (Kloosterman et al., 2003; van Haeften et al., 2003).
INTERACTIONS BETWEEN THE MEDIAL ENTORHINAL CORTEX AND HIPPOCAMPUS From Grid Cells to Place Cells Assuming that positional information in MEC is used to construct context-specific spatial maps in the hippocampus, place cells face the problem of extracting the unique position of the rat from the periodic representation of grid cells. Theoretical considerations suggest that this can be achieved by a dendritic summation of inputs from an appropriate selection of grid cells, and shed light on how the grid cell and place cell networks could be connected. Such a mechanism for place field formation is not dependent on computation by the intrahippocampal network and allows fast readout of positional information (Solstad et al., 2006). The perforant path, originating in layer III of the entorhinal cortex, terminates on the apical dendrites of hippocampal (CA1) pyramidal neurons (Witter and Amaral, 2004). Several studies indicate that hippocampal pyramidal neurons can perform dendritic summation of these inputs (Cash and Yuste 1998, 1999; Gasparini and McGee, 2006). Lesion studies also indicate that the perforant path inputs are sufficient for generating CA1 place fields (Brun et al., 2002). If we approximate the triangular grid-cell pattern by a sum of three two-dimensional cosine functions with a 608 relative angular difference (Fig. 15–4a), the selection of grid cells that a place cell must receive input from in order to form a given place field by linear summation is given by Fourier theory. Theoretically, a Gaussian place field requires input from a continuum of grid cells that cover a limited range of spacings and all orientations and share a coincident activity peak (Fig. 15–4b). Thus, the position of a given place field is defined by the position of the coincident activity peaks of the projecting grid cells, and the size of the place field depends only on the range of grid cell spacings that dominate the inputs (Solstad et al., 2006). These parameters are consistent with the observation
242
NEURAL SYSTEMS PERSPECTIVE ON PLACE FIELDS
that the spatial scale of both grid and place cells increases systematically from dorsal to ventral in the two structures (Jung et al., 1994; Hafting et al., 2005; Maurer et al., 2005; Kjelstrup et al., 2006). The model also corresponds well to the anatomical connectivity between MEC and hippocampus, as the dorsal part of MEC projects to the dorsal hippocampus, and ventral MEC projects to ventral hippocampus. Furthermore, projections from a single MEC cell cover only about 25% of the hippocampal dorsoventral axis (Dolorfo and Amaral, 1998b), which in effect limits the range of grid spacings terminating on each place cell. Building on the relationship between place field size and topography in grid spacing, the model predicts that place field size can be modulated by inactivation of either dorsal or ventral parts of the grid cell network. Going into finer anatomical detail, we can assume that grid orientation varies along the dorsoventral axis of MEC, and that the dorsal-most portion of the hip-
Figure 15–4. A model for place-field formation from grid cells. a. Grid functions are constructed from a sum of three sinusoidal grating functions with 608 and 1208 angular difference, and can take any specified spatial phase, orientation, and spacing. b. Assumed anatomical connectivity between grid cells in medial entorhinal cortex (MEC) and place cells in the hippocampus (HPC). Grid cells (blue) are illustrated with the topographic organization observed in vivo (Hafting et al., 2005) with increasing grid spacings from dorsal to ventral levels. All place cells with a place field receive input from grid cells of similar spatial phase (a common central peak) but a diversity of spacings and orientations. Hippocampal place cells with a small firing field (green) are innervated by grid cells from more dorsal parts of the MEC than place cells with a larger field (yellow). Connection weights are indicated by the thickness of the arrows. Interneurons (red) provide nonspecific inhibition to keep overall firing rates at physiological levels. The color code for the rate maps ranges from blue (0 Hz) to red (peak rate). From Solstad et al. (2006), with permission.
pocampus receives input from grid cells with spacing in the range of 28–73 cm (smallest and largest spacing reported from the dorsal-most 25% of MEC; Hafting et al., 2005). In addition, a single place cell is likely confined to being innervated by somewhere between 10 and 100 MEC neurons (Amaral et al., 1990). Investigating this biological scenario with a linear firingrate model shows that less than 50 afferent grid cells gives rise to place cells with single-peaked place fields in environments far larger than conventional experimental environments (at least 10 10 m), even if limited variation in the coincident grid activity-peak positions is allowed (Solstad et al., 2006). Thus, it seems sufficient for place cells to sum dendritic input from a relatively small set of synchronized grid cells with different spacings and orientations to produce localized place fields. The model only addresses the mechanism for transforming grid patterns into a localized position estimate,
ENTORHINAL GRID CELLS AND NAVIGATION
and does not explain how a place cell selects grid cells with coincident activity peaks. That is, how are appropriate synaptic weights between grid cells and place cells set up when entering a novel environment? Whether a local learning rule is sufficient for establishing new hippocampal maps remains an open question, however; if modeled place cells have a low selectivity for equal spatial phases, they exhibit multipeaked activity patterns reminiscent of principal cells in the dentate gyrus (Jung and McNaughton, 1993; Leutgeb et al., 2007). This at least indicates a role for the sensitivity to coincident grid cell firing in the resulting place cell representation. An alternative mechanism of producing place fields is through processing of entorhinal inputs by a competitive network. The high degree of lateral inhibitory connections in the dentate and CA1 has been taken to support the view that competitive network dynamics govern the behavior of these areas (Rolls and Treves, 1998). Simulations of such networks have been shown to produce units with naturalistic place fields in the face of hypothetical cortical cells with sensory responses to elements of the explored environment (Sharp et al., 1996). Modeling a similar network with input from a random set of grid cells gives rise to populations of place units where about one-third exhibit single-peaked place fields and the majority exhibit multiple fields, mimicking the behavior of the dentate network (Rolls et al., 2006). However, that such a network can also produce place cells resembling those of CA3 and CA1 has yet to be demonstrated (Franzius et al., 2007). If place fields constitute the result of a readout mechanism from the grid cell network, a change of MEC representation necessitates an immediate change in the hippocampal representation of space. Feedforward models of the MEC and hippocampus thus predict that spatial representations for different environments in the two networks are tightly coupled in both space and time, and that spatial changes in one network representation should be highly predictive of spatial changes in the other network. This leads us to the question of how cell ensembles in MEC and hippocampus interact in vivo.
Coherent Network Dynamics in the Medial Entorhinal Cortex and Hippocampus The ability to decorrelate overlapping input patterns before information is stored is a fundamental property of many associative memory networks (Marr, 1971). In the hippocampus, such neuronal pattern separation is expressed in the tendency of place cell ensembles to undergo extensive ‘‘remapping’’ in response to chan-
243
ges in sensory or motivational inputs to the hippocampus. The term describes the phenomenological observation that place cells have different patterns of firing in different environments or states. Hippocampal cell ensembles show two distinct kinds of remapping depending on the task used to induce the phenomenon (Leutgeb et al., 2005a,b; Fyhn et al., 2007). Global remapping describes conditions under which there is a complete reorganization of the hippocampal place code so that both place and rate of firing take statistically independent values (Leutgeb et al., 2005b). Under other conditions, the firing fields remain in the same locations, but there is a several-fold change of firing rate. This phenomenon has been coined ‘‘rate remapping’’ (Leutgeb et al., 2005b). Authors of theoretical work have proposed that the process of pattern separation is of intrahippocampal or dentate gyrus origin (Rolls and Treves, 1998). However, without knowledge of the dynamics of afferent structures, we cannot exclude the possibility that hippocampal remapping just reflects computations at earlier levels of processing such as the entorhinal cortex. This problem was recently addressed in a study by Fyhn et al. (2007), in which cell ensembles from MEC and hippocampus CA3 were recorded simultaneously in tasks that cause remapping in the hippocampus. In this study, ensemble dynamics of grid cells were fundamentally different from those of hippocampal place cells. When global remapping in CA3 was evoked by replacing a square box with a circular box at a constant location, all the simultaneously recorded grid cells shifted their firing fields coherently (Fig. 15–5). The coherent shift in the ensemble activity is visualized in the spatial cross-correlograms for the cell ensemble. The central peak of the cross-correlograms was offset from the origin when trials in different environments were compared (Fig. 15–5d). While the majority of CA3 cells that were active in one environment were not active in another environment, the same grid cells were active in all environments explored by the animal. When the animals were tested in two rooms with distinct distal landmarks, grid fields moved and rotated, and sometimes expanded or contracted between the two test locations. However, all simultaneously recorded grid cells moved in concert and by the same amount, maintaining the spatial phase relative to each other (Fig. 15–5e). This activity indicates that during global remapping in the hippocampus, the map of grid cells in MEC realigns with the cues in the environment without losing its intrinsic spatial-phase structure. This is in striking contrast to the orthogonalized representations of the CA3 area (Leutgeb et al., 2004, 2005a,b, 2007). The fact that grid cells may exhibit different spacings in different environments suggests that grid cells
244
NEURAL SYSTEMS PERSPECTIVE ON PLACE FIELDS
Figure 15–5. Coherent representation of grid cell ensemble during remapping in the hippocampus. a. Realignment of entorhinal grid fields during hippocampal global remapping between different boxes in the same location. Color-coded firing rate maps for three representative CA3 cells (top panels) and three MEC cells (lower panels) recorded during running in a square box (left), a circular box at the same location (middle), and the square box again (right). b,c. Schematic illustration of the procedure for cross-correlation of rate maps based on individual rate maps (b) or stacks of rate maps (c). Cross-correlation matrices were determined by shifting the two maps relative to one another in steps of 5 cm in both the x and y directions within the common stippled area. PV, population vector (defined for each 5 5 cm2 bin). d. Color-coded cross-correlation matrices for the individual grid cells in a during repeated testing in the same box (A vs. A0 ) or in different boxes (A vs. B). e. Crosscorrelation matrices for cell ensembles of simultaneously recorded grid cells recorded from one rat in the different remapping paradigms. Top panel is the grid cells from the rat in a, middle panel shows the cells from a rat tested in the two-room paradigm. Note that the stack in room B was rotated compared to the stack in room A to obtain the maximal correlation value. Lower panel shows cells from a rat tested in the rate-remapping paradigm where hippocampus only showed a change in firing rate of its place fields. Note that the central peak in the cross-correlogram is in the origin when comparing different boxes, indicating that the grid cells did change firing location. Adapted from Fyhn et al. (2007), with permission.
are not completely hardwired Euclidian metrics. The experience-dependent plasticity of MEC, and of grid cells in particular, demands further investigation. Comparing the spatial firing pattern in a box with that in a circle is not a trivial task; for example, the same physical point in space might be at a different distance to the walls in the two conditions. Therefore, the rigidity of ensemble activity was verified by temporal analysis. Unlike place cells in CA3, entorhinal grid cells that are coactive in one environment remain
coactive in other environments, as predicted, if their spatial phase relationship is preserved. Thus, the network of entorhinal grid cells seems likely to be quite hardwired compared to the dynamic network of CA3, which responds to each environment with a different subpopulation of active cells. Furthermore, the differentiation of the coactivity pattern of CA3 cells across environments reinforces the view that incoming sensory input is decorrelated as it enters the hippocampus (Leutgeb et al., 2007).
ENTORHINAL GRID CELLS AND NAVIGATION
When hippocampal rate-remapping was evoked by a reversal of box colors, the spatial phase and orientation of the grid fields remained stable (Fyhn et al., 2007). No systematic change in the firing rate of entorhinal subfields was detected under the rate-remapping paradigm. Thus, the orthogonalization observed as rate remapping is most likely specific to the dentate gyrus and CA3 (Leutgeb et al., 2005, 2007). Rate remapping may thus be a consequence of the convergence of spatial information from grid cells and nonspatial information from LEC (Naber et al., 1997; Hargraves et al., 2005). Since global remapping can be induced by a change of behavioral task or selected features of the local environment (Markus et al., 1995; Quirk et al., 1990; Wood et al., 2000; Wills et al., 2005), it is possible to test if global remapping in the hippocampus and grid realignment in the entorhinal cortex coincide. If global remapping is mechanistically related in the two areas one would expect strong temporal contiguity of the phenomena. It is possible to induce two independent hippocampal maps for the same recording enclosure by training rats in discrete sessions in light and darkness in the specific enclosure (Quirk et al., 1990). Fyhn et al. (2007) demonstrated a coherent displacement and rotation in grid cell representations between light and dark sessions. To evoke instantaneous global remapping on the test day, the room lights were turned on after 11 min of running in the dark condition (Fig. 15– 6). This caused two different scenarios in different rats—either instantaneous global remapping in CA3 when darkness was terminated or no remapping but an immediate reversal to the original hippocampal map after 1 min disruption of running. In both scenarios however, global remapping in the hippocampus was accompanied by a coherent shift and rotation of the population of grid cells in MEC. In spite of individual differences in the response to the sudden change in light condition, this experiment shows that, irrespective of response, hippocampal and MEC representations are strongly correlated. The coincidence of the changes in entorhinal and hippocampal population dynamics strongly suggests that global remapping in the hippocampus and realignment of the grid representation in the entorhinal cortex are part of a single integrated process. However, the strong contiguity of grid realignment and global remapping invokes the need for analysis on a finer timescale to search for causal relations between the two phenomena. The different response in MEC to the global and rate remapping paradigms points to grid cell dynamics as a possible determinant of global remapping in hippocampal place cells. However, the entorhinal cortex and hippocampus are anatomically connected in a loop and are likely to interact as an integrated system. Thus re-
245
mapping and realignment might influence each other bidirectionally. The coherent response of grid cells to the manipulations makes the network uninformative about which environment or context the rat is in, but rather provides accurate information of the rat’s position within a given environment. This rigidity of the network may be analogous to that of the head direction system, where the relative firing directions between cells are retained across environments (Taube, 1998). This analogy is also supported by the finding that simultaneously recorded grid cells from both hemispheres show a synchronous shift and rotation of the grids in the dark–light paradigm (Fyhn et al., 2007) as well as between two different rooms (Fyhn et al., 2006, 2007). Both a synchronous realignment of all grid cells in MEC and different realignments of independent modules of grid cells would induce global remapping in hippocampus in at least two alternative ways (Fyhn et al., 2007; Fig. 15–7). In the first model, the grid cell ensemble works as one coherent unit, representing the entire range of grid spacings and orientations (Fig. 15– 7a). This would represent an infinite map of which the recording environment would cover a particular portion. A sufficient change in the portion of the map due to the environmental manipulation would activate a different part of the map. If the displacement extends beyond the portion of the MEC map already activated, an unrelated population of place cells is recruited. It is possible that the rat has to walk by itself from rooms A to B to maintain this infinite map and the theory could easily be tested. In the second model, place cells receive input from several independent modules in MEC, each with a different spacing and orientation of the grids. Global remapping would be induced in the hippocampus if the modules shift, rotate, or expand independently, causing a change in the specific subset of grid cells that make a place cell reach excitation threshold (Fig. 15–7b). It remains for future studies to address whether the grid cell ensemble works as one coherent unit across all environments and contexts or whether there are independent modules that respond differentially to changes in the environment.
Models of the Grid Cell Network The remarkable spatial structure of the entorhinal grids easily makes us marvel at the brain’s abilities, albeit in other areas of biology spatially periodic patterns are nothing short of common. The seemingly universal tendency of spatially periodic organization has inspired a group of models describing biological morphogenesis in a variety of systems ranging from chemicals to animal coats and neural tissue (McNaughton et al., 2006). Local activation and long-range inhibition
Figure 15–6. Global remapping in the hippocampus coincides with grid realignment in medial entorhinal cortex (MEC). To promote the establishment of two independent maps for the same enclosure in light and darkness, rats were trained with room lights on and off on alternating discrete trials. On the test day, the room lights were turned on after 11 min of running in darkness (dark gray box). After 20 min, the rat was lifted out of the box for 1 min before it was returned to the box (double arrow). For both MEC and CA3, the stability of spatial firing is displayed as the spatial correlation between the rate maps for each block of 1 min and the total rate map for the same parts of the box during baseline recording with lights on (before darkness) (means ± SEM). Rat 11554 remapped when the lights were turned on; rat 11580 remapped after being lifted out of the box. Note that spatial correlations dropped and reversed simultaneously in MEC and CA3 in both rats, consistent with the idea that grid realignment and global remapping are mechanistically linked. Adapted from Fyhn et al. (2007), with permission. 246
ENTORHINAL GRID CELLS AND NAVIGATION
247
Figure 15–7. Two possible models for global remapping in the hippocampus during realignment of entorhinal grid fields. a. Multimap model. This model assumes independent modules in medial entorhinal cortex (MEC), each consisting of grid cells with a unique spacing and orientation. During hippocampal global remapping, these modules exhibit different degrees of rotation and translation, yielding new patterns of coactivity between the modules. If place fields are excited by convergent input from multiple entorhinal modules, the altered coactivity will activate a different subset of place cells at each location in the environment. Two environments (square boxes) and three grid fields (pink, green, and orange), each from a different module, are indicated at the top. Cross-correlations for each module are shown at the bottom. The three grid cells provide sufficient input to a particular CA3 cell (white spot), such that it fires when, and only when, the nodes of the three grids overlap. This occurs only at one location and only in one of the recording boxes (the left box) in this example. If the nodes of the grids overlapped at more than one location, the points of overlap would usually be at different places in the two boxes (global remapping). b. Single-map model. This model assumes that the entire MEC expresses a single universal map of space. The recording chambers (square boxes) can be thought of as extracting distinct locales on an infinite or periodic spatial map, related to each other by translation and rotation (white dotted arrow). Three grid fields are shown, as in a. Note that if the second chamber had been larger, it could have included the conjuction of entorhinal grid peaks that generated the CA3 place field in the original condition (the white spot). In the single-map model (b, bottom), the optimal rotation of the cross-correlations between local population vectors is identical for ensembles of grid units with different spacing and orientation. Thus, while both models predict differential shifts for grid cells in different regions of the MEC (the arrows in the cross-correlograms), differences in rotation are only predicted by the independent-modules model (a). Note that while hippocampal remapping is controlled by grid shifts in the entorhinal cortex in these models, remapping may also reversely influence the alignment of the grid map. Adapted from Fyhn et al. (2007), with permission. constitute the main requirement for stable pattern formation and have a neuronal counterpart in attractor networks implementing the Mexican hat–type connectivity. Continuous attractor networks are recurrent networks exhibiting stable activity states and imitate the behavior of content-addressable or associative memories that can represent continuous space (Samsonovich and McNaughton, 1997). For this reason, the architectural model is an especially popular model of the hippocampal area CA3 (Rolls and Treves, 1998), and it has been proposed to underlie the dynamics of the grid-cell network as well (Fuhs and Touretzky, 2006; McNaug-
hton et al., 2006). The hexagonal firing-pattern of units in the Mexican hat–type continuous attractor networks reflects the radial symmetric connections between grid cells of different spatial phases on a ‘‘neural sheet,’’ that together with the gain in speed signal determines the characteristic grid spacing of the network. Even though the continuous attractor model offers an elegant solution to the computation of path integration, the architecture forces units on the neural sheet to be topographically distributed with regard to spatial phase, and homogeneous with regard to spacing and orientation. These discrepancies from the neurophysiological data
248
NEURAL SYSTEMS PERSPECTIVE ON PLACE FIELDS
impose two additional requirements on the network. First, to implement units with different grid spacings, multiple semi-independent networks must be instantiated along the dorsoventral axis. Such a modular organization could be realized by having separate networks in separate ‘‘cortical columns’’ of the MEC (Witter and Moser, 2006). Second, although the MEC incorporates the connectivity necessary to support a Mexican hat– type network (Kumar et al., 2007), the neural sheet must be thought of as ‘‘connection space’’ to account for the lack of topography in spatial phase of biological grid cells. This means that neighboring units on the neural sheet represent neurons that are directly connected but not physical neighbors in cortex. An explanation of how the Mexican hat–type connections are set up during development therefore requires a stretch of one’s imagination. The originally simple principle of biological pattern formation needs increasingly complicating adjustments to account for the neurophysiological data, and implementation of an attractor network fully consistent with the findings reported by Hafting et al. (2005) is yet to be demonstrated (Burak and Fiete, 2006). Upcoming models of grid cell formation based on single-cell properties that can vary freely along any dimension of the cortex escape the problems of topography and unity. However, these principles have not yet been implemented at the network level (Treves et al., 2005; Burgess et al., 2007; Giocomo et al., 2007) or rest on theoretical constructs that lack experimental verification (Blair et al., 2007). The neural mechanism of grid cell formation is, therefore, still open to further inquiry.
ASSOCIATION OF SPATIAL AND NONSPATIAL INPUTS IN THE HIPPOCAMPUS The MEC and LEC represent two parallel input streams to the hippocampus conveying spatial and nonspatial information, respectively (Witter et al., 2000; Burwell et al., 2004; Fig. 15–8). In the dentate gyrus and CA3, the input streams are integrated, and metric representation from MEC can be associated with the nonspatial information from LEC into a multitude of representations that can be stored for the same environment. Through back-projections from the hippocampus to MEC, the learned memory representations can anchor the metric representation of grid cells to the environment such that the same map is retrieved every time the animal enters the same environment. As an animal enters a familiar environment, context-specific information reaching the CA3 from LEC may retrieve the memory of stored external reference points for
the particular environment, sufficient for anchoring the grid representation in MEC. The system is now ready for navigational tasks in the familiar environment. This view is in line with a study in which inactivation of the hippocampus gradually disrupted grid cell firing (Bonnevie et al., 2006), indicating that the feedback projections from the hippocampus are necessary to keep the grid cell representation from drifting (accumulating error). When the hippocampus is inactivated, no associations between space and context can be made or recalled, such that the path integrator (grid cell network) is drifting randomly and will gradually appear spatially unspecific. If the hippocampus is inactivated, MEC cannot recall which spatial representation corresponds to the current context. Accordingly, if the MEC constantly remaps, grid cells will appear spatially unspecific. It is likely that some spatial information is retained in MEC after inactivating the hippocampus, given strong inputs to MEC from upstream areas such as the pre- and parasubiculum, which contain neurons with head-direction and place properties (Caccuci et al., 2004). Fyhn et al. (2004) found that MEC neurons still showed spatial selectivity, albeit somewhat impaired, after permanent hippocampal lesioning. In addition, temporal inactivation of the hippocampus did not affect head direction cells of MEC (Bonnevie et al., 2006). The finely resolved spatial representation of MEC, however, appears to be dependent on an intact entorhinal–hippocampal loop.
Human Entorhinal Cortex and Memory Impairments in Alzheimer Disease Cross-species comparisons show that the patterns of mnemonic activity observed throughout the medial temporal lobe are largely conserved across species, and the effect of damage to these structures has been shown to be quite comparable in humans and other animals (Suzuki and Eichenbaum, 2000). In the early course of Alzheimer disease, patients commonly experience impairments in visuospatial performance, navigation, and spatial memory (Kavcic et al., 2006). The cognitive impairments can be traced to lesions in the entorhinal cortex (van Hoesen et al., 2000). Neuropathological studies have shown that during the course of aging in individuals without cognitive dementia, the number of neurons in entorhinal cortex layer II remains constant between ages 60 and 90 (Lippa et al., 1992; Gomez-Isla et al., 1996). In contrast, severe neuronal loss can be detected even in mild cases of Alzheimer disease, where symptoms are at the threshold of clinical detection of dementia. The most dramatic neuronal loss selectively targets layers II and IV of entorhinal cortex, and in severe cases of Alz-
ENTORHINAL GRID CELLS AND NAVIGATION
249
Figure 15–8. Simplified schematic presentation on how the hippocampus may associate spatial and nonspatial information from the two parallel input streams from the parahippocampal area into context-specific memory representations. The presubiculum and postrhinal cortex convey directional and visuospatial information, respectively, to the medial entorhinal cortex (MEC). The intrinsic network of MEC is thought to integrate this information onto grid and conjunctive cells to perform the process of dead reckoning. Positional information from MEC is conveyed to the hippocampus. Lateral entorhinal cortex (LEC) receives nonspatial information from perirhinal cortex, which is passed on to the hippocampal network through the lateral perforant pathway. The hippocampal network associates information from the two input streams and makes context-dependent representations of the environment or event. Stored representations in the hippocampus project back to the entorhinal cortex, thus updating incoming sensory input to the entorhinal cortex with processed hippocampal information such that the same map in entorhinal cortex is retrieved every time the animal enters the same environment.
heimer dementia the number of neurons in layer II decreases by 90% compared to that in controls (Lippa et al., 1992; Gomez-Isla et al., 1996). Given the navigational impairments that Alzheimer patients suffer and the fact that the neuronal loss is most prominent in the area homologous to where we have identified grid cells in the rat, it is tempting to speculate that that the medial entorhinal cortex plays a key function in navigation in humans as well.
CONCLUDING REMARKS The discovery of a metric representation of selflocation in MEC suggests that the primary function of the hippocampus is not the dynamic computation of location. Although the animal’s position can be predicted from the collective firing of grid cell ensembles, it remains to be determined whether readout occurs within the entorhinal cortex or in one or several of its
250
NEURAL SYSTEMS PERSPECTIVE ON PLACE FIELDS
hippocampal or parahippocampal target structures. The contextual specificity of hippocampal representations suggests that during encoding, the hippocampus associates input from the self-motion-based coordinate system in MEC with other contextual information such as information from LEC. The possible recoding of spatial information from a positional code in MEC onto statistically independent, context-sensitive cell ensembles in high-capacity networks of the hippocampus is probably crucial for the successful storage of episodic memory.
acknowledgments We thank Edvard I. Moser and Dori Derdikman for comments on an earlier version of the manuscript. This work was supported by a Centre of Excellence grant from the Norwegian Research Council.
References Amaral DG, Ishizuka N, Claiborne B (1990) Neurons, numbers and the hippocampal network. Prog Brain Res 83:1–11. Barnes CA, McNaughton BL, Mizumori SJ, Leonard BW, Lin LH (1990) Comparison of spatial and temporal characteristics of neuronal activity in sequential stages of hippocampal processing. Prog Brain Res 83:287–300. Blair HT, Welday AC, Zhang K (2007) Scale-invariant memory representations emerge from moire´ interference between grid fields that produce theta oscillations: a computational model. J Neurosci 27:3211–3229. Bonnevie T, Fyhn M, Hafting F, Moser EI, Moser M-B (2006) Misalignment of entorhinal grid fields after hippocampal inactivation. Soc Neurosci Abstr 32:68.1. Brun VH, Otnass MK, Molden S, Steffenach HA, Witter MP, Moser MB, Moser EI (2002) Place cells and place recognition maintained by direct entorhinalhippocampal circuitry. Science 296:2243–2246. Burak Y, Fiete I (2006) Do we understand the emergent dynamics of grid cell activity? J Neurosci 37:9352– 9354. Burgess N, Barry C, O’Keefe J (2007) An oscillatory interference model of grid cell firing [published online ahead of print June 27, 2007]. Hippocampus Burwell RD (2000) The parahippocampal region: corticocortical connectivity. Ann NY Acad Sci 911:25–42. Burwell RD, Amaral DG (1998) Cortical afferents of the perirhinal, postrhinal, and entorhinal cortices of the rat. J Comp Neurol 398:179–205. Burwell RD, Bucci DJ, Sanborn MR, Jutras MJ (2004) Perirhinal and postrhinal contributions to remote memory for context. J Neurosci 49:11023–11028. Cacucci F, Lever C, Wills TJ, Burgess N, O’Keefe J (2004) Theta-modulated place-by-direction cells in
the hippocampal formation in the rat. J Neurosci 38: 8265–8277. Cash S, Yuste R (1998) Input summation by cultured pyramidal neurons is linear and position-independent. J Neurosci 18:10–15. Cash S, Yuste R (1999) Linear summation of excitatory inputs by CA1 pyramidal neurons. Neuron 2:383– 394. Chen LL, Lin LH, Barnes CA, McNaughton BL (1994) Head-direction cells in the rat posterior cortex. II. Contributions of visual and ideothetic information to the directional firing. Exp Brain Res 101:24–34. Deacon TW, Eichenbaum H, Rosenberg P, Eckmann KW (1983) Afferent connections of the perirhinal cortex in the rat. J Comp Neurol 220:168–190. Dolorfo CL, Amaral DG (1998a) Entorhinal cortex of the rat: organization of intrinsic connections. J Comp Neurol 398:49–82. Dolorfo CL, Amaral DG (1998b) Entorhinal cortex of the rat: topographic organization of the cells of origin of the perforant path projection to the dentate gyrus. J Comp Neurol 398:25–48. Eichenbaum H (2000) A cortical-hippocampal system for declarative memory. Nat Rev Neurosci 1:41–50. Eichenbaum H, Dudchenko P, Wood E, Shapiro M, Tanila H (1999) The hippocampus, memory, and place cells: is it spatial memory or a memory space? Neuron 2:209–226. Frank LM, Brown EN, Wilson M (2000) Trajectory encoding in the hippocampus and entorhinal cortex. Neuron 27:169–178. Franzius M, Vollgraf R, Wiskott L (2007) From grids to places. J Comput Neurosci 22: 297–299. Fuhs MC, Touretzky DS (2006) A spin glass model of path integration in rat medial entorhinal cortex. J Neurosci 26:4366–4376. Fyhn M, Hafting T, Treves A, Moser EI, Moser MB (2006) Coherence in ensembles of grid cells. Soc Neurosci Abstr 32:68.9. Fyhn M, Hafting T, Treves A, Moser EI, Moser MB (2007) Hippocampal remapping and grid realignment in entorhinal cortex. Nature 446:190–194. Fyhn M, Molden S, Hollup S, Moser MB, Moser EI (2002) Hippocampal neurons responding to first-time dislocation of a target object. Neuron 35:555–566. Fyhn M, Molden S, Witter MP, Moser EI, Moser MB (2004) Spatial representation in the entorhinal cortex. Science 305:1258–1264. Gasparini S, Magee JC (2006) State-dependent dendritic computation in hippocampal CA1 pyramidal neurons. J Neurosci 26:2088–2100. Giocomo LM, Zilli EA, Fransen E, Hasselmo ME (2007) Temporal frequency of subtreshold oscillations scales with entorhinal grid cell spacing. Science 315:1719– 1722.
ENTORHINAL GRID CELLS AND NAVIGATION
Gomez-Isla T, Price JL, McKeel DW Jr, Morris JC, Growdon JH, Hyman BT (1996) Profound loss of layer II entorhinal cortex neurons occurs in very mild Alzheimer’s disease. J Neurosci 16:4491–4500. Hafting T, Fyhn M, Molden S, Moser M, Moser E (2005) Microstructure of a spatial map in the entorhinal cortex. Nature 436:801–806. Hargreaves EL, Rao G, Lee I, Knierim JJ (2005) Major dissociation between medial and lateral entorhinal input to dorsal hippocampus. Science 308:1792– 1794. Iijima T, Witter MP, Ichikawa M, Tominaga T, Kajiwara R, Matsumoto G (1996) Entorhinal-hippocampal interactions revealed by real-time imaging. Science 272: 1176–1179. Jung MW, McNaughton BL (1993) Spatial selectivity of unit activity in the hippocampal granular layer. Hippocampus 3:165–182. Jung MW, Wiener SI, McNaughton BL. (1994) Comparison of spatial firing characteristics of units in dorsal and ventral hippocampus of the rat. J Neurosci 14: 7347–7356. Kavcic V, Fernandez R, Logan D, Duffy CJ (2006) Neurophysiological and perceptual correlates of navigational impairment in Alzheimer’s disease. Brain 129: 736–746. Kjelstrup KG, Brun VH, Solstad T, Leutgeb S, Fyhn M, Hafting T, Moser M-B, Moser EI (2006) Spatial scale expansion along the dorsal-to-ventral axis of hippocampal area CA3 in the rat. FENS Abstr R11945. Kloosterman F, Van Haeften T, Witter MP, Lopes Da Silva FH (2003) Electrophysiological characterization of interlaminar entorhinal connections: an essential link for re-entrance in the hippocampal-entorhinal system. Eur J Neurosci 18:3037–3052. Kohler C (1986) Intrinsic connections of the retrohippocampal region in the rat brain. II. The medial entorhinal area. J Comp Neurol 246:149–169. Kumar SS, Jin X, Buckmaster PS, Huguenard JR (2007) Recurrent circuits in layer II of medial entorhinal cortex in a model of temporal lobe epilepsy. J Neurosci 27:1239–1246. Leutgeb JK, Leutgeb S, Moser M-B, Moser EI (2007) Pattern separation in dentate gyrus and CA3 of the hippocampus. Science 315:961–966. Leutgeb JK, Leutgeb S, Treves A, Meyer R, Barnes CA, McNaughton BL, Moser MB, Moser EI (2005a) Progressive transformation of hippocampal neuronal representations in ‘‘morphed’’ environments. Neuron 20: 345–358. Leutgeb S, Leutgeb JK, Barnes CA, Moser EI, McNaughton BL, Moser MB (2005b) Independent codes for spatial and episodic memory in hippocampal neuronal ensembles. Science 309:619–23.
251
Leutgeb S, Leutgeb JK, Treves A, Moser MB, Moser EI (2004) Distinct ensemble codes in hippocampal areas CA3 and CA1. Science 305:1295–1298. Lippa CF, Hamos JE, Pulaski-Salo D, DeGennaro LJ, Drachman DA (1992) Alzheimer’s disease and aging: effects on perforant pathway perikarya and synapses. Neurobiol Aging 3:405–411. Markus EJ, Qin YL, Leonard B, Skaggs WE, McNaughton BL, Barnes CA (1995) Interactions between location and task affect the spatial and directional firing of hippocampal neurons. J Neurosci 15:7079–7094. Marr D (1971) Simple memory: a theory for archicortex. Philos Trans R Soc Lond B BiolSci 262:23–81. Maurer AP, Vanrhoads SR, Sutherland GR, Lipa P, McNaughton BL (2005) Self-motion and the origin of differential spatial scaling along the septo-temporal axis of the hippocampus. Hippocampus 15:841–852. McNaughton BL, Barnes CA, Gerrard JL, Gothard K, Jung MW, Knierim JJ, Kudrimoti H, Qin Y, Skaggs WE, Suster M, Weaver KL (1996) Deciphering the hippocampal polyglot: the hippocampus as a path integration system. J Exp Biol 199:173–185. Mcnaughton BL, Battaglia FP, Jensen O, Moser EI, Moser MB (2006) Path integration and the neural basis of the ‘cognitive map’. Nat Rev Neurosci 7:663–678. Mizumori SJ, Ward KE, Lavoie AM (1992) Medial septal modulation of entorhinal single unit activity in anesthetized and freely moving rats. Brain Res 570:188– 197. Moita MA, Rosis S, Zhou Y, LeDoux JE, Blair HT (2004) Putting fear in its place: remapping of hippocampal place cells during fear conditioning. J Neurosci 24: 7015–7023. Moser M-B, Moser EI (1998) Functional differentiation in the hippocampus. Hippocampus 8:608–619. Moser E, Moser MB, Andersen P (1993) Spatial learning impairment parallels the magnitude of dorsal hippocampal lesions, but is hardly present following ventral lesions. J Neurosci 13:3916–3925. Naber PA, Caballero-Bleda M, Jorritsma-Byham B, Witter MP (1997) Parallel input to the hippocampal memory system through peri- and postrhinal cortices. Neuroreport 11:2617–2621. O’Keefe J (1999) Do hippocampal pyramidal cells signal non-spatial as well as spatial information? Hippocampus 4:352–364. O’Keefe J, Burgess N (2005) Dual phase and rate coding in hippocampal place cells: theoretical significance and relationship to entorhinal grid cells. Hippocampus 15:853–866. O’Keefe J, Nadel L (1978) The Hippocampus as a Cognitive Map. New York: Oxford University Press. Quirk GJ, Muller RU, Kubie JL (1990) The firing of hippocampal place cells in the dark depends on the rat’s recent experience. J Neurosci 10:2008–2017.
252
NEURAL SYSTEMS PERSPECTIVE ON PLACE FIELDS
Quirk GJ, Muller RU, Kubie JL, Ranck JB Jr (1992) The positional firing properties of medial entorhinal neurons: description and comparison with hippocampal place cells. J Neurosci 12:1945–1963. Redish AD (2001) The hippocampal debate: are we asking the right questions? Behav Brain Res 127:81–98. Redish AD, Touretzky DS (1997) Cognitive maps beyond the hippocampus. Hippocampus 7:15–35. Rolls ET, Stringer SM, Elliot T (2006) Entorhinal cortex grid cells can map to hippocampal place cells by competitive learning. Network 4:447–465. Rolls ET, Treves A (1998) Neural Networks and Brain Function. Oxford: Oxford University Press. Samsonovich A, McNaughton BL (1997) Path integration and cognitive mapping in a continuous attractor neural network model. J Neurosci 17:5900–5920. Sargolini F, Fyhn M, Hafting T, McNaughton BL, Witter MP, Moser MB, Moser EI (2006) Conjunctive representation of position, direction, and velocity in entorhinal cortex. Science 312:758–762. Sharp PE, Blair HT, Brown M (1996) Neural network modeling of the hippocampal formation spatial signals and their possible role in navigation: a modular approach. Hippocampus 6:720–734. Solstad T, Moser EI, Einevoll GT (2006) From grid cells to place cells: a mathematical model. Hippocampus 16:1026–1031. Suzuki WA, Eichenbaum H. (2000) The neurophysiology of memory. Ann N Y Acad Sci 911:175–191. Taube JS (1995) Place cells recorded in the parasubiculum of freely moving rats. Hippocampus 5:569–583. Taube JS (1998) Head direction cells and the neurophysiological basis for a sense of direction. Prog Neurobiol 55:225–256. Treves A, Kropff E, Biswas A (2005) On the triangular grid of entorhinal place fields. Soc Neurosci Abstr 31:198.11.
van Groen T, Wyss JM (1990) The connections of presubiculum and parasubiculum in the rat. Brain Res 518:227–243. van Haeften T, Baks-te-Bulte L, Goede PH, Wouterlood FG, Witter MP (2003) Morphological and numerical analysis of synaptic interactions between neurons in deep and superficial layers of the entorhinal cortex of the rat. Hippocampus 13:943–952. Van Hoesen GW, Augustinack JC, Dierking J, Redman SJ, Thangavel R (2000) The parahippocampal gyrus in Alzheimer’s disease. Clinical and preclinical neuroanatomical correlates. Ann N Y Acad Sci 911:254–274. Wills TJ, Lever C, Caccuci F, Burgess N, O’Keefe J (2005) Attractor dynamics in the hippocampal representation of the local environment. Science 308:873–876. Witter MP, Amaral DG. (2004) Hippocampal formation. In: The Rat Nervous System, third edition (Paxinos G, ed.). San Diego: Academic Press. Witter MP, Groenewegen HJ, Lopes da Silva FH, Lohman AH (1989a) Functional organization of the extrinsic and intrinsic circuitry of the parahippocampal region. Prog Neurobiol 33:161–253. Witter MP, Moser EI (2006) Spatial representation and the architecture of the entorhinal cortex. Trends Neurosci 29:671–678. Witter MP, Van Hoesen GW, Amaral DG (1989b) Topographical organization of the entorhinal projection to the dentate gyrus of the monkey. J Neurosci 9:216– 228. Witter MP, Wouterlood FG, Naber PA, Van Haeften T (2000) Anatomical organization of the parahippocampal-hippocampal network. Ann N Y Acad Sci 911:1–24. Wood ER, Dudchenko PA, Robitsek RJ, Eichenbaum H (2000) Hippocampal neurons encode information about different types of memory episodes occurring in the same location. Neuron 27:623–633.
16 Neocortical Influences on Hippocampal Place Cells DAVID K. BILKEY
In 1953, the patient known by his initials as H.M. underwent surgery for the purpose of controlling intractable epilepsy. Although the seizures were somewhat alleviated by the operation, which involved the bilateral resection of the medial temporal lobes, H.M was also found to have acquired a profound anterograde and temporally graded retrograde amnesia (Scoville and Milner, 1957). This event triggered an eruption of research activity aimed at determining which particular temporal lobe structures were critically involved in the memory dysfunction. By the 1970s and early 1980s it appeared as though the hippocampus and perhaps the amygdala were the key brain areas and that damage to these structures alone was sufficient to produce the amnesic syndrome (Mishkin, 1978). As a result, a period of ‘‘hippocentric’’ research activity followed, during which models of memory processing placed the hippocampus at the summit of a pyramid of cortical input (Felleman and Van Essen, 1991) from where it orchestrated various encoding, storage, and recall functions. In more recent years the hippocentric model has been reworked to include new insights into the neurobiology of memory that became available during the late 1980s and early 90s. Studies conducted during this era demonstrated that recognition memory deficits in animals that had received lesions to the cortical regions neighboring the hippocampus, but that excluded the latter structure itself, were as large or larger than those that occurred in animals that had received damage to the hippocampus alone (Meunier et al., 1993, 1996). This finding led to a renewed focus on regions such as
the parahippocampal area, encompassing the pre- and parasubiculum, the entorhinal cortex, the perirhinal cortex, and the postrhinal cortex (Witter and Wouterlood, 2002). The general importance of these regions for memory processing has now been demonstrated in a number of different studies, and as a result of an MRI scan of H.M.’s brain (Corkin et al., 1997), we now know that large portions of the parahippocampal region as well as the hippocampus were removed during his temporal lobectomy. We recognize, therefore, that the function of the hippocampus is dependent on activity occurring in a number of neighboring regions. A detailed picture of how these various regions interact during memory processing is only gradually beginning to emerge, not least because the function of the hippocampus itself has yet to be fully understood.
THEORIES OF HIPPOCAMPAL FUNCTION Most current views of hippocampal function can be categorized as belonging to one of two main groups. One set of views is centered on the theory that the primary function of the hippocampus is to process and store spatial information. This idea has been articulated most fully by O’Keefe, Nadel, and coworkers (O’Keefe and Nadel, 1978). In contrast, the second set of views cluster around the idea that the hippocampus is important for both spatial and nonspatial memory processing, playing a major role in what has been described as declarative (Squire, 1992) and/or relational
253
254
NEURAL SYSTEMS PERSPECTIVE ON PLACE FIELDS
(Eichenbaum et al., 1999) memory. Despite some continued disagreement about whether and how far the role of the hippocampus extends outside spatial memory (Colombo and Broadbent, 2000; Eichenbaum, 2000), there is a general consensus that lesions of the hippocampus disturb spatial memory function (O’Keefe and Nadel, 1978; Aggleton et al., 1986; Jarrard, 1995; Brown and Aggleton, 2001; Liu and Bilkey, 2001). This deficit is consistent with findings that variation in hippocampal structure and function is correlated with prowess in various spatial navigation tasks, in both humans and nonhuman species (Sherry et al., 1992; Clayton and Krebs, 1994, 1995; Maguire et al., 1998, 2000, 2003; Suzuki and Clayton, 2000; Lenck-Santini et al., 2001).
PLACE CELLS Certain characteristics of hippocampal neurons appear to be integral to this region’s role in spatial memory. Many hippocampal pyramidal neurons fire only when an animal is in a particular position within an environment. These cells are known as ‘‘place cells’’ and the region within which they fire is known as their ‘‘place field’’ (O’Keefe and Dostrovsky, 1971; Muller, 1996). By comparison, place cells are virtually silent outside their place field. Place cells can fire independently of the behavior that the animal is performing and, in an open field, irrespective of the direction that the animal is facing. This suggests that place cell firing reflects neural processing in some environmentcentered representation of the animal’s location in space. Because they have been studied for over 30 years and have been recorded in several species, including humans (Ekstrom et al., 2003), place cell behavior is well characterized (Muller, 1996; Poucet et al., 2004).
The Paradox of Place and Goal Representation It has been proposed that the firing of a place cell conveys information to the organism about its current location and that the hippocampus as a whole, with its many thousands of place cells, instantiates some form of representation of the environment (O’Keefe and Nadel, 1978; Wilson and McNaughton, 1993). If this is so, then it is likely that this representation has adaptive significance for the animal. One possibility is that the hippocampus allows an animal to locate some goal (and perhaps, conversely, avoid potentially dangerous regions) on the basis of previous experience of that environment. This proposal would be consistent with the vast literature showing that damage to the hippocampus impairs memory-based, goal-directed naviga-
tion, as observed in the water maze (Morris et al., 1982), and that the human hippocampus is activated when a subject is planning a route to a spatial goal (Spiers and Maguire, 2006). More generally a goal could be any object or place that is localized in space but where the current state of that object or place as a goal depends on the cognitive state of the animal (Gray and McNaughton, 2000). As an example, a food cache may or may not be a goal, depending on whether the animal is hungry or not. If the hippocampus supports goal-directed navigation, then what part could place cells play in this process? In order for a rat to solve a spatial problem, such as getting to the location of a previously encountered hidden platform in a water maze, it must have certain information available to it. First it must know where it is currently located, then it must have information about the location of the goal, or at least be able to determine what direction it should move so as to get to the goal. An initial consideration of place cell firing suggests that these cells may represent the current location of the animal. If, however, place cells also encode information about distal regions, such as goal locations, then we might also expect them to fire whenever the animal processes information about that goal location. Since this could occur relatively independently of the current location of the animal, a cell of this type would not appear to have a place field. Therefore, the idea of a cell having a place field and being able to represent a distal goal would appear to be mutually exclusive. This paradox, which has been pointed out by several authors (Tabuchi et al., 2003; Poucet et al., 2004; Bilkey and Clearwater, 2005; and Morris, 1990, cited in Eichenbaum et al., 1999), might lead one to believe that the major function of the hippocampus is to represent current position, while regions outside the hippocampus represent the goal location (Poucet et al., 2004). We have recently suggested, however, that both current position and goal information can be represented within the hippocampus by place cells in a form that can be used for navigational purposes. (Bilkey and Clearwater, 2005). This ‘‘field density’’ model is built on the premise that place fields change position as a result of experience and that some fields will shift location or develop anew at locations that have significance for the animal.
THE FIELD DENSITY MODEL The basic characteristics of the field density model are developed from standard descriptions of place cell behavior and are, therefore, similar to several other cell-level models of hippocampal function (e.g., Burgess and O’Keefe, 1996)). In this model, the hip-
NEOCORTICAL INFLUENCES ON HIPPOCAMPAL PLACE CELLS
pocampus receives input from a group of cells we call ‘‘geometry cells,’’ with the term geometry referring to the shape of the environment (Jeffery et al., 2004). These are cells that fire maximally when the animal is at a certain distance from boundaries (O’Keefe and Burgess, 1996) or major features of the environment. Such cells may relate to or depend on input from the previously described place (Quirk et al., 1992; Fyhn et al., 2004) and grid (Hafting et al., 2005) cells of the entorhinal cortex and/or putative ‘‘boundary units’’ (Jeffery and Anderson, 2003; Jeffery et al., 2004) or boundary vector cells (Barry et al., 2006). We will not concern ourselves here with the mechanisms that produce this firing activity, but we can assume that the relevant processing is completed ‘‘upstream’’ of the place cells and that it is likely to require more complex circuitry than that described in our model (e.g., Solstad et al., 2006). As a result of this input, hippocampal place cells fire only when the animal is within a subregion of the environment (i.e., the place cell has a place field) with a roughly Gaussian firing distribution. If the process by which geometry cells come to represent the environment is competitive, then the place fields of hippocampal place cells will initially be spread evenly across the environment. Although the empirical data are generally supportive of this notion (O’Keefe and Speakman, 1987), there is some suggestion that there may be an overrepresentation of place fields around edges or cues (Muller et al., 1987; Breese et al., 1989; Hetherington and Shapiro, 1997). It is unclear, however, whether this clustering represents the naive state of the network. As described above, the major purpose of the input to the hippocampus is to give place cells firing fields that will, as a population, initially be distributed evenly across the environment. For the purpose of explanation, assume that the firing activity of all hippocampal place cells is passed on to one hypothetical ‘‘output’’ cell through equally weighted connections and that the firing rate of this output cell is proportional to the sum of its afferent drive (see Fig. 1 of Bilkey and Clearwater, 2005;). If we also assume that, on average, place cells with fields in different locations of the environment have the same firing rate, then the firing rate of the output cell will simply be proportional to the number of currently active place cells within the hippocampus. If we monitor the firing rate of the output cell as an animal explores an environment where place fields are distributed evenly across that environment, then the output cell activity would not alter as the animal changed position. If, however, there was a greater density of place fields at or near a particular location in the environment, then the activity of the output cell would vary as the animal moved around, with peak activity occurring at the region where the highest density of
255
place fields was located. In this model, therefore, the neural representation of the current environment (at the level of the output cell) can be thought of as an activity surface, generated by the hypothetical output cell, that is dependent on the density of place fields across the environment. When place cells are more densely concentrated in a particular location there will be a peak in the activity surface, and as a consequence, a gradient that slopes away from this peak region. This simple configuration provides the general structure of a model that can potentially provide a signal to an animal about which direction it should move to locate a target. For example, if we assume that when an animal initially experiences a novel environment, place fields are distributed evenly across that environment, then, as we have stated already, there is no gradient in the activity surface. If, however, the animal experiences a significant event (e.g., the delivery of a reward) in the environment, then a change in place field distribution would occur so as to mark the location at which this event occurred. The shift in place field distribution involves some of the cells that have fields distal to the reward location shifting their place fields so as to be located at or near the position of the reward. This field shift could occur through plasticity in the connections between geometry cells and place cells and would be gated by brain systems activated by exposure to the reward (Fyhn et al., 2002; Bilkey and Clearwater, 2005). The shifting of place fields would produce an increase in the density of place fields near the reward location relative to other parts of the environment, as described in empirical studies of several investigators (Breese et al., 1989; Speakman and O’Keefe, 1990; Kobayashi et al., 1997, 2003; Hollup et al., 2001; Fyhn et al., 2002). Consequently, there would be a peak in the activity surface representation of the environment at this location with a gradient that sloped away from the reward location. This gradient could be used for navigational purposes and the underlying place field density structure could be stored through long-term potentiation (LTP)-type mechanisms (Bliss and Lomo, 1973). If at a later time the animal’s goal was to return to the hidden location from some distal region, then the pattern of place field density appropriate to the current situation could be recalled and the animal could obtain information about the direction to the goal by sampling the local gradient of the activity surface. This sampling could be obtained through either lateral head displacements or mechanisms that depended on the phase relationship of place cell firing with the hippocampal theta EEG rhythm, by yawing rotations of the head (see Bilkey and Clearwater, 2005, for a full description). If the animal then moved to ascend the gradient it would eventually arrive at the reward location. There would,
256
NEURAL SYSTEMS PERSPECTIVE ON PLACE FIELDS
therefore, be ongoing feedback between the environment and the animal’s brain such that movement within the environment would change the signal being generated by the hippocampus, which in turn would provide affective feedback to the animal about the value of a particular movement direction given its current state of motivation. Where the field density model differs from many previous models of hippocampal function is in the interpretation of what place cell firing ‘‘means’’ for the rest of the animal’s brain. In most previous models it was assumed that the signal from individual place cells is ‘‘read’’ to give the animal information about its current location. In the field density model, however, the firing of any individual cell, and the location at which that firing occurs, is relatively unimportant to areas downstream of the hippocampus. Rather, for the main part, place cells serve a ‘‘place-keeping’’ role only. By generating a place field at a particular location in space, they contribute to an overall representation of the environment; but what is important for hippocampal output is not the signal provided by individual cells but, rather, the summed activity of the entire group of currently active place cells, irrespective of the group of cells from which that signal originates. Thus, although information about location provided by upstream areas is important for creating an activity surface across the environment and for enabling the construction and reconstruction of gradients in that surface that peak at certain locations, it is irrelevant which cells actually contribute to which part of the surface; rather, this is a result of the vicissitudes of input and connectivity.
valence-related processing, and appears to link motivation to action (Mogenson et al., 1980; Floresco et al., 2001a,b; Goto and O’Donnell, 2001; Shibata et al., 2001; Schultz, 2004). In line with this model, recent studies have shown that locomotion toward a goal alters the synchronous firing of neurons recorded simultaneously in the subiculum, the major output area of the hippocampus, and the nucleus accumbens of rats (Martin, 2001). Thus a convergence of spatial and reward information that would be consistent with the model appears to occur at the level of single cells in both the subiculum and the nucleus accumbens (Martin and Ono, 2000). Tabuchi et al. (2000) have also reported that functional connectivity between hippocampal and accumbens neurons is stronger when a rat is near reward areas. This observation suggests that hippocampal signals influence the nucleus accumbens during goal-seeking behavior. In findings consistent with this model, it has been reported that nucleus accumbens lesions impair integration of position and reward value information (Albertin et al., 2000) and impair the classic ‘‘hippocampal’’ tasks of T-maze alternation and water maze acquisition (Annett et al., 1989). In an extension to this model, if signals were also output to a region that coded negative valence, such as the amygdala (Gao et al., 2004), place avoidance might also be driven through similar mechanisms. In this case, the balance between approach and avoidance might be mediated by the relative activation of the positive and negative valence systems, with some anatomical separation of these systems perhaps being provided through dorsal versus ventral hippocampal connectivity (Trivedi and Coover, 2004; Bertoglio et al., 2006).
Output Regions In the section above, I suggested, for the sake of explanation, that the output signal from the hippocampus is sent to a hypothetical output cell. In reality this output cell could be groups of neurons located in one or more of several regions, where activation of these neurons would result in the animal experiencing either a positive emotional state or the experience of ‘‘wanting’’ (Robinson and Berridge, 1993). This state would, therefore, be positively correlated with the magnitude of the hippocampal signal. The activity surface would become a ‘‘valence surface’’ where a peak on the surface represented a region where the animal would experience a stronger affective or ‘‘want’’ signal. As the animal roamed through the environment it would experience a greater or lesser signal, depending on where it was currently located. A valence surface of this type could potentially be generated by routing output to the nucleus accumbens, a region that receives connections from the hippocampus, is involved in
Modeling Changes in Field Density We have recently implemented a simple version of the goal-directed field density model using a three-layer, feed-forward, artificial neural network that incorporates the most general features of hippocampal functioning (Bilkey and Clearwater, 2005). The key finding from this simulation was that a systematic shift in a relatively small proportion of place fields (5%–15%) was sufficient to produce a gradient in the activity surface generated across the whole population of place cells. This gradient was steep enough to support navigation to a goal even from the most distal regions of the environment, despite the fact that the place fields of individual cells were limited in size so as to cover only a small subportion (about 12%) of the total area (Fig. 16–1). It was also shown that shifts in place field location could be produced through the operation of a simple, biologically plausible, modified Hebbian learning rule (Wang and Maler, 1997) that governed
NEOCORTICAL INFLUENCES ON HIPPOCAMPAL PLACE CELLS
257
Figure 16–1. The activity surface generated across a rectangular environment in a computer simulation of the field density model (Bilkey and Clearwater, 2005). In this model, the place fields of individual cells cover, on average, about 12% of the environment. An example place field located within a rectangular environment is shown in panel a. If these place fields are distributed evenly across the environment, the summed activity of the population of place cells that are active as the animal locomotes throughout that environment is a near-flat surface (b). If a small proportion of place fields, for example, 5% of the total population, is shifted toward the location of some important experience (the middle of the North border of the environment in this simulation), through the operation of a modified Hebbian learning rule, a gradient is generated that extends across the full extent of the activity surface (c). The surface shown in c is also shown with an expanded vertical axis in two side views (d,e).
changes in connection strength between the geometry cells and the place cells. This finding suggests that the empirical phenomenon whereby place fields have been reported to shift position toward a goal in some (Breese et al., 1989; Speakman and O’Keefe, 1990; Kobayashi et al., 1997, 2003; Hollup et al., 2001; Fyhn et al., 2002) but not all (Jeffery et al., 2003; Tabuchi et al., 2003) studies could have an important effect on hip-
pocampal output and could potentially subserve navigational processes.
Place Cell Remapping A critical feature of the field density model is that there is no need for the hippocampal ‘‘representation’’ of space to be continuous or cohesive, as the physical
258
NEURAL SYSTEMS PERSPECTIVE ON PLACE FIELDS
relationship between place cells does not need to map directly onto the spatial relationship between places in the environment as required in some other models (e.g., Muller and Stead, 1996; Skaggs et al., 1996). Thus individual place fields are not ‘‘representing’’ a particular location in space, but rather are contributing to a global representation in which their individual contribution, and the location in which they fire, is relatively unimportant. We have previously suggested that each place cell acts as a ‘‘place holder’’ for a particular location, contributing by its presence to the overall distribution of place fields that in sum creates the critical gradient (Bilkey and Clearwater, 2005). On this basis, the place field location has to be interpreted in conjunction with knowledge about what is happening to the underlying gradient instantiated in the population activity. The proposal that place cells and their respective fields are not representing a particular ‘‘place’’ is consistent with data showing that the relationship between a place field and a location in the environment can be altered quite readily. It has long been known, for example, that a change in the shape of the apparatus in which an animal is being recorded can produce a redistribution of place fields (O’Keefe and Conway, 1978; Muller and Kubie, 1987), a phenomenon that has been named ‘‘remapping.’’ More recently it has been shown that remapping can also be induced by changes in non-geometrical aspects of the environment. For example, when an animal is allowed to locomote between two geometrically identical environments, the activity of many of the place cells is significantly different in the two environments (Skaggs and McNaughton, 1998). Furthermore, remapping can also take place when changes are made to the color or odor of an environment without changes in its geometric shape or position (Anderson and Jeffery, 2003; Jeffery et al., 2004). Remapping may also occur in a stable physical environment when the position where a goal or reward is available within the environment is moved (Breese et al., 1989; Speakman and O’Keefe, 1990; Kobayashi et al., 1997, 2003; Hollup et al., 2001; Fyhn et al., 2002; although see Jeffery et al., 2003; Tabuchi et al., 2003) or where the demands of the current task change (Smith and Mizumori, 2006a). Whereas the stability of place fields can be seen as a virtue in models where place fields represent location and instability represents forgetting (Muir and Bilkey, 2001), in the field density model, place field shifts may be largely irrelevant to the overall function of the region, provided an underlying gradient is maintained. Place field shifts may, however, reflect changes in inputs and modulating factors that could reflect the differentiation of environments on the basis of both external and internal cues.
Types of Remapping In the field density model, remapping occurs for several different reasons. First, place fields might shift to locations where some significant event (e.g., reinforcement or escape) has been experienced. As we have shown in our simulations (Bilkey and Clearwater, 2005), this form of remapping can be quite subtle, involving a small proportion of cells, which may explain why some studies (Jeffery et al., 2003; Tabuchi et al., 2003) have failed to observe it. Second, place cells might shift location (or turn on or off) following a change in the physical environment (Anderson and Jeffery, 2003; Jeffery et al., 2004) or a change in the animal’s internal motivational state (Holscher et al., 2003). These second and third forms of remapping would be important because they would define the current situation in terms of both external and internal cues. That is, information about the general features of the current environment, such as its geometry and its non-geometric sensory features, would allow the discrimination of one environment from the other. Internal states could also play an important role in defining a goal-based situation such that an internal state of hunger might generate a different configuration of place fields than a state of thirst. This proposal has not been tested previously; however, it is consistent with previous suggestions that the hippocampus is involved in the control of behaviors related to food appetite (Tracy et al., 2001) and that the retrieval of memories based on internal motivational state is dependent on the hippocampus (Kennedy and Shapiro, 2004). Finally, although it has been suggested that a fourth form of remapping (rate remapping) occurs in which cells systematically change their firing rates without changing their place field location (Leutgeb et al., 2005), it is also possible that this is not a form of remapping, as described above, but instead represents the encoding of information about choice. This possibility will be discussed in a later section.
What Regions Contribute to Remapping? On the basis of the proposal described above, the purpose of remapping is to integrate situation-specific information into the hippocampal ‘‘representation’’ to allow for a different pattern of activity for each of these situations. Through plasticity mechanisms such as LTP (Bliss and Lomo, 1973; Kentros et al., 1998) the pattern of activity could be stored, ready to be retrieved at a later time given appropriate environmental and/or motivational cues. The diversity of different remappings would provide a series of underlying frameworks for the gradient, where it exists, to enable subsequent discrimination of goal location de-
NEOCORTICAL INFLUENCES ON HIPPOCAMPAL PLACE CELLS
pending on the match between the internal and external cues that occurred at the time of storage and the cues that exist at the time of retrieval. There are a variety of afferent regions that are likely to contribute to the structure of the hippocampal ‘‘framework.’’ These regions provide information about factors such as the geometric shape of the surroundings and the particular sensory features that allow two separate environments of the same shape to be differentiated from each other.
Perirhinal Cortex Influences on Place Cell Firing A large portion of the hippocampal input that is routed through medial temporal cortex passes through the perirhinal and postrhinal cortices. These regions are part of the parahippocampal region (Witter and Wouterlood, 2002). The perirhinal cortex in particular has received much attention because of its role in object-recognition memory (Zola-Morgan et al., 1989; Gaffan and Murray, 1992; Meunier et al., 1993; Wiig and Bilkey, 1995; Brown and Aggleton, 2001; Buckley, 2005). It is connected directly and indirectly via the entorhinal cortex to the hippocampus (Burwell et al., 1995; Naber et al., 1997; Burwell and Amaral, 1998) and thus serves as a conduit for information that would define many aspects of the external environment. This activity could be integrated with geometric information in the entorhinal cortex (Hafting et al., 2005) prior to use by the hippocampus. Geometric information might itself be provided by regions such as the subiculum (Barry et al., 2006) or, where external sensory cues are limited, by self-motion information provided by regions such as retrosplenial cortex (Cooper et al., 2001). In conjunction, this activity would comprise a pattern of input to the hippocampus unique to a particular location within a particular environment. Information about current motivational state could then modulate the pattern of firing generated by this input within the hippocampus. This would allow internal states to modulate a goal-based situation so that an internal state of hunger, for example, might generate a different configuration of place fields than a state of thirst, assuming that water and food were available in different locations within the same environment. Information about such internal states may be provided via subcortical inputs routed through the medial septum (Bilkey and Goddard, 1985; Freund and Antal, 1988; Dutar et al., 1995; Papp et al., 1999) and/or through the hippocampal receptors that are responsive to neurohormones, for example, leptin (Oomura et al., 2006), ghrelin (Diano et al., 2006), and vasopressin (Young et al., 2006). On the basis of the circuitry described above, one would predict that lesions of perirhinal cortex would have effects on the processing of cues that define the external environment and that this might influence
259
both the performance of hippocampal-dependent tasks and the remapping of place cells. There are now a number of studies that have investigated the effects of perirhinal lesions on tasks that appear to tax the hippocampus. The results are, however, somewhat contentious. Whereas several studies show that lesions of perirhinal cortex produce deficits in spatial memory tasks, other studies suggest that they do not (see Aggleton et al., 2004, for a recent review), and in fact, that the function of the individual structures may be dissociable on the basis of a spatial–nonspatial dichotomy (Bussey et al., 2000; Winters et al., 2004). One possible integration of these findings is that the perirhinal cortex allows animals to resolve ambiguity in visual discriminations (Bussey et al., 2002, 2006). This may depend on perirhinal cortex involvement in the processing of the relationships between cues such as those required for configural learning (Moran and Dalrymple-Alford, 2003). Feature ambiguity will occur when there are common elements in the stimuli to be discriminated, and so will be more pronounced in configural learning tasks, discriminations with large stimulus sets, and discriminations with morphed stimuli. Degraded stimuli also increase feature ambiguity and so are associated with perirhinal lesion–induced deficits. While much of the supporting evidence comes from studies with monkeys, some recent studies with rats (Eacott et al., 2001; Norman and Eacott, 2004) are consistent with this view (Aggleton et al., 2004). That said, however, one recent study of the effects of neurotoxic lesions of perirhinal cortex on a biconditional discrimination task failed to show the effects predicted by the configural model (Davies et al., 2007). Although the behavioral/lesion studies have not yet fully clarified the conditions under which the perirhinal cortex is involved in ‘‘hippocampal’’ tasks, there is some electrophysiological data showing that lesions of perirhinal cortex alter place cell remapping. In one study conducted in our lab, the firing characteristics of dorsal CA1 place cells were examined in rats with bilateral ibotenic acid lesions centered on the perirhinal cortex, or control surgeries, as they foraged freely in a recording arena (Muir and Bilkey, 2001). Recordings were made before and after the animals were removed from the recording arena for a delay period. It was determined that, although basic place field firing properties were relatively unchanged, the place fields of individual cells recorded from the lesioned animals were more likely to shift position during the delay period than those from control animals. These data indicated, therefore, that although the initial formation of place fields in the hippocampus is not dependent on the perirhinal cortex, the maintenance of place field position over time is. One possible explanation for this effect is that the instability is
260
NEURAL SYSTEMS PERSPECTIVE ON PLACE FIELDS
an erroneous ‘‘remapping’’ caused by a lesion-induced degradation of information regarding environmental cue structure. For example, if the many individual cues that define an environment are normally bound into higher-order representations that encode information about their relationships, then the removal or alteration of a single cue in the cue array may have little effect on the overall structure of the representation. Furthermore, if place cells are responsive to these high-order representations, then this manipulation will also have little effect on their place fields (O’Keefe and Conway, 1978; Nadel and Willner, 1980). If the perirhinal cortex represents information about such cue relationships, as proposed in the configural models, then the place fields in animals with perirhinal lesions may have been more dependent on individual local cues and, therefore, more susceptible to changes that occurred across a delay period (such as wiping the environment clean). A further finding from the Muir and Bilkey study was that the place fields of lesioned animals were more likely to maintain position relative to the arena but not a visual cue, when the visual cue was rotated. This was consistent with the proposal that the place cells may have been more responsive to local cues. A more recent study has also shown that the size of place fields is decreased in animals with perirhinal cortex lesions (Lu and Bilkey, 2002). It is possible that this also reflects a tendency for the lesioned animals (or their hippocampus) to be biased toward responding to local olfactory cues, such as odor trails (Wallace et al., 2002) or body waste on the floor of the apparatus. These, by their nature, would potentially provide for more accurate spatial localization (at least in the short term) and therefore, smaller place fields, than would distal visual cues. It is interesting that one other effect of the perirhinal lesions was to change the relationship between the firing rate of hippocampal place cells and the velocity of the animal. Normally these two variables are positively correlated, but in lesioned animals this relationship was degraded (Muir and Bilkey, 2003). It is unlikely that this effect was simply a result of a lesioninduced change in the animals’ locomotion, as such effects were minimal. This finding suggests that hippocampal processing of an animal’s movement through the environment might be disrupted in perirhinal cortex–lesioned rats. This self-motion information could potentially be available to the hippocampus from several different sensory systems, for example, via sensory flow, motor efferent copy, or proprioceptive signals, or through vestibular inputs, although recent findings indicate that the latter system is not the source of this signal (Hirase et al., 1999; Russell et al., 2006). The former inputs could be derived from several of the inputs to the perirhinal cortex, as anatomical and
electrophysiological evidence indicates that this region receives projections from visual (Burwell and Witter, 2002), motor (Reep et al., 1987), and barrel (Naber et al., 2000) cortex. Although it is unclear whether the hippocampus uses the velocity information encoded in the pyramidal-cell firing rate, this signal could potentially provide important feedback about the animal’s self-movement within the environment. Alternatively, however, it may reflect the fact that some forms of signals, such as those derived from inputs such as the vibrissae, may be speed modulated. We are currently investigating this possibility.
Perirhinal Cortex and Context Several authors have previously proposed that remapping is a response to context change (Anderson and Jeffery, 2003; Jeffery et al., 2004; Lee et al., 2004; Leutgeb et al., 2004), where contextual information refers to the background information (or conditions mediating the appropriate response to a given event) that is outside of the information expressed by the target event (Holland and Bouton, 1999). Since any particular feature can shift from being in the ‘‘foreground’’ (stimulus) to the ‘‘background’’ (context), depending on factors such as attention, which in turn is governed by the animal’s previous experience and current motivation, it is probable that contextual processing depends, at least in part, on the same representations used for object and feature recognition. A contextual representation may, therefore, be based on multiple cues and the complex associations and conjunctive relationships between these cues, features, and objects (Nadel and Willner, 1980). As we have described previously, perirhinal cortex may have a role in the encoding of these relationships as part of its involvement in object recognition. In findings consistent with this model, the perirhinal (and postrhinal) cortex has been implicated in the processing of contextual information, for example, in associating shock with a multifeatured context (Corodimas and LeDoux, 1995; Sacchetti et al., 1999; Bucci et al., 2000, 2002; Burwell et al., 2004a,b) and in object–context discrimination (Norman and Eacott, 2005). It is possible, therefore, that perirhinal cortex represents major cue-driven aspects of external context, not so much because its function is specifically to represent context per se, but because context is constructed out of the myriad of cue/stimulus/object representations (and their relationships) that are continually being activated within perirhinal cortex. The results of several studies suggest that the hippocampus also plays a specialized role in the processing of context (Jeffery et al., 2004). Hirsh (1974), for example, initially suggested that the hippocampus was involved in context-based retrieval of information from memory, and later work showed that hippocampal
NEOCORTICAL INFLUENCES ON HIPPOCAMPAL PLACE CELLS
lesions disrupted the acquisition and retention of fear responses to contextual cues (Kim and Fanselow, 1992; Phillips and LeDoux, 1992). There is now a considerable body of evidence indicating that this structure has a role in the processing of contextual information related to other forms of learning as well (for a review see Holland and Bouton, 1999b; Jeffery et al., 2004). Some studies, however, have failed to show an effect of hippocampal damage on contextual processing (e.g., McDonald et al., 1997; Anagnostaras et al., 2001). Some of the factors that may be involved in these discrepancies are discussed in a recent review (Jeffery et al., 2004). In light of the broader model presented in this chapter, the difference between hippocampal and perirhinal involvement in contextual processing may relate to the possibility that activity in the perirhinal cortex specifies some of the major descriptors of external context, whereas this information is then used within the hippocampus to modulate goalcentered representations. A lesion of both of these regions may, therefore, produce a deficit in contextual responses, but for different reasons. A perirhinal cortex lesion would disrupt the components that make up the ‘‘representation’’ of context, while the hippocampal lesion would disrupt the use of that representation in a task in which some goal potentially localized in space might be approached or avoided.
