e d i to r i a l
Ensuring data integrity A recent report by the National Academy of Science makes recommendations for ensuring the integrity of research data. Critically, it also highlights the urgent issues regarding the preservation of large datasets.
© 2009 Nature America, Inc. All rights reserved.
R
eports of scientific misconduct often make for sensational media reports. High-profile cases of falsification inevitably call for a re-examination of whether and how fraud can be detected before publication. The growing ease and practice of digital data transformation make this issue ever thornier. Scientists can be genuinely misled into thinking that they have found a specific result, only to later discover that they’ve been fooled by an artifact created by their digital analysis methodology. Compounding such data integrity problems is the fact that researchers can now amass huge amounts of data with relatively little effort; consider, for example, the gene lists that are generated by a single microarray experiment or the new RNA-seq technology, or large-scale detailed brain maps that typically require tens of gigabytes of memory per image. Established guidelines on best practices for data analysis, integrity, accessibility and archiving have not kept pace with this data explosion. The problem appears particularly urgent in interdisciplinary fields such as the neurosciences, in which researchers often navigate multiple layers of resolution, including gene expression datasets, cellular imaging and physiology, functional imaging and clinical data, all of which may have their own standards to safeguard data accuracy. Soon after the notorious Hwang stem cell fraud case, a group of scientific societies and publishers, including the Nature Publishing Group, approached the US National Academy of Science (NAS) to encourage a thorough study of data manipulation and preservation in the digital age and to recommend adequate best practices to ensure the accessibility of large datasets. The NAS committee, headed by cancer researcher Phillip Sharp and physicist Daniel Kleppner, published its report this July (http:// books.nap.edu/openbook.php?record_id=12615). The report’s central tenet is that the individual scientists are responsible for the truth and accuracy of their data. Most of its recommendations for ensuring data integrity and combating fraud are common-sense guidelines that are followed in most laboratories. Research institutions need to ensure that appropriate tools for management of research data are available to their scientists. The report also emphasizes the obvious, that data and experimental details must be made accessible and archived to allow for replication and consequent studies. As expected, the panel found that different disciplines have rather diverse requirements regarding data quality. Journals, as stakeholders in the research enterprise, also need to do their part, and many (including Nature Neuroscience) have taken steps to enhance the quality and reproducibility of published work. Journals may require detailed methods sections, mandate author contribution statements and many have published explicit policies on the manipulation of raw data. We, for example, ask authors to list all image-acquisition tools and image software packages, and if cropped electrophoretic gels are included in the paper, these must be indicated as nature neuroscience volume 12 | number 10 | OCTOBER 2009
such in the figure legend and uncropped gels and blots must be included in the Supplementary Information (http://www.nature.com/authors/ editorial_policies/image.html). Above and beyond recommending measures to combat mis-analysis and fraud, the NAS report also calls for an urgent evaluation of the provisions for long-term maintenance of research data. This involves the critical question of how the community (including individual scientists, universities, funding agencies and journals) can ensure that large datasets are appropriately stored, referenced and indexed for posterity. To achieve these goals, scientific disciplines and communities must first agree on the criteria as to what data should be retained, such as information about instrument calibration and proprietary tools, details of the data processing methodology, and similar nitty-gritty, but important, issues. Moreover, as the ultimate value of scientific datasets will depend on an interlinked database infrastructure, we must attempt to coordinate data standards between disciplines to ensure compatibility and avoid redundancy. In addition to agreeing on criteria for data annotation, communities must also agree on formal vocabularies for their data and concepts, to enable unambiguous description of data and to make it machine-readable. Currently, researchers have few incentives to invest much time and energy into data preservation and annotation; a situation which has to be remedied if we are to ensure the integrity of digital data. Funding must be made available to achieve lasting conservation of and access to data. Funding bodies and scientists must also work toward developing metadata management tools that would help researchers annotate data more easily and creating software that would make it possible to track individual pieces of data so as to give credit where credit is due. The increasingly important role of data-processing professionals in all scientific endeavors must also be better recognized. Better training and education of scientists in data stewardship issues is critical at this time. Many scientists have had little or no formal training in information management and are therefore simply ill-prepared to think intelligently about these matters. As the NAS panel points out, data management must start at the beginning of a project, not midway through it or as an afterthought. Institutions must put data stewardship policies in place and promote the necessary training of their employees. Such training would include an understanding of the storage and preservation of data, its annotation, some the central online databases and their organization, and an appreciation of the bioinformatic tools that are available. The NAS panel correctly notes that maintaining the integrity and accessibility of research data in this evolving digital age requires the collective efforts of individual scientists, research institutions, funding agencies, universities and journals. We urgently need to invest into our bioinformatic infrastructure to create the framework necessary to ensure that data is stored, annotated and preserved in a way that will provide maximum benefit for future studies. L
1205
book review
A neurocomputational jeremiad Memory and the Computational Brain: Why Cognitive Science will Transform Neuroscience by C R Gallistel & Adam Philip King
© 2009 Nature America, Inc. All rights reserved.
Wiley-Blackwell, 2009 336 pp, hardcover, $99.95 ISBN 1405122889
Reviewed by Peter Dayan Along with a light complement of fascinating psychological case studies of representations of space and time, and a heavy set of polemical sideswipes at neuroscientists and their hapless computational fellow travelers, this book has the simple goal of persuading us of the importance of a particular information processing mechanism that it claims does not currently occupy center stage. The authors maintain that “there must be an addressable read/write memory mechanism in brains that encodes information received by the brain into symbols (writes), locates the information when needed (addresses) and transports it to computational machinery that makes productive use of the information (reads)”. Most of the chapters are devoted to unpacking this statement, describing conventional computer science notions of representations, symbols, information processing and Turing machines. The authors stress that neuroscientists completely ignore the issue of addressable read/write memory and/or propose preposterously inadequate solutions. The book coyly forswears a solution of its own, bar the Conradian possibility that something close to the molecular heart of neurons might be involved. Some issues the book brings to center stage are spot on, notably representation: how indeed can the brain realize complex cognitive entities, sentences or even just visual scenes. The book contains good discussions about important notions such as productivity, compositionality and systematicity that were the focus of previous debates about connectionist representations, thus providing insights into how computers can be engineered and programmed to cope. However, such concerns nearly get buried by a reluctance to seriously consider the possibility that neural realizations of representation and computation might look nothing like engineered solutions. Indeed, there is altogether somewhat little regard for the structural and physiological facts of the brain. Central billing in the book’s conception of what is missing in contemporary neuroscience goes to two rather different computer architectures: the conceptually important Turing machine and something like a conventional computer, which is of more practical concern. Turing machines have two key components, a finite state machine, instantiating if-then-rules associated with transitions between states, and an infinitely Peter Dayan is in the Gatsby Computational Neuroscience Unit, University College London, London, UK. e-mail:
[email protected]
nature neuroscience volume 12 | number 10 | OCTOBER 2009
long tape storing symbols (for example, just 0 and 1), which can be read from or written to on the finite state machine’s orders. Among the hardware differences between a universal Turing machine and my desktop computer are an addressable memory and finiteness. The former makes it very easy to construct powerful data structures such as lists and trees, which in turn make it simple to write programs to solve the representational and processing puzzles posed by the book’s psychological examples. The book claims that neuroscience is ignoring symbols and addressing, for instance, maintaining that information located in synapses is somehow too inaccessible and implicit. It also suggests that Turing tape is missing, deriding the two obvious neural possibilities of plastic synapses and reverbatory activity. For the issue of addressing, the book really hoists itself on its own petard of Turing universalism. After all, a perfectly good Turing machine can also lack addressing; the tape is just read step by step without explicit (for example, numbered) locations. However, it can nevertheless simulate my computer’s addressing schemes. It seems no less plausible to interrogate a synapse via presynaptic activity than to traverse reels of tape step by step, in both cases to get to read their states. Equally, the state of an elevated or depressed synaptic conductance seems no more or less able to realize a symbol when embedded in an appropriate computational milieu than would be the state of a tape entry being 1 or 0. When, at the end of the book, the authors finally accept something like this functionalist point, the overall thesis somewhat unravels. The book suffers similar architectural blinders in suggesting that there is a fundamental difference between a Turing machine with a finite tape and a finite state machine, with the latter facing nasty combinatorial explosions in representing information. This is not true; my computer can actually be considered as a huge finite state machine with a very particular structure, in the form of stringent restrictions on the possible transitions. It is facile to demand a transparent mapping of “silicone” (sic) concepts onto biological ones; just because it doesn’t look or walk like a duck doesn’t mean that it can’t realize the quack. Finally, we can come back to the odd parts of the phrase in the quote above “locates ... and transports [information]”; this is wedded to a deeply conventional notion that there is ‘dumb’ peripheral memory (or tape) and a ‘smart’ centralized computational device. In comparison, from the relative uniformity of cortical architecture, the brain looks as if computational power and storage are generally colocated and broadly distributed; there is no reason to expect any transportation in any conventional sense. How about the mechanistic realization of something equivalent to tape? Some of the specific systems the book describes (for example, desert ants doing dead reckoning across featureless desert rocks) present compelling biophysical puzzles, but are not persuasive about the sort of general computations we expect to be enabled by tape. Outside the temporal window of reverbatory memory, we would mostly require the realization of one-shot or snapshot storage and recall. This point is not lost on neuroscientists, but is rather the intense focus of an impressive and vocal array of them. Students of Gallistel’s influential previous books, The Organization of Action and The Organization of Learning, had been eagerly awaiting an Organization of Computation. It’s not clear that this one is quite yet ready for writing to tape. L
1207
news and views
A night vision neuron gets a day job Nicholas Oesch & Jeffrey Diamond
© 2009 Nature America, Inc. All rights reserved.
During the day, certain retinal ganglion cells respond specifically to dark, approaching stimuli. A study finds that the retinal circuit that gives rise to this response makes use of an amacrine cell that was previously known for its role in night vision circuitry, demonstrating that some neurons lead double lives. To reduce the daunting complexity of the nervous system, we often presume that each neuron or circuit is dedicated to one particular function. Given the diversity of neurons and the fact than most have yet to be assigned to a specific task, this simplifying tactic seems reasonable; with so many neurons still out of work, there’s no need to give two jobs to one cell type. This socialist attitude toward the neuronal labor force may be changing, however, as mounting evidence indicates that the AII amacrine cell, an important interneuron in the retina’s night vision circuitry, may be moonlighting1–4. In this issue, Munch et al.5 demonstrate that AII cells also work the day shift as a crucial component in a newly described computational circuit for detecting approaching motion. The AII cell is best known as the link between the circuits mediating our exquisitely sensitive night vision and color-coded day vision6–8 (Fig. 1a). Our photopic (daytime) vision is mediated by cone photoreceptors, whose signals are divided into parallel ON and OFF pathways by second-order neurons, the ON and OFF cone bipolar cells. The ON pathway signals light increments and the OFF pathway signals light decrements. ON and OFF bipolar cells directly contact their respective ON and OFF ganglion cells, which transmit the visual signal through the optic nerve to the rest of the brain (Fig. 1b). Inhibitory circuitry in the inner retina shapes the inputs to ganglion cells to produce temporally precise, feature-selective signals. Under scotopic (starlight) conditions, photons are absorbed only by the more sensitive (but easily saturated) rod photoreceptors, which contact second-order rod bipolar cells. The AII cell relays the purely ON signal from rod bipolar cells to ON cone The authors are at the US National Institute of Neurological Disorders and Stroke, US National Institutes of Health, Bethesda, Maryland, USA. e-mail:
[email protected]
a
b
Night
Day
Rod Cone
Rod bipolar
ON bipolar
OFF bipolar AII
Gly
AII
Glu
Glu
Gap Gly Glu
Glu ON ganglion cell
OFF ganglion cell
Gap
Glu
Figure 1 Scotopic and photopic retinal circuitry. (a) The scotopic path of information flow from the rod → rod bipolar → AII → ON cone (gap junction, Gap) and OFF cone bipolar (glycine, Gly) → ON and OFF ganglion cell (in blue). (b) The photopic flow of information from the cone → cone bipolars → AII (gap junction) and ganglion cells (glutamate, Glu). Sign-conserving synapses are shown in green and sign-inverting synapses are shown in red.
bipolar cells via sign-conserving electrical synapses and to OFF cone bipolar cells through inhibitory (sign inverting) glycinergic synapses, thereby routing low-light information into the standard retinal circuitry mediating daytime processing (Fig. 1a). How the AII cell spends its days has long been a mystery, but early anatomical evidence that it makes chemical synapses onto OFF ganglion cells provided clues that it was up to something9,10. More recent physiological experiments have confirmed that those synapses are active and that AII cells directly inhibit some ganglion cells1–3, and others have shown that AII cells remain
nature neuroscience volume 12 | number 10 | OCTOBER 2009
responsive when the lights come up4,11,12. In this issue, Munch et al.5 show that the AII cell functions in circuitry that enables a certain ganglion cell subtype, the PV-5, to respond specifically to dark objects of increasing size, a task that probably enables would-be prey to detect approaching predators. The identification of such sophisticated processing in the retina is exciting in its own right and the authors’ elegant description of the underlying computational mechanism also provides general insights into the nature of complex feature detection. In a robust feature-selective circuit, the output neuron responds to the ‘preferred’ stimulus
1209
news and views a
Approaching
Pushpull subunit
© 2009 Nature America, Inc. All rights reserved.
b
Moving
Pushpull subunit
Figure 2 Multiple push-pull subunits form an approach detector. (a) Activation of excitation in pushpull subunits during expanding stimulus. Because the subunits are OFF subunits, the dark edge of the stimulus will activate excitation as it moves over the subunit. (b) Activation of both excitation and inhibition in push-pull subunits during moving stimuli. The leading edge of the dark stimulus will activate OFF excitation, whereas the trailing edge of the stimulus will activate ON inhibition. Excitation and inhibition are summed in the postsynaptic ganglion cell and the implementation of a spike threshold nonlinearity results in action potentials (approaching stimulus) or quiescence (moving stimulus).
over a wide range of conditions, but remains silent during inappropriate, ‘null’ stimulation. One strategy to accomplish this selectivity is to incorporate inhibition that is tuned to the anti-feature (that is, the null stimulation), as Munch et al.5 found for the PV-5 cell. When the authors measured excitatory and inhibitory inputs to approach-selective PV-5 ganglion cells, they found that approaching (expanding) stimuli elicited only excitatory inputs. In contrast, laterally moving stimuli evoked both excitatory and inhibitory inputs that cancel each other. To determine how inhibition is activated selectively by null stimuli, the authors examined the rather complex characteristics of the PV-5 cell’s receptive field. Because the approach-selective ganglion cell is an OFF cell, a dark spot elicits excitatory input; the authors also demonstrate that termination of the dark stimulus (increase in light) elicits an ON inhibitory input. This organization of excitation and inhibition in opposing pathways, often referred to as a push/pull receptive field, is a feature of other retinal neurons13. During
1210
uniform changes in illumination, the push and pull are well separated in time and may simply expand the dynamic range of the cell. The key to transforming this common organization into a feature-selective circuit is the authors’ discovery that the PV-5 cell’s receptive field comprises an array of push/pull subunits, with each subunit being capable of providing excitatory OFF input or inhibitory ON input, depending on the stimulus in the area of the visual field to which it is sensitive. The incorporation of multiple subunits into a single PV-5 receptive field gives rise to an approach detector because an expanding dark stimulus activates an increasing number of OFF subunits, thereby increasing excitatory input without activating ON inhibition. In contrast, a moving dark stimulus of fixed size activates OFF excitation and ON inhibition in the subunits beneath its leading and trailing edges, respectively (Fig. 2). Numerical simulations suggest that if the temporal characteristics of the excitation match those of the inhibition and the inputs are rectified, conditions that
the authors confirmed experimentally, then inhibition vetoes inappropriate responses over a wide range of stimulus parameters. What circuit element provides this inhibition to PV-5 cells? Standard pharmacology revealed that inhibition was indeed supplied via the ON pathway by a glycinergic amacrine cell. Surprisingly, neither AMPA nor NMDA receptor antagonists blocked this inhibition, as would be expected if the glycinergic amacrine cell signal were driven by glutamatergic input from bipolar cells. Instead, the authors found that the ON inhibition was driven through gap junctions, as approach selectivity was reduced in mice lacking Cx36, a connexin that is strongly expressed by AII amacrine cells. Further evidence of the AII cell’s involvement is provided by paired recordings demonstrating a direct glycinergic synaptic connection between presynaptic AII cells and postsynaptic PV-5 cells. Taken together, these results demonstrate that AII amacrine cells provide inhibition for the push/pull subunit. An interesting feature of the AII cell’s dual existence is that the direction of visual information flow through the neuron reverses between night and day. During night vision, the AII cell passes a depolarizing ON signal to ON cone bipolar cells (Fig. 1a). It has been speculated that the gap junction mediating this transfer might rectify or decouple during the day to prevent needless leakage in the other direction, but it now seems clear that depolarizing signals can pass in the opposite direction from ON cone bipolar cells to the AII cell (Fig. 1b). The molecular control of this bidirectional conduit remains incompletely understood. Rectification (favoring the AII → cone bipolar direction) during scotopic conditions slowly disappears as the circuit adapts to daylight4, indicating that the electrical connection is actively regulated. This could reflect parallel, independently regulated gap junctions with opposing rectification or asymmetric gap junctions comprising different connexins contributed by each cell4,14. The revelation of the AII cell’s multitasking raises many exciting questions about how this particular cell has mastered two such diverse occupations. Already, hints suggest that it may get help from the surrounding retina circuitry; the AII cell’s receptive field changes from a classical center-surround configuration under scotopic conditions to a purely ON center receptive field in daylight4. Perhaps this change in surround organization equips the AII cell to perform its daytime task of forming push/pull subunits. Important computational insights will be gained by understanding how the AII cell parses information during the daytime as well as at night. Similarly multifunctional neurons and circuitry throughout the brain would expand
volume 12 | number 10 | OCTOBER 2009 nature neuroscience
news and views the computational abilities of the nervous system well beyond previous calculations. Could certain neural pathologies be caused by an overworked neuron performing the wrong job at the wrong time? And is it possible for a neuron to perform multiple tasks simultaneously? At first glance, the AII cell’s duties appear to be functionally and temporally distinct; it passes single-photon signals at night and enables detection of an approaching falcon during the day. Perhaps future experiments will
explain what happens when an owl swoops in under the moonlight, while the AII cell is busy working the night shift. 1. Manookin, M.B., Beaudoin, D.L., Ernst, Z.R., Flagel, L.J. & Demb, J.B. J. Neurosci. 28, 4136–4150 (2008). 2. Murphy, G.J. & Rieke, F. Nat. Neurosci. 11, 318–326 (2008). 3. van Wyk, M., Wassle, H. & Taylor, W.R. Vis. Neurosci. 26, 297–308 (2009). 4. Xin, D. & Bloomfield, S.A. Vis. Neurosci. 16, 653–665 (1999). 5. Munch, T.A. et al. Nat. Neurosci. 12, 1308–1316 (2009). 6. Dacheux, R.F. & Raviola, E. J. Neurosci. 6, 331–345 (1986).
7. Famiglietti, E.V. Jr. & Kolb, H. Brain Res. 84, 293–300 (1975). 8. Kolb, H. & Famiglietti, E.V. Science 186, 47–49 (1974). 9. Kolb, H. J. Neurocytol. 8, 295–329 (1979). 10. Strettoi, E., Raviola, E. & Dacheux, R.F. J. Comp. Neurol. 325, 152–168 (1992). 11. Pang, J.J. et al. J. Physiol. (Lond.) 580, 397–410 (2007). 12. Pang, J.J., Gao, F. & Wu, S.M. J. Physiol. (Lond.) 558, 897–912 (2004). 13. McGuire, B.A., Stevens, J.K. & Sterling, P. J. Neurosci. 6, 907–918 (1986). 14. Dedek, K. et al. Eur. J. Neurosci. 30, 217–228 (2009).
Regional control of cortical lamination © 2009 Nature America, Inc. All rights reserved.
Ronald R Waclaw & Kenneth Campbell Laminar neuronal density varies between cortical areas; thus, the developmental specification of areas and layers needs to be coordinated. AP2γ turns out to be an important regulator of upper layer development in occipital cortex. The largest portion of the cerebral cortex, the neocortex, is characterized by a six-layered organization. The neocortex is also organized into areas, each of which exhibit distinct connectivity and carry out specific functions such as motor control or visual processing. Not only are the cortical areas functionally distinct, but they differ in their cytoarchitecture and laminar neuronal density1. The primate visual cortex, for example, contains 50% more pyramidal neurons in its upper layers (layers II/III) than neighboring cortical areas1. Molecular mechanisms that regulate the development of distinct cortical areas2 as well as laminar identity3 have recently been the subject of extensive investigation. However, the manner in which these two processes are integrated remains unclear. In this issue, Pinto et al.4 provide evidence that the transcription factor AP2γ is required for correct laminar development exclusively in the occipital (visual) cortex. Specifically, AP2γ is necessary and sufficient for the correct number of pyramidal neurons to be produced in layers II/III of the visual cortex. Indeed, Pinto et al. found that pyramidal neurons in layers II/III of the visual cortex, as marked by Cux1/2 staining or retrograde labeling from the opposite cortical hemisphere, were severely reduced in AP2γ (also known as Tcfap2c) conditional knockout mice (Fig. 1). In contrast, AP2γ loss did not affect the numbers of upper layer neurons in rostral cortical regions (Fig. 1). Neuronal density in the deep cortical layers V/VI, The authors are at the Division of Developmental Biology, Cincinnati Children’s Hospital Medical Center, University of Cincinnati College of Medicine, Cincinnati, Ohio, USA. e-mail:
[email protected]
as identified by the markers Er81 and Tbr1, was normal in the AP2γ mutant cortex at both rostral and caudal levels. Recent studies1,3,5–8 indicate that the neurons that occupy different cortical layers are generated from distinct progenitors in the cortical germinal zone. Deep-layer neurons arise largely from progenitors such as radial glia that divide at apical locations in the ventricular zone at early stages of corticogenesis, whereas the upper layer neurons are generated from intermediate progenitors that divide at the basal margin in the subventricular zone (SVZ) at later time points (Fig. 1a). These two progenitor types have also been termed apical and basal progenitors, respectively. The basal (intermediate) progenitors arise from radial glia and usually divide symmetrically in the SVZ to produce two neurons destined for the upper cortical layers (Fig. 1a). Because the upper layers of the occipital cortex in AP2γ mutants are specifically affected, Pinto et al.4 examined the generation of basal progenitors in the mutant cortical germinal zone. They found dividing cells at basal positions in caudal regions of the AP2γ mutant cortex and even found that the numbers of these cells were slightly increased at mid-neurogenesis stages. However, these cells lacked some typical features of basal progenitors and instead appeared to retain characteristics of the radial glial (apical) progenitors (Fig. 1c). A number of transcription factors such as Tbr2 (refs. 9,10), Insm1 (ref. 11) and Cux2 (ref. 12) have recently been shown to be required for the normal generation and neurogenic function of basal progenitors. These factors function broadly in the production of neurons from basal progenitors throughout the developing cortex, as no area-specific laminar defects have
nature neuroscience volume 12 | number 10 | OCTOBER 2009
been reported in mice that are deficient for any of these factors9–12. Pinto et al.4 found that the expression of many of these factors, including Tbr2 and Tis21 (refs. 7–10), was reduced or missing in caudal basal progenitors of the AP2γ mutant cortex (Fig. 1c). Pinto et al.4 found that AP2γ uniquely differs from other regulators of basal progenitors in a couple of ways. First, unlike Tbr2, Insm1 and Cux2 expression, AP2γ expression was restricted to the radial glial (apical) progenitors and was not maintained in the basal progenitors. Second, AP2γ only regulated the specification of basal progenitors in caudal portions of the developing cerebral (visual) cortex. Nevertheless, Pinto et al.4 observed the reduced neurogenic capacity of basal progenitors reported in Tbr2 (refs. 9,10) and Insm1 (ref. 11) mutants in the developing AP2γ mutant visual cortex. In addition, at mid-stages of cortical neurogenesis (embryonic day 14, E14), when upper-layer neurons commence generation, the number of basally dividing cells was increased in the AP2γ mutant cortex4, similar to observations in Cux2 mutants12. Indeed, both Tbr2 and Cux2 were severely downregulated in the caudal cortical germinal zone of AP2γ mutants. At late stages of cortical neurogenesis (E17), the basal progenitors were severely reduced in the AP2γ mutant visual cortex as a result of increased cell death (Fig. 1c). Thus, it appears likely that an initial misspecification of caudal basal progenitors leads to the reduced production of layer II/III neurons in the visual cortex of AP2γ mutants (Fig. 1c), indicating that AP2γ is required for the correct specification and/or generation of neurogenic basal progenitors in the caudal cortical germinal zone. Furthermore, Pinto et al.4
1211
news and views the computational abilities of the nervous system well beyond previous calculations. Could certain neural pathologies be caused by an overworked neuron performing the wrong job at the wrong time? And is it possible for a neuron to perform multiple tasks simultaneously? At first glance, the AII cell’s duties appear to be functionally and temporally distinct; it passes single-photon signals at night and enables detection of an approaching falcon during the day. Perhaps future experiments will
explain what happens when an owl swoops in under the moonlight, while the AII cell is busy working the night shift. 1. Manookin, M.B., Beaudoin, D.L., Ernst, Z.R., Flagel, L.J. & Demb, J.B. J. Neurosci. 28, 4136–4150 (2008). 2. Murphy, G.J. & Rieke, F. Nat. Neurosci. 11, 318–326 (2008). 3. van Wyk, M., Wassle, H. & Taylor, W.R. Vis. Neurosci. 26, 297–308 (2009). 4. Xin, D. & Bloomfield, S.A. Vis. Neurosci. 16, 653–665 (1999). 5. Munch, T.A. et al. Nat. Neurosci. 12, 1308–1316 (2009). 6. Dacheux, R.F. & Raviola, E. J. Neurosci. 6, 331–345 (1986).
7. Famiglietti, E.V. Jr. & Kolb, H. Brain Res. 84, 293–300 (1975). 8. Kolb, H. & Famiglietti, E.V. Science 186, 47–49 (1974). 9. Kolb, H. J. Neurocytol. 8, 295–329 (1979). 10. Strettoi, E., Raviola, E. & Dacheux, R.F. J. Comp. Neurol. 325, 152–168 (1992). 11. Pang, J.J. et al. J. Physiol. (Lond.) 580, 397–410 (2007). 12. Pang, J.J., Gao, F. & Wu, S.M. J. Physiol. (Lond.) 558, 897–912 (2004). 13. McGuire, B.A., Stevens, J.K. & Sterling, P. J. Neurosci. 6, 907–918 (1986). 14. Dedek, K. et al. Eur. J. Neurosci. 30, 217–228 (2009).
Regional control of cortical lamination © 2009 Nature America, Inc. All rights reserved.
Ronald R Waclaw & Kenneth Campbell Laminar neuronal density varies between cortical areas; thus, the developmental specification of areas and layers needs to be coordinated. AP2γ turns out to be an important regulator of upper layer development in occipital cortex. The largest portion of the cerebral cortex, the neocortex, is characterized by a six-layered organization. The neocortex is also organized into areas, each of which exhibit distinct connectivity and carry out specific functions such as motor control or visual processing. Not only are the cortical areas functionally distinct, but they differ in their cytoarchitecture and laminar neuronal density1. The primate visual cortex, for example, contains 50% more pyramidal neurons in its upper layers (layers II/III) than neighboring cortical areas1. Molecular mechanisms that regulate the development of distinct cortical areas2 as well as laminar identity3 have recently been the subject of extensive investigation. However, the manner in which these two processes are integrated remains unclear. In this issue, Pinto et al.4 provide evidence that the transcription factor AP2γ is required for correct laminar development exclusively in the occipital (visual) cortex. Specifically, AP2γ is necessary and sufficient for the correct number of pyramidal neurons to be produced in layers II/III of the visual cortex. Indeed, Pinto et al. found that pyramidal neurons in layers II/III of the visual cortex, as marked by Cux1/2 staining or retrograde labeling from the opposite cortical hemisphere, were severely reduced in AP2γ (also known as Tcfap2c) conditional knockout mice (Fig. 1). In contrast, AP2γ loss did not affect the numbers of upper layer neurons in rostral cortical regions (Fig. 1). Neuronal density in the deep cortical layers V/VI, The authors are at the Division of Developmental Biology, Cincinnati Children’s Hospital Medical Center, University of Cincinnati College of Medicine, Cincinnati, Ohio, USA. e-mail:
[email protected]
as identified by the markers Er81 and Tbr1, was normal in the AP2γ mutant cortex at both rostral and caudal levels. Recent studies1,3,5–8 indicate that the neurons that occupy different cortical layers are generated from distinct progenitors in the cortical germinal zone. Deep-layer neurons arise largely from progenitors such as radial glia that divide at apical locations in the ventricular zone at early stages of corticogenesis, whereas the upper layer neurons are generated from intermediate progenitors that divide at the basal margin in the subventricular zone (SVZ) at later time points (Fig. 1a). These two progenitor types have also been termed apical and basal progenitors, respectively. The basal (intermediate) progenitors arise from radial glia and usually divide symmetrically in the SVZ to produce two neurons destined for the upper cortical layers (Fig. 1a). Because the upper layers of the occipital cortex in AP2γ mutants are specifically affected, Pinto et al.4 examined the generation of basal progenitors in the mutant cortical germinal zone. They found dividing cells at basal positions in caudal regions of the AP2γ mutant cortex and even found that the numbers of these cells were slightly increased at mid-neurogenesis stages. However, these cells lacked some typical features of basal progenitors and instead appeared to retain characteristics of the radial glial (apical) progenitors (Fig. 1c). A number of transcription factors such as Tbr2 (refs. 9,10), Insm1 (ref. 11) and Cux2 (ref. 12) have recently been shown to be required for the normal generation and neurogenic function of basal progenitors. These factors function broadly in the production of neurons from basal progenitors throughout the developing cortex, as no area-specific laminar defects have
nature neuroscience volume 12 | number 10 | OCTOBER 2009
been reported in mice that are deficient for any of these factors9–12. Pinto et al.4 found that the expression of many of these factors, including Tbr2 and Tis21 (refs. 7–10), was reduced or missing in caudal basal progenitors of the AP2γ mutant cortex (Fig. 1c). Pinto et al.4 found that AP2γ uniquely differs from other regulators of basal progenitors in a couple of ways. First, unlike Tbr2, Insm1 and Cux2 expression, AP2γ expression was restricted to the radial glial (apical) progenitors and was not maintained in the basal progenitors. Second, AP2γ only regulated the specification of basal progenitors in caudal portions of the developing cerebral (visual) cortex. Nevertheless, Pinto et al.4 observed the reduced neurogenic capacity of basal progenitors reported in Tbr2 (refs. 9,10) and Insm1 (ref. 11) mutants in the developing AP2γ mutant visual cortex. In addition, at mid-stages of cortical neurogenesis (embryonic day 14, E14), when upper-layer neurons commence generation, the number of basally dividing cells was increased in the AP2γ mutant cortex4, similar to observations in Cux2 mutants12. Indeed, both Tbr2 and Cux2 were severely downregulated in the caudal cortical germinal zone of AP2γ mutants. At late stages of cortical neurogenesis (E17), the basal progenitors were severely reduced in the AP2γ mutant visual cortex as a result of increased cell death (Fig. 1c). Thus, it appears likely that an initial misspecification of caudal basal progenitors leads to the reduced production of layer II/III neurons in the visual cortex of AP2γ mutants (Fig. 1c), indicating that AP2γ is required for the correct specification and/or generation of neurogenic basal progenitors in the caudal cortical germinal zone. Furthermore, Pinto et al.4
1211
news and views AP2γ knockout
Wild type
a
b
c
Rostral
AP2 Tbr2, Tis21 Er81, Tbr1 Cux1 and 2
Caudal
II/III Mature cortex
IV V/VI
Cortical germinal zone
Basal (intermediate) progenitor
SVZ VZ
Apical (radial glia) progenitor
© 2009 Nature America, Inc. All rights reserved.
E12
E14 onward
E12
E14 onward
E12
E14
E17
Figure 1 Regional laminar defects in the cortex of AP2γ conditional knockout mice. (a) Schematic diagram illustrating the cellular generation of deeplayer (marked by Er81 and Tbr1) and upper-layer cortical neurons (marked by Cux1 and 2) from apical (radial glial) and basal (intermediate) progenitors, respectively, in wild-type cortex. Basal progenitors express Tbr2 and Tis21 and normally divide to generate two upper-layer neurons. (b) AP2γ is not required in rostral cortical regions for the correct generation of deep-layer neurons or for the correct specification of basal progenitors and upper-layer neurons. (c) In caudal (that is, occipital) cortical regions of the AP2γ mutants, intermediate progenitors are not correctly generated. These progenitors exhibit certain radial glial characteristics such as a process contacting the apical surface and have severely reduced or missing expression of basal progenitor regulators such as Tbr2 and Tis21. Moreover, by later stages (E17), many basal progenitors undergo cell death. These alterations in the mutant basal progenitors probably lead to the generation of fewer upper-layer cortical neurons in the occipital cortex.
provide, to the best of our knowledge for the first time, evidence that areal and laminar identity in the developing cortex may be controlled by a single molecule; for the visual cortex, this appears to be AP2γ. One major question that remains to be resolved relates to the specific mechanism that could restrict AP2γ function to the generation of upper-layer neurons in only the occipital cortex. AP2γ expression itself is not restricted to the caudal portions of the cortical ventricular zone; rather, Pinto et al.4 found it throughout the cortical ventricular zone. Moreover, at least at early stages, AP2γ appeared to be expressed at slightly lower levels in the caudal versus rostral cortex. It seems possible, therefore, that AP2γ interacts with one or several other factors that would be enriched in or restricted to caudal portions of the cortical ventricular zone. The transcription factor Emx2 is a potential candidate, as it has a high caudal–to–low rostral gradient of expression in the developing cortical ventricular zone2. Furthermore, Emx2 is required for correct area patterning of the visual cortex2. Thus, one can envision a model in which a specific threshold for AP2γ and Emx2 expression in radial glial progenitors would specify the correct production of basal progenitors in the developing occipital cortex. Unfortunately, Emx2 mutants do not survive after birth2 and whether they have occipital cortex laminar defects that might resemble those seen in the AP2γ mutants remains unknown. Alternatively, the region-specific requirements for AP2γ in the forming visual cortex could be the result of a local lack of compensation by another factor. The paired homeobox gene
1212
Pax6 is required for the correct specification of rostral cortical areas and developmental markers of caudal cortical areas expand rostrally in its absence2. Accordingly, Pax6 is expressed in a decreasing rostral to caudal gradient in the developing cortex2. Pinto et al.4 found that Pax6 directly promotes the expression of both AP2γ and the basal progenitor regulator Tbr2. Thus, the higher levels of Pax6 in rostral cortical regions could compensate for the lack of AP2γ in the specification and/or generation of basal progenitors. Pinto et al.4 also found that AP2γ is required to specify basal progenitors only from E14 onward, when the generation of upper-layer neurons’ from basal progenitors begins3. Accordingly, viral overexpression of AP2γ at E14 increased the numbers of basal progenitors and, subsequently, neurons in visual cortex layer II/III4, whereas similar overexpression initiated at E12 had no effect on basal progenitors in either rostral or caudal cortex. These results are consistent with a previous study13 that found restricted developmental potential of late-stage (that is, E15 and onward) cortical progenitors compared with mid-stage (E12) progenitors, with the restriction being dependent on cellintrinsic (that is, transcription) factors13. Thus, in addition to region-specific restrictions on AP2γ’s function in the specification of basal progenitors, temporal constraints are also active in this process. It may be, however, that temporal restrictions of developmental potential are not a region-specific phenomenon, but rather a telencephalon-wide one. Recently, we found that even when dorsal telencephalic (cortical) progenitors are ventralized at later stages of
development, they remain limited to generating late, ventral telencephalic, neuronal fates14. The laminar defects observed in the AP2γ mutant visual cortex lead to impairments in visual acuity4. Moreover, the adult AP2γ mutant visual cortex had increased plasticity, suggesting a degree of immaturity. Thus, correct neuronal density in layers II/III may be required for full functional maturation of the visual cortex. Pinto et al.4 also found that AP2γ was expressed in the germinal zone of the developing primate visual cortex, suggesting that it might be involved in the generation of the particularly large numbers of upper-layer neurons that characterize the visual cortex of higher mammals1. Future studies will help to uncover whether the coupling of area- and lamina-specific development by one transcription factor is unique to the visual cortex or whether analogous factors perform similar functions in other areas of the developing cortex. 1. Dehay, C. & Kennedy, H. Nat. Rev. Neurosci. 8, 438–450 (2007). 2. O’Leary, D.D.M., Chou, S.-J. & Sahara, S. Neuron 56, 252–269 (2007). 3. Molyneaux, B.J., Arlotta, P., Menezes, J.R.L. & Macklis, J.D. Nat. Rev. Neurosci. 8, 427–437 (2007). 4. Pinto, L. et al. Nat. Neurosci. 12, 1229–1237 (2009). 5. Noctor, S.C. Martinez-Cerdeño, V., Ivic, L. & Kriegstein, A.R. Nat. Neurosci. 7, 136–144 (2004). 6. Haubensak, W., Attardo, A., Denk, W. & Huttner, W.B. Proc. Natl. Acad. Sci. USA 101, 3196–3201 (2004). 7. Zimmer, C., Tiveron, M.C., Bodmer, R. & Cremer, H. Cereb. Cortex 14, 1408–1420 (2004). 8. Englund, C. et al. J. Neurosci. 25, 247–251 (2005). 9. Arnold, S.J. et al. Genes Dev. 22, 2479–2484 (2008). 10. Sessa, A. et al. Neuron 60, 56–69 (2008). 11. Farkas, L.M. et al. Neuron 60, 40–55 (2008). 12. Cubelos, B. et al. Cereb. Cortex 18, 1758–1770 (2008). 13. Shen, Q. et al. Nat. Neurosci. 9, 743–751 (2006). 14. Waclaw, R.R., Wang, B., Pei, Z., Ehrman, L.A. & Campbell, K. Neuron 63, 451–465 (2009).
volume 12 | number 10 | OCTOBER 2009 nature neuroscience
news and views
Calcium: an insignificant thing Christian Frøkjær-Jensen & Erik M Jorgensen
© 2009 Nature America, Inc. All rights reserved.
Fusion of synaptic vesicles upon calcium influx requires precise localization of voltage-gated calcium channels. A new study identifies a previously uncharacterized protein that mediates trafficking of CaV2 calcium channels in C. elegans. In the presynaptic terminal, a puff of calcium is an insignificant thing, a scintilla painted on the dark ceiling of the synaptic bouton. First visualized in 1992 in the synaptic terminals of the squid, these intracellular calcium increases are transient and highly local, confined to microdomains1. The portals for extracellular calcium are voltage-gated calcium channels, usually of the CaV2 class, clustered at the active zone. Depolarization of the membrane opens the pore and a surge of calcium, reaching concentrations of 100 µM, flows into the cell2. However, this rise in calcium probably only extends 20 nm or so before dissipating; calcium diffusion is limited by the action of internal buffers that are very fast acting3,4. The calcium sensor involved in the fusion of synaptic vesicles with the membrane has a low affinity for calcium; it requires every bit of that 100 µM for effective release of neurotransmitter5. If the calcium channel is not near the synaptic vesicle, then there will be no neurotransmission. So, where are the channels? Who docks them there? Who pilots the tug? There must be escorts that regulate the synthesis, transport and localization of voltage-gated calcium channels to these sites. In this issue, Saheki and Bargmann6 labeled the calcium channels with green fluorescent protein (GFP) and localized them to nematode synapses. They then used a simple in vivo visual genetic screen to identify the proteins that were required to transport and localize calcium channels to presynaptic sites in C. elegans and proposed a mechanism of calcium-channel trafficking. There is a long and difficult history for studies of calcium-channel localization. For example, one study used a combination of electrophysiology and electron microscopy2, finding that there are approximately 1,800 calcium channels in 20 discrete clusters on isolated hair cells. The study also estimated
The authors are in the Department of Biology, University of Utah, Howard Hughes Medical Institute, Salt Lake City, Utah. Christian FrøkjærJensen is also at the Danish National Research Foundation Center for Cardiac Arrhythmia, Department of Biomedical Sciences, University of Copenhagen, Denmark. e-mail:
[email protected]
a
Chaperone Folding
CaV2 α1
Unfolded CaV2 CALF-1
b
ER
Assembly
Assembly checkpoint
CaV2 α2/δ
Cytosol
GO! Assembly
CaV2 α1
c
CaV2 α2/δ
CALF-1
ER transport vesicle formation
R X R
Coat proteins
R
X
R
Figure 1 CaV2 calcium channel trafficking in C.elegans. Saheki and Bargmann6 identified two proteins that are necessary for CaV2 channel transport and localization to presynaptic sites: CALF-1 and the α2δ subunit UNC-36. CaV2 channels are retained in the endoplasmic reticulum (ER) in the absence of either CALF-1 or UNC-36. There are at least three possible models for the function of CALF-1. (a) CALF-1 permanently resides in the endoplasmatic reticulum and acts as a specific chaperone for CaV2. Chaperone functions include channel folding and subunit assembly. (b) CALF-1 serves as an endoplasmatic reticulum checkpoint that monitors the assembly of CaV2 channels. At the checkpoint, only assembled channels are allowed to exit the endoplasmatic reticulum. Although not shown here, it is possible that a fully assembled CaV2 channel occludes the RXR motif and that a CALF-1/CaV2 complex is incorporated into the transport vesicle. (c) CALF-1 stimulates CaV2 channel export from the endoplasmatic reticulum by concentrating CaV2 channels at endoplasmatic reticulum export sites and by recruiting coat proteins necessary to form transport vesicles. In this model, the main function of the RXR motif is to retrieve CALF-1 to the endoplasmatic reticulum.
an almost identical number of ion channels on the basis of freeze-fracture electron microscopy and serial-section transmission electron microscopy; from these results, the authors concluded that calcium channels are positioned within 100 nm of the presynaptic
nature neuroscience volume 12 | number 10 | OCTOBER 2009
active zone. However, these are very difficult experiments and, in the end, indirect. For studies of the mechanism and dynamics of CaV channel trafficking, it would be nice to just be able to see the channels directly in living cells.
1213
© 2009 Nature America, Inc. All rights reserved.
news and views To visualize calcium channel localization, Saheki and Bargmann6 tagged a functional CaV2 channel with GFP and expressed it in a pair of neurons that make synapses along their axons in stereotyped positions. In these experiments, the tagged calcium channel specifically localizes to presynaptic active zones. Notably, this pattern can be observed in living worms by epifluorescence. Using this, the authors were able to screen for mutants in which CaV2 channel localization is disrupted. It is not a particularly easy screen, as worm screens go; it requires that every worm be mounted on a fluorescence microscope and scored for mislocalization. Nevertheless, hard screens can pay off, and the authors isolated mutants with mislocalized CaV2 channels and identified two proteins that are necessary for correct CaV2 transport: a previously unknown protein, calcium channel localization factor 1 (CALF-1), and an α2δ subunit. CALF-1 is a small protein, composed of a single transmembrane domain and a cytosolic tail, that resides in the endoplasmatic reticulum. Saheki and Bargmann6 found that the primary function of CALF-1 is in calcium-channel biogenesis; in the absence of CALF-1, CaV2 channels were retained in the endoplasmatic reticulum, whereas other active zone and synaptic vesicle components were properly localized. Endoplasmatic reticulum retention is not a developmental defect, as expression of CALF-1 in calf-1 mutant adults promoted rapid exit of f unctional CaV2 channels from the endoplasmatic reticulum and transport to synaptic sites. How does CALF-1 promote CaV2 exit from the endoplasmatic reticulum? For most ion channels, endoplasmatic reticulum retention motifs are contained in the channels themselves. After channel assembly and maturation, outfitter proteins mask the retention signal and allow channels to exit the endoplasmatic reticulum7. In this case, however, it is not the CaV2 channel itself, but the accessory protein CALF-1, that has the endoplasmatic reticulum retention motif; the cytosolic tail of CALF-1 contains multiple arginine-x-arginine (RXR) endoplasmatic reticulum retention motifs embedded in basic and proline-rich regions. In their genetic screen, Saheki and Bargmann6 also isolated new mutant alleles of the α2δ subunit UNC-36. The α2δ subunit appears to have related CaV2 trafficking functions to CALF-1. α2δ subunits are accessory subunits to CaV channels that, in mammalian systems at least, increase the number of functional CaV channels in the cell membrane7. α2δ subunits are mainly extracellular, with the 1214
α2 subunits being tethered to the extracellular face of the membrane by the δ subunit. unc-36 mutants are uncoordinated, similar to CaV2 mutants, and GFP-tagged CaV2 is no longer detectable at presynaptic sites. Is UNC-36 mainly involved in trafficking or does it also have a functional role? One experiment in particular demonstrated that α2δ has a functional role in nematodes. In α2δ mutants, overexpression of the CALF-1 protein partially restored CaV2 channel localization to synapses. However, locomotion was not restored, arguing for a role of α2δ in both channel function and trafficking. These results are consistent with data from mammalian and Drosophila studies, although the effects in C. elegans are more severe. In mammalian cell culture, α2δ promotes CaV channel surface expression and alters subtle functional properties of calcium currents7. In flies, the α2δ mutant straightjacket has reduced neuronal transmission as a result of a reduction in CaV2 channels at the synapse8,9. These studies underscore an important point, that α2δ proteins are bona fide subunits of the calcium channel complex and assembly of these subunits is likely to be permissive for trafficking, whereas CALF-1 is more likely to be specifically involved in trafficking the complex. From these results, Saheki and Bargmann6 propose that the α2δ subunits and CALF-1 promote exit from the endoplasmatic reticulum. As an underlying mechanism, three possible processes come to mind: folding, a checkpoint for assembly and formation of transport vesicles (Fig. 1)10. In the first possibility, CALF-1 functions as a chaperone for protein folding or promotes assembly of the subunits of the calcium channel complex. Failure to assemble the complex blocks these proteins from exiting the endoplasmatic reticulum. In the second possibility, CALF-1 functions as a checkpoint protein, similar to a licensing factor, that allows the fully assembled complex to exit. In the third possibility, CALF-1 interacts with the calcium channel at the endoplasmatic reticulum exit site for the formation of transport vesicles – for example in the recruitment of coat proteins. The authors do not favor a particular mechanism, but they exclude the possibility that the endoplasmatic reticulum retention motif of CALF-1 acts as a specific brake for an unassembled complex. First, loss of CALF-1 or elimination of the endoplasmatic reticulum retention motif did not lead to constitutive exit of the calcium channel. Second, substitution of the cytosolic tail of CALF-1 with the endoplasmatic reticulum retention motif from the adrenergic receptor partially rescued channel trafficking.
Thus, the endoplasmatic reticulum retention motif probably functions to return CALF-1 to the endoplasmatic reticulum rather than being directly involved in calcium channel trafficking. Although CALF-1 does not have any obvious homologs outside of nematodes, the authors noted that gamma subunits of CaV channels in mammals share similarities, such as the RXR motifs and a proline-rich region, with CALF-1. It will be interesting to determine whether mammalian gamma subunits have similar roles in the biogenesis of CaV channels. Saheki and Bargmann’s study brings a number of questions to mind. For example, how do neurons regulate the number of CaV2 channels at synapses? At mammalian synapses, it has been proposed that there are a certain number of ‘slots’ for each type of CaV2 channel11. In Saheki and Bargmann’s study6, calcium channels at individual synapses are visible under conventional fluorescence microscope. Such a bright signal suggests that there are a substantial number of channels per synapse; however, not all of the tagged channels are necessarily inserted into the membrane. Previous experiments suggest that there may be very few calcium channels at synapses in C. elegans. It has been estimated that there are less than two CaV channels per synapse at one type of sensory neuron12. If quantitative studies bear these numbers out, calcium channels really do look like an insignificant component of the active zone, at least numerically speaking. However, Napoleon once said, “There are times when the most insignificant thing can decide the outcome of a battle.” It is possible that the placement of just a single channel determines whether a particular synapse fires or remains silent. The tiny puff of calcium from a channel is not to be dismissed lightly, as all of neurotransmission hinges on its function. We are now closer to understanding how that speck positioned itself to become so important. 1. Llinás, R., Sugimori, M. & Silver, R.B. Science 256, 677–679 (1992). 2. Roberts, W.M., Jacobs, R.A. & Hudspeth, A.J. J. Neurosci. 10, 3664–3684 (1990). 3. Yamada, W.M. & Zucker, R. Biophys. J. 61, 671–682 (1992). 4. Neher, E. Neuron 20, 389–399 (1998). 5. Heidelberger, R. et al. Nature 371, 513–515 (1994). 6. Saheki, Y. & Bargmann, C.I. Nat. Neurosci. 12, 1257–1265 (2009). 7. Jarvis, S.E. & Zamponi, G.W. Curr. Opin. Cell Biol. 19, 474–482 (2007). 8. Ly, C.V. et al. J. Cell Biol. 181, 157–170 (2008). 9. Dickman, D.K., Kurshan, P.T. & Schwarz, T.L. J. Neurosci. 28, 31–38 (2008). 10. Herrmann, J.M., Malkus, P. & Schekman, R. Trends Cell Biol. 9, 5–7 (1999). 11. Cao, Y.Q. et al. Neuron 43, 387–400 (2004). 12. Goodman, M.B. et al. Neuron 20, 763–772 (1998).
volume 12 | number 10 | OCTOBER 2009 nature neuroscience
news and views
TR(I)Pping towards treatment for ischemia David A Rempe, Takahiro Takano & Maiken Nedergaard
Stroke is the third leading cause of death in the US and is a leading cause of disability. Despite substantial progress in delineating the multiple ischemia-induced molecular cascades that contribute to neuronal death, there are still no effective stroke treatments. The contribution of excitotoxicity, mediated by glutamatergic NMDA receptors, to ischemia-induced cell death is well appreciated. However, because these receptors are widely expressed and are important for diverse neuronal functions, it is not surprising that blocking these receptors with antagonists is not well tolerated, especially in confused, elderly stroke patients. Moreover, the efficacy of NMDA receptor antagonists is restricted to a very short time window following the onset of ischemia. The discovery and subsequent targeting of mechanisms with better tolerability and longer treatment windows are desperately needed for the effective treatment of stroke. A study in this issue shows that inhibiting transient receptor potential melastatin 7 (TRPM7) channel function may fulfill this need. Using viral delivery of small hairpin RNA (shRNA) to knockdown TRPM7 channels in CA1 pyramidal neurons, Sun et al.1 found that reduced TRPM7 channel expression greatly diminishes pyramidal cell death after global ischemia, maintaining normal CA1 neuronal function and hippocampaldependent behaviors. TRPM7 channels are members of a large family of TRPM nonselective cation channels that are expressed in neurons and have diverse functions ranging from temperature sensing, taste sensation, magnesium homeostasis and cation sensing2. TRPM7 channels open when extracellular Ca2+ and/or Mg2+ concentrations decline, as is the case during ischemia (Fig. 1). Because TRPM7 channels conduct Ca2+ and open during ischemia, they could be involved in Ca2+ overload and consequent cell death during ischemia. An earlier study demonstrated that knockdown of TRPM7 markedly reduced
David A. Rempe is in the Department of Neurology, and Takahiro Takano and Maiken Nedergaard are in the Department of Neurosurgery, University of Rochester, Rochester, New York, USA. e-mail:
[email protected] or
[email protected]
neuronal cell death induced by oxygen-glucose deprivation (OGD) an in vitro model of ischemia3. This protection was observed in both the absence and presence of glutamatergic antagonists, demonstrating that TRPM7 channels mediate cell death independent of excitotoxicity during the ischemic cascade. Moreover, knockdown of TRPM7 channel function protected neurons against OGD for a longer period of time than glutamatergic antagonists, suggesting that they may have a longer treatment window compared with therapies targeting excitotoxicity alone. Given the success of targeting TRPM7 channels in preventing OGD-induced neuronal death in vitro, the next logical step was to examine the effects of TRPM7 channels on ischemia-induced neuronal death in vivo. Because knockout of TRPM7 function is lethal during embryogenesis and selective pharmacological blockers of TRPM7 channels do not exist, Sun et al.1 used AAV-1 to deliver shRNA to hippocampal CA1 pyramidal cells to knockdown TRPM7 function. As AAV-1 is neurotropic and CA1 pyramidal cells are especially vulnerable to global ischemia, this approach provides a means for examining TRPM7 function in a specific neuronal population that is particularly susceptible to ischemia-induced cell death. As in the in vitro experiments3, knockdown of TRPM7 in vivo markedly reduced CA1 pyramidal cell death following global ischemia. Notably, not only was neuronal death attenuated, but the neurons were functional when tested by electrophysiology 30 days later. Whole-cell recording of neurons lacking TRPM7 protein showed normal spike-frequency adaptation, paired-pulse facilitation and excitatory postsynaptic potentials, similar to that seen in nonischemic controls. Inhibitory postsynaptic potentials were reduced in the CA1 pyramidal cells that lacked TRPM7, but it was not clear whether this was a result of ischemia or the knockdown of TRPM7 channels. Long-term potentiation (LTP) was evoked in hippocampal slices from ischemic mice with knockdown of TRPM7, whereas ischemic controls did not maintain LTP. Thus, the preserved CA1 pyramidal cells maintained the ability to express LTP and knockdown of TRPM7 channel function did not preclude the induction of LTP. However, a subtle effect of TRPM7 loss on LTP induction
nature neuroscience volume 12 | number 10 | OCTOBER 2009
a Ca2+
[Ca2+] [Mg2+]
TRPM2
TRPM7
Ca2+
ADPR NAD+
ONOO– O2
NO NOS
Blood vessel
Mitochondria
NMDAR
TRPM4
b Microglia
Katie Vicari
© 2009 Nature America, Inc. All rights reserved.