Prefrontal Cortex Influences on Place Cell Firing As I have outlined above, there are many brain regions that contribute to the hippocampal representations that underlie navigation. The emphasis on goals in the field density model, however, focuses attention on the possible contribution of prefrontal cortex to activity in this system. The prefrontal cortex is a brain area that has been implicated in goal-directed behavior in a number of previous studies (e.g., Duncan et al., 1996; Miller et al., 2002; Hok et al., 2005; Miller and D’Esposito, 2005) as part of its role in ‘‘executive control’’ (Milner, 1982; Shallice, 1982; GoldmanRakic, 1987; Petrides, 1994; Fuster, 1995; Baddeley and Della Sala, 1996). One means by which prefrontal cortex might contribute to goal-directed navigation is by modulating the choice behavior that might occur when two or more goals are present in the same environment. This notion is consistent with the proposals of previous ‘‘executive’’ models in which prefrontal cortex is seen to have a high-order role in decision making. Many previous models of decision making, however, have tended to attribute the decision-making process to prefrontal cortex itself, sometimes in a homuncular fashion. It is possible, however, that decisions are the outcome of competition between neural representations of choices that are instantiated within the brain structures that guide particular behavioral responses, rather
261
than being computed within this single ‘‘executive’’ area (Gold and Shadlen, 2000, 2001, 2002; Shadlen, 2002). In this latter view, prefrontal cortex may contribute important information to the decision-making process, such as the relative costs and benefits of particular choices (Bannerman et al., 2002; Montague and Berns, 2002; McClure et al., 2004), but when the decision making is taking place in the spatial domain and it is dependent on hippocampal representations of hidden goals, then the decision itself may be directed by changes of activity within the hippocampus. One means by which prefrontal cortex might contribute to hippocampal representation of goals is through modulation of relative goal strength (that is, gradient slope in the field density model) within the hippocampus when two or more goals are present. This modulation could be based on assessment of the relative costs and benefits of particular choices given the animal’s current state. More than one goal representation might, therefore, be sequentially active within the hippocampus, with attentional control flipping back and forth between the two (or more) representations (Olypher et al., 2002). Each representation could then ‘‘compete’’ to coerce the animal toward one or other goal, with the strength of each of these competing representations being modulated by prefrontal cortex on the basis of the value (cost/benefit) of particular choices (Bannerman et al., 2002; Montague and Berns, 2002; McClure et al., 2004). Here cost/benefit could include the integration of information such as the probability of reward at a particular location, the reward value, the distance to the reward, and the effort required in getting to it. This information could be accumulated over previous experience of similar situations. A simple signal from prefrontal cortex, synchronized to occur when the animal is attending to a particular goal, could bias the hippocampal representation of space by differentially altering firing rates in the place cells that represent the competing goals and hence the slope of the resulting valence gradients. This hypothesis is consistent with recent findings that the correlation between neural activity in the hippocampus and prefrontal cortex is selectively enhanced during behavior that recruits spatial working memory (Jones and Wilson, 2005; Siapas et al., 2005) and that the firing rate of some neurons in the infralimbic and prelimbic regions of medial prefrontal cortex is related to specific goal locations (Hok et al., 2005). Therefore, a prefrontal-mediated increase in the firing rate of the hippocampal place cells that represent a particular movement choice would increase the probability that the animal will make a movement in that direction (Bilkey and Clearwater, 2005). In recent studies we have begun to explore the role of the prefrontal cortex in modulating activity in the hippocampus. For example, when single-unit
262
NEURAL SYSTEMS PERSPECTIVE ON PLACE FIELDS
recordings of CA1 place cells were made in freely moving rats with either prefrontal lesions or sham surgeries, the mean firing rate was significantly lower in lesion-group animals than in sham-group animals, although the size of their place fields was not affected (Kyd and Bilkey, 2003). The location of the firing fields of prefrontal-lesioned rats was also less stable across time when either 3-min or 5-hr intervals were inserted between successive recordings of the same cell. On the basis of these findings, it was hypothesised that animals with prefrontal lesions may be overly influenced by local, less stable environmental cues than sham-group rats. It is interesting that this pattern of effects is somewhat similar to that which occurs following damage to perirhinal cortex, which suggests that prefrontal cortex could influence goal representations by modulating activity in perirhinal cortex. This is consistent with previous data suggesting that these regions are functionally connected (Parker and Gaffan, 1998; Bilkey and Liu, 2000; Zironi et al., 2001; Apergis-Schoute et al., 2006). In a second study we investigated this phenomenon further by determining what effect the introduction of objects into the recording environment had on the place cell firing of control and prefrontal-lesioned rats (Kyd and Bilkey, 2005). We determined that the place fields of lesion-group cells were shifted to a greater extent than that of control cells when objects were introduced into the recording environment. This finding supports the proposal that place cell firing was overly influenced by local cues in the prefrontallesioned animals. In a second experiment, place cells were recorded while rats foraged on a circular track with access to both local and distal multimodal cues. This experiment was designed to explicitly test whether familiar, local environmental cues had increased control over place cell firing in prefrontal-lesioned animals. This hypothesis was examined by rotating a circular track containing local surface cues within a stable distal environment. It was determined that although the position of place fields in lesion-group cells was not excessively tied to local cues, a greater proportion of the fields lost their spatial selectivity following a rotation of these cues. One interpretation of this finding is that the lesion altered the ability of brain areas such as perirhinal cortex to represent the relationship between the local and distal cues, a result that would be consistent with configural models of perirhinal function. The cue-related effects were also associated with a reduction in firing rate and larger extracellular action-potential amplitudes and a greater incidence of burst-firing in lesion-group cells. Previous evidence suggests that spike amplitude, as measured during extracellular recording, may partially depend on the efficiency with which somatic action potentials
back-propagate into the dentritic arbor (Buzsaki et al., 1996; Quirk et al., 2001). Since action potential backpropagation (and hence, spike amplitude) can be increased by a reduction in the strength of inhibitory input onto the cell, provided, for example, by feedback inhibition (Tsubokawa and Ross, 1996), these data suggest that the prefrontal lesion may have resulted in a disinhibition of hippocampal place cells. The increased burst-firing in lesion-group cells is also consistent with a disinhibition model, as previous studies have shown that increased bursting occurs in the hippocampus when GABAergic inhibition is antagonized (Malouf et al., 1990). These latter findings are important because they demonstrate that the prefrontal lesion effects we observed are to some degree a result of relatively direct effects at the level of the hippocampal place cells, rather than simply being due to an alteration in upstream processing. This does not necessarily imply a direct connection between prefrontal cortex and the hippocampus, as there are a number of mechanisms through which a prefrontal lesion could result in a reduction in hippocampal inhibition. For example, a reduction in the firing rate of inhibitory interneurons (Wilson and McNaughton, 1993), a reduction in the number of active interneurons (Quirk et al., 2001), or a change in the effective coupling between pyramidal cells and interneurons (Csicsvari et al., 1998) could all be implicated. These kinds of changes could be mediated through prefrontal modulation of the ventral tegmental area (Sesack et al., 1989; Sesack and Pickel, 1992; Taber et al., 1995), which in turn projects to the CA1 layer of the hippocampus (Gasbarri et al., 1997). Alternatively, they could result via prefrontal cortex projections to parahippocampal cortices such as the perirhinal and entorhinal cortices (Groenewegen and Uylings, 2000). Since output from these latter regions reaches the hippocampus via the perforant path (Naber et al., 1999; Witter et al., 2000), and since there is some evidence to suggest that a portion of the perirhinal and entorhinal input to CA1 is directly onto inhibitory interneurons (Empson and Heinemann, 1995), lesion-induced effects on inhibitory activity could be mediated via this route. In summary, our data suggest that prefrontal lesions remove an inhibitory modulation of the hippocampus. Furthermore the data showing that place cell firing rates were decreased following prefrontal lesions suggest that an excitatory drive is also lost. This indicates that the prefrontal cortex exerts control over the hippocampus via several different mechanisms, which is consistent with previous data where both inhibition– disinhibition (Frith et al., 1991; Knight and Grabowsky, 1995; Shimamura, 2000; Meyer-Lindenberg et al., 2001; Rule et al., 2002) and excitatory (Egner and Hirsch, 2005) processes have been implicated in ‘‘ex-
NEOCORTICAL INFLUENCES ON HIPPOCAMPAL PLACE CELLS
ecutive’’ processes. Through activation of these mechanisms a relatively simple and appropriately synchronized ‘‘value’’ signal from prefrontal cortex could bias the hippocampal representation of goal-directed space by both increasing the firing rates of the place cells that represent a high-value location and decreasing the rate of those that represent a low-value option (Fig. 16–2).
Relevance to Memory I have presented a broad picture of how hippocampal place cells might contribute to spatial memory via a field density model that provides information about direction to goals. A key feature of the model is that it demonstrates a way around the apparent paradox of hippocampal cells representing local features of the
263
environment while at the same time being able to represent a distal goal (Tabuchi et al., 2003; Poucet et al., 2004; and Morris, 1990, cited in Eichenbaum et al., 1999). The model also circumvents many of the criticisms that have been made of cognitive map–type descriptions of hippocampal spatial function. For example, it does not require a homunculus to ‘‘read’’ an internal map (Vanderwolf, 2001) and does not suffer from the related ‘‘symbol grounding’’ problem (Harnad, 1990) in which the semantic interpretation of ‘‘representations’’ within other brains is dependent on our own knowledge as observers. The model explains why place fields might be modified by the insertion of a barrier into the environment (Vanderwolf, 2001; Bilkey and Clearwater, 2005), and it explains why place cells might shut down when an animal is passively moved through the environment (Foster et al.,
Figure 16–2. A diagram illustrating some of the inputs to the hippocampus that may play a role in defining the hippocampal representation of goal location. The regions are color coded on the basis of their possible involvement in providing information about external cues (blue), geometric boundaries and self-movement (green), and internal state (purple). Prefrontal cortex has a role in decision making by modulating the strength of particular representations, possibly through its influence on perirhinal cortex or through other indirect connections to the hippocampus. The output of the hippocampus is routed to regions such as the nucleus accumbens, where it produces (in this example) a positive valence signal. Given appropriate motivation (e.g., hunger, thirst) the animal behaves so as to increase the magnitude of the valence signal.
264
NEURAL SYSTEMS PERSPECTIVE ON PLACE FIELDS
1989; Vanderwolf, 2001; the hippocampus is involved in active, directed navigation, not merely in passively representing the environment). Furthermore, the model accounts for hippocampal spatial representation being possibly non-continuous, non-homogeneous, and lacking in cohesion (Tanila et al., 1997; Eichenbaum et al., 1999; Zinyuk et al., 2000; Knierim, 2002). As we have shown, place field clustering is an essential feature of the model and there is no need for the internal representation to be continuous or cohesive, as the physical relationship between place cells does not map directly onto the spatial relationship between places in the environment. In the field density model, the output of the hippocampus can be seen as contributing to what in lay terms might be referred to as ‘‘gut feeling’’ or more formally, intuitive, spatial decision-making via its valencegenerating output. Studies of non-conscious judgments and automatic evaluations in other domains show that humans can make good decisions based on the type of rapid unanalyzed reaction that likely depends on this type of emotional response (Damasio, 1996; LeDoux, 1996; Dijksterhuis, 2004; Bolte and Goschke, 2005). In some cases these decisions may actually be better (by certain criteria) than those that use introspection (Wilson and Schooler, 1991; see (Halberstadt and Wilson, in press, for a recent review). There is no reason to believe that nonhuman animals, rats included, do not also have access to this kind of information. In fact, several demonstrations of what appear to be episodiclike memory in nonhuman species could be accounted for by this kind of processing (Clayton and Dickinson, 1998; Hampton, 2001). It is possible, however, that place cells in the hippocampus may participate in ‘‘higher’’ forms of memory, particularly in humans. For example, the spatial representations within the hippocampus may provide important cue information that aids considerably in the retrieval of declarative episodic memories. For example, the unique pattern of activity within the hippocampus that corresponds to a particular internal and external context could be accessed by simultaneously activating large groups of place cells as, for example, occurs during hippocampal sharp waves or particular oscillation states (Chrobak and Buzsa´ki, 1998). If this activity were fed back into the entorhinal cortex and then into more posterior temporal cortex, it might serve to reactivate the nonspatial aspects of memories that are stored in these regions. In this way, although the hippocampus does not store episodic memories itself, it may have an important role in the retrieval of such information, based on the provision of the appropriate contextual cues (Smith and Mizumori, 2006b) that happen to be integrated in the hippocampus as a consequence of its primary role in goal-directed spatial memory.
References Aggleton JP, Hunt PR, Rawlins JNP (1986) The effects of hippocampal lesions on spatial and non- spatial tests of working memory. Behav Brain Res 19:133–146. Aggleton JP, Kyd RJ, Bilkey DK (2004) When is the perirhinal cortex necessary for the performance of spatial memory tasks? Neurosci Biobehav Rev 28:611–624. Albertin SV, Mulder AB, Tabuchi E, Zugaro MB, Wiener SI (2000) Lesions of the medial shell of the nucleus accumbens impair rats in finding larger rewards, but spare reward-seeking behavior. Behav Brain Res 117:173–183. Anagnostaras SG, Gale GD, Fanselow MS (2001) Hippocampus and contextual fear conditioning: recent controversies and advances. Hippocampus 11:8–17. Anderson MI, Jeffery KJ (2003) Heterogeneous modulation of place cell firing by changes in context. J Neurosci 23:8827–8835. Annett LE, McGregor A, Robbins TW (1989) The effects of ibotenic acid lesions of the nucleus accumbens on spatial learning and extinction in the rat. Behav Brain Res 31:231–242. Apergis-Schoute J, Pinto A, Pare D (2006) Ultrastructural organization of medial prefrontal inputs to the rhinal cortices. Eur J Neurosci 24:135–144. Baddeley A, Della Sala S (1996) Working memory and executive control. Philos Trans R Soc Lond B Biol Sci 351:1397–1403. Bannerman DM, Deacon RMJ, Offen S, Friswell J, Grubb M, Rawlins JNP (2002) Double dissociation of function within the hippocampus: spatial memory and hyponeophagia. Behav Neurosci 116:884–901. Barry C, Lever C, Hayman R, Hartley T, Burton S, O’Keefe J, Jeffery K, Burgess N (2006) The boundary vector cell model of place cell firing and spatial memory. Rev Neurosci 17:71–97. Bertoglio LJ, Joca SR, Guimaraes FS (2006) Further evidence that anxiety and memory are regionally dissociated within the hippocampus. Behav Brain Res 175:183–188. Bilkey DK, Clearwater J (2005) The dynamic nature of spatial encoding in the hippocampus. Behav Neurosci 119:1533–1545. Bilkey DK, Goddard GV (1985) Medial septal facilitation of hippocampal granule cell activity is mediated by inhibition of inhibitory interneurons. Brain Res 361: 99–106. Bilkey DK, Liu P (2000) The effects of separate and combined perirhinal and prefrontal lesions on spatial memory tasks in the rat. Psychobiology 28:12–20. Bliss TVP, Lomo T (1973) Long-lasting potentiation of synaptic transmission in the dentate area of the anaesthetized rabbit following stimulation of the perforant path. J Physiol (Lond) 232:331–356.
NEOCORTICAL INFLUENCES ON HIPPOCAMPAL PLACE CELLS
Bolte A, Goschke T (2005) On the speed of intuition: intuitive judgments of semantic coherence under different response deadlines. Mem Cognit 33:1248–1255. Breese CR, Hampson RE, Deadwyler SA (1989) Hippocampal place cells: stereotypy and plasticity. J Neurosci 9:1097–1111. Brown MW, Aggleton JP (2001) Recognition memory: what are the roles of the perirhinal cortex and hippocampus? Nat Rev Neurosci 2:51–61. Bucci DJ, Phillips RG, Burwell RD (2000) Contributions of postrhinal and perirhinal cortex to contextual information processing. Behav Neurosci 114:882–894. Bucci DJ, Saddoris MP, Burwell RD (2002) Contextual fear discrimination is impaired by damage to the postrhinal or perirhinal cortex. Behav Neurosci 116:479– 488. Buckley MJ (2005) The role of the perirhinal cortex and hippocampus in learning, memory, and perception. Q J Exp Psychol B 58:246–268. Burgess N, O’Keefe J (1996) Neuronal computations underlying the firing of place cells and their role in navigation. Hippocampus 6:749–762. Burwell RD, Amaral DG (1998) Perirhinal and postrhinal cortices of the rat: Interconnectivity and connections with the entorhinal cortex. J Comp Neurol 391:293– 321. Burwell RD, Bucci DJ, Sanborn MR, Jutras MJ (2004a) Perirhinal and postrhinal contributions to remote memory for context. J Neurosci 24:11023–11028. Burwell RD, Saddoris MP, Bucci DJ, Wiig KA (2004b) Corticohippocampal contributions to spatial and contextual learning. J Neurosci 24:3826–3836. Burwell RD, Witter MP (2002) Basic anatomy of the parahippocampal region in monkeys and rats. In: The Parahippocampal Region (Witter MP, Wouterlood FG, eds.), pp 35–59. Oxford: Oxford University Press. Burwell RD, Witter MP, Amaral DG (1995) Perirhinal and postrhinal cortices of the rat: a review of the neuroanatomical literature and comparison with findings from the monkey brain. Hippocampus 5:390–408. Bussey TJ, Duck J, Muir JL, Aggleton JP (2000) Distinct patterns of behavioural impairments resulting from fornix transection or neurotoxic lesions of the perirhinal and postrhinal cortices in the rat. Behav Brain Res 111:187–202. Bussey TJ, Saksida LM, Murray EA (2002) Perirhinal cortex resolves feature ambiguity in complex visual discriminations. Eur J Neurosci 15:365–374. Bussey TJ, Saksida LM, Murray EA (2006) Perirhinal cortex and feature-ambiguous discriminations. Learn Mem 13:103–105; author reply 106–107. Buzsaki G, Penttonen M, Nadasdy Z, Bragin A (1996) Pattern and inhibition-dependent invasion of pyramidal cell dendrites by fast spikes in the hippocampus in vivo. Proc Natl Acad Sci USA 93:9921–9925.
265
Chrobak JJ, Buzsa´ki G (1998) Operational dynamics in the hippocampal-entorhinal axis. Neurosci Biobehav Rev 22:303–310. Clayton NS, Dickinson A (1998) Episodic-like memory during cache recovery by scrub jays. Nature 395:272– 274. Clayton NS, Krebs JR (1994) Hippocampal growth and attrition in birds affected by experience. Proc Natl Acad Sci USA 91:7410–7414. Clayton NS, Krebs JR (1995) Memory in food-storing birds: From behaviour to brain. Curr Opin Neurobiol 5:149–154. Colombo M, Broadbent N (2000) Is the avian hippocampus a functional homologue of the mammalian hippocampus? Neurosci Biobehav Rev 24:465–484. Cooper BG, Manka TF, Mizumori SJY (2001) Finding your way in the dark: the retrosplenial cortex contributes to spatial memory and navigation without visual cues. Behav Neurosci 115:1012–1028. Corkin S, Amaral DG, Gonza´lez RG, Johnson KA, Hyman BT (1997) HM’s medial temporal lobe lesion: findings from magnetic resonance imaging. J Neurosci 17:3964–3979. Corodimas KP, LeDoux JE (1995) Disruptive effects of posttraining perirhinal cortex lesions on conditioned fear: contributions of contextual cues. Behav Neurosci 109:613–619. Csicsvari J, Hirase H, Czurko A, Buzsa´ki G (1998) Reliability and state dependence of pyramidal cell–interneuron synapses in the hippocampus: an ensemble approach in the behaving rat. Neuron 21:179–189. Damasio AR (1996) The somatic marker hypothesis and the possible functions of the prefrontal cortex. Philos Trans R Soc Lond Biol Sci 351:1413–1420. Davies M, Machin PE, Sanderson DJ, Pearce JM, Aggleton JP (2007) Neurotoxic lesions of the rat perirhinal and postrhinal cortices and their impact on biconditional visual discrimination tasks. Behav Brain Res 176:274–283. Diano S, Farr SA, Benoit SC, McNay EC, da Silva I, Horvath B, Gaskin FS, Nonaka N, Jaeger LB, Banks WA, Morley JE, Pinto S, Sherwin RS, Xu L, Yamada KA, Sleeman MW, Tschop MH, Horvath TL (2006) Ghrelin controls hippocampal spine synapse density and memory performance. Nat Neurosci 9: 381–388. Dijksterhuis A (2004) Think different: the merits of unconscious thought in preference development and decision making. J Pers Soc Psychol 87:586–598. Duncan J, Emslie H, Williams P, Johnson R, Freer C (1996) Intelligence and the frontal lobe: the organization of goal-directed behavior. Cognit Psychol 30:257–303. Dutar P, Bassant MH, Senut MC, Lamour Y (1995) The septohippocampal pathway: structure and function of
266
NEURAL SYSTEMS PERSPECTIVE ON PLACE FIELDS
a central cholinergic system. Physiol Rev 75:393– 427. Eacott MJ, Machin PE, Gaffan EA (2001) Elemental and configural visual discrimination learning following lesions to perirhinal cortex in the rat. Behav Brain Res 124:55–70. Egner T, Hirsch J (2005) Cognitive control mechanisms resolve conflict through cortical amplification of taskrelevant information. Nat Neurosci 8:1784–1790. Eichenbaum H (2000) Hippocampus: mapping or memory? Curr Biol 10:R785-R787. Eichenbaum H, Dudchenko P, Wood E, Shapiro M, Tanila H (1999) The hippocampus, memory, and place cells: is it spatial memory or a memory space? Neuron 23:209–226. Ekstrom AD, Kahana MJ, Caplan JB, Fields TA, Isham EA, Newman EL, Fried I (2003) Cellular networks underlying human spatial navigation. Nature 425: 184–188. Empson RM, Heinemann U (1995) Perforant path connections to area CA1 are predominantly inhibitory in the rat hippocampal–entorhinal cortex combined slice preparation. Hippocampus 5:104–107. Felleman DJ, Van Essen DC (1991) Distributed hierarchical processing in the primate cerebral cortex. Cereb Cortex 1:1–47. Floresco SB, Blaha CD, Yang CR, Phillips AG (2001a) Dopamine D1 and NMDA receptors mediate potentiation of basolateral amygdala-evoked firing of nucleus accumbens neurons. J Neurosci 21:6370–6376. Floresco SB, Blaha CD, Yang CR, Phillips AG (2001b) Modulation of hippocampal and amygdalar-evoked activity of nucleus accumbens neurons by dopamine: cellular mechanisms of input selection. J Neurosci 21:2851–2860. Foster TC, Castro CA, McNaughton BL (1989) Spatial selectivity of rat hippocampal neurons: dependence on preparedness for movement. Science 244:1580–1582. Freund TF, Antal M (1988) GABA-containing neurons in the septum control inhibitory interneurons in the hippocampus. Nature 336:170–173. Frith CD, Friston KJ, Liddle PF, Frackowiak RS (1991) A PET study of word finding. Neuropsychologia 29: 1137–1148. Fuster JM (1995) Memory and planning: two temporal perspectives of frontal lobe function. Adv Neurol 66: 9–20. Fyhn M, Molden S, Hollup S, Moser MB, Moser E (2002) Hippocampal neurons responding to first-time dislocation of a target object. Neuron 35:555–566. Fyhn M, Molden S, Witter MP, Moser EI, Moser M-B (2004) Spatial representation in the entorhinal cortex. Science 305(5688):1258–1264. Gaffan D, Murray EA (1992) Monkeys with rhinal cortex ablations succeed in object discrimination learning
despite 24-hr intervals and fail at matching to sample despite double sample presentations. Behavioural Neuroscience 106:30–38. Gao YJ, Ren WH, Zhang YQ, Zhao ZQ (2004) Contributions of the anterior cingulate cortex and amygdala to pain- and fear-conditioned place avoidance in rats. Pain 110:343–353. Gasbarri A, Sulli A, Packard MG (1997) The dopaminergic mesencephalic projections to the hippocampal formation in the rat. Prog Neuropsychopharmacol Biol Psychiatry 21:1–22. Gold JI, Shadlen MN (2000) Representation of a perceptual decision in developing oculomotor commands. Nature 404:390–394. Gold JI, Shadlen MN (2001) Neural computations that underlie decisions about sensory stimuli. Trends Cogn Sci 5:10–16. Gold JI, Shadlen MN (2002) Banburismus and the brain: decoding the relationship between sensory stimuli, decisions, and reward. Neuron 36:299–308. Goldman-Rakic PS (1987) Circuitry of primate preftonal cortex and regulation of behavior by representational memory. In: Higher Functions of the Brain, Part 1, Handbook of Physiology, Section 1: The Nervous System (Plum F, ed.), pp 373–417. Bethesda, MD: American Physiological Society. Goto Y, O’Donnell P (2001) Synchronous activity in the hippocampus and nucleus accumbens in vivo. J Neurosci 21:RC131. Gray JA, McNaughton N (2000) The Neuropsychology of Anxiety: An Enquiry into the Functions of the SeptoHippocampal System, second edition. Oxford: Oxford University Press. Groenewegen HJ, Uylings HB (2000) The prefrontal cortex and the integration of sensory, limbic and autonomic information. Prog Brain Res 126:3–28. Hafting T, Fyhn M, Molden S, Moser MB, Moser EI (2005) Microstructure of a spatial map in the entorhinal cortex. Nature 436:801–806. Halberstadt JB, Wilson T (in press) Reflections on conscious reflection. In: Reasoning: Studies of Human Inference and Its Foundations (Rips L, Adler J, eds.). Cambridge, UK: Cambridge University Press. Hampton RR (2001) Rhesus monkeys know when they remember. Proc Natl Acad Sci USA 98:5359–5362. Harnad S (1990) The symbol grounding problem. Phys D 42:335–346. Hetherington PA, Shapiro ML (1997) Hippocampal place fields are altered by the removal of single visual cues in a distance-dependent manner. Behav Neurosci 111: 20–34. Hirase H, Czurko A, Csicsvari J, Buzsaki G (1999) Firing rate and theta-phase coding by hippocampal pyramidal neurons during ‘space clamping.’ Eur J Neurosci 11:4373–4380.
NEOCORTICAL INFLUENCES ON HIPPOCAMPAL PLACE CELLS
Hirsh R (1974) The hippocampus and contextual retrieval of information from memory: a theory. Behav Biol 12: 421–444. Hok V, Save E, Lenck-Santini PP, Poucet B (2005) Coding for spatial goals in the prelimbic/infralimbic area of the rat frontal cortex. Proc Natl Acad Sci USA 102:4602–4607. Holland PC, Bouton ME (1999) Hippocampus and context in classical conditioning. [review]. Curr Opin Neurobiol 9:195–202. Hollup SA, Molden S, Donnett JG, Moser MB, Moser EI (2001) Accumulation of hippocampal place fields at the goal location in an annular watermaze task. J Neurosci 21:1635–1644. Holscher C, Jacob W, Mallot HA (2003) Reward modulates neuronal activity in the hippocampus of the rat. Behav Brain Res 142:181–191. Jarrard LE (1995) What does the hippocampus really do. Behav Brain Res 71:1–10. Jeffery KJ, Anderson MI (2003) Dissociation of the geometric and contextual influences on place cells. Hippocampus 13:868–872. Jeffery KJ, Anderson MI, Hayman R, Chakraborty S (2004) A proposed architecture for the neural representation of spatial context. Neurosci Biobehav Rev 28:201–218. Jeffery KJ, Gilbert A, Burton S, Strudwick A (2003) Preserved performance in a hippocampal-dependent spatial task despite complete place cell remapping. Hippocampus 13:175–189. Jones MW, Wilson MA (2005) Theta rhythms coordinate hippocampal-prefrontal interactions in a spatial memory task. PLoS Biol 3:e402. Kennedy PJ, Shapiro ML (2004) Retrieving memories via internal context requires the hippocampus. J Neurosci 24:6979–6985. Kentros C, Hargreaves E, Hawkins RD, Kandel ER, Shapiro M, Muller RV (1998) Abolition of long-term stability of new hippocampal place cell maps by NMDA receptor blockade. Science 280:2121–2126. Kim JJ, Fanselow MS (1992) Modality-specific retrograde amnesia of fear. Science 256:675–677. Knierim JJ (2002) Dynamic interactions between local surface cues, distal landmarks, and intrinsic circuitry in hippocampal place cells. J Neurosci 22:6254–6264. Knight RT, Grabowsky M (1995) Escape from linear time: prefrontal cortex and conscious experience. In: The Cognitive Neurosciences (Gazzaniga MS, ed.), pp 1357–1371. Cambridge, MA: MIT press. Kobayashi T, Nishijo H, Fukuda M, Bures J, Ono T (1997) Task-dependent representations in rat hippocampal place neurons. J Neurophysiol 78:597–613. Kobayashi T, Tran AH, Nishijo H, Ono T, Matsumoto G (2003) Contribution of hippocampal place cell activity to learning and formation of goal-directed navigation in rats. Neuroscience 117:1025–1035.
267
Kyd RJ, Bilkey DK (2003) Prefrontal cortex lesions modify the spatial properties of hippocampal place cells. Cereb Cortex 13:444–451. Kyd RJ, Bilkey DK (2005) Hippocampal place cells show increased sensitivity to changes in the local environment following prefrontal cortex lesions. Cereb Cortex 15:720–731. LeDoux J (1996) The Emotional Brain: The Mysterious Underpinnings of Emotional Life. New York: Simon and Schuster. Lee I, Yoganarasimha D, Rao G, Knierim JJ (2004) Comparison of population coherence of place cells in hippocampal subfields CA1 and CA3. Nature 430:456–459. Lenck-Santini PP, Save E, Poucet B (2001) Evidence for a relationship between place-cell spatial firing and spatial memory performance. Hippocampus 11:377–390. Leutgeb S, Leutgeb JK, Barnes CA, Moser EI, McNaughton BL, Moser MB (2005) Independent codes for spatial and episodic memory in hippocampal neuronal ensembles. Science 309:619–623. Leutgeb S, Leutgeb JK, Treves A, Moser M, Moser EI (2004) Distinct ensemble codes in hippocampal areas CA3 and CA1. Science 305:1295–1298. Liu P, Bilkey DK (2001) The effect of excitotoxic lesions centered on the hippocampus or perirhinal cortex in object recognition and spatial memory tasks. Behav Neurosci 115:94–111. Lu X, Bilkey DK (2002) Perirhinal cortex lesions affect place field size and may alter phase precession of hippocampal place cells. Soc Neurosci Abstr 477:8. Maguire EA, Frith CD, Burgess, N, Donnet JG, O’Keefe J (1998) Knowing where and getting there: a human navigation network. Science 280:921–924. Maguire EA, Gadian DG, Johnsrude IS, Good CD, Ashburner J, Frackowiak RS, Frith CD (2000) Navigation-related structural change in the hippocampi of taxi drivers. Proc Natl Acad Sci USA 97:4398–4403. Maguire EA, Spiers HJ, Good CD, Hartley T, Frackowiak RSJ, Burgess N (2003) Navigation expertise and the human hippocampus: A structural brain imaging analysis. Hippocampus 13:250–259. Malouf AT, Robbins CA, Schwartzkroin PA (1990) Phaclofen inhibition of the slow inhibitory postsynaptic potential in hippocampal slice cultures: a possible role for the GABAB-mediated inhibitory postsynaptic potential. Neuroscience 35:53–61. Martin PD (2001) Locomotion towards a goal alters the synchronous firing of neurons recorded simultaneously in the subiculum and nucleus accumbens of rats. Behav Brain Res 124:19–28. Martin PD, Ono T (2000) Effects of reward anticipation, reward presentation, and spatial parameters on the firing of single neurons recorded in the subiculum and nucleus accumbens of freely moving rats. Behav Brain Res 116:23–38.
268
NEURAL SYSTEMS PERSPECTIVE ON PLACE FIELDS
McClure SM, Laibson DI, Loewenstein G, Cohen JD (2004) Separate neural systems value immediate and delayed monetary rewards. Science 306:503–507. McDonald RJ, Murphy RA, Guarraci FA, Gortler JR, White NM, Baker AG (1997) Systematic comparison of the effects of hippocampal and fornix-fimbria lesions on acquisition of three configural discriminations. Hippocampus 7:371–388. Meunier M, Bachevalier J, Mishkin M, Murray EA (1993) Effects on visual recognition of combined and separate ablations of the entorhinal and perirhinal cortex in rhesus monkeys. J Neurosci 13:5418–5432. Meunier M, Hadfield W, Bachevalier J, Murray EA (1996) Effects of rhinal cortex lesions combined with hippocampectomy on visual recognition memory in rhesus monkeys. J Neurophysiol 75:1190–1205. Meyer-Lindenberg A, Poline JB, Kohn PD, Holt JL, Egan MF, Weinberger DR, Berman KF (2001) Evidence for abnormal cortical functional connectivity during working memory in schizophrenia. Am J Psychiatry 158: 1809–1817. Miller BT, D’Esposito M (2005) Searching for ‘‘the top’’ in top-down control. Neuron 48:535–538. Miller EK, Freedman DJ, Wallis JD (2002) The prefrontal cortex: categories, concepts and cognition. Philos Trans R Soc Lond B Biol Sci 357:1123–1136. Milner B (1982) Some cognitive effects of frontal lobe lesions in man. In: TheNeuropsychology of Cognitive Functions (Broadbent DE, Weiskrantz, L, eds.), pp 211–226. London: The Royal Society. Mishkin M (1978) Memory in monkeys severely impaired by combined but not separate removal of amygdala and hippocampus. Nature 273:297–298. Mogenson GJ, Jones DL, Yim CY (1980) From motivation to action: functional interface between the limbic system and the motor system. Prog Neurobiol 14:69–97. Montague PR, Berns GS (2002) Neural economics and the biological substrates of valuation. Neuron 36:265–284. Moran JP, Dalrymple-Alford JC (2003) Perirhinal cortex and anterior thalamic lesions: comparative effects on learning and memory. Behav Neurosci 117:1326–1341. Morris RGM (1990) Does the hippocampus play a disproportionate role in spatial memory. Discussions in Neuroscience VI:39–45. Morris RGM, Garrud P, Rawlins JNP, O’Keefe J (1982) Place navigation impaired in rats with hippocampal lesions. Nature 297:681–683. Muir GM, Bilkey DK (2001) Instability in the place field location of hippocampal place cells after lesions centered on the perirhinal cortex. J Neurosci 21:4016– 4025. Muir GM, Bilkey DK (2003) Theta- and movement velocity-related firing of hippocampal neurons is disrupted by lesions centered on the perirhinal cortex. Hippocampus 13:93–108.
Muller RU (1996) A quarter of a century of place cells. Neuron 17:813–822. Muller RU, Kubie JL (1987) The effects of changes in the environment on the spatial firing of hippocampal complex spike cells. J Neurosci 7:1951–1968. Muller RU, Kubie JL, Ranck JB (1987) Spatial firing patterns of hippocampal complex spike cells in a fixed environment. J Neurosci 7:1935–1950. Muller RU, Stead M (1996) Hippocampal place cells connected by Hebbian synapses can solve spatial problems. Hippocampus 6:709–719. Naber PA, Caballero-Bleda M, Jorritsma-Byham B, Witter MP (1997) Parallel input to the hippocampal memory system through peri- and postrhinal cortices. Neuroreport 8:2617–2621. Naber PA, Witter MP, Da Silva FHL (2000) Differential distribution of barrel or visual cortex—evoked responses along the rostro-caudal axis of the peri- and postrhinal cortices. Brain Res 877:298–305. Naber PA, Witter MP, Lopez da Silva FH (1999) Perirhinal cortex input to the hippocampus in the rat: evidence for parallel pathways, both direct and indirect. A combined physiological and anatomical study. Eur J Neurosci 11:4119–4133. Nadel L, Willner J (1980) Context and conditioning: a place for space. Physiol Psychol 8:218–228. Norman G, Eacott MJ (2004) Impaired object recognition with increasing levels of feature ambiguity in rats with perirhinal cortex lesions. Behav Brain Res 148:79–91. Norman G, Eacott MJ (2005) Dissociable effects of lesions to the perirhinal cortex and the postrhinal cortex on memory for context and objects in rats. Behav Neurosci 119:557–566. O’Keefe J, Burgess N (1996) Geometric determinants of the place fields of hippocampal neurons. Nature 381:425–428. O’Keefe J, Conway DH (1978) Hippocampal place units in the freely moving rat: why they fire where they fire. Exp Brain Res 31:573–590. O’Keefe J, Dostrovsky J (1971) The hippocampus as a spatial map. Preliminary evidence from unit activity in the freely moving rat. Brain Res 34:171–175. O’Keefe J, Nadel L (1978) The Hippocampus as a Cognitive Map. Oxford: Clarendon Press. O’Keefe J, Speakman A (1987) Single unit activity in the rat hippocampus during a spatial memory task. Exp Brain Res 68:1–27. Olypher AV, Lansky P, Fenton AA (2002) Properties of the extra-positional signal in hippocampal place cell discharge derived from the overdispersion in locationspecific firing. Neuroscience 111:553–566. Oomura Y, Hori N, Shiraishi T, Fukunaga K, Takeda H, Tsuji M, Matsumiya T, Ishibashi M, Aou S, Li XL, Kohno D, Uramura K, Sougawa H, Yada T, Wayner MJ, Sasaki K (2006) Leptin facilitates learning and
NEOCORTICAL INFLUENCES ON HIPPOCAMPAL PLACE CELLS
memory performance and enhances hippocampal CA1 long-term potentiation and CaMK II phosphorylation in rats. Peptides 27:2738–2749. Papp EC, Hajos N, Acsa´dy L, Freund TF (1999) Medial septal and median raphe innervation of vasoactive intestinal polypeptide-containing interneurons in the hippocampus. Neuroscience 90:369–382. Parker A, Gaffan D (1998) Interaction of frontal and perirhinal cortices in visual object recognition memory in monkeys. Eur J Neurosci 10:3044–3057. Petrides M (1994) Frontal lobes and working memory: evidence from investigations of the effects of cortical excisions in nonhuman primates. In: Handbook of Neurophysiology (Boller FSF, Hendler JA, eds.), pp 59–82. Amsterdam: Elsevier. Phillips RG, LeDoux JE (1992) Differential contribution of amygdala and hippocampus to cued and contextual fear conditioning. Behav Neurosci 106:274–285. Poucet B, Lenck-Santini PP, Hok V, Save E, Banquet JP, Gaussier P, Muller RU (2004) Spatial navigation and hippocampal place cell firing: the problem of goal encoding. Rev Neurosci 15:89–107. Quirk GJ, Muller RU, Kubie JL, Ranck JB (1992) The positional firing properties of medial entorhinal neurons—description and comparison with hippocampal place cells. J Neurosci 12:1945–1963. Quirk MC, Blum KI, Wilson MA (2001) Experiencedependent changes in extracellular spike amplitude may reflect regulation of dendritic action potential back-propagation in rat hippocampal pyramidal cells. J Neurosci 21:240–248. Reep RL, Corwin JV, Hashimoto A, Watson RT (1987) Efferent connections of the rostral portion of medial agranular cortex in rats. Brain Res Bull 19: 203–221. Robinson TE, Berridge KC (1993) The neural basis of drug craving: an incentive-sensitization theory of addiction. Brain Res Brain Res Rev 18:247–291. Rule RR, Shimamura AP, Knight RT (2002) Orbitofrontal cortex and dynamic filtering of emotional stimuli. Cogn Affect Behav Neurosci 2:264–270. Russell NA, Horii A, Smith PF, Darlington CL, Bilkey DK (2006) Lesions of the vestibular system disrupt hippocampal theta rhythm in the rat. J Neurophysiol 96:4–14. Sacchetti B, Lorenzini CA, Baldi E, Tassoni G, Bucherelli C (1999) Auditory thalamus, dorsal hippocampus, basolateral amygdala, and perirhinal cortex role in the consolidation of conditioned freezing to context and to acoustic conditioned stimulus in the rat. J Neurosci 19:9570–9578. Schultz W (2004) Neural coding of basic reward terms of animal learning theory, game theory, microeconomics and behavioural ecology. Curr Opin Neurobiol 14: 139–147.
269
Scoville WB, Milner B (1957) Loss of recent memory after bilateral hippocampal lesions. J Neurol Neurosurgpsychiatry 20:11–21. Sesack SR, Deutch AY, Roth RH, Bunney BS (1989) Topographical organization of the efferent projections of the medial prefrontal cortex in the rat: an anterograde tract-tracing study with Phaseolus vulgaris leucoagglutinin. J Comp Neurol 290:213–242. Sesack SR, Pickel VM (1992) Prefrontal cortical efferents in the rat synapse on unlabeled neuronal targets of catecholamine terminals in the nucleus accumbens septi and on dopamine neurons in the ventral tegmental area. J Comp Neurol 320:145–160. Shadlen MN (2002) Pursuing commitments. Nat Neurosci 5:819–821. Shallice T (1982) Specific impairments of planning. Philos Trans R Soc Lond B Biol Sci 299:199–209. Sherry DF, Jacobs LF, Gaulin SJC (1992) Spatial memory and adaptive specialization of the hippocampus. Trends Neurosci 15:298–303. Shibata R, Mulder AB, Trullier O, Wiener SI (2001) Position sensitivity in phasically discharging nucleus accumbens neurons of rats alternating between tasks requiring complementary types of spatial cues. Neuroscience 108:391–411. Shimamura AP (2000) The role of prefrontal cortex in dynamic filtering. Psychobiology 28:207–218. Siapas AG, Lubenov EV, Wilson MA (2005) Prefrontal phase locking to hippocampal theta oscillations. Neuron 46:141–151. Skaggs WE, McNaughton BL (1998) Spatial firing properties of hippocampal CA1 populations in an environment containing two visually identical regions. J Neurosci 18:8455–8466. Skaggs WE, McNaughton BL, Wilson MA, Barnes CA (1996) Theta phase precession in hippocampal neuronal populations and the compression of temporal sequences. Hippocampus 6:149–172. Smith DM, Mizumori SJ (2006a) Learning-related development of context-specific neuronal responses to places and events: the hippocampal role in context processing. J Neurosci 26:3154–3163. Smith DM, Mizumori SJ (2006b) Hippocampal place cells, context, and episodic memory. Hippocampus 16:716–729. Solstad T, Moser EI, Einevoll GT (2006) From grid cells to place cells: a mathematical model. Hippocampus 16:1026–1031. Speakman A, O’Keefe J (1990) Hippocampal complex spike cells do not change their place fields if the goal is moved within a cue controlled environment. Eur J Neurosci 2:544–555. Spiers HJ, Maguire EA (2006) Thoughts, behaviour, and brain dynamics during navigation in the real world. Neuroimage 31:1826–1840.
270
NEURAL SYSTEMS PERSPECTIVE ON PLACE FIELDS
Squire LR (1992) Memory and the hippocampus: a synthesis from findings with rats, monkeys and humans. Psychol Rev 99:195–231. Suzuki WA, Clayton NS (2000) The hippocampus and memory: a comparative and ethological perspective. Curr Opin Neurobiol 10:768–773. Taber MT, Das S, Fibiger HC (1995) Cortical regulation of subcortical dopamine release: mediation via the ventral tegmental area. J Neurochem 65:1407–1410. Tabuchi E, Mulder AB, Wiener SI (2000) Position and behavioral modulation of synchronization of hippocampal and accumbens neuronal discharges in freely moving rats. Hippocampus 10:717–728. Tabuchi E, Mulder AB, Wiener SI (2003) Reward value invariant place responses and reward site associated activity in hippocampal neurons of behaving rats. Hippocampus 13:117–132. Tanila H, Shapiro ML, Eichenbaum H (1997) Discordance of spatial representation in ensembles of hippocampal place cells. Hippocampus 7:613–623. Tracy AL, Jarrard LE, Davidson TL (2001) The hippocampus and motivation revisited: appetite and activity. Behav Brain Res 127:13–23. Trivedi MA, Coover GD (2004) Lesions of the ventral hippocampus, but not the dorsal hippocampus, impair conditioned fear expression and inhibitory avoidance on the elevated T-maze. Neurobiol Learn Mem 81: 172–184. Tsubokawa H, Ross RN (1996) IPSPs modulate spike backpropagation and associative [Ca2þ]I changes in the dendrites of hippocampal CA1 pyramidal neurons. J Neurosci 76:2896–2906. Vanderwolf CH (2001) The hippocampus as an olfactomotor mechanism: were the classical anatomists right after all? Behav Brain Res 127:25–47. Wallace DG, Gorny B, Whishaw IQ (2002) Rats can track odors, other rats, and themselves: implications for the study of spatial behavior. Behav Brain Res 131:185– 192. Wang D, Maler L (1997) In vitro plasticity of the direct feedbackpathwayintheelectrosensorysystemofApteronotus leptorhynchus. J Neurophysiol 78:1882–1889. Wiig KA, Bilkey DK (1995) Lesions of rat perirhinal cortex exacerbate the memory deficit observed fol-
lowing damage to the fimbria-fornix. Behav Neurosci 109:620–630. Wilson MA, McNaughton BL (1993) Dynamics of the hippocampal ensemble code for space. Science 261: 1055–1058. Wilson TD, Schooler JW (1991) Thinking too much: introspection can reduce the quality of preferences and decisions. J Pers Soc Psychol 60:181–192. Winters BD, Forwood SE, Cowell RA, Saksida LM, Bussey TJ (2004) Double dissociation between the effects of peri-postrhinal cortex and hippocampal lesions on tests of object recognition and spatial memory: heterogeneity of function within the temporal lobe. J Neurosci 24:5901–5908. Witter MP, Wouterlood FG, (Eds) (2002) The Parahippocampal Region: Organization and Role in Cognitive Function. Oxford: Oxford University Press. Witter MP, Naber PA, van Haeften T, Machielsen WC, Rombouts SA, Barkhof F, Scheltens P, Lopes da Silva FH (2000) Cortico-hippocampal communication by way of parallel parahippocampal-subicular pathways. Hippocampus 10:398–410. Young WS, Li J, Wersinger SR, Palkovits M (2006) The vasopressin 1b receptor is prominent in the hippocampal area CA2 where it is unaffected by restraint stress or adrenalectomy. Neuroscience 143: 1031–1039. Zinyuk L, Kubik S, Kaminsky Y, Fenton AA, Bures J (2000) Understanding hippocampal activity by using purposeful behavior: place navigation induces place cell discharge in both task-relevant and task-irrelevant spatial reference frames. Proc Natl Acad Sci USA 97: 3771–3776. Zironi I, Iachovelli P, Aicardi G, Liu P, Bilkey DK (2001) Lesions of prefrontal cortex augment the locationrelated firing properties of area TE/perirhinal cortex neurons in a working memory task. Cereb Cortex 11: 1093–1100. Zola-Morgan S, Squire LR, Amaral DG, Suzuki WA (1989) Lesions of perirhinal and parahippocampal cortex that spare the amygdala and hippocampal formation produce severe memory impairment. J Neurosci 9:4355–4370.
17 Spatial Learning and the Selectivity of Hippocampal Place Fields: Modulation by Dopamine KATHRYN M. GILL AND SHERI J. Y. MIZUMORI
Dopamine is a potent neuromodulator with profound consequences on learning and memory. Anatomically, midbrain dopamine projections are situated to cause widespread impact on multiple brain areas. The breadth of dopamine-related connectivity supports its alleged role in mediating different memory systems, thereby ultimately directing behavior. This chapter will compare the putative mechanisms by which dopamine selectively affects learning that is thought to be mediated by functionally distinct brain regions such as the hippocampus and dorsal striatum. First, a summary of the behavioral changes that result from alterations in normal dopamine signaling is discussed. Specifically, the way in which dopamine mediates hippocampaland dorsal striatal–dependent behaviors is outlined. Next, the synaptic plasticity mechanisms underlying the effects of dopamine on behavior are explored. Included is a discussion of the participation of dopamine in long-term potentiation (LTP) and long-term depression (LTD) and the involvement of dopamine in the activation of immediate early genes in brain regions such as the hippocampus and dorsal striatum. The third section of this chapter will discuss how the influence of dopamine on synaptic plasticity corresponds to structure-specific neural signaling, potentially contributing to functional specialization of individual regions. Research into the participation of hippocampus in learning and memory is indelibly linked to its role in the mediation of spatial behaviors. Compromised hippocampal function in human patients with damage to temporal lobe areas encompassing hip-
pocampus and parahippocampal cortices is often associated with profound anterograde amnesia and an inability to process spatial information in a meaningful way (Goldstein et al., 1989; Abrahams et al., 1997; Bohbot et al., 2002; Maguire et al., 2006). Similar decrements in spatial performance occur in rodents and primates with more focal damage isolated to the hippocampus proper (reviewed in Jarrad 1993, Chang and Gold 2003; Broadbent et al., 2006; Lavenex et al., 2006). Electrophysiological studies in rodents have provided additional insight into the processing of spatial information by the hippocampus with the discovery of place cells (O’Keefe and Dostrovsky, 1971; O’Keefe 1976; O’Keefe and Conway, 1978). By illustrating that the hippocampus is capable of producing the specialized neural signal necessary to convey spatial location in a relevant way, the importance of the hippocampus in spatial learning has been strongly supported. Lesion studies in rodents support dissociation between the hippocampus and other areas, such as dorsal striatum, in mediating allocentric behaviors. However, place cells have also been observed in these same extrahippocampal areas (Mizumori et al., 2000b; Yeshenko et al., 2004; Gill and Mizumori 2006; Eschenko and Mizumori, 2007). Just as spatial selectivity of single-unit activity extends to multiple brain regions, likewise the alteration of spatial processing by dopamine inputs may not be unique to the hippocampus. Therefore, this third section will compare the effects of dopamine on single-unit, in particular place cell,activitywithinthehippocampusanddorsalstriatum.
271
272
NEURAL SYSTEMS PERSPECTIVE ON PLACE FIELDS
The fourth section will discuss computational models that account for dopamine’s role in the modulation of hippocampal place cell activity. Specifically, we will explore how the dopamine signal conveys salient information to hippocampal place cells as a function of novelty or alterations in familiar contexts. The final section will discuss the involvement of other neuromodulatory systems, such as acetylcholine, in neural representation relevant to learning and memory.
IMPACT OF LOSS OF DOPAMINE FUNCTION ON LEARNING AND MEMORY The devastation on motor activity following substantial loss of dopamine function in Parkinson disease (PD) overshadows the sometimes subtle and complex effects on cognitive abilities. Indeed, early stages of the illness are often characterized more by these changes in memory processing than by loss of motor function (Pillon et al., 1998). Deficits include, but are not limited to, decreased working memory and probabilistic learning (Dubois and Pillon, 1997; Zgaljardic et al., 2003). Various tests of spatial memory, measured by accurate recall of associations between stimuli and arbitrary locations on a grid without the presence of cues to assist in recall, indicate that recently diagnosed patients with PD are profoundly impaired in retrieving spatial information that has been recently encoded. These spatial deficits occur without significant impact on verbal learning and memory (Pillon et al., 1997). Important to note is the observation that initial spatial learning appears to be intact, while the maintenance of these spatial representations in memory is impaired. Pillon and colleagues (1997, 1998) have proposed the idea that ‘‘strategic’’ processes are interrupted in PD patients. These processes likely involve the selection of appropriate behavioral responses based on current sensory information and expectation provided by memory. During situations when less external information is available to cue behavior, the role of the hippocampus becomes more apparent. While the pattern of deficits in Parkinson patients is hypothesized to result mainly from a loss of dopamine function in frontal cortical areas, abnormal activity within the hippocampus itself or disruption of an interaction between hippocampus and cortical areas cannot be discounted. As the illness progresses and there is a more global loss of dopamine connectivity, the impact on temporal lobe–mediated learning becomes more apparent (Owen et al., 1997). The global loss of dopamine system functioning makes it difficult to discern the importance of dopamine signaling within an individual region such as the
hippocampus. However, different tests of cognitive function in Parkinson patients could reveal potential candidate structures for the pattern of visuospatial deficits observed. Dujardin and Laurent (2003) have suggested that compromised episodic memory in PD is directly linked to an inability to establish new associations between complex contextual stimuli. In other words, patients are unimpaired in learning stimulus pairs if a prior associative link has already been established. Spatial learning is particularly demanding in the sense that navigating a new environment often requires forming seeming arbitrary associations between environmental stimuli. Thus, spatial performance in PD would be expected to be preserved in familiar environments, or certain conditions in which prior associations have already been made between stimuli, and severely compromised in novel contexts. Hippocampus may be involved in the initial formation of complex contextual representations. The loss of dopamine may subsequently interfere with the normal interaction between hippocampus and cortex, a process thought to mediate memory retrieval. Compared to studies of Parkinson disease, a more systematic approach to disrupting dopamine function within specific brain regions of primates and rodents has revealed the pervasive nature of dopamine influences in multiple brain regions and, by proxy, multiple memory systems. There are various pharmacological and neurochemical methods for examining dopamine’s impact on learning and memory in animals. To induce dopamine depletion in a manner similar to that found in PD, the neurotoxin 1-methyl-4-phenyl1,2,3,6-tetrahydropyridine (MPTP) is commonly used. Localized dopamine depletion can also be accomplished via infusions of 6-OHDA. Interestingly, destruction of the dopamine system alone within individual areas, such as the hippocampus and striatum, causes similar, functionally distinct, behavioral impairments to those when the individual structures are lesioned. The functional separation of hippocampus and dorsal striatum in lesion studies may therefore be based in part on anatomical differences in dopamine connectivity. Consistent with this view, selective lesions of the dopamine system further distinguish disparate memory systems. Efferent dopamine projections from the ventral tegmental area (VTA) and substantia nigra innervate distinct brain regions (Gasbarri et al., 1994a, b, 1997). Limbic brain regions, including the hippocampus, receive dopamine input predominately from the VTA while dopamine neurons in the substantia nigra supply mainly frontostriatal brain regions. MPTP treatment causes a relatively focal loss (40%–70%) of dopamine in the substantia nigra while leaving VTA dopamine fibers intact (Gevaerd et al., 2001). Consequently, there is a significant loss of
MODULATION OF SPATIAL LEARNING BY DOPAMINE
dopamine input to prefrontal cortical and striatal brain regions whereas limbic areas are unaffected. Lesion of dopamine fibers from the substantia nigra selectively impairs performance during the cued version of the water maze task while sparing performance during the spatial version of the task (Da Cunha et al., 2002; Miyoshi et al., 2002). When hippocampal inactivation is performed in the same animals, there is no significant interaction effect with the substantia nigra treatment. In other words, by selectively interfering with striataldependent stimulus–response learning, dopamine projections from the substantia nigra support a memory system separate from the hippocampal system. This evidence supports the idea that the dopamine system represents a common cellular mechanism that individual brain regions use to support their unique contributions to behavioral expression during learning. So, in that sense, dopamine is truly a global neuromodulator. It would be imprudent to conclude from the above evidence that dopamine is crucial for mediating dorsal striatal–dependent behaviors and inconsequential for hippocampal-dependent learning. Selective removal of hippocampal dopamine input via local 6-OHDA infusions into the subiculum and adjacent CA1 region impairs performance in the spatial version of the water maze (Gasbarri et al., 1996). Similar to the dissociation reported for substantia nigra lesions, performance in the cued version of the water maze, as well as inhibitory avoidance, is unaffected by the loss of hippocampal dopamine. Pharmacological manipulations provide further evidence of the modulatory capacity of dopamine on hippocampal- and dorsal striatal–dependent behaviors. Local post-training infusion of dopamine agonists, selective for both D1 and D2 receptors, can enhance either hippocampal- or dorsal striatal–dependent learning (Packard and White, 1991). In addition, dopamine agonist treatment in the hippocampus can reverse age-related decreases in spatial performance (Bach et al., 1999; Behr et al., 2000). A closer examination of the modulation of hippocampal function by dopamine reveals differences in D1 receptor–mediated gating of hippocampal output, via dorsal or ventral subiculum. Instrumental learning and performance, and not spontaneous motor activity, is reduced following infusions of a D1-receptor antagonist in the ventral, but not dorsal, subiculum (Andrejewski et al., 2006). This effect on learning is thought to result from an interruption of the normal interaction between NMDA and D1, or glutamate and dopamine, systems. The impact of the interaction between glutamaterigic and dopaminergic systems on plasticity mechanisms in other hippocampal subregions, CA1 and CA3, is explored further in later sections of this chapter.
273
Wolfram Schultz and colleagues have made substantial contributions to our understanding of the role of dopamine during learning (reviewed in Schultz 1998, 2004). Specifically, the details of the pattern of dopamine neural activity during classical conditioning in primates were investigated. It was found that initially dopamine neurons respond robustly to unexpected, and consequently unpredicted, presentation of reinforcement. The temporal-difference (TD) learning algorithm has been used to calculate probabilities of reinforcement based on a comparison of past experience with current behavioral responses and subsequent success. The TD model can be used to describe how neural responses to stimuli can change during learning as prediction improves linking the stimuli with its expected probability of reinforcement. The TD model predicts that dopamine neural activation will gradually shift to earlier predictors of subsequent reinforcement (Suri and Schultz, 2001; reviewed in Suri, 2002). Predictability, defined by reinforcement expectation, will also dramatically alter dopamine neural activity. Initially, presentation of reward when it is not expected, or the association between a conditioned stimulus and subsequent reward that is not yet acquired, induces robust dopamine firing. After learning, fully predicted reward omission results in a significant reduction in dopamine bursting (Schultz et al., 1993; Mirenowicz and Schultz, 1994; Hollerman and Schultz 1998). Changes in dopamine neural activity permit an evaluation of current reinforcement contingencies in a given context by simultaneously activating different neural systems, including but not limited to dorsal striatum and hippocampus. The associative activity of dopamine neurons may result in part from a feedback loop involving hippocampal input. Electrophysiological recordings from both the hippocampus and dorsal striatum demonstrate the capacity for neurons in these regions to encode contextual information by responding to changes in the expected, or familiar, environment (Mizumori et al., 2000b, 2004; Gill and Mizumori, 2006). Alterations in the pattern of dopamine activity arriving at these regions could be responsible for the detection of alterations in a familiar context.
DOPAMINE AND SYNAPTIC PLASTICITY Dopamine and Modulation of Synaptic Plasticity: Implications for Hippocampal Function and Place Cell Activity Consistent with its classification as a neuromodulator are findings that dopamine can alter the strength of neural activity, leading to changes in synaptic plasticity. In vitro studies have aided in deconstructing the
274
NEURAL SYSTEMS PERSPECTIVE ON PLACE FIELDS
influence of dopamine on hippocampal neural transmission. Hippocampal subregions vary in their response to exogenous dopamine administration and the dopamine receptor subtypes involved. Following stimulation of the Schaffer collaterals, CA1 neurons exhibit an excitatory postsynaptic potential (EPSP) that is decreased following application of a relatively low dose of dopamine (0.3 mg; Hsu, 1996). The reduction in EPSPs in CA1 was verified pharmacologically to result from changes in presynaptic D2 receptors. The reduction of excitatory, namely glutamatergic, responses in CA1 by dopamine is remarkably similar to the gating of NMDA-mediated responses in dorsal striatum medium spiny neurons (Flores-Hernandez, 2002; Lin et al., 2003). Hippocampal output via the subiculum is also modulated by dopamine afferents. A larger dose of dopamine (100 mg) than that applied to CA1 neurons (above) was used to reduce both alveus (56%) and perforant path (76%) elicited EPSPs in subiculum (Behr et al., 2000). Regardless of the dopamine receptor subtype involved, excitatory inputs to the hippocampus must surpass the inhibitory influence of low doses of dopamine. However, when even larger quantities of dopamine (250 mg) are applied to hippocampal slices, there is a facilitation of long-lasting synaptic potentiation in the CA1 region (Huang and Kandel, 1995). Therefore, it appears that changes in dopamine action may be accomplished in part by changes in the quantity of dopamine released. Essentially, dopamine acts to dose-dependently gate excitatory drive by reducing the effectiveness of potentially irrelevant inputs. By establishing the overall effectiveness of excitatory inputs within a structure, dopamine could be part of a mechanism that determines the likelihood that new or salient information is consolidated. Persistent alterations in neurotransmission, LTP and LTD, are described as plausible cellular mechanisms underlying learning and memory. The duration of LTP can vary depending on the pattern of neural activation used for induction (Frey and Morris, 1997). It has been verified immunohistochemically that there is dense localization of D1 and D5 receptors in the pyramidal cell layer of CA1 (Huang et al., 1992). Dopamine, specifically transmitted via D1 receptors, is critical for the maintenance of late-phase LTP (L-LTP) in the CA1 region of the hippocampus (Frey et al., 1990, 1991; Huang and Kandel, 1995; Matthies et al., 1997; Williams et al., 2006). Dopamine application is also capable of inducing LTP, or early-phase LTP (E-LTP), in a different subregion of the hippocampus, the dentate gyrus, following stimulation protocols that are normally insufficient (Kusuki et al., 1997). There is some indication that dopamine agonists alone may be sufficient to induce a slowly developing potentiation that is in-
dependent of any other external stimulation (Huang and Kandel, 1995; Williams et al., 2006). The effect of dopamine on the duration of signal strength can increase the associative capacity of temporally discrete events. Importantly, the dopamine agonist–induced L-LTP can be significantly attenuated by NMDA-receptor antagonism, supporting a potential interaction between these neurotransmitter systems. There is additional evidence that the interaction between glutamatergic and dopaminergic systems modulates heterosynaptic LTP (in vitro), whereby weak inputs are strongly potentiated (O’Carroll and Morris, 2004). NMDA-receptor activation in hippocampus may ‘‘set’’ or produce synaptic tags as well as synergize with neuromodulatory signals, such as dopamine, to initiate increases in mRNA and protein synthesis that is part of longlasting potentiation (Frey and Morris, 1997). In the synaptic tagging hypothesis, the transference of E-LTP to L-LTP is achieved via formation of a tag at the time potentiation was induced. Some caution is warranted when extending the conclusions from LTP/LTD experiments to learning and memory paradigms. The stimulation protocols used to induce LTP/LTD can be considered unlikely to occur during natural learning scenarios. However, there is persuasive evidence that lasting changes in synaptic plasticity in the hippocampus can result from exposure to different spatial contexts. Even relatively brief exposure to a novel environment is sufficient to lower the threshold for induction of LTP in CA1 (Li et al., 2003). Dopamine is also implicated in this context-induced change in hippocampal synaptic plasticity. Pretreatment with a D1/D5 receptor antagonist interferes with the LTP-inducing effects of spatial exploration (Li et al., 2003; Lemon and ManahanVaughan, 2006). The ability of dopamine to gate exploration-induced synaptic plasticity may be reflected in changes in spatially selective neural activity. In other words, dopamine may also act to stabilize place field properties.
Dopamine and Modulation of Synaptic Plasticity: Intracellular Mechanisms In general, the intracellular downstream effects of the dopamine signal result from its regulation of adenylate cyclase activation. Given the involvement in the protein synthesis–dependent late phase of LTP, dopamine is anticipated to be involved in the initiation of protein synthesis. That is, dopamine may be part of initiating what Clayton (2000) described as the genomic action potential, via changes in the probability of activation of transcription factors, such as cAMP-response element binding protein (CREB), and activation of various
MODULATION OF SPATIAL LEARNING BY DOPAMINE
immediate early genes (IEG). Dopamine may directly regulate the signal transduction pathways implicated in the synthesis of both regulatory (e.g., zif268, c-fos) and effector (e.g., Arc) IEGs. While effector IEGs are proposed to have a more direct effect on cellular function, both classes of IEGs can contribute to plasticity mechanisms. Each individual neuron may have a threshold for genomic activation (Lanahan and Worley, 1998). In addition, a given structure can vary in the proportion of neurons engaged in the genomic response to a given stimulus. Clayton (2000) described the IEG response in the genomic action potential as a ‘‘memory rheostat’’ that signals the likelihood that memories occurring within a given timeframe will be consolidated. Consequently, by regulating both the duration of LTP and the magnitude of the IEG response, dopamine may increase the likelihood of forming a persistent representation of associated stimuli. In this way, dopamine may be considered a molecular switch by which transient experiences are transformed to stable representations. Kang et al. (2000) showed that treatment of hippocampal slices with dopamine or a D1 agonist, SKF38393, can cause persistent increases in both c-fos mRNA, measured by reverse-transcription polymerase chain reaction (RT-PCR), and protein levels determined immunocytochemically. Extending beyond the pharmacological induction of IEGs, behaviorally relevant patterns of IEG activation can be mapped in different brain areas. There is accumulating evidence that increases in levels of the mRNA or protein products of these IEGs within individual brain regions, such as the hippocampus and dorsal striatum, are correlated with learning (Clayton, 2000; Guzowski et al., 2005). It has been shown with contextual fear conditioning that a re-instatement of the IEG response can occur with the introduction of a salient event, e.g., shock, in a familiar context (Campeau et al., 1991; Beck and Fibiger, 1995). In other words, new learning in a familiar context would be expected to trigger IEG activation. Perhaps more compelling is the potential for IEG levels to act as indicators of relative structural involvement in various learning paradigms. The evidence from spatial-learning experiments overall support the idea that increased hippocampal activation, reflected by increases in Arc mRNA or cFos protein, is associated with spatial strategy use during both Tmaze and water maze tasks (Vann et al., 2000; Guzowski et al., 2001; Colombo et al., 2003; Teather et al., 2005). These activation effects match nicely with the lesion studies implicating the hippocampus in spatial learning (Morris et al., 1990; McDonald and White, 1993; Chang and Gold, 2003). Beyond merely indicating the relative degree of activity during learning, temporal differences in IEG
275
activation in the hippocampus could reflect unique processing of hippocampal place cells in response to different spatial contexts. It has been shown that changes in the number of Arc-positive cells after exposure to a spatial context corresponds roughly to the number of cells that develop place fields in novel environments (Chawla et al., 2005). Differences in the temporal activation of Arc can also be used to elucidate separate populations of hippocampal neurons that encode separate contexts. It has been shown that Arc is linked to some degree to the behavioral history of an animal in a nonlearning situation involving repeated exposures to a familiar testing environment over many sessions (Guzowski et al., 2006). Essentially, illumination of the pattern of Arc expression within hippocampus can assist in visualizing the distributed representation of a particular context. Guzowski et al. (2001) assessed the degree of correspondence between the activation of different IEGs, both regulatory and effector, following spatial and nonspatial training in the water maze. Arc,zif268, and c-fos mRNAs were all elevated in the hippocampus after training. The levels of the different IEGs did not vary between the spatially trained and the cued-trained animals. However, the best spatial learners with faster escape latencies had greater hippocampal Arc expression than that of response learners, mainly in the CA1 and CA3 regions. Consequently, it appears that the magnitude of an IEG response might signal within an individual brain region the learning proficiency of a particular animal. Other studies have demonstrated differential IEG activation in hippocampus resulting specifically from spatial, but not cued, training in the water maze (Teather et al., 2005). Both spatial and cued training induced increases in c-Fos in several areas of hippocampus that were above levels of caged controls. The most provocative increase occurred in CA1 where spatially trained and control animals, yoked to time spent swimming, displayed significantly greater c-Fos than both caged controls and cued-trained animals. Measurement of c-Jun provided clearer spatial-induced increases in CA1 and CA3 that were unrelated to swimming. There was no difference between spatially trained and cued-trained animals in amount of c-Fos or c-Jun in dorsal striatum. These data suggest that the different IEGs may have different thresholds for exposing strategy or learning specific responses in a given structure that can vary by the specific behavior or task. Hippocampal neurons are not alone in their ability to exhibit learning-induced changes in activation via IEG expression. Explicit response training on the Tmaze can selectively elicit a significant increase in c-Fos in dorsal striatum, but only in those animals that had performed correctly in a post-training probe trial
276
NEURAL SYSTEMS PERSPECTIVE ON PLACE FIELDS
Figure 17–1. Summary of c-Fos and Zif268 activation in dorsal striatum and hippocampus resulting from explicit place and response training on the T-maze. Bars represent standardized amounts (z-scores) of c-Fos expression averaged within hippocampus (HPC) and dorsomedial (DM) and dorsolateral (DL) striatum 30 min postcriterion for place- and response-trained animals and control animals. Responsetrained animals that performed correctly during the probe trial exhibited greater c-Fos immunoreactivity in DL and DM regions than did animals. Place-trained animals and response-trained animals that failed the probe trial did not differ from controls (* p < 0.05; þ p ¼ 0.06). Also provided are examples of c-Fos immunoreactivity in dorsomedial (top) and dorsolateral (bottom) striatum from individual response-trained/ failed-probe trial (left), response-trained/passed-probe trial (middle), and control animals (right).
and required significantly fewer trials to reach a behavioral criterion during learning (Fig. 17–1; Gill et al., 2007). Consequently, one potential conclusion from these results is that proficiency with the use of a particular strategy results from optimal dopamine input to key structures, and subsequently an increase in IEG expression. Given the different pattern of IEG activation across dissociable memory systems based on task demands, it is possible that there would also be distinct
changes in the spatial selectivity of neural representations in these regions.
EFFECTS OF DOPAMINE ON SPATIAL SELECTIVITY OF SINGLE-UNIT ACTIVITY After considering evidence provided by studies of synaptic plasticity, it is not unreasonable to question
MODULATION OF SPATIAL LEARNING BY DOPAMINE
whether electrophysiological recordings from place cells exhibit the same basic principles of dopaminergic modulation of excitatory inputs. Familiar contexts are represented by particular patterns of activation of hippocampal CA1 place cells, or the arrangement of place fields, resulting from combined input from entorhinal cortex and CA3 (see Chapter 2, this volume). When a change is perceived in a familiar context, a hippocampal signal to VTA could initiate an increase in phasic dopamine release. The increase in dopamine can cause alterations in CA1 responses to glutamatergic input from entorhinal cortex, CA3, or both. Previously, we proposed that hippocampus contributes to learning or memory retrieval by enabling a mechanism to detect changes in context (Mizumori et al., 1999, 2000a; Smith and Mizumori, 2006; Chapter 2, this volume). Such a function is consistent with the view that the hippocampus responds to novel contextual stimuli. Indeed, electrophysiological recordings from CA1 in rodents exposed to novel locations in a familiar environment (radial-arm maze) revealed fairly rapid responses (Frank et al., 2004). Spatial coherence was a measure used to indicate the reliability, or stability, of field location across the session. In addition to a greater number of place fields on the novel arm, there was also little coherence in the activity observed in the novel location after repeated, brief exposures, perhaps reflecting the encoding process of spatial novelty. Place fields developed rapidly on the first exposure to the novel location ( 1–2 min). The more time spent exploring the novel arm, the greater place field stability conferred (5–6 min). In addition, place fields developed a directional component. Thus, it seemed that once a threshold of exposure to the novel location was attained, there was greater place field stability. Eventually, once the novel arm had itself become familiar, the distribution and quality of place fields did not vary between other maze locations. Stability of place fields after multiple exposures to a novel environment might be considered to reflect the learned significance of that location. If place field ‘‘consolidation’’ requires a mechanism similar to that involved in protein synthesis–dependent L-LTP, it likely also depends on the combined activity of glutamatergic and dopaminergic systems, since removal of either one of these systems interferes with protein synthesis. It could be that the temporal requirement for place field stability relies on a dopamine signal, given the importance of dopamine in L-LTP. As indicated earlier, studies examining the maintenance of LTP have supported the suggestion of an interaction between glutamatergic and dopaminergic systems in plasticity mechanisms. If place cells are models of the generation of lasting spatial representations in memory then there is the possibility that
277
the cooperativity between these neurotransmitter systems also modulates hippocampal place fields. Consequently, updated ‘‘place’’ information may be conveyed to hippocampus via glutamate, and the persistence of place fields is determined by dopamine. Introduction to novel environments has been shown to result in the development of place fields in an NMDA receptor–dependent manner (Kentros et al., 1998). If place field stabilization in a novel environment is reflective of a consolidation process, then it is also likely dependent upon new protein synthesis. The mechanisms for protein synthesis guided by dopamineactivated second messenger systems have already been summarized in this chapter. When mice were injected with anisomycin, a potent inhibitor of protein synthesis, after exposure to a novel environment, place cells were stable briefly (1 hr) but not at longer time points (6 and 24 hr; Agnihotri et al., 2004). In animals that had received the anisomycin treatment, place fields appeared to remap again on the second day in the novel environment as if the rat had never explored it, while the exact same cells maintained stable spatial representations in the familiar environment. The preliminary stability of place fields would indicate that the initiation of place field formation does not rely on new proteins. Interestingly, re-exposure to an already familiar environment also did not require protein synthesis. The result appears at odds with studies showing that the re-activation of a contextual fear memory does require new proteins (Nader et al., 2000; Debiec and Ledoux, 2004). It may be, then, that when a given context is not paired with a salient event such as shock, the representation of that context is stable, since there is no mismatch based on an expectation from previous experience. Dopamine may compute the predictability of aversive stimuli in a similar manner as rewarding stimuli. Therefore, presentation of the conditioning context in the absence of shock may represent a situation in which a fully predicted event is omitted, resulting in an inhibition in dopamine burst activity. Consequently, changes in a familiar environment sufficient to result in mismatch detection, or prediction error, may require new protein synthesis prompted by dopamine. Varying familiarity of environmental stimuli or the behavioral requirements in a familiar environment can also modulate place field stability (Kentros et al., 2004). When mice were required to learn a hippocampaldependent response in a familiar spatial context, there was an increase in the longevity of newly formed place fields. Without the introduction of behavioral significance, and in contrast to rat place fields, the place fields recorded in mouse hippocampus could sometimes remap freely after multiple exposures to the same environment. There was also a correlation
278
NEURAL SYSTEMS PERSPECTIVE ON PLACE FIELDS
between task proficiency and place field stability, suggesting that persistent hippocampal spatial representations assisted in the learning of hippocampaldependent behaviors. In order to establish a role for dopamine in the development of hippocampal place field stability, a dopamine agonist and antagonist treatment was applied in an environment in which no specific behavioral response was required. Typically, this scenario yielded unstable place fields. D1-agonist treatment was able to promote field stability, whereas a D1 antagonist destabilized fields even further. In addition to conveying or reinforcing novel information, the dopamine signal may also be crucial for detecting alterations in familiar contexts (Mizumori et al., 2004). Such a mechanism could signal changes in task demands during learning or indicate necessary switches in behavior or cognitive strategy to maintain accurate performance (Smith and Mizumori, 2006, Eschenko and Mizumori, 2007). When an animal first learns to discriminate between two contexts in the same environment, initially stable place fields can become destabilized or reorganize (Smith and Mizumori, 2006). In this case, the change in context was signaled by altering the goal location in a familiar environment without any other change in the testing environment. The orthogonalization of place fields as animals learn to effectively discriminate between the two contexts is thought to be mediated by dopamine. Different from the scenario in which a single environmental variable was altered and strategy remained the same, when animals were trained to alternate between using a spatial or response strategy on the T-maze, there were interesting changes in the spatial properties of place cells in both hippocampus and striatum (Eschenko and Mizumori, 2007). As is commonly observed with place cells, not all neurons responded to the cognitive switch between spatial and response strategies. However, for a majority of cells, dopamine could be facilitating the change in spatial firing patterns resulting from the change in task demands. A detailed description of the potential neural pathways involved will be summarized in subsequent sections on computational models of dopamine action in the hippocampus. Importantly, switching between cognitive strategies causes changes in more than one measure of spatial selectivity of hippocampal and striatal place cells (Eschenko and Mizumori, 2007). Following the switch, alterations in field stability, field reliability, infield specificity, or location specificity ensued. In other words, ‘‘remapping’’ responses are complex. It remains possible that a neuromodulator such as dopamine could regulate one of these measures of spatial selectivity, but not all. Interestingly, certain firing properties appear to be linked in their response to the cog-
nitive change. More reliable fields had consistent field locations when animals switched between spatial and response strategies while field specificity ensured location stability when a spatial strategy was still required to locate a different goal location. Like the hippocampus, similar relationships among spatial parameters were observed in the striatum. However, unlike the hippocampus, the changes observed in neural activity could be correlated with behavior. More accurate performance following a cognitive switch was associated with striatal place field stability. We can interpret this pattern of effects as different roles for dopamine processing in hippocampus and striatum. Consequently, when different memory systems are engaged, there are broad and simultaneous changes in neural activity across different brain regions. Place cells resistant to the change in response and/or to changes in behavioral context may be maintained by intrahippocampal circuitry that may be independent of any changes in the dopamine signal. Dopamine, acting via D1 receptors, may act to gate contextual changes in hippocampal and dorsal striatal neural activity (Gill and Mizumori, 2006). Both hippocampal and striatal place cells responded to changes in the visual testing environment by exhibiting alterations in place field locations while animals performed a spatial working memory task. Typically, these reorganization effects did not coincide with significant changes in field reliability or specificity. In other words, the context-induced generation of new field locations was relatively stable in terms of field reliability. However, when injection of a D1 antagonist, SCH23390, was combined with a change in the visual environment, significant alterations in hippocampal place field reliability, specificity, and reorganization occurred (Fig. 17–2). Consequently, while hippocampal and striatal place cell responses appeared similar in response to the contextual manipulation based on changes in spatial correlation values, or the degree of reorganization, hippocampal place cells exhibited unique responses to the combined D1-antagonist treatment and context manipulation. Dopamine’s facilitation of place field reorganization in response to contextual changes may also be important for maintaining behavioral flexibility in response to these changes. Early exposure to darkness within the testing environment caused significant increases in working memory errors. There was some indication that after repeated exposure to the context manipulation, some animals no longer exhibited significant disturbances in working memory performance. Figure 17–3 illustrates a hippocampal neuron recorded late in training during two separate sessions. The plots on the left side represent the place field activity during performance of five baseline working
Figure 17–2. Summary of the effects of darkness or D1 antagonism on spatial correlation values, specificity, and reliability of hippocampal and dorsal striatal place fields. Difference scores for place field reliability and specificity were calculated using the absolute value of the change across baseline and manipulation phases of testing. Spatial correlation scores are based on a pixel-by-pixel correlation analysis across training blocks. A. In hippocampus and striatum, darkness caused reorganization (i.e., low correlation) of place fields. Darkness did not cause significant changes in the place field reliability or specificity in either the striatum or hippocampus. B. The combination of D1 antagonism and darkness caused significant reorganization of both striatal and hippocampal place fields. In addition, the combination of SCH23390 and darkness produced the greatest disruption in hippocampal place-field reliability and specificity while the same measures for striatal place fields did not change. * p < 0.05.