A study in this issue found that suppressing expression of TRPM7 in hippocampal CA1 neurons conferred resistance to ischemic cell death, preserved cell function and prevented ischemia-induced deficits in memory.
Astrocyte
Figure 1 The expression and activation mechanisms of TRPM channels. (a) Neurons (blue) express TRPM2 (purple) and TRPM7 (green) channels. Reduced extracellular Ca2+ and Mg2+, which occurs during ischemia, activate TRPM7 channels, which conduct Ca2+ and other cations. TRPM2 channels are activated by ADPribose (ADPR), which is also induced during ischemia. NMDAR, NMDA receptor; NOS, nitric oxide synthase. (b) TRPM channels are expressed by other brain cells, including microglia (light green), astrocytes (beige) and vascular smooth muscle cells (red). Although activation of TRPM7 channels in neurons induces cell death during ischemia, the role of the other channels, including TRMP4 (light blue), during ischemia remains largely undefined.
and maintenance cannot be ruled out, as the effect of TRPM7 inhibition on LTP was not evaluated in nonischemic tissue. Finally, the preserved function of shRNA-treated neurons after ischemia was illustrated by preserved memory function; shRNA-treated rats subjected to ischemia performed equally well as nonischemic controls in the Morris water maze and contextual fear-conditioning tests. Rats subjected to ischemia that were not treated with shRNA or were treated with a scrambled probe were impaired in these tests. Thus, knockdown of TRPM7 function greatly reduced global ischemia–induced CA1 pyramidal cell death
1215
© 2009 Nature America, Inc. All rights reserved.
news and views and preserved function of those neurons, as demonstrated by electrophysiological and behavioral assays. Although selective shRNA-mediated knockdown of TRPM7 channel function in pyramidal neurons protects against the effects of global ischemia, it is unclear whether neuroprotection will be attainable by pharmacologically targeting TRPM7 channels. Because TRPM7 channels are expressed by non-neuronal cells, it is likely that selective pharmacological inhibition of TRPM7 channels, if achieved, will have more widespread effects. This widespread inhibition of TRPM7 channels may further augment or impede neuroprotection. For example, TRPM7 channel number is increased at the cell surface of vascular smooth muscle cells by changes in blood flow4. As TRPM7 channels are Ca2+ permeable, this could lead to contraction of the blood vessel, diminishing blood flow. Therefore, inhibition of TRPM7 channel function in the vascular smooth muscle may relieve this constriction and augment neuroprotection by increasing cerebral blood flow. In fact, activation of another member of the TRPM channel family, TRPM4, induces constriction of the cerebral vasculature5. Moreover, activation of TRPM4 channels in vascular smooth muscle during spinal cord injury induces cell death of vascular cells and contributes to injury following spinal cord damage6. Thus, pharmacological inhibition of TRPM7, TRPM4 and possibly other TRPM receptors could have the added benefit of improving cerebral blood flow and maintaining vascular integrity during ischemia. Inhibition of TRPM7 could also have adverse effects; although Sun et al.1 demonstrate that knockdown of TRPM7 does not substantially alter the electrophysiological properties of CA1 pyramidal cells, previous studies suggest that TRPM7 channels are important in cholinergic neurotransmission7, and blockage of TRPM7 could have other unknown effects on synaptic transmission in the central or peripheral nervous system. Furthermore, in contrast with vascular smooth muscle, the specific loss of TRPM7 function in the T-cell lineage disrupts thymopoiesis8. In some other cell types, loss of TRPM7 function disrupts Mg2+ homeostasis, causing cell death9. Given that TRPM7 transcript is expressed in multiple tissues, including brain, kidney, heart, liver, spleen, skeletal muscle and testis10, it is possible that temporary, systemic block of TRPM7 channel permeability could
1216
produce unacceptable side effects, including the diminished viability of non-neuronal cell types. Thus, some measure of organ or celltype specificity may need to be involved in therapeutic targeting of TRPM7. Sun et al.’s1 results lead one to question whether other TRPM isoforms also contribute to neuronal death and if their relative contribution might vary across brain regions. The protein distribution of the different TRPM channel isoforms is not well defined in the brain. Similar to TRPM7 and TRPM4, the TRPM2 transcript is widely expressed in the brain. Also similar to TRPM7, gating of TRPM2 is regulated by molecular processes that are prominent during ischemia and reperfusion10 (Fig. 1). For example, reactive oxygen species increase TRPM7 channel conductance. In addition, activation of the nuclear enzyme poly-adenosine diphosphate polymerase (PARP-1) phosphorylates ADPribose, which opens the TRPM2 channel. As PARP-1 is highly activated during ischemia, PARP-1 probably contributes to TRPM2 channel activation during stroke. Although the role of TRPM2 in ischemic cell death is unknown, activation of TRPM2 channels induces death of striatal neurons (which are damaged in both global and focal ischemia) via oxidative stress11. Functional TRPM2 channels have also been identified in hippocampal pyramidal cells12. It will be interesting to determine whether TRPM2 and TRPM7 channels have similar effects on ischemiainduced neuronal death and in which brain regions they are most prominent. Moreover, beyond neurons, multiple other cell types contribute to the pathology of brain ischemia, including microglia and astrocytes. TRPM2 channels are functional in microglia13, and oxidative stress induces an increase in TRPM2 channels in astrocytes14. The effect of TRPM2 channels on microglia and astrocyte function is unknown; thus, the potential effects of glial TRPM2 activation during stroke are unknown and represent an important future avenue of investigation. The role of TRPM channels in glial function will be particularly important when considering potential pharmacological inhibitors of TRPM channels. Will targeting TRPM channels provide a treatment window for ischemic stroke extended beyond that of NMDA receptor antagonists? TRPM7 channel gating is integrally linked to NMDA channel activation (Fig. 1). NMDA channel activity induces production of nitric oxide and reactive
itrogen species during ischemia. The reactive n nitrogen species, in turn, activate TRPM7. NMDA channel activity also activates TRPM2 channels through generation of ADP-ribose. Because TRPM2 and TRPM7 channels are activated by processes downstream of NMDA channels, targeting TRPM channels may extend the therapeutic window for ischemic stroke. Neurons in which TRPM7 channels have been knocked down can withstand OGD for 3 hours, compared to only 1 hour with antagonism of glutamatergic receptors alone3. This observation suggests that the therapeutic w indow is longer for TRPM channel antagonism than for glutamatergic channel antagonism. However, the development of selective pharmacological agents to inhibit TRPM7 or TRPM2 function will be necessary to determine the therapeutic window of antagonizing TRPM channels in vivo. Beyond TRPM channels, other channels that dysfunction during ischemia, such as volume-regulated anion channels, gap junctions/ hemichannels, sodium/calcium exchanger and acid-sensing channels are also targets that hold promise for stroke treatment15. In particular, antagonism of acid-sensing channels reduces stroke-related damage when administered up to 5 h after the onset of ischemia15. Nonetheless, Sun et al.1 show the importance of antagonizing TRMP7 channel function for preventing cell death and maintaining CA1 pyramidal cell function during global ischemia in vivo. This approach demonstrates the encouraging potential of targeting nonglutamatergic molecular processes as new neuroprotective therapies for stroke. Demonstrations of the efficacy of selective inhibitors of the different TRPM family members will hopefully be forthcoming. Sun, H.-S. et al. Nat. Neurosci. 12, 1300–1307 (2009). Nilius, B. & Voets, T. Pflugers Arch. 451, 1–10 (2005). Aarts, M. et al. Cell 115, 863–877 (2003). Oancea, E., Wolfe, J.T. & Clapham, D.E. Circ. Res. 98, 245–253 (2006). 5. Reading, S.A. & Brayden, J.E. Stroke 38, 2322–2328 (2007). 6. Gerzanich, V. et al. Nat. Med. 15, 185–191 (2009). 7. Krapivinsky, G. et al. Neuron 52, 485–496 (2006). 8. Jin, J. et al. Science 322, 756–760 (2008). 9. Schmitz, C. et al. Cell 114, 191–200 (2003). 10. McNulty, S. & Fonfria, E. Pflugers Arch. 451, 235–242 (2005). 11. Fonfria, E. et al. J. Neurochem. 95, 715–723 (2005). 12. Olah, M.E. et al. J. Physiol. (Lond.) 587, 965–979 (2009). 13. Kraft, R. et al. Am. J. Physiol. Cell Physiol. 286, C129–C137 (2004). 14. Bond, C.E. & Greenfield, S.A. Glia 55, 1348–1361 (2007). 15. Besancon, E. et al. Trends Pharmacol. Sci. 29, 268–275 (2008).
1. 2. 3. 4.
volume 12 | number 10 | OCTOBER 2009 nature neuroscience
news and views
Six degrees of separation: the amygdala regulates social behavior and perception Rebecca M Todd & Adam K Anderson
Traditionally, the amygdala has gotten a lot of ‘bad’ press. Popular wisdom has portrayed the human amygdala as the center of an ancient animal id that drives us to rapid impulsive action before our more reasoned judgments can kick in. For a long time, it was considered to be a fear center or threat detector that is instrumental in allocating processing resources to potentially harmful events. This was in part because, thanks to research in nonhuman animals, the amygdala’s role in fear learning was extremely well mapped. More recent studies in humans suggest that it is responsive to positive and arousing rather than to strictly negative events, as well as to ambiguous events1,2. In this issue, two case studies of an individual with bilateral amygdala damage indicate that ideas about amygdala function may need even further reconsideration. The connectivity of the amygdala places it at the center of the brain, a physical hub linking numerous distant regions, and it is positioned to allow emotions to influence how the rest of the brain works, from the first stages of stimulus encoding to regulating social behavior. Adolphs and colleagues examined these two functions and found that the amygdala may be important for regulating social distance3 and influencing slower, explicit responses, as opposed to rapid automatic alerting to social signals of threat4. Although the amygdala has been studied in numerous ways, from molecular manipulations in mice all the way up to functional imaging in humans, patient SM’s complete bilateral amygdala injury represents a unique opportunity to causally link the function of this well-studied structure to human behavior. Adolphs and colleagues took advantage of this opportunity to test both a previously unknown function of the amygdala and a well-established one. Kennedy et al.3 studied
The authors are in the Department of Psychology, University of Toronto, and the Rotman Research Institute, Toronto, Ontario, Canada. e-mail:
[email protected] or
[email protected]
Marina Corral
© 2009 Nature America, Inc. All rights reserved.
Recent human imaging work has expanded the view of amygdala function beyond early findings in animals, but two studies of an individual with bilateral amygdala damage now suggest that we should be thinking even more broadly.
Figure 1 A new study by Kennedy et al.3 suggests that the human amygdala may be crucial for those feelings of discomfort with close physical proximity that help maintain appropriate social distances.
the amygdala’s role in regulating interpersonal distance. People automatically regulate the distance between themselves and others on the basis of feelings of personal comfort. Crowding unnecessarily close to a stranger in an uncrowded subway car or to a colleague at a meeting feels prohibitively uncomfortable and this reaction may serve as a powerful repulsive force in adjusting interpersonal distances (Fig. 1). In a series of elegant experiments, the authors demonstrate that SM fails to show evidence of these invisible social force fields that regulate close physical proximity, suggesting that the amygdala is crucial for the sense of interpersonal space. In this study, SM was asked to face an experimenter at the interpersonal distance at which she felt most comfortable or to rate how comfortable she felt (ranging from perfectly comfortable to extremely uncomfortable) while standing at different distances from an experimenter. SM’s preferred distance was consistently smaller than that of control subjects and she claimed to be comfortable
nature neuroscience volume 12 | number 10 | OCTOBER 2009
even when standing nose to nose with the experimenter. This effect was robust in a number of variations of the procedure that controlled for alternate explanations of the main finding, including familiarity with the experimenter, the gender of the experimenter and the presence of eye contact during the experiment. Notably, SM reported that she was perfectly aware that other people had a sense of interpersonal space and she did not. She did not differ from the controls at the level of rational explanation; she simply did not feel the same discomfort with proximity that they did. On the basis of their findings with SM, Kennedy et al.3 predicted that the amygdala should be more active at close interpersonal distances in normal subjects. To test this prediction, they measured amygdala activation in control subjects using functional magnetic resonance imaging in response to reports that an experimenter was nearby or far away. As predicted, subjects showed greater amygdala activation at close distances. These data,
1217
© 2009 Nature America, Inc. All rights reserved.
news and views together with SM’s lack of aversion to close physical proximity, provide evidence that the amygdala is crucial for the sense of interpersonal space and may mediate emotional responses to personal space violation. In addition to directly mediating the sense of discomfort, the importance of the amygdala in regulating social distance may also reflect emotional learning of social conventions. Parents socialize their children by using reward and punishment to teach them the emotional importance of social norms. Indeed, SM’s disregard for interpersonal distance despite knowledge of the typical rules of social engagement parallels the established role of the amygdala in fear-conditioned responses, where autonomic responses are abolished, but factual knowledge of events that predict unpleasant events remains intact5. Moreover, there are different rules for social proximity for different relationships (parents or lovers versus strangers) and these vary across cultures. As these rules are culturally learned, amygdala responses to specific social stimuli in specific contexts may arise from developmental experience. Recent research has shown that the amygdala responds preferentially to different stimuli depending on their social context6. As the amygdala is an important hub in networks implicated in social learning7, socialization may partly be a process of tuning amygdala responses during development. Another potential account of the finding that SM has no instinctive discomfort at close social distances is that she lacks a rapid amygdala response that directs attention to socially and emotionally salient events. In support of this explanation, there is evidence that fearful faces, which are thought to signal danger in the environment, are processed more quickly than other facial expressions and activate the amygdala more than neutral faces, even when presented subliminally8. This evidence has often been interpreted in terms of an evolutionary advantage for responding to threatening events faster than the speed of thought. There is also conflicting evidence that awareness is required for the amygdala to respond to fearful faces9. A study of SM by Tsuchiya et al.4 challenges the notion that the amygdala is responsible for rapid processing of fearful faces by showing that SM has a normal capacity for detecting fearful faces when they are presented rapidly or at the threshold of awareness. SM was shown images flashed on a screen (for 40 ms), with a fearful face on one side and a neutral face on the other. Although she showed the same degree of speed and accuracy as controls, her intensity ratings of fearful expressions were much lower, suggesting that, for her, they lacked their intense
1218
f earful quality. Two additional experiments employing visual search and interocular suppression further supported the notion that SM demonstrated enhanced awareness of fearful facial expressions. From these results, Tsuchiya et al.4 conclude that the amygdala is not necessary for rapid implicit detection of fearful facial expressions. Instead, they propose that it contributes to slower processes involved in explicit recognition of facial expression in particular and in social judgment in general. Although Kennedy et al.3 and Tsuchiya et al.4 propose a seductively simple account of amygdala function, the results from these two case studies paint a complex picture of intact and spared capacities following amygdala damage. Although amygdala damage leaves explicit knowledge of interpersonal distance norms intact, it impairs the potentially implicit and automatic use of these norms in regulating behavior. Conversely, although amygdala damage impairs direct explicit knowledge about the fear quality of faces, it leaves their rapid and relatively automatic processing intact. Whether the amygdala is necessary for the rapid perceptual processing of social and emotional salience in general, rather than for fear faces in particular, may still be an open question. First, the finding that SM has a normal ability to more rapidly detect fear faces may not generalize to more intense emotional stimuli. This may be because fearful faces in experimental contexts fail to elicit sufficient autonomic arousal10, which has been shown to be a critical dimension in accounting for emotional salience. Indeed, amygdala lesions reduce autonomic arousal responses to subliminally presented emotionally arousing images11. Amygdala lesions also impair the enhanced awareness associated with arousing stimuli when attentional resources are limited, but leave intact the enhanced awareness associated with visually distinctive events12. These results fit with nonhuman animal studies demonstrating that amygdala stimulation results in altered cortical arousal13. Given that amygdala damage leaves enhanced awareness associated with visually distinctive events intact12, the rapid detection of fear faces may be independent of the amygdala because the fear face advantage is related to low-level visual characteristics of fearful faces. Fear faces are characterized by the exaggerated whites of the eyes, and this information is extracted in the early stages of visual processing of faces14. Thus, more rapid detection of fearful faces may be a result of particularities of visual features: their visual rather than motivational salience. Tsuchiya et al.4 go to substantial lengths to account for these potential contributions using face-morphing procedures that control
the global physical distance between neutral and fear faces, but this process may not equate physical differences in the especially visually salient region around the eyes, particularly the whites of the eyes. As such, it remains possible that, although SM’s difficulties in explicit fear recognition stem from a failure to use emotional information from the eyes15, these features may still serve as a source of physical salience in the visual cortex that can be used in the absence of rapid amygdala modulatory influences on perceptual encoding12. Thus, it might be too early to discount the early modulation account. It is important to answer whether social and emotional influences on stimulus encoding reflect predominantly early feedforward inputs from the amygdala or later re-entrant influences onto the amygdala from the cortex and back again. However, the studies appearing in this issue highlight the idea that the broader salience of social and emotional events, and the amygdala’s role therein, is not confined to those first fleeting moments of stimulus encoding. Instead, it may extend to time scales across multiple orders of magnitude, from stimulus consolidation in the present moment to the differential consolidation of these moments into our neural and social networks. The old image of the amygdala as an automatic threat detector may come to be replaced with a picture of the amygdala as a hub of distributed networks that mediate rapid and extended responses to the emotional salience of people, objects and events. Furthermore, just as it is physically central in the brain, the amygdala may serve as a hub of social networks, influencing the literal degrees of separation between ourselves and the social world around us. 1. Anderson, A.K. et al. Nat. Neurosci. 6, 196–202 (2003). 2. Sergerie, K., Chochol, C. & Armony, J.L. Neurosci. Biobehav. Rev. 32, 811–830 (2008). 3. Kennedy, D.P., Gläscher, J., Tyszka, J.M. & Adolphs, R. Nat. Neurosci. 12, 1226–1227 (2009). 4. Tsuchiya, N., Moradi, F., Felsen, C., Yamazaki, M. & Adolphs, R. Nat. Neurosci. 12, 1224–1225 (2009). 5. Bechara, A. et al. Science 269, 1115–1118 (1995). 6. Van Bavel, J.J., Packer, D.J. & Cunningham, W.A. Psychol. Sci. 19, 1131–1139 (2008). 7. Davis, F.C. et al. Cereb. Cortex published online, doi:10.1093/cercor/bhp126 (25 June 2009). 8. Whalen, P.J. et al. J. Neurosci. 18, 411–418 (1998). 9. Pessoa, L., Kastner, S. & Ungerleider, L.G. Brain Res. Cogn. Brain Res. 15, 31–45 (2002). 10. Anderson, A.K. et al. Learn. Mem. 13, 711–718 (2006). 11. Gläscher, J. & Adolphs, R. J. Neurosci. 23, 10274–10282 (2003). 12. Anderson, A.K. & Phelps, E.A. Nature 411, 305–309 (2001). 13. Kapp, B.S., Supple, W.F. & Whalen, P.J. Behav. Neurosci. 108, 81–93 (1994). 14. Schyns, P.G., Petro, L.S. & Smith, M.L. PLoS One 4, e5625 (2009). 15. Adolphs, R. et al. Nature 433, 68–72 (2005).
volume 12 | number 10 | OCTOBER 2009 nature neuroscience
B RIE F CO M M UNICATIONS
Spike timing–dependent plasticity (STDP) is a ubiquitous Hebbian learning rule1 in which synaptic modification depends on the order of pre- and postsynaptic spiking in time windows of a few tens of milliseconds. If the presynaptic input is active before the postsynaptic spike, then potentiation occurs, as was originally predicted by Hebb2,
a
b
c
d
TA stim
Rec
4 nS 20 mV
200 ms
Advancement/delay (ms)
The method by which local networks in the brain store information from extrinsic afferent inputs is not well understood. We found that the timing of afferent input can bidirectionally control the sign of spike timing–dependent plasticity at local synapses in rat hippocampus. This mechanism provides a means by which temporal information in external input can be encoded in the local matrix of synaptic weights.
4 nS 20 mV
© 2009 Nature America, Inc. All rights reserved.
Jeehyun Kwag & Ole Paulsen
whereas synaptic depression is induced if this order is reversed 3–5. The computational consequences of this local learning rule depends on the architecture and circuit dynamics of the network in which the synapses are embedded. The hippocampus, which has an established role in memory, is an attractive experimental system in which to study such interactions, as both the network architecture and circuit dynamics are well characterized6,7. CA1 pyramidal neurons receive local input via the Schaffer collaterals from CA3 and external input from the entorhinal cortex via perforant path fibers (the ‘direct’ tempero-ammonic pathway)8. During spatial learning, the hippocampal network engages in rhythmic theta activity, during which hippo campal principal neurons receive rhythmic perisomatic inhibition at 4–6 Hz9. To mimic this network state, we subjected individual CA1 pyramidal neurons to a rhythmic inhibitory conductance using dynamic clamp while depolarizing the cell to fire a single action potential at the peak of each theta cycle (Fig. 1a,b; see Supplementary Methods). To test how the external tempero-ammonic input controls spike timing in CA1 pyramidal cells during theta oscillations, we stimulated the tempero-ammonic input at different theta phases and recorded the effects on postsynaptic spike timing. We found that, depending on the timing of the tempero-ammonic input, the spike times
Advancement/delay (ms)
The timing of external input controls the sign of plasticity at local synapses
40
20
30 20 10 0 0 –10 –20 –30
0
–20
–40 –200 –200
–100 Time (ms)
0
–150
–100
–50
0
Time of EPSP (ms)
Figure 1 Tempero-ammonic input controls postsynaptic spike timing of CA1 pyramidal neurons during theta oscillation. (a) Experimental set-up. A CA1 neuron with a recording electrode at the soma (Rec) and an extracellular electrode stimulating tempero-ammonic input (TA stim) is shown. (b) Example voltage traces during theta oscillation produced by conductance clamp (black trace, minimum inhibitory conductance upwards). Without synaptic perturbation, neuron spiked near the peak of oscillation (gray, dashed line). Tempero-ammonic input stimulation on the ascending phase of oscillation (light gray bar) advanced postsynaptic spikes (light gray trace). Tempero-ammonic input stimulation on the descending phase of oscillation (black bar) delayed postsynaptic spikes (black trace). (c) Superimposed voltage traces of postsynaptic spikes (light gray to black bars, time of postsynaptic spike) with tempero-ammonic input stimulation at different times during theta oscillation (black to light gray bars). Note the reversal and time compression of output relative to input (gray scale). (d) Plot of spike time advancement and delay as a function of time of tempero-ammonic stimulation for the cell shown in b and c. Data are mean ± s.d. of ten postsynaptic spike times for each tempero-ammonic stimulation time. Inset, maximum spike time delay (black bar) and advancement (light gray bar) induced by tempero-ammonic stimulation (n = 7). EPSP, excitatory postsynaptic potential.