280
NEURAL SYSTEMS PERSPECTIVE ON PLACE FIELDS
memory trials during which no manipulations were conducted. The plots on the right side represent the changes in place field activity that occurred following either darkness or the combination of darkness and D1-antagonist treatment. The typical place cell response to darkness was a shifting of field location (Fig. 17–3A). Interestingly, during this particular session, there was noteworthy place cell reorganization without a significant behavioral impairment resulting from darkness. The reorganization effects on place field activity may coincide with activation of different neural systems or the selection of a different cognitive strategy in response to the darkness. The administration of a D1 antagonist blocked this effect on place field reorganization (Fig. 17–3B). Perhaps it was no coincidence that there was also an increase in working memory errors. It has already been discussed how
significant changes in place field locations may be important for distinguishing between two different contexts (Smith and Mizumori, 2006). Therefore, systematic changes in hippocampal spatial representations, promoted by D1-receptor activity, may be important for adaptive behavioral responses to contextual changes.
COMPUTATIONAL MODELS THAT ACCOUNT FOR DOPAMINE’S INFLUENCE ON HIPPOCAMPAL MEDIATED LEARNING AND MEMORY Several noteworthy models account for the regulation by dopamine of the processing and storage of information in the hippocampus. As a neuromodulatory in-
Figure 17–3. Color spatial-density plots illustrating the effects of darkness alone or in combination with D1 antagonism on HPC place fields recorded from the same neuron as the animal performed the spatial working memory task. For each day, colors represent areas associated with the maximum firing (shown in red), as well as proportions of the maximum firing in 25% increments (from blue to red). A. Hippocampal place fields reorganized following the presentation of darkness and performance was not greatly impaired during this session. B. When combined with D1 antagonism, the same hippocampal place field reorganized in a different pattern than that from darkness alone and working memory performance was significantly impaired.
MODULATION OF SPATIAL LEARNING BY DOPAMINE
fluence on hippocampal pyramidal cell activity, dopamine could permit place cell remapping in a way that is necessary to maintain behavior following contextual changes. It has been postulated that dopamine can selectively alter some inputs to CA1 while leaving others unaffected (Lisman and Otmakhova, 2001). According to this model, CA1 is proposed to have two potential functions: (1) identification of novel stimuli and (2) alteration of signals provided by CA3 to be conveyed to cortex. Entorhinal cortex (perforant path) and CA3 (Schaffer collateral) afferents provide the main input to CA1. Dopamine can shift responses of CA1 neurons from entorhinal cortical input to the autoassociative and heteroassociative hippocampal networks represented by connectivity within dentate gyrus and CA3. Evidence for this shift is that dopamine application significantly reduces perforant path, but not Schaffer collateral, responses in CA1 (Otmakhova and Lisman, 1999, 2000). It has also been demonstrated that CA1 place cell activity remains stable after reversible blockade of CA3 input, which suggests that the spatial representations are maintained by cortical input alone (Mizumori et al., 1989b). In addition, the incorporation of new information into the autoassociative networks in dentate gyrus and CA3 is achieved by subthreshold perforant path input from the entorhinal cortex. This would ‘‘enable’’ only a select population of cells, thereby priming them for subsequent consolidation of new information. This view of dopamine function corresponds with the description of dopamine action as a selective filter. Consequently, dopamine may act to facilitate learning by enhancing signals processed through the dentate–CA3 buffer and decreasing potentially irrelevant sensory input from cortex. The hippocampus may initiate a novelty circuit that ultimately contributes to the consolidation of new, salient information (Lisman and Grace, 2005). The hippocampus can perform match–mismatch comparisons through rapid neural responses to differences in expected and unexpected conditioned stimuli (Brankack et al., 1996; Grunwald et al., 1998; Mizumori et al., 1999, 2000a, 2004; Vinogradova, 2001). As indicated previously, CA1 neurons compute novelty by comparing sensory information supplied by perforant path input with predictions of the current context made by CA3. This novelty signal may then be sent on to other regions, such as nucleus accumbens and prefrontal cortex, which in turn modulate VTA activity to affect widespread dopamine release. Exposing animals to a novel cage increases dopamine release (Ihalainen, 1999). Novelty-induced release of dopamine in the nucleus accumbens is dependent upon output from the hippocampus transmitted via the ventral subiculum (Legault and Wise, 2001). In addition,
281
stimulating the subiculum or artificially increasing hippocampal output increases dopamine release from the VTA, based on the number of active neurons (Floresco et al., 2001b, 2003). The nucleus accumbens, via its inhibition of the ventral pallidum and subsequent disinhibition of the VTA, seems especially important in transmitting the novelty signal detected in the hippocampus to VTA (Legault et al., 2000; Floresco et al., 2001a,b). When the ventral pallidum is inhibited by a GABA A/B agonist, there is also an increase in the number of active dopamine neurons (Floresco et al., 2003). It has been suggested that the hippocampus– prefrontal cortex circuit conveys stimulus saliency to the VTA via the pedunculopontine nucleus. The effect of this circuitry is an increased dopamine efflux from the VTA creating a feedback loop to hippocampus. Ultimately, hippocampal–VTA loops may act to ensure that only relevant information is encoded by CA1 neurons. The increased dopamine release acts as a filter within CA1 to selectively enhance responses to contextually relevant, perhaps novel, stimuli. The automatic representations in the hippocampus can be rapid and transient as part of the encoding of attended experience (Morris 2001; Morris et al., 2003; reviewed in Morris, 2006). Without a continuing signal, hippocampal traces would decay rapidly and not last past an initial encoding phase. Dopamine may act to promote cellular consolidation in recently activated synapses, as proposed by the synaptic tagging hypothesis (Frey and Morris, 1998). Consistent with this hypothesis, it has recently been demonstrated that D1-receptor activation within the hippocampus is necessary during the encoding phase of a one-trial hippocampal-dependent task (O’Carroll et al., 2006). If dopamine signaling is important for establishing hippocampal memory traces or incorporating novel but relevant information into existing schemas, a similar dopamine mechanism may be supported in other brain areas such as dorsal striatum. Common among models of dopamine function in dorsal striatum is the reinforcement of appropriate behavioral responses in a particular context. The actor–critic model describes how different brain regions interact to result in flexible behavior. Typically, one region acts as the ‘‘critic’’ to shape neural responses in the ‘‘actor’’ region. Ultimately, the output signal of the actor is what initiates changes in behavioral responses. The actor–critic model is complementary to the TD model in that the TD algorithm is used by the critic to ‘‘teach’’ the actor sensorimotor associations. The actor–critic model assigns a role for dopamine, transmitted from the pars compacta region of substantia nigra, in shaping dorsal striatal responses. The coordinated activity of the indirect and direct pathways originating from dorsal striatum is believed important for flexible
282
NEURAL SYSTEMS PERSPECTIVE ON PLACE FIELDS
behavioral responses. Participating as the critic, dopamine input can suppress inappropriate responses while exciting pathways responsible for response generation.
BEYOND DOPAMINE: THE INFLUENCE OF CHOLINERGIC ACTIVITY ON REGULATING SPATIAL BEHAVIORS AND PROCESSING The deficits observed in PD patients can extend beyond dopamine involvement. Septohippocampal fibers are also reduced by 60%–80% and lesions of tegmental and laterodorsal pedunculopontine areas have been reported. Alteration of other neuromodulatory systems in hippocampus, such as acetylcholine (ACh), can result in the pattern of visuospatial deficits of PD patients. Therefore, it is important to take into consideration these other neuromodulatory systems when discussing hippocampal place cell activity. The final section of this chapter focuses on the cholinergic influence on hippocampal- and dorsal striatal–dependent learning, and the implications for place cell activity. Cholinergic activity in hippocampus may regulate the gating of cortical and intrahippocampal inputs to CA1 and, thus, place cell codes. The assumption is that with modification in place cell codes, there can be a complementary change in spatial learning. There is some debate regarding the requirement of ACh in spatial learning. While blockade of ACh transmission within ventral hippocampus interferes with spatial working memory performance, ACh agonists have no effect (Kim and Levin, 1996). Transient changes in ACh levels during learning can be used to predict the strategy employed during various maze tasks. As animals engaged in hippocampal-dependent behaviors, there were observable increases in ACh release in the hippocampus (Fadda et al., 1996; Ragozzino et al., 1996, 1998). With continued testing on a standard Tmaze task, animals switched from relying on a spatial strategy to a response strategy (Chang and Gold, 2003). This alternation between two proposed independent memory systems was also correlated with changes in hippocampal and dorsal striatal ACh levels. While hippocampal ACh levels remained elevated, there was a gradual increase in dorsal striatal ACh. When animals were explicitly trained to perform a place or response task, a similar pattern was observed of sustained elevation in hippocampal ACh levels during both tasks and significantly greater dorsal striatal ACh levels during response testing only (Pych et al., 2005). Despite the compelling evidence that increases in ACh output correlate with different stages of learning, nearly complete removal of septohippocampal ACh input often leaves spatial learning intact (Baxter et al.,
1995; Cahill and Baxter, 2001). Therefore, it is likely that cholinergic activity can act to promote or modulate hippocampal plasticity without being required to mediate hippocampal-dependent behaviors. One proposed function of acetylcholine in the hippocampus is mediation of encoding of new information without interference of previously stored input (Hasselmo and Wyble, 1997). Most models of cholinergic regulation of hippocampal neural activity describe its importance in theta-oscillation modulation. For example, theta modulation may be vital in separating encoding and retrieval processes in hippocampus (Hasselmo, 2005). By alternating these memory processes, only relevant information is incorporated into existing hippocampal representations while minimizing interference from potential irrelevant inputs. There have also been studies exploring how ACh input impacts place cell activity. Reversible inactivation of medial septum caused significant reorganization in hippocampal place fields corresponding to impaired spatial memory (Mizumori et al., 1989a). Electrophysiological recordings from animals with permanent lesions of the septohippocampal cholinergic system revealed interference in reorganization of place cells in a new environment as animals performed a well-learned spatial task (Leutgeb and Mizumori, 1999). A similar study explored the effects of the loss of cholinergic input on spatial selectivity of neurons as animals foraged in either a familiar or novel arena (Ikonen et al., 2002). As animals explored the familiar environment, there was no effect of lesioning on common measures of place cell activity: in-field rate, number of place fields, place field area, or spatial selectivity. Place cells recorded in control animals developed new field locations in the novel environment, while the place cells of lesioned animals failed to reorganize. Indeed, the overall place cell response in lesioned animals appeared to be greater stability, or maintenance, of original field configuration with successive exposures to the new environment. Similar disruptions in place field properties in familiar arenas were seen following intraventricular infusion of ACh muscarinic antagonists (Brazhnik et al., 2003, 2004). In general, the place fields exhibited weak and disorganized spatial activity after scopolamine treatment, as illustrated by a decrease in spatial coherence values. Cholinergic agonists such as carbachol are able to differentially suppress synaptic transmission in CA1 such that inputs arising from autoassociative networks represented by dentate gyrus–CA3 are suppressed to a greater extent than inputs arriving from entorhinal cortex (Hasselmo and Schnell, 1994; Hasselmo et al., 1995; Hasselmo and Wyble, 1997). Thus, like dopamine, ACh is able to shift CA1 pyramidal cell responses, but in the opposing direction so that perforant
MODULATION OF SPATIAL LEARNING BY DOPAMINE
283
Figure 17–4. Proposed circuit representing updating of spatial-context network via dopamine. Changes in hippocampal processing in CA1, prompted by entorhinal cortical or CA3 input, can act to indirectly alter the activity of dopamine neurons in the ventral tegmental area (VTA). Subsequent projections from dopamine neurons back to the hippocampus can selectively enhance or attenuate certain inputs, either cortical or intrahippocampal, and modify the strength of the responses of hippocampal place cells in response to context changes. Nucleus accumbens, NAc; pedunculopontine tegmental nucleus, PPtg; ventral pallidum, VP. path inputs have greater influence than associational Schaffer collateral inputs. In this case, CA1 responses rely more on new afferent input patterns instead of on previously stored information. This situation should be characterized by less stable place fields. The combined activity of dopamine and acetylcholine modulatory systems may act to ensure consistent behavioral responses following changes in context. Flexible behavioral responses are supported by the actions of dopamine and acetylcholine following the successful updating of context representations in the hippocampus, as measured by systematic changes in place cell activity.
CONCLUSIONS The general conclusion could be made that by selectively filtering multiple inputs, dopamine acts to increase the efficiency of processing within a given brain region. More intriguing is the possibility that
dopamine orchestrates the relative efficiency of neural processing in different neural systems. In this way, dopamine serves a key role in the implementation of, and coordination within, fundamentally important functional domains (Mizumori et al., 2004). An example of one such domain is the distributed spatial-context network that provides a framework onto which new associations, sensory elements, and behavioral responses can be coordinated. As learning progresses or changes occur in a new context, the entire spatialcontext network must be updated. Presumably dopamine contributes to this updating process. Although unique functionality of different brain areas emerges in part from their distinct intrinsic local networks operation, detection of a mismatch between current and expected (context) information is a fundamental functional consequence for many brain systems. Consequently, while similar types of information are being processed across the global spatial-context network, there is disparity in the ways in which each region uses this information to represent the current context (Fig. 17–4).
284
NEURAL SYSTEMS PERSPECTIVE ON PLACE FIELDS
No matter how one defines context, whether it includes only the particular constellation of environmental stimuli or in addition the internal representations shaped by motivation and expectation that direct an animal’s behavior, the process of selectively filtering and enhancing only relevant inputs for a given context is an important process for maintaining flexible behavior. The action by dopamine in facilitating this aspect of hippocampal function might be separated into two potential mechanisms. First, dopamine release may contribute to the stabilization of spatial representations in novel environments via its interaction with glutamatergic and cholinergic systems. Second, according to the context discrimination hypothesis (Chapter 2, this volume), systematic changes in hippocampal neural activity prompted by dopamine may be vital in supporting the orthogonalization of spatial representations. A significant implication of these proposed features of dopamine activity is that the function of dopamine extends beyond simple reward processing. Rather, dopamine appears to detect changes in the consequences of behavioral acts more generally. When doing so, dopamine functions to guide novelty detection mechanisms across a distributed neural system (e.g., Fig. 17–4).
acknowledgments This work was supported by NIMH grant 58755 to S.J.Y.M. We also thank Emily Wood and Ilene Bernstein for their valuable contribution to the studies described in this chapter.
References Abrahams S, Pickering A, Polkey CE, Morris RG (1997) Spatial memory deficits in patients with unilateral damage to the right hippocampal formation. Neuropsychologia 35:11–24. Agnihotri NT, Hawkins RD, Kandel ER, Kentros C (2004) The long-term stability of new hippocampal place fields requires new protein synthesis. Proc Natl Acad Sci USA 101:3656–3661. Andrzejewski ME, Spencer RC, Kelley AE (2006) Dissociating ventral and dorsal subicular dopamine Dsub-1 receptor involvement in instrumental learning, spontaneous motor behavior, and motivation. Behav Neurosci 120:542–553. Bach ME, Barad M, Son H, Zhuo M, Lu Y, Shih R, Mansuy I, Hawkins RD, Kandel ER (1999) Agerelated defects in spatial memory are correlated with defects in the late phase of hippocampal long-term potentiation in vitro and are attenuated by drugs that enhance the cAMP signaling pathway. Proc Natl Acad Sci USA 96:5280–5285.
Baxter MG, Bucci DJ, Gorman LK, Wiley RG, Gallagher M (1995) Selective immunotoxic lesions of basal forebrain cholinergic cells: effects on learning and memory in rats. Behav Neurosci 109:714–722. Beck CH, Fibiger HC (1995) Conditioned fear-induced changes in behavior and in the expression of the immediate early gene c-fos: with and without diazepam pretreatment. J Neurosci 15:709–720. Behr J, Gloveli T, Schmitz D, Heinemann U (2000) Dopamine depresses excitatory synaptic transmission onto rat subicular neurons via presynaptic D1like dopamine receptors. J Neurophysiol 84:112– 119. Bohbot VD, Jech R, Ruzicka E, Nadel L, Kalina M, Stepankova K, Bures J (2002) Rat spatial memory tasks adapted for humans: characterization in subjects with intact brain and subjects with selective medial temporal lobe thermal lesions. Physiol Res 51 Suppl 1:S49–S65. Brankack J, Seidenbecher T, Muller-Gartner HW (1996) Task-relevant late positive component in rats: is it related to hippocampal theta rhythm? Hippocampus 6:475–482. Brazhnik E, Borgnis R, Muller RU, Fox SE (2004) The effects on place cells of local scopolamine dialysis are mimicked by a mixture of two specific muscarinic antagonists. J Neurosci 24:9313–9323. Brazhnik ES, Muller RU, Fox SE (2003) Muscarinic blockade slows and degrades the location-specific firing of hippocampal pyramidal cells. J Neurosci 23: 611–621. Broadbent NJ, Squire LR, Clark RE (2006) Reversible hippocampal lesions disrupt water maze performance during both recent and remote memory tests. Learn Mem 13:187–191. Cahill JF, Baxter MG (2001) Cholinergic and noncholinergic septal neurons modulate strategy selection in spatial learning. Eur J Neurosci 14:1856–1864. Campeau S, Hayward MD, Hope BT, Rosen JB, Nestler EJ, Davis M (1991) Induction of the c-fos proto-oncogene in rat amygdala during unconditioned and conditioned fear. Brain Res 565:349–352. Chang Q, Gold PE (2003) Switching memory systems during learning: changes in patterns of brain acetylcholine release in the hippocampus and striatum in rats. J Neurosci 23:3001–3005. Chawla MK, Guzowski JF, Ramirez-Amaya V, Lipa P, Hoffman KL, Marriott LK, Worley PF, McNaughton BL, Barnes CA (2005) Sparse, environmentally selective expression of Arc RNA in the upper blade of the rodent fascia dentata by brief spatial experience. Hippocampus 15:579–586. Clayton DF (2000) The genomic action potential. Neurobiol Learn Mem 74:185–216.
MODULATION OF SPATIAL LEARNING BY DOPAMINE
Colombo PJ, Brightwell JJ, Countryman RA (2003) Cognitivestrategy-specific increasesinphosphorylated cAMP response element-binding protein and c-Fos in the hippocampus and dorsal striatum. J Neurosci 23: 3547–3554. Da Cunha C, Angelucci ME, Canteras NS, Wonnacott S, Takahashi RN (2002) The lesion of the rat substantia nigra pars compacta dopaminergic neurons as a model for Parkinson’s disease memory disabilities. Cell Mol Neurobiol 22:227–237. Debiec J, Ledoux JE (2004) Disruption of reconsolidation but not consolidation of auditory fear conditioning by noradrenergic blockade in the amygdala. Neuroscience 129:267–272. Dubois B, Pillon B (1997) Cognitive deficits in Parkinson’s disease. J Neurol 244:2–8. Dujardin K, Laurent B (2003) Dysfunction of the human memory systems: role of the dopaminergic transmission. Curr Opin Neurol Suppl 2:S11–6. Eschenko O, Mizumori SJ (2007) Memory influences on hippocampal and striatal neural codes: effects of a shift between task rules. Neurobiol Learn Mem 87:495–509. Fadda F, Melis F, Stancampiano R (1996) Increased hippocampal acetylcholine release during a working memory task. Eur J Pharmacol 307:R1–2. Floresco SB, Blaha CD, Yang CR, Phillips AG (2001a) Modulation of hippocampal and amygdalar-evoked activity of nucleus accumbens neurons by dopamine: cellular mechanisms of input selection. J Neurosci 21: 2851–2860. Floresco SB, Todd CL, Grace AA (2001b) Glutamatergic afferents from the hippocampus to the nucleus accumbens regulate activity of ventral tegmental area dopamine neurons. J Neurosci 21:4915–4922. Floresco SB, West AR, Ash B, Moore H, Grace AA (2003) Afferent modulation of dopamine neuron firing differentially regulates tonic and phasic dopamine transmission. Nat Neurosci 6:968–973. Flores-Hernandez J, Cepeda C, Hernandez-Echeagaray E, Calvert CR, Jokel ES, Fienberg AA, Greengard P, Levine MS (2002) Dopamine enhancement of NMDA currents in dissociated medium-sized striatal neurons: role of D1 receptors and DARPP-32. J Neurophysiol 88:3010–3020. Frank LM, Stanley GB, Brown EN (2004) Hippocampal plasticity across multiple days of exposure to novel environments. J Neurosci 24:7681–7689. Frey U, Matthies H, Reymann KG, Matthies H (1991) The effect of dopaminergic D1 receptor blockade during tetanization on the expression of long-term potentiation in the rat CA1 region in vitro. Neurosci Lett 129:111–114. Frey U, Morris RG (1997) Synaptic tagging and longterm potentiation. Nature 385:533–536.Frey U,
285
Morris RG (1998) Synaptic tagging: implications for late maintenance ofhippocampal long-term potentiation. Trends Neurosci 21:181–188. Frey U, Schroeder H, Matthies H (1990) Dopaminergic antagonists prevent long-term maintenance of posttetanic LTP in the CA1 region of rat hippocampal slices. Brain Res 522:69–75. Gasbarri A, Packard MG, Campana E, Pacitti C (1994a) Anterograde and retrograde tracing of projections from the ventral tegmental area to the hippocampal formation in the rat. Brain Res Bull 33:445–452. Gasbarri A, Sulli A, Innocenzi R, Pacitti C, Brioni JD (1996) Spatial memory impairment induced by lesion of the mesohippocampal dopaminergic system in the rat. Neuroscience 74:1037–1044. Gasbarri A, Sulli A, Packard MG (1997) The dopaminergic mesencephalic projections to the hippocampal formation in the rat. Prog Neuropsychopharmacol BiolPsychiatry 21:1–22. Gasbarri A, Verney C, Innocenzi R, Campana E, Pacitti C (1994b) Mesolimbic dopaminergic neurons innervating the hippocampal formation in the rat: a combined retrograde tracing and immunohistochemical study. Brain Res 668:71–79. Gevaerd MS, Takahashi RN, Silveira R, Da Cunha C (2001) Caffeine reverses the memory disruption induced by intra-nigral MPTP-injection in rats. Brain Res Bull 55:101–106. Gill KM, Bernstein IL, Mizumori SJ (2007) Immediate early gene activation in hippocampus and dorsal striatum: effects of explicit place and response training. Neurobiol Learn Mem 87:583–596. Gill KM, Mizumori SJ (2006) Context-dependent modulation by D(1) receptors: differential effects in hippocampus and striatum. Behav Neurosci 120:377– 392. Goldstein LH, Canavan AG, Polkey CE (1989) Cognitive mapping after unilateral temporal lobectomy. Neuropsychologia 27:167–177. Grunwald T, Lehnertz K, Heinze HJ, Helmstaedter C, Elger CE (1998) Verbal novelty detection within the human hippocampus proper. Proc Natl Acad Sci USA 95:3193–3197. Guzowski JF, Miyashita T, Chawla MK, Sanderson J, Maes LI, Houston FP, Lipa P, McNaughton BL, Worley PF, Barnes CA (2006) Recent behavioral history modifies coupling between cell activity and Arc gene transcription in hippocampal CA1 neurons. Proc Natl Acad Sci USA 103:1077–1082. Guzowski JF, Setlow B, Wagner EK, McGaugh JL (2001) Experience-dependent gene expression in the rat hippocampus after spatial learning: a comparison of the immediate-early genes Arc, c-fos, and zif268. J Neurosci 21:5089–5098.
286
NEURAL SYSTEMS PERSPECTIVE ON PLACE FIELDS
Guzowski JF, Timlin JA, Roysam B, McNaughton BL, Worley PF, Barnes CA (2005) Mapping behaviorally relevant neural circuits with immediate-early gene expression. Curr Opin Neurobiol 15:599–606. Hasselmo ME (2005) What is the function of hippocampal theta rhythm?—Linking behavioral data to phasic properties of field potential and unit recording data. Hippocampus 15:936–949. Hasselmo ME, Schnell E (1994) Laminar selectivity of the cholinergic suppression of synaptic transmission in rat hippocampal region CA1: computational modeling and brain slice physiology. J Neurosci 14:3898– 3914. Hasselmo ME, Schnell E, Barkai E (1995) Dynamics of learning and recall at excitatory recurrent synapses and cholinergic modulation in rat hippocampal region CA3. J Neurosci 15:5249–5262. Hasselmo ME, Wyble BP (1997) Free recall and recognition in a network model of the hippocampus: simulating effects of scopolamine on human memory function. Behav Brain Res 89:1–34. Hollerman JR, Schultz W (1998) Dopamine neurons report an error in the temporal prediction of reward during learning. Nat Neurosci 1:304–309. Hsu KS (1996) Characterization of dopamine receptors mediating inhibition of excitatory synaptic transmission in the rat hippocampal slice. J Neurophysiol 76:1887–1895. Huang Q, Zhou D, Chase K, Gusella JF, Aronin N, DiFiglia M (1992) Immunohistochemical localization of the D1 dopamine receptor in rat brain reveals its axonal transport, pre- and postsynaptic localization, and prevalence in the basal ganglia, limbic system, and thalamic reticular nucleus. Proc Natl Acad Sci USA 89:11988–11992. Huang Y, Kandel ER (1995) D1/D5 receptor agonists induce a protein synthesis-dependent late potentiation in the CA1 region of the hippocampus. Proc Natl Acad Sci USA 92:2446–2450. Ihalainen JA, Riekkinen P Jr, Feenstra MG (1999) Comparison of dopamine and noradrenaline release in mouse prefrontal cortex, striatum and hippocampus using microdialysis. Neurosci Lett 277:71–74. Ikonen S, McMahan R, Gallagher M, Eichenbaum H, Tanila H (2002) Cholinergic system regulation of spatial representation by the hippocampus. Hippocampus 12:386–397. Jarrard LE (1993) On the role of the hippocampus in learning and memory in the rat. Behav Neural Biol 60: 9–26. Kang DK, Kim KO, Lee SH, Lee YS, Son H (2000) c-Fos expression by dopaminergic receptor activation in rat hippocampal neurons. Mol Cells 10:546–551. Kentros CG, Agnihotri NT, Streater S, Hawkins RD, Kandel ER (2004) Increased attention to spatial con-
text increases both place field stability and spatial memory. Neuron 42:283–295. Kentros C, Hargreaves E, Hawkins RD, Kandel ER, Shapiro M, Muller RV (1998) Abolition of long-term stability of new hippocampal place cell maps by NMDA receptor blockade. Science 280:2121–2126. Kim JS, Levin ED (1996) Nicotinic, muscarinic and dopaminergic actions in the ventral hippocampus and the nucleus accumbens: effects on spatial working memory in rats. Brain Res 725:231–240. Kusuki T, Imahori Y, Ueda S, Inokuchi K (1997) Dopaminergic modulation of LTP induction in the dentate gyrus of intact brain. Neuroreport 8:2037–2040. Lanahan A, Worley P (1998) Immediate-early genes and synaptic function. Neurobiol Learn Mem 70:37–43. Lavenex PB, Amaral DG, Lavenex P (2006) Hippocampal lesion prevents spatial relational learning in adult macaque monkeys. J Neurosci 26:4546–4558. Legault M, Rompre PP, Wise RA (2000) Chemical stimulation of the ventral hippocampus elevates nucleus accumbens dopamine by activating dopaminergic neurons of the ventral tegmental area. J Neurosci 20: 1635–1642. Legault M, Wise RA (2001) Novelty-evoked elevations of nucleus accumbens dopamine: dependence on impulse flow from the ventral subiculum and glutamatergic neurotransmission in the ventral tegmental area. Eur J Neurosci 13:819–828. Lemon N, Manahan-Vaughan D (2006) Dopamine D1/ D5 receptors gate the acquisition of novel information through hippocampal long-term potentiation and long-term depression. J Neurosci 26:7723–7729. Leutgeb S, Mizumori SJ (1999) Excitotoxic septal lesions result in spatial memory deficits and altered flexibility of hippocampal single-unit representations. J Neurosci 19:6661–6672. Li S, Cullen WK, Anwyl R, Rowan MJ (2003) Dopamine-dependent facilitation of LTP induction in hippocampal CA1 by exposure to spatial novelty. Nat Neurosci 6:526–531. Lin JY, Dubey R, Funk GD, Lipski J (2003) Receptor subtype-specific modulation by dopamine of glutamatergic responses in striatal medium spiny neurons. Brain Res 959:251–262. Lisman JE, Grace AA (2005) The hippocampal-VTA loop: controlling the entry of information into longterm memory. Neuron 46:703–713. Lisman JE, Otmakhova NA (2001) Storage, recall, and novelty detection of sequences by the hippocampus: elaborating on the SOCRATIC model to account for normal and aberrant effects of dopamine. Hippocampus 11:551–568. Maguire EA, Nannery R, Spiers HJ (2006) Navigation around London by a taxi driver with bilateral hippocampal lesions. Brain 129:2894–2907.
MODULATION OF SPATIAL LEARNING BY DOPAMINE
Matthies H, Becker A, Schroeder H, Kraus J, Hollt V, Krug M (1997) Dopamine D1-deficient mutant mice do not express the late phase of hippocampal longterm potentiation. Neuroreport 8:3533–3535. McDonald RJ, White NM (1993) A triple dissociation of memory systems: hippocampus, amygdala, and dorsal striatum. Behav Neurosci 107:3–22. Mirenowicz J, Schultz W (1994) Importance of unpredictability for reward responses in primate dopamine neurons. J Neurophysiol 72:1024–1027. Miyoshi E, Wietzikoski S, Camplessei M, Silveira R, Takahashi RN, Da Cunha C (2002) Impaired learning in a spatial working memory version and in a cued version of the water maze in rats with MPTP-induced mesencephalic dopaminergic lesions. Brain Res Bull 58:41–47. Mizumori SJ, Barnes CA, McNaughton BL (1989a) Reversible inactivation of the medial septum: selective effects on the spontaneous unit activity of different hippocampal cell types. Brain Res 500:99– 106. Mizumori SJ, Cooper BG, Leutgeb S, Pratt WE (2000a) A neural systems analysis of adaptive navigation. Mol Neurobiol 21:57–82. Mizumori SJ, McNaughton BL, Barnes CA, Fox KB (1989b) Preserved spatial coding in hippocampal CA1 pyramidal cells during reversible suppression of CA3 output: evidence for pattern completion in hippocampus. J Neurosci 9:3915–3928. Mizumori SJ, Ragozzino KE, Cooper BG (2000b) Location and head direction representations in the dorsal striatum of rats. Psychobiology 28:441–462. Mizumori SJY, Ragozzino KE, Cooper BG, Leutgeb S (1999) Hippocampal representational organization and spatial context. Hippocampus 9:444–451. Mizumori SJ, Yeshenko O, Gill KM, Davis DM (2004) Parallel processing across neural systems: implications for a multiple memory system hypothesis. Neurobiol Learn Mem 82:278–298. Morris RGM (2001) Episodic-like memory in animals: psychologicalcriteria,neuralmechanismsandthevalue of episodic-like tasks to investigate animal models of neurodegenerative disease. Philos Trans R Soc Lond B Biol Sci 356:1453–1465. Morris RGM (2006) Elements of a neurobiological theory of hippocampal function: the role of synaptic plasticity, synaptic tagging and schemas. Eur J Neurosci 23:2829–2846. Morris RG, Moser EI, Riedel G, Martin SJ, Sandin J, Day M, O’Carroll C (2003) Elements of a neurobiological theory of the hippocampus: the role of activitydependent synaptic plasticity in memory. Philos Trans R Soc Lond B Biol Sci 358:773–786. Morris RG, Schenk F, Tweedie F, Jarrard LE (1990) Ibotenate lesions of hippocampus and/or subiculum:
287
dissociating components of allocentric spatial learning. Eur J Neurosci 2:1016–1028. Nader K, Schafe GE, Le Doux JE (2000) Fear memories require protein synthesis in the amygdala for reconsolidation after retrieval. Nature 406:722–726. O’Carroll CM, Martin SJ, Sandin J, Frenguelli B, Morris RG (2006) Dopaminergic modulation of the persistence of one-trial hippocampus-dependent memory. Learn Mem 13:760–769. O’Carroll CM, Morris RG (2004) Heterosynaptic co-activation of glutamatergic and dopaminergic afferents is required to induce persistent long-term potentiation. Neuropharmacology 47:324–332. O’Keefe J (1976) Place units in the hippocampus of the freely moving rat. Exp Neurol 51:78–109. O’Keefe J, Conway DH (1978) Hippocampal place units in the freely moving rat: why they fire where they fire. Exp Brain Res 31:573–590. O’Keefe J, Dostrovsky J (1971) The hippocampus as a spatial map. Preliminary evidence from unit activity in the freely moving rat. Brain Res 34:171–175. Otmakhova NA, Lisman JE (1999) Dopamine selectively inhibits the direct cortical pathway to the CA1 hippocampal region. J Neurosci 19:1437–1445. Otmakhova NA, Lisman JE (2000) Dopamine, serotonin, and noradrenaline strongly inhibit the direct perforant path-CA1 synaptic input, but have little effect on the Schaffer collateral input. Ann N Y Acad Sci 911:462– 464. Owen AM, Iddon JL, Hodges JR, Summers BA, Robbins TW (1997) Spatial and non-spatial working memory at different stages of Parkinson’s disease. Neuropsychologia 35:519–532. Packard MG, White NM (1991) Dissociation of hippocampus and caudate nucleus memory systems by post-training intracerebral injection of dopamine agonists. Behav Neurosci 105:295–306. Pillon B, Deweer B, Vidailhet M, Bonnet AM, HahnBarma V, Dubois B (1998) Is impaired memory for spatial location in Parkinson’s disease domain specific or dependent on ‘strategic’ processes? Neuropsychologia 36:1–9. Pillon B, Ertle S, Deweer B, Bonnet AM, Vidailhet M, Dubois B (1997) Memory for spatial location in ‘de novo’ parkinsonian patients. Neuropsychologia 35: 221–228. Pych JC, Chang Q, Colon-Rivera C, Haag R, Gold PE (2005) Acetylcholine release in the hippocampus and striatum during place and response training. Learn Mem 12:564–572. Ragozzino ME, Pal SN, Unick K, Stefani MR, Gold PE (1998) Modulation of hippocampal acetylcholine release and spontaneous alternation scores by intrahippocampal glucose injections. J Neurosci 18: 1595–1601.
288
NEURAL SYSTEMS PERSPECTIVE ON PLACE FIELDS
Ragozzino ME, Unick KE, Gold PE (1996) Hippocampal acetylcholine release during memory testing in rats: augmentation by glucose. Proc Natl Acad Sci USA 93:4693–4698. Schultz W (1998) Predictive reward signal of dopamine neurons. J Neurophysiol 80:1–27. Schultz W (2004) Neural coding of basic reward terms of animal learning theory, game theory, microeconomics and behavioural ecology. Curr Opin Neurobiol 14: 139–147. Schultz W, Apicella P, Ljungberg T (1993) Responses of monkey dopamine neurons to reward and conditioned stimuli during successive steps of learning a delayed response task. J Neurosci 13:900–913. Smith DM, Mizumori SJ (2006) Learning-related development of context-specific neuronal responses to places and events: the hippocampal role in context processing. J Neurosci 26:3154–3163. Suri RE (2002) TD models of reward predictive responses in dopamine neurons. Neural Netw 15:523–533. Suri RE, Schultz W (2001) Temporal difference model reproduces anticipatory neural activity. Neural Comput 13:841–862. Teather LA, Packard MG, Smith DE, Ellis-Behnke RG, Bazan NG (2005) Differential induction of c-Jun and
Fos-like proteins in rat hippocampus and dorsal striatum after training in two water maze tasks. Neurobiol Learn Mem 84:75–84. Vann SD, Brown MW, Erichsen JT, Aggleton JP (2000) Fos imaging reveals differential patterns of hippocampal and parahippocampal subfield activation in rats in response to different spatial memory tests. J Neurosci 20:2711–2718. Vinogradova OS (2001) Hippocampus as comparator: role of the two input and two output systems of the hippocampus in selection and registration of information. Hippocampus 11:578–598. Williams S, Mmbaga N, Chirwa S (2006) Dopaminergic D1 receptor agonist SKF 38393 induces GAP-43 expression and long-term potentiation in hippocampus in vivo. Neurosci Lett 402:46–50. Yeshenko O, Guazzelli A, Mizumori SJ (2004) Contextdependent reorganization of spatial and movement representations by simultaneously recorded hippocampal and striatal neurons during performance of allocentric and egocentric tasks. Behav Neurosci 118:751–769. Zgaljardic DJ, Borod JC, Foldi NS, Mattis P (2003) A review of the cognitive and behavioral sequelae of Parkinson’s disease: relationship to frontostriatal circuitry. Cogn Behav Neurol 16:193–210.
18 Spatial Decisions and Neuronal Activity in Hippocampal Projection Zones in Prefrontal Cortex and Striatum FRANCESCO P. BATTAGLIA, ADRIEN PEYRACHE, MEHDI KHAMASSI, AND SIDNEY I. WIENER
While place and grid responses in the hippocampal system are textbook examples of the cognitive basis of representations of abstract information on a single-cell level, their functional significance can only be evaluated by investigating how they impact other brain areas and behavior. This assertion is underscored by the observation that hippocampal lesions induce only anterograde amnesia while leaving previously consolidated memories intact, indicating a role for the hippocampus in the formation (and renewal) of memories that are successively stored elsewhere in the brain (Dudai, 2004; Frankland and Bontempi, 2005). Furthermore, since the hippocampal system has no direct projections to motor areas, its outputs must pass through other areas (whose cells do not have place responses) prior to reaching and influencing premotor, motor, autonomic motor, and neuromodulatory control areas and hence making an impact on ongoing behavior. The principal areas receiving hippocampal system place-related signals can be crudely divided into four main functional groups: (1) hypothalamus, lateral septum, amygdala; (2) anteromedial thalamic nucleus/medial mammillary nucleus (see Hopkins, 2005), midline thalamic nuclei, supramammillary nucleus; (3) ventral striatum, medial prefrontal cortex; and (4) subiculum, pre- and parasubiculum, entorhinal, and parahippocampal cortices, which have reciprocal connections with diverse cortical areas. These areas may be broadly summarized as respectively being associated with (1) autonomic, visceral, and emotional functions, together implicated in higher cognitive functions leading to consciousness (Damasio, 1999), (2)
theta-rhythm modulation and coordination of brain circuit dynamics in general, (3) learning of goal-directed behaviors, planning, sequencing, and action selection, and (4) information reprocessing, in particular memory consolidation. This chapter will focus primarily on the third system, but will also examine some aspects of the fourth, concerning relations with cortical areas. Our research has concentrated on these as most likely to be informative about the influence of hippocampal activity on goal-directed spatial orientation.
PREFRONTAL CORTEX: THE MAIN GENERATOR OF PLANNED BEHAVIOR Everyday life constantly requires choices among alternatives. From choosing the right time to switch lanes when driving a car, or selecting menu items for dinner, right up to making the most important choices in life, we engage some of the brain’s most important features: the circuits involved in making decisions. Decision processes involve the evaluation and prioritization of one’s needs and goals, and the analysis of information referring to the present, the immediate past, and our learned experience and knowledge relevant to the current context. The brain has to make predictions about possible upcoming events and action outcomes. Those predictions need to be evaluated for risks and benefits in order to select the more advantageous option; all of this unfolds continuously, at
289
290
NEURAL SYSTEMS PERSPECTIVE ON PLACE FIELDS
multiple cognitive and computational levels, nested and intertwined with each other. The prefrontal cortices are the central component of the ‘‘generator of planned behavior’’ (Dickinson, 1985). Behavioral planning and sequencing require diverse types of information concerning the current situation to be stored in a short-term memory buffer, or working memory, which has been shown to depend preponderantly on the prefrontal cortex (for reviews see, e.g., Goldman-Rakic, 1987; Fuster, 2000). The decisions are then implemented via its direct lines of command to motor structures (which express behavior) as well as to autonomic and neuromodulatory structures that regulate the internal state of the brain and the body (Goldman-Rakic, 1987; Pandya and Yeterian, 1990; Barbas, 2000; Rolls, 2004). From a functional point of view, the interaction between prefrontal cortex and the hippocampus has been implicated in both working and episodic memory processing. Working memory maintains online the different pieces of information needed to assess and formulate decisions. The prefrontal neural basis of this sort of memory, in the form of self-sustained delay activity, has been extensively studied from the experimental (Fuster, 2000; Goldman-Rakic, 1987) and theoretical point of view (e.g., in Amit and Brunel, 1997; Wang, 1999; Compte et al., 2000). Evidence for such memory-supporting activity has recently been observed in the rat prefrontal cortex as well (Baeg et al., 2003; Jones and Wilson, 2005). The prefrontal cortex’s role in working memory for diverse modalities could be implemented through specific temporary functional linkages with other modality-specific cortical regions relevant to current cognitive demands (see, e.g., Postle, 2006). In rodents, for example, it has been shown that experimental interference with hippocampal function affects working memory (Aggleton et al., 1986; Knowlton et al., 1989; Wan et al., 1994). Moreover, the integrity of the hippocampal–prefrontal network seems necessary for normal working memory performance (Floresco et al., 1997). In this network, the two structures would play different but complementary roles: Winocur (1992) showed that hippocampal lesions cause delay-dependent working memory deficits, whereas prefrontal lesions result in delay-independent impairment. Taken together, these findings point to a role for the hippocampus that is more centered on the actual maintenance of information, with the prefrontal cortex being implicated more in the monitoring and flexible use of such information (Moscovitch and Winocur, 2002). Thus, hippocampal and prefrontal activity would be coordinated for working memory in domains of hippocampal specialization, such as cue configurations distributed in space and/or time.
The hippocampus and prefrontal cortex are likely to collaborate in episodicmemory processing as well, although this may engage other mechanisms. Complex, flexible, goal-directed behavior, supported by prefrontal function, requires rich contextual information, that is, the ability to process a large amount of relational information, and reconstruct previous episodes that may be relevant for shaping the decision at hand. The hippocampal system can perform this task by informing the combinatorial tangle of interrelationships between sensory inputs into a topological structure, reflecting the arrangements of cues in the external space. This is an important component of episodic memory, which indeed has been found to be largely dependent on hippocampal activity. The prefrontal cortex, besides monitoring and processing episodic memories, may also play an active role in the formation and maintenance of episodic memory (Ferbinteanu et al., 2006), for example, encoding the temporal aspects of an episode (Funahashi et al., 1993; Ninokura et al., 2003, 2004).
The Hippocampal–Prefrontal Network Many anatomical and physiological lines of evidence also point to a close collaboration between the hippocampus and the prefrontal cortex: the prefrontal cortex is distinguished within the neocortex in that it receives direct hippocampal innervation (Swanson, 1981; Goldman-Rakic et al., 1984; Jay et al., 1989). In the rat, this projection originates in the ventral (temporal) portion of the CA1 hippocampal subfield and the subiculum and reaches prefrontal areas including infralimbic (Swanson, 1981), prelimbic (Ferino et al., 1987; Jay et al., 1989; Jay and Witter, 1991), medial orbital (Jay and Witter, 1991), and agranular insular (Verwer et al., 1997) areas. Notably, the dorsal hippocampus, whose neurons have the smallest firing fields and hence the most precise spatial representations, sends no direct connections to the prefrontal cortex; rather, the ventral hippocampus, with a lower incidence of spatially selective cells with larger fields (Jung et al., 1994; Poucet et al., 1994; Maurer et al., 2005), provides input to the prefrontal areas. Furthermore, Tabuchi et al. (2003) observed a greater incidence of discharges during immobility at reward sites in ventral (temporal) than in dorsal (septal) hippocampal neurons. This is not consistent with interpretation as hippocampal place responses, which generally cease during immobility (when thetafrequency rhythmic local field potentials are diminished). More recently, Hok et al. (2007) have observed comparable, albeit with distinct properties, goal siterelated activity in dorsal hippocampus. Thus hippo-
PREFRONTAL CORTEX AND STRIATUM IN SPATIAL DECISIONS
campal input to the prefrontal cortex can convey diverse aspects of the current context. The hippocampal-to-prefrontal connection has been extensively studied with electrophysiological methods: pulse stimulation of the ventral hippocampus elicits a short excitatory response in neurons in the prefrontal cortex, followed by a period of inhibition, reflecting both short- and long-term plasticity (Laroche et al., 1990; Mulder et al., 1997). Monosynaptic hippocampal inputs have been shown for prefrontal GABAergic interneurons as well (Tierney et al., 2004), so the effect of hippocampal projections on the prefrontal cortex can be described as a combination of excitation and feed-forward inhibition. The inverse pathway, from the prefrontal cortex to the hippocampus, is mediated by the parahippocampal cortices (see, e.g., Burwell and Witter, 2002). An interesting alternate relay between the two structures is represented by the nucleus reuniens of the thalamus (see Vertes, 2006, for a review). The interaction between the prefrontal cortex and hippocampus serving as a major information relay for the ventral striatum and the midbrain dopaminergic structures will be discussed below.
Spatial Representations and the Prefrontal Cortex The strong hippocampal input and the importance of spatial contextual information for decision making hint at the possibility that the prefrontal cortex contains some sort of representation of space, as a part of a more generic encoding and processing of a behavioral situation. Such a representation would not likely be in the form of a stable, cognitive-map representation of the spatial environment, such as that proposed for the computational basis of navigation. Indeed, an intact prefrontal cortex is not needed for the acquisition and performance of navigational tasks such as the Morris water maze (Maaswinkel et al., 1996; de Bruin et al., 1997; Kesner, 2000). Rather, the prefrontal cortex may contain representations of higher-order interactions (Banquet et al., 2005) between place and other factors relevant to behavior, such as the presence or absence of a cue, a reward, the location of a goal, changing environmental contingencies, or motor patterns. The results of several experimental studies support this view. In a study by Jung et al. (1998) in which rats performed behavioral tasks on mazes, the activity of prelimbic cortex cells was selective for particular task phases (approaching, leaving reward sites, etc.), particular behavioral patterns such as body turns, or conjunctions of these same factors with place. In a delayed alternation task on a figure-eight maze, pre-
291
limbic cells were active on the central arm, depending on the previous choice arm, consistent with a role in maintaining a working memory–related trace (Baeg et al., 2003). Medial prefrontal cells show rewardrelated anticipatory activity (Pratt and Mizumori, 2001). In a place preference task (Hok et al., 2005), prelimbic and infralimbic cells showed place-related responses, with the goal area (which was dissociated from reward) disproportionately represented, so that the goal location was encoded by the overall ensemble activity. Another perspective on the function of the prefrontal cortex, in particular the prelimbic/infralimbic (PL/IL) areas, has emerged from lesion experiments in rats (Ragozzino et al., 1999a,b; Birrell and Brown, 2000; Delatour and Gisquet-Verrier, 2000; Granon and Poucet, 2000), which show evidence for flexible attentional shift between contingency rules related to different sensory and behavioral dimensions. Hence, it may be argued that prefrontal place representations should be greatly influenced by the current task contingency rule, or by the information domains currently being attended to. For example, for items such as the predicted goal/reward location, the position of a currently attended cue may be encoded in a prefrontal representation, and such encoding may be radically modified by a change in the contingency rule. In our work we have used extracellular neurophysiological recordings of single-unit activity and local field potentials in behaving animals to investigate the mechanisms underlying cognitive function in terms of the dynamic signal-processing algorithms implemented by brain circuitry. The behavioral correlates of single-unit activity show the types of information discernible in the neural activity of hippocampal efferent structures, permitting comparison with the hippocampal system. Deductive reasoning then indicates what these structures may have received from the hippocampus as well as what other information might have arrived from other structures or via internal processing. Analyses of cross-correlations between discharges of neurons in hippocampus and downstream structures can indicate common sources of inputs and may even suggest transmission of signals from one structure to the next. Moreover, such cross-correlations can be tested for behavioral modulation, hence providing evidence for selective gating of hippocampal output signals according to self-organizing principles of brain circuitry. The comparison of local field potentials (LFPs) in the hippocampus and downstream structures can demonstrate the coherence or, alternatively, decoupling of the respective structures. Selective coherence of LFPs is hypothesized to be a mechanism of functional coupling of modular subsystems of the brain.
292
NEURAL SYSTEMS PERSPECTIVE ON PLACE FIELDS
Battaglia et al. (2005b, 2006) tested these hypotheses using multiple tetrode recordings of the activity of prelimbic cell ensembles simultaneously with hippocampal LFP recording. This was done as the rats learned and performed a contingency shift task on a Ymaze, formed by three arms separated by 1208, with a circular platform at the center. Rats started all trials from the same departure arm, and after the central barrier was lowered, they had to select between the two choice arms, then go to the end to receive a chocolate milk reward. Concomitantly with the lowering of the barrier, one of the two arms, randomly selected for each trial, was illuminated. For each trial, the reward was available on only one arm. The baited arm was determined on the basis of one of four possible contingency rules: (1) the right arm was always baited (spatial-orienting right rule), (2) the left arm was always baited (spatial-orienting left rule), (3) reward was available on the illuminated arm (visual cue– based light rule), (4) reward was available on the non-illuminated arm (visual cue–based dark rule; Fig. 18–1A). Once the rat acquired the current rule (i.e., performance reached a criterion level), the rule was then switched (always in an extradimensional fashion, that is, from a spatial-orienting rule to a visual cue– based rule or vice versa). The switch was not explicitly signaled to the rat in any way, so that it had to be inferred by the pattern of unexpectedly unrewarded trials. Rats typically took 3–6 experimental sessions, or 60–100 trials, to acquire the rule when the switch was toward a visual cue–based trial; much less time (typically one session) was required for a switch toward a spatial-orienting rule. The rat prelimbic cortex cells showed a variety of behavioral correlates (Battaglia et al., 2006); for example, certain neurons were active preferentially at the beginning of the trial, during the run on the arm, or
at the end of the trial, in either a reward-dependent or reward-independent fashion. At the beginning of the trial (thus prior to the animal’s selection of a maze arm), some cells encoded the position of the intended arm choice. In this way, these cells were encoding a response relevant to either a visual cue–based or a spatial-orienting rule. If we analyze the activity of these cells from the viewpoint of their spatial correlates, we have a picture that reflects this high degree of flexible encoding. Similar to the results of Jung et al. (1998), these cells showed spatial correlates, but not (even transiently) resembling a static map of an environment, as is found in the hippocampal formation and entorhinal cortex. For the most part, space is represented as a function of its role in the task. Figures 18–1B–E shows some examples of firing-rate maps for spatially selective prelimbic cells recorded in the Y-maze task. For example, Figure 18–1B shows a neuron with selective firing immediately after choosing the left (Fig. 18– 1B2), but not right (Fig. 18–1B1), arm during reward site approaches. The low firing rate is characteristic of prefrontal principal neurons. Figure 18–1C shows an example of a cell firing selectively at the reward sites at the end of the left arm, perhaps associating the behavioral choice and a possible negative or positive outcome. Figure 18–1D depicts a cell with a similar response on the right arm. For most cells, spatial information is encoded in combination with task phases and other behavioral variables—Figure 18–1E shows a cell that fired preferentially after arrival at the left reward site. Moreover, after a contingency shift was imposed, the prelimbic cortex responded with dramatic changes in the activity correlate of many of its cells: certain cells acquired or lost responses in specific task phases, either abruptly or in a way that closely followed the
" Figure 18–1. A. Schematic depiction of the contingency rules used for the Y-maze task and recording of prelimbic cortex neurons. The rat started each trial on the ‘‘departure’’ arm (at bottom) and had to select between the other two arms. In the spatial-orienting rules, the rat had to always choose the right arm (spatial-orienting right rule, left diagrams) or the left arm (spatial-orienting left rule, not shown), regardless of which arm was illuminated. With the visual cue–based light rules (middle diagrams), the rat had to direct itself toward the illuminated arm. With the visual cue–based dark rule (right diagrams), the rat had to choose the unlit arm. B1, B2. Occupancy-normalized firing rate maps for a prelimbic neuron for trajectories from the start arm to the goal arm for right- and left-arm choices, respectively. Activity was significantly different between left and right arms in the 2.5-s period following trial start (Kruskal-Wallis p < 0.002, cell 150712–1-7). C. Prelimbic neuron with selective firing on the left arm for the 2.5-s period prior to and following reward site arrival (Kruskal-Wallis p < 0.0001; cell 201227–1-5). D. From the same tetrode as the neuron in C, this neuron fired selectively on the right arm (prior to and after reward site arrival; Kruskal-Wallis p < 0.006; cell 201227–1-16). E. Another prelimbic neuron with spatial selectivity (200111–2-6). There was a significant left–right difference in the post-arrival period (Kruskal-Wallis p < 0.0001). This difference does not correspond to a spatial asymmetry of rewarded and unrewarded trials (not shown). Scale is in impulses per second
293
294
NEURAL SYSTEMS PERSPECTIVE ON PLACE FIELDS
changes in the rat’s strategy (Wiener et al., 2006). These shifts in response properties bear similarities to the dramatic modulation of behavioral correlates described below in accumbens neurons in rats performing task shifts or after maze rotation (Shibata et al., 2001). Here shifts were also found for the spatial correlates of activity. Figure 18–2 depicts examples of firing-rate maps for prefrontal cortex cells recorded in sessions when the rat reached the performance criterion mid-session and the contingency rule was then changed. These cells with position selective responses demonstrate dramatic changes following the rule shift. Figure 18–2A shows a cell with a prominent response on the left arm before the shift (while this was under a spatial-orienting left rule there were five visits to the right arm, and activity is shown only prior to rewards, not after.). This activity moves to the start arm after the shift. Figure 18–2B shows a cell that had an increase in firing rate on the start arm after the contingency change. In Figure 18–2C there is a decrease after the contingency shift. To summarize, we have shown here prefrontal responses that carry information about place, even though place per se is not considered to be the most important correlate of prefrontal cell activity. These extended representations, compared to the hippocampal or entorhinal ones, are of little utility for the animal to precisely locate itself in an environment. They seem more adapted for guiding the goal-directed behavior of the rat and encoding potentially relevant variables in the task, providing candidate templates for learning new reward contingencies. No finegrained place-field response was found in the prefrontal cortex, for example, covering only a restricted portion of one arm, although activity was observed at points with specific behavioral significance, such as the decision point or goal sites. This may reflect the hippocampal input (from the ventral portion of the hippocampus) that contains only a rather coarse place representation (Jung et al., 1994). The prefrontal cortex could select such ‘‘relevant’’ input information on
the basis of temporal relations between activity in its afferent regions (including hippocampus, see, e.g., Battaglia et al., 2005a) and reward onsets. In some way the hippocampus and prefrontal cortex may be coordinated to permit the brain to compare rewards on successive types of trials (left–right, light–dark) and elaborate a new rule—perhaps with the participation of the striatum and its learning mechanisms described below. The ventral hippocampal input would then provide spatial information in a form particularly suitable for selecting among a limited number of paths at successive crucial choice points. A reduced representation of the environment is all that is necessary; precise topographic localization would render computations unnecessarily complex (Trullier et al., 1997). It would be interesting to probe the same cells in a task situation where place (that is, identification of a given arm) has no particular relevance for reward acquisition, and see whether that selectivity is lost. Finally, we would like to point out a possible parallel between these prefrontal activity shifts and the way the hippocampus can remap, or switch to, a partially or completely different representation of an environment (e.g., Barnes et al., 1997; Leutgeb et al., 2005), prompted by new or conflicting environmental inputs or the creation of a different context. Since the evidence presented above does not support the view that the prefrontal cortex encodes a ‘‘topographic map’’ of the environment, the word remapping is inappropriate. Nonetheless, it is intriguing that these two phenomena take place in structures so closely related to each other. Wiener et al. (1989) showed that the behavioral correlates of hippocampal neurons changed from coding olfactory cue configurations to spatial position when the rats shifted between an odor discrimination task and a water search task in the same arena (see also Markus et al., 1995). Remapping in the hippocampus, whether caused by changes in the spatial context or by changes in the behavioral task, can be characterized as a transition between two spatial representations (or, in any event, representations that
" Figure 18–2. Occupancy-normalized firing-rate maps for prelimbic neurons recorded on the Y-maze task of Figure 18–1. The firing-rate map obtained for the periods up until the contingency switch and thereafter are depicted, respectively, to the left and the right of each row. The contingency rule is indicated above each panel; the number corresponds to the order of presentation of this rule in the recording session. In A, there was activity on the left arm for the visual cue–based dark-rule task. This disappeared when the contingency was changed to the spatial-orienting left rule (A2). Note that this activity was not correlated with whether trials were rewarded (correct) or not. In B, the shift in contingency gave rise to an increase in activity in the start arm. In C, there was a change in spatial firing after the contingency shift. This is difficult to see in the firing rate maps because of the low firing rates of this neuron; it is more evident in C10 and C20 , where the trajectories are shown in gray and locations where the neuron discharged are shown as red points. There is more activity at the end of the right arm in the spatial-orienting right rule (C1) than in the visual cue–based light rule.
296
NEURAL SYSTEMS PERSPECTIVE ON PLACE FIELDS
reflect the relationship between stimuli and locations). Thus hippocampal activity shifts correspond to changes in the behavioral context of the organism. The prefrontal code, by contrast, changes according to changing contingencies, which require altering the way the animal responds to stimuli and directs its behavior. This computation incorporates hippocampal information, as well as reward contingencies, and the relation between behavior patterns and its impact on the internal state. A second type of shift we observed in prefrontal neurons occurred when the animal shifted its behavioral pattern, adapting to the new contingency rules. It seems likely that these two codes and their dynamics are not completely uncoupled, rather, that they interact in a sophisticated way, representing a fascinating subject of research for the future.
VENTRAL STRIATAL AND HIPPOCAMPAL ACTIVITY The substrate of these cognitive processes is not limited to the prefrontal cortex and hippocampus, rather it is distributed in a network including structures intimately connected to them, such as the amygdala, the striatum, particularly the nucleus accumbens, as well as midbrain dopaminergic structures. The striatum is the entry point to the basal ganglia, an ensemble of subcortical nuclei organized in parallel loops receiving inputs from different cortical areas including the hippocampal system and recurrently sending processed signals back to the prefrontal cortex through the mediodorsal thalamic nucleus (Alexander and Crutcher, 1990). It has been proposed that this architecture favors the selection of a motor or cognitive (decision) response on the basis of the present context (sensory, spatial, and/or motivational; Graybiel, 1998; Redgrave et al., 1999). The input nucleus to the basal ganglia, the striatum, contains large segregated territories corresponding to these loops (Uylings et al., 2003). To broadly summarize, all of the neuron groups in a given territory can be characterized by the prefrontal subregion and cortical regions from which they receive input. The dorsal striatum (putamen and dorsolateral caudate) receives primarily sensorimotor inputs, while its counterpart, the nucleus accumbens (in ventral striatum), receives hippocampal and amygdalar inputs (e.g., Pennartz et al., 1994). Other intermediate subdivisions (between dorsal and ventral) correspond to different modalities, with unique convergences of diverse cortical inputs arriving in distinct foci. Prescott et al. (1999) proposed that the basal ganglia carry out action selection, which can be considered a type of decision making. They cited the ana-
tomical architecture as well as neurophysiological data as implicating this system in motor decisionmaking (e.g., Gulley et al., 2002). The respective roles of the prefrontal cortex and the striatum in decision making are not clear, however. Some authors suggest that they are in competition for the control of behavior, with the striatum controlling behavior requiring simpler decisions than the prefrontal cortex (Daw et al., 2005). Other authors propose that the striatum and the prefrontal cortex are cooperating together, the striatum taking into account candidates of possible responses ‘‘biased’’ by the prefrontal cortex and making the final decision (Redgrave et al., 1999). Pasupathy and Miller (2005) recently showed in electrophysiological experiments in monkeys that while both the prefrontal cortex and the striatum adapted their neural responses with learning, the prefrontal cortex adapted more slowly than the striatum and was more correlated to changes in the animal’s behavior. This finding is consistent with the hypothesis that basal ganglia gate the prefrontal cortex to bias the decisionmaking process taking place cortically (Frank et al., 2001). Learning to adapt these selections could involve dopaminergic signals (Schultz et al., 1997). Dopaminergic neurons in monkey ventral tegmental area (VTA) and substantia nigra pars compacta (SNc) fired at higher rates when an unexpected reward was encountered, but the response vanished as the reward became predictable and was inhibited when an expected reward was omitted (Schultz, 1998; Schultz et al., 2000). This finding supports the idea that dopamine carries reward-prediction error signals (Schultz et al., 1997). Long-term potentiation (LTP) and longterm depression (LTD) have been observed at corticostriatal synapses after exposure to dopamine (Centonze et al., 2001; Reynolds et al., 2001), supporting the hypothesis that these signals are implicated in learning processes taking place in the striatum (Houk et al., 1994). The dorsal and ventral striatal territories mentioned above receive dopaminergic input from distinct zones, the VTA and the SNc, respectively (Haber et al., 2000; Joel and Wiener, 2000; Thierry et al., 2000; Ikemoto, 2002). These data suggest that dopamine release in the dorsolateral striatum would affect the learning of stimulus–response associations, permitting the establishment of a repertory of stereotyped, habitual, and automatic actions (Houk et al., 1995; Graybiel, 1998; Miyachi et al., 2002). However, the role of the ventral striatum in learning is less clear. Hypotheses include driving stimulus–response learning in the dorsal striatum (Niv et al., 2005), learning to select trajectories or routes based on hippocampal afferences (Arleo and Gerstner, 2000), learning to select among navigation strategies (Girard et al., 2005),
PREFRONTAL CORTEX AND STRIATUM IN SPATIAL DECISIONS
or learning to integrate amygdalar afferences to drive motivationally modulated goal-directed behavior (Dayan and Balleine, 2002). Much remains to be understood about the precise involvement of the ventral striatum in behavior and cognitive function. While these ventral striatal regions have attracted our interest because of their rich interconnection with the hippocampal formation, they are also well placed to play an important role in cognitive functions by virtue of their inputs from prefrontal cortex and amygdala, and their projections to premotor and neuromodulatory areas and to basal ganglia loops transiting through mediodorsal thalamic nucleus back to prefrontal cortex. In the rat, the caudal dorsomedial striatum receives bilateral inputs from lateral entorhinal cortical layers III and IV (McGeorge and Faull, 1989)—the former region also has head direction cell activity (Wiener, 1993). The nucleus accumbens receives inputs from medial and lateral entorhinal cortex layers II, III, IV, and VI (therefore directly acquiring head direction and path integration information, the latter in the form of ‘‘grid cells’’; Hafting et al., 2005; Sargolini et al., 2006). Furthermore, the nucleus accumbens shell receives a topographic projection from the hippocampus via the subiculum, with the ventrolateral and rostral shell receiving dorsal (septal) hippocampal inputs and caudal and medial shell receiving ventral (temporal) hippocampal inputs (see Groenewegen et al., 1996). Since dorsal hippocampal neurons have a higher incidence of place-correlated responses and smaller firing fields (Jung et al., 1994), it might be expected that this trend be reflected in the response properties of neurons of the respective parts of the accumbens shell. This is, however, difficult to ascertain, since the shell, named for its distribution as a thin (300–500 micron) layer surrounding the accumbens core, is rather difficult to access, particularly in the ventrolateral part.
Experimental Data Comparing Ventral Striatal and Hippocampal Responses In our previous studies we examined the simultaneous activity of hippocampal and ventral striatal neurons in rats performing spatial orientation tasks. In one series of experiments, recordings were made in rats shuttling between one of two pairs of reward dispensers, each located on diametrically opposite edges of a circular platform (the sites were located at northeast, southeast, southwest, and northwest positions). While hippocampal neurons demonstrated place (or, in a small subset of neurons, position-independent behavioral) correlates, ventral striatal neurons demonstrated selective responses for particular task-related behaviors (Shibata et al., 2001; also see Korshunov et al., 1996;
297
Wiener, 1996). Many of these responses occurred immediately prior to reward release. Interestingly, these responses were often spatially modulated, that is, they were greater prior to arrivals at certain reward sites. While the identity of the preferred sites varied among neurons, there were always at least two. Thus this activity was distinct from place responses of hippocampal neurons, which were selective for single sites only (Trullier et al., 1999). When the circular platform was rotated by 908 or the animals were required to shift between cue approach and spatial mapping strategies, the place responses of the hippocampal neurons invariably remained stably fixed relative to the experimental room (see Fig. 18–3). The ventral striatal neurons, by contrast, underwent dramatic shifts in the amplitude of responses following task contingency changes (see Fig. 18–3). We deduced that this task strategy selectivity was not a result of the hippocampal input. However, the spatial modulation of the ventral striatal responses could well have derived from the hippocampal system inputs to this region. In a subsequent study we found significant crosscorrelations between simultaneously ipsilaterally recorded accumbens shell and hippocampal neurons in rats performing in a four-arm (plus) maze (Tabuchi et al., 2000). The cross-correlations were preferentially observed as the rats arrived at the reward arms, rather than after reward delivery or during arrivals at the maze center (see Fig. 18–4). This result points to a task-related modulation of the relation between hippocampal and accumbens shell activity. The most frequently observed latency for these cross-correlations was that of the hippocampal neuron discharging 10– 20 ms prior to the accumbens shell neuron, which corresponds precisely to the latency of evoked potentials in the accumbens shell following electrical stimulation of the hippocampus (e.g., Albertin et al., 2000). It is unlikely that single hippocampal neurons were monosynaptically evoking action potentials in single accumbens neurons, for various reasons (such as the low peak firing rates of individual hippocampal neurons and the diffuseness of its axonal arborizations). However, another explanation derives from observation of the timing of significant cross-correlations in our entire sample of recordings. As shown in Figure 18–5, the peak incidences of cross-correlations occurred at 200, 80, 20, 130, and 240 ms. (accumbens neuron discharges are shown at time zero, and hippocampal cross-correlation peak incidences are distributed along the x-axis). The latency between the successive peaks is 120 ms, corresponding to the theta rhythm. This correspondence suggests, then, that the correlated activity between these structures occurs preferentially during the theta rhythm, and correlates nicely with the relatively few cross-correlations
Figure 18–3. Comparison of striatal and hippocampal recordings in rats alternating between performing beacon approach and place navigation tasks on the same platform (overhead view shown in inset in left histogram). The rats had to shuttle between diametrically opposite reward sites. In the beacon approach task (B) the two active reward sites were cued by lights at the reward sites, while in the place navigation task (P) only northeast and southwest sites were rewarded—and these were unmarked. The platform was rotated after five to eight trials as a challenge. Adjacent to the raster plots, B or P indicates the task and the number indicates a rotated position of the platform—different task configurations (epochs) are separated by horizontal bars. Left: Pre-reward response in a neuron of the ventral shell of the nucleus accumbens. The cell activity increased gradually as the animal arrived at the reward box, then stopped suddenly as the drop of water was delivered (at time zero). The cell did not fire at all during visits to the northeast or northwest corners, and had a maximum firing rate of only 4 impulses/s as the animal arrived at the southeast corner (not shown). Thus this pre-reward-correlated firing was spatially modulated. This neuron showed significant changes in firing rate (asterisks and arrows in right column) after changes in the task contingency and after a different reward box was rotated into this southwest corner. Right: Hippocampal place responses show no such sensitivity to task configuration changes. The inset shows a spatial plot of this activity. The green lines indicate the trajectories of the rat, the red squares show that the neuron fired when the rat left the reward box in the upper right corner. In the raster plot (synchronized with arrivals at the southwest reward site), this spatial response remained unchanged even after the animal was required to change its orientation strategy to place navigation. Adapted with permission from Shibata et al. (2001) and Trullier et al. (1999).
298
PREFRONTAL CORTEX AND STRIATUM IN SPATIAL DECISIONS
299
Figure 18–4. In recordings from a rat performing a plus-maze task, incidence of significant peaks in crosscorrelations (CCR) histograms computed for three different task-related behaviors in a group of 18 cell pairs selected for having significant peaks (>3.0 SD of confidence limits) corresponding to the latency of the hippocampus to nucleus accumbens (NAcb) pathway (10–20 ms). For analyses 1-s observation periods were selected before or after three task events: arrival at reward sites, departure from reward sites, and center arrivals. Note that there are fewer peaks in analyses for center arrivals than for reward box arrivals or departures, and a greater number of peaks are found for latencies corresponding to connections from hippocampus to accumbens. Shaded columns are data from paired recordings of dorsal hippocampus vs. NAcb ventrolateral shell neurons, while open columns are from pairs of ventral hippocampus vs. NAcb medial shell neurons. Reproduced with permission from Tabuchi et al. (2000).
following reward delivery when the rat was immobile and no theta would be expected. But differences in the expression of behaviors associated with theta rhythmic activity cannot account for the greater incidence of cross-correlations for runs toward the goals rather than toward the center. One possible explanation is that the hippocampal input to accumbens was gated
off during this post-reward period, perhaps permitting the accumbens to more effectively attend to other inputs or internal processing. This plus-maze task was designed to address the controversial issue of how goal sites are represented in the hippocampostriatal system. Increasing evidence is becoming available about an effect of goal location on
300
NEURAL SYSTEMS PERSPECTIVE ON PLACE FIELDS
Figure 18–5. Incidences (counts) of significant peaks in cross-correlations of 154 hippocampus–nucleus accumbens cell pairs for analyses of 1-s observation periods prior to the instant that the rats arrived at any of the boxes. The histogram shows the number of cell pairs with significant peaks of hippocampal activity in a time window ranging 300 ms prior to and after the occurrence of each nucleus accumbens action potential (bin width, 10 ms). Most frequently occurring peaks were at 20 ms (hippocampus firing prior to accumbens) and at intervals of 100–110 ms, corresponding to the hippocampal theta rhythm. Reproduced with permission from Tabuchi et al. (2000). hippocampal activity, for example, in the form of a clustering of place fields around the goal location (Hollup et al., 2001), a prospective encoding of the goal selected by the animal (Lee et al., 2006; Hok et al. 2007), or subtle, transient remapping effects (Jackson, 2006). The issue was raised in a neural network model of Burgess et al. (1994), in which hippocampal place activity was used for navigating toward goal sites. The authors hypothesized the existence of ‘‘goal cells’’ that would discharge from the moment the goal was selected until the model avatar arrived at the goal site. Since the ventral striatum was proposed to be a limbicmotor interface (Mogenson et al., 1993), this area was recorded in our study. We found neurons that discharged along trajectories between task-relevant points in the plus maze. For example, some neurons discharged action potentials from the instant the rat departed from the four reward sites until it arrived at the
center of the maze. This was not simply a motor correlate, since these neurons did not discharge during outward trajectories on the arms. In contrast, other neurons discharged preferentially during the outward, but not inward, trajectories. The latter group showed preferential firing for certain subsets of maze arms (see Fig. 18–6). However, as found in Shibata et al.’s (2001) study, no striatal neurons were selective for only a single arm in the manner of hippocampal neurons. Thus these neurons could well mediate goaldirected spatial-orienting behaviors on the basis of hippocampal input signals. This striatal activity could act to implement the prospective coding of hippocampal neurons, since these neurons do not continue to discharge until the animal arrives at the goal site. Hok et al. (2005) have shown comparable responses in the rat medial prefrontal cortex, which also projects to the striatum.
PREFRONTAL CORTEX AND STRIATUM IN SPATIAL DECISIONS
301
Figure 18–6. Goal coding in ventral striatal neurons. Top: This ventromedial caudate neuron discharged as the rat walked from the maze center to the reward sites at the ends of the arms. In these overhead views of the plus maze, cell firing is indicated by red crosses and position samples with no firing as blue dots. This activity is spatially modulated: ANOVA post-hoc tests ( p < 0.05) confirmed that the firing rate during the period from 1.0 to 0.5 s prior to arrival at boxes 1 and 2 is significantly greater than that for box 4 and that the rate during approaches to box 2 was greater than that for box 3. (Box 1 is to the lower right and numbering continues counterclockwise.) Bottom: Another ventromedial caudate neuron continuously active during displacements to the goal boxes but not the maze center. This activity was also spatially modulated, but with a different profile than for the neuron above. Adapted with permission from Mulder et al. (2004).
HIPPOCAMPAL–CORTICAL INTERACTION AND MEMORY Sleep has been recognized as fundamental for the successful consolidation and storage of recently acquired memories. Brain activity during sleep (and behavioral inactive states) is highly structured, carries informationally relevant content, and is likely a crucial player
in such consolidation processes. In the rat hippocampus, several features of activity configuration taking place during a behavioral task are preserved during subsequent sleep (Wilson and McNaughton, 1994; Skaggs and McNaughton, 1996; Lee and Wilson, 2002; Battaglia et al., 2005b), and decay during the first half hour of sleep after the task (Kudrimoti et al., 1999). In humans, the same hippocampal areas active
302
NEURAL SYSTEMS PERSPECTIVE ON PLACE FIELDS
during learning of a navigation task are also active during sleep, in a way that covaries with the improvement in navigation performance, as shown, for example, by functional imaging studies of Peigneux et al. (2004). Outside the hippocampus, memory trace– carrying reactivation has also been found in rat and monkey neocortex (Hoffman and McNaughton, 2002; Ji and Wilson, 2007) and in rat striatum (Pennartz et al., 2004). The communication between the hippocampus and the neocortex is an important component of such mechanisms taking place during sleep. Marr (1970, 1971) was the first to theorize that hippocampal memory reactivation during sleep has the role of orchestrating neocortical activity so that a cortical memory trace can be formed, encompassing several different regions (for a review, see McNaughton et al., 2003). In fact, human neuropsychological data suggest a gradual transition of memories from a hippocampaldependent state to a hippocampal-independent (presumably neocortically supported) state (Dudai, 2004). Activity-related gene expression techniques have also confirmed that as memories become more and more remote, the neocortical role in maintaining memories increases and the hippocampal role decreases (Bontempi et al., 1999; Ross and Eichenbaum, 2006). Taken together, these findings strongly suggest that the hippocampus and the neocortex should exchange information during sleep. This communication may be mediated by the distinctive patterns of brain activity observed during sleep. During slow-wave sleep, in particular, the neocortex alternates between states of generalized high activity (‘‘up’’-states) and periods of global neural silence (‘‘down’’-states; see, e.g., Steriade and Buzsaki, 1990; Cowan and Wilson, 1994; Petersen et al., 2003). Up- and down-states repeat themselves with a period of hundreds of milliseconds up to seconds (Steriade et al., 1993) in anesthetized and natural sleep states, and organize faster oscillatory phenomena such as sleep spindles (Amzica and Steriade, 1998). Upstates seem to be the result of the action of recurrent cortical excitation (Sanchez-Vives and McCormick, 2000) and inhibition (Timofeev et al., 2001). The hippocampus shows a very different pattern of activity in slow-wave sleep: most of the activity is concentrated in brief (50–100 ms), burst-like episodes of activity, breaking a state of almost complete silence with little regularity—the sharp waves. Sharp waves are probably generated by positive feedback in the recurrent collaterals of the CA3 hippocampal subfield. In CA1, they are accompanied by a 150- to 200-Hz oscillation in the local field potential (ripple oscillations; Buzsa´ki et al., 1992; Csicsvari et al., 2000). Sharp waves are a potent influence on cortical activity in the entorhinal cortex (Chrobak and Buzsa´ki, 1994,
1996) and are correlated to spindle oscillations in the prefrontal cortex (Siapas and Wilson, 1998). Sirota et al. (2003) showed that cortical oscillations in somatosensory cortex and hippocampal activity are correlated over a short time scale; sharp waves are correlated with the hippocampal activity 50–100 ms earlier. They also showed that spindle and delta oscillations in the cortex affect hippocampal activity, with hippocampal synaptic inputs phase-locked to the cortical oscillations, possibly influencing the generation of sharp waves. Thus the communication between the hippocampus and the neocortex can be bidirectional (see also Hahn et al., 2006). Because hippocampal memory trace reactivation is strongest during sharp waves (Kudrimoti et al., 1999), it is attractive to think that sharp waves represent an input from the hippocampus to the rest of the brain that is powerful and synchronized enough to orchestrate reactivation in the brain. In fact, the ventral striatal neurons modulated by hippocampal sharp waves show a higher degree of reactivation than their peers (Pennartz et al., 2004). Using an extensive array of electrodes over a large portion of the rat neocortex, Battaglia et al. (2004) demonstrated that up- and down-states are characterized by a large degree of coherence across the entire neocortex, with virtually all cortical neurons shutting down during a down-state and global simultaneous activation during up-states. Moreover, by looking at the sharp wave–triggered average of global cortical activity, they were able to show that there was a probabilistic relationship between the occurrence of sharp waves and the transitions from down- to up-states. Figure 18–7A–D reproduces the sharp wave–triggered averages. Both in periods of high-delta/slow-EEG activity (probably corresponding to deeper sleep) and in periods of low-delta/slow-EEG activity (probably corresponding to lighter sleep or quieter wakefulness), cortical activity showed a peak for approximately the duration of the sharp wave, shown in the expanded scale in Figure 18–7B,D. The sharp wave–triggered averages also showed a slower trend, with the overall cortical activity before a sharp wave being on average smaller than that after the sharp waves. Interestingly, in high-delta periods, the cortical activity showed a relatively flat baseline up until the sharp wave, then a sudden increase, and then a decline over the next 10 s (Fig. 18–7A). A different picture emerged from the low-delta EEG periods: here the activity declined in the 10 s preceding the sharp waves, returned to baseline at the time of sharp-wave onset, and remained rather constant thereafter (Fig. 18–7C). This picture is compatible with the idea that a fraction of the downto-up transition corresponds with the times of the sharp waves (in a time scale of seconds): in high-delta
PREFRONTAL CORTEX AND STRIATUM IN SPATIAL DECISIONS
303
Figure 18–7. A. Peri-event time histogram (PETH) of cortical population activity during periods of identified global oscillations in the delta/slow range, centered on hippocampal sharp-wave events (error bars ¼ SEM; bin size, 100 ms). B. Same PETH, with an expanded scale (bin size, 20 ms). C. Sharp-wave-triggered PETH (bin size, 100 ms) of cortical firing during periods in which oscillations in the delta/slow range were absent. D. Same PETH shown on an enlarged scale (bin size, 20 ms). E. Recording of prelimbic cortex local field potentials (solid black line, depicted negative-up) and global instantaneous unit activity (gray dashed line), and hippocampal sharp wave/ripple events (inverted triangles). Cortical down-states were evident in the local field potential trace (indicated by arrows), with successive transition to up-states. A–D adapted with permission from Battaglia et al. (2004); E from Battaglia FP, Khamassi M, Peyrache A, Tierney P, and Wiener SI (unpublished results).
periods, the up-states are much shorter than the downstates (causing the decline visible after the sharp wave in Fig. 18–7A), whereas the converse is true in lowdelta states (corresponding to a more excited brain state). In this functional regime, down-states are rarer with shorter occurrence, whose transition into upstates is probabilistically related to sharp waves,
yielding the trough in the sharp wave–triggered average prior to the sharp wave in Figure 18–7C. A 10-s excerpt (Fig. 18–7E) from our medial prefrontal recordings during sleep shows an example of how this probabilistic relationship is expressed over the time course of prefrontal and hippocampal activity. Here we highlight the clustering of hippocampal
304
NEURAL SYSTEMS PERSPECTIVE ON PLACE FIELDS
sharp-wave events after prefrontal down-states, indicated by downward—positive—fluctuations in the prefrontal local field potentials (solid line) and corresponding drops in the global prefrontal unit activity (dashed line). These lines of evidence point toward an important correlation between the major collective phenomena characterizing the slow wave sleep dynamics of the neocortex (up-/down-states) and hippocampus (sharp waves). The interaction is probably bidirectional: the results from stimulation experiments we reviewed above indicate that the hippocampal activity can measurably affect the prefrontal cortex; on the other hand, the results by Sirota et al. (2003) and Hahn et al. (2006), for example, suggest an influence of cortical up-states on the hippocampus. A theoretical possibility is that this reciprocal action is at the basis of a functional loop that subserves memory maintenance and consolidation: a hippocampal sharp wave may influence cortical activity in initiating an up-state, and instating the reactivation of a memory item in the cortex. In turn, the up-state may affect the activity in the subsequent hippocampal sharp wave, helping the hippocampus to maintain continuity of information processing in the silent period between sharp waves.
CONCLUSIONS We have presented here a brief overview of the neural representations in the rat prefrontal cortex and striatum in spatial tasks. While both structures appear to be strongly affected by position information (and possibly hippocampal input), the nature of their representation is deeply different from the map-like structure of hippocampal-system activity. Both structures seem to be affected by combinations of factors such as task phase, current motor behavior, reward or lack thereof, and task-relevant locations. The prefrontal cortex is sensitive to the task-relevant cues, and exhibits a large deal of flexibility in changing its response characteristics when the reward contingency is changed, in accordance with its proposed role in attentional shifts. Spatial modulation of activity is often present in these two structures, but most likely in conjunction with some other influence, such as the factors mentioned above. When present, the wide scale of the spatial modulation is more reminiscent of the responses in the ventral hippocampus (where the hippocampal projections to the prefrontal cortex originate). It seems to us that these data indicate the importance of the interactions among structures more implicated in self-location (such as the hippocampus) and structures associated with planning, behavioral selection, and decision making (the prefrontal cortex,
the striatum). Such interactions seem to be relevant also during sleep, thus suggesting a role for coordinated hippocampal–cortical–striatal activity for memory consolidation. All of these data present an invitation to undertake more study of multiple brain structures and their interaction, which may prove to be crucial for our understanding of the relationship between brain activity and behavior.
acknowledgments We thank Patrick Tierney for his invaluable participation in the prefrontal recording experiments, Alain Berthoz for indispensable support throughout all stages of this work, F. Maloumian for figure preparation, S. Doutremer for histology, and Drs. A.-M. Thierry and J.-M. Deniau for helpful discussions. Thanks also go to the Fondation Fyssen for a fellowship to F.P.B., and European Community Integrated Projects ICEA, BACS, and Neuroprobes.
References Aggleton JP, Hunt PR, Rawlins JN (1986) The effects of hippocampal lesions upon spatial and non-spatial tests of working memory. Behav Brain Res 19:133–146. Albertin SV, Mulder AB, Tabuchi E, Zugaro MB, Wiener SI (2000) Lesions of the medial shell of the nucleus accumbens impair rats in finding larger rewards, but spare reward-seeking behavior. Behav Brain Res 117: 173–183. Alexander GE, Crutcher MD (1990) Functional architecture of basal ganglia circuits: neural substrates of parallel processing. Trends Neurosci 13:266–271. Amit DJ, Brunel N (1997) Model of global spontaneous activity and local structured activity during delay periods in the cerebral cortex. Cereb Cortex 7:237–252. Amzica F, Steriade M (1998) Cellular substrates and laminar profile of sleep K-complex. Neuroscience 82: 671–686. Arleo A, Gerstner W (2000) Spatial cognition and neuromimetic navigation: a model of hippocampal place cell activity. Biol Cybern 83:287–299. Baeg EH, Kim YB, Huh K, Mook-Jung I, Kim HT, Jung MW (2003) Dynamics of population code for working memory in the prefrontal cortex. Neuron 40:177–188. Banquet JP, Gaussier P, Quoy M, Revel A, Burnod Y (2005) A hierarchy of associations in hippocampocortical systems: cognitive maps and navigation strategies. Neural Comput 17:1339–1384. Barbas H (2000) Complementary roles of prefrontal cortical regions in cognition, memory, and emotion in primates. Adv Neurol 84:87–110. Barnes CA, Suster MS, Shen J, McNaughton BL (1997) Multistability of cognitive maps in the hippocampus of old rats. Nature 388:272–275.
PREFRONTAL CORTEX AND STRIATUM IN SPATIAL DECISIONS
Battaglia F, Khamassi M, Douchamps V, Tierney PL, Wiener SI (2005a) EEG correlations between hippocampus and prefrontal cortex in rats performing a decision making spatial task. Society for Neuroscience Abstracts, Washington DC. Battaglia FP, Khamassi M, Peyrache A, Douchamps V, Tierney PL, Wiener SI. (2006) Spatial and reward correlates in medial prefrontal neurons of rats acquiring and switching strategies in a Y-maze. Society for Neuroscience Abstracts, Washington DC. Battaglia FP, Sutherland GR, Cowen SL, Mc Naughton BL, Harris KD (2005b) Firing rate modulation: a simple statistical view of memory trace reactivation. Neural Netw 18:1280–1291. Battaglia FP, Sutherland GR, McNaughton BL (2004) Hippocampal sharp wave bursts coincide with neocortical ‘‘up-state’’ transitions. Learn Mem 11:697–704. Birrell JM, Brown VJ (2000) Medial frontal cortex mediates perceptual attentional set shifting in the rat. J Neurosci 20:4320–4324. Bontempi B, Laurent-Demir C, Destrade C, Jaffard R (1999) Time-dependent reorganization of brain circuitry underlying long-term memory storage. Nature 400:671–675. Burgess N, Recce ML, O’Keefe J (1994) A model of hippocampal function. Neural Netw 7:1065–1081. Burwell RD, Witter MP (2002) Basic anatomy of the parahippocampal region in monkeys and rats. In: The Parahippocampal Region (Witter MP, Wouterlood FG, eds.), pp 35–59. Oxford: Oxford University Press. Buzsa´ki G, Horvath Z, Urioste R, Hetke J, Wise K (1992) High-frequency network oscillation in the hippocampus. Science 256:1025–1027. Centonze D, Picconi B, Gubellini P, Bernardi G, Calabresi P (2001) Dopaminergic control of synaptic plasticity in the dorsal striatum. Eur J Neurosci 13:1071– 1077. Chrobak JJ, Buzsa´ki G (1994) Selective activation of deep layer (V-VI) retrohippocampal cortical neurons during hippocampal sharp waves in the behaving rat. J Neurosci 14:6160–6170. Chrobak JJ, Buzsa´ki G (1996) High-frequency oscillations in the output networks of the hippocampalentorhinal axis of the freely behaving rat. J Neurosci 16:3056–3066. Compte A, Brunel N, Goldman-Rakic PS, Wang XJ (2000) Synaptic mechanisms and network dynamics underlying spatial working memory in a cortical network model. Cereb Cortex 10:910–923. Cowan RL, Wilson CJ (1994) Spontaneous firing patterns and axonal projections of single corticostriatal neurons in the rat medial agranular cortex. J Neurophysiol 71:17–32. Csicsvari J, Hirase H, Mamiya A, Buzsa´ki G (2000) Ensemble patterns of hippocampal CA3-CA1 neu-
305
rons during sharp wave-associated population events. Neuron 28:585–594. Damasio AR (1999) The Feeling of What Happens: Body and Emotion in the Making of Consciousness. New York: Harcourt Brace & Co. Daw ND, Niv Y, Dayan P (2005) Uncertainty-based competition between prefrontal and dorsolateral striatal systems for behavioral control. Nat Neurosci 8:1704–1711. Dayan P, Balleine BW (2002) Reward, motivation, and reinforcement learning. Neuron 36:285–298. de Bruin JP, Swinkels WA, de Brabander JM (1997) Response learning of rats in a Morris water maze: involvement of the medical prefrontal cortex. Behav Brain Res 85:47–55. Delatour B, Gisquet-Verrier P (2000) Functional role of rat prelimbic-infralimbic cortices in spatial memory: evidence for their involvement in attention and behavioural flexibility. Behav Brain Res 109:113–128. Dickinson A (1985) Actions and habits: the development of behavioral autonomy. Philos Trans R Soc Lond B Biol Sci 308:67–78. Dudai Y (2004) The neurobiology of consolidations, or, how stable is the engram? Annu Rev Psychol 55:51–86. Ferbinteanu J, Kennedy PJ, Shapiro ML (2006) Episodic memory—from brain to mind. Hippocampus 16:691– 703. Ferino F, Thierry AM, Glowinski J (1987) Anatomical and electrophysiological evidence for a direct projection from Ammon’s horn to the medial prefrontal cortex in the rat. Exp Brain Res 65:421–426. Floresco SB, Seamans JK, Phillips AG (1997) Selective roles for hippocampal, prefrontal cortical, and ventral striatal circuits in radial-arm maze tasks with or without a delay. J Neurosci 17:1880–1890. Frank MJ, Loughry B, O’Reilly RC (2001) Interactions between frontal cortex and basal ganglia in working memory: a computational model. Cogn Affect Behav Neurosci 1:137–160. Frankland PW, Bontempi B (2005) The organization of recent and remote memories. Nat Rev Neurosci 6: 119–130. Funahashi S, Inoue M, Kubota K (1993) Delay-related activity in the primate prefrontal cortex during sequential reaching tasks with delay. Neurosci Res 18:171– 175. Fuster JM (2000) Memory networks in the prefrontal cortex. Prog Brain Res 122:309–316.Girard B, Filliat D, Meyer JA, Berthoz A, Guillot A (2005) Integration of navigation and action selection functionalities in a computational model of cortico-basal gangliathalamo-cortical loops. Adaptive Behavior 13:115– 130. Goldman-Rakic PS (1987) Circuitry of primate prefrontal cortex and regulation of behavior by representational
306
NEURAL SYSTEMS PERSPECTIVE ON PLACE FIELDS
memory. In: The Handbook of Physiology—The Nervous System V (Plum F, ed.). New York: Oxford University Press. Goldman-Rakic PS, Selemon LD, Schwartz ML (1984) Dual pathways connecting the dorsolateral prefrontal cortex with the hippocampal formation and parahippocampal cortex in the rhesus monkey. Neuroscience 12:719–743. Granon S, Poucet B (2000). Involvement of the rat prefrontal cortex in cognitive functions: a central role for the prelimbic area. Psychobiology 28, 219–228. Graybiel AM (1998) The basal ganglia and chunking of action repertoires. Neurobiol Learn Mem 70:119–136. Groenewegen HJ, Wright CI, Beijer AV (1996) The nucleus accumbens: gateway for limbic structures to reach the motor system? Prog Brain Res 107:485– 511. Gulley JM, Kosobud AE, Rebec GV (2002) Behaviorrelated modulation of substantia nigra pars reticulata neurons in rats performing a conditioned reinforcement task. Neuroscience 111:337–349. Haber SN, Fudge JL, McFarland NR (2000) Striatonigrostriatal pathways in primates form an ascending spiral from the shell to the dorsolateral striatum. J Neurosci 20:2369–2382. Hafting T, Fyhn M, Molden S, Moser MB, Moser EI (2005) Microstructure of a spatial map in the entorhinal cortex. Nature 436:801–806. Hahn TT, Sakmann B, Mehta MR (2006) Phase-locking of hippocampal interneurons’ membrane potential to neocortical up-down states. Nat Neurosci 9:1359–1361. Hoffman KL, McNaughton BL (2002) Coordinated reactivation of distributed memory traces in primate neocortex. Science 297:2070–2073. Hok V, Lenck-Santini PP, Roux S, Save E, Muller RU, Poucet B. (2007) Goal-related activity in hippocampal place cells. J Neurosci 27:472–482. Hok V, Save E, Lenck-Santini PP, Poucet B (2005) Coding for spatial goals in the prelimbic/infralimbic area of the rat frontal cortex. Proc Natl Acad Sci USA 102:4602–4607. Hollup SA, Molden S, Donnett JG, Moser MB, Moser EI (2001) Accumulation of hippocampal place fields at the goal location in an annular watermaze task. J Neurosci 21:1635–1644. Hopkins DA (2005) Neuroanatomy of head direction cell circuits. In: Head Direction Cells and the Neural Mechanisms of Spatial Orientation (Wiener SI, Taube JS, eds.). Cambridge, MA: MIT Press. Houk JC, Adams JL, Barto AG (1994) A model of how the basal ganglia generate and use neural signals that predict reinforcement. In: Models of Information Processing in the Basal Ganglia (Houk JC, Davis JL, Beiser DG, eds.). Cambridge, MA: MIT Press.
Ikemoto S (2002) Ventral striatal anatomy of locomotor activity induced by cocaine, D-amphetamine, dopamine and D1/D2 agonists. Neuroscience 113:939–955. Jackson J (2006) Network consistency and hippocampal dynamics: using the properties of cell assemblies to probe the hippocampal representation of space. Ph.D. thesis, University of Minnesota. Jay TM, Glowinski J, Thierry AM (1989) Selectivity of the hippocampal projection to the prelimbic area of the prefrontal cortex in the rat. Brain Res 505:337– 340. Jay TM, Witter MP (1991) Distribution of hippocampal CA1 and subicular efferents in the prefrontal cortex of the rat studied by means of anterograde transport of Phaseolus vulgaris-leucoagglutinin. J Comp Neurol 313:574–586. Ji D, Wilson MA (2007) Coordinated memory replay in the visual cortex and hippocampus during sleep. Nat Neurosci 10:100–107. Joel D, Weiner I (2000) The connections of the dopaminergic system with the striatum in rats and primates: an analysis with respect to the functional and compartmental organization of the striatum. Neuroscience 96:451–474. Jones MW, Wilson MA (2005) Theta rhythms coordinate hippocampal-prefrontal interactions in a spatial memory task. PLoS Biol 3:e402. Jung MW, Qin Y, McNaughton BL, Barnes CA (1998) Firing characteristics of deep layer neurons in prefrontal cortex in rats performing spatial working memory tasks. Cereb Cortex 8:437–450. Jung MW, Wiener SI, McNaughton BL (1994) Comparison of spatial firing characteristics of units in dorsal and ventral hippocampus of the rat. J Neurosci 14:7347–7356. Kesner RP (2000) Subregional analysis of mnemonic functions of the prefrontal cortex in the rat. Psychobiology 28:219–228. Knowlton BJ, Shapiro ML, Olton DS (1989) Hippocampal seizures disrupt working memory performance but not reference memory acquisition. Behav Neurosci 103:1144–1147. Korshunov VA, Wiener SI, Korshunova TA, Berthoz A (1996) Place- and behavior-independent sensory triggered discharges in rat hippocampal CA1 complex spike cells. Exp Brain Res 109:169–173. Kudrimoti HS, Barnes CA, McNaughton BL (1999) Reactivation of hippocampal cell assemblies: effects of behavioral state, experience, and EEG dynamics. J Neurosci 19:4090–4101. Laroche S, Jay TM, Thierry AM (1990) Long-term potentiation in the prefrontal cortex following stimulation of the hippocampal CA1/subicular region. Neurosci Lett 114:184–190.
PREFRONTAL CORTEX AND STRIATUM IN SPATIAL DECISIONS
Lee AK, Wilson MA (2002) Memory of sequential experience in the hippocampus during slow wave sleep. Neuron 36:1183–1194. Lee I, Griffin AL, Zilli EA, Eichenbaum H, Hasselmo ME (2006) Gradual translocation of spatial correlates of neuronal firing in the hippocampus toward prospective reward locations. Neuron 51:639–650. Leutgeb S, Leutgeb JK, Barnes CA, Moser EI, McNaughton BL, Moser MB (2005) Independent codes for spatial and episodic memory in hippocampal neuronal ensembles. Science 309:619–623. Maaswinkel H, Gispen WH, Spruijt BM (1996) Effects of an electrolytic lesion of the prelimbic area on anxietyrelated and cognitive tasks in the rat. Behav Brain Res 79:51–59. Markus EJ, Qin YL, Leonard B, Skaggs WE, McNaughton BL, Barnes CA (1995) Interactions between location and task affect the spatial and directional firing of hippocampal neurons. J Neurosci 15:7079–7094. Marr D (1970) A theory for cerebral neocortex. Proc R Soc Lond B Biol Sci 176:161–234. Marr D (1971) Simple memory: a theory for archicortex. Philos Trans R Soc Lond B Biol Sci 262:23–81. Maurer AP, Vanrhoads SR, Sutherland GR, Lipa P, McNaughton BL (2005) Self-motion and the origin of differential spatial scaling along the septo-temporal axis of the hippocampus. Hippocampus 15:841–852. McGeorge AJ, Faull RLM (1989) The organization of the projection from the cerebral cortex to the striatum in the rat. Neuroscience 29:503–538. McNaughton BL, Barnes CA, Battaglia FP, Bower MR, Cowen SL,, Ekstrom AD, Gerrard JL, Hoffman KL, Houston FP, Karten Y, Lipa P, Pennartz CM, Sutherland GR (2003) Off-line reprocessing of recent memory and its role in memory consolidation: a progress report. In: Sleep and Brain Plasticity (Maquet P, Stickgold B, Smith C, eds.), pp 225–246. Oxford: Oxford University Press. Miyachi S, Hikosaka O, Lu X (2002) Differential activation of monkey striatal neurons in the early and late stages of procedural learning. Exp Brain Res 146: 122–126. Mogenson GJ, Brudzynski SM, Wu M, Yang CR, Yim CY (1993) From motivation to action: a review of dopaminergic regulation of limbic-nucleus accumbensventral pallidum-pedunculopontine nucleus circuitries involved with limbic-motor integration. In: LimbicMotor Circuits and Neuropsychiatry (Kalivas PW, Barnes CD, eds.), pp 193–263. Boca Raton, FL: CRC Press. Moscovitch M, Winocur G (2002) The frontal cortex and working with memory. In: Principles of Frontal Lobe Function (Stuss DT, Knight RT, eds.). Oxford: Oxford University Press.
307
Mulder AB, Arts MP, Lopes da Silva FH (1997) Shortand long-term plasticity of the hippocampus to nucleus accumbens and prefrontal cortex pathways in the rat, in vivo. Eur J Neurosci 9:1603–1611. Ninokura Y, Mushiake H, Tanji J (2003) Representation of the temporal order of visual objects in the primate lateral prefrontal cortex. J Neurophysiol 89:2868–2873. Ninokura Y, Mushiake H, Tanji J (2004) Integration of temporal order and object information in the monkey lateral prefrontal cortex. J Neurophysiol 91:555–560. Niv Y, Daw, ND, Dayan P (2005) How fast to work: response vigor, motivation and tonic dopamine. In: Neural Information Processing Systems 18 (Weiss Y, Scho¨lkopf B, Platt J, eds.) pp 1019–1026. Cambridge, MA: MIT Press. Pandya DN, Yeterian EH (1990) Prefrontal cortex in relation to other cortical areas in rhesus monkey: architecture and connections. Prog Brain Res 85:63–94. Pasupathy A, Miller EK (2005) Different time courses of learning-related activity in the prefrontal cortex and striatum. Nature 433:873–876. Peigneux P, Laureys S, Fuchs S, Collette F, Perrin F, Reggers J, Phillips C, Degueldre C, Del Fiore G, Aerts J, Luxen A, Maquet P (2004) Are spatial memories strengthened in the human hippocampus during slow wave sleep? Neuron 44:535–545. Pennartz CM, Groenewegen HJ, Lopes da Silva FH (1994) The nucleus accumbens as a complex of functionally distinct neuronal ensembles: an integration of behavioural, electrophysiological and anatomical data. Prog Neurobiol 42:719–761. Pennartz CM, Lee E, Verheul J, Lipa P, Barnes CA, McNaughton BL (2004) The ventral striatum in offline processing: ensemble reactivation during sleep and modulation by hippocampal ripples. J Neurosci 24:6446–6456. Petersen CC, Hahn TT, Mehta M, Grinvald A, Sakmann B (2003) Interaction of sensory responses with spontaneous depolarization in layer 2/3 barrel cortex. Proc Natl Acad Sci USA 100:13638–13643. Postle BR (2006) Working memory as an emergent property of the mind and brain. Neuroscience 139:23–38. Poucet B, Thinus-Blanc C, Muller RU (1994) Place cells in the ventral hippocampus of rats. Neuroreport 5: 2045–2048. Pratt WE, Mizumori SJ (2001) Neurons in rat medial prefrontal cortex show anticipatory rate changes to predictable differential rewards in a spatial memory task. Behav Brain Res 123:165–183. Prescott TJ, Redgrave P, Gurney K (1999) Layered control architectures in robots and vertebrates. Adapt Behav 7:99–127. Ragozzino ME, Detrick S, Kesner RP (1999a) Involvement of the prelimbic-infralimbic areas of the rodent
308
NEURAL SYSTEMS PERSPECTIVE ON PLACE FIELDS
prefrontal cortex in behavioral flexibility for place and response learning. J Neurosci 19:4585–4594. Ragozzino ME, Wilcox C, Raso M, Kesner RP (1999b) Involvement of rodent prefrontal cortex subregions in strategy switching. Behav Neurosci 113:32–41. Redgrave P, Prescott TJ, Gurney K (1999) The basal ganglia: a vertebrate solution to the selection problem? Neuroscience 89:1009–1023. Reynolds JN, Hyland BI, Wickens JR (2001) A cellular mechanism of reward-related learning. Nature 413: 67–70. Rolls ET (2004) The functions of the orbitofrontal cortex. Brain Cogn 55:11–29. Ross RS, Eichenbaum H (2006) Dynamics of hippocampal and cortical activation during consolidation of a nonspatial memory. J Neurosci 26:4852–4859. Sanchez-Vives MV, McCormick DA (2000) Cellular and network mechanisms of rhythmic recurrent activity in neocortex. Nat Neurosci 3:1027–1034. Sargolini F, Fyhn M, Hafting T, McNaughton BL, Witter MP, Moser MB, Moser EI (2006) Conjunctive representation of position, direction, and velocity in entorhinal cortex. Science 312:758–762. Schultz W (1998) Predictive reward signal of dopamine neurons. J Neurophysiol 80:1–27. Schultz W, Dayan P, Montague PR (1997) A neural substrate of prediction and reward. Science 275:1593– 1599. Schultz W, Tremblay L, Hollerman JR (2000) Reward processing in primate orbitofrontal cortex and basal ganglia. Cereb Cortex 10:272–284. Shibata R, Mulder AB, Trullier O, Wiener SI (2001) Position sensitivity in phasically discharging nucleus accumbens neurons of rats alternating between tasks requiring complementary types of spatial cues. Neuroscience 108:391–411. Siapas AG, Wilson MA (1998) Coordinated interactions between hippocampal ripples and cortical spindles during slow-wave sleep. Neuron 21:1123–1128. Sirota A, Csicsvari J, Buhl D, Buzsaki G (2003) Communication between neocortex and hippocampus during sleep in rodents. Proc Natl Acad Sci USA 100: 2065–2069. Skaggs WE, McNaughton BL (1996) Replay of neuronal firing sequences in rat hippocampus during sleep following spatial experience. Science 271:1870– 1873. Steriade M, Buzsaki G (1990) Parallel activation of thalamic and cortical neurons by brainstem and basal forebrain cholinergic systems. In: Brain Cholinergic Systems (Steriade M, Biesold D, eds.), pp 3–64. Oxford: Oxford University Press. Steriade M, Nunez A, Amzica F (1993) A novel slow (<1 Hz) oscillation of neocortical neurons in vivo:
depolarizing and hyperpolarizing components. J Neurosci 13:3252–3265. Swanson LW (1981) A direct projection from Ammon’s horn to prefrontal cortex in the rat. Brain Res 217: 150–154. Tabuchi ET, Mulder AB, Wiener SI (2000) Position and behavioral modulation of synchronization of hippocampal and accumbens neuronal discharges in freely moving rats. Hippocampus 10:717–728. Tabuchi E, Mulder AB, Wiener SI (2003) Reward value invariant place responses and reward site associated activity in hippocampal neurons of behaving rats. Hippocampus 13:117–132. Thierry AM, Gioanni Y, Degenetais E, Glowinski J (2000) Hippocampo-prefrontal cortex pathway: anatomical and electrophysiological characteristics. Hippocampus 10:411–419. Tierney PL, Degenetais E, Thierry AM, Glowinski J, Gioanni Y (2004) Influence of the hippocampus on interneurons of the rat prefrontal cortex. Eur J Neurosci 20:514–524. Timofeev I, Grenier F, Steriade M (2001) Disfacilitation and active inhibition in the neocortex during the natural sleep-wake cycle: an intracellular study. Proc Natl Acad Sci USA 98:1924–1929. Trullier O, Wiener SI, Berthoz A, Meyer JA (1997) Biologically based artificial navigation systems: review and prospects. Prog Neurobiol 51:483–544. Trullier O, Shibata R, Mulder AB, Wiener SI (1999) Hippocampal neuronal position selectivity remains fixed to room cues only in rats alternating between place navigation and beacon approach tasks. Eur J Neurosci 11:4381–4388. Uylings HB, Groenewegen HJ, Kolb B (2003) Do rats have a prefrontal cortex? Behav Brain Res 146: 3–17. Vertes RP (2006) Interactions among the medial prefrontal cortex, hippocampus and midline thalamus in emotional and cognitive processing in the rat. Neuroscience 142:1–20. Verwer RW, Meijer RJ, Van Uum HF, Witter MP (1997) Collateral projections from the rat hippocampal formation to the lateral and medial prefrontal cortex. Hippocampus 7:397–402. Wan RQ, Pang K, Olton DS (1994) Hippocampal and amygdaloid involvement in nonspatial and spatial working memory in rats: effects of delay and interference. Behav Neurosci 108:866–882. Wang XJ (1999) Synaptic basis of cortical persistent activity: the importance of NMDA receptors to working memory. J Neurosci 19:9587–9603. Wiener SI (1993) Spatial and behavioral correlates of striatal neurons in rats performing a self-initiated navigation task. J Neurosci 13:3802–3817.
PREFRONTAL CORTEX AND STRIATUM IN SPATIAL DECISIONS
Wiener SI (1996) Spatial, behavioral and sensory correlates of hippocampal CA1 complex spike cell activity: Implications for information processing functions. Prog Neurobiol 49:335–361. Wiener SI, Khamassi M, Peyrache A, Douchamps V, Tierney P, Battaglia FP (2006) Transitions in behaviorally correlated activity in medial prefrontal neurons of rats acquiring and switching strategies in a Y-maze. Society for Neuroscience Abstracts, Atlanta, GA.
309
Wiener SI, Paul CA, Eichenbaum H (1989) Spatial and behavioral correlates of hippocampal neuronal activity. J Neurosci 9:2737–2763. Wilson MA, McNaughton BL (1994) Reactivation of hippocampal ensemble memories during sleep. Science 265:676–679. Winocur G (1992) A comparison of normal old rats and young adult rats with lesions to the hippocampus or prefrontal cortex on a test of matching-to-sample. Neuropsychologia 30:769–781.
This page intentionally left blank
IV THEORETICAL SIGNIFICANCE OF PLACE FIELDS
This page intentionally left blank
19 Hippocampal Theta Rhythm and Memory-Guided Behavior AMY L. GRIFFIN, HOWARD EICHENBAUM, AND MICHAEL E. HASSELMO
Along with place cells the hippocampal theta rhythm, thought to be an ‘‘online’’ state of hippocampus, is one of the most predominant and well-studied physiological patterns in the hippocampal literature. This chapter first outlines the major behavioral correlates of theta rhythm, particularly pertaining to learning and memory, followed by a discussion of the relationship between theta rhythm and the activity of individual neurons (including place cells) in hippocampus. Finally, we will review computational models that have been useful in linking learning and memory functions to theta rhythm and theta-related hippocampal unit firing with the goal of generating hypotheses for future experimental studies.
BEHAVIORAL CORRELATES OF THE THETA RHYTHM The hippocampal theta rhythm is a slow, largeamplitude oscillation often associated with attention to environmental stimuli and exploration (Green and Arduini, 1954; Vanderwolf, 1971). Green and Arduini (1954) were among the first to describe the theta rhythm and its behavioral correlates. They observed that arousal elicited by sensory or brain stem stimulation induced cortical desynchronization accompanied by rhythmic synchronous slow waves in the hippocampus. Later, Vanderwolf (1969, 1971) recorded hippocampal field potentials during exploratory activity in rats and concluded that theta rhythm appeared during what he called ‘‘voluntary’’ behaviors: locomotion, orienting, rearing, and sniffing. Interestingly, recent work shows that theta
oscillations in the human hippocampus increase during virtual navigation in much the same way as during real locomotion in rats (Ekstrom et al., 2005). The fact that theta appears during running and other types of motor output has led to many hypotheses about the role of the theta rhythm in integrating sensory and motor information (Bland and Oddie, 2001). However, theta appears in immobile rabbits during both aversive eyeblink conditioning and appetitive conditioning (Asaka et al., 2000; Berry and Seager, 2001; Seager et al., 2002; Griffin et al., 2004) and in rats and mice during fear conditioning (Whishaw, 1972; Sainsbury et al., 1987a; Seidenbecher et al., 2003), or attention to predators (Sainsbury et al., 1987b). This immobility-related theta is in a lower frequency band (3 –7 Hz) than movementrelated theta commonly seen in locomoting rats (8– 12 Hz) and is known to be sensitive to cholinergic manipulations (Kramis et al., 1975; Stewart and Fox, 1989a,b). The theta rhythm’s most intriguing behavioral correlate is its proposed role in learning and memory (Berry and Thompson, 1978; Winson, 1978; Givens and Olton, 1990; Vertes and Kocsis, 1997; Berry and Seager, 2001). Experiments that have made a convincing case for the direct correlation between theta and learning and memory are those that use a combination of lesion and recording techniques and behavior to demonstrate that manipulations that disrupt or eliminate theta cause learning and memory impairments. Theta rhythm is abolished by lesions of the medial septum (MS) or dorsal fornix (Green and Arduini, 1954; Rawlins et al., 1979). For this reason, the MS is
313
314 THEORETICAL SIGNIFICANCE OF PLACE FIELDS often referred to as the pacemaker of the theta rhythm. This idea is strengthened by the observation that MS neurons burst in phase with hippocampal theta (Petsche et al., 1962; Stewart and Fox, 1990). In a pioneering study, medial septal lesions caused learning impairments in a spatial memory task in rats that correlated with the reduction of the theta rhythm (Winson, 1978). Similarly, medial septal lesions slowed the learning rate of rabbit eyeblink conditioning and concomitantly reduced theta power and hippocampal conditioning-related unit responses (Berry and Thompson, 1978). While lesion techniques have been informative in correlating theta activity with behavior, pharmacological manipulation of the MS has revealed important details about the generation of theta and has given researchers more sophisticated tools for pinpointing the role of theta rhythm in behavior. The MS sends cholinergic and GABAergic projections to hippocampus, with cholinergic septohippocampal (SH) neurons innervating both pyramidal cells and inhibitory interneurons, and GABAergic SH neurons selectively innervating inhibitory interneurons (Freund and Antal, 1988). The somata of SH neurons are known to express both GABAergic and cholinergic receptors, with a larger density of cholinergic receptors on GABAergic SH neurons (Van der Zee and Luiten, 1994). Administration of cholinergic agents to the MS, therefore, should have a more dramatic effect on the GABAergic SH neurons than cholinergic SH neurons and thus would directly affect inhibitory interneurons in the hippocampus. Indeed, the cholinergic agonist muscarine was shown to selectively excite GABAergic but not cholinergic SH neurons (Wu et al., 2000). The importance of the SH projection in theta generation has been shown with injections of the neurotoxin IgG saporin into the MS, a procedure that has been established to selectively lesion the cholinergic SH neurons and to dose-dependently reduce theta power. It is important to note that these lesions leave theta frequency intact, a result suggesting that the SH cholinergic projection is important for enhancing the magnitude of the theta oscillation, whereas the GABAergic projection is important for generating rhythmic inhibition of hippocampal pyramidal cells, which gives rise to the theta oscillatory pattern (Lee et al., 1994). In one behavioral study, the selective cholinergic lesion produced dose-dependent working memory deficits in a delayed-nonmatch-to-sample (DNMS) task in an eight-arm radial maze. The behavioral deficits were accompanied by reductions in highaffinity choline uptake in the hippocampus, a measure of cholinergic function (Walsh et al., 1996). In a direct measurement of the effect of MS manipulation on both theta and working memory, choice accuracy in a continuous spatial alternation in a T-maze was reduced by
reversibly inactivating sodium channels by intraseptal infusions of tetracaine, by blocking GABAA receptors with muscimol and blocking acetylcholine (ACh) receptors with scopolamine. These drug manipulations also reduced theta power (Givens and Olton, 1990). It was still unclear if the disruption of theta interfered with encoding or retrieval. To answer this question, rats were given reversible septal lesions, which led to a dramatic reduction in theta power, before the sample phase of a DNMS task in a radial-arm maze. The lesion caused an increase of errors in the test phase. However, if the lesion was made after the sample phase, there was no effect on subsequent choice accuracy (Mizumori et al., 1990). These results suggest that theta activity is important during the encoding of new information. Similarly, intraseptal microinfusions of scopolamine suppressed theta activity and caused a corresponding deficit in acquisition of trace appetitive conditioning in rabbits (Asaka et al., 2000). Another more naturalistic approach used to explore the relationship between theta rhythm and learning is to examine the relationship between naturally occurring theta and learning. In an early study, Berry and Thompson (1978) recorded a 2-min sample of hippocampal EEG immediately prior to the beginning of a delay eyeblink conditioning procedure. Frequency analysis of these pretraining samples showed a range of predominance of theta oscillations, with some samples dominated by slow waves in the theta range and some samples showing a mixture of theta and higher frequencies. Interestingly, there was a high positive correlation (0.72) between learning rate and the index of hippocampal frequencies (nontheta/theta), a measure of the amount of theta in the hippocampal EEG. In addition to predicting behavioral acquisition rate, the amount of theta in the pretraining recording also predicted the degree of responsiveness of hippocampal neurons to the conditioned stimulus and unconditioned stimulus. Animals that learned quickly showed a large proportion of EEG waves in the theta range prior to conditioning and developed robust conditioned unit responses sooner than the slower learning cohorts (Berry, 1982). A subsequent study used the presence or absence of theta to directly manipulate learning rate. Hippocampal EEG was monitored ‘‘online’’ and the presence or absence of theta in the hippocampal EEG was detected by spectral analysis. For one group of rabbits, delay eyeblink conditioning trials were given only when theta epochs were detected. For a second group of rabbits, trials were given during nontheta states. The theta group learned significantly faster than the nontheta group (Seager et al., 2002). Similarly, in the hippocampus-dependent trace eyeblink conditioning paradigm, the theta group learned four times faster and also showed significantly larger conditioned unit
THETA RHYTHM AND MEMORY-GUIDED BEHAVIOR
responses than the nontheta group (Griffin et al., 2004). These results suggest that the presence of theta has a facilitory effect on hippocampal plasticity. There are a number of studies that have directly investigated the role of theta rhythm in synaptic plasticity. Long-term potentiation (LTP) was induced in the dentate gyrus when a tetanic stimulus was given on the positive phase of theta in urethane-anesthetized (Pavlides et al., 1988) and freely moving rats (Orr et al., 2001). Bidirectional modulation of hippocampal plasticity has also been demonstrated in CA1. First, in slice preparations, theta was induced by bath application of cholinergic agonists, stimulation delivered at the peak of theta caused LTP, and stimulation at the trough of theta caused long-term depression (LTD; (Huerta and Lisman, 1995). This same effect was seen in the intact animal under urethane anesthesia (Holscher et al., 1997). In the awake animal, LTP was induced by timing stimulation to coincide with the peak of theta, and LTD was induced by delivering stimulation during theta troughs (Hyman et al., 2003). Together these data support a computational model that proposes that encoding and retrieval occur at separate phases of the theta rhythm (Hasselmo et al., 2002b). These results suggest that induction of LTP in stratum radiatum occurs when transmission is weak but dendrites are depolarized by entorhinal input. LTP does not depend on spiking at the soma, as dendritic spikes can induce LTP even when the soma is hyperpolarized (Golding et al., 2002).
RELATIONSHIP BETWEEN HIPPOCAMPAL THETA RHYTHM AND HIPPOCAMPAL UNIT FIRING Place cell discharge and theta rhythm are intimately connected. At the transition from large irregular activity (LIA) to theta states, the in-field firing rate increases and the out-of-field firing rate decreases, which enhances the signal-to-noise ratio of place cell firing (Kubie et al., 1985). Also, ACh release increases twofold in hippocampus during theta compared to LIA (Dudar et al., 1979). Muscarinic blockade mimics the reduction in ACh release produced by switching from theta to LIA. This manipulation not only decreases the in-field firing rate but decreases the precision in the spatial representation by place cells(Brazhnik et al., 2003). Additionally, the flexibility of spatial representations (i.e., remapping after changes in the sensory cues in the testing environment) is decreased after septal lesioning (Leutgeb and Mizumori, 1999). Together, these results suggest that cholinergic modulation of hippocampal pyramidal cells during theta states enhances both the specificity and accuracy of the spatial representation.
315
In contrast to simple phase-locking, hippocampal place cells show the phenomenon of theta-phase precession, firing late in the theta cycle when a rat first enters the place field of the cell, and moving to earlier phases as the rat moves through the place field (O’Keefe and Recce, 1993; Skaggs et al., 1996; Mehta et al., 2002; Huxter et al., 2003). The phase relationship of each spike to the cycle of the hippocampal theta rhythm has been proposed as a temporal code that provides spatial information about an animal’s environment. There is evidence that the ‘‘rate code’’ (i.e., the firing rate of the place cell) is independent of this temporal code, with the firing rate signaling nonspatial information (i.e., running speed) and the phase of firing signaling the location of the rat within the place field (Huxter et al., 2003). Alternatively, another study found phase precession during both spatial and nonspatial behaviors and argued that phase precession is a result of the strength of dendritic depolarization(Harris et al., 2002). A more recent study examined the effects of transient hippocampal spike suppression on phase precession. Stimulation of the hippocampal commissure during traversals through a place field caused a brief cessation of neuronal firing in hippocampus. The phase precession, however, was not disrupted and commenced with the precession pattern after the stimulationinduced suppression of spiking (Zugaro et al., 2005). These results suggest that the phase of each spike carries information about the rat’s location within the place field and that this information is updated during each theta cycle by cortical input structures.
COMPUTATIONAL MODELS LINKING MEMORY-GUIDED BEHAVIOR TO HIPPOCAMPAL PHYSIOLOGY As shown in Figures 19–1 and 19–2, physiological data have demonstrated the specific changes in current sources and sinks associated with different phases of the theta-rhythm oscillations. A full understanding of the behavioral role of theta rhythm will require linking these cyclical physiological changes to behavioral function. Linking behavior to physiology requires explicit modeling of both the memory-guided actions of a rat during behavior and the physiological mechanisms underlying these actions. Network simulations of cortical structures have been used to guide the movements of a virtual rat in a virtual environment (Hasselmo et al., 2002a; Hasselmo et al., 2002c; Hasselmo, 2005; Hasselmo and Eichenbaum, 2005; Koene and Hasselmo, 2005), in tasks including spatial reversal (Hasselmo et al., 2002b), spatial alternation and linear tracks (Hasselmo and Eichenbaum, 2005), and delayed nonmatch to position (Hasselmo and Zilli,
316 THEORETICAL SIGNIFICANCE OF PLACE FIELDS
Figure 19–1. A. Anatomical diagram of CA1 showing synaptic inputs to region CA1 that contribute to thetarhythm oscillations in the local field potential. These include input from entorhinal cortex (EC) to stratum lacunosum-moleculare (s.1-m), input from region CA3 to stratum radiatum (s. rad), and input from medial septum to inhibitory interneurons (int) and pyramidal cells in stratum pyramidale (s. pyr). B. Example recording of two cycles of the theta-rhythm oscillation in the local field potential recorded at the hippocampal fissure. C. Current source density (CSD) plot adapted from Brankack, Stewart, and Fox (1993) showing rhythmic changes in current sinks due to synaptic input to different layers of region CA1. Darker colors indicate stronger inward currents. Top: summed depolarizing currents in region CA1 s.pyr that are associated with CA1 cell firing; middle: sinks caused by synaptic input from region CA3 to s.rad, which are strong near the time of cell firing; bottom: sinks caused by synaptic input from ECIII to s.1-m., which are strongly out of phase with cell firing. D. Phases of theta-rhythmic activity computed as optimal in a model of associative memory function with long-term potentiation (Has-
2005). The behavior in these modeled tasks has been guided by network simulations that simulate features of the physiological data. As shown in Figure 19–3, the structure of these simulations is based on the anatomy and physiology of the hippocampus, entorhinal cortex, and prefrontal cortex. As the virtual rat moves through the task, information about its state in the environment (place) and its receipt of food reward (reward) is sent from the virtual rat to the neural simulation. In the model, the prefrontal cortex performs goal-directed selection of next action (motor output) (Hasselmo, 2005; Koene and Hasselmo, 2005), performing functions similar to reinforcement learning algorithms (Sutton and Barto, 1998). Action selection depends on both the current state and episodic retrieval of previous responses from circuits representing the hippocampus and entorhinal cortex (Hasselmo and Eichenbaum, 2005). These models extend previous work that used network simulations of the hippocampal formation to guide the movements of a virtual rat in spatial tasks (Sharp et al., 1996; Burgess et al., 1997; Redish and Touretzky, 1998), but focuses on a more detailed representation of dynamical changes during thetarhythm oscillations. The series of models described here have progressed through a number of stages. Earlier models used associations between place cells in the hippocampus and entorhinal cortex to store pathways and select between possible pathways (Hasselmo et al., 2002c), similar to the hypothesis that Hebbian modification between place cells could provide a distance metric (Muller and Stead, 1996). These models used a backward spread of activity from goal location in entorhinal cortex layer III to guide action selection. The pattern of backward spread from goal location in those models resembles the backward spread of activity seen in the hippocampus in recent unit recording studies (Foster and Wilson, 2006). The same pattern of backward spread from goal locations has also been used in later models of action selection proposed to
selmo et al., 2002a,b;(Hasselmo, 2003a); Judge and Hasselmo, 2004). Each band shows the rhythmic synaptic activity across two cycles of modeled theta activity plotted horizontally. Top: dark shading indicates depolarizing currents in cell bodies; middle: dark shading indicates excitatory synaptic input from region CA3; bottom: dark shading indicates excitatory synaptic input from EC layer III. Note that best function in the model occurs with phase relationships matching CSD, including the slight offset between cell body depolarization and region CA3 input.
THETA RHYTHM AND MEMORY-GUIDED BEHAVIOR
317
Figure 19–2. Theoretical separation of encoding and retrieval during theta rhythm (Hasselmo et al., 2002b). Left: The encoding phase occurs at the trough of fissure theta, when synaptic currents arising from entorhinal cortex (EC) are strong (Brankack et al., 1993), due to greater depolarization and greater spread of activity in of EC layer III. During this phase, transmission from CA3 is weak (Wyble et al., 2000), preventing retrieval of previously encoded information, but long-term potentiation (LTP) in these synapses is very strong (Holscher et al., 1997; Hyman et al., 2003), which allows encoding of associations between EC inputs. Right: The retrieval phase is at the peak of fissure theta, when synaptic currents arising from EC are weak, but synaptic currents arising from CA3 are strong (Brankack et al., 1993), allowing effective retrieval of previously encoded sequences. During this phase synapses do not reflect the retrieval because they do not undergo LTP, instead they undergo long-term depression (LTD) or depotentiation.
occur in prefrontal cortical circuits (Hasselmo, 2005; Koene and Hasselmo, 2005). This backward spread is related to more simplified representations of action selection using the algorithms of reinforcement learning theory. The model described in most of the figures in this chapter performs action selection in prefrontal cortex based on encoding and retrieval of episodic sequences using the primary subregions of the hippocampal formation, including entorhinal cortex layer II, entorhinal cortex layer III, the dentate gyrus, and regions CA3 and CA1 of the hippocampus (Hasselmo and Eichenbaum, 2005). This network encodes episodes consisting of sequences of visits to different states (places) in the environment. Thus, encoding is based on sequential activation of ‘‘place cells’’ in the hippocampus (O’Keefe and Dostrovsky, 1971; O’Keefe, 1976; McNaughton et al., 1983; Muller et al., 1987; Eichenbaum et al., 1989; Muller and Kubie, 1989; Wiener et al., 1989; Skaggs et al., 1996; Wood et al., 2000; Huxter et al., 2003) and entorhinal cortex
(Barnes et al., 1990; Frank et al., 2000). This framework has been used to model encoding and retrieval of sequences in prior work as well. The construction of these models demonstrates the potential functional requirements for theta rhythm to time activity in multiple different regions during the encoding and retrieval required for performing memory-guided behavior. One of the problems encountered in linking the hippocampus to memory function is that it is often unclear if hippocampal neuronal spiking is associated with the encoding of a new memory or the retrieval of an existing memory. A more general question is: how does the hippocampus encode new information without interference from old memories? Recent computational models use changing network dynamics during separate phases of the theta rhythm to address this question (Hasselmo et al., 2002b). As seen in Figures 19–1 and 19–2, encoding is proposed to occur at the trough of fissure theta, which is associated with strong input from the entorhinal cortex (Brankack et al., 1993). During this encoding phase, synaptic transmission is
318 THEORETICAL SIGNIFICANCE OF PLACE FIELDS
Figure 19–3. A. Schematic overview of the neuralsimulation guiding actions of a virtual rat in a virtual spatial-alternation task. The simulation receives input about location and proximity to food reward. The modeled prefrontal cortex performs action selection on the basis of selective sequential retrieval in the hippocampal formation of the previous turn response in the task. Sequential retrieval involves forward associations in entorhinal cortex layer III (ECIII) that drive activity in region CA1, which is gated by the temporal context arriving from EC layer II via region CA3. This causes selective retrieval of the most recent episode, which allows correct memory-guided selection of actions by the prefrontal cortex. B. Behavioral context showing two trial types: B1: post R—return from the right arm going left, and B2: post L—return from left arm going right. C. Simulation of ‘‘splitter cell’’ response. This example splitter cell will fire as part of sequences retrieved in the stem after the virtual rat performs a right-turn response (left side), but will not fire after a left-turn response (right side). Color coding represents number of spikes fired by simulations when virtual rat is in particular locations (more saturated colors indicate higher firing rate). Here encoding is assumed to be induced with dendritic spiking only.
weak between neurons in CA3 and between CA3 and CA1, which prevents interference. Conversely, retrieval occurs near the peak of fissure theta when synaptic input from entorhinal cortex is weak, but there is strong synaptic input within CA3 and from CA3 to CA1 (Brankack et al., 1993). Mathematical analysis of network simulations of associative memory function in the hippocampus show the phases necessary for best function of the model. When both LTP and LTD are included, the phase of synaptic input from CA3 should match the phase of greatest depolarization in region CA1 cell bodies, and be 1808 out of phase with entorhinal cortex layer III input (Hasselmo et al., 2002b). When only LTP is allowed, then the phase of CA1 depolarization must overlap more with entorhinal cortex layer III input, and the CA3 input must correspondingly be offset to a different phase from CA1 depolarization (Hasselmo et al., 2002c; Hasselmo, 2003a; Judge and Hasselmo, 2004). These phase differences are shown in Figure 19–1D, which shows how they compare with the actual phase offsets observed in the experimental current source density data. This framework can account for behavioral data showing that fornix lesions (which reduce theta rhythm) cause an increase in the number of errors after reversal in a T-maze task (M’Harzi et al., 1987). Specifically, rats with fornix lesions persist in visiting an arm that was previously rewarded but is currently unrewarded. This impairment could result from the loss of theta rhythm, allowing the induction of LTP and synaptic transmission in stratum radiatum to be strong at the same time. After reversal, the rat makes erroneous visits to the previously rewarded location. In this case, strong synaptic transmission allows the rat to retrieve postsynaptic activity corresponding to memory of food at the now unrewarded location. This retrieval activity could cause further LTP, thus strengthening associations with the memory of food despite the fact that the location is now unrewarded. This mechanism could slow the extinction of the old association and increase the period of error generation before reversal (Hasselmo et al., 2002b). The behavioral deficits following fornix lesions might also result from loss of slow modulatory effects of ACh, but appear to depend on combined block of both cholinergic and GABAergic input (Pang et al., 2001). Another interference problem is that we are often required to recall specific sequences of events that may overlap with other highly similar sequences. The physiological changes during the theta rhythm may enhance the selective context-dependent retrieval of individual sequences without interference from other sequences (Sohal and Hasselmo, 1998a,b; Hasselmo and Eichenbaum, 2005). This hypothesis builds from earlier models of sequence encoding in hippocampal
THETA RHYTHM AND MEMORY-GUIDED BEHAVIOR
circuits. Marr (1971) initially proposed that excitatory recurrent connections in region CA3 could provide sequential associations between individual patterns in a sequence, and this mechanism has been used in many models (McNaughton and Morris, 1987; Blum and Abbott, 1996; Jensen and Lisman, 1996b; Levy, 1996; Wallenstein and Hasselmo, 1997; Lisman, 1999; Hasselmo and Eichenbaum, 2005). During encoding in these models, each pattern in a sequence activates a set of neurons shortly before the next pattern, and spike timing–dependent plasticity (STDP) (Levy and Steward, 1983) strengthens synapses between the sequential patterns. During retrieval, input of the first pattern will cause activity to spread across strengthened synapses to cause sequential spiking in other patterns, reading out the full sequence (Levy, 1996; Wallenstein and Hasselmo, 1997). This simple sequence retrieval does not have to occur in region CA3. It could occur in any network in which retrieval output can cue another retrieval step. Thus, the same mechanism could occur at recurrent synapses in entorhinal cortex layer II or III (Hasselmo et al., 2002c; Hasselmo and Eichenbaum, 2005), in a loop involving dentate gyrus, region CA3, and mossy cells in the hilus (Lisman, 1999), or in a loop involving the full hippocampal circuit from entorhinal cortex back to entorhinal cortex. In addition, evidence suggests that whereas the hippocampus may encode episodic sequences, the basal ganglia may encode highly familiar sequences (White and McDonald, 2002). Simple sequence-encoding models cannot encode and selectively retrieve highly overlapping sequences (Levy, 1996). Selective retrieval requires some additional mechanism for context-dependent retrieval of one out of many highly overlapping sequences, through gating of retrieval output by additional synaptic activity. In one network model of CA3, disambiguation of overlapping sequences has been obtained by having retrieval of the end of a sequence depend on synaptic input from persistent firing of additional CA3 neurons termed local context units (Levy, 1996). As an alternative, context-dependent retrieval could also be obtained by gating the output of retrieval with input from another region. For example, the retrieval of region CA3 could be gated by activity from entorhinal cortex layer II, or output from CA3 to CA1 could be gated by activity from entorhinal cortex layer III. In the model presented here, output from entorhinal cortex layer III to CA1 is gated by context activity in the dentate gyrus and region CA3. These mechanisms of context-dependent retrieval require a balance between the forward sequence retrieval and gating by context. Theta rhythm could provide a mechanism for sampling across different magnitudes of network variables. Once sequence re-
319
trieval activity occurs, feedback inhibition can ensure that the first, best-matching sequence is selectively retrieved. Early models of this process used phasic changes in magnitude of synaptic transmission to allow retrieval of single associations due to a global context signal representing specific environmental cues selective for one episode (Sohal and Hasselmo, 1998a,b). The simulations of episodic retrieval reviewed here provide a more detailed model of the role of theta rhythm in allowing global context to regulate selective retrieval (Hasselmo and Eichenbaum, 2005). Figure 19–4 summarizes the mechanism for context-dependent retrieval in a spatial alternation task in the most recent model (Hasselmo and Eichenbaum,
Figure 19–4. Example of context-dependent retrieval for guidance of the virtual rat in a spatial alternation task (here the task is represented with rectangular return arms). Darker-shaded rectangles indicate stronger activity. A. Activity in entorhinal cortex layer III (ECIII) retrieves both a left- and a right-turn response. B. ECII holds activity representing temporal context, which is stronger for more recently visited locations. This activity spreads through dentate gyrus (DG) and CA3 to CA1. C. CA1 activity is driven by the multiplicative interaction between ECIII and CA3 inputs. For this particular example, the ECIII activity in both directions interacts with the stronger temporal context for the right side, allowing retrieval of the sequence representing the prior right-turn response. This guides the rat to make a left turn in order to receive reward.
320 THEORETICAL SIGNIFICANCE OF PLACE FIELDS 2005). On each trial in this task, the rat must retrieve memory of its previous response, in order to generate a response to the opposite arm of the maze. Based on previous behavior, the rat has a memory of sequences going both left and right from the choice point. As shown in Figure 19–4A, in the component of the model representing entorhinal cortex layer III, activity spreads across modified connections to retrieve sequences in both directions. The context-dependent retrieval of a single episode is obtained by gating the retrieval of forward associations with activity representing temporal context. As shown in Figure 19–4B, the model has a representation of layer II of entorhinal cortex that contains gradually decaying persistent activity representing previous states encountered in the task. This provides a global temporal context in which the representation of an individual location grows weaker with greater time since the visit to that location. This resembles the temporal context used in models of human episodic memory (Howard and Kahana, 2002; Howard et al., 2004). As shown in Figure 19–4B, this temporal context is stronger for the more recently visited arm. This temporal context signal spreads through the dentate gyrus and region CA3 to converge in region CA1 with the forward retrieval of both sequences from layer III of entorhinal cortex. The multiplicative interaction of these two inputs allows selective retrieval of the most recent sequence, because the forward retrieval going to the right interacts with the strong temporal context representation of the right arm to cause spiking activity. Thresholding of the activity in region CA1 results in selective retrieval of only the most recently encountered sequence, as shown in Figure 19–4C. As shown in Figure 19–3C, the model also simulates the phenomenon of some ‘‘splitter cells,’’ shown in experimental data, which fire on the stem only for a specific trial (left versus right) even though all the cues are the same (Wood et al., 2000). This effect appears in the virtual rat because of selective retrieval on the stem of only the most recently performed sequence. In the model this occurs for neurons that encode a location in one arm (e.g., the right arm). When sequence retrieval of the most recent episode occurs on the stem, the sequence will activate these neurons selectively, for example, after a right-turn response, but not after a left-turn response (Hasselmo and Eichenbaum, 2005). This process must deal with the problem that memories must be retrieved over a range of different temporal delays. The hippocampal theta rhythm provides a solution to this problem by allowing scanning for the first good match by phasically increasing context input due to changes in postsynaptic depolarization (Fox, 1989) or strength of excitatory synaptic
transmission (Wyble et al., 2000). The activity would then be read out by a multiplicative interaction of increasing entorhinal input and decreasing CA3 input, resulting in sequential retrieval that is equivalent in magnitude for each element of a sequence, but different in magnitude for different sequences, and strongest for the sequence best matching the current context (Hasselmo and Eichenbaum, 2005). In addition to the ‘‘splitter cell’’ phenomenon, the model presented here also addresses the phenomenon of theta-phase precession (O’Keefe and Recce, 1993; Skaggs et al., 1996). Previously, retrieval of sequences was used to model the phenomenon of theta-phase precession (Jensen and Lisman, 1996b; Tsodyks et al., 1996; Wallenstein and Hasselmo, 1997). In these models, entry to location 1 causes the readout of locations 2–3–4–5. As shown in Figure 19–5, if one observes the response of a single cell (coding for location 5), it will initially occur late in theta at the end of the readout sequence, and as the rat moves through the locations 2, 3, and 4, it will move to earlier phases until it is driven by sensory input at the start of the cycle. Models shown in Figure 19–5A require that input for the start of the sequence occur only at one phase of the theta cycle. In contrast, the model of the contextdependent retrieval process can keep the input cue present during the full theta cycle (Hasselmo and Eichenbaum, 2005), as shown in Figure 19–5B. The phase of firing of region CA1 neurons in the new model depends on the relative strength of forward association retrieval (Fig. 19–5B1) and the backward temporal-context retrieval (Fig. 19–5B2) at different phases of theta, as shown in Figure 19–5B and Figure 19–4. In addition, the forward retrieval in entorhinal cortex only occurs during half of the theta cycle. During earlier phases, entorhinal cortex responds primarily to afferent input, resulting in a firing response for current location across a broad range of phases (appearing as an increase in phase variance in later portions of the place field). The work modeling both the splitter-cell phenomenon and the phenomenon of theta-phase precession automatically resulted in an exciting capability of the model to explain a puzzling feature of unit-recording data, which other models do not effectively address. As shown in Figure 19–6, the model also generates context-dependent properties of theta-phase precession. In experimental data, theta-phase precession is very weak on the first trial of a testing day and becomes more prominent over subsequent trials on each day (Mehta et al., 2002). Previous models of this phenomenon assumed that the weakening at the start of each day would have to require LTD of synapses each night (Mehta, 2004). However, in our simulations, the absence of theta-phase precession results from the
THETA RHYTHM AND MEMORY-GUIDED BEHAVIOR
321
Figure 19–5. A. Simple model of phase precession based on sequence retrieval (Jensen and Lisman, 1996a; Tsodyks et al., 1996). A1. The upper plots show the retrieval of sequences of place cell activity over time at each new place in the environment. At place 1, places 2, 3, 4, and 5 are retrieved. At place 3, places 4, 5, 6, and 7 are retrieved. Note that the model only functions if the input (I) is present only at a single early phase. A2. When observing a single cell (5), firing initially appears late in the cycle, and then moves to earlier phases of theta. B. Newer model of phase precession based on contextdependent retrieval of sequences (Hasselmo and Eichenbaum, 2005). B1. Forward retrieval in entorhinal cortex layer III (ECIII) that spreads to more neurons as theta phase increases. Note that input is present at all phases of theta, but the length of forward spread increases during later phases. B2. Temporal context input from ECII via CA3. Temporal context is stronger for more recent locations (to right), and weaker for more distant places (to left). For locations to the left of the dotted line, temporal context is too weak and CA1 activity falls below threshold. This input also decreases over phases within each cycle, causing activity to fall below threshold for more locations. B3. CA1 activity depends on the multiplicative gating of EC input by temporal context from CA3. The input cue is present in EC and dentate gyrus during the full cycle, but only results in retrieval activity in region CA1 when EC input converges with strong temporal context. B4. When observing a single cell (5), CA1 activity appears as wide, slightly scalloped distribution of spiking activity over phase and position similar to that seen in experimental data (Skaggs et al., 1996).
absence of temporal context on the first run of the day. Once the virtual rat has run all the way around the track, then temporal context is present throughout the track and the model shows context-dependent retrieval of the sequence, which shows up as theta-phase precession (Hasselmo and Eichenbaum, 2005). Thus, the model effectively accounts for the weakness of thetaphase precession on the first trial relative to later trials. This same mechanism in the model also effectively simulates the backward expansion of place fields that is observed on each day of recording (Mehta et al., 1997). Behavioral transitions in the environment occur over a time course of several hundred milliseconds or
even seconds. For example, a rat may take several seconds to complete a full trial in a behavioral task. The time course of behavior is much slower than the time intervals important for spike timing–dependent synaptic plasticity in the hippocampus, which results in strengthened synapses only when a presynaptic spike precedes a postsynaptic spike by less than about 40 ms (Levy and Steward, 1983; Bi and Poo, 1998). A rat does not move very far in less than 40 ms, so the input from the environment does not change very much, raising the question of how a rat could form associations between neurons spiking at much slower intervals during sequential visits over longer periods in a task.
322 THEORETICAL SIGNIFICANCE OF PLACE FIELDS
Figure 19–6. Theta-phase precession of hippocampal place cells. A1. Experimental data (Skaggs et al., 1996) shows that the theta phase of firing (y-axis) moves to earlier phases as the rat moves from left to right (x-axis). A2. The simulation shows the same pattern for activation of a single place cell in different locations as the virtual rat moves in one direction around a rectangular track. Early firing for a place field occurs when it is retrieved as a later portion of the retrieved sequence at late phases of theta. Late firing in the place field occurs when afferent input drives spiking activity at early phases of theta. B1. Experimental data (Mehta et al., 2002) showing that thetaphase precession is not strong on the first pass through a location on a given day (top), but is stronger on later passes through the same location (bottom). B2. Simulation showing that absence of temporal context on first pass results in absence of phase precession on the first pass (top). On later passes (bottom), temporal context results in increased theta-phase precession. C. Schematic illustration of phase precession. As the rat enters the place field of a cell, the cell fires at late phases of theta. Firing moves to earlier phases as the rat moves through the place field.
In order to hold information about prior location over hundreds of milliseconds or longer, encoding of sequences in the model uses buffering of input activity based on the Lisman-Jensen model (Lisman and Idiart, 1995; Jensen and Lisman, 1996a). Simulations have shown how buffering could result from input eliciting sustained spiking activity in the entorhinal cortex (Franse´n et al., 2002; Hasselmo et al., 2002a; Koene et al., 2003; Koene and Hasselmo, 2007), where cholinergic modulation activates after-depolarization currents (Klink and Alonso, 1997; Egorov et al., 2002; Franse´n et al., 2006) that allow rhythmic reactivation of the elements of the input sequence. Thus, theta rhythm in entorhinal cortex of these models serves to time the reactivation and updating of the working memory buffer. This buffer maintains place representations for a period of time sufficient for synapses to be modified between place cells activated by adjacent locations, forming the basis for episodic memory of specific sequences traversed through the environment. Lesions that impair hippocampal theta rhythm do not prevent general goal-directed behavior, but only specifically the memory-guided aspects of behavior. For example, in spatial reversal, fornix lesions do not impair learning of the initial reward location, but impairs learning of the new reward location. Thus, general processes of action selection must be outside of the hippocampus, but should interact strongly with hippocampal function. The prefrontal cortex component of the model provides a link to data on the association of theta rhythm with voluntary movement (Bland and Oddie, 2001). The model of prefrontal cortex requires separate phases for the encoding of new associations between states and actions and for the spread of activity from the goal through previous associations during retrieval (Hasselmo, 2005; Koene and Hasselmo, 2005). In the spatial alternation task, strengthening of synaptic connections in the prefrontal cortex forms associations between episodic memories retrieved by the hippocampus and the motor plans for specific actions. For example, a population of neurons in the prefrontal cortex responds to hippocampal activity for the memory of a previous left turn; this activity then spreads across strengthened synapses to activate prefrontal neurons representing a right-turn response, which then activates the appropriate output. These interactions of hippocampus and prefrontal cortex require phasic timing of the input from hippocampus to prefrontal cortex. These timing requirements could be provided by the phasic timing of prefrontal cortex firing relative to the phase of theta rhythm in the hippocampus (Manns et al., 2000a,b; Siapas et al., 2000; Hyman et al., 2002; Hyman and Hasselmo, 2004; Hyman et al., 2005; Jones and Wilson, 2005; Siapas et al., 2005).
THETA RHYTHM AND MEMORY-GUIDED BEHAVIOR
EXPERIMENTAL EVIDENCE CONSISTENT WITH COMPUTATIONAL MODELS AND PREDICTIONS FOR FUTURE EXPERIMENTS The computational models presented here have generated a number of new predictions about the timing of spikes relative to hippocampal theta rhythm in behavioral tasks. Spikes that signify item or event encoding should occur on separate phases of the theta rhythm from those of spikes that signify later retrieval of those items. In line with the model predictions, a recent study found systematic differences in the preferred phase of firing for individual hippocampal neurons as rats inspected items that were familiar (i.e., conditions that favored retrieval of previous encounters with the item) and items that were new (i.e., conditions that favored encoding of the new item) (Manns et al., 2007). These results are consistent with the Hasselmo model (Hasselmo et al., 2002b; Zilli and Hasselmo, 2006) and suggest that spikes that occur when there is strong input from entorhinal cortex to CA3 and CA1 are associated with encoding, and spikes that occur when input from CA3 to CA1 is strong are associated with retrieval. Experiments currently under way in the Hasselmo and Eichenbaum laboratories are testing the prediction that splitter-cell responses occur during the retrieval phase of theta-rhythm oscillations (Griffin et al., 2005). Thus, they should occur during the late phases of theta during which precession occurs. In contrast, splitter cells should reflect activity encoded in one arm of the maze on the preceding trial, thus encoding activity should occur during the early encoding phase of theta rhythm. This same prediction applies to a delayed nonmatch-to-position task, in which firing on the sample trial (forced choice of one arm) should occur during the encoding phase, whereas firing on the test trial (both arms open for choice) should occur during the retrieval phase of theta rhythm. Simulations also demonstrate an interesting need to separate sequences of rewarded behavior into separate epochs. This separation could be provided by the suppression of theta rhythm during the receipt of reward, which has been described in both operant and spatial tasks (Wyble et al., 2004). Results from recent experiments indicate that the sequential retrieval of episodes can be supplemented by mechanisms for performance of spatial alternation that depend on working memory of the most recent response (Lee et al., 2006). Specifically, a rat could perform the spatial alternation task by maintaining spiking activity representing the current return arm in order to guide subsequent choice at the choice point. This persistent activity could be provided by cellular mechanisms for persistent firing shown in entorhinal
323
cortex (Egorov et al., 2002; Franse´n et al., 2006). Thus, a rat might hold activity representing the right return arm while it runs on the stem and this activity could guide selection of a left reward arm response at the choice point. Then it might hold activity representing the left return arm in order to guide a right-turn response. The persistence of activity could cause formation of associations between later sensory input and the effect of persistent activity on region CA1 neuronal firing. This would result in later input gradually dominating in causing the spiking of neurons that represent earlier locations (e.g., the return arm). This mechanism could underlie the apparent forward shift in place cell firing in recent recordings in the spatial alternation task (Lee et al., 2006). The mechanism of theta-phase precession and splitter cells proposed here should also be relevant to the encoding and retrieval of sequences of odor stimuli. In a separate project, recording during an odor sequence disambiguation task (Agster et al., 2002) tests whether individual cells show theta-phase precession of odor responses, in which the phase of firing of a neuron moves to earlier phases of theta as the sequence progresses (activity is plotted versus odor number, rather than position in a place field). In addition, this same task can be used to test for splitter-cell phenomena during the overlapping component of the odor sequence, analogous to splitter-cell phenomena during the overlapping spatial component (the stem) of the spatial alternation task. In a delayed matching task, responses to odors should occur on different phases during the sample period and test periods, or during match trials (which involve a previously encountered odor) versus nonmatch trials. The generation of such predictions provides an opportunity to test and modify the models to address the full range of experimental data.