Department of Physiology, Anatomy and Genetics, University of Oxford, Oxford, UK. Correspondence should be addressed to O.P. (
[email protected]). Received 25 May; accepted 22 July; published online 6 September 2009; doi:10.1038/nn.2388
nature neuroscience VOLUME 12 | NUMBER 10 | october 2009
1219
b r i e f c o mm u n i c at i o n s a
b
d
e
2
5
200
100
1 0
Normalized EPSP slope (%)
400
TA stim SC stim
h
20 Time (min)
30
2 300
j
1
20 ms
200
100
1
0 0
2 10
20 Time (min)
30
40
3
2
1
0 –40
40
2
1
4
–20 0 20 tpost – t pre (ms)
40
–20 0 20 t post – t pre (ms)
40
5
4
3
2
1
0 –40
Figure 2 Tempero-ammonic input modulates the sign of STDP at hippocampal Schaffer collateral–CA1 synapses during theta oscillation. (a) Experimental setup. Extracellular electrodes stimulating tempero-ammonic (TA stim, light gray) and Schaffer collateral input (SC stim, gray). (b) Oscillatory conductance (black, minimum inhibitory conductance upwards) and voltage response (gray) during pre-before-post pairing, Schaffer collateral stimulation is indicated with the gray bar and dashed line. (c) Adding tempero-ammonic stimulation on the ascending phase of oscillation (light gray bar) advanced the postsynaptic spike time (tpost), changing pairing order to post-before-pre pairing. (d) The time course of the normalized Schaffer collateral EPSP slope. The induction protocol in b (arrowhead) resulted in LTP (gray symbols, n = 10) and the protocol in c led to LTD (light gray symbols, n = 8). Data are mean ± s.e.m. Insets, example Schaffer collateral EPSP traces before (1) and after pairing (2). (e) Histogram of mean spike time difference (tpost − tpre) for pairing protocol in b (gray) and c (light gray) for all cells. (f) Experimental setup as in a, with tempero-ammonic stimulation (black) on descending phase of oscillation. (g) Schaffer collateral input (gray bar) and postsynaptic spikes during post-before-pre pairing, Schaffer collateral stimulation is indicated with the gray bar and dashed line. (h) Adding tempero-ammonic stimulation on the descending phase of oscillation (black bar) delayed the postsynaptic spike, changing pairing order to pre-before-post pairing. (i) Time course of Schaffer collateral EPSP slope. The induction protocol in g (arrowhead) resulted in LTD (gray symbols, n = 8) and that in h led to LTP (black symbols, n = 8). (j) Histogram of spike time difference (tpost − tpre) for pairing protocol in g (gray) and h (black) for all cells.
of CA1 pyramidal neurons could be either advanced or delayed. Stimulation of tempero-ammonic input on the ascending phase of the oscillation advanced postsynaptic spike timing (Fig. 1b), whereas stimulation on the descending phase delayed postsynaptic spike timing (Fig. 1b). That is, depending on the timing of the stimulation of tempero-ammonic input, it could either advance or delay the postsynaptic spike time relative to the oscillation (Fig. 1c,d). The maximum spike
a
time advancement and delay that we observed, without altering firing rate, were −25.8 ± 1.8 ms and 16.0 ± 1.5 ms, respectively (mean ± s.e.m., n = 10; Fig. 1d). The spike time delay was more prominent with distal tempero-ammonic input compared with Schaffer collateral input, which could be attributed to the preferential recruitment of GABAB receptor–mediated inhibition by tempero-ammonic stimulation8 (Supplementary Fig. 1).
b
c
Figure 3 Enforcement of potentiation and 10 10 depression by external input during theta oscillation. (a) Schematic of Schaffer collateral 5 5 inputs (black bars) near the peak of theta oscillation and effect of external temperoLTP LTD 0 0 ammonic input on postsynaptic spike time LTP LTD 20 ms (dashed line) and sign of STDP. Without –5 –5 tempero-ammonic input, postsynaptic spikes occurred near the peak (gray line). Advancing –10 –10 the postsynaptic spike (light gray line) by LTP LTD –80 –40 0 40 –60 –40 –20 0 20 40 60 tempero-ammonic input enforced LTD. Delaying t peak – t pre (ms) t post – t pre (ms) the postsynaptic spike (black line) enforced LTP. (b,c) Synaptic weight change against time of Schaffer collateral input relative to theta peak (tpeak − tpre) (b) and relative to the postsynaptic spike (tpost − tpre) (c) without tempero-ammonic input (gray), with tempero-ammonic input advancing the spike (light gray) and with tempero-ammonic input delaying the spike (black). ∆W (×10–3)
© 2009 Nature America, Inc. All rights reserved.
2 10
i
g
Rec
2 mV
20 ms
0
f
2 Number of experiments
c
100 ms
1 300
Number of experiments
Rec
5 nS 20 mV
SC stim
1
2 mV
TA stim
Normalized EPSP slope (%)
400
1220
VOLUME 12 | NUMBER 10 | october 2009 nature neuroscience
© 2009 Nature America, Inc. All rights reserved.
b r i e f c o mm u n i c at i o n s The similarity of the range of spike time advancement and delay to spike time windows reported for STDP at hippocampal synapses 4 prompted us to investigate whether the timing of tempero-ammonic input can control STDP at Schaffer collateral–CA1 synapses. Indeed, we found that, depending on the timing, stimulation of tempero-ammonic input could convert potentiation at the Schaffer collateral–CA1 synapse into depression or vice versa (Fig. 2). After 10 min of stable baseline recording, 200 pairings of presynaptic Schaffer collateral input stimulation ~10 ms before the postsynaptic spike during theta oscillation (pre-before-post pairing; Fig. 2a,b) induced long-term potentiation (LTP, +154.1 ± 4.3%, P < 0.05, n = 10; Fig. 2d). However, prior stimulation of tempero-ammonic input on the ascending phase of oscillation converted synaptic potentiation into depression (−49.6 ± 1.7%, P < 0.05, n = 8; Fig. 2c,d). Conversely, stimulation of presynaptic Schaffer collateral input ~15 ms after the postsynaptic spike (post-before-pre pairing; Fig. 2f,g) induced long-term depression (LTD) in the control condition (−53.9 ± 2.1%, P < 0.05, n = 8; Fig. 2i). However, prior stimulation of tempero-ammonic input on the descending phase of oscillation converted synaptic depression into potentiation (+100.6 ± 3.9%, P < 0.05, n = 8; Fig. 2h,i). The reversal of the sign of plasticity was associated with a reversal of the order of pre- and postsynaptic spike times during pairing (Fig. 2e,j). Thus, stimulation of tempero-ammonic input on the ascending phase of oscillation advanced the postsynaptic spike and the spike order during pairing became post-before-pre pairing (Fig. 2e), whereas stimulation of tempero-ammonic input on the descending phase of oscillation delayed the postsynaptic spike, reversing the spike order to pre-before-post pairing (Fig. 2j). Reversal of the sign of STDP by tempero-ammonic input was also observed with 60 pairings (see Supplementary Fig. 2), under intact inhibition (Supplementary Fig. 3) and required activation of NMDA receptors (Supplementary Fig. 4). A simple spike timing–based learning model (Supplementary Methods) showed that, for randomly activated Schaffer collateral inputs across the theta cycle, tempero-ammonic input could enforce either LTP or LTD on the network depending on the timing of tempero-ammonic activation relative to theta oscillation (Fig. 3). These results indicate that activation of tempero-ammonic input can reverse the sign of plasticity and enforce either LTP or LTD at the Schaffer collateral–CA1 synapse by prospectively controlling postsynaptic spike time. As few as ten pairing events could induce tempero-ammonic–enforced LTP in young adult tissue, making it likely that similar mechanisms operate in vivo
(Supplementary Fig. 5). Notably, activation of Schaffer collateral input instead of tempero-ammonic input primarily enforced LTD (Supplementary Fig. 6). The control of sign of plasticity by external input strengthens earlier experimental links between synaptic plasticity and network oscillations in the hippocampus10 and emphasizes the importance of interaction between different inputs in controlling synaptic plasticity11. Thus, firing correlations between local neurons determine whether plasticity will occur, whereas the sign of that plasticity can be determined by information encoded in the timing of an external input relative to the local network dynamics. Because there is a time-varying difference in the phase of synaptic input from the Schaffer collateral and tempero-ammonic input during physiological theta oscillations12,13, and because these inputs are under differential neuromodulatory control14, this neurocomputational principle is likely to contribute to hippocampal memory processing in vivo. Notably, STDP during oscillation in mushroom bodies in the locust has recently been implicated in odor learning15, suggesting that oscillatory control of STDP is a fundamental facet of neural network operations in many species.
nature neuroscience VOLUME 12 | NUMBER 10 | october 2009
1221
Note: Supplementary information is available on the Nature Neuroscience website. Acknowledgments This work was supported by the Biotechnology and Biological Sciences Research Council (grant number BB/D0157581) and the Kwanjeong Foundation. AUTHOR CONTRIBUTIONS J.K. conducted the experiments and analyzed the data. J.K. and O.P. designed the experiments and wrote the manuscript. Published online at http://www.nature.com/natureneuroscience/. Reprints and permissions information is available online at http://www.nature.com/ reprintsandpermissions/. 1. Caporale, N. & Dan, Y. Annu. Rev. Neurosci. 31, 25–46 (2008). 2. Hebb, D.O. The Organization of Behavior (John Wiley, New York, 1949). 3. Markram, H. et al. Science 275, 213–215 (1997). 4. Bi, G.Q. & Poo, M.M. J. Neurosci. 18, 10464–10472 (1998). 5. Debanne, D. et al. J. Physiol. (Lond.) 507, 237–247 (1998). 6. Buzsáki, G. Neuron 33, 325–340 (2002). 7. Neves, G. et al. Nat. Rev. Neurosci. 9, 65–75 (2008). 8. Otmakhova, N.A. & Lisman, J.E. J. Neurophysiol. 92, 2027–2039 (2004). 9. Soltesz, I. & Deschênes, M. J. Neurophysiol. 70, 97–116 (1993). 10. Huerta, P.T. & Lisman, J.E. Nature 364, 723–725 (1993). 11. Remondes, M. & Schuman, E.M. Nature 416, 736–740 (2002). 12. Kocsis, B. et al. J. Neurosci. 19, 6200–6212 (1999). 13. Buzsáki, G. Rhythms of the Brain (Oxford University Press, Oxford, 2006). 14. Hasselmo, M.E. & Schnell, E. J. Neurosci. 14, 3898–3914 (1994). 15. Cassenaer, S. & Laurent, G. Nature 448, 709–713 (2007).
B r i e f c o m m u n i c at i o n s
Selective suppression of hippocampal ripples impairs spatial memory Sharp wave–ripple (SPW-R) complexes in the hippocampusentorhinal cortex are believed to be important for transferring labile memories from the hippocampus to the neocortex for long-term storage. We found that selective elimination of SPW-Rs during post-training consolidation periods resulted in performance impairment in rats trained on a hippocampus-dependent spatial memory task. Our results provide evidence for a prominent role of hippocampal SPW-Rs in memory consolidation. Memory consolidation refers to the stabilization of labile memory traces, possibly including intrahippocampal synaptic reinforcement
b
a
Field EPSP amplitude (mV)
e
REM/SWS ratio
No. of ripples
No. of ripples
d
Cortex Pyr Int
c
Hippocampus Pyr Int
Figure 1 Ventral hippocampal commissural 1.4 stimulation interrupts SPW-Rs and hippocampal 0.2 mV cell discharges without changing global sleep 0 architecture. (a) Interruption of SPW-R and 1.4 spiking activity in the hippocampus. Local field potential (LFP, black) in the hippocampus and 0 spiking activity (vertical ticks) of pyramidal cells 1.4 (pyr; hippocampus, dark blue; sensorimotor 0 cortex, red) and interneurons (int; hippocampus, 1.4 light blue; sensorimotor cortex, orange). Left, an intact ripple and the associated spiking activity. 50 ms 0 50 ms –100 0 200 Vertical dashed lines and arrowheads represent Time (ms) stimulation times. (b) Duration of unit activity suppression as a function of the magnitude of 250 Test Test the evoked field response. Pseudo-color plots 200 0.08 150 show the z scores of multiple unit activity with 100 increasing levels of stimulation (ordinate). Note 0.06 50 the transient, evoked response magnitude– 0 250 Control dependent suppression of spiking activity in 0.04 Control 200 the hippocampus with no observable effect on 150 0.02 global neocortical activity. The increased activity 100 before the stimulus is a result of the buildup of 50 0 0 ripple-associated discharge. (c) SPW-R blocking –1,000 –100 –10 0 10 100 1,000 Test Control by ventral hippocampal commissural stimulation Time (ms) (arrowheads). Example SPW-R in a test rat and a control rat (left). SPW-R was blocked after a few cycles in the test rat (upper right). For illustration purposes, the SPW-R–detection threshold was set higher for this example than in sleep sessions (inset). In the control rat (lower right), stimulation was triggered after a delay. Scale bars represent 20 ms and 0.2 mV. (d) Cross-correlograms of stimulations and offline-detected SPW-Rs in test and control rats. Virtually all SPW-Rs were suppressed in test rats, but were preserved in control rats, as a result of the 80–120-ms delay introduced between the ripples (blue peak) and the stimulations (time zero). (e) Average random eye movement (REM) sleep/slow-wave sleep (SWS) ratios in a random subset of test and control sessions (n = 24 and n = 27, respectively; t test, not significant (P > 0.05), error bars represent s.e.m.). Hippocampus Hippocampus Cortex LFP Pyr Int Pyr Int
© 2009 Nature America, Inc. All rights reserved.
Gabrielle Girardeau1,3, Karim Benchenane1,3, Sidney I Wiener1, György Buzsáki2 & Michaël B Zugaro1
and the transfer of information initially encoded in the hippocampal system to the neocortex for long-term storage1–3. The consolidation process has been proposed to occur during post-learning rest or sleep3,4 by reactivation of memory traces in short bouts of neuronal activity associated with SPW-R events3,5–7, which can be temporally biased by neocortical slow oscillations8–10. Although numerous studies provide compelling correlative links between hippocampal SPW-Rs and memory consolidation3,5–7, a causal relationship has not yet been demonstrated. To examine the consequences of SPW-R elimination on performance in a hippocampus-dependent, spatial-reference memory task11 (Supplementary Methods and Supplementary Fig. 1), we selectively suppressed SPW-Rs during post-learning sleep. All of our experiments were conducted in accordance with institutional (CNRS Comité Opérationnel pour l’Ethique dans les Sciences de la Vie) and international (US National Institutes of Health guidelines) standards and legal regulations (certificat no. 7186, Ministère de l’Agriculture et de la Pêche). The onset of SPW-Rs was detected online by filtering the signal in the ripple-band and thresholding it. Threshold crossing triggered single-pulse stimulation of the ventral hippocampal commissure12 (n = 17 rats). This
1Laboratoire
de Physiologie de la Perception et de l’Action, Collège de France, CNRS, Paris, France. 2Center for Molecular and Behavioral Neuroscience, Rutgers University, Newark, New Jersey, USA. 3These authors contributed equally to this work. Correspondence should be addressed to M.B.Z. (
[email protected]) or G.B. (
[email protected]). Received 20 May; accepted 9 July; published online 13 September 2009; doi:10.1038/nn.2384
1222
VOLUME 12 | NUMBER 10 | october 2009 nature neuroscience
b r i e f c o m m u n i c at i o n s 1.0
Stimulated control Unimplanted control Test
Performance index
0.9 0.8 0.7 0.6 0.5
Chance
0.4
© 2009 Nature America, Inc. All rights reserved.
0.3
1
2
3
4
5
6
7
8 9 10 11 12 13 14 15 Days
Figure 2 Suppression of SPW-Rs interferes with memory consolidation. Test rats (red: n = 7) were significantly impaired in the radial maze task compared with control rats (blue, n = 7 stimulated controls; black, n = 12 unimplanted controls; error bars represent s.e.m.). Grey shading indicates the chance zone. Although performance increased in the three groups, rats with ripple suppression took more days to perform above upper chance level (t tests) and their performance remained consistently below that of the control groups. The average s.d. of the performance index were 0.18 (unimplanted controls), 0.18 (stimulated controls) and 0.19 (tests).
blocked further development of the oscillation and transiently silenced hippocampal spiking activity12 (Fig. 1a,b and Supplementary Fig. 2), thus preventing potential replay of place-cell13 sequences5,6 previously activated during waking. In contrast with hippocampal cells, firing of neocortical neurons was not interrupted at the stimulus intensities that we used for abolishing ripples (Fig. 1a,b and Supplementary Fig. 3 shows data from anterior cingulated and prelimbic/infralimbic prefrontal cortices, two major candidates of hippocampal-neocortical information transfer during spatial memory consolidation2). Next, we tested the role of SPW-Rs on memory consolidation. Three groups of rats (test group, n = 7; stimulated controls, n = 7; unimplanted controls, n = 12) were trained to find food rewards on an eight-arm radial maze in which the same three arms were baited every day (Supplementary Fig. 1). The rats performed three trials per day, after which they were allowed to sleep for 1 h. During post-training rest and sleep, all of the online-detected ripples were suppressed by commissural stimulations in test rats (average online detection rate was 86.0 ± 1.3% (s.e.m.) of post hoc detected SPW-Rs; Fig. 1c,d and Supplementary Methods). Stimulated control rats underwent the same protocol, except that a random delay (80–120 ms) was introduced between SPW-R detection and stimulation, ensuring that the stimulations occurred mainly outside of the ripple episodes (Fig. 1c,d). Thus, these control rats received the same number of stimulations as test rats (t test, not significant, P > 0.05), but their hippocampal ripples were left largely intact. The global architecture of sleep and the local field potential power in distinct sleep stages were not modified by the suppression of SPW-Rs (Fig. 1e and Supplementary Fig. 4). Because reactivations of previously active cell assemblies occur preferentially during the first half-hour of sleep after exploration5, we blocked SPW-Rs for 1 h following training sessions. As stimulation outside SPW-Rs had no detectable effect on task performance (no significant difference between stimulated control and unimplanted rats, two-factor ANOVA, day × group, P > 0.05; Fig. 2), the two control groups were pooled and compared with test rats. Performance of the test rats was significantly impaired compared with control rats (two-factor ANOVA day × group, P < 0.001 for main factors, P < 0.01 for interaction; Fig. 2). In control rats (stimulated and unimplanted groups combined), performance exceeded the upper chance level after 5 d of training, whereas test rats nature neuroscience VOLUME 12 | NUMBER 10 | october 2009
continued to perform at chance level until the eighth day of training (t tests, P < 0.05). Test rats did not develop stereotyped turning strategies (Supplementary Fig. 5) and working memory errors remained very low (less than one error per trial on average) in the three groups (Supplementary Fig. 6). Suppression of SPW-Rs and associated neuronal discharges resulted in deterioration of memory consolidation. The behavioral effect was specifically related to suppression of SPW-Rs, rather than to nonspecific consequences of the stimulation, as SPW-R–yoked control stimulation had no detectable effect on behavior. The observed deficit is all the more notable, as we suppressed the SPW-Rs for only 1 h and ripple incidence returned to normal levels after the stimulation period (Supplementary Fig. 4). The magnitude of impairment in the ripple-suppressed rats was comparable to that reported in a previous study on hippocampus-lesioned rats11. The slight performance improvement in the test group could be the result of the spared small-amplitude SPW-Rs, of the SPW-Rs occurring after the stimulation period or of other, nonhippocampal learning mechanisms, as has been reported previously11. Our findings therefore indicate that SPW-Rs are critical for memory consolidation, possibly because, by temporally compressing reactivations of waking firing sequences 3,5,6 in the hippocampus, they allow spikes to occur in a time window that is compatible with activation of the NMDA receptors and spike timing–dependent plasticity. In addition or alternatively, they would enable the reactivated ensembles to exert a strong effect on downstream target neurons7. Moreover, hippocampal SPW-Rs are coordinated in time with neocortical unit firing, slow oscillations and sleep spindles8–10,14,15, suggesting that they have a widespread effect on cortical function underlying long-term memory consolidation. Note: Supplementary information is available on the Nature Neuroscience website. Acknowledgments We thank S. Sara, A. Larrieu, L. Hazan, A. Fruchart, P. Ruther, S. Kisban, S. Herwik, S. Doutremer, M.-A. Thomas, Y. Dupraz, M. Ehrette and S. Rateau for advice and technical support. This work was supported by the International Human Frontiers Science Program Organization (CDA0061/2007-C), the US National Institutes of Health (NS34994), European Projects NeuroProbes (IST-027017) and Integrating Cognition, Emotion and Autonomy (FP6-IST 027819), and a Collège de France Visiting Professor Fellowship. AUTHOR CONTRIBUTIONS G.B. and M.B.Z. designed the study. G.G. carried out the majority of the experiments, K.B. and M.B.Z. carried out the remaining experiments. M.B.Z., G.G. and K.B. analyzed the data. All of the authors contributed to writing the manuscript. Published online at http://www.nature.com/natureneuroscience/. Reprints and permissions information is available online at http://www.nature.com/ reprintsandpermissions/. 1. Squire, L.R. & Bayley, P.J. Curr. Opin. Neurobiol. 17, 185–196 (2007). 2. Maviel, T., Durkin, T.P., Menzaghi, F. & Bontempi, B. Science 305, 96–99 (2004). 3. Buzsáki, G. Neuroscience 31, 551–570 (1989). 4. Marshall, L., Helgadóttir, H., Mölle, M. & Born, J. Nature 444, 610–613 (2006). 5. Kudrimoti, H.S., Barnes, C.A. & McNaughton, B.L. J. Neurosci. 19, 4090–4101 (1999). 6. Wilson, M.A. & McNaughton, B.L. Science 265, 676–679 (1994). 7. Chrobak, J.J. & Buzsáki, G. J. Neurosci. 16, 3056–3066 (1996). 8. Battaglia, F.P., Sutherland, G.R. & McNaughton, B.L. Learn. Mem. 11, 697–704 (2004). 9. Siapas, A.G. & Wilson, M.A. Neuron 21, 1123–1128 (1998). 10. Sirota, A., Csicsvari, J., Buhl, D. & Buzsáki, G. Proc. Natl. Acad. Sci. USA 100, 2065–2069 (2003). 11. Jarrard, L.E. Behav. Brain Res. 71, 1–10 (1995). 12. Zugaro, M.B., Monconduit, L. & Buzsáki, G. Nat. Neurosci. 8, 67–71 (2005). 13. O’Keefe, J. & Nadel, L. The Hippocampus as a Cognitive Map (Oxford University Press, Oxford, 1978). 14. Eschenko, O., Ramadan, W., Mölle, M., Born, J. & Sara, S.J. Learn. Mem. 15, 222–228 (2008). 15. Peyrache, A., Khamassi, M., Benchenane, K., Wiener, S.I. & Battaglia, F. Nat. Neurosci. 12, 919–926 (2009).
1223
B r i e f c o m m u n i c at i o n s
Intact rapid detection of fearful faces in the absence of the amygdala The amygdala is thought to process fear-related stimuli rapidly and nonconsciously. We found that an individual with complete bilateral amygdala lesions, who cannot recognize fear from faces, nonetheless showed normal rapid detection and nonconscious processing of those same fearful faces. We conclude that the amygdala is not essential for early stages of fear processing but, instead, modulates recognition and social judgment. Subject SM has complete bilateral lesions of the amygdala and is impaired in her recognition of fear1, an impairment that is consistent with previous studies showing activation of the amygdala to overt and masked fear faces2. These studies have suggested that the amygdala is involved in pre-attentive, rapid processing, whereby the amygdala receives subcortical visual information via the superior colliculus and pulvinar thalamus3. Such a picture is similar to the known subcortical route for the amygdala in auditory fear conditioning, as demonstrated in rats, and is consistent with blood oxygen level–dependent
Figure 1 Intact rapid, automatic and nonconscious detection of fearful Threat scenes Fear faces Angry faces faces in the absence of the amygdala. (a) Rapid detection of fear- and 3 threat-related images. Viewers were shown two images side by side for 2 40 ms (unmasked), one neutral and the other showing fear, anger or threat. 1 We carried out three experiments, one with fearful versus neutral faces, the second with angry versus neutral faces and the third with threat-related 0 200 400 600 800 1,000 200 400 600 800 1,000 200 400 600 800 1,000 Reaction time (ms) images versus neutral images. Subjects were asked to push a button as quickly as possible to indicate if the target image that showed more fear/anger or was more threatening was on the left or the right. Speed and Fear/sad accuracy tradeoffs in discriminating fear were normal in two sessions for SM (red) compared with 12 controls (solid black line indicates the mean Fear/happy and the dotted lines indicate the 95% confidence interval). Accuracy was quantified by d′, the difference between the z-transformed hit and the false alarm rate. (b) Visual search for fear. Subjects detected an oddball target Fear/calm SM among distractors; both were perceived as belonging either to the same –0.1 0 –0.1 0 0.1 0.2 0.3 0.4 0.1 0.2 0.3 0.4 category (for example, mild and extreme fear) or to different categories Normalized reaction time improvement Normalized reaction time difference (for example, neutral and mild fear), even though they differed geometrically Happy Fear by the same degree. SM showed normal category boundary effects in faster faster reaction time. The black bars indicate the average across three age-matched controls and the white bars indicate SM’s performance. (c) Breaking into consciousness probed by continuous flash suppression. Fearful faces broke interocular suppression faster than happy faces in SM (white) to the same degree as in controls (black). Subjects clicked a mouse as soon as any part of the face became visible. The dot and error bar indicate the mean and the s.d. for seven control subjects. All subjects gave written informed consent as approved by the institutional review board of the California Institute of Technology.
a
d′
© 2009 Nature America, Inc. All rights reserved.
Naotsugu Tsuchiya1,2,5, Farshad Moradi1,4,5, Csilla Felsen1, Madoka Yamazaki3 & Ralph Adolphs1,3
activation of the amygdala by nonconscious fearful faces in humans4. However, there are discrepancies with this view of amygdala function. Some neuroimaging studies have found that the amygdala’s response to fearful faces is strongly modulated by conscious detectability, at least when backward masking is used5. Electrophysiological latencies recorded in the amygdala are, by and large, inconsistent with rapid visual processing6 and there is no direct anatomical evidence to support the rapid visual subcortical route that has been hypothesized7. These discrepancies suggest that the amygdala modulates social judgments of fear, rather than initial pre-attentive detection. To help resolve this debate regarding the amygdala’s contribution to fear processing, we tested subject SM on rapid detection of fear- and threat-related stimuli. In our first experiment, subjects saw a target stimulus (fearful face, angry face or scene showing threat) next to neutral stimuli for 40 ms (unmasked) and had to push a button as rapidly as possible to indicate which face showed more fear/anger or which scene was more threatening (Supplementary Fig. 1). SM’s performance on this task was completely normal for all three threatrelated categories (Fig. 1a). As reported previously1, SM rated the intensity of fear shown in the same face stimuli substantially lower than did the controls (2.8 and 3.7 s.d. below the normal mean on the two testing sessions). Control experiments ruled out several possible interpretations (Supplementary Methods and Supplementary Table 1). First, we used backward masking in the control experiments, as it might be required to prevent afterimages to demonstrate the amygdala’s rapid fear detection. Second, the control experiments compared fearful faces with sad and happy faces, rather than just
b
c
1Division of Humanities and Social Sciences, California Institute of Technology, Pasadena, California, USA. 2Brain Science Institute, Tamagawa University, Machida, Tokyo, Japan. 3Division of Biology, California Institute of Technology, Pasadena, California, USA. 4Present address: Department of Radiology, University of California San Diego, San Diego, California, USA. 5These authors contributed equally to this study. Correspondence should be addressed to N.T. (
[email protected]).
Received 20 April; accepted 8 July; published online 30 August 2009; doi:10.1038/nn.2380
1224
VOLUME 12 | NUMBER 10 | october 2009 nature neuroscience
© 2009 Nature America, Inc. All rights reserved.
b r i e f c o m m u n i c at i o n s neutral faces, as SM might simply have been discriminating ‘emotional’ from ‘neutral’ using specific low-level features of neutral faces, rather than detecting fear in particular. Third, the control experiments used NimStim faces, which SM had never seen before, as SM may have been overtrained with the Ekman faces, which she saw many times in various experiments. Across all of these tasks, SM showed entirely normal rapid detection of fearful faces. In a second experiment, we found an analogous pattern; despite impaired categorization of fearful faces when given unlimited time, SM showed normal effects of category boundaries on speeded visual search. In this experiment, we dissociated physical from psychological similarity by showing subjects faces that had been morphed between neutral and fearful expressions (Supplementary Fig. 2). We first asked subjects to categorize these morphs as being neutral or afraid; as expected, subjects showed sharp category boundaries for the morphs, a categorical perception effect that has been well documented8. SM’s category boundary was significantly shifted (P < 0.0005), and she required a greater intensity of fear to categorize a morph as being afraid (Supplementary Fig. 3). We then gave subjects a visual search task in which they were asked to detect, as rapidly as possible, which face in an array was different from the rest (no specific information was given about the basis of that difference). Two versions of this search task showed subjects targets (more fearful morphs) and distractors (less fearful morphs) that always differed by the same physical morph distance, but which either spanned the mean normal category boundary or did not (Supplementary Methods). All subjects showed faster search times for targets and distractors that spanned the neutral/fear category boundary, than for those that did not span the category boundary, as did SM (the controls were, on average, 15% faster and SM was 26% faster; their 95% confidence intervals overlapped). Thus, SM showed normal effects of the neutralfear category boundary (as derived from the controls) on implicit rapid visual search for fearful faces, despite impaired overt categorization of the same faces. To show that these findings were not limited to fear-neutral discriminations, we carried out an identical experiment with morphs for happy/fear and sad/fear using the Karolinska directed faces rather than the Ekman faces; the controls were, on average, 18% and 27% faster, respectively, and SM was 25% and 33% faster, respectively (Fig. 1b). Moreover, SM’s accuracy was >99% in all conditions. Thus, in a search task, SM implicitly discriminated between fearful and other expressions with the same fear category effects as normal subjects. In a third experiment, we focused more specifically on the amyg dala’s role in nonconscious processing of fear. We used continuous flash suppression9,10 to measure the potency of a fearful face in overcoming strong interocular suppression. Subjects were presented with a stream of flashes of Mondrian patterns at 10 Hz to the right eye while an emotional face was gradually introduced into one quadrant to the left eye; we used both Ekman face set stimuli and NimStim stimuli to ensure that there was no idiosyncratic effect of the Ekman faces as a result of SM’s greater familiarity with them (Supplementary Methods and Supplementary Fig. 4). In our normal subjects, we found that fearful facial expressions break through into consciousness more rapidly than happy expressions (Fig. 1c), as has been seen previously10. To our surprise, SM showed exactly the same fear advantage in breaking interocular suppression. Thus, fearful faces gain access to consciousness in SM just as rapidly as in control subjects. There was one qualitative exception to SM’s otherwise entirely intact processing of fearful faces. SM performed somewhat worse on fear-sad discriminations (z score = −1.18) than on other discriminations (fear-neutral z score = −0.77 and fear-happy z score = −0.62;
Supplementary Methods and Supplementary Table 1), although this difference was small. It may be that SM can distinguish fear on the basis of valence (for example, from happiness), but has relatively greater difficulty for more subordinate-level discriminations between fear and other negatively valenced emotions (for example, from sad), a possibility that could be probed in greater detail in future experiments with larger numbers of trials. Taken together, our findings suggest that the amygdala is not essential for nonconscious, rapid fear detection. It is still possible that the amygdala participates in such processing, provided that it results from indirect modulation. Our interpretation is also consistent with earlier findings that SM’s impaired explicit fear recognition can be rescued if she is instructed to look at the eyes in faces, something she fails to do spontaneously11. In the absence of the amygdala, explicit fear recognition may be impaired as a result of an absence of the amygdala’s normal modulation of information processing (for example, directing visual attention to the eyes in faces). More puzzling is that individuals with blindsight resulting from primary visual cortex lesions show amygdala activation by fearful faces12, a finding that could be taken as evidence for a subcortical visual route to the amygdala involved in nonconscious fear processing. However, although cortical damage is sufficient to prevent conscious vision in such individuals, it may be incomplete and could permit sparse cortical input to the amygdala that is sufficient to account for the observed activation. Alternatively, a retino-collicular-pulvinarcortical pathway might indirectly route visual information to the amyg dala via spared extrastriate cortical areas. Our conclusions are also not inconsistent with an early role for emotion in driving attention3, even in the absence of conscious awareness13, but argue that the amygdala is not a necessary substrate for this role. Instead, we favor the idea that the amygdala modulates other cognitive processes on the basis of an appraisal-like evaluation of the biological relevance of stimuli14 and contributes to explicit judgments about the fear shown in fearful faces once substantial cortical processing has already taken place.
nature neuroscience VOLUME 12 | NUMBER 10 | october 2009
1225
Note: Supplementary information is available on the Nature Neuroscience website. Acknowledgments We thank C. Holcomb for help with recruiting and testing subjects. This work was funded by grants from the National Institute of Mental Health and the Simons Foundation to R.A. and was supported, in part, by Tamagawa University global Center of Excellence program of the Japanese Ministry of Education, Culture, Sports and Technology. N.T. is supported by the Japan Society for the Promotion of Science. AUTHOR CONTRIBUTIONS N.T., F.M. and R.A. designed the study and wrote the paper. N.T. and F.M. executed most of the study and analyzed all of the data. C.F. and M.Y. helped with aspects of the data collection. Published online at http://www.nature.com/natureneuroscience/. Reprints and permissions information is available online at http://www.nature.com/ reprintsandpermissions/. 1. Adolphs, R., Tranel, D., Damasio, H. & Damasio, A. Nature 372, 669–672 (1994). 2. Whalen, P.J. et al. J. Neurosci. 18, 411–418 (1998). 3. Ohman, A., Carlsson, K., Lundqvist, D. & Ingvar, M. Physiol. Behav. 92, 180–185 (2007). 4. Jiang, Y. & He, S. Curr. Biol. 16, 2023–2029 (2006). 5. Pessoa, L., Japee, S., Sturman, D. & Ungerleider, L.G. Cereb. Cortex 16, 366–375 (2006). 6. Mormann, F. et al. J. Neurosci. 28, 8865–8872 (2008). 7. Adolphs, R. Curr. Opin. Neurobiol. 18, 166–172 (2008). 8. Rotshtein, P., Henson, R.N., Treves, A., Driver, J. & Dolan, R.J. Nat. Neurosci. 8, 107–113 (2005). 9. Tsuchiya, N. & Koch, C. Nat. Neurosci. 8, 1096–1101 (2005). 10. Yang, E., Zald, D.H. & Blake, R. Emotion 7, 882–886 (2007). 11. Adolphs, R. et al. Nature 433, 68–72 (2005). 12. Pegna, A.J., Khateb, A., Lazeyras, F. & Seghier, M.L. Nat. Neurosci. 8, 24–25 (2005). 13. Lin, J.Y., Murray, S.O. & Boynton, G.M. Curr. Biol. 19, 1118–1122 (2009). 14. Sander, D., Grafman, J. & Zalla, T. Rev. Neurosci. 14, 303–316 (2003).
B r i e f c o m m u n i c at i o n s
Personal space regulation by the human amygdala
© 2009 Nature America, Inc. All rights reserved.
Daniel P Kennedy1, Jan Gläscher1, J Michael Tyszka2 & Ralph Adolphs1,2 The amygdala plays key roles in emotion and social cognition, but how this translates to face-to-face interactions involving real people remains unknown. We found that an individual with complete amygdala lesions lacked any sense of personal space. Furthermore, healthy individuals showed amygdala activation upon close personal proximity. The amygdala may be required to trigger the strong emotional reactions normally following personal space violations, thus regulating interpersonal distance in humans.
People automatically and reliably regulate the distance maintained between themselves and others during social interaction 1. Personal space, defined as the area individuals maintain around themselves into which intrusion by others causes discomfort2, is one mechanism by which this automatic regulation of interpersonal distance is achieved. However, little is known regarding the neural substrates of personal space. One candidate brain region is the amygdala, as studies in nonhuman primates have found that this structure is involved in social approach and avoidance3–5. Here we show that one’s sense of personal space is dependent on the amygdala. We studied S.M., a 42-year-old woman with complete bilateral amygdalar damage we have described extensively6,7. S.M. indicated the position at which she felt most comfortable as a female experimenter approached her from 4.7 m across the room; chin-to-chin distance was recorded using a digital laser measurer. We repeated this procedure four times (counterbalanced with other trial types; see Supplementary Text). S.M.’s preferred distance (0.34 ± 0.02 m; mean and s.d.) was smaller than the smallest preferred distance on any trial of any comparison subject (0.76 ± 0.34 m, range = 0.44–1.52 m, N = 20; Fig. 1) and statistically significantly smaller than that of the comparison group (after excluding the three outliers with the largest distance preferences, a mean comparison-subject distance of 0.64 ± 0.13 m, Z = −2.20, P = 0.014, one-tailed; with a modified t-test, t16 = −2.14, P = 0.024.) This highly abnormal pattern was found reliably across various experimental manipulations (gaze direct or averted; subject being approached or approaching; starting close or far; a total of 32 trials per subject; Z = −2.38, P = 0.009, one-tailed; t16 = −2.31, P = 0.017, one-tailed, excluding three outliers) and when S.M.’s distance preferences were compared to female controls alone (Z = −1.93, P = 0.027; t11 = −1.86, P = 0.045). Furthermore, it could not be accounted for by S.M.’s degree of familiarity with the experimenter (see Supplementary Text for detailed results).
Throughout the experiment, S.M. demonstrated a notable lack of discomfort at close distances. For example, on one trial, she walked all the way toward the experimenter to the point of touching, and she repeatedly stated that any distance felt comfortable. We quantified this by asking her to rate her degree of discomfort (1, perfectly comfortable; 10, extremely uncomfortable) while one of us stood facing her at various distances. Even when nose-to-nose with direct eye contact, S.M. rated the experience a 1. In a more natural and unexpected context, a completely unfamiliar male confederate stood abnormally close to her while engaging in conversation; S.M. again rated the experience a 1. By contrast, the confederate rated his experience a 7. Although S.M. indicated afterward that she knew we were “up to something,” awareness that this was an experiment cannot explain her lack of discomfort, since the confederate had complete awareness yet still found the experience to be highly uncomfortable. At a cognitive level, S.M. understood the concept of personal space. She spontaneously stated that she did not want to make the experimenter uncomfortable by standing too close, and also stated that she believed her personal space was smaller than most. Furthermore, we asked S.M. to position the experimenter at the distance she judged other people might feel most comfortable. Although she considerably underestimated this distance (0.47 ± 0.03 m), her estimation was 38% greater than her own personal preference, thus demonstrating that she is aware that other people have personal space requirements different from her own. The fact that S.M. had a nonzero distance preference at all may simply reflect typical sensory processing constraints (for example, too close makes it more difficult to focus on the person). Our findings in S.M. made a clear prediction regarding the amygdala in healthy individuals: its activity should be modulated by
a 0
0.2
0.4
0.6 0.8 1.0 Distance from experimenter (m)
b
1.2
1.4
1.6
c
0.34 m
0.64 m
Figure 1 Lesion study: mean preferred distances from the experimenter. (a) Preference of S.M. (red) was the closest distance to the experimenter (black), among age-, gender-, race- and education-matched controls (purple, n = 5), as well as general comparison subjects (blue, n = 15). (b) S.M.’s mean preferred distance from the experimenter (image drawn to scale). (c) Control participants’ mean preferred distance from the experimenter, excluding the three largest outliers (image drawn to scale).
1Division
of Humanities and Social Sciences and 2Division of Biology, California Institute of Technology, Pasadena, California, USA. Correspondence should be addressed to D.P.K. (
[email protected]). Received 26 March; accepted 8 July; published online 30 August 2009; doi:10.1038/nn.2381
1226
VOLUME 12 | NUMBER 10 | october 2009 nature neuroscience
b r i e f c o m m u n i c at i o n s
interpersonal distance. As a preliminary test of this prediction, and to obtain corroborating evidence, we conducted a functional magnetic resonance imaging (fMRI) study in eight healthy participants. We found that the amygdala responded to a greater degree when the participants knew an experimenter was maintaining a close distance to them (standing immediately next to the scanner) than when they knew an experimenter was maintaining a far distance. This effect was statistically significant at the group level (Fig. 2; see Supplementary Text for details). Although we did not collect ratings of subjective comfort from S.M. or control subjects on the protocol used in this fMRI study, our interpretation of the observed amygdala activation is that it reflects the same amygdala-dependent mechanism that comes into play when our personal space is noticeably violated. In sum, we found that the amygdala was differentially activated by proximity to another person, and that complete bilateral damage to this structure in S.M. resulted in the absence of a detectable personal space boundary and an abnormally small interpersonal distance preference. In various animal species, many social behaviors (including collective group organization and consensus decision-making) can be modeled as a balance between attractive and repulsive forces between individual members of a group8,9. Our findings suggest that the amygdala may mediate the repulsive force that helps to maintain a minimum distance between people. Further, our findings are consistent with those in monkeys with bilateral amygdala lesions, who stay within closer proximity to other monkeys or people4,5, an effect we suggest arises from the absence of strong emotional responses to personal space violation. One open question concerns how this mechanism might develop in infants and young children. It is possible that the amygdala is necessary for learning the association between close distances and aversive outcomes rather than triggering innate emotional responses to close others. As the developmental course of S.M.’s lesion is unknown, her data cannot distinguish between these two possibilities. A second open question is how this mechanism can accommodate modulation by situational context, personal familiarity and other factors2,10. Furthermore, there are variations in social distance between individuals, and gross dysregulation in disorders such as autism and Williams syndrome. These effects could arise in part through modulation of the amygdala from the prefrontal cortex, an effect of considerable recent interest in explaining individual differences and psychiatric disease11.
nature neuroscience VOLUME 12 | NUMBER 10 | october 2009
a
Left amygdala [–18, –3, –15]
0
b Contrast estimate (close – far)
© 2009 Nature America, Inc. All rights reserved.
Figure 2 fMRI study: activation of the amygdala by close (relative to far) interpersonal distance. (a) Coronal slices showing significantly activated voxels in the dorsal amygdala (cluster-level significance, P < 0.05); scale shows t-value. (b) Contrast parameters (arbitrary units) for each of the eight subjects who participated in the experiment (extracted from and averaged across all significant voxels in a; blue dots), along with the group mean (black line). Coordinates for the peak voxel are shown. Subjects were unable to see the position of the experimenter, but were informed of his location at all times. All experiments were approved by the California Institute of Technology’s Institutional Review Board, and informed written consent was obtained from all participants. See Supplementary Text for a detailed description of the experiment.
Right amygdala [18, 3, –18]
5
0
7
1.5
1.5
1.0
1.0
0.5
0.5
0
0
–0.5
–0.5
Note: Supplementary information is available on the Nature Neuroscience website. Acknowledgments We thank C. Holcomb for behavioral data collection, R. Nair and V. Chib for help with the fMRI study, and M. Spezio for discussions. Supported by US National Institute of Mental Health and the Simons Foundation (R.A.), the Della Martin Foundation (D.P.K.) and the Tamagawa University global Centers of Excellence program of the Japanese Ministry of Education, Culture, Sports and Technology. AUTHOR CONTRIBUTIONS D.P.K. and R.A. designed the experiment and wrote the paper; D.P.K. executed the studies; D.P.K., J.G. and J.M.T. analyzed the data. Published online at http://www.nature.com/natureneuroscience/. Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/. 1. 2. 3. 4. 5.
Hall, E. The Hidden Dimension (Doubleday, Garden City, New York, USA, 1966). Hayduk, L.A. Psychol. Bull. 85, 117–134 (1978). Kluver, H. & Bucy, P.C. Am. J. Physiol. 119, 352–353 (1937). Emery, N.J. et al. Behav. Neurosci. 115, 515–544 (2001). Mason, W.A., Capitanio, J.P., Machado, C.J., Mendoza, S.P. & Amaral, D.G. Emotion 6, 73–81 (2006). 6. Adolphs, R., Tranel, D. & Damasio, A.R. Nature 393, 470–474 (1998). 7. Buchanan, T.W., Tranel, D. & Adolphs, R. in The Human Amygdala (eds. Whalen, P.J. & Phelps, E.A.) 289–320 (Oxford Univ. Press, New York, 2009). 8. Couzin, I.D., Krause, J., James, R., Ruxton, G.D. & Franks, N.R. J. Theor. Biol. 218, 1–11 (2002). 9. Couzin, I.D., Krause, J., Franks, N.R. & Levin, S.A. Nature 433, 513–516 (2005). 10. Hayduk, L.A. Psychol. Bull. 94, 293–335 (1983). 11. Pezawas, L. et al. Nat. Neurosci. 8, 828–834 (2005).
1227
a r t ic l e s
AP2 regulates basal progenitor fate in a region- and layer-specific manner in the developing cortex
© 2009 Nature America, Inc. All rights reserved.
Luisa Pinto1,14, Daniela Drechsel1,2, Marie-Theres Schmid1, Jovica Ninkovic1, Martin Irmler3, Monika S Brill1,4, Laura Restani5, Laura Gianfranceschi5, Chiara Cerri5, Susanne N Weber6, Victor Tarabykin7, Kristin Baer8, François Guillemot2, Johannes Beckers3,9, Nada Zecevic10, Colette Dehay11,12, Matteo Caleo5, Hubert Schorle6 & Magdalena Götz1,4,13 An important feature of the cerebral cortex is its layered organization, which is modulated in an area-specific manner. We found that the transcription factor AP2 regulates laminar fate in a region-specific manner. Deletion of AP2 (also known as Tcfap2c) during development resulted in a specific reduction of upper layer neurons in the occipital cortex, leading to impaired function and enhanced plasticity of the adult visual cortex. AP2 functions in apical progenitors, and its absence resulted in misspecification of basal progenitors in the occipital cortex at the time at which upper layer neurons were generated. AP2 directly regulated the basal progenitor fate determinants Math3 (also known as Neurod4) and Tbr2, and its overexpression promoted the generation of layer II/III neurons in a time- and region-specific manner. Thus, AP2 acts as a regulator of basal progenitor fate, linking regional and laminar specification in the mouse developing cerebral cortex. Recent years have seen great progress in understanding the mecha nisms regulating area and laminar specificity of neurons in the devel oping cerebral cortex1–4. This is of particular relevance as distinct cortical areas differ in their laminar composition1,2,5,6. However, it remains elusive as to whether and how molecular fate determinants regulate laminar composition at the progenitor level and how this is integrated with factors regulating cortical area identity. The cortex is generated largely from two sets of progenitors: basal progenitors, which reside in the subventricular zone (SVZ), a layer that forms above the ventricular zone hosting apical progenitors, and the neuro epithelial cells (NECs) and radial glial cells (RGCs)7. In mice, basal progenitors differ in their mode of division from the apically dividing NECs or RGCs; they usually generate two postmitotic neurons8–10, with only a minority dividing more than once11. Thus, basal progeni tors are mostly direct neuronal progenitors and may serve to expand specific populations of neurons12. The increased number of basal pro genitors at the time of upper neuron generation 6,12,13 and a common transcriptional code with the upper layer neurons 14–17 suggest that basal progenitors at mid-neurogenesis are involved in the generation of upper layer neurons. However, basal progenitors seem to contribute to all cortex layers, as has been shown by video microscopy of Tbr2– green fluorescent protein (GFP) labeled cells18 and the conditional deletion of the transcription factor Tbr2 (also known as Eomes) that
is expressed in basal progenitors from the onset of neurogenesis. Tbr2 deletion results in reduced production of basal progenitors through out the cortex at early stages and consequently leads to a decrease in overall neuron numbers18,19 (but see ref. 20). Conversely, Cux2 is only expressed in basal progenitors at mid-neurogenesis stages and affects only the proliferation of late basal progenitors, thereby increasing upper layer neuron numbers17. Although previous work has shown that some transcription fac tors act as basal progenitor fate determinants, nothing is known about their regulators, particularly the extent to which regulators of basal progenitor fate may differ during the generation of dif ferent cortex layers. Given that basal progenitors arise from apical progenitors8–10,21, it is likely that specification of basal progeni tor fate occurs in the apical progenitors. Indeed, Tbr2 expression seems to be initiated in a subset of apical progenitors, but its upstream regulator is not yet known. Pax6, a transcription fac tor that is exclusively expressed in apical ventricular zone pro genitors and is absent from basal progenitors22,23 may be such an upstream regulator, as basal progenitors are severely misspecified in the Pax6 mutants. Although, the number of basal mitoses is even increased in the Pax6 mutant cortex24, these lack their characteristic expression of Svet1 (ref. 16), Tbr2, Cux1 and Cux2 (refs. 14,15,23). However, these defects may also be secondary to the telencephalic
1Helmholtz
Center Munich, German Research Center for Environmental Health, Institute for Stem Cell Research, Neuherberg/Munich, Germany. 2Division of Molecular Neurobiology, National Institute for Medical Research, Mill Hill, London, UK. 3Helmholtz Center Munich German, Research Center for Environmental Health, Institute of Experimental Genetics, Neuherberg, Germany. 4Physiological Genomics, University of Munich, Munich, Germany. 5Istituto di Neuroscienze, Consiglio Nazionale delle Ricerche, Pisa, Italy. 6Department of Developmental Pathology, Institute for Pathology, University of Bonn Medical School, Bonn, Germany. 7Max-Planck-Institute for Experimental Medicine, Göttingen, Germany. 8Molecular Neuroscience, School of Medicine, Institute of Life Science, Swansea University, Swansea, UK. 9Technical University Munich, Center of Life and Food Sciences Weihenstephan, Freising, Germany. 10University of Connecticut Health Center, Department of Neuroscience, Farmington, Connecticut, USA. 11INSERM, U846, Stem Cell and Brain Research Institute, Bron, France. 12Université de Lyon, UMR-S 846, Lyon, France. 13Munich Center for Integrated Protein Science Munich, CIPSM, Munich, Germany. 14Present address: Life and Health Sciences Research Institute, School of Health Sciences, University of Minho, Braga, Portugal. Correspondence should be addressed to M.G. (
[email protected]) or L.P. (
[email protected]). Received 15 May; accepted 14 July; published online 13 September 2009; corrected after print 25 September 2009; doi:10.1038/nn.2399
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1229
a
b
AP2�
f
d
AP2�
E12
E14
h
Ctx
e
AP2�
AP2�
Ki67 Ctx
v v GE
GE
i
Pax6
AP2γ/Dapi
GE
Adult Ctx
Ctx
v
GE
E18
v v
g
E14
RESULTS AP2 expression in the mouse forebrain AP2γ was specifically expressed in the cerebral cortex of the devel oping mouse embryo, whereas other members of this family were not expressed in this region (Fig. 1a–e; see also refs. 29,30). In the cerebral cortex, AP2γ expression starts around embryonic day 12 (E12), with higher expression levels laterally and rostrally, follow ing the gradient of neurogenesis (Fig. 1f). Expression levels further increased to mid-neurogenesis (E14) in the progenitor layer (Fig. 1f,g) and declined thereafter (Fig. 1f,g). Consistent with our previous data showing an enrichment of AP2γ expression in ventricular zone cells generating basal progenitors21, only a subset of dividing cells labeled by Ki67 were AP2γ immuno positive (Fig. 1h). Most AP2γ-positive cells contained Pax6 (Fig. 1i). We were able to detect Tbr2 and AP2γ double-positive cells in the ventricular zone (Fig. 1j), whereas Tbr2-positive cells in the SVZ were AP2γ negative (Fig. 1j). Thus, AP2γ was expressed in a subset of apical ventricular zone progenitors including the population that starts to express Tbr2 (Supplementary Fig. 1).
v E18
v
patterning defects of the Pax6 mutant telencephalon 25, resulting in aberrant gen eration of GABAergic neurons26. Because many of the molecular characteristics of the basal progenitors are linked to the glutamatergic neuronal lineage23, the loss of the typical basal progenitor transcripts may be secondary to the switch from glutamatergic to GABAergic neuron production in the Pax6 mutant cortex. To identify candidate upstream regulators of basal progenitor fate determinants, we previously established a protocol for enriching RGCs that generate basal progenitors21. Amongst the mRNAs enriched in this RGCs subset, the transcription factor AP2γ caught our attention, as it is also downregulated in the Pax6 mutant cerebral cortex27 and is important for fate determination of the trophecto derm, where Tbr2 is also expressed28. We found that this transcription factor directly regulates basal progenitor fate and does so in a regionspecific manner.