References Agster KL, Fortin NJ, Eichenbaum H (2002) The hippocampus and disambiguation of overlapping sequences. J Neurosci 22:5760–5768. Asaka Y, Seager MA, Griffin AL, Berry SD (2000) Medial septal microinfusion of scopolamine disrupts hippocampal activity and trace jaw movement conditioning. Behav Neurosci 114:1068–1077. Barnes CA, McNaughton BL, Mizumori SJ, Leonard BW, Lin LH (1990) Comparison of spatial and temporal characteristics of neuronal activity in sequential stages of hippocampal processing. Prog Brain Res 83:287–300. Berry SD (1982) Septo-hippocampal activity and learning rate. In: Conditioning: Representation of Involved Neural Functions (Woody CD, ed.), pp 417–431. New York: Plenum Press.
324 THEORETICAL SIGNIFICANCE OF PLACE FIELDS Berry SD, Seager MA (2001) Hippocampal theta oscillations and classical conditioning. Neurobiol Learn Mem 76:298–313. Berry SD, Thompson RF (1978) Prediction of learning rate from the hippocampal electroencephalogram. Science 200:1298–1300. Bi GQ, Poo MM (1998) Synaptic modifications in cultured hippocampal neurons: dependence on spike timing, synaptic strength, and postsynaptic cell type. J Neurosci 18:10464–10472. Bland BH, Oddie SD (2001) Theta band oscillation and synchrony in the hippocampal formation and associated structures: the case for its role in sensorimotor integration. Behav Brain Res 127:119–136. Blum KI, Abbott LF (1996) A model of spatial map formation in the hippocampus of the rat. Neural Comput 8:85–93. Brankack J, Stewart M, Fox SE (1993) Current source density analysis of the hippocampal theta rhythm: associated sustained potentials and candidate synaptic generators. Brain Res 615:310–327. Brazhnik ES, Muller RU, Fox SE (2003) Muscarinic blockade slows and degrades the location-specific firing of hippocampal pyramidal cells. J Neurosci 23: 611–621. Burgess N, Donnett JG, Jeffery KJ, O’Keefe J (1997) Robotic and neuronal simulation of the hippocampus and rat navigation. Philos Trans R Soc Lond B Biol Sci 352:1535–1543. Dudar JD, Whishaw IQ, Szerb JC (1979) Release of acetylcholine from the hippocampus of freely moving rats during sensory stimulation and running. Neuropharmacology 18:673–678. Egorov AV, Hamam BN, Fransen E, Hasselmo ME, Alonso AA (2002) Graded persistent activity in entorhinal cortex neurons. Nature 420:173–178. Eichenbaum H, Wiener SI, Shapiro ML, Cohen NJ (1989) The organization of spatial coding in the hippocampus: a study of neural ensemble activity. J Neurosci 9:2764–2775. Ekstrom AD, Caplan JB, Ho E, Shattuck K, Fried I, Kahana MJ (2005) Human hippocampal theta activity during virtual navigation. Hippocampus 15:881–889. Foster DJ, Wilson MA (2006) Reverse replay of behavioural sequences in hippocampal place cells during the awake state. Nature 440:680–683. Fox SE (1989) Membrane potential and impedance changes in hippocampal pyramidal cells during theta rhythm. Exp Brain Res 77:283–294. Frank LM, Brown EN, Wilson M (2000) Trajectory encoding in the hippocampus and entorhinal cortex. Neuron 27:169–178. Franse´n E, Alonso AA, Hasselmo ME (2002) Simulations of the role of the muscarinic-activated calciumsensitive nonspecific cation current INCM in en-
torhinal neuronal activity during delayed matching tasks. J Neurosci 22:1081–1097. Franse´n E, Tahvildari B, Egorov AV, Hasselmo ME, Alonso AA (2006) Mechanism of graded persistent cellular activity of entorhinal cortex layer v neurons. Neuron 49:735–746. Freund TF, Antal M (1988) GABA-containing neurons in the septum control inhibitoryinterneurons in the hippocampus. Nature 336:170–173. Givens BS, Olton DS (1990) Cholinergic and GABAergic modulation of the medial septal area: Effect on working memory. Behav Neurosci 104:849–855. Golding N, Staff N, Spruston N (2002) Dendritic spikes as a mechanism for cooperative long-term potentiation. Nature 418:326–331. Green JD, Arduini AA (1954) Hippocampal electrical activity and arousal. J Neurophysiol 17:533–557. Griffin AL, Asaka Y, Darling RD, Berry SD (2004) Thetacontingent trial presentation accelerates learning rate and enhances hippocampal plasticity during trace eyeblink conditioning. Behav Neurosci 118:403–411. Griffin AL, Lee I, Eichenbaum H, Hasselmo ME (2005) Phase relationship between single unit firing in CA1 and theta rhythm on a continuous T-maze alternation task. Soc Neurosci Abstr 31:72.7. Harris KD, Henze DA, Hirase H, Leinekugel X, Dragoi G, Czurko A, Buzsaki G (2002) Spike train dynamics predicts theta-related phase precession in hippocampal pyramidal cells. Nature 417:738–741. Hasselmo M, Cannon RC, Koene RA (2002a) A simulation of parahippocampal and hippocampal structures guiding spatial navigation of a virtual rat in a virtual environment: A functional framework for theta theory. In: The Parahippocampal Region: Organisation and Role in Cognitive Functions (Witter MP, Wouterlood FG, eds.), pp 139–161. Oxford: Oxford University Press. Hasselmo ME (2003a) Theta theory: requirements for encoding events and task rules explain theta phase relationships in hippocampus and neocortex. Proceedings of the International Joint Conference on Neural Networks 2:1470–1475. Hasselmo ME (2005) A model of prefrontal cortical mechanisms for goal-directed behavior. J Cogn Neurosci 17:1115–1129 Hasselmo ME, Bodelon C, Wyble BP (2002b) A proposed function for hippocampal theta rhythm: Separate phases of encoding and retrieval enhance reversal of prior learning. Neural Comput 14:793–817. Hasselmo ME, Eichenbaum H (2005) Hippocampal mechanisms for the context-dependent retrieval of episodes. Neural Netw 15:689–707. Hasselmo ME, Hay J, Ilyn M, Gorchetchnikov A (2002c) Neuromodulation, theta rhythm and rat spatial navigation. Neural Netw 15:689–707.
THETA RHYTHM AND MEMORY-GUIDED BEHAVIOR
Hasselmo ME, Zilli E (2005) Hebbian synaptic modification in cortical circuits and memory-guided behavior in spatial alternation and delayed non-match to position. In: Neural Networks, 2005. IJCNN 2005. Proceedings. 2005 International Joint Conference on Neural Networks, pp 2754–2759. Montreal, Canada: IEEE Press. Holscher C, Anwyl R, Rowan MJ (1997) Stimulation on the positive phase of hippocampal theta rhythm induces long-term potentiation that can be depotentiated by stimulation on the negative phase in area CA1 in vivo. J Neurosci 17:6470–6477. Howard MW, Fotedar MS, Datey AS, Hasselmo M (2004) The temporal context model in spatial navigation and relational learning: Explaining medial temporal lobe function across domains. Psychol Rev 112:75–116. Howard MW, Kahana MJ (2002) A distributed representation of temporal context. J Math Psychol 46: 269–299. Huerta PT, Lisman JE (1995) Bidirectional synaptic plasticity induced by a single burst during cholinergic theta oscillation in CA1 in vitro. Neuron 15:1053–1063. Huxter J, Burgess N, O’Keefe J (2003) Independent rate and temporal coding in hippocampal pyramidal cells. Nature 425:828–832. Hyman JM, Hasselmo M (2004) Medial prefrontal cortex cells fire with a phase relationship to the hippocampal theta rhythm. Soc Neurosci Abstr 30:551.512. Hyman JM, Wyble BP, Goyal V, Rossi CA, Hasselmo M (2003) Stimulation in hippocampal region CA1 in behaving rats yields LTP when delivered to the peak of theta and LTD when delivered to the trough. J Neurosci 23:11725–11731. Hyman JM, Wyble BP, Rossi CA, Hasselmo ME (2002) Coherence between theta rhythm in rat medial prefrontal cortex and hippocampus. Soc Neurosci Abstr 28:477.476. Hyman JM, Zilli EA, Paley AM, Hasselmo ME (2005) Medial prefrontalcortex cells show dynamic modulation with the hippocampal theta rhythm dependent on behavior. Hippocampus 15:739–749. Jensen O, Lisman JE (1996a) Novel lists of 7þ/-2 known items can be reliably stored in an oscillatory shortterm memory network: interaction with long-term memory. Learn Mem 3:257–263. Jensen O, Lisman JE (1996b) Hippocampal CA3 region predicts memory sequences: accounting for the phase precession of place cells. Learn Mem 3:279–287. Jones MW, Wilson MA (2005) Theta rhythms coordinate hippocampal-prefrontal interactions in a spatial memory task. PLoS Biol 3:e402. Judge SJ, Hasselmo ME (2004) Theta rhythmic stimulation of stratum lacunosum-moleculare in rat hippocampus contributes to associative LTP at a phase offset in stratum radiatum. J Neurophysiol 92:1615–1624.
325
Klink R, Alonso A (1997) Muscarinic modulation of the oscillatory and repetitive firing properties of entorhinal cortex layer II neurons. J Neurophysiol 77:1813– 1828. Koene RA, Gorchetchnikov A, Cannon RC, Hasselmo ME (2003) Modeling goal-directed spatial navigation in the rat based on physiological data from the hippocampal formation. Neural Netw 16:577–584. Koene RA, Hasselmo ME (2005) An integrate and fire model of prefrontal cortex neuronal activity during performance of goal directed decision-making. Cereb Cortex 15:1964–1981. Koene RA, Hasselmo ME (2007) First-in-first-out item replacement in a model of short-term memory based on persistent spiking. Cereb Cortex 17:1766– 1781. Kramis R, Vanderwolf CH, Bland BH (1975) Two types of hippocampal rhythmical slow activity in both the rabbit and the rat: relations to behavior and effects of atropine, diethyl ether, urethane, and pentobarbital. Exp Neurol 49:58–85. Kubie JL, Muller RU, Fox SE (1985) Firing fields of hippocampal place cells: interim report. In: Electrical Activity of the Archicortex (Buzsaki G, Vanderwolf CH, eds.), pp 221–231. Budapest: Hungarian Academy of Science. Lee I, Griffin AL, Zilli EA, Eichenbaum H, Hasselmo ME (2006) Gradual translocation of spatial correlates of neuronal firing in the hippocampus toward prospective reward locations. Neuron 51:639–650. Lee MG, Chrobak JJ, Sik A, Wiley RG, Buzsaki G (1994) Hippocampal theta activity following selective lesion of the septal cholinergic system. Neuroscience 62: 1033–1047. Leutgeb S, Mizumori SJ (1999) Excitotoxic septal lesions result in spatial memory deficits and altered flexibility of hippocampal single-unit representations. J Neurosci 19:6661–6672. Levy WB (1996) A sequence predicting CA3 is a flexible associator that learns and uses context to solve hippocampal-like tasks. Hippocampus 6:579–590. Levy WB, Steward O (1983) Temporal contiguity requirements for long-term associative potentiation/depression in the hippocampus. Neuroscience 8:791–797. Lisman JE (1999) Relating hippocampal circuitry to function: recall of memory sequences by reciprocal dentate-CA3 interactions. Neuron 22:233–242. Lisman JE, Idiart MA (1995) Storage of 7 þ/- 2 shortterm memories in oscillatory subcycles. Science 267: 1512–1515. Manns ID, Alonso A, Jones BE (2000a) Discharge profiles of juxtacellularly labeled and immunohistochemically identified GABAergic basal forebrain neurons recorded in association with the electroencephalogram in anesthetized rats. J Neurosci 20:9252–9263.
326 THEORETICAL SIGNIFICANCE OF PLACE FIELDS Manns ID, Alonso A, Jones BE (2000b) Discharge properties of juxtacellularly labeled and immunohistochemically identified cholinergic basal forebrain neurons recorded in association with the electroencephalogram in anesthetized rats. J Neurosci 20: 1505–1518. Manns JR, Zilli EA, Ong KC, Hasselmo ME, Eichenbaum H (2007) Hippocampal CA1 spiking during encoding and retrieval: Relation to theta phase. Neurobiol Learn Mem 87:9–20. Marr D (1971) Simple memory: a theory for archicortex. Philos Trans R Soc Lond B Biol Sci 262:23–81. McNaughton BL, Barnes CA, O’Keefe J (1983) The contributions of position, direction, and velocity to single unit-activity in the hippocampus of freelymoving rats. Exp Brain Res 52:41–49. McNaughton BL, Morris RGM (1987) Hippocampal synaptic enhancement and information storage within a distributed memory system. Trends Neurosci 10: 408–415. Mehta MR (2004) Cooperative LTP can map memory sequences on dendritic branches. Trends Neurosci 27:69–72. Mehta MR, Barnes CA, McNaughton BL (1997) Experience-dependent, asymmetric expansion of hippocampal place fields. Proc Natl Acad Sci USA 94: 8918–8921. Mehta MR, Lee AK, Wilson MA (2002) Role of experience and oscillations in transforming a rate code into a temporal code. Nature 417:741–746. M’Harzi M, Palacios A, Monmaur P, Willig F, Houcine O, Delacour J (1987) Effects of selective lesions of fimbria-fornix on learning set in the rat. Physiol Behav 40:181–188. Mizumori SJY, Perez GM, Alvarado MC, Barnes CA, Mcnaughton BL (1990) Reversible inactivation of the medial septum differentially affects 2 forms of learning in rats. Brain Res 528:12–20. Muller RU, Kubie JL (1989) The firing of hippocampal place cells predicts the future position of freely moving rats. J Neurosci 9:4101–4110. Muller RU, Kubie JL, Ranck Jr. JB (1987) Spatial firing patterns of hippocampal complex-spike cells in a fixed environment. J Neurosci 7:1935–1950. Muller RU, Stead M (1996) Hippocampal place cells connected by Hebbian synapses can solve spatial problems. Hippocampus 6:709–719. O’Keefe J (1976) Place units in the hippocampus of the freely moving rat. Exp Neurol 51:78–109. O’Keefe J, Dostrovsky J (1971) The hippocampus as a spatial map. Preliminary evidence from unit activity in the freely-moving rat. Brain Res 34:171–175. O’Keefe J, Recce ML (1993) Phase relationship between hippocampal place units and the EEG theta rhythm. Hippocampus 3:317–330.
Orr G, Rao G, Houston FP, McNaughton BL, Barnes CA (2001) Hippocampal synaptic plasticity is modulated by theta rhythm in the fascia dentata of adult and aged freely behaving rats. Hippocampus 11:647–654. Pang KC, Nocera R, Secor AJ, Yoder RM (2001) GABAergic septohippocampal neurons are not necessary for spatial memory. Hippocampus 11:814–827. Pavlides C, Greenstein YJ, Grudman M, Winson J (1988) Long-term potentiation in the dentate gyrus is induced preferentially on the positive phase of theta rhythm. Brain Res 439:383–387. Petsche H, Stumpf C, Gogolak G (1962) The significance of the rabbit’s septum as a relay station between the midbrain and the hippocampus. II. The differential influence of drugs upon both the septal cell firing pattern and the hippocampus theta activity. Electroencephalogr Clin Neurophysiol 14:202–211. Rawlins JN, Feldon J, Gray JA (1979) Septo-hippocampal connections and the hippocampal theta rhythm. Exp Brain Res 37:49–63. Redish AD, Touretzky DS (1998) The role of the hippocampus in solving the Morris water maze. Neural Comput 10:73–111. Sainsbury RS, Harris JL, Rowland GL (1987a) Sensitization and hippocampal type 2 theta in the rat. Physiol Behav 41:489–493. Sainsbury RS, Heynen A, Montoya CP (1987b) Behavioral correlates of hippocampal type 2 theta in the rat. Physiol Behav 39:513–519. Seager MA, Johnson LD, Chabot ES, Asaka Y, Berry SD (2002) Oscillatory brain states and learning: Impact of hippocampal theta-contingent training. Proc Natl Acad Sci USA 99:1616–1620. Seidenbecher T, Laxmi TR, Stork O, Pape HC (2003) Amygdalar and hippocampal theta rhythm synchronization during fear memory retrieval. Science 301:846– 850. Sharp PE, Blair HT, Brown M (1996) Neural network modeling of the hippocampal formation spatial signals and their possible role in navigation: a modular approach. Hippocampus 6:720–734. Siapas AG, Lee AK, Lubenov EV, Wilson MA (2000) Prefrontal phase-locking to hippocampal theta oscillations. Soc Neurosci Abstr 26:467.461. Siapas AG, Lubenov EV, Wilson MA (2005) Prefrontal phase locking to hippocampal theta oscillations. Neuron 46:141–151. Skaggs WE, McNaughton BL, Wilson MA, Barnes CA (1996) Theta phase precession in hippocampal neuronal populations and the compression of temporal sequences. Hippocampus 6:149–172. Sohal VS, Hasselmo ME (1998a) Changes in GABAB modulation during a theta cycle may be analogous to the fall of temperature during annealing. Neural Comput 10:869–882.
THETA RHYTHM AND MEMORY-GUIDED BEHAVIOR
Sohal VS, Hasselmo ME (1998b) GABA(B) modulation improves sequence disambiguation in computational models of hippocampal region CA3. Hippocampus 8:171–193. Stewart M, Fox SE (1989a) Two populations of rhythmically bursting neurons in rat medial septum are revealed by atropine. J Neurophysiol 61:982–993. Stewart M, Fox SE (1989b) Detection of an atropineresistant component of the hippocampal theta rhythm in urethane-anesthetized rats. Brain Res 500:55–60. Stewart M, Fox SE (1990) Do septal neurons pace the hippocampal theta rhythm? Trends Neurosci 13:163–168. Sutton RS, Barto AG (1998) Reinforcement Learning (Adaptive Computation and Machine Learning). Cambridge, MA: MIT Press. Tsodyks MV, Skaggs WE, Sejnowski TJ, McNaughton BL (1996) Population dynamics and theta rhythm phase precession of hippocampal place cell firing: a spiking neuron model. Hippocampus 6:271–280. Vanderwolf CH (1969) Hippocampal electrical activity and voluntary movement in the rat. Electroencephalogr Clin Neurophysiol 26:407–418. Vanderwolf CH (1971) Limbic-diencephalic mechanisms of voluntary movement. Psychol Rev 78:83–113. Van der Zee EA, Luiten PGM (1994) Cholinergic and GABAergic neurons in the rat medial septum express muscarinic acetylchoine receptors. Brain Res 652:263– 272. Vertes RP, Kocsis B (1997) Brainstem-diencephaloseptohippocampal systems controlling the theta rhythm of the hippocampus. Neuroscience 81:893– 926. Wallenstein GV, Hasselmo ME (1997) GABAergic modulation of hippocampal population activity: sequence learning, place field development, and the phase precession effect. J Neurophysiol 78:393–408. Walsh TJ, Herzog CD, Gandhi C, Stackman RW, Wiley RG (1996) Injection of IgG 192-saporin into the medial septum produces cholinergic hypofunction
327
and dose-dependent working memory deficits. Brain Res 726:69–79. Whishaw IQ (1972) Hippocampal electroencephalographic activity in the Mongolian gerbil during natural behaviours and wheel running and in the rat during wheel running and conditioned immobility. Can J Psychol 26:219–239. White NM, McDonald RJ (2002) Multiple parallel memory systems in the brain of the rat. Neurobiol Learn Mem 77:125–184. Wiener SI, Paul CA, Eichenbaum H (1989) Spatial and behavioral-correlates of hippocampal neuronalactivity. J Neurosci 9:2737–2763. Winson J (1978) Loss of hippocampal theta rhythm results in spatial memory deficit in the rat. Science 201:160–163. Wood ER, Dudchenko PA, Robitsek RJ, Eichenbaum H (2000) Hippocampal neurons encode information about different types of memory episodes occurring in the same location. Neuron 27:623–633. Wu M, Shanabrough M, Leranth C, Alreja M (2000) Cholinergic excitation of septohippocampal GABA but not cholinergic neurons: implications for learning and memory. J Neurosci 20:3900–3908. Wyble BP, Hyman JM, Rossi CA, Hasselmo M (2004) Analysis of theta power in hippocampal EEG during bar pressing and running behavior in rats during distinct behavioral contexts. Hippocampus 14:368– 384. Wyble BP, Linster C, Hasselmo ME (2000) Size of CA1evoked synaptic potentials is related to theta rhythm phase in rat hippocampus. J Neurophysiol 83:2138– 2144. Zilli EA, Hasselmo ME (2006) An analysis of the mean theta phase of population activity in a model of hippocampal region CA1. Network 17:277–297. Zugaro MB, Monconduit L, Buzsaki G (2005) Spike phase precession persists after transient intrahippocampal perturbation. Nat Neurosci 8:67–71.
20 Network Analysis of the Significance of Hippocampal Subfields GERGELY PAPP AND ALESSANDRO TREVES
MAMMALS ARE GENERATED BY GRANULATION Mammals originate from the therapsids, one order among the first amniotes, or early reptiles, as they are commonly referred to. They are estimated to have radiated away from other early reptilian lineages, including the anapsids (the progenitors of modern turtles) and diapsids (from which other modern reptilians, as well as birds, derive) some three hundred million years ago (Carroll, 1988). Perhaps mammals emerged as a fully differentiated class out of the third-to-last of the great extinctions, in the Triassic period. The changes in the organization of the nervous system, which mark the transition from protoreptilian ancestors to early mammals, can be reconstructed only indirectly. Along with supporting arguments from the examination of endocasts (the inside of fossil skulls; Jerison, 1990) and of presumed behavioral patterns (Wilson, 1975), the main line of evidence is the comparative anatomy of present-day species (Diamond and Hall, 1969). Among a variety of quantitative changes in the relative development of different structures, changes that have been extended, accelerated, and diversified during the entire course of mammalian evolution (Finlay and Darlington, 1995), two major qualitative changes stand out in the forebrain, two new features that, once established, characterize the cortex of mammals as distinct from that of reptilians and birds. Both
of these changes involve the introduction of a new ‘‘input’’ layer of granule cells. In the first change, it is the medial pallium (the medial part of the upper surface of each cerebral hemisphere, as it bulges out of the forebrain) that reorganizes into the modern-day mammalian hippocampus. The crucial step is the detachment of the most medial portion, which loses both its continuity with the rest of the cortex at the hippocampal sulcus, and its projections to dorsolateral cortex (Ulinski, 1990). The rest of the medial cortex becomes Ammon’s horn, and retains the distinctly cortical pyramidal cells, while the detached cortex becomes the dentate gyrus, with its population of granule cells, that project now, as a sort of pre-processing stage, to the pyramidal cells of field CA3 (Amaral et al., 1990). In the second change, the dorsal pallium (the central part of the upper surface) reorganizes internally, in those areas that process topographic modalities, to become the cerebral neocortex. Aside from special cases, most mammalian neocortices display the characteristic isocortical pattern of lamination, or organization into distinct layers of cells (traditionally classified as six, in some cases with sublayers). A prominent step in lamination is granulation, whereby the formerly unique principal layer of pyramidal cells is split by the insertion of a new layer of excitatory, but intrinsic, granule cells, in between the pyramidal cells of the infragranular and supragranular layers. This is
328
NETWORK ANALYSIS OF HIPPOCAMPAL PLACE FIELDS
layer IV, where the main ascending inputs to cortex terminate (Diamond et al., 1985).
Cortical Lamination May Try to Preserve Topography while Building up Memory In the context of analyzing spatial correlates of hippocampal activity, it may be useful to consider first the significance of the ‘‘other’’ granulation, i.e., the lamination of the neocortex. A hypothesis has been formulated (Treves, 2003) that accounts for isocortical granulation and for the differentiation between supraand infragranular pyramidal layers, as advantageous to support fine topography in the sensory maps that mammals have evolved beyond the gross topography of sensory maps in reptiles. Fine topography implies a generic distinction between ‘‘where’’ information, explicitly mapped on the cortical sheet, and ‘‘what’’ information, represented in a distributed fashion as a distinct firing pattern across neurons. Memory patterns can be stored on recurrent collaterals in the cortex, and such memory can help substantially in the analysis of current sensory input. The effective use of recurrent collaterals, because of the crucial limit on memory storage load, requires afferent projections to the cortex that are spread over a large patch; whereas the precise localization of a stimulus on the sensory map requires narrowly focused afferents (see Treves, 2003, for the complete argument, and Roudi and Treves, 2006, for analytical treatment of a single-layer model). Simulation of a simplified network model demonstrates that a nonlaminated patch of cortex, with a single characteristic spread of afferent connections, must compromise between transmitting ‘‘where’’ information and retrieving ‘‘what’’ information. The differentiation of a granular layer affords a quantitative advantage, by allowing focused afferents to the granular units together with widespread afferents to pyramidal units. For this purely anatomical differentiation to be effective, however, it must be accompanied by a physiological differentiation: pyramidal units must adapt their firing, that is, decrease their response to steady inputs, much more than granular units. With this further difference, the pyramidal layers can select the correct attractor for memory retrieval before the granular layer, which adapts less, partially takes over the dynamics, and focuses activity on the cortical spot that most accurately reflects the position of the sensory input. Adaptation thus effectively separates out in time, albeit only partially, two information-processing operations that occur in different spaces: the retrieval of memories in the abstract space of attractors, and the accurate relay of stimulus position in the physical space of the cortical surface. The advantage of the differentiation is quantitatively minor. The unconven-
329
tional hypothesis is that a major qualitative step, the transition from a simpler paleocortex to a more elaborate isocortex, came about just in order to gain a few percent more bits in the average combined value of ‘‘what’’ and ‘‘where’’ information.
Medial Entorhinal Cortex Codes for Space, but not Topographically The discovery of grid cells in the medial entorhinal cortex (MEC) of the rat (Fyhn et al., 2004) and of the precise triangular pattern of their firing fields (Hafting et al., 2005) offers an illuminating perspective on the relation between granulation and spatial correlates of neural activity. Entorhinal cortex, the so-called gateway to the hippocampus, is poorly granulated, i.e., its layer IV is almost absent, in contrast to sensory cortices. In parallel, its newly discovered spatial correlates are organized in a distinctly nontopographic fashion: individual grid cells have ‘‘place’’ fields dispersed over the entire surface of any environment used for testing, in two dimensions, at some of the vertices of a triangular grid, and cells very close to each other in tissue have grids shifted by a seemingly random phase (Hafting et al., 2005). The only remnant of topography appears to be a gradual increase in grid spacing from more dorsal to more ventral portions of MEC. By combining the responses of ensembles of grid units at different dorsoventral levels one may estimate the position of the animal in space with great accuracy, which suggests that grid units may be part of a mechanism for path integration; on the other hand very, little or no information can be extracted from MEC activity about which environment the animal is in (Fyhn et al., 2007). It now appears, therefore, that spatial computations per se are largely performed by the rat brain before ever accessing the hippocampus, and culminate in a sort of universal map of allocentric space, in MEC layer II. Only a portion of entorhinal cortex participates in such a map, which is applied and used irrespective of context, but in combination with context-specific signals that may determine the activity of other parts of entorhinal cortex. The hippocampus operates on the universal map and on context-specific signals to create context-specific metric representations of space, which are stored in memory. The capacity of the hippocampus to rapidly switch between the representations of different contexts is illustrated by hippocampal global ‘‘remapping,’’ i.e., the transition to new, unrelated arrangements of place fields by the same population of recorded cells after suitable behavioral manipulations. During hippocampal remapping, local ensembles of MEC grid cells exhibit a rigid, coherent shift of their map. Understanding the circuitry of the hippocampus, therefore, crucially involves understanding
330 THEORETICAL SIGNIFICANCE OF PLACE FIELDS this capacity for decorrelating in the hippocampus spatial representations that are rigidly coherent in MEC— at least in rodents. It could well be that in other species complex memories of a less spatial nature take a more prominent role, in which case it would be even more appropriate to approach hippocampal decorrelation and memory processes at an abstract level, and independently of the possibly species-specific spatial processes so finely investigated with the rat model.
Looking at Granules through a Spin Glass Which approaches can take us beyond a mere functional description of the role of these different networks and lead us to understand, in evolutionary terms, their design principles? In recent years, the same computational viewpoint, which can be labeled a ‘‘spin-glass’’ viewpoint, has been applied to three apparently disparate topics: the lamination of sensory cortex, mentioned above; the differentiation of the mammalian hippocampus into subfields, which is the focus of this chapter; and the neuronal dynamics that might underlie the faculty for language in the human frontal lobes (Treves, 2005). These studies share a common perspective: they all discuss the evolution of cortical networks in terms of their computations, quantified by simulating simplified formal models. They all dwell on the interrelationship between qualitative and quantitative change. Finally, they all include, as a necessary ingredient of the relevant computational mechanisms, a simple feature of pyramidal cell biophysics: firing rate adaptation. A belief motivating the approach is that the most important steps in the evolution of the nervous system are those that address computational demands— demands that are part of the ‘‘job specification’’ of the brain as an information-processing system, rather than those steps that address, say, physiological or anatomical constraints. Among genuine information-processing problems, one that has been quantified through the use of formal models is the limit on the storage of memories that is imposed by the connectivity of a system of neuron-like units. Consideration of this limit is partly motivated by the observation that most gray matter volume appears to be devoted to synaptic contacts (Braitenberg and Schu¨z, 1991), as if the cortex had evolved to maximize connectivity and, ultimately, memory storage. The mathematical procedures that have been used to obtain a proper quantification of the relation between connectivity and memory were originally developed to analyze the physics of a class of materials known as spin glasses (see, e.g., Amit, 1989). Spin glasses are endowed with interactions that can be characterized as disordered and hence as interfering with each other, somewhat like, in a neural network,
distinct memory representations interfering with each other at retrieval. Although spin glasses have nothing deeper in common with memory systems than this analogy and the mathematical procedures useful in analyzing them, the effectiveness and generality of these procedures have lead some investigators to approach many information-processing problems by relying on the analysis of spin glasses as a basic paradigm. Unwrapped from its technicalities, the ‘‘spin glass approach’’ reduces essentially to the idea that cortical systems face a crucial connectivity constraint on extensive memory storage, that the constraint results from interference among memories, and that to analyze such interference we can borrow techniques from statistical physics. The three evolutionary problems cited above are all, to some extent, ‘‘spin glass problems’’ in disguise. The first two involve the insertion of a new layer of granule cells into a pristine associative memory network, and they both relate in some way to spatial representations. Yet, the relation appears to be completely different in the two cases.
DIFFERENTIATION OF THE HIPPOCAMPUS Focusing now on the hippocampus, one may ask what evolutionary advantage for mammals was brought about by granulation. Attempts to account for its remarkable differentiation into three main subfields have been mostly based, since the seminal paper by David Marr (1971)—and well before awareness developed among modelers of the evolutionary specificity of such organization—on a computational analysis of the role of the hippocampus in memory. With the simultaneous discovery of place cells (O’Keefe and Dostrovsky, 1971), the rodent model seemed to point toward a special hippocampal role for spatial representation and spatial memory. Although an accumulating body of evidence has indicated that hippocampal activity is not exclusively related to space (Eichenbaum, 2000), the prevalence of spatial correlates in the rat has encouraged speculations on the evolution of the hippocampus based on spatial function, reinforced by the very recent discovery of place cells in a rather different mammal, the echolocating bat (Ulanovsky and Moss, 2007). The hippocampus, however, is important for spatial memory in birds as well (Bingman and Jones, 1994; Clayton and Krebs, 1995; Clayton et al., 2003). The avian and mammalian hippocampi are structurally very different, with birds having stayed close to their reptilian progenitors, and mammals having detached the dentate gyrus from Ammon’s horn, as mentioned above. A reasonable hypothesis may then be that the new mammalian de-
NETWORK ANALYSIS OF HIPPOCAMPAL PLACE FIELDS
sign somehow enhances the capability of the hippocampus to serve as a memory store, perhaps with the nuance of a prevailingly spatial memory store. It is plausible that the primitive cortical tissue in early reptilian-like ancestors to both mammals and birds was rich in recurrent collaterals, much like region CA3 in the modern mammalian hippocampus. Simplified models show how a recurrent network can naturally retrieve distributions of activity from partial cues as an autoassociative memory (Hopfield, 1982), provided the synapses on the recurrent connections among its pyramidal cells are endowed, as likely was the case for primitive cortex with associative, ‘‘Hebbian,’’ plasticity, such as that based on NMDA receptors (Collingridge and Bliss, 1995). That cortex can then be conceptualized as having operated at least as a content-addressable memory for distributed activity patterns—provided it had an effective way of distinguishing its operating modes. A generic problem with associative memories based on recurrent connections is distinguishing a storage mode from a retrieval mode. To be effective, recurrent connections should dominate the dynamics of the system when it is operating in retrieval mode; whereas while storing new information the dynamics should be primarily determined by afferent inputs, with limited interference from the memories already stored in the recurrent connections, which instead should modify their weights to store the new information (Treves and Rolls, 1992).
Distinguishing Storage from Retrieval The most phylogenetically primitive solution to effect the dual operating mode is to use a modulator that acts differentially on the afferent inputs (originally, those arriving at the apical dendrites) and on the recurrent connections (predominantly lower on the dendritic tree). Acetylcholine (ACh) can achieve this effect, exploiting the orderly arrangement of pyramidal cell dendrites in the cortex (Hasselmo and Schnell, 1994). Acetylcholine is one of several very ancient neuromodulating systems, well conserved across vertebrates, and it is likely that it operated in this way already in the early reptilian cortex, throughout its subdivisions. Mike Hasselmo has emphasized this role of ACh in memory, through a combination of slice work and neural network modeling (Hasselmo et al., 1995, 1996). This work has been focused on the hippocampus—originally, the medial wall—and on piriform cortex—originally, the lateral wall. The proposed mechanism, however, has no reason to be circumscribed to these regions, and it could well operate across cortical systems involved in memory storage. One flaw of an Ach-based mechanism is that it requires an active process that distinguishes storage
331
from retrieval periods and regulates ACh-release accordingly. In the hippocampus, however, it appears that mammals have devised a more refined expedient to separate storage from retrieval, which can efficiently perform both in a passive mode, by inserting a preprocessor before the CA3 memory network. The preprocessor should instruct which units in CA3 should fire in a new distribution of activity to be stored as the memory representation of a new item to be remembered. As a simple model, one can think of a preprocessing network without recurrent connections, which simply forms new arbitrarily determined representations on the fly, and through a system of one-toone connections (‘‘detonator’’ synapses) imposes these new representations onto CA3 (McNaughton and Morris, 1987). In fact, the one-to-one correspondence is not needed; what enriches the representation to be stored of meaningful content, against the interference of recurrent connections, is just a system of sparse and strong connections from a sparsely coded feed-forward network. In developing the preprocessor notion, a quantitative argument was derived, revolving around the amount of new information that could be encoded in CA3 representations with different input systems (Treves and Rolls, 1992; Fig. 20–1). The argument is based on the ‘‘quasi-theorem,’’ which has never been satisfactorily proven but empirically holds true, that an associative memory can hold up to I=(NC) 0:2 0:3 bits
(1)
of information per synapse, where N is the number of units and C the average number of connections, or synapses, per unit. Since the storage capacity, or maximum number of discrete patterns of activity that can be individually retrieved, is estimated as pc 0:2 0:3 C=[a ln(1=a)];
(2)
where a is the sparsity of the stored representation (see Treves and Rolls, 1991), information theoretic efficiency requires that each such representation should contain at least roughly i N a ln (1=a) bits
(3)
of new information per unit. This is, apart from an ln(2) factor, the amount of information per unit of a binary variable (in the sparse, a << 1 regime). Thus efficient storage requires that CA3 pyramidal units be as informative about new contents as, roughly, binary units. The challenge for afferent inputs is to prevail over the recurrent connections, which do not impart new contents to a pattern of activity to be stored. Figure 20–1 shows that this challenge can be met by
332 THEORETICAL SIGNIFICANCE OF PLACE FIELDS
Figure 20–1. Information per unit in a model CA3 representation, as a function of its sparsity. The solid line is the amount that can be associatively retrieved, the dash-dotted line is what can be stored by inputs with the characteristics of the perforant path (pp) to CA3, and the three dotted lines are three estimates of what can be stored through mossy-fiber (mf) inputs (the estimates differ in the sparsity of the dentate). Analytical derivation in Treves and Rolls (1992).
afferent inputs with the characteristics of the mossy fibers, but not by those with the characteristics of the perforant path to CA3 (Treves and Rolls, 1992).
Mapping Continuous Spatial Attractors onto Discrete Memories The argument above has been worked out for the case of discrete memory items, which can be taken as a model of episodic memory. Initially, the neural network approach, aiming at quantifying the capacity of associative memories, was formulated in terms of fully connected recurrent architectures and discrete memory states, conceived, in the limit of no fluctuations, as points in the multidimensional space in which each component corresponds to the firing rate or in general to the activity of one unit (Hopfield, 1982). This formulation, which was the starting point for physicists interested in applying powerful mathematical analysis techniques, was preceded by the more rudimentary analysis by David Marr. Marr also thought in terms of discrete memory states, and had
guessed the importance of recurrent collaterals, a prominent feature of the CA3 subfield (Amaral et al., 1990), even though his own model was not really affected by the presence of such collaterals, as shown later (Willshaw and Buckingham, 1990). Although the article by Marr (1971) was nearly simultaneous with two of the most exciting experimental discoveries related to the hippocampus, that of place cells (O’Keefe and Dostrovsky, 1971) and that of long-term synaptic potentiation (Bliss and Lomo, 1973), for a long time it did not seem to inspire further theoretical analyses—with the exception of an interesting discussion of the collateral effect in a neural network model (Gardner-Medwin, 1976). One factor was probably the mathematical ‘‘technology’’ available to Marr, inadequate to really investigate his models quantitatively. Marr himself become disillusioned with his youthful enthusiasm for unraveling brain circuits, and in his mature years took a much more sedate—and less neural—interest in vision. From the inspiring 1987 review by McNaughton and Morris, however, an increasing number of other investigators rediscovered the young Marr, and tried to elaborate those ideas in order to understand the operation of hippocampal circuits. Edmund Rolls (1989) and others again emphasized the crucial role probably played by the CA3 recurrent collaterals, and made explicit the relation to the autoassociative memory networks studied quantitatively by the physicists (Amit et al., 1987). In establishing the relation, the salient spatial character of hippocampal memory correlates was provisionally neglected, to take advantage of the formal models based on discrete attractor states. As a matter of fact, an autoassociator may subserve both the storage of discrete memories as point-like attractor states or of more complex memories, for example ‘‘synfire chains’’ (see Abeles, 1991), which can be individually distinct and discrete or organized in arbitrary branching patterns, or continuous attractors, when network dynamics converges to fixed points that are a continuous arrangement on some manifold in the high-dimensional activity space. Simple examples of continuous attractors are present in models of orientation selectivity by horizontal interaction in visual cortex (Sompolinsky and Shapley, 1997) or of the head direction system (Skaggs et al., 1993). These models do not store information in longterm memory, and their fixed points comprise a single (in these particular cases, one-dimensional) manifold. Samsonovich and McNaughton’s multiple-chart model (1997) demonstrated instead, in the context of a model for path integration, how one could conceive of fixed points organized in multiple two-dimensional (2D) continuous manifolds, each of which maps the animal position in a distinct environment. Exploration of a
NETWORK ANALYSIS OF HIPPOCAMPAL PLACE FIELDS
new environment leads to the formation of a new chart (ab initio or using some prewired connectivity; it may be difficult to distinguish the two possibilities). The question then arises of how many charts a given recurrent network can hold in long-term memory. The discrete memory states models point at sparsity as the crucial representational parameter that influences memory capacity (besides the anatomical one of the connectivity per unit), as demonstrated in the above equations and in Figure 20–1. Hippocampal space-related activity has been known for many years to be quite sparse, both in rats and in monkeys, which fits in with the notion of a memory store for large numbers of items. Exactly how sparse it is depends on the measure used to quantify sparseness. One measure used in formal network models is the parameter a, defined as the square of the mean firing rate across a population of units at a given instant, divided by the mean of the square firing rate (Treves and Rolls, 1991). Application of such measure to cells recorded by Jill and Stefan Leutgeb in the Moser lab gives the values in Figure 20–2. As one can see, there is some arbitrary element in applying the measure to real data, like choosing the temporal or spatial bins; however, values are in a rather sparse range, a&0.02–0.07, and CA3 activity tends to be sparser than CA1 activity, as reported with other measures (Leutgeb et al., 2004), but not in all rats. The storage capacity of a multichart recurrent autoassociator was analyzed by Battaglia and Treves (1998), who extracted a simple rule-of-thumb for assessing the memory load of a chart. A chart that maps a finite environment onto the activity of place cell–like units is equivalent, capacity-wise, to as many discrete attractor states as there are locations in the environment, for which the activity vectors are pair-wise decorrelated. If the 2D environment is represented by place cell–like units, which are quiescent outside their place field, the decorrelation radius is roughly the radius of the typical place field, which is itself proportional to the linear size of the environment times the square root of the sparsity of the neural representation. Thus, if some dozen typical CA3 fields ‘‘fit,’’ once properly juxtaposed, in a typical rat recording box, the memory load of the chart corresponding to that box is roughly equivalent to a dozen discrete memories of equal sparsity. The number of such charts, or distinct environments of the size of a typical recording box, that can be held simultaneously in the network, is limited by the critical value pcharts 0:1 C = ln (1=a)
(4)
(see Figs. 1, 2 in Battaglia and Treves, 1998). This is still a huge number, of the order of several thousands,
333
given the recurrent connectivity of the CA3 network in rodents (Amaral et al., 1990). The apparent paradox that fewer charts can be stored if they are sparser (a lower a parameter makes the denominator larger) can be understood by considering that sparser activity in a large net leads to better spatial resolution, and hence requires more discrete fixed points attractors to cover, as effectively smaller tiles, the whole environment. This chart capacity again respects the unproven associative memory theorem mentioned above, in that the maximum amount of information that can be retrieved per synapse is about 0.15 bits, as shown in Figure 5 of Battaglia and Treves (1998).
Decoding Position on a Continuous Attractor As shown by Wilson and McNaughton (1993), one can easily decode, with reasonable accuracy, the position of the rat in a recording box from a sufficient number of simultaneously recorded units. One constructs firing-rate vectors at any time t, first as vectors {ri(t)}, where each component ri is the firing rate of one unit. Then one may simply average all vectors expressed when the rat is in position (x,y), to extract a ‘‘template’’ or mean vector {ri(x,y)}. Finally, by finding the best match for a vector at time t among all templates, one can assign a decoded position to time t. More sophisticated procedures yield somewhat better performance. In applying the same data as that used to extract a measure of sparsity in Figure 20–2, the simple decoding procedure yields the results in Figure 20–3. One can appreciate that localization benefits from decoding larger samples, with a trend that is approximately linear, and a y-intercept at the 0.01 chance level (the recording box was divided into 10 10 spatial bins). Decoding from small samples of fewer than about a dozen units leads to large fluctuations, depending on exactly which units are included in the sample; those data points are omitted from Figure 20–3. The very same decoding procedure can be applied to data obtained by simulating simple network models. In that case, one is not limited by the number of units that could be recorded simultaneously, and one may as well construct templates with all the units that simulate a given hippocampal subfield. The localization accuracy thus obtained then reflects an intrinsic limitation of the spatial representation in terms of a continuous attractor, i.e., the fact that the attractor, continuous and smooth in a conceptual limit, is in practice fractious and wrinkled. This more ‘‘absolute’’ measure of localization accuracy obtained in simulations would require, in experiments, extrapolating the trends of Figure 20–3 to very large sample sizes, a task
334 THEORETICAL SIGNIFICANCE OF PLACE FIELDS
Figure 20–2. Sparsity parameter a measured from samples of simultaneously recorded units in CA3 (red) and CA1 (blue). The average sparsity of population vectors recorded over a 10-min session at each of 10 10 spatial bins is plotted against the number of cells in each sample (filled circles are means over samples of each size, ± SD). In the limit of infinite sample size, sparsity values are shown to converge, approximately, with the average sparsity of individual cells, each measured by dividing the recording session into 1-s temporal bins (bars, ± SD across cells). Shorter temporal bins result in sparser averages. Data courtesy of Stefan Leutgeb.
that is impractical, given that one does not notice any strong tendency of the initial trends to deviate from linearity and saturate. Therefore, while the correspondence with experimental data is essential to keep the models tied to the real world, assessment of the real limits on network performance does require the simulation of (suitably formulated) network models. One can assess the localization afforded by decoding population vectors, real or simulated, also in terms of information, of how many bits are embedded in the correspondence between actual and decoded position.
Again, one finds increasing values with decoding larger samples. The increase, which for the fraction of accurately decoded temporal bins is obviously limited by 1, for information is limited by the log2 of the number of spatial bins used. In practice, even when decoding an entire simulated population using many spatial bins, the information values saturate at a lower level, which is a measure of the intrinsic spatial-information content of the representation (if one uses sufficiently many spatial bins in a simulation, this saturation level becomes insensitive to the number of spatial bins used).
NETWORK ANALYSIS OF HIPPOCAMPAL PLACE FIELDS
335
Figure 20–3. Fraction of correct localization obtained dividing the recording box (same raw data as for Fig. 20–2) into 10 10 spatial bins, constructing a template population vector in each, and assigning to each temporal bin the spatial bin with the best matching template. Performance is shown for simultaneously recorded CA3 (red) and CA1 (blue) units in three different rats, as a function of population size. Each data point is the mean over several samples of a given size, randomly selected from the total number of units (the largest appearing in each series).
With the relatively small samples available from experiments, one can instead see the initial linear trend, which for information has a y-intercept of 0, and thus obtain a measure of bits/unit, which can be compared, with some additional subtleties, to the graph in Figure 20–1. Without discussing those subtleties, we report that, in general, single hippocampal units convey substantial amounts of information, in the region of Figure 20–1, consistent with their having been instructed by mossy fibers. This point is further discussed below, using simulations. Finally, if a rat, real or simulated, has explored several environments, one may measure the accuracy of decoding both its position in each environment and which environment the rat is in, with the same procedure and using multiple sets of templates, one per environment.
VIRTUAL RAT SIMULATIONS Simulations are not constrained to follow the anatomy of the real system. One of their most useful roles can be in assessing the implications of a given network architecture, by contrasting its performance with that of alternative architectures. In Treves’s (2004) study, simulations were developed of a virtual rat exploring a virtual environment, realized as a square box with toroidal boundary conditions, while two arrays of units, modeling CA3 and CA1, developed representations of the environment. The unsupervised self-organization of a chart was produced solely by a ‘‘Hebbian learning rule’’ for modifying connection weights that models associative NMDA-dependent long-term plasticity at synapses between pyramidal cells. Two
336 THEORETICAL SIGNIFICANCE OF PLACE FIELDS
Figure 20–4. The uniform and the differentiated CA3–CA1 networks used with the virtual rat simulations. DG, dentate gyrus; MF, mossy fibers; RC, recurrent collaterals; SC, Schaeffer collaterals. Simulation details in Treves (2004).
architectures were contrasted, in particular, shown in Figure 20–4. With either architecture, place cell–like responses emerge spontaneously in both the CA3 and CA1 arrays. The initially randomly distributed spatial responses are gradually concentrated and refined into a single region with smooth contours. The size of the fields is determined by the ‘‘competitive’’ mechanism, which models inhibitory interneurons that regulate the sparsity of the representation. In simulations, to avoid having to deal with the finite size effects associated with simulating networks of size and connectivity well below the real system—effects that are more severe with sparser representations—the sparsity of CA fields is set in the conservative (not so sparse) range of a & 0.2 (see Fig. 20–5).
The Dentate Gyrus as a Chart Preprocessor With simulations that demonstrate the spontaneous emergence of charts and a formula that quantifies the retrieval capacity for charts, one is in a position to begin investigating, once more, whether enough new information can be stored in each representation to fully exploit the network capacity for information retrieval. In other words, a quantitative analysis of information storage in a model CA3 network, operating with and without the dentate gyrus, is needed, to assess any information-theoretic advantage in forming new representations, this time in the form of charts of place cell–like units. Unfortunately, the very 2D nature of charts makes a simplified mathematical
NETWORK ANALYSIS OF HIPPOCAMPAL PLACE FIELDS
337
Figure 20–5. The self-organization of place fields in the model CA1 network. Three examples of individual units are shown, before and after three learning sessions roughly equivalent to 10 min of real rat exploration of a novel environment. The sparsity parameter was set to a ¼ 0.2.
analysis of information storage like the one used by Treves and Rolls (1992) not applicable, because neighboring locations on one new chart generate correlated activity, which cannot be easily dissected from the interfering correlations with other, unrelated charts. While more sophisticated mathematical tools may yet be developed, computer simulations offer a straightforward approach to the issue. Within their limits, the simulations confirm the essential role of the inputs from the dentate gyrus to CA3 in guiding the learning of a new chart (see Fig. 20–6). It is worth noting that the beneficial ‘‘forcing’’ effect of mossy fiber inputs in driving the establishment of novel CA3 representations is even more salient when assessed indirectly in the information content or localization accuracy afforded by representations in CA1, whose units are only indirectly influenced by dentate gyrus activity. If CA3 and CA1 are not differentiated, which implies that CA1, too, receives the model mossy fiber projection (the ‘‘uniform’’ model of Fig. 20–6), the beneficial effect of the mossy fibers is reduced, counterintuitively perhaps. The simple simulations used by Treves (2004), while supportive of a specific role of the dentate gyrus in information storage, did not take into consideration
two major recent findings. One is the difference emerging between CA3 and CA1 in the representation of similar, correlated environments; these will be discussed more below. The other is the novel observation that the activity of dentate units, previously deemed to be very sparse, seems to be concentrated on a relative small fraction of newly generated granule cells (Ramirez-Amaya et al., 2006), which are biophysically indistinguishable from older neurons (Laplagne et al., 2006), but appear to ‘‘take care’’ of representing new information, much more than older and perhaps already committed neurons. New recording experiments (Leutgeb et al., 2006) may help clarify how space is coded by dentate granule cells. At any rate, the crucial prediction of the argument based on the analysis of discrete memories, if applied to charts, is that the inactivation of the mossy fiber synapses should impair the formation of new charts, but not the retrieval of previously stored ones. This prediction has recently been supported at the behavioral level (Lassalle et al., 2000): mice with a temporary inactivation supposedly selective for the mossy synapses were impaired in finding the hidden platform in a Morris water maze, but not if they had learned its location the previous week. A consistent result was
338 THEORETICAL SIGNIFICANCE OF PLACE FIELDS
Figure 20–6. Localization accuracy as a function of training session in hippocampal network models with and without the differentiation between CA3 and CA1, and with and without mossy fiber (mf) inputs. In the differentiated model, the position of the virtual rat is decoded from the CA1 representation. With reference to Figure 20–7, cue size here is Q ¼ 0.2. Other details are in Treves (2004).
more recently obtained in rats with a different, irreversible procedure of selective lesions and using indicators that only very approximately dissociate storage from retrieval (Lee and Kesner, 2004); the strong double dissociation found between perforant path and dentate lesions is remarkable, given the overlapping nature of the behavioral measures. While waiting for neurophysiological experiments to test the prediction at the neural level, the tentative conclusion from behavioral tests in rodents is that the dentate gyrus may have evolved in order to facilitate the storage of new information in the recurrent CA3 network. If validated, this hypothesis suggests that a quantitative, information-theoretical advantage may have favored a qualitative change, such as the insertion of the dentate gyrus in the hippocampal circuitry.
CA1 as a Clean-up Device? The dentate gyrus argument does not itself address the CA3–CA1 differentiation, which is equally prominent in the mammalian hippocampus. If the dentate gyrus can be understood as a CA3 preprocessor, perhaps CA1 should be understood as a CA3 postprocessor.
Structurally, CA3 and CA1 are contiguous portions of the dorsomedial cortex in reptiles. As this is reorganized into the mammalian hippocampus, CA3 and CA1 differentiate in two important ways, as shown in Figure 20–4. First, only CA3 receives the projections from the dentate gyrus, the mossy fibers. Second, only CA3 is dominated by recurrent collaterals, while most of the inputs to CA1 cells are the projections from CA3, the Schaffer collaterals (Amaral et al., 1990). The simplest version of the postprocessor notion is that it may be useful to add a further feed-forward associative network to clean-up memory representations already retrieved, but in incomplete form, by the CA3 network. The extra stage of recoding, if based on more neurons (there are more pyramidal cells in CA1 than in CA3 across all species in which numbers have been estimated), could also add robustness to the retrieved representation. Yet a mathematical network analysis of the clean-up notion, in the framework of discrete ‘‘episodic memory’’ fixed-point attractors and neglecting the separate entorhinal cortex inputs directly to CA1, has failed to illustrate impressive advantages to adding such a postprocessing stage (Treves, 1995). Information content was shown to grow from CA3 to CA1 in that study, but only by a minor amount. A more interesting suggestion comes from a review of neuropsychological studies in rats (Kesner et al., 2002), indicating a more salient role for CA1 along the temporal dimension. CA3 may specialize in associating information that was experienced strictly at the same time, whereas CA1 may link together, more than CA3, information across adjacent times (Rawlins, 1985). This may lead to the storage of sequences of instantaneous events that together build up an episode, or, if the events are not parsed, to the storage of continuous attractors along the temporal dimension, effectively short ‘‘movies’’ of episodes past. A way to formulate a qualitative implication of such a putative functional differentiation is to state that CA1 is important for prediction, i.e., for producing an output representation of what happened just after whatever is represented by the pattern of activity retrieved at the CA3 stage. In Kesner et al.’s (2002) review, however, their Figure 31–2 suggests that CA3 may be involved in temporal pattern separation just as much as CA1. Moreover, the role of either the dentate gyrus or CA3 in temporal pattern association has not been satisfactorily assessed. Available studies on the role of CA1 fail to make a clear distinction between tasks in which massive hippocampal outputs to the cortex are crucial and tasks in which a more limited hippocampal influence on the cortex may be sufficient. In the first case, lesioning CA1 should have an effect independent of what CA1 specifically contributes to information processing, simply because the main hip-
NETWORK ANALYSIS OF HIPPOCAMPAL PLACE FIELDS
pocampocortical output pathway is severed. In the second, CA3 outputs through the fimbria/fornix could enable hippocampal-mediated influences to be felt, even if deprived of the specific CA1 contribution.
Testing the Prediction of Predictive Coding The hypothesis that the differentiation between CA3 and CA1 may help solve precisely the computational conflict between pattern completion—integrating current sensory information on the basis of memory—and prediction—moving from one pattern to the next in a stored continuous sequence—can be assessed with computer simulations. To obtain results comparable to those of typical rat experiments, the same neural network simulations used to analyze the role of dentate gyrus inputs to CA3 can be used of a virtual rat exploring a small, toroidal environment (Treves, 2004). The network model was thus trained to acquire a chart representation of the explored environment as a spatially continuous attractor. Temporal continuity along each trajectory was used to assess the extent to which CA3 would take care of (spatial) pattern completion and CA1 would concentrate on prediction (i.e., temporal pattern completion). Simply put, activity in both subfields was decoded into a spatial position that was then compared to the past, current, or future position of the virtual rat. With the simulations, at the price of some necessary simplification, one can compare the performance of the differentiated circuit with a ‘‘uniform,’’ nondifferentiated circuit of equal number and type of components (one in which CA3 and CA1 have identical properties, e.g., both receive mossy fibers and are interconnected with recurrent collaterals). Lesion studies, instead, can only compare the normal circuit with others with missing components, thus it is difficult to assess the significance of a differentiation. The functional-differentiation hypothesis was not convincingly supported by neural network simulations. The conflict between spatial pattern completion, as quantified by localization accuracy, and temporal prediction indeed exists, but two mechanisms that would more directly relate to a functional CA3–CA1 differentiation were found unable to produce genuine prediction. Instead, a simple mechanism based on firing-frequency adaptation in pyramidal cells was found to be sufficient for prediction, with the degree of adaptation being the crucial parameter balancing retrieval with prediction. This is evident already from the simulations of the nondifferentiated model. The differentiation between the connectivity of CA3 and CA1 has a significant positive effect on this balance; in particular for each given anticipatory interval, it significantly increases, in the model, the information
339
content of hippocampal outputs, making the CA1 representation more informative than the CA3 one (or than the nondifferentiated one) when used to decode the position of the virtual rat (see Fig. 20–7). The CA1 representation is not shifted forward in time, however, with respect to the CA3 one. Different degrees of adaptation in CA3 and CA1 cells, moreover, were not found to lead to better performance, further undermining the notion of a full qualitative functional dissociation. There may, therefore, be just a plain quantitative advantage in differentiating the connectivity of the two fields, just as the hypothesis on isocortical lamination holds that there may be just a plain quantitative advantage in differentiating connectivity across layers. In a sense, the outcome of the simulations supports a revised version of the postprocessing clean-up notion. As Figure 20–4 shows, the information content in CA1, in the differentiated model, is higher than in CA3, with the nondifferentiated model midway between the two.
Correlated Environments Stimulate Orthogonal Ideas As for the lamination study, the analysis of this hypothesis about the differentiation of hippocampal
Figure 20–7. Localization accuracy as a function of cue size, after extensive training, in hippocampal network models with and without the differentiation between CA3 and CA1 (and with mossy fiber inputs). In the differentiated model, the position of the virtual rat is better decoded from the CA1 than from the CA3 representation. Simulation details are in Treves (2004).
340 THEORETICAL SIGNIFICANCE OF PLACE FIELDS subfields was based on the simulation of two simplified models, uniform and differentiated, tested on the same task—in this case, acquiring a memory chart for a single spatial environment. The accuracy of spatial memory retrieval is subject to the general ‘‘spin glass’’ limit, and it is further modulated by connectivity details. In particular, the spin glass limit implies that while more charts can be retrieved if they are sparser, their individual information content, and hence the localization accuracy they afford, is larger for less sparse charts. Thus, one expects that if CA3 representations were sparser than CA1 ones, each one could be retrieved more effectively from partial cues, but it would allow less accurate localization. A simulation in which CA3 and CA1 not only differ in connectivity but also in sparsity (aCA3 ¼ 0.16 and aCA1 ¼ 0.24) confirms this expectation (Fig. 20–8). Recording the activity of multiple hippocampal cells in the lab of Edvard and May-Britt Moser, Leutgeb et al. (2004) demonstrated a quantitative difference in sparsity (see Fig. 20–2) that goes in the direction of enhancing the retrieval of individual representations in CA3 and of increasing the localization accuracy offered by CA1 output activity. Their results, however, together with those obtained in the lab of James Knierim (Lee
Figure 20–8. Localization accuracy as a function of training session in hippocampal network models with and without the differentiation between CA3 and CA1, and with and without further differentiation in the sparsity of CA3 and CA1 representations. With reference to Figure 20–7, cue size here is Q ¼ 0.2. Other details are in Treves (2004).
et al., 2004), indicate a potentially much more dramatic differentiation between CA3 and CA1 units, which has to do with their ability to distinguish among several spatial environments. Activity in CA3 and CA1 was found to differ remarkably when rats were asked to explore environments that some cues suggested were the same, while others indicated they were different. CA3 appears to take an all-or-none decision, usually allocating nearly orthogonal neural representations to even very similar environments, and switching to essentially identical representations only above a high threshold of physical similarity. Activity in CA1, instead, varies smoothly to reflect the degree of similarity. This functional differentiation and the finding that new representations in CA3 emerge slowly, presumably through iterative processing, are entirely consistent with the recurrent character of the CA3 network and the prevailing feed-forward character of the CA1 network. Further surprises have come from applying a ‘‘morphing’’ paradigm to test spatial representations in environments quasi-continuously changed between two well-learned extremes (Leutgeb et al., 2005; Wills et al., 2005). In their original form (Treves, 2004), the connectivity differentiation models addressed the mechanism linking firing-rate adaptation to the prediction of spatial position within a single environment, and maybe because of that could not capture the essential advantage brought about by the connectivity differentiation, if it has to do with multiple maps. The experimental results have then stimulated the development of more elaborate computational models, which still have to satisfactorily find their way around the spin glass limit on memory retrieval. In fact, training virtual rats on several virtual environments, correlated or not, requires them to be endowed with large virtual brains. Simulations with networks of a thousand units or so, which were adequate for the single-environment case, have to be extended to networks larger by one or two orders of magnitude, which have become time consuming to simulate extensively. Even then, because of the heavy memory load for multiple environments (Battaglia and Treves, 1998), the representations tend to collapse on each other, making the comparison with real rat data more problematic (Papp and Treves, 2006). One observation that emerges from this study already at this stage is that CA3 representations tend to be more fragmented, in the sense that neural activity in pairs of separate locations can be identical or quite different, in violation of the metric nature of the environment. Comparatively, CA1 representations tend to be smoother, with a higher match between the distance among locations and the difference among their neural activity vectors. This smoothing function for CA1 resembles very much the clean-up notion originally investigated for discrete memories that was also relevant to the notion of prediction, implying
NETWORK ANALYSIS OF HIPPOCAMPAL PLACE FIELDS
continuity in time, but that now finds a more interesting role in reproducing the continuity of physical space. Thanks to the experimental findings with correlated environments, therefore, we may be beginning to finally understand CA1 and to make some (spatial) sense of the events that drastically altered the structure of our medial pallium hundreds of millions of years ago.
QUALITY VERSUS QUANTITY, AND THE NEED TO ADAPT As for the separation between the processing of ‘‘what’’ and ‘‘where’’ information in the sensory cortex model mentioned at the beginning, the hippocampal models require firing-rate adaptation, in this case, for producing a time-shifted localization, i.e., the prediction of future locations in a spatial environment. As for the sensory cortex model, memory retrieval is limited by the spin glass constraint. As for the sensory cortex model, the hypothesis is that a major qualitative structural change may have served to produce solely a quantitative functional advantage. In fact, it has been noted (Carroll, 1988) that often in evolution major steps may subserve only ‘‘small’’ improvements in survival ability. Although these hypotheses require much further testing, they serve to underscore the often subtle relations between structure and function that can apply to cortical networks, mediated by the collective, emergent dynamics of large populations of neurons.
acknowledgments We are grateful to all our colleagues in the lab of Edvard and May-Britt Moser for letting us participate in their investigations, and in particular to Stefan Leutgeb, who gave us the data used in Figures 20–2 and 20–3.
References Abeles M (1991) Corticonics: Neural Circuits of the Cerebral Cortex. Cambridge, UK: Cambridge University Press. Amaral DG, Ishizuka N, Claiborne B (1990) Neurons, numbers and the hippocampal network. Prog Brain Res 83:1–11. Amit DJ (1989) Modeling Brain Function. Cambridge, UK: Cambridge University Press. Amit DJ, Gutfreund H, Sompolinsky H (1987) Statistical mechanics of neural networks near saturation. Ann Phys (NY) 173:30–67. Battaglia FP, Treves A (1998) Attractor neural networks storing multiple space representations: a model for hippocampal place fields. Phys Rev E 58:7738–7753.
341
Bingman VP, Jones TJ (1994) Sun-compass based spatial learning impaired in homing pigeons with hippocampal lesions. J Neurosci 14:6687–6694. Bliss TV, Lomo T (1973) Long-lasting potentiation of synaptic transmission in the dentate area of the anaesthetized rabbit following stimulation of the perforant path. J Physiol (Lond) 232:331–356. Braitenberg V, Schu¨z A (1991) Anatomy of the Cortex. Berlin: Springer-Verlag. Carroll RL (1988) Vertebrate Paleontology and Evolution. New York: W. H. Freeman. Clayton NS, Bussey TJ, Dickinson A (2003) Can animals recall the past and plan for the future? Nat Rev Neurosci 4:685–691. Clayton NS, Krebs JR (1995) Memory in food-storing birds: from behaviour to brain. Curr Opin Neurobiol 5:149–154. Collingridge GL, Bliss TV (1995) Memories of NMDA receptors and LTP. Trends Neurosci 18:54–66. Diamond IT, Conley M, Itoh K, Fitzpatrick D (1985) Laminar organization of geniculocortical projections in Galago senegalensis and Aotus trivirgatus. J Comp Neurol 242:584–610. Diamond IT, Hall WC (1969) Evolution of neocortex. Science 164:251–262. Eichenbaum H (2000) A cortical-hippocampal system for declarative memory. Nat Rev Neurosci 1:41–50. Finlay BL, Darlington RB (1995) Linked regularities in the development and evolution of mammalian brains. Science 268:1578–1584. Fyhn M, Molden S, Witter MP, Moser EI, Moser M-B (2004) Spatial representation in the entorhinal cortex. Science 305:1258–1264. Fyhn M, Hafting T, Treves A, Moser MB, Moser EI (2007) Hippocampal remapping and grid realignment in entorhinal cortex. Nature 446:190–194. Gardner-Medwin AR (1976) The recall of events through the learning of associations between their parts. Proc R Soc Lond B 194:375–402. Hafting T, Fyhn M, Molden S, Moser M-B, Moser EI (2005) Microstructure of a spatial map in the entorhinal cortex. Nature 436:801–806. Hasselmo ME, Schnell E (1994) Laminar selectivity of the cholinergic suppression of synaptic transmission in rat hippocampal region CA1: computational modeling and brain slice physiology. J Neurosci 14:3898– 3914. Hasselmo ME, Schnell E, Barkai E (1995) Dynamics of learning and recall at excitatory recurrent synapses and cholinergic modulation in rat hippocampal region CA3. J Neurosci 15:5249–5262. Hasselmo ME, Wyble B, Wallenstein G (1996) Encoding and retrieval of episodic memories: role of cholinergic and GABAergic modulation in the hippocampus. Hippocampus 6:693–708.
342 THEORETICAL SIGNIFICANCE OF PLACE FIELDS Hopfield JJ (1982) Neural networks and physical systems with emergent collective computational abilities. Proc Natl Acad Sci USA 79:2554–2558. Jerison HJ (1990) Fossil evidence on the evolution of the neocortex. In: Cerebral Cortex, Volume 8A: Comparative Structure and Evolution of Cerebral Cortex, Part I. (Jones EG, Peters A, eds.), pp 285–309. New York: Plenum Press. Kesner RP, Gilbert PE, Lee I (2002) Subregional analysis of hippocampal function in the rat. In: Neuropsychology of Memory, 3rd ed. (Squire LR, Schacter DL, eds.), pp 395–411. New York: Guilford Press. Laplagne DA, Esposito MS, Piatti VC, Morgenstern NA, Zhao C, van Praag H, Gage FH, Schinder AF (2006) Functional convergence of neurons generated in the developing and adult hippocampus. PLoS Biol 4:e409. Lassalle J-M, Bataille T, Halley H (2000) Reversible inactivation of the hippocampal mossy fiber synapses in mice impairs spatial learning, but neither consolidation nor memory retrieval, in the Morris navigation task. Neurobiol Learn Mem 73:243–257. Lee I, Kesner RP (2004) Encoding versus retrieval of spatial memory: double dissociation between the dentate gyrus and the perforant path inputs into CA3 in the dorsal hippocampus. Hippocampus 14:66–76. Lee I, Yoganarasimha D, Rao G, Knierim JJ (2004) Comparison of population coherence of place cells in hippocampal subfields CA1 and CA3. Nature 430:456–459. Leutgeb JK, Leutgeb S, Treves A, Meyer R, Barnes CA, McNaughton BL, Moser M-B, Moser EI (2005) Progressive transformation of hippocampal neuronal representations in ‘‘morphed’’ environments. Neuron 48:345–358. Leutgeb JK, Leutgeb S, Moser M-B, Moser EI (2006) Expansion recoding is expressed in CA3 but not in the dentate gyrus. Abstr Soc Neurosci 32:60.3. Leutgeb S, Leutgeb JK, Treves A, Moser M-B, Moser EI (2004) Distinct ensemble codes in hippocampal areas CA3 and CA1. Science 305:1295–1298. Marr D (1971) Simple memory: a theory for archicortex. Philos Trans R Soc Lond B Biol Sci 262:23–81. McNaughton BL, Morris RGM (1987) Hippocampal synaptic enhancement and information storage within a distributed memory system. Trends Neurosci 10:408–415. O’Keefe J, Dostrovsky J (1971) The hippocampus as a spatial map: preliminary evidence from unit activity in the freely moving rat. Brain Res 34:171–175. Papp G, Treves A. (2006) Continuous attractors come fragmented. Soc Neurosci Abs 68:13. Rawlins JNP (1985) Associations across time: the hippocampus as a temporary memory store. Behav Brain Sci 8:479–528. Ramirez-Amaya V, Marrone DF, Gage FH, Worley PF, Barnes CA (2006) Integration of new neurons into functional neural networks. J Neurosci 26:12237–12241.
Rolls ET (1989) Functions of neuronal networks in the hippocampus and cerebral cortex in memory. In: Models of Brain Function (Cotterill R, ed.), pp 15–33. Cambridge, UK: Cambridge University Press. Roudi Y, Treves A (2006) Localized activity profiles and storage capacity of rate-based autoassociative networks. Phys Rev E Stat Nonlin Soft Matter Phys 73(6 Pt 1):061904. Samsonovich A, McNaughton BL (1997) Path integration and cognitive mapping in a continuous attractor neural network model. J Neurosci 17:5900. Skaggs WE, McNaughton BL, Gothard K, Markus E (1993) An information theoretic approach to deciphering the hippocampal code. Adv Neural Inf Proc Sys 5:1030–1037. Sompolinsky H, Shapley R (1997) New perspectives on the mechanisms for orientation selectivity. Curr Opin Neurobiol 7:514–522. Treves A (1995) Quantitative estimate of the information relayed by the Schaffer collaterals. J Comput Neurosci 2:259–272. Treves A (2003) Computational constraints that may have favoured the lamination of sensory cortex. J Comput Neurosci 14:271–282. Treves A (2004) Computational constraints between retrieving the past and predicting the future, and the CA3-CA1 differentiation. Hippocampus 14:539–556. Treves A (2005) Frontal latching networks: a possible neural basis for infinite recursion. Cogn Neuropsychol 22:276–291. Treves A, Rolls ET (1991) What determines the capacity of autoassociative memories in the brain? Network 2:371–397. Treves A, Rolls ET (1992) Computational constraints suggest the need for two distinct input systems to the hippocampal CA3 network. Hippocampus 2:189–199. Ulanovsky N, Moss CF (2007) Hippocampal cellular and network activity in freely moving echolocating bats. Nat Neurosci 10:224–233. Ulinski PS (1990) The cerebral cortex of reptiles. In: Cerebral Cortex, Vol. 8A: Comparative Structure and Evolution of Cerebral Cortex (Jones EG, Peters A, eds.), pp 139–215. New York: Plenum Press. Wills TJ, Lever C, Cacucci F, Burgess N, O’Keefe J (2005) Attractor dynamics in the hippocampal representation of the local environment. Science 308:873–876. Willshaw D, Buckingham J (1990) An assessment of Marr’s theory of the hippocampus as a temporary memory store. Philos Trans R Soc Lond B Biol Sci 329:205– 215. Wilson EO (1975) Sociobiology. The New Synthesis. Cambridge, MA: Harvard University Press. Wilson MA, McNaughton BL (1993) Dynamics of the hippocampal ensemble code for space. Science 261: 1055–1058.