1230
c
AP2�
Ctx
AP2 �
Figure 1 AP2γ expression in the mouse brain. (a–e) ISH of all of the members of the AP2 family of transcription factors, AP2α–ε, on sagittal sections of the embryo at E14. Highmagnification images are shown below. (f) ISH of AP2γ on coronal sections of the cortex at E12, E14 and E18 and from adult cortex. (g) Coronal sections of E14 and E18 cortices immunolabeled for AP2γ. (h–j) Sagittal sections of the E14 cortex immunolabeled for AP2γ and Ki67 (h, high-magnification images of Ki67 and AP2γ double-positive cells depicted by a white arrow and Ki67-positive, AP2γ-negative cells by a green arrow), Pax6 (i, high-magnification images of Pax6-positive, AP2γ-negative cells depicted by a green arrow, AP2γ-positive, Pax6-negative cells depicted by a red arrow and AP2γ and Pax6 double-positive cells depicted by a white arrow) or Tbr2 (j, high-magnification images of AP2γ and Tbr2 double-positive cells depicted by a white arrow and AP2γ-positive, Tbr2-negative cells by a red arrow). Scale bars represent 100 µm (a–f) and 50 µm (g–j). Ctx, cortex; GE, ganglionic eminences; V, ventricle; VZ, ventricular zone.
AP2γ/Dapi
© 2009 Nature America, Inc. All rights reserved.
a r t ic l e s
j
SVZ
VZ v
Tbr2
v
Neurons are reduced in the caudal AP2−/− cortex at E14 The expression pattern of AP2γ suggests that it is involved in Tbr2 upregulation and initial basal progenitor fate specification. To examine this possibility, we inactivated AP2γ from its onset of expression in the developing cortex using mice that express Cre recombinase in the Emx1 locus31 and mice in which exon 5 of the AP2γ gene is flanked by loxP sites (AP2γloxP/loxP)32. In situ hybridization (ISH) and immuno staining revealed that AP2γ mRNA and protein were absent by E12 in the Emx1-cre; AP2γloxP/loxP (henceforth referred to as AP2γ−/−) tel encephalon (Supplementary Fig. 2). Despite the absence of the protein, E12 AP2γ−/− forebrains had a normal cytoarchitecture and normal numbers of progenitors and neurons (Supplementary Fig. 3). Only at E14 did we observe a reduced thickness of the AP2γ −/− cortex in caudal regions (Fig. 2a). No significant differences were detected at rostral levels (P = 0.06; Fig. 2a–g). Although the progenitor layer labeled by Ki67 was comparable in wild-type and AP2γ−/− cortex (Fig. 2b), the band of β3-tubulin–positive neurons was significantly reduced in the caudal (15% reduction, P = 0.02; Fig. 2c and Supplementary Fig. 4), but not rostral, AP2γ−/− cortex compared with wild type (Supplementary Fig. 4). Cell proliferation is not affected by AP2 deletion This reduction in neuron numbers may be a result of progenitors continuing to proliferate instead of generating neurons, defects in the production of the specific progenitor subtypes that generate the neurons (such as the basal progenitors) or defects in fate specifica tion of these progenitors. To discriminate between these possibilities, we first examined proliferation by quantifying all cells in G2/M phase that were labeled by the phosphorylated form of histone 3 (PH3) or by fluorescence-activated cell sorting (FACS)-based measurement of DNA content indicating the phase of the cell cycle. Both analyses revealed that there was no change in the overall proli feration between E14 wild-type and AP2γ−/− cortical cells with VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
© 2009 Nature America, Inc. All rights reserved.
a r t ic l e s
equal numbers of PH3-positive cells (Fig. 2h) and equal numbers of cells in S or G2/M phase (Supplementary Fig. 5). We then quantified the number of cells undergoing cell division at apical or basal positions separately in sections stained for PH3 (Fig. 2d,e). The number of apical mitoses was not affected in the mutant (neither at rostral nor caudal positions), although we observed a slight increase in the number of basally located PH3-positive cells at caudal levels in the AP2γ−/− cortex, that is, in the region where we observed the reduction in neurons (Fig. 2d,e,i,j). As basal progeni tors are only a small subset of all progenitors (30% of all mitoses, n = 1,127), the relatively minor increase in their number (to 40%, n = 1,258) was not sufficient to cause an overall significant increase in the number of PH3-positive cells (P = 0.09; Fig. 2h,j). Moreover, as the number of apical PH3-positive cells was not reduced by the deletion of AP2γ, we conclude that cells dividing basally were not increased in number by a larger production at the expense of apical progenitors. Thus, neither proliferation of apical progenitors nor the production of basal progenitors were affected in the absence of AP2γ.
the AP2γ−/− cortex (Supplementary Fig. 6). Labeling cells with GFPcontaining plasmids from the ventricle by electroporation at E14 revealed an increased proportion of GFP and Tbr2 double-positive cells with an apical process contacting the ventricle in the AP2γ−/− cortex compared with wild-type littermates (note that wild-type levels were normalized to 100%;
AP2 regulates basal progenitor fate Most basal progenitors generate two postmitotic neurons and express the pan-neurogenic gene Tis21 (also known as Btg2)8, indicating their direct neurogenic fate. To examine whether reduced neurogenesis might be explained by a reduction in the number of neurogenic basal progenitors, we examined the expression of Tis21GFP (Tis21 knockin mice expressing GFP) in AP2γ mutants and their wild-type litter mates. The amount of basally dividing cells (labeled with PH3) that also coexpressed Tis21GFP was significantly reduced in the caudal cortex of AP2γ−/− mice compared with wild types both in absolute numbers (wild type = 7 ± 1.2 and AP2γ−/− = 5.5 ± 0.7 PH3 and Tis21GFP double-positive cells per 500-µm-wide radial stripe) and in the percentage of all basal PH3-positive cells (82% of PH3-posi tive cells were Tis21GFP positive in wild type and 63% in AP2γ−/−; Supplementary Fig. 6). Thus, reduced Tis21 expression of basal pro genitors in the caudal AP2γ−/− cortex may explain the reduction in neurogenesis that we observed at this position. To determine the identity of these non-neurogenic basal pro genitors in the AP2γ−/− cortex, we examined the cells for two hall marks of RGCs, Glast (also known as Slc1a3) and an apical process contacting the ventricle. Although virtually no basally located PH3-positive cells were GLAST positive in wild-type cortex (1%, n = 573), almost a quarter (23%, n = 690) were GLAST positive in nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1231
a r t ic l e s a
c
f
d
b
Math3
/WT linear fold change
AP2�
–/–
Fluc/Rluc (normalized)
Lmo4
/WT linear fold change
Figure 3 Regulation of basal progenitor –/– Tbr2 promoter AP2� Wild type Math3 Svet1 Tbr2 Lmo4 Neurod1 Cux2 transcripts by AP2γ. (a,b) mRNAs predicted 5 Math3 promoter ** –1 3.9 to be significantly changed in their expression 4.5 –2 * in the caudal (a) and rostral (b) cortex by 3.4 * ** 4 * 3.1 –3 Affymetrix analysis (light gray bars) and by 3.5 * real-time PCR (dark gray bars in panels a and –4 3 b depict the linear fold change of mRNA levels ** RT-PCR Affymetrix –5 −/− 2.5 between wild-type and AP2γ cortices after ** 1.9 2 * normalization to Gapdh mRNA levels) comparing * 1.5 1.6 Math3 Svet1 Tbr2 Lmo4 Cux2 Neurod1 –1 wild-type and AP2γ−/− cortices. The values for 1.5 1.0 genes expressed at lower levels in the AP2γ−/− 1 –2 compared with wild-type cortices were set to 0.5 –3 be negative. We carried out t tests for all of 0 –4 the genes that we analyzed (four wild-type and four AP2γ−/− mice, both three litters, Student’s –5 * RT-PCR Affymetrix t test, * P ≤ 0.05, ** P ≤ 0.01). (c–e) ISH on −/− E14 wild-type and AP2γ cortex sections of three selected mRNAs with different expression levels between wild-type and AP2γ−/− cortices in the microarrays analysis. (f) Histogram depicting the luciferase luminescence intensity normalized to Renilla intensity from Neuro2A cells transduced with the firefly or Renilla luciferase constructs (Fluc or Rluc, respectively) using either Tbr2 or Math3 promoters. Values were normalized to the pMXIG empty vector containing only GFP (four independent experiments). Error bars represent s.e.m. Scale bars represent 50 µm.
c LTR
IRES
gfp
LTR
b LTR
d
AP2 �
IRES
gfp
e
IRES-gfp at E14–17
LTR
Percentage of Tbr2+ cells to total GFP+ cells
a
40%
AP2 �-IRES-gfp
IRES-gfp
**
30% 20% 10% 0%
Rostral
Caudal
f
AP2�-IRES-gfp at E14–17
AP2�-IRES-gfp at E14–17
GFP/Tbr2/Dapi
V V
AP2�-IRES-gfp at E14–P2
Satb2/GFP
IRES-gfp at E14–P2
Tbr1/GFP
h
1232
100%
**
80% 60% 40% 20% 0%
j
In utero OE at E14 to P2
** IRES-gfp
AP2 �-IRES -gfp
Layers 2/3 Layer 4 Layers 5/6
Layer markers+ per GFP+ cells (500 µm) (normalized to WT)
i
6
+
Pa x
6
Pa x
To examine whether AP2γ may directly regulate these basal progeni tors transcripts, we transfected mouse neuroblastoma cells (Neuro2A) with a pGL3 luciferase vector containing either Math3 or Tbr2 promoters (which contain AP2γ binding sites) and with a control or AP2γ cDNA expression vector. Transfection with AP2γ led to a significant
V
g
AP 2γ
pM XI G
AP 2γ
these basal progenitor–specific transcripts was significantly affected in its expression in rostral regions of the AP2γ−/− cortex (P > 0.05; Fig. 3b). Notably, all of these changes occurred in the absence of alterations in telencephalic patterning (Supplementary Fig. 7 and Supplementary Table 1).
Percentage of GFP+ in layers 2/3/4/5/6 per total GFP+ cells (500 µm)
© 2009 Nature America, Inc. All rights reserved.
AP2�
–/–
Svet1
e
In utero OE at E14 to P2 250% 200%
**
150%
**
IRES-gfp AP2 �-IRES-gfp
100%
**
50% 0% Cux1
Satb2
Ctip2
Tbr1
Figure 4 AP2γ overexpression in the developing cortex. (a,b) Schematic drawing of the retroviral constructs used for control (IRES-gfp in a) and overexpression of AP2γ (AP2γ-IRES-gfp in b). (c–j) Injections of the above constructs (a,b) into the ventricle of E14 wild-type embryos and analysis at E17 (d–f) or P2 (g–h). The histogram in c depicts the quantification of the percentage of GFP-positive control cells and cells infected with AP2γ-IRES-gfp that were immunopositive for Tbr2 in E17 cortices (rostral: IRES-gfp, n = 773 cells; AP2γ-IRES-gfp, n = 1,179 cells; caudal: IRES-gfp, n = 894; AP2γIRES-gfp, n = 1,241; five mice per genotype, three litters). Coronal sections of the E17 cortices were immunolabeled for GFP, Tbr2 and DAPI (GFP and Tbr2 double-positive cells are indicated by white arrows, shown at higher magnifications below; d–f). Sagittal sections of the P2 cortices were immunolabeled for GFP, Satb2 (g) and Tbr1 (h) (GFP-positive cells colocalized with neuronal layer markers are indicated by white arrows and examples are shown at a higher magnification to the right). The histograms in i and j depict the percentage of control and AP2γ-IRESgfp–transduced GFP-positive cells localized in the cortical layers II–VI (i; IRES-gfp, n = 645; AP2γ-IRES-gfp, n = 887; four mice per genotype, three litters) and double-staining for antigens specific for neurons in the respective layers II–VI (j, values were normalized to those obtained in control infected GFP-positive cells; IRES-gfp, n = 452; AP2γ-IRES-gfp, n = 372; both four mice, three litters). Error bars represent s.e.m. Student’s t test, * P ≤ 0.05, ** P ≤ 0.01. Scale bars represent 50 µm (d–f) and 100 µm (g,h).
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
a r t ic l e s a
d
Cux2
Cux2
Figure 5 Upper layer neuron defects in adult AP2�–/– AP2�–/– Wild type Wild type AP2γ−/− mice. (a,b) Sagittal sections of Fr Fr 2-month-old wild-type and AP2γ−/− mouse Par Par cortices labeled for Cux2 mRNA in layers II/III (a) and for Er81 in layer V (b) (white bars M1 M1 S1 S1 represent the width of the Cux2-expressing Ins Ins layers in wild type, and the red bar in a depicts the width of the Cux2-expressing layers in the occipital cortex of the AP2γ−/− cortex). Fr Fr Par (c) Histogram depicts the measurement of Par Cing Cing S1 the width of the Cux2-positive band of layer S1 II/II neurons in the primary motor (M1), Ins somatosensory (S1), auditory (A1) and visual Ins −/− (V1) areas in wild-type and AP2γ cortices 1.5 0.95 1.06 1.03 as a ratio between mutant and control * 1 0.69 (three mice per genotype, three litters). 0.5 (d–f) Coronal sections of wild-type and 0 AP2γ−/− adult cortices at rostral (d) to occipital (f) V1 V1 M1 S1 A1 V1 regions labeled for Cux2 mRNA (white A1 A1 Occipital cortex and yellow bars represent the width of the Hippocampus Cux2-expressing layers in wild type, the red bar Fr in f depicts the width of the Cux2-expressing layers in the primary visual area of the AP2γ−/− cortex and the red arrows delineate distinct cortex areas). Error bars represent s.e.m. Student’s t test, * P ≤ 0.05. Scale bars represent 100 µm. Cing, cingulate cortex; Fr, frontal cortex; Ins, insular cortex; Par, parietal cortex.
b
Cux2
Er81
e
Cux2 width AP2�–/– versus WT
c
activation of both promoters (Tbr2, P = 0.009; Math3, P = 0.04; Fig. 3f). Consistent with the downregulation of Tbr2 and Math3 in the Pax6 mutant cortex, we also observed significant activation of these promoters by Pax6, but no cooperative effects with AP2γ (Tbr2, P = 0.005; Math3, P = 0.04; Fig. 3f). To examine whether AP2γ may be sufficient to upregulate basal progenitor transcripts in vivo, we injected Moloney murine leukemia virus–based retroviral vectors containing IRES-gfp or AP2γ-IRES-gfp (Fig. 4a,b) into the telencephalic ventricle of E14 embryos. By 3 d postinjection, the proportion of GFP-positive cells that coexpressed Tbr2 on overexpression of AP2γ in the caudal cortex had almost doubled, whereas Tbr2 expression was not increased rostrally (Fig. 4c–f). When GFP-positive cells were examined at postnatal stages (Fig. 4g–j), more GFP-positive cells were located in layers II/III, as defined by DAPI and Cux1 or Satb2 expression3, and expressed these transcription factors after AP2γ overexpression compared with controls (Fig. 4g,i,j and data not shown). This increase occurred mostly at the expense of neurons in BrdU at E14–P7 AP2�–/–
e
vGlut2/BrdU Sorla/BrdU
g
Dapi/Tbr1
Layers V–VI
d
f Layers II–IV
c
AP2�–/–
Wild type
Dapi/Cux1
b
Percentage of BrdU+ cells per 500 µm (normalized to WT)
Wild type
Layers II/III
Cux2/BrdU
a
Er81/BrdU
© 2009 Nature America, Inc. All rights reserved.
Cux2
f
120%
Layers II–IV Layers V–VI BrdU negative
100%
*
80% 60% 40% 20% 0%
WT
AP2�–/–
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
the lower layers V/VI (IRES-gfp, 5.3 ± 1.9 GFP and Cux1 double-positive cells per 500-µm-wide radial stripe; AP2γ-IRES-gfp, 11.6 ± 4.3; IRES-gfp, 3 ± 0.1 GFP and Tbr1 double-positive cells per 500-µmwide radial stripe; AP2γ-IRES-gfp, 1.8 ± 0.6; Fig. 4h–j). Thus, AP2γ overexpression at mid-neurogenesis was sufficient to increase the number of Tbr2-positive progenitors and induce increased generation of upper layer neurons in vivo in the caudal, but not rostral, cortex. Consistent with the region- and time-specific defects that we observed in AP2γ−/− cortices, AP2γ overexpression at E12 had no effect on basal progenitors (Supplementary Fig. 8). Layer II/III neurons are reduced in the caudal AP2−/− cortex Given the effects of AP2γ overexpression on laminar fate, we exam ined the later consequences of conditional AP2γ deletion. Notably, the width of the Cux2- (Fig. 5a) and Cux1-expressing (Supplementary Fig. 9) layers II/III was reduced in the adult occipital AP2γ−/− cor tices, whereas layer II/III neurons at other rostro-caudal levels were not affected (Fig. 5). In contrast with the significant reduction that we observed in the band of layer II/III neurons in V1 of the AP2γ−/− cortex (P = 0.03), we found no difference in the primary motor (M1), somatosensory (S1) and auditory (A1) cortices as compared with wild type (Fig. 5c–f, areas were defined as described in ref. 33). The lower layers that were labeled by Er81 and Tbr1 were of equal thickness in wild-type and AP2γ−/− occipital cortices (Fig. 5b and Supplementary Fig. 9), suggesting that the decrease in the upper layers is not a result of an increase in the lower layers. Figure 6 Neurogenesis at E14 in the cerebral cortex of AP2γ−/− mice. (a–d) ISH in sagittal sections of P7 wild-type and AP2γ−/− cortices in caudal regions, with BrdU staining in red (BrdU injected at E14). (e,f) Immunolabeling for Cux1 (e) or Tbr1 (f) in P7 wild-type and AP2γ−/− caudal cortices. Note the thinner band of Cux1-positive cells in the AP2γ−/− (red bar in e) compared with the wild-type cortex (white bar in e). (g) The total number of BrdU-positive cells in P7 AP2γ−/− cortices normalized to wild type counted in 500-µm-wide radial stripes and the percentage of BrdU-positive cells colocalized with markers for layer II–VI neurons (wild type, n = 2,639 cells; AP2γ−/−, n = 1,686 cells; three mice per genotype, three litters). Error bars represent s.e.m. Student’s t test, * P ≤ 0.05, ** P ≤ 0.01. Scale bars represent 100 µm.
1233
© 2009 Nature America, Inc. All rights reserved.
a r t ic l e s
Many layer II/III neurons project via the corpus callosum to the contralateral cortex. Consistent with their reduced number, both the size of the corpus callosum (Supplementary Fig. 9) and staining of L1-labeled callosal fibers crossing the midline (Supplementary Fig. 9) were reduced in the AP2γ−/− cortices at medial, but not rostral, levels compared with wild type. To examine whether this only occurs in layer II/III callosal neurons and spares layer V callosal neurons, we injected fluorescent beads into the occipital visual cortex (see Online Methods) and analyzed the retrogradely labeled cells in the contral ateral occipital cortex 1 week later (Supplementary Fig. 9). Although the proportion of retrogradely labeled neurons was significantly reduced in the upper layers of AP2γ−/− cortices (by 50%, P = 0.04), layer V callosal projection neurons were present in normal numbers (Supplementary Fig. 9). These results further confirm that only layer II/III neurons are affected by the loss of AP2γ.
different cell types or whether their mis-specification is so severe that they succumb to cell death. To measure cell death, we stained dissociated cells with propidium iodide, which only labels dead cells. The number of propidium iodide–labeled cells that we found in AP2γ−/− cortices was twofold larger than that in wild-type cortices (Supplementary Fig. 10). TUNEL (Supplementary Fig. 10) and activated caspase 3 immunoreactivity showed that cell death was particularly increased in the ventricular zone/SVZ in the caudal regions of the AP2γ−/− cor tices (Supplementary Fig. 10). Colocalization with basal progenitors and SVZ markers (Neurod1, Svet1 and Cux2; Supplementary Fig. 10) further suggested that basal progenitors or their immediate progeny undergo apoptosis in the AP2γ−/− cortex (Supplementary Fig. 10). Consistently, the percentage of basally located mitotic cells was sig nificantly decreased in the E17 AP2γ−/− cortices at caudal regions (P = 0.04; Supplementary Fig. 10), which is in contrast with the increased basal mitosis that we observed at E14, thereby explaining the reduced neuronal output during the generation of upper layer neurons without any converse increase in other neuronal subtypes.
AP2 regulates neuron production via basal progenitors The reduction that we observed in the number of upper layer neu rons without an increase in other neuronal subtypes indicates that there is an overall reduction in neuronal output that occurs largely, but not exclusively, in the second half of corticogenesis, when most of the upper layer neurons are generated. We birth dated neurons by injecting BrdU at E14 and examining the colocalization of BrdU pro tein with cortical layer marker mRNAs at postnatal day 7 (P7). The overall number of neurons generated in AP2γ−/− cortices was lower than in wild-type cortices (Fig. 6). Notably, however, upper layer neu rons are also generated in small numbers at earlier stages34,35, which is consistent with the small decrease in neuron numbers that we found at E14 (Fig. 2c). Furthermore, the proportion of cells undergoing S phase at E14 and exiting the cell cycle (BrdU positive, Ki67 negative) at E15 was reduced by 19% in the occipital AP2γ−/− cortex (wild type, n = 1,709; AP2γ−/−, n = 1,301), which is consistent with the idea that the decrease in neuron generation occurs at the times when most upper layer neurons are generated. This decrease in neurogenesis occurred at the expense of upper layer neurons that expressed Cux2, vGlut2 or Sorla (also known as Sorl1) (Fig. 6a–c,e,g) and did not affect lower layer neurons the expressed Er81 or Tbr1 (Fig. 6d,f,g). Thus, loss of AP2γ results in mis-specification of basal progenitors at mid-neurogenesis and, consequently, in a reduction in the number of upper layer neuron numbers in the occipital cortex (Supplementary Fig. 1). This reduced neuronal output resulting from aberrantly specified basal progenitors prompts the question of whether they generate
Functional defects in the AP2−/− visual cortex The marked reduction of upper layer neuron numbers in the primary visual cortex of AP2γ−/− mice provided us with an opportunity to examine the functional consequences of these alterations in visual processing. Visual evoked potentials (VEPs) from the binocular visual cortex were recorded blind to genotype at a depth of 400 µm into the cortex in response to patterned stimuli. The overall amplitude of VEPs was not substantially different between the genotypes (wild type = 198 ± 13 µV; AP2γ−/− = 171 ± 17 µV), whereas the cortical spatial resolution (visual acuity), a well-established measure of overall visual function in mammals36, was severely impaired in the AP2γ−/− visual cortex (Fig. 7a,b). In contrast, other physiological parameters, such as retinotopy, contrast threshold and temporal resolution, were in the normal range (Supplementary Table 2). Notably, AP2γ−/− mice also showed alterations in cortical binocularity (Fig. 7c,d and Supplementary Table 2) and a tendency toward an increased latency of visual response (wild type = 110.0 ± 3.8 ms, AP2γ−/− = 127.2 ± 6.4 ms; t-test, P = 0.05; Supplementary Table 2). As the alterations of these visual parameters were reminiscent of a physiologically more immature state of the visual cortex37, we tested the hypothesis of whether this is also accompanied by mainte nance of a higher degree of plasticity, as is the case at more immature stages. Indeed, monocular deprivation for 3 d caused a significant
1234
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
a r t ic l e s change in binocularity in adult AP2γ−/− (P = 0.01), but not wild-type (P = 0.365), mice (Fig. 7d). Closure of the contralateral eye produced a robust shift of ocular dominance toward the open, ipsilateral eye in the AP2γ−/− cortex (Fig. 7d), whereas no significant variation could be detected in wild-type mice. These data suggest that cortical con nections are more malleable in the cortex of adult AP2γ−/− mice.
© 2009 Nature America, Inc. All rights reserved.
DISCUSSION AP2γ deletion caused a phenotype in which upper layer neuron defects were restricted to the visual cortex. This specificity of the AP2γ−/− cor tex phenotype allowed us to examine the functional consequences of a selective alteration in the number of upper layer neurons in the visual cortex and to determine the uniquely specific role of AP2γ in cortex development.
rather in a subset of RGCs before basal progenitor fate acquisition. AP2γ expression levels are increased in the subset of apical progenitors that generate Tbr2-expressing progenitors21 and the basal progeni tor deficits in the AP2γ−/− cortex highlight the crucial role of AP2γ in the upregulation of various basal progenitor fate determinants at the stages when upper layer neurons are generated. Loss of AP2γ impaired upregulation of virtually all known basal progenitor and SVZ transcripts, such as Tbr2, Math3, Svet1, Lmo4, Cux2 and Neurod1, and both our luciferase and gain-of-function experiments indicate that AP2γ directly regulates the basal progenitor fate determinants Tbr2 and Math3.
AP2 as upstream regulator of basal progenitor fate The mis-specification of basal progenitors in the absence of AP2γ is a result of its role as an upstream regulator of basal progenitor fate deter minants in the apical ventricular zone progenitors (Supplementary Fig. 1). AP2γ was not expressed in basal progenitors themselves, but
Region- and layer-specific functions of AP2 Both the functional defects and the basal progenitor transcript reduc tion were restricted to the occipital cortex. These data suggest that the reduced expression of the single basal progenitor transcript Math3 in rostral regions of AP2γ−/− cortices had little effect on the specification and survival of basal progenitors in rostral areas. No phenotype could be detected in the rostral AP2γ−/− cortex during development, which is consistent with the fact that there are a normal number of neurons in these regions in adulthood. These data evidence a certain degree of redundancy amongst the basal progenitor fate determinants, as only the coordinated reduction of several of these transcription factors results in mis-specification of basal progenitors and the consequent development of a layer phenotype. Indeed, the basal progenitor phe notype observed in the AP2γ−/− cortex comprises aspects observed when Cux2 is deleted (increased basal mitoses17) and parts of the phenotype observed after Tbr2 and Insm1 deletion (reduced neuro genic basal progenitors18–20,39 and reduced expression of additional basal progenitor and SVZ transcripts such as Svet1 and Neurod1)18–20. However, the phenotype of the latter two mouse mutants comprises all stages of cortex development and thus affects all of the layers and regions of the cortex, a pronounced difference to the highly specific phenotype of AP2γ−/− cortices, which were restricted in time to the upper layers and in region to the occipital cortex. Moreover, although Tbr2 and Cux2 were expressed and thus acted in the basal progenitors themselves, AP2γ was only expressed in apical progenitors and acted upstream of the above basal progenitor fate determinants. AP2γ itself is regulated by Pax6 (Supplementary Fig. 11) and Pax6 can directly promote expression of Tbr2 and Math3. Our data sug gest a model of partial redundancy between Pax6 and AP2γ in the regulation of basal progenitor fate determinants, indicating that the absence of AP2γ may be more compensated in rostral regions where Pax6 expression is high25, but less so in caudal regions where its expression is low (Supplementary Fig. 1). Such a model would also be consistent with the region-specific bias in the function of AP2γ in overexpression experiments, with a much weaker effect in rostral regions of the cortex, possibly as a result of the saturating effects of Pax6 and/or other transcription factors in rostral cortex. This model is also consistent with the lack of compensation in the Pax6 mutant cortex, which results in a much more severe phenotype. As both AP2γ and Ngn2 are also targets of Pax6 and are severely downregulated in the Pax6 mutant cortex25,27, they can no longer compensate for defects in the regulation of basal progenitor fate determinants, such as Tbr2 (ref. 23), Svet1 (ref. 16) and Math3 (ref. 40). Moreover, these basal progenitor fate determinants may be repressed by transcription factors that are normally restricted to the ventral telencephalon, such as Gsh1, Gsh2 and Mash1, the latter of which inhibits expression of AP2γ (Supplementary Fig. 11). Although the available data are so far consistent with such a model of redundancy, the region- and time-specific cofactors required for
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1235
The role of AP2 in specifying basal progenitor fate Detailed analysis allowed us to discriminate between the effects of AP2γ on proliferation, production and the fate of basal progenitors. Cell cycle analysis and quantification of the cells in mitoses revealed that there were a normal number of progenitors without any overall cell cycle aberrations. Thus, in contrast with its role in carcinoma cells38, AP2γ appears to not regulate cell cycle parameters in the developing brain. The production of cells dividing at basal posi tions was also not impaired; there was even a slight increase in the number and proportion of basal mitoses in occipital AP2γ−/− cortex at mid-neurogenesis. However, these basally located progenitors were severely mis-speci fied and their mis-specification also explains their initial increase in number. Both, the number of Tbr2-positive progenitors and the expression of Tbr2 and several other basal progenitor–specific tran scripts were reduced in the occipital AP2γ−/− cortex, with hardly any transcriptional alterations being detected at rostral regions in the AP2γ−/− cortex. Basal progenitors at occipital positions also aberrantly maintained RGCs characteristics, such as Glast immunoreactivity and an apical process contacting the ventricle, which is rarely the case in wild-type cortex. These basal progenitors were impaired in neu rogenesis, as evidenced by the severe downregulation of a reporter monitoring expression of the neurogenic transcription factor Tis21. Indeed, we observed a decrease in cell cycle exit and in generation of postmitotic cells, indicating that there was a substantial reduction in neuronal output starting around E13–14 in the occipital AP2γ−/− cortex. Thus, mis-specified basal progenitors in the absence of AP2γ first increase in number as they generate fewer neurons and aber rantly re-enter the cell cycle. This is consistent with the reduction of neurons that has been observed at E14 and the generation of at least small numbers of upper layer neurons by E13 (ref. 34). Notably, immunostaining for Tbr1 and Cux1 at E14 indicated that there was a selective reduction of Cux1-positive neurons in the cortical plate, whereas Tbr1-positive neurons were not affected (data not shown). At later stages, the number of basal progenitors undergoing apoptosis in the occipital AP2γ−/− cortex ultimately (E16–17) led to a depletion of basally dividing cells. Thus, our analysis established that the cause for selective reduction in upper layer neurons in the occipital cortex is the mis-specification of basal progenitors, which generated fewer neurons at and after mid-neurogenesis.
© 2009 Nature America, Inc. All rights reserved.
a r t ic l e s transcriptional regulation by AP2γ or differential availability of promoter elements may equally contribute to the specific function of AP2γ. For example, changes in DNA methylation occurring during corticogenesis may not only affect the Gfap promoter and glial differentiation41, but could also control the timing of the generation of different neuronal subtypes and the activity of the transcription factors involved, potentially in cooperation with other changes in chromatin organization. Thus, AP2γ binding sites in several target gene promoter regions could be inaccessible at earlier stages of corti cal development. Moreover, the region- and time-specific role of AP2γ function, and thus its transcriptional activity, may also be influenced by extrinsic signaling pathways, for example, resulting in the differ ential phosphorylation of AP2γ, as this has been shown to influence the transcriptional activity of several transcription factors including Pax6 (ref. 42) and neurogenin43. A variety of mechanisms that remain to be further elucidated may thus contribute to the region- and timespecific function of AP2γ in the developing cerebral cortex. However, this is also not unprecedented, as both Pax6 and Ngn2 are expressed throughout corticogenesis, but exert only the development of specific cortex neurons located in the upper and lower layers, respectively44. AP2γ not only acts in a region-specific manner, but is also time or layer specific. The loss of AP2γ only affects basal progenitor fate at mid-neurogenesis (E14) and the same effect was observed on overex pression of AP2γ. At early developmental stages (E12), AP2γ does not affect basal progenitors or neuronal fate, thus acting in a temporalspecific manner. This is consistent with the effects of AP2γ loss and gain of function on upper layer neurons and implies that this molecu lar network can only influence basal progenitor fate around mid-neu rogenesis. Our data suggest a mechanism by which basal progenitor fate is altered when upper layer neurons are generated. This molecular network influencing basal progenitor fate in their ventricular zone ancestors differs profoundly from the transcriptional mechanisms regulating the fate of specific types of projection neurons that mostly act in the postmitotic neurons at later stages of development3 Quantitative changes in neuron numbers during phylogeny A hallmark of the context-dependent function of AP2γ is that its absence results in quantitative changes in neurons of specific laminar phenotypes in a region-specific manner. Such quantitative changes are of functional relevance, as highlighted by the alterations in visual processing in the AP2γ−/− mice, but also by the fact that the numbers of specific neuronal subtypes, but not their identities, are altered during evolution. The selective reduction in the number of upper layer neurons in the visual cortex of AP2γ−/− mice resulted in a profound reduction in vis ual acuity and an abnormal binocularity. These functional deficits are reminiscent of a more immature state, as visual acuity and VEP con tralateral-ipsilateral ratios are lower in juvenile mice37,45. Moreover, the enhanced plasticity of the adult mutant mice is typical of a juvenile cortex and indicates that sensitivity to monocular deprivation is not downregulated in the adult AP2γ−/− cortices to the same extent as in wild-type cortex that contains more upper layer neurons. Although it was previously thought that these plastic processes are limited to the critical period, it has now become clear that alterations in cortical binocularity can still occur in adult rodents46. This plasticity has been linked to inhibitory circuits in the cortex46. Our data also implicate layer II/III callosal projection neurons in this phenomenon. Indeed, the physiological phenotype of AP2γ−/− mice resembles the alterations observed after inactivation of interhemispheric connections during development47, suggesting that the number of callosal neurons may act as an important determinant in the functional maturation of visual 1236
processing, a feature that is particularly critical in higher mammals with a huge increase in upper layer neurons and a further elaborated visual system. Notably, AP2γ was also expressed in primate and human cortical progenitors in the visual cortex when upper layer neurons are generated and its expression levels were substantially higher in progenitors of area 17 compared to 18 (Supplementary Fig. 12). Moreover, different from the mouse, AP2γ was expressed specifically in upper layer neurons of adult monkey (data not shown) and human cortex (Supplementary Fig. 13). As upper layer neuron numbers are particularly increased in area 17 of primate and human cortex 6, it is tempting to speculate that AP2γ may be one of the molecular compo nents regulating the number of upper layer neurons during ontogeny and phylogeny in the visual cortex. Methods Methods and any associated references are available in the online version of the paper at http://www.nature.com/natureneuroscience/. Accession numbers. We submitted our microarray data to Gene Expression Omnibus (GSE12134). Note: Supplementary information is available on the Nature Neuroscience website. Acknowledgments We are very grateful to M. Moser, S. Pfaff, C. Schuurmans and Y. Gotoh for in situ probes and to R. Jäger for providing reagents. We thank T. Öztürk, A. Steiner, A. Waiser and D. Franzen for excellent technical assistance. H.S. was supported by the Deutsche Forschungsgemeinschaft. M.G. was supported by the Deutsche Forschungsgemeinschaft, Bundesministerium für Bildung und Forschung and the Bavarian government. L.P. is supported by the Portuguese Fundaçäo para a Ciência e Tecnologia/European Social Fund. AUTHOR CONTRIBUTIONS L.P. did most of the experimental work. D.D. contributed to the in vitro studies and to immunostaining in the embryonic cortex. M.-T.S. contributed to the in utero injections. J.N. contributed to the luciferase assay. M.I. and J.B. conducted the microarrays analyses. M.S.B. contributed to the beads injections. L.R., L.G., C.C. and M.C. conducted the visual functional analysis. S.N.W. and H.S. generated the AP2γ conditional knockout mice. V.T. contributed with antibodies. K.B. conducted the adult human analysis. F.G. prepared the Mash1 construct and provided the Ngn2KiMash1 mice. N.Z. conducted the embryonic human analyses. C.D. conducted the embryonic monkey analyses. M.G. supervised the project and wrote the manuscript together with L.P. Published online at http://www.nature.com/natureneuroscience/. Reprints and permissions information is available online at http://www.nature.com/ reprintsandpermissions/.
1. Rash, B.G. & Grove, E.A. Area and layer patterning in the developing cerebral cortex. Curr. Opin. Neurobiol. 16, 25–34 (2006). 2. O’Leary, D.D., Chou, S.J. & Sahara, S. Area patterning of the mammalian cortex. Neuron 56, 252–269 (2007). 3. Molyneaux, B.J., Arlotta, P., Menezes, J.R. & Macklis, J.D. Neuronal subtype specification in the cerebral cortex. Nat. Rev. Neurosci. 8, 427–437 (2007). 4. Sur, M. & Rubenstein, J.L. Patterning and plasticity of the cerebral cortex. Science 310, 805–810 (2005). 5. Polleux, F., Dehay, C., Goffinet, A. & Kennedy, H. Pre- and post-mitotic events contribute to the progressive acquisition of area-specific connectional fate in the neocortex. Cereb. Cortex 11, 1027–1039 (2001). 6. Dehay, C. & Kennedy, H. Cell-cycle control and cortical development. Nat. Rev. Neurosci. 8, 438–450 (2007). 7. Götz, M. & Huttner, W.B. The cell biology of neurogenesis. Nat. Rev. Mol. Cell Biol. 6, 777–788 (2005). 8. Haubensak, W., Attardo, A., Denk, W. & Huttner, W.B. Neurons arise in the basal neuroepithelium of the early mammalian telencephalon: a major site of neurogenesis. Proc. Natl. Acad. Sci. USA 101, 3196–3201 (2004). 9. Miyata, T. et al. Asymmetric production of surface-dividing and non–surface dividing cortical progenitor cells. Development 131, 3133–3145 (2004). 10. Noctor, S.C., Martinez-Cerdeno, V., Ivic, L. & Kriegstein, A.R. Cortical neurons arise in symmetric and asymmetric division zones and migrate through specific phases. Nat. Neurosci. 7, 136–144 (2004).
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
© 2009 Nature America, Inc. All rights reserved.
a r t ic l e s 11. Wu, S.X. et al. Pyramidal neurons of upper cortical layers generated by NEX-positive progenitor cells in the subventricular zone. Proc. Natl. Acad. Sci. USA 102, 17172– 17177 (2005). 12. Martínez-Cerdeño, V., Noctor, S.C. & Kriegstein, A.R. The role of intermediate progenitor cells in the evolutionary expansion of the cerebral cortex. Cereb. Cortex 16, 152–161 (2006). 13. Lukaszewicz, A. et al. The concerted modulation of proliferation and migration contributes to the specification of the cytoarchitecture and dimensions of cortical areas. Cereb. Cortex 16 (Suppl 1): i26–i34 (2006). 14. Nieto, M. et al. Expression of Cux-1 and Cux-2 in the subventricular zone and upper layers II–IV of the cerebral cortex. J. Comp. Neurol. 479, 168–180 (2004). 15. Zimmer, C., Tiveron, M.C., Bodmer, R. & Cremer, H. Dynamics of Cux2 expression suggests that an early pool of SVZ precursors is fated to become upper cortical layer neurons. Cereb. Cortex 14, 1408–1420 (2004). 16. Tarabykin, V., Stoykova, A., Usman, N. & Gruss, P. Cortical upper layer neurons derive from the subventricular zone as indicated by Svet1 gene expression. Development 128, 1983–1993 (2001). 17. Cubelos, B. et al. Cux-2 controls the proliferation of neuronal intermediate precursors of the cortical subventricular zone. Cereb. Cortex 18, 1758–1770 (2008). 18. Kowalczyk, T. et al. Intermediate neuronal progenitors (basal progenitors) produce pyramidal-projection neurons for all layers of cerebral cortex. Cereb. Cortex published online, doi:10.1093/cercor/bhn260 (23 January 2009). 19. Sessa, A., Mao, C.A., Hadjantonakis, A.K., Klein, W.H. & Broccoli, V. Tbr2 directs conversion of radial glia into basal precursors and guides neuronal amplification by indirect neurogenesis in the developing neocortex. Neuron 60, 56–69 (2008). 20. Arnold, S.J. et al. The T-box transcription factor Eomes/Tbr2 regulates neurogenesis in the cortical subventricular zone. Genes Dev. 22, 2479–2484 (2008). 21. Pinto, L. et al. Prospective isolation of functionally distinct radial glial subtypes— lineage and transcriptome analysis. Mol. Cell Neurosci. 38, 15–42 (2008). 22. Götz, M., Stoykova, A. & Gruss, P. Pax6 controls radial glia differentiation in the cerebral cortex. Neuron 21, 1031–1044 (1998). 23. Englund, C. et al. Pax6, Tbr2, and Tbr1 are expressed sequentially by radial glia, intermediate progenitor cells and postmitotic neurons in developing neocortex. J. Neurosci. 25, 247–251 (2005). 24. Haubst, N. et al. Molecular dissection of Pax6 function: the specific roles of the paired domain and homeodomain in brain development. Development 131, 6131–6140 (2004). 25. Stoykova, A., Treichel, D., Hallonet, M. & Gruss, P. Pax6 modulates the dorsoventral patterning of the mammalian telencephalon. J. Neurosci. 20, 8042–8050 (2000). 26. Kroll, T.T. & O’Leary, D.D. Ventralized dorsal telencephalic progenitors in Pax6 mutant mice generate GABA interneurons of a lateral ganglionic eminence fate. Proc. Natl. Acad. Sci. USA 102, 7374–7379 (2005). 27. Holm, P.C. et al. Loss- and gain-of-function analyses reveal targets of Pax6 in the developing mouse telencephalon. Mol. Cell. Neurosci. 34, 99–119 (2007).
28. Werling, U. & Schorle, H. Transcription factor gene AP-2 gamma essential for early murine development. Mol. Cell. Biol. 22, 3149–3156 (2002). 29. Zhao, F., Lufkin, T. & Gelb, B.D. Expression of Tfap2d, the gene encoding the transcription factor Ap-2 delta, during mouse embryogenesis. Gene Expr. Patterns 3, 213–217 (2003). 30. Chazaud, C. et al. AP-2.2, a novel gene related to AP-2, is expressed in the forebrain, limbs and face during mouse embryogenesis. Mech. Dev. 54, 83–94 (1996). 31. Iwasato, T. et al. Cortex-restricted disruption of NMDAR1 impairs neuronal patterns in the barrel cortex. Nature 406, 726–731 (2000). 32. Werling, U. & Schorle, H. Conditional inactivation of transcription factor AP-2gamma by using the Cre/loxP recombination system. Genesis 32, 127–129 (2002). 33. Ferrere, A., Vitalis, T., Gingras, H., Gaspar, P. & Cases, O. Expression of Cux-1 and Cux-2 in the developing somatosensory cortex of normal and barrel-defective mice. Anat. Rec. A Discov. Mol. Cell. Evol. Biol. 288, 158–165 (2006). 34. Roy, K. et al. The Tlx gene regulates the timing of neurogenesis in the cortex. J. Neurosci. 24, 8333–8345 (2004). 35. Gillies, K. & Price, D.J. The fates of cells in the developing cerebral cortex of normal and methylazoxymethanol acetate-lesioned mice. Eur. J. Neurosci. 5, 73–84 (1993). 36. Gianfranceschi, L., Fiorentini, A. & Maffei, L. Behavioural visual acuity of wild type and bcl2 transgenic mouse. Vision Res. 39, 569–574 (1999). 37. Fagiolini, M., Pizzorusso, T., Berardi, N., Domenici, L. & Maffei, L. Functional postnatal development of the rat primary visual cortex and the role of visual experience: dark rearing and monocular deprivation. Vision Res. 34, 709–720 (1994). 38. Li, H., Goswami, P.C. & Domann, F.E. AP-2gamma induces p21 expression, arrests cell cycle, and inhibits the tumor growth of human carcinoma cells. Neoplasia 8, 568–577 (2006). 39. Farkas, L.M. et al. Insulinoma-associated 1 has a panneurogenic role and promotes the generation and expansion of basal progenitors in the developing mouse neocortex. Neuron 60, 40–55 (2008). 40. Mattar, P. et al. Basic helix-loop-helix transcription factors cooperate to specify a cortical projection neuron identity. Mol. Cell. Biol. 28, 1456–1469 (2008). 41. Fan, G. et al. DNA methylation controls the timing of astrogliogenesis through regulation of JAK-STAT signaling. Development 132, 3345–3356 (2005). 42. Kim, E.A. et al. Phosphorylation and transactivation of Pax6 by homeodomaininteracting protein kinase 2. J. Biol. Chem. 281, 7489–7497 (2006). 43. Cundiff, P. et al. ERK5 MAP kinase regulates Neurogenin1 during cortical neurogenesis. PLoS One 4, e5204 (2009). 44. Schuurmans, C. et al. Sequential phases of cortical specification involve neurogenindependent and -independent pathways. EMBO J. 23, 2892–2902 (2004). 45. Cancedda, L. et al. Acceleration of visual system development by environmental enrichment. J. Neurosci. 24, 4840–4848 (2004). 46. Spolidoro, M., Sale, A., Berardi, N. & Maffei, L. Plasticity in the adult brain: lessons from the visual system. Exp. Brain Res. 192, 335–341 (2008). 47. Caleo, M. et al. Transient synaptic silencing of developing striate cortex has persistent effects on visual function and plasticity. J. Neurosci. 27, 4530–4540 (2007).
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1237
ONLINE METHODS
© 2009 Nature America, Inc. All rights reserved.