21 Storage of the Distance between Place Cell Firing Fields in the Strength of Plastic Synapses with a Novel Learning Rule ¨ RGY CSIZMADIA AND ROBERT U. MULLER GYO
A cognitive map is a stored neural representation of an environment that permits an animal to solve navigational problems using information about the structure or geometry of the environment (see Gallistel, 1990). According to a very influential theory (O’Keefe and Nadel, 1978), place cells in the hippocampus are essential units of the cognitive map in rats and mice. Anatomically, place cells are pyramidal cells of the CA1 and CA3 regions of the hippocampus and are characterized by location-specific firing; they are intensely active only when the rat’s head is in a cellspecific part of the environment, called the ‘‘firing field,’’ and are virtually silent when the head is elsewhere. Place cells are stable over long periods of time (Muller et al., 1987; Thompson and Best, 1990) and in open areas their firing is independent of the direction in which the rat’s head points (Muller et al., 1994). How can place cells be combined to form a cognitive map? Several theories have been advanced (Touretzky and Redish, 1996; Wiener, 1996; Burgess et al., 1997; Gerstner and Abbott, 1997; Samsonovich and McNaughton, 1997; Redish, 1999; Koene et al., 2003; O’Keefe, 2007), but here we specifically address one in which the highly interconnected pyramidal cells of CA3 store a map in the strengths of the recurrent, long-term potentiation (LTP)-modifiable synapses made by pairs of CA3 place cells (Muller et al, 1991; Hetherington and Shapiro, 1993; Muller et al., 1996). In this model, the distance between the firing fields of connected pairs of CA3 cells is encoded as synaptic resistance (reciprocal synaptic strength). The encoding occurs because pairs of cells with coincident or over-
lapping fields will tend to fire together in time, thereby causing a decrease in synaptic resistance via LTP (Bliss and Collingridge, 1993). Cells with widely separated fields will tend never to fire together, causing no change in the synaptic resistance. If the encoding of distance as synaptic resistance is ideal (i.e., the synaptic resistance is a monotonic function of distance), it is possible to use graph theory to demonstrate that the CA3 network allows the shortest path between any pair of points to be calculated, so that the network has an essential property of a Euclidean map.1 Assume that the environment is covered with overlapping firing fields, and choose any place in the environment as the starting point and any other place as the end point of a path. Each of these places corresponds to the firing-field center of (at least) one place cell. Now identify a cell whose field is at the start and another cell whose field is at the goal and compute (using Dijkstra’s algorithm; see, for example, Cormen et al., 1992) the sequence of cells in the network that minimizes the sum of the synaptic resistances from the start cell to the goal cell. Since each of these cells is a place cell, the selected sequence of cells corresponds to a sequence of firing field positions in the environment or, in other words, to a path through the environment from start to end. It was previously shown that such a path will be indistinguishable from a straightline segment, so long as the density of place cells and the density of connections between place cell pairs are great enough (Muller et al., 1996). The idea that the hippocampal map has the form of a Euclidean graph of the environment is far from
343
344 THEORETICAL SIGNIFICANCE OF PLACE FIELDS complete. For example, it is plausible that the start of a path is signaled by place cells whose fields are at the rat’s current location, but it remains to be shown if the excess activity seen in many place cells while a rat is at a goal location (Hok et al., 2007) can be used to create a graph-searching algorithm based on realistic neural mechanisms. Our purpose in this chapter is to address a second major problem with the original theory, namely, the way in which information acquired during exploration is translated into the static structure of synaptic resistances. In previous models, nonuniform exploration distorted the map so that the synaptic resistance for a given distance in a familiar part of the environment was always lower than the synaptic resistance for the same distance in a relatively unexplored part of the environment. As a result of this distortion, paths found by the graph-searching algorithm were biased toward familiar places and routes, as illustrated in Figure 21–1. Here we propose a scheme that drastically reduces the bias of producing paths that go through more highly explored parts of the environment. By use of a modified version of the Hebbian learning rule, the representation of the environment always tends to become more accurate with increasing exploration time so that eventually the representation asymptotically becomes exact. In addition to its ability to solve an interesting problem, the new learning rule is biologically plausible. In the discussion we argue that the novel requirement is the existence of a negative feedback pathway from AMPA receptors to NMDA receptors such that additional strengthening via NMDA receptor activation is inhibited in already strengthened synapses. We cite several example phenomena that can be explained in this way and cite a study that indicates a form of the necessary mechanism exists.
A NEW SYNAPTIC STRENGTHENING RULE We will first consider difficulties with the Hebbian rule for strengthening synapses that connect pairs of place cells, and then present the modified rule that solves the problems. The firing rate, f, of a place cell is defined as the number of spikes discharged per unit time over a convenient averaging interval; we imagine the interval is on the order of 0.1 s, an estimate of the amount of time it takes a rat to move by an amount that has an important effect on the time-averaged firing rate as a function of position. If fpre is the firing frequency of the presynaptic place cell and fpost the firing frequency of the postsynaptic place cell when the rat is at a particular place in the apparatus, the Hebbian rule for synaptic strength (S) is (Johnston and Wu, 1995; Kohonen, 1995):
S ¼ Sold þ fpre fpost
(1)
The product fpre fpost depends, in a so far unstated way, on the distance to the firing fields of the presynaptic and postsynaptic cells. For any dependency in which the rate is a nonincreasing function of the distance between the rat and the field center, however, the greatest possible product will occur when the distance between the fields is zero and the rat’s head is at the position of the coincident centers. Moreover, the product must approach zero when the distance between the field centers becomes large, because the firing rate of one or both cells must be zero regardless of the rat’s position. One difficulty with the Hebbian rule is that S can increase without limit unless a negative correction term is included (Kohonen, 1995). Moreover, in the case of place cells, the magnitude of the synaptic strength depends not just on distance but on the time spent in the vicinity of the two firing fields so that synapses strengthened according to the Hebbian rule cannot uniquely represent distance. We now propose that synaptic strength (S) changes occur according to a novel version of the Riccati learning law (Kohonen, 1995): DS ¼
0 if S > f pre f post a(f pre f post S) otherwise
(2)
Thus, if the product of the frequencies of the presynaptic and postsynaptic cells is less than the current synaptic strength, the strength is unchanged. On the other hand, if the product of the frequencies is greater than the current synaptic strength, the strength is increased, so it gets closer to the product of the frequencies. As shown below, as exploration time increases, this ‘‘threshold’’ strengthening rule ensures that synaptic resistance asymptotically approaches the correct value and, therefore, provides a unique encoding of distance as synaptic resistance.2 An additional asset of the rule is that ‘‘reading’’ the strength of the synapse with fpre fpost < S can occur without altering the synaptic strength, a highly desirable property. To use any strengthening rule to build a map, it is necessary to specify how firing rate depends on distance from the field center. Muller et al. (1996) showed that the exact form of this relationship had little effect on the ability to find near-ideal paths in the environment by searching the graph. For this reason and because place cell recordings show that firing rates decrease in an approximately exponential way with distance from the field center, we take the rate-distance function to be f ¼ e kd
PLACE CELL FIRING FIELDS DURING NOVEL LEARNING
345
Figure 21–1. Schematic drawing of how excess dwell time of a rat in a preferred region can interfere with computing optimum paths from start to goal. A. With homogeneous exploration (top) the graphsearching algorithm finds a straight-line path from start to goal (bottom), even if weights are set by the conventional Hebbian synaptic-modification rule. B. If the rat prefers to spend more of its time in a certain region of the apparatus (top), paths computed by graph search will tend to bend toward the preferred region (bottom). C. If synaptic weights are set by the modified (‘‘threshold’’) learning rule, the presence of a preferred region (top) does not distort paths computed by graph search.
where d is the rat’s distance from the firing field center. One advantage of this approximation is that with the proper choice of the constant, k, the firing rate rapidly falls to zero outside the immediate region of the firing field center.3 How does S change according to the threshold rule as the rat explores the environment? With the exponential rate-distance function, the rule becomes DS ¼
0 a(e ka e kb S)
if e ka e kb < S otherwise:
where a is the distance from the rat’s position to the center of the presynaptic cell’s firing field and b is the distance from the rat’s position to the center of the postsynaptic cell’s firing field, as shown in Figure 21–2. Note that e ka e kb ¼ e kða þ bÞ , so the synaptic strength is a function of the sum of these distances.
Before exploration, S . 0, which is equivalent to saying that the representation of the summed distance from the rat to the two field centers is very large. When the rat gets into the vicinity of both fields (which is possible if the centers are not very far apart), the summed distance a þ b to the field centers decreases and the synaptic strength increases. The current value of the distance between the two field centers is computed as a þ b ¼ ln(S)k: Under what circumstances does S continue to increase? If the real distance between the fields of the presynaptic and postsynaptic cell is d, the synaptic strength will reach a maximum when a þ b ¼ d, or, in other words, if the rat’s head stays on the line segment between the field centers long enough for the
346 THEORETICAL SIGNIFICANCE OF PLACE FIELDS gion very well, S approaches Sideal . The time it takes to reach Sideal depends on the averaging time for computing the product of the presynaptic and postsynaptic frequencies. If fpre fpost S is computed continuously, S will reach Sideal as soon as the rat happens to stay on the line segment connecting the field centers for a short time. If DS is computed periodically, it may never reach Sideal if the rat happens not to be on the line segment at computation time. Nevertheless, the difference between S and Sideal will get arbitrarily small with continued exploration.
Rat
b a
d
DISCUSSION Center of Field 2 Center of Field 1
Figure 21–2. Setting synaptic strength according to the distance of a rat’s head from the centers of two nearby firing fields. The head is distance a from the center of field 1 and distance b from the center of field 2. If the firing rate decreases exponentially away from each field center with a distance constant k, the rate of each cell is determined. The product of presynaptic and postsynaptic rates is then given by ek(a þ b). If this quantity is greater than the current synaptic strength S, the change in strength DS is set to a constant times ek(a þ b) S; otherwise S is unchanged. If the rat’s head position is on the segment that connects the two field centers and stays there long enough, the synaptic strength reaches a maximum value ekd since d is the minimum sum for a and b.
presynaptic and postsynaptic firing rates to begin to represent the distances to the field centers. The requirement for dwell on the line segment arises because the firing frequencies are averaged over a short time interval on the order of 0.1 s so that, for a continuously moving rat, firing frequencies represent an average of the distances in the time interval. When this happens, S ¼ Sideal and synaptic strength can get no greater. In addition, if the firing rate is set to zero when the distance from the field center is greater than some constant, Sideal can be reached only if the fields overlap; otherwise, S remains 0. One other noteworthy feature of the threshold rule is that S tends toward a maximum even though there is no provision for decreasing S due to processes like long-term depression. In fact, if the rat explores a re-
In previous work on neural navigational systems, we analyzed recurrently connected (CA3-like) networks of place cells. The place cells in these networks were connected by synapses whose resistances increased nondecrementally with the distance between the firing fields of the presynaptic and postsynaptic cells. We found that such networks could be used to compute optimal paths through the environment using the methods of graph theory. To calculate such paths, an arbitrary position in the environment is selected as a starting point and a second arbitrary point is selected as a goal. Next, a place cell whose field is near the starting point and a second place cell whose field is near the goal are chosen. The network is then exhaustively searched (using Dijkstra’s algorithm) for all sequences of cells that connect the starting cell to the goal cell. Finally, the sequence along which the sum of the synaptic resistances is lowest is identified. Since each cell along the sequence is a place cell, the sequence corresponds to a path in space from the starting point to the goal. We showed that the path in space asymptotically approached a straight-line segment as the average divergence (outdegree) of the network increased. Even accepting the premise that the nervous system can compute optimal paths in this way, there are several major problems with this model. First, the synaptic resistances in the original model were set using explicit functions of the distance between the firing fields of each connected cell pair. The simple relationship between distance and synaptic resistance could be achieved with a Hebbian learning rule only by the unrealistic assumption that exploration of the environment is uniform. This requirement for uniform exploration arises because Hebbian schemes are integrative, so that synaptic strength depends not only on distance but also on the amount of time spent in the vicinity of the firing fields of the presynaptic and postsynaptic cells. The work here shows it is possible to set synaptic weights according to distance using a modified Hebbian scheme. In the novel synaptic
PLACE CELL FIRING FIELDS DURING NOVEL LEARNING
strengthening rule, the new weight is equal to either the current weight or to the product of the presynaptic and postsynaptic firing rates, which ever is higher. This rule reduces the dependence of synaptic strength on tendencies of the rat to prefer certain regions of the environment, ensures that the accuracy of distance encoding improves with increasing exploration, and reduces encoding errors to zero as exploration increases without limit. Having solved the problem of uniform exploration, a new difficulty arises concerning the uniformity of firing-field properties. In the model developed here, the ability to represent distance as synaptic strength assumes that all firing fields have the same circular shape, the same peak rate, and the same rate contours away from the peak. In reality, however, field shape, intensity, and size all vary considerably, so that the final values of synaptic strengths will not precisely encode distance. We imagine, however, that a graphsearching algorithm can find optimal paths without precise encoding if there are many place cells with centers in each small region in the environment. A second difficulty with the graph scheme concerns the issue of goal representation. For a rat to run along a best path that minimizes the sum of synaptic resistances, the goal location must in some way be taken into account while the rat is at the start of the path. It was recently shown that a large fraction of place cells with fields away from an unmarked goal discharge an excess number of spikes (i.e., have a secondary field) at the goal, while a hungry rat pauses there to cause a food pellet to drop into the apparatus. This cross-cell signal may be an internal confirmation that the rat thinks it is at the correct location of goal location (Hok et al., 2007). It may also be a means of enhancing the excitability of place cells whose fields happen to be at the goal location, with the result that such cells could discharge at the moment the rat decided to go to toward the goal, regardless of its current location. Discharge of this kind might be revealed by an analysis similar to that used by Hok et al. (2007). We close by briefly addressing a different kind of biological plausibility, that of the threshold learning rule. We argue that this rule has an interesting and testable interpretation in terms of known properties of NMDA-based synaptic modification mechanisms in the hippocampus. It is accepted that NMDA channels open when presynaptically released glutamate binds to their receptors only if the postsynaptic cell is depolarized enough to relieve the Mg2þ blockade of the channel lumen. Thus NMDA channels act as coincidence detectors for presynaptic and postsynaptic activity and thereby compute an approximation to the product fpre fpost (Bliss et al., 2007).
347
A consequence of NMDA channel opening is to allow extracellular Ca2þ to enter the postsynaptic cell, beginning a series of reactions that culminate in increased synaptic strength. The end point of this process is an increase of the conductance due to AMPA channels, a second type of glutamate channel whose opening depends only on the extracellular glutamate released by the presynaptic cell. It is unclear if the AMPA-based conductance increases because there are more postsynaptic AMPA channels, because the conductance per channel is enhanced, because of increased glutamate release, or a combination of these mechanisms, but in any case the AMPA conductance is an approximation to the current synaptic strength S in equation 2. If the product fpre fpost reflects the degree of activation of NMDA channels and S reflects the degree of activation of AMPA channels, the difference fpre fpost S can be computed at individual synapses if activation of AMPA channels can reduce or prevent the consequences of NMDA channel activation. The learning rule in equation 2 can therefore be realized if activation of AMPA conductance generates a signal that inhibits the opening of the NMDA channels (or the consequences of their opening) to an extent proportional to the AMPA conductance. This picture is in agreement with several recent observations on the properties of NMDA-based synaptic modification. First, Bi and Poo (1998) find that the ability to potentiate synapses is an inverse function of the initial strength of the synapse; it is easier to potentiate weak synapses. This effect was taken by Bi and Poo to reflect saturation of the potentiation process, but it may instead be a consequence of our proposed learning rule. In support of the second interpretation are the results of Jia et al. (1996), who measured LTP in slices from mice that lack the GluR2 subunit. AMPA channels without the GluR2 subunit are permeable to Ca2þ, so that LTP can be initiated by extracellular glutamate regardless of whether NMDA channels open. In slices from these mice, separate trains of high-frequency stimulation progressively increased synaptic strength by up to nine-fold, which suggests that the limit of LTP in normal hippocampal slices is not due to saturation. Another indication that AMPA conductance may serve to limit the effects of NMDA channel opening is provided by the effects of blocking AMPA channels in developing cultures of hippocampal neurons. Liao et al. (1999) found that such blockade increases the number and size of AMPA channel clusters in such cultures. This unexpected effect is explained by proposing that the lack of AMPA channel opening allows the processes started by the opening NMDA channels to go on in an unfettered way. Finally, there is direct evidence that AMPA channel activity can inhibit NMDA channel opening (Bai et al.,
348 THEORETICAL SIGNIFICANCE OF PLACE FIELDS 2002). Thus, application of AMPA receptor agonists decreases NMDA-based conductance, whereas application of antagonists has the opposite effect. This coupling was shown to be non-ionotropic, implying the existence of a chemical signal from AMPA channels to NMDA channels of the sort required by the thresholdstrengthening rule. We conclude, therefore, that the proposed synaptic learning rule may in fact exist at CA3 ! CA3 recurrent connections. If so, it may be that optimal paths through the environment are computed via a neural implementation of a graph theoretic algorithm that minimizes synaptic resistance along place cell sequences that connect a starting location to a goal.
acknowledgments The work presented here was supported by NIH NS20686.
Notes 1. A map formed by place cells connected by synapses that encode distance can also be used to find detours and shortcuts (Muller et al., 1996). 2. The generality of the threshold rule can be enhanced by making a a function of S such that 1 a 0. By choosing a to be a monotonically decreasing function of S (such as 1/S), we can model the observation of Bi and Poo (1998) that it is easier to potentiate weak than strong synapses. 3. Any function in which frequency decreases monotonically with distance could be used instead of the exponential, as long as it has an inverse; the exponential function was chosen as a matter of convenience. For example, we could use the function f ¼ 1/d and set the synaptic strength differently, such as S ¼ eð 1=f 1Þ eð 1=f 2Þ . Then ln(S) is proportional to the distance between the place cell field centers.
References Bai D, Muller RU, Roder JC (2002) Non-ionotropic crosstalk between AMPA and NMDA receptors in rodent hippocampal neurones. J Physiol 543(Pt 1):23–33. Bi G, Poo MM (1998) Synaptic modifications in cultured hippocampal neurons: dependence of spike timing, synaptic strength and postsynaptic cell type. J Neurosci 24:10464–10472. Bliss TV, Collingridge GL (1993). A synaptic model of memory: long-term potentiation in the hippocampus. Nature 361:31–39. Bliss T, Collingridge G, Morris R (2007) Synaptic plasticity in the hippocampus. In: The Hippocampus Book (Andersen P, Andersen R, Morris, R, Amaral D, Bliss T, O’Keefe J, eds.). New York: Oxford University Press.
Burgess N, Donnett JG, Jeffery KJ, O’Keefe J (1997) Robotic and neuronal simulation of the hippocampus and rat navigation. Philos Trans R Soc Lond B Biol Sci 352:1535–1543. Cormen TH, Leiserson CE, Rivest RL (1992) Introduction to Algorithms. Cambridge, MA: MIT Press. Gallistel CR (1990) The Organization of Learning. Cambridge, MA: MIT Press. Gerstner W, Abbott LF (1997) Learning navigational maps through potentiation and modulation of hippocampal place cells. J Comput Neurosci 4:79–94. Hetherington PA, Shapiro ML (1993) A simple network model simulates hippocampal place fields: II. Computing goal-direct trajectories and memory fields. Behav Neurosci 107:434–443. Hok V, Lenck-Santini PP, Roux S, Save E, Muller RU, Poucet B (2007) Goal-related activity in hippocampal place cells. J Neurosci 27:472–482. Jia Z, Agopyan N, Miu P, Xiong Z, Henderson J, Gerlai R, Taverna FA, Velumian A, MacDonald J, Carlen P, Abramow-Newerly W, Roder J (1996) Enhanced LTP in mice deficient in the AMPA receptor GluR2. Neuron 17:945–956. Johnston D, Wu SM (1995) Foundations of Cellular Neurophysiology. Cambridge, MA: MIT Press. Koene RA, Gorcehtchnikov A, Cannon RC, Hasselmo ME (2003) Modeling goal-directed spatial navigation in the rat based on physiological data from the hippocampal formation. Neural Netw 16:577–584. Kohonen T (1995) Self-Organizing Maps. Berlin: SpringerVerlag. Liao D, Zhang X, O’Brien R, Ehlers M, Huganir R (1999) Regulation of morphological postsynaptic synapses in developing hippocampal neurons. Nat Neurosci 2: 37–43. Muller RU, Bostock E, Taube JS, Kubie JL (1994) On the directional firing properties of hippocampal place cells. J Neurosci 14:7235–7251. Muller RU, Kubie JL, Ranck JB Jr (1987) Spatial firing properties of hippocampal complex-spike cells in a fixed environment. J Neurosci 7:1935–1950. Muller RU, Kubie JL, Saypoff R (1991) The hippocampus as a cognitive graph (abridged version). Hippocampus 1:243–246. Muller RU, Stead M, Pach J (1996) The hippocampus as a cognitive graph. J Gen Physiol 107:663–694. O’Keefe J (2007) Hippocampal physiology in the behaving animal. In: The Hippocampus Book (Andersen P, Andersen R, Morris, R, Amaral D, Bliss T, O’Keefe J, eds.). New York: Oxford University Press. O’Keefe J, Nadel L (1978) The Hippocampus as a Cognitive Map. Oxford: Clarendon Press. Redish AD (1999) Beyond the Cognitive Map: From Place Cells to Episodic Memory. Cambridge, MA: MIT Press.
PLACE CELL FIRING FIELDS DURING NOVEL LEARNING
Samsonovich A, McNaughton BL (1997) Path integration and cognitive mapping in a continuous attractor neural network model. J Neurosci 17:5900–5920. Thompson LT, Best PJ (1990) Long-term stability of the place-field activity of single units recorded from the dorsal hippocampus of freely moving rats. Brain Res 509:299–308.
349
Touretzky DS, Redish AD (1996) Theory of rodent navigation based on interacting representations of space. Hippocampus 6:247–270. Wiener SI (1996) Spatial, behavioral and sensory correlates of hippocampal CA1 complex spike cell activity: implications for information processing functions. Prog Neurobiol 49:335–361.
This page intentionally left blank
V PLACE FIELDS AND AGE-RELATED CHANGES IN MEMORY
This page intentionally left blank
22 Hippocampal Place Cells as a Window into Age-related Memory Impairments IAIN A. WILSON AND HEIKKI TANILA
Following the pioneering discovery of hippocampal place cells by John O’Keefe and colleagues over 30 years ago (O’Keefe and Dostrovsky, 1971; O’Keefe and Nadel, 1978), several characteristics have emerged that make these cells a remarkable tool for investigating how memories are made. First, place cells show distinct spatial representations for distinct environments, and second, they demonstrate a striking longterm stability of these representations. Recently researchers have begun to take full advantage of the many potential applications of this unique phenomenon. There are several approaches to exploring the mechanisms underlying memory formation. Some place cell studies start with individuals who have learned a task and investigate the mechanisms that enable memory formation (O’Keefe and Speakman, 1987; Lenck-Santini et al., 2001, 2002). Another approach is to study individuals who naturally have difficulties forming memories and ask what causes the memory failure. Because the cognitive abilities of normal aged individuals show a large degree of variability, aging presents the fascinating possibility to study both those individuals with good memory abilities and those with poor abilities. In our own studies we have used hippocampal place cells as a tool to assess changes in hippocampal information processing that may underlie failing spatial memory in aged individuals. We will show here that place cells indeed can provide a window into the individual differences associated with cognitive aging.
HUMAN AGING Although the normal aging brain may undergo a general slowing of information-processing speed (Salthouse, 1996), not all kinds of memory processing are impaired with normal aging (Gallagher and Rapp, 1997). Indeed, to store information the brain makes use of multiple long-term memory systems (Cohen and Eichenbaum, 1993; Milner et al., 1998). Procedural memories are skills or habits learned through repetition of trials, and memories of this class remain intact with aging (Rodrigue et al., 2005; Smith et al., 2005). In contrast, declarative memories concern facts, events, or relationships that can be consciously recalled (Cohen and Eichenbaum, 1993), and memories of this class often fail with aging (Gallagher and Rapp, 1997). Much of the research in human declarative memory has used experiments based on language, but with verbal tests it is not possible to use other species to examine the mechanisms of age-related memory decline. Spatial memory, by contrast, is a fundamental cross-species function and therefore represents a key means of studying memory impairments through animal models. Human aging is associated with declines in the ability to remember spatial relationships among landmarks(reviewedbyRosenzweigandBarnes,2003). For example, aged subjects are impaired at drawing from memory a studied spatial map and have difficulties in navigating the route memorized from that studied map (Wilkniss et al., 1997).
353
354 PLACE FIELDS AND AGE-RELATED CHANGES IN MEMORY A key feature of normal aging in humans is the large degree of variation between individuals in their cognitive abilities: some aged subjects reliably perform successfully on tests of spatial memory, whereas other aged subjects consistently fail on tasks dependent on spatial memories (Monacelli et al., 2003). Monacelli and colleagues (2003) tested young, middle-aged, and healthy elderly adults and patients with Alzheimer’s disease on their ability to navigate in a familiar hospital by memory. Almost all Alzheimer’s disease patients and 38% of normal aged subjects made errors of spatial navigation, but other normal aged subjects as well as all middle-aged and younger subjects had no navigation difficulties. This wide spectrum of cognitive abilities within the aged population, ranging from memory abilities comparable to those of young adults to memory impairments, is a phenomenon that has also become well-established in a rat model of neurocognitive aging, allowing investigations into the mechanisms underlying age-related memory impairments (Gallagher et al., 1993, 2006).
WATER MAZE IN AGING AND PLACE CELL RECORDINGS The animal model of cognitive aging that our research group has focused on evaluates the spatial memory of rats through the Morris water maze task. The water maze requires the rat to use distal landmarks to navigate in an open pool to an escape platform hidden under the water surface. This task has become the most widely used means to evaluate the spatial memory of rats for several reasons. First, the rats search for the hidden platform with a motivation that is uniform and almost guaranteed on every trial. Second, learning occurs relatively fast in the water maze; rats can navigate successfully to the platform within a few training trials (Morris et al., 1982). Third, impairments on the spatial water maze task can be firmly attributed to spatial memory deficits rather than procedural deficits; for example, rats with spatial memory difficulties in finding the hidden platform are capable of learning to navigate to a cued platform (Morris et al., 1982). Fourth, ability on the water maze shows a strong brain structure– function relationship; hippocampal lesions—and, more precisely, septal hippocampal lesions—impair spatial memory performance on the task without affecting procedural aspects (Morris et al., 1982; Moser and Moser, 1998). Aged rodents show clear deficits in the spatial water maze, and, supporting the validity of the model, aged people are also significantly impaired on dry-land versions of the water maze. Aged human subjects were impaired relative to young ones in identifying the
target location based on the relationship of spatial cues (Newman and Kaszniak, 2000). In a virtual computer version of the water maze, the young subjects focused their searches nearer to the platform than did the aged subjects. Furthermore, the aged subjects successfully used objects proximal to the target but not distal spatial cues to aid navigation (Moffat and Resnick, 2002). With robust impairments in aging humans and animals, the spatial water maze has become the goldstandard screening evaluation for aging research. Our collaborators have developed a systematic water maze testing routine that quantifies the cognitive ability of young and aged rats and highlights robust individual differences between aged Long-Evans rats reminiscent of the individual differences seen in aged humans (Gallagher et al., 1993). Some of the aged rats learn as well as the young rats to locate the submerged escape platform, whereas other aged rats are severely impaired in learning the platform location (Gallagher et al., 1993). For each rat, a spatial learning index is derived from the average swimming distance from the location of the hidden platform during probe trials (see Fig. 22–1). Importantly, these individual differences in hippocampal-dependent performance are reliable. If tested in a new environment weeks later, rats that were impaired in the initial assessment again perform poorly, whereas the rats that demonstrated preserved function remain proficient at the task (Gallagher and Burwell, 1989; Gallagher et al., 2006). The comparison between good aged performers and poor aged performers on a hippocampus-dependent water maze task allows insights into the origins of both cognitive aging and consistent behavioral outcomes. A key question, therefore, is how hippocampal information processing differs between aged impaired, aged unimpaired, and young rats. Hippocampal place cells serve as an important window into how these individuals process spatial information. The goal of the research discussed here has been to find characteristics of place cells that correlate strongly with memory abilities of the rats. To investigate this, ideally one would record place cells during performance of the spatial water maze task itself. Unfortunately, electrophysiological recordings during the actual water maze task are impractical, because of two main problems. First, water provides a noisy electrophysiological recording environment as it is difficult to ground. Second, sampling of the entire environment is critical for place cell recordings because this allows visualization of all active place cells. In a well-learned water maze, such thorough sampling is unlikely to occur; as the rat learns the platform location and begins to swim directly to it, sampling of the maze environment becomes limited. The Moser laboratory in Trondheim, Norway, has cleverly used an annular water
PLACE CELLS AND AGE-RELATED MEMORY IMPAIRMENTS
Figure 22–1. Learning index scores from behavioral characterization of individual rats used in the recording studies. The learning index is derived from interpolated probe trials during place learning in the water maze (Gallagher et al., 1993). Lower values represent more rapid and accurate acquisition of a search for the escape platform. Animals with learning index scores below 240 possess intact learning ability (open symbols), whereas animals with learning index scores above 240 possess impaired learning ability (filled symbols). Adapted from Wilson et al. (2004).
maze to allow adequate sampling of the entire arena during place cell recordings. This approach has given valuable information on the means by which goaldirected behavior modulates hippocampal place fields (Hollup et al., 2001; Fyhn et al., 2002). For our purposes, however, navigation in an annular water maze cannot be considered the same as that in the open water pool. When the freedom of movement is restricted, navigation can be largely based on single landmarks and not require the triangulation using multiple landmarks necessary for solving the open water maze task. Thus, the annular water maze recording technique has proved particularly appropriate for studying place recognition memory (Brun et al., 2002), but not for age-related deficits in navigation based on memories of spatial relationships. For these reasons, our research group has taken another approach to relating place cell recordings to water maze learning. Rather than mimicking the conditions of the water maze testing, we addressed the question of which kinds of deficiencies in spatial information processing could result in defective navigation. The crux of the Morris water maze is that the
355
rat is released from various starting points, making it ineffective to navigate to the hidden platform by using just one guiding landmark or by relying on a simple response strategy from a constant start location (Eichenbaum et al., 1990). By this reasoning, we hypothesized that failure by some aged rats to navigate to the platform derives from impaired processing of the spatial relationships between the extramaze visual landmarks. We therefore recorded from rats in an open-field dry-land environment featuring prominent landmarks that the rat could use to orient itself. By encouraging the rats to move steadily to receive rewarding stimulation of the lateral hypothalamus, we ensured full sampling of the entire arena. In order to test how the hippocampus processes spatial information pertaining to the visual cues, we recorded cells prior to and during manipulations of the visual environment. These manipulations included exposure to a familiar and a visually distinct new arena, to cue rotations of the new arena, and to two visually identical compartments reached by walking.
AGING PLACE CELLS: RIGID REPRESENTATIONS Recordings of hippocampal place cells in our laboratory as well as in several other laboratories have yielded a consistent but surprising finding: even the most memory-impaired aged rats possess hippocampal place cells with distinct spatial firing fields (Markus et al., 1994; Mizumori et al., 1996; Barnes et al., 1997; Shen et al., 1997; Tanila et al., 1997a; Rosenzweig et al., 2003; Wilson et al., 2003, 2004, 2005a,b). One might expect place fields in aged memory-impaired rats to be degraded and a dearth of spatial information within the hippocampus to contribute to spatial difficulties. This is not the case, however. As Figure 22–2 illustrates, place fields of aged rats are equally sharp as those of young rats. What then could account for information-processing failures in the aged hippocampus? The key point is that in a constant environment, hippocampal place cells of aged animals are just as specific and as stable as those of young animals (Barnes et al., 1997). However, a change in environment brings out encoding difficulties for the aged hippocampus. In our first series of experiments, we challenged the hippocampus of young and aged rats to encode novelty in an environment. The rats explored a highly familiar open-field environment, were removed to a holding bucket, and then were placed into a novel open-field environment. After sufficient exploration of the novel environment, the rats were always recorded a final time in the familiar arena. The extent of the novelty,
356 PLACE FIELDS AND AGE-RELATED CHANGES IN MEMORY
Figure 22–2. Place cells of aged memory-impaired rats fail to rapidly create new spatial representations. The top row depicts the experimental setup with rats exploring familiar (fam) and new environments successively. Each subsequent row represents the activity of one cell over the entire experiment; individual grids represent the area where the rat was moving. Firing-rate scales are provided on the left such that darker pixels indicate areas in which more action potentials occurred. Tetrode waveforms are also shown. Two example cells from young rats (A) and from aged rats (B) are shown. Adapted from Wilson et al. (2005a).
that is, the difference between the familiar and novel environments, was graded over experiments. The first experiment consisted of a minor change, from being a familiar cylindrical environment to a square one with the same three prominent wall landmarks in a rotated position with respect to the room. In this situation the place cells of young and aged memory-intact rats de-
veloped distinct patterns of activity for the two environments, whereas spatial representations of aged memory-impaired rats often used similar firing patterns in both environments (Wilson et al., 2003, 2005a) (see Fig. 22–2). Despite changes in the surrounding environment, the hippocampal representations of aged memory-impaired failed to change, indicating that
PLACE CELLS AND AGE-RELATED MEMORY IMPAIRMENTS
hippocampal processing in aged memory-impaired rats is beset by a rigidity of representation. Furthermore, the extent of this rigidity of place cells correlated with the spatial learning ability of the rats (Wilson et al., 2003) (see Fig. 22–3). Clearly, this finding indicates that the memory-impaired aged rats showed less recognition of the change to the environment. In this cylinder–square paradigm there were many overlapping features in the environment; only the geometry differed. This leads to the question of whether hippocampal representations of aged memory-impaired rats can distinguish between familiar and novel environments that are more distinct. When the rats were moved from the familiar cylindrical environment to a novel hexagonal environment with different wall background color and landmarks, place cells of young rats created new representations for the new environment (Wilson et al., 2004). Surprisingly, place cells of some aged rats still maintained the same representations between environments that had distinctly different wall color, landmarks, and shape (Wilson et al., 2004). We then tested if additional exposure to the novel environment on the following day could overcome the rigidity of hippocampal cells in aged animals. Indeed, during a second session, even hippocampal cells of aged rats created distinct representations for the familiar and novel environments. Do the aged rats simply need more experience to learn the details of the new environment? To test whether the representations of the new hexagonal arena of aged animals were indeed processed equally well to those of the young animals, we made a closer examination of the factors that determine the place field location. The novel arena and its landmarks
Figure 22–3. The extent of place cell rigidity predicts the magnitude of spatial learning impairment. Adapted from Wilson et al. (2003).
357
were rotated and the place fields of young and aged memory-intact rats rotated with them, whereas the place fields of aged memory-impaired rats failed to rotate with the landmarks (Wilson et al., 2004) (see Fig. 22–4). This was reflected in a strong correlation between the extent of place field rotation and the spatial learning index (Wilson et al., 2004). In subsequent sessions the rotations of the novel arena were repeated, and the spatial representations of young rats invariably rotated with the arena rotations, but the spatial representations of aged memory-impaired rats rotated on some occasions but not on others (Fig. 22–4). This finding suggests that the aged hippocampus is indeed quite capable of processing visual information (such as landmark rotations), but that the information is processed inconsistently. This inconsistent processing of the visual landmarks may well make the aged rats with rigid place fields poor learners on the water maze when the visual landmarks are the only reliable source of information about the platform location. Thus far we have argued on the basis of data from our own recent experiments that aged memoryimpaired rats have hippocampal representations that fail to encode changes to the environment: they are rigid in response to novelty and fail to rotate with landmarks. These data from our own experiments are also in agreement with earlier experiments. First, Barnes and colleagues (1997) showed that within a constant environment, place cells of aged rats are as stable as those of young rats. When rats were taken out of a recording environment, exposed to other novel environments, and then replaced back into the recording environment, the place cells of young rats successfully reinstated the representations of the recording environment (after they had presumably used other representations in the novel exposures). In the return to the recording environment the place cells of the aged rats, by contrast, often displayed representations that were completely different from their original representation. Second, Tanila and colleagues (1997a,b) demonstrated that the hippocampus of aged animals failed to encode changes to the external environment as deeply as young animals did; the place cells only encoded part of the available information. Third, Oler and Markus (2000) have shown that it is not only environmental changes that are poorly represented by the aged hippocampus. When the task context changes within a constant external environment, place cells of young rats often create new spatial representations, but place cells of aged rats fail acknowledge this kind of change (Oler and Markus, 2000). In summary, we now have strong evidence that the hippocampal cells of aged individuals often fail to process changes in both the external environment and the task context.
358 PLACE FIELDS AND AGE-RELATED CHANGES IN MEMORY
Figure 22–4. Place field responses of a young rat and an aged memory-impaired rat to arena rotations over many exposures to a new environment. Rats explored the novel environment in original and rotated orientations. Sessions shown in rows are abbreviated S1–S6, and arrows indicate the direction of the arena rotation. The locations of action potentials for each simultaneously recorded cell are identified with a unique color. Tetrode waveforms are shown beside the spike maps. See Wilson et al. (2004) for details.
This conclusion leads to the question of whether the aged hippocampus is beset by rigid processing of all kinds of information, or whether some domains are preserved (or even enhanced) in how the aged hippocampus processes them. An interesting angle to this question comes from studies of self-motion processing. Barnes and colleagues (1980) demonstrated that aged rats learn capably in a T-maze to navigate to the goal by means of self-movement, whereas young rats in the same circumstances rely on place information to navigate. Although the hippocampus is not necessarily required for navigating by self-movement (response learning), the place cells of the hippocampus do reflect
self-motion information in balance with external environmental information (Quirk et al., 1990; McNaughton et al., 1996; Redish and Touretzky, 1997; Knierim et al., 1998; Terrazas et al., 2005). Thus, the hippocampus of aged rats may rely more on self-motion information than on external landmark information. To explore how this information processing is balanced and whether rigid aged place cells extend to selfmotion processing, we challenged the place cells of young and aged rats to process a change indicated by changes in self-motion information rather than changes in visual information (Wilson et al., 2005b). In the experiment, the rats walked between two visually
PLACE CELLS AND AGE-RELATED MEMORY IMPAIRMENTS
359
Figure 22–5. Spatial representations in visually identical compartments. Boxes A and B are separated by a wall with a hidden door. Dashed arrows indicate transitions between the trials: walking from A1 to B2, lifted passively by the experimenter from B2 into a hanging bucket and then to B3, and walking from B3 to A4. Each row depicts the spatial firing patterns of one cell across the experiment; note that only the explored arena (box A or B) is shown for each trial. A. Young cell 1 shows different spatial representations in trials A1 and B2. Young cell 2 shows a spatial representation that did not change across the trials. B. Aged cell 1 was silent in environment A, but active in box B. Aged cell 2 had a similar firing pattern on all trials. Adapted from Wilson et al. (2005b).
identical compartments, pitting self-motion information against visual information (Fig. 22–5). In contrast to the earlier experiments in which external visual cues indicated the change in environment, the place cells of aged memory-impaired rats created new spatial rep-
resentations as often as those of the young and aged memory-intact rats (Wilson et al., 2005b). These results suggest that the aged hippocampus processes internal cues more sufficiently than it processes external cues. However, there must still exist a bias
360 PLACE FIELDS AND AGE-RELATED CHANGES IN MEMORY toward rigid representations in the aged hippocampus, because place cells of aged rats did sometimes use similar representations for the two identical compartments despite impaired hippocampal processing of visual information. Without inherent rigidity, the place cells of aged rats would have created new spatial representations more often than those of young rats because of the stronger influence of self-motion processing. This was not the case, however. Thus, rigid representation seems to be a dominating feature of information processing by the aged hippocampus.
1996). Drawing from the neurophysiological evidence discussed above and from neurobiological evidence, we have proposed that aging in the hippocampus may upset the balance between these two processes of pattern completion and pattern separation. The excessive rigidity of aged hippocampal CA3 representations may arise from an aged CA3 that is performing too much pattern completion or from an aged dentate gyrus that is not performing enough pattern separation (Wilson et al., 2006).
DIRECTIONS FOR THE FUTURE CA3 ORIGIN OF RIGIDITY? Recent studies have successfully teased out differences between the CA1 and CA3 subregions of the young hippocampus in how they process novel information (Lee et al., 2004a,b; Leutgeb et al., 2004; Vazdarjanova and Guzowski, 2004). These studies indicate that the CA3 subregion is particularly critical for rapid encoding of new information (Nakazawa et al., 2002; Lee et al., 2004a; Leutgeb et al., 2004). In light of this finding, our post-hoc analysis of how place cells encode a novel environment revealed a striking difference between the CA1 and CA3 subregions of aged animals. The failure to rapidly create new spatial representations for the novel environment was evident significantly more often in the CA3 subregion of the aged hippocampus than in the CA1 subregion (Wilson et al., 2005a). Thus, the rigidity of representations in aged hippocampus may arise predominantly from the CA3 subregion and not the CA1 subregion. This dissociation between CA3 and CA1 responses firmly suggests that the impaired processing, and associated behavioral impairments, is inherent to the hippocampal region itself and not due to a more general deficit, such as a sensory, perceptual, motivational, or motor deficit, or to deficient processing of spatial information by cortical structures projecting to the hippocampus. According to theoretical and recent experimental evidence, the hippocampus may function as an attractor network that maintains patterns of output activity despite small alterations in the inputs (pattern completion) until a threshold for change is reached and the network then switches to distinguishing between the two patterns (pattern separation; Marr, 1971; McNaughton et al., 1996; Redish and Touretzky, 1997; Guzowski et al., 2004; Leutgeb et al., 2005; Wills et al., 2005). The CA3 subregion, with its extensive recurrent collateral connections, has been strongly implicated in pattern completion (Guzowski et al., 2004), and the dentate gyrus subregion, with its large number of neurons, has been implicated in pattern separation (McClelland and Goddard, 1996; McNaughton et al.,
The experiments discussed here have shown that hippocampal place cells of aged memory-impaired rats are impaired in processing changes in inputs during a random exploration task. This observation correlates with the degree of impairment in using external cues during an earlier spatial water maze task. Of course, for further insight into age-related memory impairments, it is of critical importance to record during actual performance of memory tasks. Rosenzweig and colleagues (2003) have taken a major step in this direction by recording hippocampal cells from young and aged animals as they performed a simple spatial navigation task. In this task, rats ran on a linear track, motivated by rewarding brain stimulation to find an unmarked goal whose position was fixed relative to the external room landmarks but not to the self-motion cues. The start box was shifted for each trial, causing the walking distance to the goal location at the opposite end of the track to vary. On each trial, place fields were initially determined by self-motion cues, and successful performance entailed a switch in the control of place fields from self-motion cues to external landmark cues well before the goal area was reached (Gothard et al., 1996). Young rats successfully updated their spatial representations and slowed their running speed in the goal areas to receive the reward, whereas aged rats failed to update their contextual information and thus failed to slow their speed in the correct area (Rosenzweig et al., 2003). This study shows that in this situation of cue conflict, spatial representations are linked to concurrent behavior of young and aged individuals. In order to further understand how hippocampal representations participate in spatial memory, it will ultimately be important to record spatial representations during performance of a long-term memory task. Spatial tests have recently been designed for young animals specifically for this purpose. These tasks contain both random exploration times to provide sampling of the entire arena and reward times to provide directed search (or avoidance) for particular spatial
PLACE CELLS AND AGE-RELATED MEMORY IMPAIRMENTS
zones. The reward time can be likened to a dry-land water maze task, as the rats learn to use the spatial cues to navigate to a specific zone. These experiments done with young animals have tightened the link between place cells and spatial memory, with indications that the relationship between goal and place fields is a critical one (Zinyuk et al., 2000; Kentros et al., 2004; Lenck-Santini et al., 2002). With respect to aging, these tasks represent exciting prospects; they may be sensitive to aging and appropriate for recordings. Place cells provide a unique window into how memories are made. By using them to investigate agerelated memory impairments, we can also understand how memories fail to be made. They illustrate, for example, that at least with aging, memory success requires a flexible hippocampus, whereas memory failures are characterized by an inflexible hippocampus poorly prepared for making new memories. With continued development of new behavioral tasks, especially those suitable for genetically engineered mice to investigate molecular mechanisms of memory, place cells will provide an important and insightful tool. Indeed, we are now learning to take advantage of the robust signal of place cells. The era of place cell applications is only beginning.
acknowledgments This work was supported by the National Institute on Aging, by the Academy of Finland, by the Northern Savonia Cultural Foundation, and by the Research and Science Foundation of Farmos. We thank Howard Eichenbaum and Michela Gallagher for their collaboration and organization of the Aging Program project.
References Barnes CA, Nadel L, Honig WK (1980) Spatial memory deficit in senescent rats. Can J Psychol 34:29–39. Barnes CA, Suster MS, Shen J, McNaughton BL (1997) Multistability of cognitive maps in the hippocampus of old rats. Nature 388:272–275. Brun VH, Otnass MK, Molden S, Steffenach HA, Witter MP, Moser MB, Moser EI (2002) Place cells and place recognition maintained by direct entorhinalhippocampal circuitry. Science 296:2243–2246. Cohen NJ, Eichenbaum H (1993) Memory, Amnesia, and the Hippocampal System. Cambridge, MA: MIT Press. Eichenbaum H, Stewart C, Morris RG (1990) Hippocampal representation in place learning. J Neurosci 10:3531–3542. Fyhn M, Molden S, Hollup S, Moser MB, Moser E (2002) Hippocampal neurons responding to first-time dislocation of a target object. Neuron 35:555–566.
361
Gallagher M, Burwell RD (1989) Relationship of agerelated decline across several behavioral domains. Neurobiol Aging 10:691–708. Gallagher M, Burwell R, Burchinal M (1993) Severity of spatial learning impairment in aging: development of a learning index for performance in the Morris water maze. Behav Neurosci 107:618–626. Gallagher M, Colantuoni C, Eichenbaum H, Haberman R, Rapp PR, Tanila H, Wilson IA (2006) Individual differences in neurocognitive aging of the medial temporal lobe. Age 28:221–233. Gallagher M, Rapp PR (1997) The use of animal models to study the effects of aging on cognition. Annu Rev Psychol 48:339–370. Gothard KM, Skaggs WE, McNaughton BL (1996) Dynamics of mismatch correction in the hippocampal ensemble code for space: interaction between path integration and environmental cues. J Neurosci 16:8027– 8040. Guzowski JF, Knierim JJ, Moser EI (2004) Ensemble dynamics of hippocampal regions CA3 and CA1. Neuron 44:581–584. Hollup SA, Molden S, Donnett JG, Moser MB, Moser EI (2001) Accumulation of hippocampal place fields at the goal location in an annular watermaze task. J Neurosci 21:1635–1644. Kentros CG, Agnithori NT, Streater S, Hawkins RD, Kandel ER (2004) Increased attention to spatial context increases both place field stability and spatial memory. Neuron 42:283–295. Knierim JJ, Kudrimoti HS, McNaughton BL (1998) Interactions between idiothetic cues and external landmarks in the control of place cells and head direction cells. J Neurophysiol 80:425–446. Lee I, Rao G, Knierim JJ (2004a) A double dissociation between hippocampal subfields: differential time course of CA3 and CA1 place cells for processing changed environments. Neuron 42:803–815. Lee I, Yoganarasimha D, Rao G, Knierim JJ (2004b) Comparison of population coherence of place cells in hippocampal subfields CA1 and CA3. Nature 430:456–459. Lenck-Santini PP, Muller RU, Save E, Poucet B (2002) Relationships between place cell firing fields and navigational decisions by rats. J Neurosci 22:9035–9047. Lenck-Santini PP, Save E, Poucet B (2001) Evidence for a relationship between place-cell spatial firing and spatial memory performance. Hippocampus 11:377–390. Leutgeb JK, Leutgeb S, Treves A, Meyer R, Barnes CA, McNaughton BL, Moser MB, Moser EI (2005) Progressive transformation of hippocampal neuronal representations in ‘‘morphed’’ environments. Neuron 48: 345–358. Leutgeb S, Leutgeb JK, Treves A, Moser MB, Moser EI (2004) Distinct ensemble codes in hippocampal areas CA3 and CA1. Science 305:1295–1298.
362 PLACE FIELDS AND AGE-RELATED CHANGES IN MEMORY Markus EJ, Barnes CA, McNaughton BL, Gladden VL, Skaggs WE (1994) Spatial information content and reliability of hippocampal CA1 neurons: effects of visual input. Hippocampus 4:410–421. Marr D (1971) Simple memory: a theory for archicortex. Philos Trans R Soc Lond B Biol Sci 262:23–81. McClelland JL, Goddard NH (1996) Considerations arising from a complementary learning systems perspective on hippocampus and neocortex. Hippocampus 6:654–665. McNaughton BL, Barnes CA, Gerrard JL, Gothard K, Jung MW, Knierim JJ, Kudrimoti H, Qin Y, Skaggs WE, Suster M, Weaver KL (1996) Deciphering the hippocampal polyglot: the hippocampus as a path integration system. J Exp Biol 199:173–185. Milner B, Squire LR, Kandel ER (1998) Cognitive neuroscience and the study of memory. Neuron 20:445– 468. Mizumori SJ, Lavoie AM, Kalyani A (1996) Redistribution of spatial representation in the hippocampus of aged rats performing a spatial memory task. Behav Neurosci 110:1006–1016. Moffat SD, Resnick SM (2002) Effects of age on virtual environment place navigation and allocentric cognitive mapping. Behav Neurosci 116:851–859. Monacelli AM, Cushman LA, Kavcic V, Duffy CJ (2003) Spatial disorientation in Alzheimer’s disease: the remembrance of things passed. Neurology 61:1491–1497. Morris RG, Garrud P, Rawlins JN, O’Keefe J (1982) Place navigation impaired in rats with hippocampal lesions. Nature 297:681–683. Moser MB, Moser EI (1998) Functional differentiation in the hippocampus. Hippocampus 8:608–619. Nakazawa K, Quirk MC, Chitwood RA, Watanabe M, Yeckel MF, Sun LD, Kato A, Carr CA, Johnston D, Wilson MA, Tonegawa S (2002) Requirement for hippocampal CA3 NMDA receptors in associative memory recall. Science 297:211–218. Newman MC, Kazniak AW (2000) Spatial memory and aging: performance in a human analog of the Morris water maze. Aging Neuropsychol Cogn 7:86–93. O’Keefe J, Dostrovsky J (1971) The hippocampus as a spatial map. Preliminary evidence from unit activity in the freely moving rat. Brain Res 34:171–175. O’Keefe J, Nadel L (1978) The Hippocampus as a Cognitive Map. Oxford: Clarendon Press. O’Keefe J, Speakman A (1987) Single unit activity in the rat hippocampus during a spatial memory task. Exp Brain Res 68:1–27. Oler JA, Markus EJ (2000) Age-related deficits in the ability to encode contextual change: a place cell analysis. Hippocampus 10:338–350. Quirk GJ, Muller RU, Kubie JL (1990) The firing of hippocampal place cells in the dark depends on the rat’s recent experience. J Neurosci 10:2008–2017.
Redish AD, Touretzky DS (1997) Cognitive maps beyond the hippocampus. Hippocampus 7:15–35. Rodrigue KM, Kennedy KM, Raz N (2005) Aging and longitudinal change in perceptual-motor skill acquisition in healthy adults. J Gerontol B Psychol Sci Soc Sci 60:P174–P181. Rosenzweig ES, Barnes CA (2003) Impact of aging on hippocampal function: plasticity, network dynamics, and cognition. Prog Neurobiol 69:143–179. Rosenzweig ES, Redish AD, McNaughton BL, Barnes CA (2003) Hippocampal map realignment and spatial learning. Nat Neurosci 6:609–615. Salthouse TA (1996) The processing-speed theory of adult age differences in cognition. Psychol Rev 103: 403–428. Shen J, Barnes CA, McNaughton BL, Skaggs WE, Weaver KL (1997) The effect of aging on experiencedependent plasticity of hippocampal place cells. J Neurosci 17:6769–6782. Smith CD, Walton A, Loveland AD, Umberger GH, Kryscio RJ, Gash DM (2005) Memories that last in old age: motor skill learning and memory preservation. Neurobiol Aging 26:883–890. Tanila H, Shapiro M, Gallagher M, Eichenbaum H (1997a) Brain aging: changes in the nature of information coding by the hippocampus. J Neurosci 17: 5155–5166. Tanila H, Sipila P, Shapiro M, Eichenbaum H (1997b) Brain aging: impaired coding of novel environmental cues. J Neurosci 17:5167–5174. Terrazas A, Krause M, Lipa P, Gothard KM, Barnes CA, McNaughton BL (2005) Self-motion and the hippocampal spatial metric. J Neurosci 25:8085–8096. Vazdarjanova A, Guzowski JF (2004) Differences in hippocampal neuronal population responses to modifications of an environmental context: evidence for distinct, yet complementary, functions of CA3 and CA1 ensembles. J Neurosci 24:6489–6496. Wilkniss SM, Jones MG, Korol DL, Gold PE, Manning CA (1997) Age-related differences in an ecologically based study of route learning. Psychol Aging 12:372– 375. Wills TJ, Lever C, Cacucci F, Burgess N, O’Keefe J (2005) Attractor dynamics in the hippocampal representation of the local environment. Science 308:873–876. Wilson IA, Gallagher M, Eichenbaum H, Tanila H (2006) Neurocognitive aging: prior memories hinder new hippocampal encoding. Trends Neurosci 29:662– 670. Wilson IA, Ikonen S, Gallagher M, Eichenbaum H, Tanila H (2005a) Age-associated alterations of hippocampal place cells are subregion specific. J Neurosci 25:6877–6886. Wilson IA, Ikonen S, Gureviciene I, McMahan RW, Gallagher M, Eichenbaum H, Tanila H (2004) Cog-
PLACE CELLS AND AGE-RELATED MEMORY IMPAIRMENTS
nitive aging and the hippocampus: how old rats represent new environments. J Neurosci 24:3870–3878. Wilson IA, Ikonen S, Gurevicius K, McMahan RW, Gallagher M, Eichenbaum H, Tanila H (2005b) Place cells of aged rats in two visually identical compartments. Neurobiol Aging 26:1099–1106. Wilson IA, Ikonen S, McMahan RW, Gallagher M, Eichenbaum H, Tanila H (2003) Place cell rigidity
363
correlates with impaired spatial learning in aged rats. Neurobiol Aging 24:297–305. Zinyuk L, Kubik S, Kaminsky Y, Fenton AA, Bures J (2000) Understanding hippocampal activity by using purposeful behavior: place navigation induces place cell discharge in both task-relevant and task-irrelevant spatial reference frames. Proc Natl Acad Sci USA 97: 3771–3776.
23 Aging Ensembles: Circuit Contributions to Memory Deficits SARA N. BURKE AND CAROL A. BARNES
The normal aging process is accompanied by a decline in spatial and other forms of episodic memory. It is well documented that lesions of the hippocampus lead to profound deficits in spatial memory in mice (Farr et al., 2000), rats (Morris et al., 1982), dogs (Kowalska, 1995), and monkeys (Lavenex et al., 2006). Additionally, in humans both episodic and spatial memory is severely disrupted in patients with hippocampal damage (Scoville and Milner, 1957; Squire et al., 1993; Teng and Squire, 1999). Together these data imply that functional changes in the hippocampus contribute to age-related alterations in memory (for a review, see Rosenzweig and Barnes, 2003; Wilson et al., 2006). Thus, uncovering the details of ageassociated changes in temporal lobe structures, which are responsible for memory impairment, is of considerable importance. For ‘‘normal’’ age-related memory loss, it is known that the neurobiological underpinnings involve subtle, region-specific changes in synapses, and the dynamics of neural ensembles, rather than gross changes in cell number or neuron morphology. Specifically, it is now known that in humans (West et al., 1994; Pakkenberg and Gundersen, 1997), nonhuman primates (Peters et al., 1994; Gazzaley et al., 1997; Merrill et al., 2000; Keuker et al., 2003), and rodents (Rapp and Gallagher, 1996; Rasmussen et al., 1996; Merrill et al., 2001), there is no significant cell death in most principal cell groups in the hippocampus or surrounding neocortex during normal aging. Moreover, dendritic branching and length do not decrease during normal aging in CA1 (Hanks and Flood, 1991), CA3 (Flood et al.,
1987), or the subiculum (Flood, 1991) of humans. Similar stability in cell morphology has been reported for the granule cells in the rat dentate gyrus (Flood, 1993) and pyramidal cells in the CA1 subregion (Turner and Deupree, 1991; Pyapali and Turner, 1996), with some data showing an increase in basilar dendritic length and branching in 24-month-old rats compared with 2-month-old rats (Pyapali and Turner, 1996; but see Markham et al., 2005). These observations imply that it will be productive to examine the interactions among and temporal dynamics of the neural ensembles in brain structures such as the hippocampus that ultimately support cognition. Age-associated alterations in the temporal dynamics of neural circuits that support memory can be investigated by measuring differences in experimentally induced plasticity in young and old animals (for review, see Rosenzweig and Barnes, 2003; Burke and Barnes, 2006). Such experiments have been essential for understanding the functional changes in aged neural networks; however, they are traditionally conducted in vitro or in anesthetized animals. Moreover, it is unclear how the parameters used to induce plasticity experimentally relate to the plasticity mechanisms engaged by neural activity elicited by behavior. Another approach to investigating the impact of advanced age on neural circuits is to monitor the activity of neurons in awake, behaving animals. Methods have been developed that allow for the activity of a hundred or more individual neurons to be recorded in awake, behaving young and old rats (Wilson and McNaughton, 1993). Such experiments have begun to uncover
364
CIRCUIT CONTRIBUTIONS TO MEMORY DEFICITS
the impact of age on the neurobiology of the medial temporal lobe. This chapter will review what is currently known about aged neural ensembles in the hippocampus and how alterations in the dynamics of these circuits are linked to memory decline.
FUNDAMENTAL PROPERTIES OF PLACE CELLS IN YOUNG AND OLD RATS Neuronal recordings from the hippocampus of adult rats reveal that when a rat explores an environment, pyramidal (O’Keefe and Dostrovsky, 1971) and granule cells (Jung and McNaughton, 1993; Leutgeb et al., In press) show patterned neural activity that is highly correlated with a rat’s position in space (the ‘‘place field’’ of the cell). Approximately 30%–50% of CA1 pyramidal cells show place-specific firing in a given environment of intermediate size (e.g., Muller et al., 1987; Wilson and McNaughton, 1993; Gothard et al., 1996b), earning these neurons the name place cells. When the firing patterns of many hippocampal neurons are recorded simultaneously, it is possible to reconstruct the position of a rat within an environment from the place cell firing patterns alone (Wilson and McNaughton, 1993). The composite cell activity is ‘‘map-like’’ and, in different environments, hippocampal place maps change markedly. Although these maps can be driven by external environmental features, internal events are also important and a new map may be generated in the same environment if the demands of the task change (McNaughton et al., 1983; Muller et al., 1994; Markus et al., 1995). Given the extensive age-associated impairments in behaviors that are hippocampal dependent (for review, see Rosenzweig and Barnes, 2003), it is surprising that the basic characteristics of CA1 place fields such as spike amplitude, spike width, mean and maximum firing rates, and inter-spike interval (ISI) distributions are remarkably preserved in advanced age (e.g., Barnes et al., 1983; Markus et al., 1994; Mizumori et al., 1996; Barnes et al., 1997; Shen et al., 1997; Oler and Markus, 2000). Moreover, upon the initial pass through a place field, or during a task in which the rat’s trajectory is not restricted (i.e., random foraging), the place fields of aged rats (Fig. 23–1) are just as specific as those of young rats (Markus et al., 1994; Mizumori et al., 1996; Barnes et al., 1997; Shen et al., 1997; Tanila et al., 1997a; Oler and Markus, 2000). In contrast, the basic properties of CA3 place fields show at least two age-related changes. First, the firing rates of CA3 pyramidal cells are higher in aged rats than in young animals (Wilson et al., 2005). Additionally, during a random foraging task, the CA3 place fields are larger in old rats than in young animals. The
365
behavioral implications of these age-related changes in CA3 pyramidal cell firing are just beginning to be elucidated (Wilson et al., 2005; for review, see Wilson et al., 2006). While there are no reports of electrophysiological recordings from granule cells in awake behaving old rats, it is possible to label the neurons that were activated during an epoch of behavior by monitoring the expression of immediate early genes (IEGs), a class of genes that do not require the preceding activation of any other responsive genes or de novo protein synthesis to be expressed. Consequently, the transcription of IEGs is thought to be dynamically regulated by specific forms of patterned synaptic activity believed to underlie information storage (Cole et al., 1989), and it has been shown that IEGs are expressed by cells activated during spatial exploration (Guzowski et al., 1999). One of the IEGs of interest is the gene for activity-regulated cytoskeletal (Arc) protein. Arc codes for an effector protein that is involved in AMPA receptor trafficking (Chowdhury et al., 2006; Rial Verde et al., 2006; Shepherd et al., 2006) and has been shown to be necessary for the maintenance of longterm memory, long-term potentiation (Guzowski et al., 2000), and the induction and maintenance of longterm depression (Plath et al., 2006). Changes in the proportion of cells that express a gene during behavior can be assessed using fluorescence in situ hybridization. This allows exact determination of which individual cells are expressing which genes. Using this method, it was determined that granule cells of the dentate gyrus, but not the pyramidal cells of CA1 and CA3, in aged rats have a significantly smaller proportion of neurons that express Arc following spatial exploration (Small et al., 2004). These data suggest that fewer granule cells would express place fields in a given environment. Alternatively, it is also possible that the same proportion of dentate gyrus granule cells express place fields but that during advanced age the transcription of Arc becomes decoupled from the granule cell activity, as is known to occur under some conditions in CA1 (Fletcher et al., 2006; Guzowski et al., 2006). Neural recordings from the dentate gyrus of awake and behaving young and old rats are required, however, to discriminate between these two possibilities (Fig. 23–2). Interestingly, using other imaging methods, in humans and monkeys, the granule cells also seem to be particularly vulnerable to the effects of normal aging (Small et al., 2002; Small et al., 2004). In all subregions of the hippocampus the timing of spikes shows a dynamic relationship to the hippocampal theta rhythm, a 5–10 Hz oscillation that is prominent in the rat hippocampus during active exploration (Vanderwolf, 1969) and REM sleep (Vanderwolf et al.,
366 PLACE FIELDS AND AGE-RELATED CHANGES IN MEMORY
Figure 23–1. CA1 place fields in old rats. Each panel shows a firing rate map for 1 of 16 representative CA1 pyramidal cells that were active while an aged rat (27 months old) randomly foraged for chocolate sprinkles in a familiar environment. Each box reflects the environment explored by the rat in this experiment. Red indicates portions of the environment where the firing rate was high, while blue indicates areas where the neuron did not fire. The CA1 place field characteristics of old rats during a random foraging task are not qualitatively different from those of a young rat.
1977). As a rat passes through a hippocampal principal cell’s place field, the timing of the spikes shows a clear shift relative to the local theta rhythm such that the spike timing occurs at earlier phases of the theta cycle (O’Keefe and Recce, 1993). Regardless of the direction from which the animal enters the field, the first spikes appear late in the theta cycle. As the rat continues to advance through the place field, however, the spikes shift progressively in relative theta phase so as to appear early in the last theta cycle as the rat exits the field (Fig. 23–3A). For the majority of place fields recorded from CA1 pyramidal cells, this ‘‘theta phase precession’’ (O’Keefe and Recce, 1993) covers almost 3608 of phase shift, but not more (O’Keefe and Recce, 1993; Skaggs et al., 1996; Shen et al., 1997; Ekstrom et al., 2001; Maurer et al., 2006). This observation indicates that the rate of theta phase precession must be dynamically coupled to the size of a neuron’s place field, such that smaller place fields show a faster rate of precession than that of larger place fields (Shen
et al., 1997; Ekstrom et al., 2001; Terrazas et al., 2005; Maurer et al., 2006). Therefore, a single 3608 cycle of phase precession can be used as an objective measure of place field size, as this definition does not require an arbitrary experimenter-defined threshold as with the traditional firing rate measures of place field size (Maurer et al., 2006). Moreover, it is possible for a single CA1 neuron to have multiple place fields that overlap in space. These overlapping place fields can be distinguished from one another on the basis of their discreet and independent cycles of theta phase precession. This feature of the phase precession definition of a place field is particularly advantageous compared with other methods that use only firing rates to determine place field boundaries, because the latter methods cannot easily distinguish between overlapping place fields (Maurer et al., 2006). In the CA1 subregion of the hippocampus, pyramidal cells in both old and young rats exhibit theta phase precession (Shen et al., 1997). Moreover, there
CIRCUIT CONTRIBUTIONS TO MEMORY DEFICITS
Figure 23–2. Arc expression in the dentate gyrus. The proportion of cells expressing the immediateearly gene Arc following behavioral induction can be measured in specific brain structures with fluorescence in situ hybridization (from Small et al., 2004). Shown here are confocal images of fluorescence in situ hybridization for Arc mRNA in the dentate gyrus of a young rat (top panel) and an old rat (bottom panel). Granule cells are shown in red and Arc mRNA in yellow. More granule cells are positively labeled for Arc in young rats than in the old rats after spatial exploration.
is no difference between the total phase change of place fields in old rats compared to young rats such that the place field spikes begin and end at approximately the same phase of theta in both young and old rats (Shen et al., 1997). Additionally, theta phase precession continues to occur in the presence of an NMDA
367
receptor antagonist, and the magnitude of the total phase advance is not altered by the absence of NMDA receptor activity (Ekstrom et al., 2001). Interestingly, the place fields of CA1 pyramidal cells are smaller in old rats than those in young rats when the animal repeatedly takes the same path. This difference in place field size is presumably due to a lack of NMDA receptor–dependent place field expansion plasticity in old rats, which will be discussed more extensively in a later section. Because there is no difference in the total magnitude of phase advance between young and old rats, this indicates that the rate of phase precession must be greater in the old rats in order for the same magnitude of phase advance to be achieved in a smaller portion of space. Empirically, this has been shown, and the slope of a phase-versus-location plot is significantly steeper in old rats (Fig. 23–3B), indicating a more rapid phase precession with distance (Shen et al., 1997). A similar observation has been made in young rats given an NMDA receptor antagonist (Ekstrom et al., 2001). Therefore, phase precession is a basic property of CA1 place fields that does not require NMDA receptor–dependent plasticity. In addition to theta phase precession, other basic properties of place fields are maintained in advanced age. It is well documented that as the running speed of an animal increases the firing rate of dorsal CA1 neurons also increases proportionally (McNaughton et al., 1983). Even though old rats run at significantly slower velocities than young animals, the relationship between firing rate and the velocity is similar between young and old rats (Shen et al., 1997; Burke et al., 2005). While changes have been observed in properties of place cells in both the dentate gyrus and CA3, it is still unclear how an increase in size and firing rate of CA3 place fields and a decrease in the proportion of active granule cells during spatial exploration result in behavioral impairments. Conversely, reports of agerelated changes in the ensemble dynamics of CA1 place fields have been correlated with observed behavioral impairments (Barnes et al., 1997; Rosenzweig et al., 2003) in aged rats, but these changes can only be measured and understood by monitoring the activity of many hippocampal neurons over time.
ADVANCED AGE AND THE DYNAMIC PROPERTIES OF HIPPOCAMPAL PLACE CELLS It is widely agreed that modifiable neuronal ensembles support cognition. Therefore, alterations in how these networks adapt in response to stimuli or behavioral experience could be responsible for the cognitive
368 PLACE FIELDS AND AGE-RELATED CHANGES IN MEMORY
Figure 23–3. Theta phase precession in young and old rats. A. Schematic diagram of theta phase precession. As a rat passes through a principal cell’s place field, the timing of the spikes shows a monotonic shift in their phase of firing such that, as the rat first enters the field, the spikes occur late in phase, but as the rat exits the field, the spikes occur relatively early in phase. For most CA1 pyramidal cells, place fields exhibit 3608 of phase advance and not more. B. Phase by position plots for a young (open circles) and an old (filled circles) rat. The slope of this phase by position plot is significantly greater for the old rat than for the young rat, indicating that while both young and old rats exhibit theta phase precession, the rate of precession is greater in old rats than in young rats because the size of the field is larger in the young rats after the first traversals of the route. When the size of the place field is smaller, as in old rats, the spikes of the pyramidal cells need to precess, on average, more degrees per centimeter traveled by the rat in order to achieve 3608 of precession during a single pass through the place field. The result is a greater rate of theta phase precession.
impairments observed with aging. The physiological mechanisms by which synaptic strength can be altered in response to environmental stimuli and/or experience is similar to the experimentally induced forms of plasticity known as long-term potentiation (LTP) and long-term depression (LTD). LTP and LTD represent mechanisms by which the synaptic weight matrix can adapt, which ultimately is thought to lead to modifications that reflect learning in the system. LTP can be divided into an induction phase (early-phase LTP) and a maintenance phase (late-phase LTP). The induction phase involves the temporal association of presynaptic glutamate release with postsynaptic depolarization (necessary to eject Mg2þ ions from the pores of NMDA receptors), which results in an increase in intracellular Ca2þ (for review, see Bliss and Collingridge, 1993). LTP maintenance is the continued expression of increased synaptic efficacy that persists after induction. This involves changes in gene expression and insertion of AMPA receptors into the postsynaptic membrane (for review, see Malinow and Malenka, 2002; Malenka, 2003). Aged rats have deficits in both LTP induction and maintenance. These deficits, however, are complex, protocol dependent, and region specific (for review, see Burke and Barnes, 2006). Within the hippocampus there are age-related changes in experimentally induced plasticity in all subregions (Fig. 23–4). Briefly, at the perforant path– dentate gyrus granule cell synapse there is a decrease in the maintenance of LTP (Barnes, 1979) and an increase in the threshold to induce LTP (Barnes et al., 2000). Additionally, at the perforant path–CA3 pyramidal cell synapse there is a decrease in the maintenance of LTP in aged rats compared with that in young rats (Dieguez and Barea-Rodriguez, 2004). At the CA3-to-CA1 pyramidal cell–Schaffer collateral synapse, when peri-threshold stimulation parameters are used, the level of LTP induction in aged rats is less than in young rats (Deupree et al., 1991; Moore et al., 1993; Barnes et al., 1996; Rosenzweig et al., 1997; Tombaugh et al., 2002). Finally, when low-frequency stimulation parameters are used at the CA3-to-CA1 Schaffer collateral synapse, aged rats show an increase in the induction of LTD and LTP reversal (Norris et al., 1996; but see, Lee et al., 2005), which can be reversed by blocking the release of Ca2þ from internal stores (Kumar and Foster, 2005). For more extensive reviews see Rosenzweig and Barnes (2003) and Burke and Barnes (2006). Findings of age-associated impairments in experimentally induced LTP and LTD suggest that while the basic properties of CA1 place fields are similar between young and aged rats, the dynamic properties of place fields that require NMDA receptor–dependent plasticity could be altered during the aging process.
CIRCUIT CONTRIBUTIONS TO MEMORY DEFICITS
369
Figure 23–4. Summary of age-related alterations in long-term potentiation (LTP) and long-term depression (LTD) between young and aged animals. The y-axes show the change in EPSP slope following LTP or LTD induction and the x-axes show the retention intervals for maintenance of LTP or LTD. Dashed lines, young rats; solid black lines, aged rats. A. When supra-threshold stimulation parameters are used, LTP induction is intact at old hippocampal synapses (from left, Barnes, 1979; Landfield et al., 1979; Dieguez and Barea-Rodriguez, 2004), but decay over days in the dentate gyrus (DG) (Barnes, 1979) and CA3 (Dieguez and Barea-Rodriguez, 2004) is faster in aged rats. PP, perforant path. B. When peri-threshold stimulation parameters are used, aged rats can show LTP induction deficits (from left, Moore et al., 1993; Barnes et al., 1996; Deupree et al., 1991). C. In CA1, aged rats are more susceptible to LTD induction (Norris et al., 1996, but see Lee et al., 2005). LTD induction with low-frequency stimulation (LFS) in old rats, however, can be attenuated by agents, such as cyclopiazonic acid, that prevent the release of Ca2þ from internal Ca2 stores (Kumar and Foster, 2005). D. Aged rats are also more susceptible than young rats to the reversal of LTP. The increase in EPSP slope that results from LTP-inducing stimuli can be attenuated by the application of LFS to the potentiated pathway. In young rats, LTP is not completely reversed by LFS and there is still some residual potentiation. In old rats, however, LFS returns the EPSP slope to the baseline pre-LTP levels (Norris et al., 1996).
Moreover, these alterations could be related to hippocampal-dependent age-related memory decline. The initial expression of CA1 place fields within an environment does not require NMDA receptor– dependent plasticity (Kentros et al., 1998). Similarly, in young and old rats the same proportion of CA1 neurons are active (Small et al., 2004) and express place fields in a single environment (Shen et al., 1997).
There are, however, dynamic properties of place fields that are disrupted both when the NMDA receptor is blocked and in advanced age. Among these dynamic place field characteristics are experience-dependent place field expansion and place map stability. In young rats, CA1 place fields expand asymmetrically during repeated route following (for example, traversing a circular track), resulting in a shift in the
370 PLACE FIELDS AND AGE-RELATED CHANGES IN MEMORY center of mass of place fields in the direction opposite the rat’s trajectory (Fig. 23–5A) (Mehta et al., 1997). This observation is consistent with neural network models dating back to Hebb’s (1949) concept of the ‘‘phase sequence’’ of cell assemblies, which have suggested that an associative, temporally asymmetric synaptic plasticity mechanism could serve to encode sequences or episodes of experience (Hebb, 1949). The magnitude of this place field expansion, however, significantly decreases in aged rats (Fig. 23–5B) (Shen et al., 1997). It is likely that this age-associated reduction in experience-dependent plasticity is due to LTP-like deficits, as place field expansion does not occur when young rats are administered the NMDA receptor antagonist CPP (Ekstrom et al., 2001). Specifically, plasticity deficits at the Schaffer collateral CA3–CA1 synapse in old animals (Moore et al., 1993; Norris et al., 1996; Rosenzweig et al., 1997; Tombaugh et al., 2002; Dieguez and Barea-Rodriguez, 2004; for review, see Burke and Barnes, 2006) could be responsible for the lack of place field expansion in aged rats. In young animals, rapid behaviorally induced potentiation of feed-forward synapses from CA3 to CA1 can explain the experience-dependent expansion of CA1 place fields (Mehta et al., 2000). During the first pass through a place field, a dorsal CA1 neuron may inherit its place-specific firing from neurons in layer III of the medial entorhinal cortex (Brun et al., 2002), and input to the CA1 neuron from CA3 would be relatively symmetric; that is, the CA1 neuron receives equal input from CA3 neurons with place fields just before and just after the location of the CA1 neuron place field. After repeated traversals, in which both directionally selective CA3 and CA1 neurons are activated, a spike timing–dependent plasticity mechanism would strengthen synapses from those CA3 neurons that had a place field just before that of a CA1 neuron (Mehta et al., 2000; Lee et al., 2004a). The result is a backward, asymmetric shift of CA1 place fields (Fig. 23–5A). If behaviorally induced plasticity of the CA3-to-CA1 Schaffer collateral synapse is the primary mechanism for place field expansion, then the decrease in LTP induction and the increase in LTD induction at this synapse could explain why aged rats fail to show this phenomenon under normal conditions. Interestingly, it has recently been shown that memantine (approved for treatment of the cognitive disorders associated with Alzheimer’s disease) can, at least partially, reinstate experience-dependent place field expansion plasticity in aged rats (Burke et al., 2005). Specifically, acute administration of memantine 20 min prior to an episode of track running results in an increase in the size of place fields, a shift in the center of mass of place fields, and an increase in spike number over the first several laps of tracking running
Figure 23–5. Experience-dependent place field expansion in young and old rats. A. Schematic diagram of experience-dependent place field expansion plasticity that occurs in young rats as they repeatedly traverse the same route traveling through a CA1 pryamidal cell’s place field multiple times. During the first pass through the place field, the field is symmetrical (shaded distribution). As the rat passes through the field multiple times, however, the field expands in the direction opposite the rat’s trajectory, leading to a shift in the center of mass (COM) of the place field’s spikes. B. Plot of place field size by lap. Young rats show significant place field expansion plasticity (open circles) while old rats fail to show this form of experience-dependent plasticity robustly (filled circles). C. Old rats given memantine (10 mg/kg; open circles) show improved place field expansion plasticity compared to saline control (filled circles; data from Burke et al., 2007).
CIRCUIT CONTRIBUTIONS TO MEMORY DEFICITS
(Fig. 23–5C). The efficacy of memantine at reinstating experience-dependent place field expansion plasticity may be attributed to memantine’s pharmacokinetics and binding affinity for the NMDA receptor (Parsons et al., 1995; Rogawski and Wenk, 2003). The observed plasticity deficits at the CA3–CA1 Schaffer collateral synapse in aged rats can be attributed in part to age-associated alterations in the regulation of intracellular Ca2þ levels. Aged CA1 pyramidal cells have increased Ca2þ conductance due to a higher density of L-type Ca2þ channels (Thibault and Landfield, 1996). This may lead to disruptions in Ca2þ homeostasis (for review, see Toescu et al., 2004) that ultimately contribute to age-related plasticity deficits (Landfield, 1988; Foster and Norris, 1997; Kumar and Foster, 2005). Moreover, it has been hypothesized that postsynaptic intracellular levels of Ca2þ are involved in setting the synaptic modification threshold. This threshold may then affect the probability that a synapse will be depressed or potentiated at a given time (Bear et al., 1987; Bear and Malenka, 1994; Foster and Norris, 1997). Ca2þ dyshomeostatis in aged animals (Landfield, 1988; Thibault and Landfield, 1996; Foster and Norris, 1997) could therefore alter the probability that synaptic activity will induce either LTP or LTD. In line with the Ca2þ hypothesis of age-related plasticity impairments is the finding that the inhibition of Ca2þ-induced Ca2þ release from intracellular Ca2þ stores attenuates LTD induction in aged CA1 neurons (Kumar and Foster, 2005). If the disruption in experience-dependent CA1 place field expansion in aged rats is related to Ca2þ dyshomeostasis, then it is possible that reducing the amount of Ca2þ entering a neuron through the activated NMDA receptor may reinstate this behaviorally induced plasticity phenomenon. Memantine has low to moderate affinity for the NMDA receptor channel, strong voltage-dependent channel-blocking characteristics (similar to Mg2þ), and fast channel-unblocking kinetics (Parsons et al., 1995). Therefore, it is possible that memantine restores place field expansion through the mechanism just described. In addition to age-related alterations in experiencedependent place field expansion, the maintenance of hippocampal maps (that is, the distribution of place fields in an environment) also differs between young and old animals. In normal young rats, a map for a given environment can remain stable for months (Thompson and Best, 1990), that is, when a rat is repeatedly returned to the same environment, the same map is retrieved. A similar stability of CA1 maps in aged rats is observed within and between episodes of behavior in the same environment. Occasionally, however, if the old rat is removed from the environment and returned later, the original map is not re-
371
trieved and an independent population of place cells may be activated even in a familiar room (Fig. 23–6) (Barnes et al., 1997; Gerrard et al., 2001). This ‘‘remapping’’ predicts that old rats should show bimodal performance on tasks that require the functional integrity of the hippocampus. For spatial tasks, good performance should correspond to retrieval of the original map, and poor performance should correspond to retrieval of an incorrect map. This prediction is at least consistent with empirical observation. When trained on the spatial version of the Morris swim task, in early trials for both young and aged rats, performance is bimodal. This means that for some trials rats find the hidden escape platform with a short path but for other trials the rats do not recall the location of the platform and take a longer path. By the final training trials, however, the young rats’ performance is unimodal, with most rats taking a direct path to the platform. By contrast, the aged rats’ performance remains bimodal. The trials on which the old rats fail to correctly remember the location of the hidden escape platform could correspond to map retrieval failures (Barnes et al., 1997). A probable mechanism for map retrieval failures is defective LTP in aged rats. Although map stability within an episode does not require plasticity, the maintenance of maps between episodes depends on an LTP-like mechanism. It has been shown in young rats that blockade of NMDA receptors (Kentros et al., 1998) or protein synthesis inhibition (Agnihotri et al., 2004) results in map retrieval errors when the rat is returned to the same environment. Thus, effective long-term plasticity mechanisms are required for the map to become stabilized, possibly through binding the external features of the environment to the map. If this binding does not occur or is weakened by defective plasticity mechanisms, the likelihood that the correct map will be recalled in a familiar environment decreases. In support of the idea that external features of the environment become bound to the ‘‘map’’ is the phenomenon of hippocampal map realignment. This can be demonstrated by training a young rat to shuttle back and forth on a linear track between a start box mounted on a sliding rail and a reward site at the opposite end of the track. On the initial part of all journeys out from the start box, CA1 place cells fire at fixed distances from the start box. As the rat approaches the reward site, however, CA1 place cells fire at fixed distances from the destination. Thus, on outward journeys from the start point, the position representation is updated by path integration. Farther along the journey, however, the place map becomes aligned on the basis of external stimuli. If the path integration information (i.e., distance traveled from the start box) is put into conflict with the external
372 PLACE FIELDS AND AGE-RELATED CHANGES IN MEMORY
Figure 23–6. CA1 map stability in young and old rats. Place fields recorded from one young rat (upper panel) and one old rat (lower panel) on two sessions of track running recorded about 1 hr apart. Only a subset of cells recorded is shown, each cell is a different color in each rat. The dots represent spikes of a given CA1 pyramidal cell. Note the changes in the distribution of individual place fields across sessions for the old rat, suggesting ‘‘remapping,’’ but stability of the fields from one session to the next for the young rat.
features of the environment by moving the start box closer to the reward site during the rat’s outbound journey so that the return journey back to the start box is shorter, the CA1 map realigns during the rat’s journey from an alignment with the origin to an alignment with the destination (Gothard et al., 1996a). Hippocampal map realignment can also be observed in young rats required to run from a movable start box and to pause at a hidden goal location at a fixed location relative to stable distal cues (roomaligned coordinate frame) in order to receive a medial forebrain bundle stimulation reward. If the origin of each lap is varied by shifting the start box, the CA1 place field activity realigns from a representation that was box aligned to one that was room aligned (Fig. 23–7). If this realignment did not occur, the rat would be unable to find the goal location, which requires using a room-aligned coordinate frame. Importantly, the accuracy of young rats at pausing at the correct goal location is correlated with the position on the track of the map realignment from a start box reference frame to a room-aligned reference frame (Rosenzweig et al., 2003). Using this same paradigm, it has been shown that old rats are delayed at the point on the track where the
map realigns to the fixed environmental cues; that is, the old rats move further from the start box before the CA1 place map shifts to a room-aligned reference frame. Consistent with these physiological findings, old rats are also impaired at finding the goal location and there is a significant correlation between the ability to find the goal location and the location on the track of the map realignment (Rosenzweig et al., 2003). While there are no reports of CA1 map realignment in young rats given an NMDA receptor antagonist, one would predict that cessation of NMDA receptor activity would result in a hippocampal map realignment deficit. When CA1 and CA3 place cell recordings are pooled, it seems that spatial representations in old rats do not change when they should (for example, in response to major changes in the environment) (Tanila et al., 1997a; Tanila et al., 1997b). Combining results across areas, however, is problematic, since ensemble activity in these different subregions is dissociable (Lee et al., 2004b; Leutgeb et al., 2004; Vazdarjanova and Guzowski, 2004). This dissociation could reflect two competing functions of the hippocampal network: pattern completion versus pattern separation (e.g., Marr, 1971; McNaughton and Morris, 1987). More-
CIRCUIT CONTRIBUTIONS TO MEMORY DEFICITS
373
Figure 23–7. Hippocampal map realignment. A schematic of the apparatus and procedure used for measuring CA1 hippocampal map realignment in young and old rats. A. The rat exits a start box and traverses a linear track. B. At a fixed position in the room (aligned to distal room cues) is a hidden reward location. If the rat pauses at the correct location he receives a medial forebrain bundle stimulation reward. C. For some trials the track and start box are moved but the reward location remains fixed. In order for the rat to find the goal location he must use a room-aligned reference frame rather than one that is box aligned. Old rats are delayed at shifting from a boxaligned to a room-aligned reference frame and are thus impaired at finding the goal location, compared with young rats.
over, there is evidence of a dissociation of the effects of aging on CA1 and CA3 ensembles. While maps in CA1 and CA3 of aged rats are both stable within a single session of environmental exploration, in CA1 spatial representations are less stable across sessions in aged than in young rats (Barnes et al., 1997). By contrast, spatial representations in CA3 seem to be more rigid in aged rats. When an aged rat explores a familiar environment for 7 min and is then placed into a novel environment, spatial representations in CA3 remain the same even though the environment has changed (Wilson et al., 2005). In young rats, however, CA3 place maps are independent between familiar and novel environments (Leutgeb et al., 2004; Wilson et al., 2005). A disruption in the ensemble characteristics of dentate gyrus granule cells, a structure known to be particularly vulnerable to the aging process (Small et al., 2002; Small et al., 2004), could con-
tribute to the failure of aged CA3 networks to form new spatial representations. It is believed that the dentate gyrus makes information stored in hippocampal networks more dissimilar (i.e., pattern separation), thereby increasing storage capacity (Marr, 1971). Because the transfer of information between granule cells and CA3 pyramidal cells may be degraded, this may contribute to the inability of the aged CA3 network to form new spatial representations when it should (for review, see Wilson et al., 2006).
FROM THE MAP TO THE MEMORY: AGING ENSEMBLES AND MEMORY DECLINE Since the seminal observation that hippocampal pyramidal cells are active at discrete locations in an
374 PLACE FIELDS AND AGE-RELATED CHANGES IN MEMORY environment (O’Keefe and Dostrovsky, 1971), a central question has been what, in fact, place cells encode. Because the external sensory input to the rat is not necessarily the input to the hippocampus, it might not be possible to determine what place cell firing reflects without knowing how external input is transformed and delivered to the hippocampus, and what internal inputs are received by the hippocampus (e.g., motivational states of the animal). Moreover, it has been suggested that place cell firing does not represent any single stimulus or even the configuration of stimuli. Rather, it has been proposed that hippocampal output provides an index code for each memory (Marr, 1971; Teyler and DiScenna, 1986). In this model, the hippocampal ‘‘code’’ is the distribution of place field activity. These hippocampal activity patterns become linked to the patterns of activity in other areas of the cortex, thereby acting as an index. Thus, reactivating the hippocampal ‘‘code’’ will evoke the associated patterns of activity in other brain structures (for review, see Sutherland and McNaughton, 2000). Understanding the process by which hippocampal activity becomes associated with patterns of cortical activity can provide insight into how memory is consolidated and retrieved, and why memory impairments are present in aged humans and other animals. It is clear from both human (Scoville and Milner, 1957; Squire et al., 1993; Teng and Squire, 1999) and animal lesion studies (Winocur, 1990; Zola-Morgan and Squire, 1990; Kim and Fanselow, 1992; Kim et al., 1995) that the hippocampus is necessary for the initial acquisition of certain types of memory but only temporarily necessary for the retrieval of these memories (but see Nadel and Moscovitch, 1997; Ryan et al., 2001). This supports standard theories of memory consolidation, which contend that while the neocortex is the site of storage for long-term memory, the hippocampus is necessary for indirectly associating independent regions of the neocortex. The hypothesis is that the hippocampus serves as an indirect linking system for neocortical regions that can maintain the memory more efficiently and for a longer period of time. This indirect association between independent cortical sites is possible because the hippocampus receives highly processed, polymodal input from most cortical areas and returns projections, via the entorhinal, perirhinal, and postrhinal cortices back to these areas (Swanson and Kohler, 1986; Felleman and Van Essen, 1991; Amaral and Witter, 1995; McIntyre et al., 1996; Insausti et al., 1997; Insausti and Munoz, 2001; Lavenex et al., 2002). It is these back-projections from the hippocampus to the neocortex that provide a rapid linking system for associations distributed across the brain, which lack the intrinsic connectivity to form these associations without the hippocampus (e.g., Mc
Clelland et al., 1995). The physiological mechanism by which the formation of novel connections between the neocortical regions encoding different aspects of a memory occurs is likely the reactivation of memory traces during periods when the brain is not acquiring new information. Such ‘‘off-line’’ reactivation during quiet wakefulness or sleep is hypothesized to lead to a reorganization of cortico–cortical connections and the strengthening of the memory trace (Marr, 1971; for review, see Sutherland and McNaughton, 2000). Pavlides and Winson (1989) were the first to report that the firing rate activity patterns of CA1 neurons during behavior influenced the firing-rate activity patterns during the subsequent sleep episode (Pavlides and Winson, 1989). Since this initial report, additional empirical evidence that memories are reactivated during sleep has been obtained from ensemble recordings of multiple neurons in the hippocampus (Wilson and McNaughton, 1994; Shen et al., 1998; Kudrimoti et al., 1999; Hirase et al., 2001) and neocortex (Qin et al., 1997; Hoffman and McNaughton, 2002; Ji and Wilson, 2007). When a young rat explores an environment, CA1 pyramidal cells with overlapping place fields have correlated firing patterns. It has been shown that cells with correlated firing patterns due to overlapping place fields maintain this correlation in the subsequent slow-wave sleep period following this episode (Wilson and McNaughton, 1994; Shen et al., 1998). Therefore, a significant amount of the variance in correlated firing rates of a population of CA1 neurons during slow-wave sleep following the behavior, but not preceding that behavior, can be explained by the correlated firing rates of CA1 neurons during the behavior (Wilson and McNaughton, 1994). This phenomenon can be quantified using the measure of explained variance (Fig. 23–8), which describes how much of the variability in the firing patterns of neurons during a rest period can be explained by the activity patterns of the neurons during the preceding behavioral experience. Therefore, an explained variance value that is ‘‘significant’’ (or greater than expected by chance) reflects the case where the activity patterns established during behavior are recapitulated during the subsequent rest episode. Interestingly, it has been shown that activity pattern reactivation in the hippocampus occurs primarily during large, irregular-amplitude local field potentials (Vanderwolf, 1969) called sharp wave events (Buzsaki, 1986). Superimposed on the sharp-wave events are high-frequency ( 200 Hz) oscillations called ‘‘ripples’’ (O’Keefe and Nadel, 1978). Ripples are associated with hippocampal pyramidal cell discharge and can be detected near the CA1 stratum pyramidale (Ylinen et al., 1995). Sharp waves occur about once or twice a second during immobilization and slow-wave
Figure 23–8. Procedure for calculating the explained-variance (EV) measure of activity pattern reactivation. The matrix of correlations of firing rates between all possible cell pairs that were active during an episode of behavior is computed (omitting within-tetrode correlations, black squares) for the rest 1 (pre, left panel), behavior (middle panel), and rest 2 (post, right panel) epochs. These are referred to as the R matrices. Both the xand the y-axes represent the cell ID. Red indicates cell pairs that have high firing-rate correlations (r ¼ ~0.4), while dark blue indicates that there was no correlation (see color bar) between the cell pair (r ¼ 0.0). The correlations between each pair of R matrices are then calculated. Next, the explained variance between the behavior and subsequent rest epoch is calculated by using the partial correlation coefficient to eliminate the contribution of the pre-behavior rest on the rest 2–behavior correlation. This value is squared to yield the explained variance between the rest 2 and maze-running epochs. Notice that the patterns of correlations in the R matrices are more similar between rest 2 and behavior compared to the patterns between rest 1 and behavior.
376 PLACE FIELDS AND AGE-RELATED CHANGES IN MEMORY sleep (Buzsaki, 1986) and result from a depolarization of CA1 pyramidal cells and interneurons by the CA3 afferents (Buzsaki, 1986, 1989; Chrobak and Buzsaki, 1994; Ylinen et al., 1995; Chrobak and Buzsaki, 1996). Simultaneous recordings of hippocampal EEG and multiple pyramidal cells have shown that the explained variance measure of activity pattern reactivation is significantly higher during the sharp wave/ripple event than during the inter-sharp wave interval (Kudrimoti et al., 1999). This finding is consistent with the idea that the sharp waves, and the associated ripple event, correspond to a convergence of the network onto an ‘‘attractor state’’ that represents a stored memory. These events may thus lead to the reactivation of the memory trace (Shen and McNaughton, 1996). Moreover, sharp wave events during sleep correlate with the transition from periods of low cortical activity (‘‘down-state’’) to periods of synchronous high cortical activity (‘‘up-state’’) (Battaglia et al., 2004; but see, Sirota et al., 2003). The observation that sharp waves facilitate transitions from neocortical down-states to up-states may reflect joint memory trace reactivation in the hippocampus and in the neocortex, possibly contributing to the formation of cortico–cortical connections and the consolidation of long-term memory. Importantly, the incidence and amplitude of sharp waves and the frequency of the corresponding ripple event is the same in young and old rats (Smith et al., 2000; Gerrard et al., 2001). Although activity pattern reactivation in young rats is blocked by antagonism of the NMDA receptor (Stanis et al., 2004), it remains intact in the aged rat (Gerrard et al., 2001). For a single epoch of running on a maze, aged rats show significantly greater explained variance for the subsequent sleep episode and the behavior preceding it,than the observed explained variance between the behavior and the pre-running sleep episode. Moreover, the magnitude of the explained variance is similar between the young and the old rats (Gerrard et al., 2001). Importantly, akin to the young rats, aged rats also show greater activity pattern reactivation during the sharp wave/ripple epochs than during the inter-sharp-wave intervals (Gerrard et al., 2001). These data imply that the mechanisms necessary to reactivate patterns of neural ensembles that were involved in an episode of behavior remain relatively intact in advanced age. In addition to activity pattern reactivation, CA1 neurons have a temporal order in their firing patterns in relation to one another during behavior, and they retain this temporal order during subsequent sleep (Skaggs and McNaughton, 1996; Nadasdy et al., 1999; Lee and Wilson, 2002). Quantification of the extent of preservation of the temporal pattern of cell firing during behavior can be extracted from cross-correlograms
among cells. That is, the cross-correlations of neurons that were active during an episode of behavior are compared with the cross-correlations between cells during both the preceding and the subsequent sleep episodes (Fig. 23–9A). From this comparison, one can calculate the ‘‘temporal bias’’ or tendency for cell pairs to fire in the same sequence during post-behavior rest episodes as during the actual behavior. For example, if neuron A fires before neuron B during behavior, this is reflected in the shape of the cross-correlogram. During subsequent sleep, the cross-correlogram for cells A and B will retain a shape similar to what was observed during behavior. If the similarity between the cross-correlograms from the behavior and subsequent rest episode is greater compared to the similarity of the cross-correlograms between the behavior and the preceding rest episode, then this is taken as evidence that the temporal order of the activity among CA1 neurons during behavior is, in fact, preserved during sleep (Skaggs and McNaughton, 1996). In old rats, there is a significant decrease in the amount of temporal bias reactivation compared to that of young rats (Fig. 23–9B). This finding suggests that while the cell pairs active together during behavior in old rats reactivate during subsequent rest periods, they do not necessarily reactivate in the same temporal order as that during behavior (Gerrard et al., 2002). An impairment in sequence reactivation could be related to the age-related plasticity deficits and the lack of experience-dependent place field expansion plasticity in CA1 neurons of aged rats (Shen et al., 1997). As discussed in a previous section, place field expansion plasticity could be a necessary mechanism for sequence encoding. In support of this idea, aged rats show deficits in sequence learning (Winter, 1997; Oler and Markus, 1998). Additionally, in both young (Fig. 23– 9C) and old (Fig. 23–9D) rats the amount of temporal bias reactivation was significantly correlated with performance on the spatial version of the Morris swim task (Gerrard et al., 2002). Thus, sequence reactivation could be an essential mechanism for the learning and/or maintenance of hippocampal-dependent tasks. Not surprisingly, in young rats, temporal bias reactivation is also blocked by antagonism of the NMDA receptor (Stanis et al., 2004). Another mechanism that could promote the reactivation of the sequential firing patterns of neurons in sleep episodes could be sharp wave/ripple events that occur during the behavior. Recently, it has been shown that cell assemblies active during the acquisition of a memory are repeated during sharp wave/ ripple events that occur during behavior or wakeful rest (Foster and Wilson, 2006; Jackson et al., 2006; O’Neill et al., 2006). While an animal is performing a behavior, the theta rhythm dominant in the EEG is
CIRCUIT CONTRIBUTIONS TO MEMORY DEFICITS
interrupted when the animal stops at a food dish to eat or stops to groom. When this occurs, sharp wave/ ripple events are observed (Foster and Wilson, 2006; Jackson et al., 2006; O’Neill et al., 2006). Additionally, theta during active exploration can be interrupted by sharp wave/ripples events (O’Neill et al., 2006). These
377
non-rest sharp wave/ripples events contain the firing patterns of recently visited spatial locations along with the cell assembly active in the current location (Foster and Wilson, 2006; O’Neill et al., 2006). Therefore, in addition to experience-dependent place field expansion plasticity, non-rest sharp wave/ripple events may
Figure 23–9. Sequence reactivation in young and old rats. A. Representative cross-correlation histograms used to calculate the temporal bias measure of sequence reactivation. In each of the three plots, the temporal bias of a cell pair is defined as the difference between the cross-correlagram integrated from 0 to 200 ms (shown in dark gray—POST) and the cross-correlagram integrated from – 200 to 0 ms (shown in light gray—PRE). Notice that there is no temporal bias in the cell pair during REST1. Prior to track-running behavior the area of light gray and dark gray are similar (top panel). The middle panel shows the crosscorrelation for the same pair of cells during the behavior. In this cell pair, cell 2 tends to fire after cell 1 during track running, as the dark gray area is larger than the light gray area. Finally, the bottom panel shows the cross-correlation histogram for the cell pair during REST2. Notice that the tendency for cell 2 to fire after cell 1 that was formed during track running is maintained during REST2, as seen by the much larger area shaded dark gray (the 0 to 200 ms window) versus the area shaded light gray (the –200 to 0 ms window). B. Using this measure of sequence reactivation, it has been observed that old rats show a significantly smaller correlation of cell pair cross-correlations between the behavior and the subsequent sleep episode (REST 2) compared to young rats. In both young (C) and old (D) rats, measures of temporal bias are significantly correlated with performance on the spatial version of the Morris swim task. For both C and D, the x-axis shows the normalized performance scores for all rats and the y-axis shows the corresponding normalized temporal bias score (data from Gerrard et al., 2002).
378 PLACE FIELDS AND AGE-RELATED CHANGES IN MEMORY
Figure 23–9. (Continued )
also facilitate the formation of place-related cell assemblies through facilitating connections between cells with overlapping place fields. During repetitive behaviors, a phase sequence of cell assemblies could develop (Hebb, 1949) and reactivate during sleep (Foster and Wilson, 2006; Jackson et al., 2006; O’Neill et al., 2006). Consistent with this idea, the number of sharp wave/ripple events emitted has been shown to increase with experience during behaviors that require repetition. It has been suggested that these non-sleep sharp wave/ripple events arise from experience-dependent plasticity that occurs during behavior (Jackson et al., 2006). Since old rats have deficits in this form of plasticity (Shen et al., 1997),
these sharp wave/ripple events during behavior may not occur with the same incidence in old animals as they do in young ones. This prediction needs to be examined directly, however.
CONCLUSIONS Old rats have notable differences in the dynamic properties of CA1 place fields, and several of these differences correspond with observed age-associated behavioral deficits. Aged rats fail to show experiencedependent place field expansion plasticity to the same extent as young rats (Shen et al., 1997). Because it has
CIRCUIT CONTRIBUTIONS TO MEMORY DEFICITS
been suggested that a mechanism similar to this phenomenon could serve to encode sequences of events (Hebb, 1949), it is not surprising that aged animals are impaired at learning sequences (Winter, 1997; Oler and Markus, 1998). Between episodes of experience in a single environment, aged rats are also impaired at maintaining stable spatial representations in the CA1 subregion of the hippocampus. This observation is consistent with the finding that the performance of old rats on tasks that require an allocentric spatial reference frame to be solved is impaired (Barnes et al., 1997; Rosenzweig et al., 2003). Additionally, once a memory trace has been formed in the hippocampus, aged rats are able to reactivate the activity patterns of the CA1 neurons associated with the memory (Gerrard et al., 2001), but cannot accurately reactivate the sequential patterns of the neural activity (Gerrard et al., 2002). Again, this electrophysiological correlate of consolidation is significantly associated with a given animal’s behavior on a spatial task (Gerrard et al., 2002). Thus, there is mounting evidence that some of the cognitive capabilities of old and young animals are reflected by the integrity of information processing in the hippocampus, and by improving how aged hippocampal circuits function one might be able to improve some areas of cognition. Considering that the average lifespan is increasing worldwide, an understanding of the brain mechanisms that are responsible for age-related cognitive impairment and development of therapeutic agents that might curb this decline are becoming increasingly urgent. acknowledgments We would like to thank Michelle Carroll and Luann Snyder for their administrative assistance. Additionally, we extend our deep thanks to Andrew Maurer for help with the figures, and to Zaneta Navritilova and Thabelo Khoboko for insightful editorial comments. This work was supported by Arizona Department of Health Services Grant AGR2007–37, the Evelyn F. McKnight Brain Research Foundation, and NIH grants AG003376, AG012609, NS054465.
References Agnihotri NT, Hawkins RD, Kandel ER, Kentros C (2004) The long-term stability of new hippocampal place fields requires new protein synthesis. PNAS 101: 3656–3661. Amaral D, Witter M (1995) Hippocampal formation. In: The Rat Nervous System, second edition (Paxinos G, ed.), pp 443–486. San Diego: Academic Press. Barnes CA (1979) Memory deficits associated with senescence: a neurophysiological and behavioral study in the rat. J Comp Physiol Psychol 93:74–104.
379
Barnes CA, McNaughton BL, O’Keefe J (1983) Loss of place specificity in hippocampal complex spike cells of senescent rat. Neurobiol Aging 4:113–119. Barnes CA, Rao G, McNaughton BL (1996) Functional integrity of NMDA-dependent LTP induction mechanisms across the lifespan of F-344 rats. Learn Mem 3:124–137. Barnes CA, Rao G, Houston FP (2000) LTP induction threshold change in old rats at the perforant path—granule cell synapse. Neurobiol Aging 21:613– 620. Barnes CA, Suster MS, Shen J, McNaughton BL (1997) Multistability of cognitive maps in the hippocampus of old rats. Nature 388:272–275. Bear MF, Malenka RC (1994) Synaptic plasticity: LTP and LTD. Curr Opin Neurobiol 4:389–399. Bear MF, Cooper LN, Ebner FF (1987) A physiological basis for a theory of synapse modification. Science 237:42–48. Bliss TV, Collingridge GL (1993) A synaptic model of memory: long-term potentiation in the hippocampus. Nature 361:31–39. Brun VH, Otnass MK, Molden S, Steffenach HA, Witter MP, Moser MB, Moser EI (2002) Place cells and place recognition maintained by direct entorhinalhippocampal circuitry. Science 296:2243–2246. Burke SN, Barnes CA (2006) Neural plasticity in the ageing brain. Nat Rev Neurosci 7:30–40. Burke SN, Maurer AP, Navratilova Z, McNaughton BL, Barnes CA (2005) The non-competitive NMDA receptor antagonist memantine reinstates experiencedependent plasticity in aged rats. Soc Neurosci Abstr 317:12. Buzsaki G (1986) Hippocampal sharp waves: their origin and significance. Brain Res 398:242–252. Buzsaki G (1989) Two-stage model of memory trace formation: a role for ‘‘noisy’’ brain states. Neuroscience 31:551–570. Chowdhury S, Shepherd JD, Okuno H, Lyford G, Petralia RS, Plath N, Kuhl D, Huganir RL, Worley PF (2006) Arc/Arg3.1 interacts with the endocytic machinery to regulate AMPA receptor trafficking. Neuron 52:445– 459. Chrobak JJ, Buzsaki G (1994) Selective activation of deep layer (V-VI) retrohippocampal cortical neurons during hippocampal sharp waves in the behaving rat. J Neurosci 14:6160–6170. Chrobak JJ, Buzsaki G (1996) High-frequency oscillations in the output networks of the hippocampalentorhinal axis of the freely behaving rat. J Neurosci 16:3056–3066. Cole AJ, Saffen DW, Baraban JM, Worley PF (1989) Rapid increase of an immediate early gene messenger RNA in hippocampal neurons by synaptic NMDA receptor activation. Nature 340:474–476.
380 PLACE FIELDS AND AGE-RELATED CHANGES IN MEMORY Deupree DL, Turner DA, Watters CL (1991) Spatial performance correlates with in vitro potentiation in young and aged Fischer 344 rats. Brain Res 554:1–9. Dieguez D, Jr., Barea-Rodriguez EJ (2004) Aging impairs the late phase of long-term potentiation at the medial perforant path-CA3 synapse in awake rats. Synapse 52:53–61. Ekstrom AD, Meltzer J, McNaughton BL, Barnes CA (2001) NMDA receptor antagonism blocks experiencedependent expansion of hippocampal ‘‘place fields’’. Neuron 31:631–638. Farr SA, Banks WA, La Scola ME, Flood JF, Morley JE (2000) Permanent and temporary inactivation of the hippocampus impairs T-maze footshock avoidance acquisition and retention. Brain Res 872:242–249. Felleman DJ, Van Essen DC (1991) Distributed hierarchical processing in the primate cerebral cortex. Cereb Cortex 1:1–47. Fletcher BR, Calhoun ME, Rapp PR, Shapiro ML (2006) Fornix lesions decouple the induction of hippocampal arc transcription from behavior but not plasticity. J Neurosci 26:1507–1515. Flood DG (1991) Region-specific stability of dendritic extent in normal human aging and regression in Alzheimer’s disease. II. Subiculum. Brain Res 540:83–95. Flood DG (1993) Critical issues in the analysis of dendritic extent in aging humans, primates, and rodents. Neurobiol Aging 14:649–654. Flood DG, Guarnaccia M, Coleman PD (1987) Dendritic extent in human CA2-3 hippocampal pyramidal neurons in normal aging and senile dementia. Brain Res 409:88–96. Foster DJ, Wilson MA (2006) Reverse replay of behavioural sequences in hippocampal place cells during the awake state. Nature 440:680–683. Foster TC, Norris CM (1997) Age-associated changes in Ca(2þ)-dependent processes: relation to hippocampal synaptic plasticity. Hippocampus 7:602–612. Gazzaley AH, Thakker MM, Hof PR, Morrison JH (1997) Preserved number of entorhinal cortex layer II neurons in aged macaque monkeys. Neurobiol Aging 18: 549–553. Gerrard JL, Kudrimoti H, McNaughton BL, Barnes CA (2001) Reactivation of hippocampal ensemble activity patterns in the aging rat. Behav Neurosci 115: 1180–1192. Gerrard JL, Kudrimoti HK, Cowen SL, Redish AD, Rosenzweig ES, Barnes CA, McNaughton BL (2002) Dissociation of pattern and sequence reactivation efficiency in the aged rat hippocampus. Soc Neurosci Abstr 678:7. Gothard KM, Skaggs WE, McNaughton BL (1996a) Dynamics of mismatch correction in the hippocampal ensemble code for space: interaction between path
integration and environmental cues. J Neurosci 16: 8027–8040. Gothard KM, Skaggs WE, Moore KM, McNaughton BL (1996b) Binding of hippocampal CA1 neural activity to multiple reference frames in a landmark-based navigation task. J Neurosci 16:823–835. Guzowski JF, McNaughton BL, Barnes CA, Worley PF (1999) Environment-specific expression of the immediate-early gene Arc in hippocampal neuronal ensembles. Nat Neurosci 2:1120–1124. Guzowski JF, Lyford GL, Stevenson GD, Houston FP, McGaugh JL, Worley PF, Barnes CA (2000) Inhibition of activity-dependent arc protein expression in the rat hippocampus impairs the maintenance of long-term potentiation and the consolidation of long-term memory. J Neurosci 20:3993–4001. Guzowski JF, Miyashita T, Chawla MK, Sanderson J, Maes LI, Houston FP, Lipa P, McNaughton BL, Worley PF, Barnes CA (2006) Recent behavioral history modifies coupling between cell activity and Arc gene transcription in hippocampal CA1 neurons. PNAS 103:1077–1082. Hanks SD, Flood DG (1991) Region-specific stability of dendritic extent in normal human aging and regression in Alzheimer’s disease. I. CA1 of hippocampus. Brain Res 540:63–82. Hebb D (1949) The Organization of Behavior: A Neurophysiological Theory.New York: Wiley. Hirase H, Leinekugel X, Czurko A, Csicsvari J, Buzsaki G (2001) Firing rates of hippocampal neurons are preserved during subsequent sleep episodes and modified by novel awake experience. PNAS 98:9386– 9390. Hoffman KL, McNaughton BL (2002) Coordinated reactivation of distributed memory traces in primate neocortex. Science 297:2070–2073. Insausti R, Munoz M (2001) Cortical projections of the non-entorhinal hippocampal formation in the cynomolgus monkey (Macaca fascicularis). Eur J Neurosci 14:435–451. Insausti R, Herrero MT, Witter MP (1997) Entorhinal cortex of the rat: cytoarchitectonic subdivisions and the origin and distribution of cortical efferents. Hippocampus 7:146–183. Jackson JC, Johnson A, Redish AD (2006) Hippocampal sharp waves and reactivation during awake states depend on repeated sequential experience. J Neurosci 26:12415–12426. Ji D, Wilson MA (2007) Coordinated memory replay in the visual cortex and hippocampus during sleep. Nat Neurosci 10:100–107. Jung MW, McNaughton BL (1993) Spatial selectivity of unit activity in the hippocampal granular layer. Hippocampus 3:165–182.
CIRCUIT CONTRIBUTIONS TO MEMORY DEFICITS
Kentros C, Hargreaves E, Hawkins RD, Kandel ER, Shapiro M, Muller RV (1998) Abolition of long-term stability of new hippocampal place cell maps by NMDA receptor blockade. Science 280:2121–2126. Keuker JI, Luiten PG, Fuchs E (2003) Preservation of hippocampal neuron numbers in aged rhesus monkeys. Neurobiol Aging 24:157–165. Kim JJ, Fanselow MS (1992) Modality-specific retrograde amnesia of fear. Science 256:675–677. Kim JJ, Clark RE, Thompson RF (1995) Hippocampectomy impairs the memory of recently, but not remotely, acquired trace eyeblink conditioned responses. Behav Neurosci 109:195–203. Kowalska DM (1995) Effects of hippocampal lesions on spatial delayed responses in dog. Hippocampus 5: 363–370. Kudrimoti HS, Barnes CA, McNaughton BL (1999) Reactivation of hippocampal cell assemblies: effects of behavioral state, experience, and EEG dynamics. J Neurosci 19:4090–4101. Kumar A, Foster TC (2005) Intracellular calcium stores contribute to increased susceptibility to LTD induction during aging. Brain Res 1031:125–128. Landfield PW (1988) Hippocampal neurobiological mechanisms of age-related memory dysfunction. Neurobiol Aging 9:571–579. Lavenex P, Suzuki WA, Amaral DG (2002) Perirhinal and parahippocampal cortices of the macaque monkey: projections to the neocortex. J Comp Neurol 447:394–420. Lavenex PB, Amaral DG, Lavenex P (2006) Hippocampal lesion prevents spatial relational learning in adult macaque monkeys. J Neurosci 26:4546–4558. Lee AK, Wilson MA (2002) Memory of sequential experience in the hippocampus during slow wave sleep. Neuron 36:1183–1194. Lee HK, Min SS, Gallagher M, Kirkwood A (2005) NMDA receptor-independent long-term depression correlates with successful aging in rats. Nat Neurosci 8:1657–1659. Lee I, Rao G, Knierim JJ (2004a) A double dissociation between hippocampal subfields: differential time course of CA3 and CA1 place cells for processing changed environments. Neuron 42:803–815. Lee I, Yoganarasimha D, Rao G, Knierim JJ (2004b) Comparison of population coherence of place cells in hippocampal subfields CA1 and CA3. Nature 430: 456–459. Leutgeb JK, Leutgeb S, Moser E, Moser MB (2007) Pattern separation in the dentate gyrus. Science 315: 961–966. Leutgeb S, Leutgeb JK, Treves A, Moser MB, Moser EI (2004) Distinct ensemble codes in hippocampal areas CA3 and CA1. Science 305:1295–1298.
381
Malenka RC (2003) Synaptic plasticity and AMPA receptor trafficking. Ann N Y Acad Sci 1003:1–11. Malinow R, Malenka RC (2002) AMPA receptor trafficking and synaptic plasticity. Annu Rev Neurosci 25:103–126. Markham JA, McKian KP, Stroup TS, Juraska JM (2005) Sexually dimorphic aging of dendritic morphology in CA1 of hippocampus. Hippocampus 15: 97–103. Markus EJ, Barnes CA, McNaughton BL, Gladden VL, Skaggs WE (1994) Spatial information content and reliability of hippocampal CA1 neurons: effects of visual input. Hippocampus 4:410–421. Markus EJ, Qin YL, Leonard B, Skaggs WE, McNaughton BL, Barnes CA (1995) Interactions between location and task affect the spatial and directional firing of hippocampal neurons. J Neurosci 15: 7079–7094. Marr D (1971) Simple memory: a theory for archicortex. Philos Trans R Soc Lond B Biol Sci 262:23–81. Maurer AP, Cowen SL, Burke SN, Barnes CA, McNaughton BL (2006) Organization of hippocampal cell assemblies based on theta phase precession. Hippocampus 16:785–794. McClelland JL, McNaughton BL, O’Reilly RC (1995) Why there are complementary learning systems in the hippocampus and neocortex: insights from the successes and failures of connectionist models of learning and memory. Psychol Rev 102:419–457. McIntyre DC, Kelly ME, Staines WA (1996) Efferent projections of the anterior perirhinal cortex in the rat. J Comp Neurol 369:302–318. McNaughton BL, Morris RG (1987) Hippocampal synaptic enhancement and information storage within a distributed memory system. Trends in Neurosciences 10:408–415. McNaughton BL, Barnes CA, O’Keefe J (1983) The contributions of position, direction, and velocity to single unit activity in the hippocampus of freelymoving rats. Exp Brain Res 52:41–49. Mehta MR, Barnes CA, McNaughton BL (1997) Experience-dependent, asymmetric expansion of hippocampal place fields. PNAS 94:8918–8921. Mehta MR, Quirk MC, Wilson MA (2000) Experiencedependent asymmetric shape of hippocampal receptive fields. Neuron 25:707–715. Merrill DA, Roberts JA, Tuszynski MH (2000) Conservation of neuron number and size in entorhinal cortex layers II, III, and V/VI of aged primates. J Comp Neurol 422:396–401. Merrill DA, Chiba AA, Tuszynski MH (2001) Conservation of neuronal number and size in the entorhinal cortex of behaviorally characterized aged rats. J Comp Neurol 438:445–456.
382 PLACE FIELDS AND AGE-RELATED CHANGES IN MEMORY Mizumori SJ, Lavoie AM, Kalyani A (1996) Redistribution of spatial representation in the hippocampus of aged rats performing a spatial memory task. Behav Neurosci 110:1006–1016. Moore CI, Browning MD, Rose GM (1993) Hippocampal plasticity induced by primed burst, but not long-term potentiation, stimulation is impaired in area CA1 of aged Fischer 344 rats. Hippocampus 3:57–66. Morris RG, Garrud P, Rawlins JN, O’Keefe J (1982) Place navigation impaired in rats with hippocampal lesions. Nature 297:681–683. Muller RU, Kubie JL, Ranck JB, Jr. (1987) Spatial firing patterns of hippocampal complex-spike cells in a fixed environment. J Neurosci 7:1935–1950. Muller RU, Bostock E, Taube JS, Kubie JL (1994) On the directional firing properties of hippocampal place cells. J Neurosci 14:7235–7251. Nadasdy Z, Hirase H, Czurko A, Csicsvari J, Buzsaki G (1999) Replay and time compression of recurring spike sequences in the hippocampus. J Neurosci 19:9497– 9507. Nadel L, Moscovitch M (1997) Memory consolidation, retrograde amnesia and the hippocampal complex. Curr Opin Neurobiol 7:217–227. Norris CM, Korol DL, Foster TC (1996) Increased susceptibility to induction of long-term depression and long-term potentiation reversal during aging. J Neurosci 16:5382–5392. O’Keefe J, Dostrovsky J (1971) The hippocampus as a spatial map. Preliminary evidence from unit activity in the freely-moving rat. Brain Res 34:171–175. O’Keefe J, Nadel L (1978) The Hippocampus as a Cognitive Map. Oxford: Claredon Press. O’Keefe J, Recce ML (1993) Phase relationship between hippocampal place units and the EEG theta rhythm. Hippocampus 3:317–330. Oler JA, Markus EJ (1998) Age-related deficits on the radial maze and in fear conditioning: hippocampal processing and consolidation. Hippocampus 8:402–415. Oler JA, Markus EJ (2000) Age-related deficits in the ability to encode contextual change: a place cell analysis. Hippocampus 10:338–350. O’Neill J, Senior T, Csicsvari J (2006) Place-selective firing of CA1 pyramidal cells during sharp wave/ ripple network patterns in exploratory behavior. Neuron 49:143–155. Pakkenberg B, Gundersen HJ (1997) Neocortical neuron number in humans: effect of sex and age. J Comp Neurol 384:312–320. Parsons CG, Quack G, Bresink I, Baran L, Przegalinski E, Kostowski W, Krzascik P, Hartmann S, Danysz W (1995) Comparison of the potency, kinetics and voltage-dependency of a series of uncompetitive NMDA receptor antagonists in vitro with anticon-
vulsive and motor impairment activity in vivo. Neuropharmacology 34:1239–1258. Pavlides C, Winson J (1989) Influences of hippocampal place cell firing in the awake state on the activity of these cells during subsequent sleep episodes. J Neurosci 9:2907–2918. Peters A, Leahu D, Moss MB, McNally KJ (1994) The effects of aging on area 46 of the frontal cortex of the rhesus monkey. Cereb Cortex 4:621–635. Plath N, Ohana O, Dammermann B, Errington ML, Schmitz D, Gross C, Mao X, Engelsberg A, Mahlke C, Welzl H, Kobalz U, Stawrakakis A, Fernandez E, Waltereit R, Bick-Sander A, Therstappen E, Cooke SF, Blanquet V, Wurst W, Salmen B, Bosl MR, Lipp HP, Grant SG, Bliss TV, Wolfer DP, Kuhl D (2006) Arc/Arg3.1 is essential for the consolidation of synaptic plasticity and memories. Neuron 52:437–444. Pyapali GK, Turner DA (1996) Increased dendritic extent in hippocampal CA1 neurons from aged F344 rats. Neurobiol Aging 17:601–611. Qin YL, McNaughton BL, Skaggs WE, Barnes CA (1997) Memory reprocessing in corticocortical and hippocampocortical neuronal ensembles. Philos Trans R Soc Lond B Biol Sci 352:1525–1533. Rapp PR, Gallagher M (1996) Preserved neuron number in the hippocampus of aged rats with spatial learning deficits. Proc Natl Acad Sci U S A 93:9926–9930. Rasmussen T, Schliemann T, Sorensen JC, Zimmer J, West MJ (1996) Memory impaired aged rats: no loss of principal hippocampal and subicular neurons. Neurobiol Aging 17:143–147. Rial Verde EM, Lee-Osbourne J, Worley PF, Malinow R, Cline HT (2006) Increased expression of the immediate-early gene arc/arg3.1 reduces AMPA receptormediated synaptic transmission. Neuron 52:461–474. Rogawski MA, Wenk GL (2003) The neuropharmacological basis for the use of memantine in the treatment of Alzheimer’s disease. CNS Drug Rev 9:275–308. Rosenzweig ES, Barnes CA (2003) Impact of aging on hippocampal function: plasticity, network dynamics, and cognition. Prog Neurobiol 69:143–179. Rosenzweig ES, Rao G, McNaughton BL, Barnes CA (1997) Role of temporal summation in age-related long-term potentiation-induction deficits. Hippocampus 7:549–558. Rosenzweig ES, Redish AD, McNaughton BL, Barnes CA (2003) Hippocampal map realignment and spatial learning. Nat Neurosci 6:609–615. Ryan L, Nadel L, Keil K, Putnam K, Schnyer D, Trouard T, Moscovitch M (2001) Hippocampal complex and retrieval of recent and very remote autobiographical memories: evidence from functional magnetic resonance imaging in neurologically intact people. Hippocampus 11:707–714.
CIRCUIT CONTRIBUTIONS TO MEMORY DEFICITS
Scoville WB, Milner B (1957) Loss of recent memory after bilateral hippocampal lesions. J Neurochem 20:11–21. Shen B, McNaughton BL (1996) Modeling the spontaneous reactivation of experience-specific hippocampal cell assembles during sleep. Hippocampus 6:685–692. Shen J, Kudrimoti HS, McNaughton BL, Barnes CA (1998) Reactivation of neuronal ensembles in hippocampal dentate gyrus during sleep after spatial experience. J Sleep Res 7 Suppl 1:6–16. Shen J, Barnes CA, McNaughton BL, Skaggs WE, Weaver KL (1997) The effect of aging on experiencedependent plasticity of hippocampal place cells. J Neurosci 17:6769–6782. Shepherd JD, Rumbaugh G, Wu J, Chowdhury S, Plath N, Kuhl D, Huganir RL, Worley PF (2006) Arc/Arg3.1 Mediates homeostatic synaptic scaling of AMPA receptors. Neuron 52:475–484. Sirota A, Csicsvari J, Buhl D, Buzsaki G (2003) Communication between neocortex and hippocampus during sleep in rodents. PNAS 100:2065–2069. Skaggs WE, McNaughton BL (1996) Replay of neuronal firing sequences in rat hippocampus during sleep following spatial experience. Science 271:1870–1873. Skaggs WE, McNaughton BL, Wilson MA, Barnes CA (1996) Theta phase precession in hippocampal neuronal populations and the compression of temporal sequences. Hippocampus 6:149–172. Small SA, Tsai WY, DeLaPaz R, Mayeux R, Stern Y (2002) Imaging hippocampal function across the human life span: is memory decline normal or not? Ann Neurol 51:290–295. Small SA, Chawla MK, Buonocore M, Rapp PR, Barnes CA (2004) Imaging correlates of brain function in monkeys and rats isolates a hippocampal subregion differentially vulnerable to aging. PNAS 101:7181– 7186. Smith AC, Gerrard JL, Barnes CA, McNaughton BL (2000) Effect of age on burst firing characteristics of rat hippocampal pyramidal cells. Neuroreport 11: 3865–3871. Squire LR, Knowlton B, Musen G (1993) The structure and organization of memory. Annu Rev Psychol 44: 453–495. Stanis J, Dees J, Gerrard J, Lipa P, VanRhoads S, McNaughton B, Barnes C (2004) Reactivation of hippocampal neural ensemble patterns is NMDA receptor-dependent. Soc Neurosci Abstr 329:20. Sutherland GR, McNaughton B (2000) Memory trace reactivation in hippocampal and neocortical neuronal ensembles. Curr Opin Neurobiol 10:180–186. Swanson LW, Kohler C (1986) Anatomical evidence for direct projections from the entorhinal area to the entire cortical mantle in the rat. J Neurosci 6:3010– 3023.
383
Tanila H, Shapiro M, Gallagher M, Eichenbaum H (1997a) Brain aging: changes in the nature of information coding by the hippocampus. J Neurosci 17:5155–5166. Tanila H, Sipila P, Shapiro M, Eichenbaum H (1997b) Brain aging: impaired coding of novel environmental cues. J Neurosci 17:5167–5174. Teng E, Squire LR (1999) Memory for places learned long ago is intact after hippocampal damage. Nature 400:675–677. Terrazas A, Krause M, Lipa P, Gothard KM, Barnes CA, McNaughton BL (2005) Self-motion and the hippocampal spatial metric. J Neurosci 25:8085–8096. Teyler TJ, DiScenna P (1986) The hippocampal memory indexing theory. Behav Neurosci 100:147–154. Thibault O, Landfield PW (1996) Increase in single Ltype calcium channels in hippocampal neurons during aging. Science 272:1017–1020. Thompson LT, Best PJ (1990) Long-term stability of the place-field activity of single units recorded from the dorsal hippocampus of freely behaving rats. Brain Res 509:299–308. Toescu EC, Verkhratsky A, Landfield PW (2004) Ca2þ regulation and gene expression in normal brain aging. Trends Neurosci 27:614–620. Tombaugh GC, Rowe WB, Chow AR, Michael TH, Rose GM (2002) Theta-frequency synaptic potentiation in CA1 in vitro distinguishes cognitively impaired from unimpaired aged Fischer 344 rats. J Neurosci 22: 9932–9940. Turner DA, Deupree DL (1991) Functional elongation of CA1 hippocampal neurons with aging in Fischer 344 rats. Neurobiol Aging 12:201–210. Vanderwolf CH (1969) Hippocampal electrical activity and voluntary movement in the rat. Electroencephalogr Clin Neurophysiol 26:407–418. Vanderwolf CH, Kramis R, Robinson TE (1977) Hippocampal electrical activity during waking behaviour and sleep: analyses using centrally acting drugs. Ciba Found Symp (58):199–226. Vazdarjanova A, Guzowski JF (2004) Differences in hippocampal neuronal population responses to modifications of an environmental context: evidence for distinct, yet complementary, functions of CA3 and CA1 ensembles. J Neurosci 24:6489–6496. West MJ, Coleman PD, Flood DG, Troncoso JC (1994) Differences in the pattern of hippocampal neuronal loss in normal ageing and Alzheimer’s disease. Lancet 344:769–772. Wilson I, S I, Gallagher M, Eichenbaum H, Tanila H (2005) Age-associated alterarions in place cells are subregion specific. J Neurosci 25:6877–6886. Wilson IA, Gallagher M, Eichenbaum H, Tanila H (2006) Neurocognitive aging: prior memories hinder new hippocampal encoding. Trends Neurosci 29:662–670.
384 PLACE FIELDS AND AGE-RELATED CHANGES IN MEMORY Wilson MA, McNaughton BL (1993) Dynamics of the hippocampal ensemble code for space. Science 261: 1055–1058. Wilson MA, McNaughton BL (1994) Reactivation of hippocampal ensemble memories during sleep. Science 265:676–679. Winocur G (1990) Anterograde and retrograde amnesia in rats with dorsal hippocampal or dorsomedial thalamic lesions. Behav Brain Res 38: 145–154.
Winter JC (1997) The effects of age on continuous learning in the radial maze. Physiol Behav 61:609–612. Ylinen A, Bragin A, Nadasdy Z, Jando G, Szabo I, Sik A, Buzsaki G (1995) Sharp wave-associated highfrequency oscillation (200 Hz) in the intact hippocampus: network and intracellular mechanisms. J Neurosci 15:30–46. Zola-Morgan SM, Squire LR (1990) The primate hippocampal formation: evidence for a time-limited role in memory storage. Science 250:288–290.
Author Index
Abbott, LF, 96, 319, 343 Abeles, M, 332 Abrahams, S, 177, 271 Adcock, RA, 111, 114 Addis, DR, 83, 153, 165 Agamanolis, D, 218 Aggleton, JP, 59, 68, 82, 97, 254, 259, 290 Agnihotri, NT, 277, 371 Agster, KL, 323 Aguirre, GK, 156, 177, 181, 185 Ainge, JA, 45, 47, 48, 49, 50, 51, 52, 53, 54, 55 Albertin, SV, 256, 297 Alexander, GE, 223, 296 Alonso, A, 322 Alvarez, P, 82, 97 Alvarez-Leefmans, FJ, 27 Alyan, S, 153 Alyan, SH, 162 Amaral, D, 374 Amaral, DG, 55, 84, 87, 192, 206, 237, 238, 241, 242, 259, 328, 332, 333, 338 Amit, DJ, 213, 290, 330, 332 Amzica, F, 302 Anagnostaras, SG, 5, 18, 261 Andersen, P, 84 Andersen, RA, 64, 156, 195 Anderson, AK, 117 Anderson, M, 62, 63, 83 Anderson, MI, 8, 18, 62, 66, 119, 120, 255, 258, 260 Anderson, RA, 62 Andrzejewski, ME, 273 Angeli, SJ, 83 Annett, LE, 256 Antal, M, 27, 259, 314 Apergis-Schoute, J, 262
Arduini, AA, 313 Arleo, A, 296 Asaka, Y, 313, 314 Asratyan, EA, 4 Assaf, SY, 27 Aston-Jones, G, 115, 116 Astur, RS, 185 Averbeck, BB, 127, 134 Bach, ME, 114, 115, 273 Baddeley, A, 261 Baeg, EH, 35, 55, 290, 291 Bai, D, 347 Bains, JS, 87 Baker, CL, 228 Balleine, BW, 297 Balogh, SA, 5 Bannerman, DM, 114, 117, 261 Banquet, JP, 291 Banta, Lavenex P, 83 Barbas, H, 290 Barea-Rodriguez, EJ, 368, 369, 370 Barnes, CA, 29, 93, 113, 182, 183, 187, 238, 294, 317, 353, 355, 357, 358, 364, 365, 367, 368, 369, 370, 371, 373, 379 Barnes, TD, 54 Barry, C, 29, 67, 151, 152, 255, 259 Bartlett, FC, 156 Barto, AG, 316 Battaglia, FP, 292, 294, 301, 302, 303, 333, 340, 376 Baxter, MG, 117, 282 Bayley, PJ, 218, 229 Beane, M, 115 Bear, MF, 371 Beason-Held, L, 219
385
386
AUTHOR INDEX
Beck, CH, 275 Beers, DR, 82, 97 Behr, J, 273, 274 Behrmann, M, 157 Bennett, AT, 60 Berger, TW, 45, 46, 65 Beritoff, JS, 4 Bernabeu, R, 114 Berns, GS, 261 Berridge, CW, 115, 116 Berridge, KC, 256 Berry, SD, 313, 314 Bertoglio, LJ, 256 Best, PJ, xvii, xviii, 16, 17, 74, 161, 343, 371 Bi, G, 347 Bi, GQ, 17, 321 Bienenstock, EL, 29 Bilkey, DK, 27, 254, 255, 256, 257, 258, 259, 260, 261, 262 Bingman, VP, 330 Bird, CM, 59 Birrell, JM, 291 Blades, M, 170 Blair, HT, 79, 248 Blanchard, DC, 83 Blanchard, RJ, 83 Bland, BH, 117, 313, 322 Bliss, TVP, 84, 127, 255, 258, 331, 332, 343, 347, 368 Blum, KI, 96, 319 Bohbot, VD, 271 Bolles, RC, 4 Bolte, A, 264 Bonnevie, T, 248 Bontempi, B, 6, 289, 302 Bostock, E, 28, 62, 154 Bouma, H, xvii Bouton, ME, 4, 260, 261 Bower, MR, 29, 44, 45, 47, 52, 53, 144, 166 Braitenberg, V, 330 Brankack, J, 281, 316, 317, 318 Brasted, PJ, 226 Brazhnik, ES, 115, 117, 282, 315 Breese, CR, 64, 91, 98, 144, 187, 255, 257, 258 Broadbent, N, 254 Broadbent, NJ, 271 Brown, JE, 62, 136, 189 Brown, MW, 59, 68, 220, 254, 259 Brown, VJ, 291 Brun, VH, 100, 238, 241, 355, 370 Brunel, N, 290 Bucci, DJ, 260 Buckingham, J, 332 Buckley, MJ, 259 Buckmaster, CA, 222 Bunsey, M, 167, 222 Bunsey, MD, 98 Burak, Y, 248 Bures, J, 131 Burgess, N, 19, 24, 33, 66, 67, 82, 83, 84, 85, 97, 100, 109, 151, 152, 154, 155, 156, 161, 162, 163, 164, 169, 171, 183, 212, 213, 248, 254, 255, 300, 316, 343
Burke, SN, 364, 367, 368, 370 Burnham, WH, 9 Burton, HL, 68 Burwell, RD, 237, 248, 259, 260, 291, 354 Bussey, TJ, 259 Buzsaki, G, 17, 19, 25, 26, 27, 28, 29, 30, 64, 108, 109, 110, 115, 117, 145, 262, 264, 302, 374, 376 Byrne, P, 152, 156, 157 Cacucci, F, 78, 248 Cahill, JF, 282 Cahusac, PMB, 177, 181, 194, 206, 220, 223, 226, 228, 230 Campeau, S, 275 Carmichael, ST, 192, 206 Carroll, RL, 328, 341 Cash, S, 241 Centonze, D, 296 Chang, Q, 271, 275, 282 Chawla, MK, 275 Chen, LL, 156, 238 Chiba, AA, 98 Chowdhury, S, 365 Chrobak, JJ, 264, 302, 376 Chun, MM, 114 Churchland, PS, 85 Cimadevilla, JM, 129 Cipolotti, L, 59 Clayton, DF, 274, 275 Clayton, NS, 82, 97, 254, 264, 330 Clearwater, J, 254, 255, 256, 257, 258, 261 Cohen, JD, 115, 116 Cohen, NJ, 18, 19, 82, 84, 97, 135, 162, 163, 164, 165, 167, 219, 222, 224, 353 Colby, CL, 35 Cole, AJ, 365 Collingridge, GL, 127, 331, 343, 368 Colombo, M, 220, 222, 223, 254 Colombo, PJ, 275 Compte, A, 290 Conner, JM, 117 Conway, DH, xviii, 16, 74, 78, 92, 93, 138, 151, 258, 260, 271 Cook, EP, 108 Cooper, BG, 30, 153, 259 Coover, GD, 256 Corcoran, KA, 6, 83 Corkin, S, 162, 163, 253 Cormen, TH, 343 Corodimas, KP, 260 Cowan, RL, 302 Cowan WM, 27 Cressant, A, 65, 151 Crutcher, MD, 296 Csicsvari, J, 262, 302 Czurko´, A, 24 Da Cunha, C, 273 Dahl, D, 27 Dalley, JW, 117 Dalrymple-Alford, JC, 259
AUTHOR INDEX Damasio, AR, 264, 289 Dan, Y, 17 Darlington, RB, 328 Davachi, L, 83, 85, 165 Davidson, TL, 98 Davies, M, 259 Daw, ND, 296 Day, M, 201 Dayan, P, 119, 120, 297 Deacon, TW, 237 Deadwyler, SA, 65, 164 de Araujo, IET, 206, 210 Debiec, J, 6, 277 de Bruin, JP, 291 deCharms, RC, 36 Deco, G, 201, 205, 207, 208, 210, 211, 212, 213 de Curtis, M, 119 Delatour, B, 291 Della Sala, S, 261 Derdikman, D, 68 Desimone, R, 115, 118 D’Esposito, M, 156, 177, 181, 185, 224, 261 Deupree, DL, 364, 368, 369 Diamond, IT, 328, 329 Diano, S, 259 Dickinson, A, 82, 114, 115, 264, 290 Dieguez, D Jr, 368, 369, 370 Dijksterhuis, A, 264 DiScenna, P, 152, 374 Disterhoft, JF, 45, 46, 115, 228 Disterhoft, JG, 16 Doane, B, xvii Dobbins, IG, 83 Dodson, CS, 156 Dolorfo, CL, 238, 242 Dostrovsky, J, vii, xviii, 3, 16, 73, 84, 91, 108, 138, 161, 177, 183, 219, 254, 271, 317, 330, 332, 353, 365, 374 Downs, RM, 60 Dragoi, G, 28, 29, 66, 110 Dubois, B, 272 Dudai, Y, 289, 302 Dudar, JD, 315 Dudchenko, PA, 46, 61, 68 Dujardin, K, 272 Duncan, J, 115, 261 Dusek, J, 167 Dusek, JA, 82, 97 Dutar, P, 259 Eacott, MJ, 259, 260 Egner, T, 262 Egorov, AV, 322, 323 Eichenbaum, H, 16, 18, 19, 23, 29, 64, 65, 74, 82, 97, 98, 119, 135, 139, 144, 161, 162, 163, 164, 165, 166, 167, 169, 177, 183, 185, 219, 220, 222, 224, 225, 228, 229, 237, 248, 254, 263, 264, 302, 315, 316, 317, 318, 319, 320, 321, 330, 353, 355 Eifuku, S, 177, 183 Ekstrom, AD, 24, 25, 68, 96, 165, 168, 185, 200, 254, 313, 366, 367, 370
387
Eldridge, LL, 83, 165 Elliot, T, 211 Emerich, DF, 87 Empson, RM, 262 Epstein, R, 170, 212, 213 Eschenko, O, 17, 19, 23, 34, 271, 278 Etienne, AS, 17, 152 Evarts, EV, xvii Everitt, BJ, 115 Fadda, F, 281 Fanselow, MS, 4, 5, 18, 82, 83, 84, 85, 97, 155, 261, 374 Farr, SA, 364 Faull, RLM, 297 Feigenbaum, JD, 91, 177, 181, 195 Felleman, DJ, 253, 374 Fenton, AA, 44, 118, 129, 130, 131, 141 Ferbinteanu, J, 17, 19, 21, 25, 27, 29, 44, 45, 49, 50, 52, 67, 82, 91, 97, 100, 111, 144, 166, 290 Ferino, F, 290 Feustle, WA, 162 Fibiger, HC, 275 Fiete, I, 248 Finlay, BL, 328 Fletcher, BR, 365 Fletcher, PC, 156 Flitman, S, 170 Flood, DG, 364 Floresco, SB, 55, 256, 281, 290 Flores-Hernandez, J, 274 Forbes, BR, xvii Formisano, E, 156 Fortin, NJ, 18, 83 Forwood, SE, 83 Foster, DJ, 26, 316, 376, 377, 378 Foster, T, 108 Foster, TC, 18, 24, 187, 196, 263–264, 368, 369, 371 Fox, SE, 17, 313, 314, 320 Franck, LM, 144 Frank, LM, 17, 24, 33, 44, 45, 53, 56, 66, 91, 108, 109, 111, 166, 228, 230, 238, 277, 317 Frank, MJ, 296 Frankland, PW, 289 Franse´n, E, 322, 323 Franzius, M, 243 Freeman, JH Jr, 18 Freund, TF, 27, 259, 314 Frey, S, 116 Frey, U, 33, 54, 110, 111, 219, 274, 281 Fried, I, 24, 200 Frith, CD, 262 Fritz, J, 118 Fruhmann-Berger, M, 157 Fu, Y-X, 108 Fuhs, MC, 77, 154, 247 Funahashi, S, 223, 290 Furusawa, AA, 178 Fuster, JM, 223, 261, 290 Fyhn, M, 7, 18, 77, 79, 91, 98, 111, 127, 237, 238, 239, 240, 243, 244, 245, 246, 247, 248, 255, 257, 258, 329, 355
388
AUTHOR INDEX
Gabrieli, JDE, 171 Gaffan, D, 193, 194, 226, 259, 262 Gallagher, M, 353, 354, 355, 364 Galletti, C, 153, 156 Gallistel, CR, 343 Gao, YJ, 256 Gardiner-Medwin, AR, 27, 332 Gasbarri, A, 262, 272, 273 Gasparini, S, 241 Gavrilov, VV, 17, 24 Gawne, T, 134 Gazzaley, AH, 364 Gemmell, C, 35 Gengler, S, 55 Georges-Franc¸ois, P, 93, 196, 197, 199, 200, 212, 213, 223 Georgopoulos, AP, 134 Gerrard, JL, 371, 376, 377, 379 Gerstner, W, 296, 343 Gevaerd, MS, 272 Gewirtz, JC, 6 Ghaem, O, 157, 177 Gil, Z, 117 Gilbert, PE, 30, 82, 83, 84, 89, 92 Gill, KM, 54, 113, 115, 271, 273, 276, 278 Giocomo, L, 117 Giocomo, LM, 248 Giovanello, KS, 165 Girard, B, 296 Gisquet-Verrier, P, 291 Givens, BS, 313, 314 Gluck, M, 18 Goddard, GV, 27, 259 Goddard, NH, 82, 84, 86, 93, 97, 360 Gold, AE, 89, 91 Gold, E, 29 Gold, JI, 261 Gold, PE, 27, 271, 275, 282 Goldberg, ME, 35, 107 Golding, N, 315 Goldman-Rakic, PS, 261, 290 Goldstein, LH, 271 Golob, EJ, 63, 147 Gomez-Isla, T, 248, 249 Good, M, 6, 19, 82 Gorny, B, 17 Goschke, T, 264 Gothard, KM, 23, 25, 91, 93, 98, 112, 118, 133, 138, 144, 166, 168, 360, 365, 372 Goto, Y, 256 Grabowsky, M, 262 Grace, AA, 28, 29, 111, 115, 281 Granon, S, 35, 291 Gray, JA, 18, 254 Graybiel, AM, 296 Green, C, 77, 78 Green, JD, 313 Griffin, AL, 313, 315, 323 Griffiths, D, 82, 97 Groenewegen, HJ, 262, 297 Gross, CG, 220, 222, 223
Grunwald, T, 281 Gulley, JM, 296 Gundersen, HJ, 364 Guzowski, JF, 12, 13, 29, 56, 62, 84, 85, 95, 96, 111, 177, 274, 275, 360, 365, 372 Haber, SN, 296 Habib, M, 181 Hafting, T, 12, 23, 33, 68, 77, 97, 108, 109, 147, 154, 210, 237, 238, 239, 240, 241, 242, 248, 255, 259, 297, 329 Hagler, JDJ, 109 Hahn, TT, 302, 304 Hall, G, 6 Hall, WC, 328 Hampson, RE, 30, 65, 109, 134, 136, 164, 166, 168, 220, 222, 223, 224, 225 Hampton, RR, 185, 264 Hanks, SD, 364 Hannula, DE, 154, 224 Hargreaves, EL, 119, 237, 245 Harley, CW, 116 Harnad, S, 263 Harris, KD, 118, 134, 152, 315 Harrison, S, 194 Hartley, T, 67, 151, 152, 154 Harzi, MM, 27 Hassabis, D, 12, 153, 155 Hasselmo, ME, 18, 26, 28, 29, 84, 86, 87, 115, 117, 282, 315–316, 317, 318, 319, 320, 321, 322, 323, 331 Hayes-Roth, B, 181, 185 Hayman, RM, 189 Hayward, A, 3 Healy, SD, 60 Hebb, D, 370, 378, 379 Hebb, DO, 9, 134 Heckers, S, 167 Heinemann, U, 262 Heit, G, 24 Henke, K, 165 Herrmann, T, 35 Hertz, J, 85, 89 Hetherington, PA, 24, 33, 64, 92, 255, 343 Hill, AJ, 16, 17 Hirase, H, 260, 374 Hirsch, J, 262 Hirsh, R, 18, 82, 85, 96, 98, 182, 260 Hitch, GJ, 155 Hobin, JA, 6 Hock, BJ Jr, 98 Hoffman, KL, 26, 302, 374 Hok, V, 64, 145, 146, 261, 290, 291, 300, 347 Holland, PC, 4, 260, 261 Hollerman, JR, 273 Hollup, SA, 64, 111, 144, 145, 166, 255, 257, 258, 300, 354 Holscher, C, 45, 50, 206, 258, 315, 317 Holt, W, 6, 83 Holt, WG, 83, 85 Honey, RC, 6, 19, 82, 85 Hopfield, JJ, 28, 85, 87, 89, 332 Hopkins, DA, 289
AUTHOR INDEX Hopkins, RO, 82, 84 Hori, E, 68, 177, 183, 185, 223 Houk, JC, 36, 296 Houston, FP, 5 Howard, M, 155 Howard, MW, 320 Hsu, KS, 274 Huang, Q, 274 Huang, Y, 111, 114, 115, 274 Hubel, D, 108 Hubel, DH, xvii Huerta, PT, 117, 315 Hunsaker, MR, 87 Hupbach, A, 8, 9, 10 Huxter, J, 315, 317 Hyman, JM, 26, 315, 317, 322 Idiart, MA, 322 Ihalainen, JA, 281 Iijima, T, 241 Ikemoto, S, 296 Ikonen, S, 115, 117, 282 Ikonen, SH, 27 Ino, T, 157 Insausti, R, 374 Isomura, Y, 100 Jackson, J, 300 Jackson, JC, 376, 377, 378 Jakab, RL, 27, 87 James, W, 8, 107 Jarrard, LE, 27, 98, 185, 254, 271 Jasper, H, xvii Jay, TM, 30, 34, 290 Jeffery, KJ, 8, 16, 17, 18, 23, 30, 52, 62, 63, 64, 66, 67, 85, 109, 119, 120, 143, 151, 152, 153, 155, 177, 189, 255, 257, 258, 260, 261 Jenkins, TA, 19, 85 Jensen, O, 319, 320, 321, 322 Jerison, HJ, 328 Jerman, T, 84 Ji, D, 302, 374 Jia, Z, 347 Joel, D, 296 Jog, MS, 54 Johnson, A, 114 Johnston, D, 344 Jones, MW, 26, 55, 56, 115, 261, 290, 322 Jones, TJ, 330 Jonides, J, 156 Judge, SJ, 316, 318 Jung, MW, 17, 25, 55, 56, 242, 243, 290, 291, 292, 294, 297, 365 Just, MA, 181 Kahana, MJ, 155, 320 Kalyani, A, 33, 93 Kandel, ER, 111, 114, 115, 274 Kang, DK, 275 Kanwisher, N, 170, 212, 213
389
Karnath, HO, 157 Kastner, S, 107 Katsuki, H, 115 Kavcic, V, 248 Kazniak, AW, 354 Kennedy, PJ, 8, 27, 83, 98, 258 Kentros, CG, 61, 110, 112, 111, 113, 114, 115, 119, 139, 140, 228, 258, 277, 361, 369, 371 Kesner, RP, 29, 30, 55, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91, 96, 97, 100, 201, 204, 210, 212, 213, 291, 338 Keuker, JI, 364 Kim, JJ, 4, 5, 18, 82, 83, 85, 97, 155, 261, 374 Kim, JS, 282 King, JA, 154, 155, 163 Kirwan, CB, 83 Kita, T, 185 Kitchin, R, 170 Kjelstrup, KG, 238, 242 Klein, SB, 12 Klement, D, 128 Klink, R, 322 Kloosterman, F, 238, 241 Knierim, JJ, 7, 17, 28, 91, 96, 98, 100, 112, 119, 120, 151, 153, 177, 264, 358 Knight, R, 85 Knight, RT, 111, 156, 262 Knowlton, BJ, 50, 290 Kobatake, E, 228 Kobayashi, T, 64, 91, 98, 139, 147, 178, 183, 187, 255, 257, 258 Kocsis, B, 313 Koene, RA, 315, 316, 317, 322, 343 Kohler, C, 238, 374 Kohler, S, 170 Kohonen, T, 87, 344 Konorski, J, 4 Korshunov, VA, 98, 297 Koulakov, AA, 26 Kowalska, DM, 364 Kramis, R, 313 Krebs, JR, 254, 330 Kreiman, G, 200 Kreiman, K, 168 Kronforst-Collins, MA, 115 Kruskal, JB, 183 Kubie, JL, xix, 16, 19, 24, 25, 33, 61, 62, 67, 73, 74, 78, 91, 92, 93, 108, 132, 136, 138, 139, 141, 151, 154, 183, 187, 210, 258, 315, 317 Kubik, S, 129, 131 Kudrimoti, HS, 301, 302, 374 Kumar, A, 368, 369, 371 Kumar, SS, 241, 248 Kumaran, D, 36, 111 Kusuki, T, 274 Kyd, RJ, 262 LaBar, KS, 6 Lanahan, A, 275 Land, C, 6 Landfield, PW, 369, 371
390
AUTHOR INDEX
Laplagne, DA, 337 Laroche, S, 291 Lassalle, J-M, 337 Latham, PE, 134 Laurent, B, 272 Lavenex, P, 374 Lavenex, PB, 271, 364 LeDoux, J, 264 LeDoux, JE, 4, 18, 83, 117, 155, 260, 261, 277 Lee, AC, 154 Lee, AK, 166, 301, 376 Lee, AW, 115, 116 Lee, D, 127, 134 Lee, HK, 368, 369 Lee, I, 25, 29, 45, 64, 82, 83, 84, 85, 87, 88, 89, 90, 91, 93, 94, 95, 96, 97, 98, 99, 100, 112, 144, 167, 260, 300, 323, 338, 340, 360, 370, 372 Lee, MG, 314 Legault, M, 281 Lehmann, H, 6 Lemon, N, 274 Lenck-Santini, PP, 7, 16, 45, 53, 61, 140, 141, 142, 143, 144, 145, 151, 254, 353, 361 Lepage, M, 83 Leranth, C, 27, 87 Leutgeb, JK, 7, 28, 66, 237, 243, 244, 245, 337, 340, 360, 365 Leutgeb, S, 19, 23, 27, 29, 56, 84, 85, 95, 96, 98, 108, 109, 117, 120, 237, 239, 243, 258, 260, 282, 294, 315, 333, 340, 360, 372, 373 Lever, C, 8, 28, 66, 109, 136, 155, 164 Levin, ED, 282 Levy, WB, 84, 86, 319, 321 Lewis, DJ, 9 Li, S, 111, 114, 115, 274 Li, W, 108 Li, XG, 84 Li, Y, 67 Liao, D, 347 Lin, JY, 274 Lippa, CF, 248, 249 Lipton, PA, 45, 55 Lisman, JE, 18, 28, 29, 84, 86, 111, 114, 115, 117, 280, 281, 315, 319, 320, 321, 322 Liu, P, 254, 262 Liu, Z, 225 Loftus, EF, 9, 156 Logothetis, NK, 228 Lomo, T, 84, 255, 258, 332 Lorente de No, R, 84 Louie, K, 166 Loy, R, 116 Lu, X, 260 Luiten, PGM, 313 Luria, A, 107 Ma, YY, 185 Maaswinkel, H, 291 MacArdy, EA, 5 Magee, JC, 241
Maguire, EA, 34, 36, 59, 111, 156, 157, 165, 171, 177, 181, 188, 254, 271 Majchrzak, M, 12 Makintosh, N, 119, 120 Malenka, RC, 368, 371 Maler, L, 256 Malinow, R, 368 Malkova, L, 193, 201, 315 Malouf, AT, 262 Manahan-Vaughan, D, 274 Manns, ID, 322 Manns, JR, 18, 36, 218, 323 Maren, S, 6, 18, 83, 85 Markham, JA, 364 Markowitsch, HJ, 82, 97, 155 Markram, H, 17 Markus, EJ, 17, 24, 28, 33, 66, 91, 93, 98, 111, 119, 128, 139, 168, 177, 192, 237, 239, 245, 294, 355, 357, 365, 376, 379 Marr, D, 29, 62, 84, 85, 86, 87, 109, 152, 156, 243, 302, 319, 330, 332, 360, 372, 373, 374 Marrocco, RT, 115 Martin, PD, 256 Martin, SJ, 29 Matsumoto, K, 147 Matsumura, N, 177, 178, 183, 185, 189, 200, 223 Matthies, H, 274 Matus-Amat, P, 5 Matzel, LD, 18 Maunsell, JHR, 108 Maurer, AP, 17, 242, 290, 366 McAllister, DE, 5 McAllister, WR, 5 McClelland, JL, 29, 33, 62, 82, 84, 85, 86, 87, 89, 93, 97, 152, 167, 360, 374 McClure, SM, 261 McCormick, DA, 302 McDonald, RJ, 50, 110, 111, 131, 261, 275, 319 McEchron, MD, 16, 45, 46, 66, 228 McGaugh, JL, 60, 115, 116, 117, 138 McGaughy, J, 28, 29, 117 McGeorge, AJ, 297 McHugh, TJ, 25, 90, 110 McIntyre, DC, 374 McLamb, RL, 87 McNamara, TP, 170 McNaughton, BL, 7, 16, 17, 23, 24, 26, 28, 29, 33, 74, 77, 79, 82, 84, 85, 86, 87, 89, 91, 93, 97, 98, 108, 109, 111, 116, 118, 119, 120, 134, 151, 152, 153, 154, 161, 168, 169, 177, 178, 182, 183, 184, 189, 192, 195, 208, 210, 213, 228, 237, 241, 243, 245, 247, 254, 258, 262, 301, 302, 317, 319, 332, 333, 343, 358, 360, 364, 365, 367, 372, 374, 376 McNaughton, N, 254 Mehta, MR, 25, 96, 112, 315, 320, 321, 322, 370 Mellet, E, 170, 171, 188 Merrill, DA, 364 Merzenich, MM, 36 Meunier, M, 219, 253, 259 Meyer-Lindenberg, A, 262
AUTHOR INDEX M’Harzi, M, 185, 318 Miller, BT, 261 Miller, EK, 223, 261, 296 Miller, JJ, 27 Milner, AD, 156 Milner, B, 9, 59, 82, 96, 193, 218, 253, 261, 353, 364, 374 Mirenowicz, J, 273 Misanin, JR, 9 Mishkin, M, 82, 96, 97, 193, 201, 218, 219, 253, 315 Mittelstaedt, H, 17, 153 Mittelstaedt, MK, 17 Mittelstaedt, M-L, 153 Miyachi, S, 296 Miyashita, Y, 177, 181, 206, 225, 228, 230 Miyoshi, E, 273 Mizumori, SJY, 4, 8, 16, 17, 18, 19, 23, 27, 28, 29, 30, 33, 34, 35, 36, 44, 45, 49, 52, 54, 60, 67, 85, 93, 98, 100, 113, 115, 117, 140, 147, 177, 238, 258, 264, 271, 273, 277, 278, 280, 281, 282, 283, 291, 314, 315, 355, 365 Moffat, SD, 354 Mogami, T, 225 Mogenson, GJ, 256, 300 Moita, MA, 19, 65, 98, 237, 239 Moita, MAP, 120, 165, 168 Monacelli, AM CJ, 354 Montague, PR, 261 Moody, SL, 229 Moore, CI, 368, 369, 370 Moores, E, 114 Moran, JP, 259 Morris, RG, 54, 59, 82, 83, 97, 185, 274, 275, 281, 354, 364, 372 Morris, RGM, 19, 28, 29, 32, 33, 82, 84, 85, 86, 87, 89, 93, 110, 111, 115, 128, 138, 162, 219, 254, 263, 281, 319, 332 Morrison, A, xvii Mort, DJ, 156 Moscovitch, M, 5, 35, 290, 374 Moser, EI, 18, 83, 109, 115, 238, 248, 354 Moser, MB, 238, 354 Moss, CF, 330 Mountcastle, VB, 109 Movshon, JA, 134 Muir, GM, 63, 258, 259, 260 Mulder, AB, 291, 301 Muller, GE, 9 Muller, RU, xix, 16, 17, 19, 24, 25, 33, 44, 61, 62, 67, 73, 74, 78, 91, 92, 93, 108, 110, 118, 128, 132, 136, 138, 139, 151, 154, 161, 163, 164, 166, 168, 177, 183, 187, 192, 195, 207, 210, 212, 254, 255, 258, 316, 317, 343, 344, 348, 365 Mumby, DG, 85 Munoz, M, 374 Murchison, CF, 115 Murray, EA, 193, 226, 229, 259 Muzzio, IA, 119 Myers, CE, 18 Myhrer, T, 85 Naber, PA, 245, 259, 260, 262 Nadasdy, Z, 166, 376
391
Nadel, L, 3, 5, 6, 7, 8, 9, 12, 16, 17, 18, 23, 29, 59, 60, 73, 74, 82, 85, 86, 87, 91, 96, 98, 127, 132, 134, 138, 150, 151, 153, 155, 161, 162, 181, 183, 185, 189, 212, 218, 219, 225, 237, 253, 254, 260, 343, 353, 374 Nader, K, 9, 277 Nakazawa, K, 87, 89, 110, 152, 360 Newman, MC, 354 Niki, H, 177, 194, 220, 222, 223, 225 Ninokura, Y, 290 Nirenberg, S, 134 Nishijo, H, 177, 178, 183, 185 Nishijo, HT, 185 Niv, Y, 296 Norman, G, 259, 260 Norris, CM, 368, 369, 370, 371 Nyberg, L, 229 O’Carroll, CM, 274, 281 Oddie, SD, 313, 322 O’Donnell, P, 256 O’Keefe, J, vii, xvii, xviii, 3, 6, 7, 8, 16, 17, 18, 21, 23, 24, 29, 33, 44, 59, 60, 61, 64, 66, 67, 73, 74, 78, 82, 84, 91, 92, 93, 96, 98, 108, 127, 132, 134, 138, 140, 141, 144, 150, 151, 153, 154, 155, 161, 164, 177, 181, 183, 185, 187, 192, 195, 210, 212, 213, 218, 219, 225, 237, 253, 254, 255, 257, 258, 260, 271, 315, 317, 320, 330, 332, 343, 353, 365, 366, 374 O’Keefe, JA, 161, 163, 164 O’Keefe, JM, 151, 153 Olds, J, xvii Oler, JA, 357, 365, 376, 379 Olson, IR, 154, 224 Olton, DS, 23, 30, 50, 82, 96, 162, 168, 169, 177, 183, 313, 314 Olypher, A, 128 Olypher, AV, 118, 140, 164, 261 O’Mara, S, 195, 196 O’Mara, SM, 16, 35, 177, 195, 196, 207, 210, 223 O’Neill, J, 376, 377, 378 Ono, T, 93, 177, 183, 185, 196, 200, 256 Oomura, Y, 259 O’Reilly, RC, 5, 18, 19, 29, 62, 82, 84, 85, 86, 87, 89, 93, 97, 167 Orr, G, 315 Otmakhova, NA, 18, 86, 114, 115, 280, 281 Otto, T, 224 Owen, AM, 272 Packard, MG, 50, 60, 116, 138, 273 Pakkenberg, B, 364 Pandya, DN, 290 Pang, KC, 318 Panzeri, S, 200 Papez, JW, 156 Papp, EC, 259 Papp, G, 340 Pare, D, 116, 119 Parker, A, 262 Parkinson, JK, 83, 193 Parron, C, 100
392
AUTHOR INDEX
Parsons, CG, 371 Passetti, F, 117 Pastalkova, E, 127 Pasupathy, A, 296 Pauls, J, 228 Paulsen, O, 18 Pavlides, C, 315, 374 Payne, JD, 18 Paz, R, 117 Paz-Villagran, V, 68 Pearce, JM, 82, 97 Peigneux, P, 301 Pelletier, JG, 116 Penick, S, 4, 6, 18 Pennartz, CM, 26, 296, 302 Peters, A, 364 Petersen, CC, 302 Petrides, M, 193, 261 Petrovich, GD, 100 Petsche, H, 314 Pham, K, 108 Phelps, EA, 6, 117 Phillips, RG, 4, 18, 83, 155, 261 Pickel, VM, 262 Pillon, B, 272 Pilzecker, A, 9 Pitkanen, A, 192, 206 Pittenger, C, 114 Plath, N, 365 Poo, MM, 17, 321, 347 Posner, MI, 113, 114, 115, 116 Postle, BR, 157, 290 Poucet, B, 16, 35, 67, 138, 143, 147, 254, 263, 290, 291 Pouget, A, 156 Poulos, AM, 18 Power, AE, 115 Pratt, WE, 35, 291 Prescott, TJ, 296 Preston, A, 167 Preston, AR, 222, 229 Pribram, KH, 223 Price, JL, 192, 206 Przybyslawski, J, 9 Pugh, CR, 5 Puryear, CB, 19, 143 Pyapali, GK, 364 Pych, JC, 282 Qin, YL, 374 Quirk, GJ, 16, 28, 66, 73, 78, 108, 119, 128, 238, 245, 255, 262, 358 Quiroga, RQ, 168, 222 Ragozzino, ME, 27, 35, 282, 291 Ramirez-Amaya, V, 337 Ramon Y Cahal, S, 84 Ranck, JB Jr, xviii, xix, 17, 61, 74, 164, 219, 225 Ranck, JR Jr, 16, 17 Ranganath, C, 165, 224
Rao, G, 177 Rapp, PR, 353, 364 Rasmussen, T, 364 Rawlins, JN, 313 Rawlins, JNP, 30, 338 Recce, M, 152 Recce, ML, 24, 315, 320, 366 Redgrave, P, 296 Redish, AD, 16, 19, 54, 75, 112, 114, 118, 119, 134, 136, 178, 181, 182, 189, 237, 241, 316, 343, 358, 360 Reep, RL, 260 Rempel-Clower, NL, 82, 86, 97 Renshaw, BA, xvii Rescorla, RA, 3 Resnick, SM, 354 Reynolds, JH, 118 Reynolds, JN, 296 Rial Verde, EM, 365 Ricci, GF, xvii Riccio, DC, 5 Riches, IP, 220, 222 Richmond, B, 134 Richmond, BJ, 225 Risold, PY, 27, 185 Rivard, B, 67, 166, 168 Robbe, D, 118, 119 Robbins, TW, 115 Robertson, RG, 93, 196, 197, 199, 200, 201, 207, 212, 213 Robinson, TE, 256 Robitsek, RJ, 45, 47 Rodrigue, KM, 353 Rogawski, MA, 371 Rogers, J, 100 Rolls, ET, 24, 29, 68, 84, 85, 86, 87, 89, 90, 91, 93, 96, 97, 98, 177, 181, 194, 195, 196, 197, 199, 200, 201, 203, 204, 205, 206, 207, 208, 209, 210, 211, 212, 213, 220, 223, 225, 243, 247, 290, 331, 332, 333, 337 Rondi-Reig, L, 87 Rosenbaum, RS, 163 Rosenzweig, ES, 112, 113, 353, 355, 360, 364, 365, 367, 368, 370, 372, 379 Ross, RN, 262 Ross, RS, 302 Ross, RT, 182 Rossier, J, 141, 142, 145 Rotenberg, A, 128 Rothbart, MK, 113, 115, 116 Roudi, Y, 329 Rudy, JW, 4, 5, 18, 19, 82, 83, 84, 85, 86, 97, 167 Rule, RR, 262 Rupniak, NM, 226 Rupniak, NMJ, 193 Russell, NA, 260 Ryan, L, 374 Sacchetti, B, 260 Sack, AT, 156
AUTHOR INDEX Sainsbury, RS, 313 Sakurai, Y, 16, 23 Salthouse, TA, 353 Samsonovich, A, 7, 84, 152, 178, 182, 189, 208, 247, 332, 343 Sanchez-Vives, MV, 302 Sara, SJ, 9 Sargolini, F, 23, 75, 77, 78, 100, 108, 237, 238, 239, 241, 297 Sarter, M, 115, 117 Sato, C, 4 Saunders, RC, 193 Save, E, 7, 16, 19, 138, 139, 143, 153 Schacter, DL, 156, 229 Schautzer, F, 188 Schendan, HE, 34 Schmitzer-Torbert, N, 54 Schnell, E, 282, 331 Schooler, JW, 264 Schu¨z, A, 330 Schultz, S, 200, 213 Schultz, W, 114, 115, 256, 273, 296 Scoville, WB, 9, 59, 82, 96, 218, 253, 364, 374 Seager, MA, 313, 314 Seamans, JK, 35, 55 Segal, M, 27 Seidenbecher, T, 313 Sejnowski, TJ, 85, 156 Selden, NRW, 4 Sereno, MI, 109 Sesack, SR, 262 Shadlen, MN, 134, 261 Shallice, T, 261 Shapiro, ML, 8, 17, 19, 21, 24, 25, 27, 29, 33, 44, 45, 49, 50, 52, 62, 64, 67, 83, 84, 91, 92, 93, 97, 98, 111, 144, 166, 169, 255, 258, 343 Shapley, R, 332 Sharp, PE, 16, 17, 63, 74, 75, 77, 78, 138, 151, 243, 316 Shelton, AL, 171 Shen, B, 376 Shen, J, 355, 365, 366, 367, 369, 370, 374, 376, 378 Shepard, RN, 183 Shepherd, JD, 365 Sherry, DF, 254 Shibata, R, 256, 294, 297, 298, 300 Shimamura, AP, 262 Shinoe, T, 115 Shors, TJ, 18 Shrager, Y, 154 Siapas, AG, 26, 261, 302, 322 Sigala, N, 228 Silva, AJ, 5 Simons, DJ, 154 Singer, W, 134 Sirigu, A, 181 Sirota, A, 302, 304, 376 Skaggs, WE, 25, 28, 62, 76, 132–133, 136, 169, 184, 189, 209, 258, 301, 315, 317, 320, 321, 322, 332, 366, 376 Small, SA, 83, 165, 365, 367, 369, 373
393
Smith, AC, 376 Smith, CD, 353 Smith, DM, 4, 8, 16, 18, 19, 23, 33, 44, 45, 49, 52, 67, 98, 140, 177, 258, 264, 277, 278, 280 Smith, EE, 156 Smith, ML, 193 Snyder, LH, 156 Sohal, VS, 318, 319 Solomon, PR, 4, 6, 18 Solstad, T, 238, 241, 242, 255 Sompolinsky, H, 332 Song, EY, 24 Song, PC, 76 Speakman, A, 21, 61, 64, 73, 92, 140, 141, 144, 151, 187, 255, 257, 258, 353 Spelke, ES, 170 Sperling, R, 165 Spiers, HJ, 82, 97, 254 Squire, LR, 82, 97, 163, 219, 223, 253, 364, 374 Stackman, RW, 187 Stanis, J, 376 Staubli, U, 225 Stea, D, 60 Stead, M, 258, 316 Stefanacci, L, 192, 206 Steinmetz, PN, 118 Steriade, M, 302 Steward, O, 319, 321 Stewart, M, 313, 314 Stringer, SM, 205, 209, 211, 213 Stumwasser, F, xvii Stuss, DT, 156 Summerfield, C, 107 Suri, RE, 273 Sutherland, GR, 374 Sutherland, RJ, 82, 84, 97 Sutton, RS, 316 Suzuki, WA, 82, 97, 177, 192, 206, 219, 225, 248, 254 Swanson, LW, 27, 185, 290, 374 Taber, MT, 262 Tabuchi, E, 206, 254, 256, 257, 258, 263, 290 Tabuchi, ET, 144, 297, 299, 300 Tanaka, K, 225 Tanila, H, 16, 28, 62, 92, 115, 116, 169, 189, 264, 355, 357, 365, 372 Taube, JS, xix, 17, 61, 63, 68, 74, 75, 78, 147, 151, 156, 207, 238, 245 Teather, LA, 275 Teng, E, 163, 364, 374 Terrazas, A, 358, 366 Teyler, TJ, 152, 374 Thibault, O, 371 Thierry, AM, 296 Thompson, LT, 74, 343, 371 Thompson, RF, 74, 313, 314 Thorndyke, PW, 181, 185 Tierney, PL, 291 Tilson, HA, 87
394
AUTHOR INDEX
Timofeev, I, 302 Toescu, EC, 371 Tolman, EC, 60, 83, 138 Tombaugh, GC, 370 Tomie, J, 162 Touretzky, DS, 16, 77, 118, 119, 154, 178, 181, 182, 189, 241, 247, 316, 343, 358, 360 Tracy, AL, 258 Traub, RD, 117 Treves, A, 29, 30, 84, 85, 86, 87, 93, 97, 195, 200, 201, 204, 205, 206, 208, 210, 212, 213, 243, 247, 248, 329, 330, 331, 332, 333, 335, 336, 337, 338, 339, 340 Trinkler, I, 155 Trivedi, MA, 256 Trullier, O, 143, 294, 297, 298 Tsodyks, MV, 320, 321 Tsubokawa, H, 262 Tulving, E, 12, 18, 82, 97, 155, 164 Turner, DA, 364 Ulanovsky, N, 330 Ulinski, PS, 328 Uncapher, MR, 114 Ungerleider, LG, 107 Uylings, HB, 262, 296 Vanderwolf, C, 17 Vanderwolf, CH, xviii, 84, 263, 313, 365, 365–366, 374 Van der Zee, EA, 313 Van Elzakker, M, 167 Van Essen, DC, 253, 374 van Groen T, 237, 238 van Haeften, T, 238, 241 Van Hoesen, GW, 192, 193, 248 Vann, SD, 275 Vargha-Khadem, F, 82, 83, 97 Vazdarjanova, A, 12, 84, 85, 95, 96, 360, 372 Vertes, RP, 26, 291, 313 Verwer, RW, 290 Victor, M, 218 Vinogradova, OS, 18, 85, 111, 118, 281 von Bechterew, WV, 218 von der Malsburg, C, 134 Wagner, AD, 229 Wagner, AG, 165 Wagner, AR, 3 Wall, P, xvii–xviii Wallace, DG, 260 Wallenstein, GV, 84, 93, 155, 319, 320 Wallentin, M, 156 Walling, SG, 116 Walsh, TJ, 87, 314 Wan, RQ, 290 Wang, D, 256 Wang, RF, 154, 170 Wang, XJ, 76, 89, 290 Watanabe, T, 177, 194, 220, 222, 223, 225 Waterhouse, BD, 115, 116
Weible, AP, 115 Weinberger, NM, 117 Weiner, I, 296 Weiss, C, 74 Wenk, GL, 371 Wesierska, M, 129 West, MJ, 364 Westmacott, R, 34 Wheeler, MA, 155 Whishaw, IQ, 17, 162, 313 White, AM, xvii, xviii White, NM, 50, 131, 273, 275, 319 Whittington, MA, 117 Wible, CG, 19 Wiebe, S, 225 Wiener, SI, 16, 17, 23, 30, 64, 65, 74, 91, 93, 98, 168, 169, 183, 294, 297, 317, 343 Wiesel, T, 108 Wiig, KA, 259 Wilkniss, SM, 353 Williams, JD, 17 Williams, S, 274 Willner, J, 3, 5, 6, 189, 260 Wills, TJ, 7, 66, 154, 155, 237, 239, 245, 340, 360 Willshaw, D, 332 Wilner, J, 18 Wilson, CJ, 302 Wilson, EO, 328 Wilson, FAW, 220, 224, 225 Wilson, I, 365, 373 Wilson, IA, 354, 355, 356, 357, 358, 359, 360, 364, 373 Wilson, M, 118 Wilson, MA, 24, 26, 33, 55, 56, 93, 115, 134, 151, 166, 183, 210, 228, 254, 261, 262, 290, 301, 302, 316, 322, 333, 364, 365, 374, 376, 377, 378 Wilson, TD, 264 Wiltgen, BJ, 5 Winocur, G, 5, 182, 290, 374 Winson, J, 27, 313, 314, 374 Winter, JC, 376, 379 Winters, BD, 259 Wirth, S, 28, 165, 207, 226, 228, 230 Wise, RA, 281 Wise, SP, 226, 229 Witte, EA, 117 Witter, M, 374 Witter, MP, 30, 34, 55, 84, 87, 109, 237, 238, 241, 248, 253, 259, 260, 262, 290, 291 Wittmann, BC, 114, 115 Wood, E, 166, 169, 170 Wood, ER, 17, 18, 19, 21, 44, 45, 46, 50, 52, 55, 65, 67, 74, 98, 111, 143, 164, 165, 168, 225, 237, 239, 245, 317, 320 Worden, R, 67 Worley, P, 275 Wouterlood, FG, 253, 259 Wu, M, 314 Wu, SM, 344
AUTHOR INDEX
395
Wyble, BP, 282, 317, 320, 323 Wyss JM, 237, 238
Yu, X, 96 Yuste, R, 241
Xavier, GF, 87 Xiang, J-Z, 98, 201, 203, 206, 212, 213, 225 Xu, L, 109
Zeineh, MM, 83, 165 Zgaljardic, DJ, 272 Zhang, K, 209 Zhou, TL, 27 Zilli, E, 315–316 Zilli, EA, 323 Zinyuk, L, 61, 62, 112, 139, 140, 141, 264, 361 Zinyuk, LE, 68 Zipser, D, 156 Zironi, I, 262 Zola, SM, 219 Zola-Morgan, S, 82, 86, 97, 218, 219, 223, 225, 258 Zola-Morgan, SM, 374 Zugaro, MB, 25, 315
Yan, J, 110 Yanike, M, 165, 228, 230 Yeshenko, O, 19, 23, 24, 34, 54, 271 Yeterian, EH, 290 Ylinen, A, 374, 376 Yoganarasimha, D, 76 Yoshimura, Y, 109 Young, BJ, 16, 62, 91, 164, 183 Young, FW, 183 Young, WS, 259
This page intentionally left blank
Subject Index
Acceleration, location-sensitive firing and, 17 Acetylcholine (ACh) in mediation of orienting, 113, 117 in neuromodulation of memory storage vs. retrieval, 331 in novelty detection mechanism, 28, 29 in regulation of spatial behaviors and processing, 282–283 theta rhythms and, 314, 315 Adrenergic system, and place field stability, 115–116 Affective state, episodic memory and, 204–206 Age-associated alterations in neural circuits dynamics of place cells and, 367–373 and memory impairment, 373–378 overview, 364–365, 378–379 properties of place cells and, 365–367 Age-related memory impairments CA3 subfield and rigidity, 360 future research directions, 360–361 human aging and, 353–354 rigid place cell encoding and, 355–360 water maze testing, 354–355 Aging, normal changes in, 364 Alerting, 113 Allocentric representations. See also Cognitive maps in amnesia patients, 163 vs. egocentric representations, 150–151, 152, 156–157 hippocampus and, 161–162, 177, 189 medial temporal lobe and, 177 place cells in support of, 163–164 in primate hippocampus, 195 and primate spatial view cells, 192 Alzheimer disease memory impairments entorhinal cortex and, 248–249 normal aging compared to, 354 Ambiguous conditions. See Disambiguation Ammon’s horn, 328, 330
Amnesia animal models of, 83, 87–90 differential roles of subfields in, 86–90 and episodic memory impairment, 82, 83 and formation of coherent mental images, 152–153 relational memory loss in, 224 spatial information processing and, 271 spatial memory and, 163 AMPA, and negative feedback to NMDA receptors, 344, 347–348 AMPA receptors long-term potentiation, and age-related changes, 368 place field formation and, 110 Amygdala attention and, 117 in emotionally charged memories, 116–117 entorhinal input from, 192 hippocampal output to, in field density model, 256 lesions of, and memory deficits, 218–219 Animal models of aging, 353, 354 of cognitive aging, 354–355 of hippocampal amnesia, 83, 87–90 for study of medial temporal lobe and memory, 218, 229–230 Anisomycin, 277 Arc gene expression age-related changes and, 365 dopamine and, 275–276 environmental exploration and, 96 Archicortex theory, 85 Associations hippocampal encoding of, 165–166 of object and spatial representations, hippocampus in, 192, 201
397
398 SUBJECT INDEX Associations (continued) between visual stimuli and spatial responses, 206–207 well-learned, monkey hippocampal activity and, 228–229, 230 Associative learning, monkey hippocampus activity and, 226–228 Associative memories, updating of, 28–29 Associative memory networks, place cell remapping and, 243–245 Attention alteration of hippocampal representations by, 118–119, 120–121 amygdala and, 117 and context definition, 8 in long-term memory, 107–108, 114, 120 neuromodulators of, 114–118 selective, vs. general arousal in memory consolidation, 119–120 Attentional modulation, of place cell activity, 111–114, 120 Attentional networks, human, 113–114 Attractor neural networks autoassociative, 208, 213, 332–333 continuous, 207–208 and path integration, 208–210, 213 and episodic memory, 207–208, 213 hippocampus as, 360 Mexican hat–type, 247–248 Auditory stimuli responses, during memory tasks with short delay intervals, 222 Autoassociative attractor neural networks, 208, 213, 332–333 Autobiographical experiences, hippocampal activity during retrieval of, 165 Avian hippocampus, 330–331 Background context conditioning, fear-conditioning paradigm and, 4–5 Basal ganglia, 296–297, 319 Behavior. See also Random foraging; Spatial behavior correlation with theta rhythms, 313–315 goal-directed navigational strategies for, 138 place cell firing during, 139–143, 147 place cell function in, 254 prefrontal cortex and, 115, 261–263, 290, 291, 294 striatum and, 299–300 hippocampal and dorsal striatal–dependent, dopamine and, 271, 272–273 memory-guided, computational models of, 315–323 planned, prefrontal cortex and, 289–290, 294 Behavioral conditions context discrimination in, 30–34, 35 neuronal firing and, 16 Behavioral context hippocampal activity in vs. hippocampal dependence of, 44–46. See also Context-dependent hippocampal activity hippocampal encoding of, 164 place cell firing variations and, 139–140 and place field modulation, 111–113
Behavioral effects on spatial context processing, hippocampal subfield lesions and, 86–90 Behavioral responses context-dependent place cell activity and, 44, 50–54 learned, and place fields, 24 Birds, hippocampus of, 330–331 Boundaries, place fields and, 67–68, 69 Boundary vector cells (BVCs), 151 Boundary vector cells (BVC) model, 151, 152, 157 Brain regions. See also names of specific regions, e.g., Amygdala functional specialization of, structure-specific dopamine signaling and, 271, 272–273 mammalian, evolution of, 328–329 receiving hippocampal place-related signals, 289 BVC (boundary vector cells) model, 151, 152, 157 BVCs (boundary vector cells), 151 CA1 subfield age-related changes in, 365–367, 368–373, 374–376, 378–379 in amnesia, 86–87 vs. CA3 subfield, in age-related memory impairments, 360 as CA3 subfield postprocessor, 338–339 cholinergic modulation of activity of, 282–283 context-dependent activity in, 45, 47, 49–50, 53, 54–56 dopaminergic modulation of activity of, 274 interconnections of, 84, 241 in mechanisms of context comparison, 29–30 in pattern completion, 93–95 place field formation time frame in, 109 potential functions of, 280–281 in prediction, 338, 339 in primates, with obscured vision, 199–200 in response to contextual change, 12n6 selective dopamine alteration of inputs to, 280–281 simultaneous monitoring of, with CA3 subfield, 93–95 CA3–CA1 Schaffer collateral synapses, age-related changes in, 368–371 CA3 subfield age-related changes in, 360, 365, 372–373 in amnesia, 86–87 association of spatial and nonspatial inputs in, 248 in autoassociative memory, 204–205 computational models of, 85–86, 319 differentiation of, from CA1 subfield, 338–339, 340 global remapping and, 243–244 interconnections of, 84 in location-specific retrieval of memories, 152 in mechanisms of context comparison, 29–30, 62 in novel spatial context representation, 87–89, 96 in pattern completion, 29, 62, 85–86, 89, 93–95, 96, 152, 360 in pattern separation, 85–86, 95, 96 place field formation time frame in, 109 preprocessor of dentate gyrus as, 336–338 and memory storage vs. retrieval, 331–332 in primates, with obscured vision, 199–200
SUBJECT INDEX rate remapping in, 56 recurrent collaterals of, in network analysis, 332–333 in response to contextual change, 12n6 simultaneous monitoring of, with CA1 subfield, 93–95 spatial map of, Euclidian graph model, 343–348 Calcium, intracellular regulation of, and aging, 371 Category recognition, hippocampus and, 168, 222 CDH. See Context discrimination hypothesis Cell assembly coding, in segregation of representations, 134 Cerebral association cortex, hippocampal input from, in primates, 192 c-fos gene, 275–276 Chlorophenylpiperazine (CPP), 110 Cholinergic activity. See Acetylcholine Cholinergic septohippocampal neurons, 314 Coding population, in segregation of representations, 134 prospective behavioral context and, 111 cognitive demand and, 17 in context comparisons, 29, 30, 33 place cell firing and, 144 sequential representation of episodes and, 166 task-specific events and, 25–26 retrospective cognitive demand and, 17 in context comparisons, 29, 30, 33 sequential representation of episodes and, 166 task-specific events and, 25–26 Cognitive abilities, dopamine depletion and, 272–273 Cognitive demand, retrospective/prospective coding and, 17 Cognitive maps alternative locations of, 63–64 configurational context as, 6–7 definitions of, 60 discordant representations and, 61–63 Euclidian graph model of, 343–348 hippocampus and, 60–64, 161 mosaic map system for, 67–68, 69 place cells as, 60–61, 73–74 previously acquired, in amnesia patients, 163 spatial behaviors and, 138 spatial relationships between environmental cues in primates and, 182–185 task solution without use of, 63 vs. temporal relationship as organizing principle, 23 theory of, vs. relational memory theory, 169–171 theory proposed, 59, 60, 219 updating of, 7–8 Cognitive processes, and reshaping of sensory outputs, 119 Cognitive strategy, place fields stability and, dopamine in, 278–280 Cognitivism, 59 Computational models of dopamine’s role in place cell activity, 272, 280–281
399
of hippocampal subfields, in spatial context processing, 84–86, 93, 96 of hippocampus in memory, 330–335, 340–341 linking memory-guided behavior to hippocampal physiology, 315–323 spin glass approach, 330, 340 of storage of distance between place cell firing fields, 343–348 Computer-based research systems, development of, xix Conditional left–right discrimination learning, fornix lesions and impairment of, 193–194 Conditional spatial response learning, hippocampus in, 206 Conditional stimuli, context differentiated from, 4 Configural context, 5–8 Conjunctive cells description of, 238 in updating of spatial representation with movement, 241 Context alterations of determination of, 18, 27–28 dopamine in detection of, 278–279 behavioral hippocampal activity in vs. hippocampal dependence of, 44–46. See also Context-dependent hippocampal activity hippocampal encoding of, 164 place cell firing variations and, 139–140 and place field modulation, 111–113 configural, 5–8 definition of, 8–9, 154–155 discrimination between, 18, 30–34. See also Context discrimination hypothesis elemental, 5–7 environmental grid cells and, 77 head direction cells and, 75–76 hippocampal place cells and, 74, 78, 79–80 of subicular place cells, 77–78 hippocampal role in, 3–4, 18 literature on hippocampus in, 4–6 and memory recall, 204–206 multiple representations of, 4–5 novel. See Novel spatial context representation in past and future representations, 12 recognition of, 12n5 spatial. See Spatial context updating of, 7–8. See also Remapping Context comparisons. See also Mismatch detection hierarchial model and, 30–34 mechanisms of, 29–30 Context conditioning, fear-conditioning paradigm and, 4–5 Context-dependent hippocampal activity behavioral responses and, 44, 50–54 in episodic memory formation, 79–80 and hippocampus-dependent tasks, 46–50, 52–53 and hippocampus-independent tasks, 44–46, 50–51 memory tasks and, 44, 49–50 nature of, 54–56 Context-dependent retrieval. See Contextual retrieval
400 SUBJECT INDEX Context discrimination hypothesis (CDH). See also Context processing behavioral and learning situations and, 30–34 dopamine in orthogonalization of spatial representations, 284 functional networks and, 18–19 level 1 organization, 21–27 level 2 organization, 21, 27–28 level 3 organization, 21, 28–29 memory inputs and, 16 summary and implications of, 34–36 Context fear, requirements for development of, 5 Context invariant properties of limbic system cells, 74–80 Context processing. See also Context discrimination hypothesis behavioral and learning situations and, 30–34 hippocampal lesions and, 260–261 memory and, 16, 19, 21–23 perirhinal cortex in, 260–261 population efferent code in, 28–29, 33 spatial. See Spatial context processing spatial information and, 19, 23–24 temporal organization and, 19, 24–26 visual, by primates, 24, 93 Context recognition, reconsolidation and, 6, 9–11 Context-sensitive vs. persistent place cells, 27–28 Context specificity hippocampal damage and, 3–4 of hippocampal place cells, 74, 78, 79–80 Contextual fear-conditioning paradigm, 83, 89 Contextual retrieval computational model of, 318–321 definition of, 82 hippocampus in, 82–85 Continuous alteration task, 44–46, 47–48, 50, 54–55 Continuous attractor neural networks, 207–208 and path integration, 208–210, 213 Continuous attractors decoding position on, 333–335 examples of, 332 CPP (chlorophenylpiperazine), 110 Cues extramaze distal, and navigation, 185–189 idiothetic, 98, 177, 210 internal sensory, 17, 359 local vs. distal, manipulation of, 92–93 removal and manipulation of, and place cell activity, 92–93 spatial context learning differentiated from, 3 spatial relationships between, in primates, 182–185 DA. See Dopamine Dead reckoning. See Path integration Decision making prefrontal cortex and, 289–290, 291–294 striatum and, 296–297 Declarative memory aging and, 353 medial temporal lobe and, 218–219 Delayed spatial-response task, 194
Delay interval of memory tasks, hippocampus activity during, 223–224 Dentate gyrus in amnesia, 86–87 as CA3 subfield preprocessor, 336–338 computational models of pattern completion/pattern separation in, 85–86 evolution of, 328 granule cells of, age-related changes in, 365, 367, 368, 373 interconnections of, 84 in pattern completion, 29, 30 in pattern separation, 29, 89–90, 360 Directional bias, as property of locational cells, 78 Directional generation and representation, head direction cells and, 17 Directionality of place cells, alternative explanation of, 168–169 Disambiguation contextual control and, 4 hippocampal cells in, 169 perirhinal cortex and, 259 Discrete attractor neural networks. See Autoassociative attractor neural networks Discrete memory states model, 332–333 Discrete stimuli, place cell response to, 65 Distal cues vs. local, 92–93 navigation and, 185–189 Distance between place cell firing fields, encoding as synaptic resistance, 343, 346–348 Dopamine (DA) computational models of role of, 272, 280–281 in hippocampal and dorsal striatal–dependent behaviors, 271, 272–273 mechanisms selectively affecting learning, 271–272, 284 in mediation of executive attention, 113 in novelty detection mechanism, 28, 29, 111 and place field stability, 113, 277–280, 284 in plasticity and memory, 114–115 reward prediction and, 296 selective alteration of CA1 subfield inputs by, 280–281 spatial selectivity of single-unit activity, 271, 276–280 synaptic plasticity mechanisms and, 114–115, 271, 273–276 Dopamine receptors, 273, 274, 280, 281, 287 Dopaminergic system–glutaminergic system interactions, 273, 274, 277, 284 Dorsal striatal–dependent behaviors, dopamine and, 271, 272–273 EC. See Entorhinal cortex Egocentric representations vs. allocentric representations, 150–151, 152, 156–157 in amnesia patients, 163 in image formation, 152–153 Elemental context, 5–7 Emotionally charged memories, amygdala and, 116–117 Emotional state, and memory recall, 204–206
SUBJECT INDEX Entorhinal cortex (EC). See also Grid cells Alzheimer disease memory impairments and, 248–249 anatomy of, 238 context conditioning and, 12n3 in formation of representations for episodic memory, 210–212 in generation of context-dependent activity, 55–56 hippocampal input pathways of, 84, 86 hippocampal inputs from and episodic memory formation, 79–80 in primates, 192 spatial and nonspatial, 119–120, 248–249 instantaneous response to novelty by, 109 lateral innervation of, 237 and nonspatial inputs to hippocampus, 248 rate remapping and, 245 medial coherent network dynamics in, 243–245 grid cell–place cell networks, 241–243 grid cell network models, 245–248 innervation and anatomy of, 237–238, 241–242 path integration–based spatial map and, 97 spatial representation in, 237–238, 329–330 as a universal spatial map, 238–241 spatial representation by, 108–109 stimulus-selective match–nonmatch signals and, 225 Environmental changes navigational strategies and, 138 pattern separation and, 62 and place cell encoding in aged individuals, 355–360 Environmental context grid cells and, 77 head direction cells and, 75–76 hippocampal place cells and, 74, 78, 79–80 subicular place cells and, 77–78 Environmental exploration, immediate early gene expression and, 96 Environmental features, context definition and, 8 Episodes reconstruction of, location-specific mental images and, 150, 155–156, 157 sequential representation of, 166–167, 169, 170, 171 memory space hypothesis and, 219 Episodic memory. See also Memory affective state associations and, 204–206 attention and, 107–108 attractor neural networks and, 207–208 context as critical component of, 11 entorhinal cortex and hippocampus in representation formation for, 210–212 extrahippocampal modulation of, 98–100 grid cell input in formation of, 79–80, 210–212 hippocampal–prefrontal network and, 290 hippocampal lesions in humans and, 59, 82, 83 hippocampal subfield functions in, 83–85, 96–98 hippocampal system and, 3, 79–80, 82, 152–153, 154 hippocampus in retrieval of, field density model and, 264 impairment of, 82, 83
401
location-specific mental images and, 150, 152–153, 154 network model of, 332, 338 object–place associations and, 201 in past and future representations, 12 place cell firing and, 138–139, 143–144, 147 place cell maps and, 68–69 place cell nonspatial information associations and, 237, 248 recall of, in primates, 201–204. See also Spatial view cells reconstruction of episodes and, 150, 155–156, 157 relational memory hypothesis and, 164–171 spatial context and, 83, 177, 189 spatial view cells in, 212–213 Error-driven computations. See Match–mismatch computations Euclidian graph model of cognitive maps, 343–348 Evolutionary history of mammals, 328–329 Executive attention, dopamine in mediation of, 113 Executive selection, of hippocampal neuron firing patterns, 121 Exploration in creation of internal models of external world, 8 development of place field stability during, dopamine and, 277 in novel context exposure, 33 and spatial organization of hippocampal representations, 23–24 Extended stimuli, place cell response to, 65–67 Extramaze distal cues. See Distal cues Fear conditioning foreground and background context conditioning and, 4–5 memory consolidation and, 5–6 FET (field effect transistors), xvii Field density model changes in field density and, 256–257 description of, 254–256 goal-directed behavior and, 261–263 memory and, 263–264 output regions and, 256 remapping and, 257–258 Field effect transistors (FET), xvii Field expansion plasticity, 368–371, 376–377, 378 Focal stimuli, place cell response to, 65 Foreground context conditioning, fear-conditioning paradigm and, 4–5 Fornix lesions impairments produced by, 193–194, 318, 322 theta rhythms and, 313 Functional grouping, in avoidance of superposition catastrophe, 134, 136 Functional networks, context discrimination hypothesis and, 18–19 Future, imagining of, episodic memory in, 12 Future, representation of, context and, 12 GABAergic septohippocampal neurons, 314 GABAergic system, in novelty detection mechanism, 28 Gain-field modulation, in goal encoding, 64
402 SUBJECT INDEX General arousal, in memory consolidation, 119 Genetic studies of hippocampal plasticity, 110 immediate early gene expression age-related changes in, 365 dopamine and, 275–276 and environmental exploration, 96 of receptor expression in hippocampal subfields, 87, 89 Genomic action potential, 274–275 Geometric navigation, 239–241 Geometry cells, 255 Global remapping alternate models of, 245 changes in context and, 7, 8 grid cells and, 243–245 Glutaminergic system–dopaminergic system interactions, 273, 274, 277, 284 Goal-directed behavior navigational strategies for, 138 place cell firing during, 139–143, 147 place cell function in, 254 prefrontal cortex and, 115, 261–263, 290, 291, 294 striatum and, 299–300 Goal-related place cell firing, 145, 146–147 Goals, place cell encoding of, 64, 144–145, 146–147 Granulation, in evolution of mammalian brain, 328, 330–335 Granule cells, age-related changes in, 365, 367, 368, 373 Grid cell–place cell networks, 241–243 Grid cell network, models of, 245–248 Grid cells context invariant properties of, 77 description of, 238 in formation of representations for episodic memory, 210–212 idiothetic cues and, 210 instantaneous response to novelty by, 109 nontopographic spatial encoding by, 329–330 in path integration, 79 spatial behavior and, 147 in spatial frame formation, 12n4 in updating of spatial representation with movement, 241 voluntary movement and, 23 Grid fields description of, 239–240 in processing of memory and temporal information, 23 Head direction cells context invariant properties of, 75–76 description of, 238 directional generation and representation and, 17 discovery of, xix, 61 functional possibilities of, 213 limbic system locations of, 74–75 mosaic maps, and orientation of, 68 navigation and, 61, 63 and Papez’s circuit, 156 path integration and, 76–77 in primates vs. rodents, 207 spatial behavior and, 147 spatial view cells compared to, 201
Hippocampal–prefrontal network anatomy and physiology of, 290–291 in working and episodic memory, 290 Hippocampal interconnections, 84 in primates, 192–193 Hippocampal subfields. See also names of specific subfields, e.g., CA1 subfield differentiation of, in mammals, 330–335 functional-differentiation hypothesis of, 330–331, 339–341 input and output pathways of, 84 in spatial context processing, 82–100, 360 computational models of, 84–86, 93, 96 conclusions, 96–100 overview, 82–84 perturbation studies and, 86–90 physiological studies and, 91–96 subfield differential roles in, 93–96, 98 Historical perspective, xvii–xix H.M. case, 9, 82, 163, 218, 253 Homer1a gene expression, and environmental exploration, 96 Human aging. See also Age-associated alterations in neural circuits; Age-related memory impairments normal changes in, 364 and performance on virtual water maze task, 354 spatial memory impairment in, 353–354 Hypothalamus, and motivational state, 27 Idiothetic cues grid cells and, 210 place cell modulation by, 98, 177 IEG. See Immediate early genes Images. See Mental images Immediate early genes (IEG) expression age-related changes in, 365 dopamine and, 275–276 and environmental exploration, 96 Immobility-related theta rhythms, 313 Index code, 152, 374 Internal cognitive processes, and reshaping of sensory outputs, 119 Internal sensory cues processing of, in aged individuals, 359 visual information association with, 17 Internal variables, place cell response to, 66–67, 98–100 Intracellular mechanisms of synaptic plasticity, 274–276 Lateral entorhinal cortex (LEC). See Entorhinal cortex, lateral Lateral septum. See Septum, lateral LC-NE system. See Locus coeruleus–norepinephrine system Learning. See also Memory acetylcholine and, 282–283 associative, monkey hippocampus activity and, 226–228 conditional spatial response, hippocampus in, 206 context discrimination hypothesis and, 16, 18–19, 30–36 dopamine and, 271–273, 281–284 fornix lesions and impairment of, 193–194 hippocampal activity and, 166–167
SUBJECT INDEX hippocampal neuron firing patterns and, 108–109, 118–119 immediate early gene expression and, 275–276 in novel contexts, 33–34 reinforcement, as model for place field modulation, 114 segregation of relevant vs. irrelevant information in, 129 theta rhythms and, 313–315 LEC (lateral entorhinal cortex). See Entorhinal cortex, lateral Limbic system in path integration, 78–79 spatial cells in, 74–78 Location-specific mental images brain regions involved in, 156–157 episodic memory and, 150, 152–153, 154 overview, 150–151 path-integrative influences on, 150, 153–154 perceptual influence on place cell firing, 150, 151 retrieval of, 150, 152 Locomotion and place cell firing, 196 and primate spatial view cells, 200 Locus coeruleus–norepinephrine (LC-NE) system, 116 Long-term depression (LTD) age-related changes and, 368–371 at corticostriatal synapses, 296 dopamine and, 274 Long-term memory attention in formation of, 113–114, 120 location-specific mental images in retrieval from, 152–153, 156 monkey hippocampal activity and, 228–229, 230 research on age-related changes in, 360–361 Long-term potentiation (LTP) age-related changes and, 368–371 at corticostriatal synapses, 296 dopamine and, 114, 274, 277 spatial representations and, 109 theta rhythms and, 315 LTD. See Long-term depression LTP. See Long-term potentiation Mammals differentiation of hippocampus in, 330–335 evolutionary history of, 328–329 Match–mismatch computations. See also Mismatch detection; Novelty detection as fundamental operating principle, 36 in memory retrieval, 85 Match–nonmatch signals, in memory tasks with short delay intervals, 224–225 MDS (multidimensional scaling), 183–185 Medial entorhinal cortex (MEC). See Entorhinal cortex, medial Medial septum. See Septum, medial Medial temporal lobe. See Temporal lobe, medial Memantine, 370–371 Memory. See also Learning; Relational memory hypothesis context-dependent place cell activity and, 44, 49–50 context-dependent retrieval of, 82–85, 318–321
403
context discrimination and, 16, 34–36 cortical lamination development and, 329 dopamine loss and, 272–273 dynamic, and motion-related spatial updating, 154 emotionally charged, amygdala and, 116–117 encoding and retrieval of, theta rhythms and, 315, 317–321 episodic. See Episodic memory field density model and, 263–264 hippocampal–cortical interactions and, 300–304 and hippocampal context processing, 19, 21–23 hippocampal function in early experimental studies on, 218–219 evidence for, 34, 64–67, 68–69 memory space hypothesis, 219–220, 229–230 primate studies, 192–214, 218–230 relational memory hypothesis, 164–171 spatial processing and, 161–164, 171 immediate early gene expression and, 275–276 location-specific retrieval of, 150, 152–153, 156 long-term attention in formation of, 113–114, 120 location-specific mental images in retrieval from, 152–153, 156 monkey hippocampal activity and, 228–229, 230 research on age-related changes in, 360–361 procedural, aging and, 353, 354 recognition, 218, 220–221, 224–225 retrieval of. See Memory retrieval segregation of multiple relevant memories coordinated use of two spatial representations and, 129–131, 136 hippocampal involvement in, 128–129, 135–136 place cell discharge and, 131–135 population coding and, 134 spatial knowledge system and, 127–128 short-term. See Working memory spatial representations of places viewed and. See Spatial view cells storage vs. retrieval, network analysis of, 331–332 theta rhythms and, 313–315 working. See Working memory Memory consolidation amygdala in, 116–117 malleability of, 9–11 memory transformation as term for, 5–6, 9, 12n1 neuromodulators of, 114–118 selective attention vs. general arousal in, 119–120 sleep and, 301–304, 374–378 Memory-guided behavior, computational models of, 315–323 Memory impairments, age-related. See Age-associated alterations in neural circuits; Age-related memory impairments Memory networks. See also Neural networks associative, place cell remapping and, 243–245 hippocampus and, 167–168 relational memory and, 164–165, 171 updating of, 28–29 Memory reconsolidation, 6, 9–11, 12n1
404 SUBJECT INDEX Memory retrieval computational model of, 318–321 context-dependent, 82–85, 318–321 field density model and, 264 location-specific, 150, 152–153, 156 match–mismatch computations in, 85 in primates, 201–204 vs. storage, network analysis of, 331–332 theta rhythms and, 315, 317–321 Memory space hypothesis, 219–220, 222, 226, 228, 229–230 Memory tasks with short delay intervals, hippocampal responses during delay period activity, 223–224 in memory phase, 224–225 motor-related activity, 225 reward-related signals, 225–226 to spatial stimuli, 222–223 to visual and auditory stimuli, 222 Memory transformation, 5–6, 9–11, 12n1. See also Memory consolidation Memory updating, 9–11, 12n1 Mental images allocentric vs. egocentric representations in, 156–157 formation of, 152–153 location-specific brain regions involved in, 156–157 episodic memory and, 150, 152–153, 154 overview, 150–151 path-integrative influences on, 150, 153–154 perceptual influence on place cell firing, 150, 151 retrieval of, 150, 152 retrieval of, 155–156, 157 1-Methyl-4-phenyl-1,2,3,6-tetrahydropyridine (MPTP), 272 Metric features, in orientation and navigation, 8 Mexican hat–type networks, 247–248 Mismatch detection. See also Match–mismatch computations; Novelty detection dentate gyrus–CA3 circuit in, 29 hierarchial model and, 30–31 in novel contexts, 33 in regulation of ongoing behaviors, 28–29 Mnemonic selection, 107–108 Monkeys. See Primates Morris water maze task, 354–355 Mosaic maps, place cells as, 67–68, 69 Motion, whole-body, hippocampal response to, 207, 213 Motion-related spatial updating. See also Locomotion in aged individuals, 358–360 medial entorhinal cortex and, 239–241 motor-efference signals and, 154, 157 perirhinal cortex and, 260 Motivational state as internal contextual cue, 82 modulation by, 27 remapping and, 258–259 Motor-efference signals, and motion-related spatial updating, 154, 157 Motor programs, stereotyped, vs. context-dependent activity, 50
Motor-related activity, in memory tasks with short delay intervals, 225 Movement, translational, 17 Movement-related theta rhythms, 313 MPTP (1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine), 272 Multidimensional scaling (MDS), 183–185 Navigation. See also Path integration in amnesia patients, 163 association of spatial and nonspatial inputs in, 248 extramaze distal cues and, primate neuronal responses to, 185–189 geometric, 239–241 goal-directed. See Goal-directed behavior head direction cells and, 61, 63 hippocampal neuron firing and, 168 large-scale environmental maps and, 170–171 metric features in, 8 mosaic maps and, 68, 69 nonvisual information and, 17 place cell activity and, 61, 63 place-differential neural responses during, 178–182 vs. random foraging, place cell firing during, 139–140 response-based, 60 spatial organization and, 23, 24, 34 strategies used for, 138 virtual, 168, 171, 185–189 voluntary. See Exploration NE (norepinephrine), 113, 116 Neocortex evolution of, 328–329 hippocampus in linking of regions of, 374 influence on hippocampal place cells, 253–264. See also Field density model lamination of, function and, 329 in memory consolidation, 5–6, 12n2 in memory processing, 253 in storage of accessible information, 156 Networks. See Memory networks; Neural networks Neural networks. See also Memory networks attractor, 207–208, 213 autoassociative, 208, 213, 332–333 continuous, 207–210, 213 and episodic memory, 207–208, 213 hippocampus as, 360 Mexican hat–type, 247–248 coherent, in medial entorhinal cortex and hippocampus, 243–245 grid cell–place cell networks, 241–243 grid cell network models, 245–248 hippocampal–prefrontal network, 290–291 Neuronal plasticity. See Synaptic plasticity Neurophysiological studies, of hippocampal subfields in spatial context processing, 91–96 NMDA channels, in synaptic modification, 347–348 NMDA receptor–based synapse potentiation, 87 NMDA receptor blockade experiments, 87–88, 96, 110 NMDA receptors AMPA negative feedback pathway and, 344 genetic knock-out studies of, 87, 89
SUBJECT INDEX long-term potentiation, and age-related changes, 368–371 in modulation of late-phase LTP, 274, 277 in novel spatial context representation, 87–89 phase precession and, 367 plasticity mediation by, 110–111 Nodal cells, memory space hypothesis and, 219, 230 Nonmetric features, in context definition, 8 Nonspatial information. See also Context-dependent hippocampal activity association with spatial information by space cells, 237, 248 entorhinal cortex input of, 119–120 hippocampal encoding of, 164, 170, 219 place cell response to, 65–67, 98–100 Nonvisual information mediation of use of, 18 visual information association with, 17 Norepinephrine (NE) in mediation of alerting, 113 in response to novelty, 116 Novel environment exploration, development of place field stability during, dopamine and, 277–278 Novel spatial context representation CA3 subfield in, 87–88, 96 NMDA receptor in, 87–89 Novelty context adjustment and, 7 context discrimination and, 31–34 Novelty detection. See also Match–mismatch computations; Mismatch detection in aged individuals, 355–360 CA3 subfield in, 360 and context-dependent learning, 18–19, 28, 36 context discrimination in, 31–34 dopamine and, 114–115 in memory retrieval, 85 NMDA receptors and, 110–111 Novelty-induced dopamine release, in consolidation of new information, 280–281 Nucleus accumbens, hippocampal output to, in field density model, 256 Object, definition of, 201 Object–place memory, fornix lesions and impairment of, 193 Object–place memory tasks, 194–195 Object–place neurons, in primate hippocampus, 201 Odor stimuli, encoding and retrieval of, 323 6-OHDA, 272, 273 Orbitofrontal cortex, entorhinal input from, 192 Orientation and navigation. See Navigation Orienting, 113, 117 Pallium, evolution of, 328 Papez’s circuit, 151, 156 Parahippocampal gyrus hippocampal input from, 192 in spatial processing of mental images, 156
405
Parietal cortex in egocentric mental imagery, 156 hippocampal input from, in primates, 192 in path integration, 153 in spatial information processing, 171 Parkinson’s disease (PD), 272, 282 Partial remapping, 17, 62–63 Past, representation of, context and, 12 Path integration continuous attractor neural networks and, 208–210, 213 grid cells and, 77, 78 head direction cells and, 76–77 hippocampal place cells and, 73–74 limbic system in, 78–79 medial entorhinal cortex and, 97, 241 nonvisual information and, 17 place cell firing and, 150, 153–154 subicular place cells and, 78 Pattern completion as basic computational algorithm, 36 CA3 subfield in, 29, 62, 85–86, 89, 93–95, 96, 152, 360 computational models of, 85–86 cue removal and, 92 hippocampal system and, 12n6 in learning, 33 simultaneous monitoring of CA1 and CA3 subfields and, 93–95 Pattern generalization, 95 Pattern separation as basic computational algorithm, 36 computational models of, 85–86 cue manipulation and, 92 dentate gyrus in, 29, 89–90, 360 environmental changes and, 62 hippocampal system and, 12n6 in learning, 33 simultaneous monitoring of CA1 and CA3 subfields and, 95 PD (Parkinson’s disease), 272, 282 Perceptual influence on place cell firing, location-specific mental images and, 150, 151 Performance of spatial tasks, place cell firing and, 140–143, 145–146 Perirhinal cortex and context processing, 260–261 hippocampal input from, in primates, 192 influences on place cell firing, field density model and, 259–260 reward-related information and, 225 stimulus-selective match–nonmatch signals and, 225 PFC. See Prefrontal cortex Phase precession. See Theta-phase precession Place, definition of, 201 Place cells description of, 254 early experimental studies on, xvii–xix, 219 primate spatial view cells compared to, 194, 195–196, 200, 210–212
406 SUBJECT INDEX Place-differential neural responses change in, remapping and, 188–189 of primates cognitive map and, 182–185 task dependent, during real and virtual navigation, 178–182 in virtual spaces, 186–188 Place fields boundaries and, 67–68, 69 of CA3 vs. CA1, relative contribution of, 29–30 cognitive map theory and, 60–61 correlation of, across tasks, 181 determinants of, 16 as experience-dependent sensory and behavioral information, 16–18 formation of NMDA/AMPA receptor mediation in, 110–111 time frame for, 109 forward or backward expansion of, 25 grid cell input and, 79 of lateral septum, 27 learned behavioral responses and, 24 sensory receptive fields compared to, 108 spatial memory performance correlation of, 120 stability of adrenergic system and, 115–116 attentional modulation and, 111–114 dopaminergic modulation of, 113, 277–280, 284 Planned behavior, prefrontal cortex and, 289–290, 294 Plasticity. See Synaptic plasticity Pointer translocation task (PTT), 186 Population coding, in segregation of representations, 134 Population efferent code, in context processing, 28–29, 33 Position decoding, on continuous attractor, 333–335 Prediction, network model of, 338, 339 Pre-exposure effect, 5, 6 Prefrontal–hippocampal network anatomy and physiology of, 290–291 in working and episodic memory, 290 Prefrontal cortex (PFC) in computational models of memory-guided behavior, 316–317, 322 in delay interval of memory tasks, 223–224 in generation of context-dependent activity, 55–56 goal-directed behavior and, 115, 261–263, 290, 291, 295 hippocampal input from, in primates, 192 hippocampal interactions of, and memory, 300–304 and hippocampal memory processing, 34–35 lesions of, and hippocampal inhibition, 262–263 and planned behavior, 289–290, 294 prelimbic//infralimbic function in spatial representation, 291–293 in spatial information processing, 171 spatial representations in, 291–296, 304 in spatial working memory, 156–157 theta rhythms and, 26 Prelimbic/infralimbic prefrontal cortex function in spatial representation, 291–293 Presubicular head direction cells, primate vs. rodent, 207
Primates attentional studies in, compared to rodents, 107–108 hippocampus in learning and memory, 218–230 place-differential neural responses during real and virtual navigation, 177–189 spatial view cells and memory in, 192–214 visual context processing of, 24, 93 Procedural memory, aging and, 353, 354 Prospective coding behavioral context and, 111 cognitive demand and, 17 in context comparisons, 29, 30, 33 place cell firing and, 144 sequential representation of episodes and, 166 task-specific events and, 25–26 PTT (pointer translocation task), 186 Pyramidal cells, age-related changes in, 365–368, 371 Random foraging, place cell firing during, 139–140, 147 Rate remapping in CA3 subfield, 56 and encoding of choice, 258 entorhinal cortex and, 245 novelty and, 7–8 Recognition memory, 218, 220–221, 224–225 ‘‘Re-experience’’ of events, 150, 155–156, 157 Reference associations, 228–229 Reference frame. See Context Reinforcement learning, as model for place field modulation, 114 Relational memory, hippocampus in, 224 Relational memory hypothesis, 164–171 Remapping acetylcholine and, 117 as change in place-differential activity, 188–189 coherent network dynamics in, 243–245 complete, 17 context changes and, 7–8, 154 environmental changes and, 61–62 as evidence for soft-wired system, 108–109 field density model and, 257–258 global alternate models of, 245 context changes and, 7, 8 grid cells and, 243–245 grid cell map shift during, 329–330 incremental, 155 neuromodulation of, by dopamine, 280–281 nonspatial stimuli and, 66 partial, 17, 62–63 prefrontal cortex activity shifts compared to, 294–296 rate remapping in CA3 subfield, 56 and encoding of choice, 258 entorhinal cortex and, 245 novelty and, 7–8 regions contributing to, 258–263 perirhinal cortex, 259–261 prefrontal cortex, 261–263 time course of, 109, 155
SUBJECT INDEX Reorganization. See Remapping Replay of neural activation, 26 Representations, activation of relevant. See Segregation of memories Reptilian brain, 328, 330–331 Response position, memory of, 193, 194 Retrospective coding cognitive demand and, 17 in context comparisons, 29, 30, 33 sequential representation of episodes and, 166 task-specific events and, 25–26 Retrosplenial cortex, in path integration, 153 Reward–place neurons, in primate hippocampus, 204–206 Reward prediction, dopamine and, 296 Reward-related information input to hippocampus, 192 in memory tasks with short delay intervals, 225–226 Reward value, place cell encoding of, 144 Rodent place cells, primate spatial view cells compared to, 194, 195–196, 200, 210–212 Rodents, attentional studies in, compared to primates, 108 Route knowledge, 170, 171 Route retrieval, hippocampus in, 170–171 Schaffer collateral synapses, age-related changes in, 368–371 Segregation of memories coordinated use of two spatial representations and, 129–131, 136 hippocampal involvement in, 128–129, 135–136 place cell discharge and, 131–135 population coding and, 134 spatial knowledge system and, 127–128 Selective attention, in memory consolidation, 119–120 Self-motion information. See Locomotion; Motion-related spatial updating Self-motion processing, in aged individuals, 358–360 Semantic information, hippocampus in storage of, 219, 222, 224, 230 Semantic memory structures, cognitive maps as, 169, 170 Sensory information, neuronal firing and, 16–17 Sensory maps, cortical lamination and, 329 Sensory receptive fields, place fields compared to, 108 Sensory systems, spatial organization of, 24 Septohippocampal neurons, 314 Septum amnesia and, 87 lateral place fields of, 27 spatial-context information relay and, 27 medial inactivation of, and CA1 vs. CA3 place field disruption, 29–30 motivational modulation and, 27 theta rhythms and, 313–314 motivational modulation and, 27 place-differential neurons in, 185 Sequential representation of episodes, 166–167, 169, 170, 171 memory space hypothesis and, 219
407
Shape stimulus dimension, place cell response to, 66 Shortcut behavior, 141–142 Short-term memory. See Working memory Single-unit activity, spatial selectivity of, 271, 276–280 Sleep CA1 neuron activity and, 374–378 hippocampal–cortical interactions during, 301–304 Slow-wave sleep CA1 neuron activity and, 374–378 hippocampal and cortical activity during, 302–304 replay during, 26 Spatial behavior acetylcholine and, 282–283 cognitive maps and, 138 dopamine and. See Dopamine functional and anatomic network for, 147 goal encoding by place cells and, 144–145, 146–147 hippocampal and dorsal striatal–dependent, 271 hippocampal place cells and, 138–139 memory of, and place cell firing, 143–144, 147 performance of, and place cell firing, 140–143, 145–147 Spatial configurations, in context definition, 8, 12n5 Spatial context. See also Context configurational nature of, 3 distributed network of, 35 entorhinal grid cells in, 12n4 and episodic memory, 83 hippocampal system and, 3–4 information transmission to lateral septum, 27 navigation and, 23, 24, 34 place cell encoding of, 67 remapping and, 155 Spatial context processing hippocampal subfields in, 82–100, 360 computational models of, 84–86, 93, 96 conclusions, 96–100 differential roles of subfields in, 93–96, 98 overview, 82–84 perturbation studies and, 86–90 physiological studies and, 91–96 theories of, 82 Spatial context representation, novel CA3 subfield in, 87–88, 96 NMDA receptor in, 87–89 Spatial information entorhinal cortex input of, 119–120 retrieval of, in Parkinson’s disease, 272 Spatial information processing cortical function in, 171 and hippocampal context processing, 19, 23–24 and hippocampal function in memory, 161–164, 171, 264 hippocampal lesions and, 162 Spatial knowledge system, place cells as part of, 127–128 Spatial maps allocentric. See Allocentric representations; Cognitive maps universal, 63–64, 238–241, 329
408 SUBJECT INDEX Spatial memory hippocampal amnesia and, 163, 271 impairments of, in aged individuals. See Age-associated alterations in neural circuits; Age-related memory impairments place field correlation with performance, 120 Spatial moving task, 177 Spatial representation. See also Allocentric representations; Cognitive maps; Egocentric representations ambiguity of, 61–63 in amnesia patients, 163, 271 attentional modulation of, 111–113 coordinated use of two, 129–131, 136 by entorhinal cortex vs. place cells, 108–109 field density model and, 263–264 hippocampal, in memory segregation, 127 hippocampal lesions and, 162–163, 254 hippocampus in, 64–67, 219 in medial entorhinal cortex, 237–241, 248 in monkey hippocampal formation, 182–185 mosaic map representations and, 67–68, 69 multiple forms of, 170 place cell encoding of, 68–69 of places viewed. See Spatial view cells in prefrontal cortex, 291–296, 304 rigidity of, in aged individuals, 355–360, 372–373 Spatial scene memory, fornix lesions and impairment of, 193 Spatial stimuli responses, during memory tasks with short delay intervals, 222–223 Spatial tasks, performance of, place cell firing and, 140–143, 145–146 Spatial updating, motion-related. See also Locomotion in aged individuals, 358–360 medial entorhinal cortex and, 239–241 motor-efference signals and, 154, 157 perirhinal cortex and, 260 Spatial view cells activity of, with obscured vision, 196–200 and allocentric representation, 192, 195 characteristics of, 200–201 in memory, 210, 212–213 object–place associations and, 201 overlapping fields of, 200 rodent place cells compared to, 194, 195–196, 200, 210–212 Spike timing age-related changes in, 366 phase precession effect, 24–25 synchronization of place cells and, 25–26 Spike timing–dependent plasticity, 319, 321, 323 Spin glass analysis, 330, 340 Splitter-cell phenomenon, 320, 323 Stimuli familiar, response to, 229, 230 hippocampal response to various types of, 65–67 odor, encoding and retrieval of, 230 spatial, response to, 222–223 visual and auditory, responses to, 222 visual, and spatial responses, 206–207, 228 Stimulus–response learning, hippocampal lesions and, 34
Stimulus-responsive cells, effects of memory on, 224–225 Stimulus-selective memory responses, as prototypical memory signal, 220, 223–224, 228 Storage capacity, computational model of, 85–86 Striatum in decision making, 296–297 dorsal striatal–dependent behaviors, dopamine and, 271, 272–273 in generation of context-dependent activity, 54–55, 56 in stereotyped motor programs, 50 ventral, and hippocampal activity, 296–300 Subcortical systems. See also specific subcortical structures in amnesia, 87 hippocampal input from, 192 modulation by, 27 Subfields. See Hippocampal subfields Subicular place cells, context invariant properties of, 77–78 Subiculum, dopamine modulation in, 274 Substantia nigra, brain regions receiving dopaminergic innervation from, 272–273 Superposition catastrophe, 134, 136 Survey knowledge, 170, 171 Synaptic changes, age-related, 368–371 Synaptic plasticity age-related changes in. See Age-associated alterations in neural circuits of CA3–CA1 Schaffer collateral synapse, 368–371 dopamine and, 114–115, 271, 273–276 of hippocampal neurons, 109–111 neuromodulators of, 114–118 NMDA and, 347–348 spike timing–dependent, 319, 321, 323 theta rhythms and, 315 Synaptic resistance, in storage of distance between place cell firing fields, 343, 346–348 Systems-level memory consolidation, development of theory, 9 Task demand, place cell firing and, 139–140 Task-dependent hippocampal responses, 178–182 Task-specific events, spike timing and, 25–26 Temporal coding, hippocampal neural responses and, 65–66 Temporal cortex visual areas, in primates, 192 visual representations and, 171 Temporal-difference learning algorithm, 273, 281 Temporal lobe, medial allocentric representations and, 177 declarative memory and, 218–219, 226 in delay interval of memory tasks, 223–224 hippocampal input from, in primates, 193 memory over a short delay period and, 223–224 recognition memory and, 224–225 reward-related information and, 225 visual and auditory association areas, in primates, 192 Temporally regulated neural firing, prospective codes and, 17 Temporal organization in avoidance of superposition catastrophe, 134 CA1 subfield in, 30
SUBJECT INDEX cognitive maps vs., 23 and hippocampal context processing, 19, 24–26, 34 incremental remapping and, 155 memory space hypothesis and, 219 place cells and, 65–66 Temporal pattern association, hippocampal subfields in, 338 Temporal stamping, 170, 171 Tetrodotoxin (TTX), 128–129 Theta cells, velocity and, 17 Theta-phase precession. See also Spike timing age-related changes in, 366–367 computational model of, 320–321 in encoding and retrieval of odor stimuli, 323 importance of discovery of, 24–25 long-term potentiation induction and, 315 Theta rhythms acetylcholine and, 117, 282 age-related changes in, 365–367 and attention, 108 behavioral correlates of, 313–315 computational models of memory-guided behavior and, 315–323 and hippocampal unit firing, 315 in learning and memory, 313–315 medial septum and, 27 movement-related vs. immobility-related, 313 septohippocampal neurons and, 314 spike timing relative to, 24–25, 26 Threshold learning rule, 344–346, 347 Topographical organization grid cells and, 109, 238, 240, 241, 242, 247–248 place cells and, 91–93, 109 Topography of sensory and motor systems, 24, 109 Transient stimuli, place cell response to, 65 Transitive inference, hippocampus implicated in, 222 TTX (tetrodotoxin), 128–129 TVTA. See Ventral tegmental area Universal spatial map, 63–64, 238–241, 329 Velocity, location-sensitive firing and, 17 Ventral tegmental area (VTA) brain regions receiving dopaminergic innervation from, 272–273
409
dopaminergic neurons of, and reinforcement learning, 114 and novelty-induced dopamine release, 281 View neurons. See Spatial view cells Viewpoints, shift of, hippocampus in, 154 Virtual navigation extramaze distal cues, and primate neuronal responses, 185–189 human, hippocampal neuron firing and, 168, 171 place-differential neural responses during, 178–182 Virtual rat simulations of memory-guided behavior, 315–322 modeling of CA1 and CA3 subfield differentiation, 335–341 Virtual translocation task, place-differential neural responses and, 178–179, 181 Vision relative importance of, in primates, 192, 196–199. See also Spatial view cells in rodents vs. primates, 210 Visual context processing, by primates, 24, 93 Visual information associations with spatial responses, in primates, 206–207 nonvisual information association with, 17 sensitivity of place fields to, 16–17 Visual stimuli responses, during memory tasks with short delay intervals, 222 Voluntary navigation. See Exploration Water maze task, 354–355 Whole-body motion, hippocampal response to, 207, 213 Working memory hippocampal–prefrontal network and, 290 location-specific mental images and, 156–157 Working memory tasks delay period activity, 223–224 in memory phase, 224–225 motor-related activity, 225 reward-related signals, 225–226 to spatial stimuli, 222–223 to visual and auditory stimuli, 222 zif268 gene, dopamine in expression of, 275–276