Animal and human samples. We maintained AP2γloxP/loxP and Emx1-cre on a C57Bl/6J background (also used as wild type) and identified the genotypes by PCR of genomic DNA32. loxP sites flanked exon 5, whose deletion caused a loss of the helix-span-helix domain near the protein carboxy terminus. We crossed AP2γloxP/loxP mice with Emx1-cre and Tis21-gfp knockin mice8 and considered the day of vaginal plug as E0 and the day of birth as P0. We used parasagittal sections of E80 cynomolgus macaque monkey cortex and human sections of the occipital pole with 20–22 gestational weeks (mid-gestation) for antibody staining. We carried out staining on 25-µm-thick sections and fixed the brains with 4% paraformaldehyde (wt/vol) and cryopro tected with 20% sucrose (wt/vol). We obtained the postmortem adult human brain tissue for this study from the Neurological Foundation of New Zealand Human Brain Bank. The University of Auckland Human Subject Ethics Committee approved the protocols used in these studies, and all tissue was obtained with the full consent of the subjects’ families. We are very grateful to R. Faull (University of Auckland) for providing the samples. Immunocytochemistry. We fixed cultures in 4% paraformaldehyde for 15 min at 20–23 °C and performed staining as described previously22. We used primary antibodies to β3-tubulin 1:100, (Sigma, mouse IgG2b), GFP (1:500, RDI, rabbit), Pax6 (1:300, Chemicon, rabbit), PH3 (1:200, Upstate, rabbit), GLAST (1:400, Chemicon, guinea pig), Ki67/Tec3 (1:50, DAKO, rat), Tbr2 (rabbit, 1:2,000, kind gift of R. Hevner, Seattle Children’s Hospital Research Institute), NeuN (1:50, Chemicon, mouse IgG), CTIP2 (1:100, Abcam, rabbit), Tbr1 (1:100, Abcam, rab bit), Satb2 (1:1,000, provided by V.T.), Cux1 (1:100, Santa Cruz, rabbit), BrdU (1:100, Abcam, rat IgG2a), AP2γ (specific to an N-terminal epitope, 1:100, Santa Cruz, rabbit; Abcam, mouse IgG1), activated caspase-3 (1:500, Promega, rabbit) and L1 (1:100, kind gift of T. Tilling, Universitätsklinikum Hamburg-Eppendorf, rabbit). We incubated primary antibodies overnight at 4 °C and used subclassspecific secondary antibodies labeled with Cy2/FITC, Cy3/TRITC (Jackson ImmunoResearch) or biotin (detected by AMCA-coupled streptavidin, Cy3coupled streptavidin or Alexa 488-coupled streptavidin; Vector Laboratories) for detection. Plasmids and viral production. We cloned the full-length cDNA of mouse AP2γ (1,816 bp, EMBL accession number 94694, kindly provided by M. Moser, Max Planck Institute of Biochemistry) into the EcoRI unique restriction site of the retroviral vector pMXIG between the upstream long terminal repeat and the IRES sequence. We used the empty retroviral vector pMXIG as a control. We transiently transfected the packaging cell line GPG-23 with the retroviral expres sion plasmid to produce replication-incompetent virus. Viral titers were typically 106−107 per ml. Real-time PCR analysis. We synthesized cDNA from the RNA samples used for microarray analysis using the SuperScript II First-Strand Synthesis System (Invitrogen). Real-time PCR was performed as described previously21. We car ried out real-time PCR according to the manufacturer’s recommendations using SYBRGreen master mixes from Qiagen on a DNA Engine Opticon machine (Biorad) and determined the primer dimer melting temperatures to exclude primer dimers from the analysis. We used Gapdh as the housekeeping gene to normalize the target gene’s expression and calculated the relative expression as E = ½(–∆Ct), where Ct is the difference between the threshold of cycle number of Gapdh and the gene analyzed for reaching the threshold. The threshold is defined at the onset of the exponential phase. ISH. We carried out ISH on frozen tissue with digoxigenin-labeled riboprobes as described previously21. The in situ probe that we used to detect AP2γ (1,816 bp, EMBL accession number 94694) was kindly provided by M. Moser. Dual-luciferase reporter assay. We constructed expression plasmids for our luci ferase reporter assay using full-length cDNAs of mouse AP2γ, Pax6, Pax5a and PDless cloned into the pMXIG vector and the full-length cDNA of Mash1 cloned into the pcDNA expression vector. We used the respective empty vectors as con trols. The promoters of Tbr2, Math3 and AP2γ were cloned into the pGL3 vector (Promega). The Tbr2 promoter was constructed using the sequence GP_683 from Genomatix and the coding sequence (816 bp) removed, the Math3 promoter
nature NEUROSCIENCE
was constructed using the complete sequence from the Genomatix promoter GP_216893 (775 bp) and the AP2γ promoter was constructed using the pro moter GP_203628 from Genomatix and the coding region (789 bp). We plated mouse Neuro2A and human embryonic kidney cells on DMEM media with 10% fetal calf serum (vol/vol), 1:100 penicillin/streptomycin and 25 mM HEPES (1 × 105 cells per 24 wells) and transfected them with expression plasmids (3 µg), with 1.5 µg of the firefly (Photinus pyralis) luciferase reporter under either the Tbr2 or Math3 or AP2γ promoter and 0.1 µg of the pRL-TK plasmid encoding Renilla (sea pansy, Renilla reniformis) luciferase (Promega). We changed media after 24 h and prepared the cell extracts on the following day. We normalized the relative light units to Renilla luciferase activity and expressed the results as the ratio of firefly to Renilla luciferase activity. The luciferase activity was measured with the Dual-Luciferase Reporter Assay System (Promega) using a luminometer (Berthold Centro LB 960). Values represent the mean ± s.e.m. of three experiments. FACS analysis. Preparation of cells was performed as described in ref. 21. For cell death analysis, we immediately stained the cells with propidium iodide for 5 min and analyzed in the FACS purity mode. For β3-tubulin and cell cycle analysis, we fixed cells in 70% methanol (vol/vol) overnight at 20 °C, washed with phosphatebuffered saline and re-suspended in phosphate-buffered saline containing 1% FCS. We stained the DNA of cells with propidium iodide (final concentration of 1 mg ml–1) for 5 min and analyzed them via FACS in the purity mode. RNA isolation and microarray analysis. We analyzed three biological replicates (wild-type and AP2γ−/− cortices of three different litters divided into rostral and caudal cortex regions). We isolated total RNA from cortical tissue of E14 wild-type and AP2γ−/− embryos using RNeasy Mini kit (Qiagen) including DNase treatment. RNA quality examination and microarray analysis were performed as described previously21. We examined RNA quality with the Agilent 2100 Bioanalyzer and only used high-quality RNA for amplification with the MessageAmp II-Biotin Enhanced Single Round aRNA Amplification kit (Ambion). We did hybridization of Affymetrix MOE430 2.0 arrays and scanning according to standard protocols provided by Affymetrix (http://www.affymetrix.com). For statistical analysis of the expression data, we used the ChipInspector software (Genomatix) and further filtered the significantly regulated genes (false discovery rate < 5%) for expres sion level (>50 arbitrary units) and fold change (>1.5-fold difference), as we found this approach to be useful in previous experiments. Filtering was done on normalized Mas5 data of significant genes, because the ChipInspector software does not provide absolute expression levels. Any provided ratio and expression level is derived from Mas5 data. In utero injections. We anesthetized and injected with viral vectors timed preg nant mice with embryos at a gestational age of E14, as described in ref. 48. We killed the injected embryos 3 d after injection or as pups 2–5 d after birth. We subjected E14 wild-type and AP2γ−/− embryos to in utero electroporation using the pCIG2-gfp plasmid49 and carried out the analyses 1 d later (E15). For electro poration conditions, we used five 50-V pulses spaced by 200 ms, which were applied with 5-mm tweezer-style electrodes (Protech) using a BT square-wave electroporator (Harvard Apparatus). Beads injections. We injected red RetroBeads (100 nl, LumaFluor) into the mid line representation of the visual cortex (stereotaxic coordinates, 2.8 mm lateral to midline on the apical progenitor level of Lambda) of wild-type and AP2γ−/− mice. Each injection was first delivered 600 µm below the cortical surface with small amounts injected continually until a level 200 µm below the surface was reached. We killed the mice by perfusion 1 week later and analyzed the contra lateral hemisphere. BrdU labeling. We injected timed pregnant mice (E14) intraperitoneally with BrdU and analyzed them 1 or 3 d later or at postnatal stages P2 or P5 (birth dating analysis). Electrophysiology. We used 14 naive adult wild-type and AP2γ−/− mice and per formed the measurements blind to genotype. Transient VEPs were recorded as described previously50. In all experiments, we inserted the electrode 2.6–3.5 mm lateral to the lambda to map the entire binocular visual field. The electrode was advanced 400 µm below the cortical surface, where VEPs had their maximal
doi:10.1038/nn.2399
amplitude. We quantified the VEP amplitude by measuring the peak through amplitude and collected the responses to 0% contrast to measure noise level. Visual acuity was assessed after presentation of gratings of variable spatial frequencies (90% contrast). Contrast threshold was evaluated in response to 0.06 cycles per degree gratings. We measured temporal resolution with gratings of 0.06 cycles per degree and 90% contrast. We measured acuity, temporal reso lution and contrast threshold as the highest spatial frequency, highest temporal frequency and lowest contrast, respectively, that evoked a VEP response greater than the mean value of the noise. We determined VEP depth profile, cortical retinotopy and ocularity as described previously50. For monocular deprivation, one eyelid was sutured under avertin (tribromo ethanol, Sigma) anesthesia as previously described47 in an additional group of ten adult (3.5–4.5-months-old) wild-type and AP2γ−/− mice. We checked the mice daily to ensure that the suture remained intact. We prepared the mice for recording as described above after 3 d of monocular deprivation. We determined contralateral-ipsilateral VEP ratios in three tracks made in correspondence with the cortical representation of the vertical meridian.
48. Costa, M.R., Bucholz, O., Schroeder, T. & Gotz, M. Late origin of glia-restricted progenitors in the developing mouse cerebral cortex. Cereb. Cortex 19, 135–143 (2009). 49. Hand, R. et al. Phosphorylation of Neurogenin2 specifies the migration properties and the dendritic morphology of pyramidal neurons in the neocortex. Neuron 48, 45–62 (2005). 50. Porciatti, V., Pizzorusso, T. & Maffei, L. The visual physiology of the wild-type mouse determined with pattern VEPs. Vision Res. 39, 3071–3081 (1999).
© 2009 Nature America, Inc. All rights reserved.
Data analysis. We carried out all quantifications in the almost entire mediolateral extent of the caudal cortex, reaching from the ventricular curve at the border to the ganglionic eminence to the ventricular curve to the very medial cortex,
hippocampal anlage and hem. We considered mitoses as basal when they occurred more than five cell diameters from the apical surface. We quantified the number of PH3-positive cells using Neurolucida software 6.0 (MicroBrightField). We con sidered rostral sections as the first cortical slides not containing the hem and cau dal sections contained the hippocampus. We carried out all other quantifications using single confocal optical sections. We counted and calculated the number of PH3-immunopositive cells and caspase 3–positive cells as the number of cells per µm2. In the overexpression experiments and beads injections, we analyzed only cryosections from the dorsal telencephalon that contained targeted cells. For most of the quantifications, we analyzed at least three different litters of mice. For each single embryo, the quantification included 6–20 sections from rostral or caudal levels. We determined statistical significance using Student’s t test.
doi:10.1038/nn.2399
nature NEUROSCIENCE
CORRI G EN D U M
Corrigendum: AP2γ regulates basal progenitor fate in a region- and layer-specific manner in the developing cortex Luisa Pinto, Daniela Drechsel, Marie-Theres Schmid, Jovica Ninkovic, Martin Irmler, Monika S Brill, Laura Restani, Laura Gianfranceschi, Chiara Cerri, Susanne N Weber, Victor Tarabykin, Kristin Baer, François Guillemot, Johannes Beckers, Nada Zecevic, Colette Dehay, Matteo Caleo, Hubert Schorle & Magdalena Götz Nat. Neurosci. 12, 1229–1237 (2009); published online 13 September 2009; corrected after print 25 September 2009.
© 2009 Nature America, Inc. All rights reserved.
In the version of this article initially published, one of the corresponding authors’ email addresses was misspelled. It should be luisapinto@ecsaude. uminho.pt. In addition, errors occurred in some of the numbers listed in the last subsection of the Results section. Instead of “Notably, AP2γ−/− mice also showed alterations in cortical binocularity (Fig. 7c,d and Supplementary Table 2) and a tendency toward an increased latency of visual response (wild type = 109.95 ms, AP2γ−/− = 127.19 ms; Supplementary Table 2). […] Indeed, monocular deprivation for 3 d caused a significant change in binocularity in adult AP2γ−/− (P = 0.027), but not wild-type (P = 0.365), mice (Fig. 7d),” the affected sentences should read, “Notably, AP2γ−/− mice also showed alterations in cortical binocularity (Fig. 7c,d and Supplementary Table 2) and a tendency toward an increased latency of visual response (wild type = 110.0 ± 3.8 ms, AP2γ−/− = 127.2 ± 6.4 ms; t-test, P = 0.05; Supplementary Table 2). […] Indeed, monocular deprivation for 3 d caused a significant change in binocularity in adult AP2γ−/− (P = 0.01), but not wild-type (P = 0.365), mice (Fig. 7d).” The errors have been corrected in the HTML and PDF versions of the article.
nature neuroscience
articles
SOX6 controls dorsal progenitor identity and interneuron diversity during neocortical development
© 2009 Nature America, Inc. All rights reserved.
Eiman Azim1–3, Denis Jabaudon1–3,5, Ryann M Fame1–4 & Jeffrey D Macklis1–3 The neuronal diversity of the CNS emerges largely from controlled spatial and temporal segregation of cell type-specific molecular regulators. We found that the transcription factor SOX6 controls the molecular segregation of dorsal (pallial) from ventral (subpallial) telencephalic progenitors and the differentiation of cortical interneurons, regulating forebrain progenitor and interneuron heterogeneity. During corticogenesis in mice, SOX6 and SOX5 were largely mutually exclusively expressed in pallial and subpallial progenitors, respectively, and remained mutually exclusive in a reverse pattern in postmitotic neuronal progeny. Loss of SOX6 from pallial progenitors caused their inappropriate expression of normally subpallium-restricted developmental controls, conferring mixed dorsal-ventral identity. In postmitotic cortical interneurons, loss of SOX6 disrupted the differentiation and diversity of cortical interneuron subtypes, analogous to SOX5 control over cortical projection neuron development. These data indicate that SOX6 is a central regulator of both progenitor and cortical interneuron diversity during neocortical development.
Two broad functional classes of cortical neurons, excitatory projection neurons and inhibitory interneurons, arise from spatially and molecularly segregated pallial (dorsal) and subpallial (ventral) proliferative ventricular zones of the telencephalon, respectively1–3. Parcellation of these proliferative regions into molecularly segregated domains separated at the pallial-subpallial boundary (PSB) is critical for the generation of these distinct classes of neurons. In these broad excitatory and inhibitory neuronal classes, tremendous subtype diversity arises largely from the dynamic temporal expression of progenitor and postmitotic transcriptional regulators. Both of these developmental mechanisms (inter- and intra-domain segregation of molecular regulators) combine to give rise to the extraordinary neuronal diversity of the adult mammalian brain. The parcellation of the proliferative neuroepithelium at the PSB is defined and maintained by the interactions of several critical early patterning transcription factors, exemplified by the repressive interaction of pallium-expressed Neurogenin2 (Ngn2, also known as Neurog2) on the generally subpallium-expressed Mash1 (also known as Ascl1)1. Accordingly, loss of Ngn2 function results in dorsal expansion of Mash1 expression and a consequent ventralization of pallial progenitors, which aberrantly give rise to subpallial-like neurons4,5. The dynamic interaction between this pair of transcription factors exemplifies the delicate balance of molecular regulators that are required to establish and maintain the PSB. Throughout corticogenesis, these pallial and subpallial progenitors give rise to neurons whose fates depend largely on the location and time
at which they are born3,6–9. In the pallium, distinct excitatory projection neuron subtypes are born sequentially under the control of temporally coordinated programs that guide their subtype specification and differentiation3. Simultaneously, inhibitory cortical interneurons, which constitute approximately 25% of all cortical neurons, are primarily born in the subpallial medial (MGE) and caudal ganglionic eminences (CGE)2. Acquisition of distinct interneuron subtype identities, distinguishable by molecular, morphological and electrophysiological phenotypes, depends on both the place and time of birth in the MGE and CGE2,6–12. Differentiating interneurons then migrate tangentially toward and then radially into the cortex to populate their final laminar destinations alongside concurrently born pallium-derived excitatory projection neurons2,13. Because cortical interneurons are implicated in several developmental disorders14, including epilepsy15, autism16 and schizophrenia17, understanding the molecular controls over their subtype diversity might clarify some causes of and potential therapeutic approaches to these important disorders. Although major progress has been made in understanding the regulation of broad aspects of neuronal heterogeneity during development1, only recently have specific controls over excitatory3,18–26 and inhibitory27–31 cortical neuron subtype differentiation been characterized. We recently reported that SOX5 postmitotically controls the sequential generation of distinct pallium-derived excitatory corticofugal projection neuron populations, regulating their subtype diversity22,26. Motivated by the complementary and largely redundant functions of SOX5 and SOX6 in other systems32,33, we hypothesized that
1Massachusetts General Hospital–Harvard Medical School Center for Nervous System Repair, Departments of Neurosurgery and Neurology, Program in Neuroscience, Harvard Medical School, Boston, Massachusetts, USA. 2Nayef Al-Rodhan Laboratories, Massachusetts General Hospital, Boston, Massachusetts, USA. 3Department of Stem Cell and Regenerative Biology, and Harvard Stem Cell Institute, Harvard University, Cambridge, Massachusetts, USA. 4Department of Molecular and Cellular Biology, Harvard University, Cambridge, Massachusetts, USA. 5Present address: Department of Basic Neurosciences and Clinic of Neurology, University of Geneva, Switzerland. Correspondence should be addressed to J.D.M. (
[email protected]).
Received 26 May; accepted 22 July; published online 5 August 2009; doi:10.1038/nn.2387
1238
volume 12 | number 10 | OCTOBER 2009 nature neuroscience
articles
E13.5 E15.5
E10.5
E13.5 E15.5
© 2009 Nature America, Inc. All rights reserved.
E10.5
SOX6 might also function in the generation of Sox6 SOX5 forebrain neuronal diversity. a b c SOX6 and SOX5 belong to the SRY-type HMG D box (SOX)-containing transcription factor A P Ctx family, which is composed of 20 members in V mammals, many of which have precise temporal LGE CGE and spatial functions in cell-fate specification MGE and differentiation in multiple organ systems including the CNS34,35. SOX6 and SOX5, which share 93% identity in their HMG DNA-binding domains and 61% overall identity36, interact d e and functionally overlap during chondrogenesis and oligodendroglial development in the spinal cord. During chondrogenesis, SOX6 and SOX5 are coexpressed in prechondrocytes, where they have overlapping and additive roles in promoting appropriate and timely differentiation into chondroblasts. The loss of either gene alone produces mild skeletal defects and perinatal death, and the loss of both genes results in major cartilage dysgenesis and death during late gestation32. Similarly, SOX6 and SOX5 are coexpressed in developing oligodendroglia in the spinal cord, where they act as functionally equivalent repressors of specification and terminal differentiation33. SOX6 is expressed in the forebrain during mid-gestation, as seen by whole-mount in situ hybridization36, and in the early postnatal brain, as determined by northern Figure 1 SOX6 and SOX5 are expressed in complementary populations of telencephalic progenitors blot and real-time quantitative RT-PCR37, but and neuronal progeny during corticogenesis. (a) Schematic illustrating the relative positions of the neocortex (Ctx), LGE, MGE and CGE in the developing brain. A, anterior; D, dorsal; P, posterior; its cell type–specific expression and function in V, ventral. (b,c) Sox6 (b, black arrow) was expressed in a slight dorsal-high ventral-low gradient the brain have not been investigated. (analyzed by in situ hybridization) and SOX5 (c, black arrow) was expressed in a ventral-high dorsalWe found that, in contrast with their low gradient (analyzed by immunocytochemistry) in the telencephalon (white dotted circles) at overlapping expression and largely redundant E10.5, as corticogenesis is beginning. Insets show higher-magnification view of the telencephalon. functions in other systems, SOX6 and SOX5 (d,e) During corticogenesis, shown here at E13.5 and E15.5, Sox6 (d, in situ hybridization) was were almost entirely mutually exclusively expressed in progenitors of the pallial ventricular zone (red arrows) and in postmitotic neurons in the MGE and CGE mantle zones (red arrowheads), but it was not expressed in subpallial ventricular zone expressed in the forebrain and had distinct, progenitors (blue arrows). SOX5 (e, immunocytochemistry) was expressed in subpallial ventricular complementary functions. SOX6 and SOX5 zone progenitors (blue arrows) and postmitotic neurons in the cortical plate (blue arrowheads), but it were complementarily expressed in pallial was not expressed in pallial ventricular zone progenitors (red arrows). Panel a is adapted from ref. 13. and subpallial progenitors, respectively, and Scale bars represent 100 µm and 150 µm for E.13.5 and E.15.5, respectively. this expression was reversed in differentiating postmitotic neurons, as progeny of subpallial progenitors (at least largely composed of cortical interneurons) postmitotic subpallial neurons, and SOX5 in subpallial progenitors and expressed SOX6, and corticofugal projection neuron progeny of pallial postmitotic pallial corticofugal projection neurons22 (Fig. 1a–e). Early progenitors expressed SOX5. During development, SOX6 controlled in corticogenesis, SOX6 was expressed in the telencephalon in a slight the segregation of pallial from subpallial progenitors by repressing the dorsal-high to ventral-low gradient (Fig. 1b), whereas SOX5 was expressed expression of Mash1 and downstream subpallium-specific programs in a spatially reciprocal ventral-high to dorsal-low gradient (Fig. 1c). in pallial progenitors. Postmitotically, SOX6 regulated multiple aspects During mid- to late corticogenesis, SOX6 and SOX5 were mutually of cortical interneuron differentiation, ultimately controlling the exclusively expressed in pallial (SOX6) and subpallial (SOX5) ventricular molecular diversity of cortical interneuron subtypes. We conclude that zone progenitors (Fig. 1d,e and Supplementary Fig. 1). Their expression SOX6 and SOX5 have independent and complementary roles in the overlapped exclusively in a discrete portion of the dorsal subpallial generation of neuronal diversity during neocortical development. ventricular zone at the PSB (Supplementary Fig. 1), a region that gives rise to the lateral cortical stream, populating basal telencephalic structures, RESULTS including the amygdala and piriform cortex38,39. SOX6 and SOX5 are mutually exclusively expressed The postmitotic progeny of pallial and subpallial progenitors mutually To determine whether SOX6 and SOX5 have complementary or exclusively expressed SOX6 and SOX5 in a reverse pattern; SOX6 was interactive roles during neocortical development, we characterized their expressed in the MGE and CGE mantle zones, which contain, among expression at important stages of corticogenesis. In situ hybridization and other neuronal populations, developing cortical interneurons that immunocytochemistry revealed that SOX6 and SOX5 were expressed maintain SOX6 expression as they mature in the neocortex, whereas in complementary and almost entirely mutually exclusive populations SOX5 was expressed by corticofugal projection neurons in the cortical of progenitors and cortical neurons: SOX6 in pallial progenitors and plate22 (Fig. 1d,e and Supplementary Fig. 1). Notably, SOX6 was not nature neuroscience volume 12 | number 10 | OCTOBER 2009
1239
articles Figure 2 SOX6 and SOX5 are cross-repressive SOX5 Sox6 in pallial and subpallial telencephalic progenitor –/– WT Sox6 WT Sox5 –/– domains. (a) SOX5 expression (analyzed by immunocytochemistry), which normally extended to the ventral edge of the pallial-subpallial boundary (blue arrows in wild type, WT) and was absent from pallial ventricular zone progenitors, ectopically expanded into Sox6–/– pallial ventricular zone progenitors (blue arrows in Sox6–/–) at E13.5 and P0. This SOX5 expansion was most pronounced near the PSB at E13.5 and extended evenly throughout the entire pallial ventricular zone by P0. (b) Conversely, Sox6 expression (analyzed by in situ hybridization), which normally extended to the dorsal edge of the PSB (red arrows in wild type) and was absent from subpallial progenitors, ectopically expanded into Sox5–/– subpallial ventricular zone progenitors (red arrows in Sox5–/–) at E13.5 and P0. This Sox6 expansion was most pronounced near the PSB in the LGE at E13.5 and extended throughout the entire subpallial ventricular zone by P0. Scale bars represent 50 µm.
P0
expressed in the mantle zone of the lateral ganglionic eminence (LGE), where medium spiny neurons that populate the striatum will later mature (Fig. 1d). Immunocytochemical analysis of the S-phase marker BrdU, the pan-mitotic marker PCNA, and the M-phase marker phospho-histone 3 (PH3) revealed that subpallial expression of SOX6 was overwhelmingly postmitotic (Supplementary Fig. 2). Taken together, these data indicate that SOX6 and SOX5 are expressed in spatially abutting, almost entirely non-overlapping populations of progenitors and postmitotic neurons, suggesting cross-repressive interactions during development. SOX6 and SOX5 progenitor expression is cross-repressive We hypothesized that if cross-repressive interactions exist, either direct or indirect, loss of either SOX6 or SOX5 would result in the corresponding
a
Mash1 E13.5 Sox6 –/–
WT
Sox6 –/–
Sox6 –/–
WT
Sox6 –/–
ectopic expression of the other. Accordingly, loss of SOX6 function in Sox6–/– mice32 resulted in an expansion of SOX5 expression into the pallial ventricular zone that normally expresses SOX6 (Fig. 2a). During corticogenesis, this ectopic SOX5 expression in Sox6–/– pallium gradually expanded from the lateral to the medial extent of the pallial ventricular zone, suggesting a developmental gradient of SOX5 expression. Conversely, in Sox5–/– mice32, SOX6 expression expanded ventrally into the subpallial ventricular zone during neocortical development (Fig. 2b). This crossrepressive interaction was restricted to progenitors and was not apparent in postmitotic neurons that expressed SOX6 or SOX5, indicating that progenitor-specific, and possibly indirect, interactions are occurring between these two transcription factors, most likely in coordination with other progenitor patterning genes. Notably, the dorsal subpallial ventricular zone coexpressed SOX6 and SOX5 (Supplementary Fig. 1), indicating that there is a unique relationship between the two transcription factors in this distinct developmental domain and suggesting that the expression of either one alone is not sufficient to repress the expression of the other. To directly investigate whether SOX6 and SOX5 are sufficient to repress the expression of each other in progenitors, we mis-expressed Sox6 in the subpallial ventricular zone and Sox5 in the pallial ventricular zone via in utero electroporation at embryonic day 12.5 (E12.5) for analysis at E16.5. Many progenitors transfected with one of the genes continued to express the other gene, which is not surprising, given their normal coexpression in a discrete
CGE
MGE/LGE
Ngn2 E13.5
c
Mash1
Sox6
Ngn2 –/–; Ngn1 –/–
WT
E14.5
WT
1240
WT
CGE
MGE/LGE
WT
b
E14.5
© 2009 Nature America, Inc. All rights reserved.
P0
E13.5
b
E13.5
a
Ngn2 –/–; Ngn1–/–
Figure 3 Loss of SOX6 function results in ectopic proneural gene expression in pallial progenitors and subpallial mantle zones. (a) Mash1, which was normally restricted to subpallial ventricular zone progenitors and was not expressed by pallial ventricular zone progenitors (red arrows) or by postmitotic neurons in subpallial mantle zones (red arrowheads), was ectopically expressed in Sox6–/– pallial ventricular zone progenitors and in the MGE and CGE mantle zones at E13.5. (b) Ngn2, which was normally restricted to pallial ventricular zone progenitors and was not expressed by postmitotic neurons in subpallial mantle zones (red arrowheads), was ectopically expressed in Sox6–/– MGE and CGE mantle zones at E13.5. (c) As previously reported4, Mash1, which is normally not strongly expressed in pallial ventricular zone progenitors, was ectopically expressed in progenitors of the Ngn2–/–; Ngn1–/– pallial ventricular zone at E14.5 (red arrows). (d) As in the wild type, Sox6 continued to be expressed in Ngn2–/–; Ngn1–/– pallial ventricular zone progenitors (red arrows). Sox6 was ectopically expressed in Ngn2–/–; Ngn1–/– postmitotic pallium-derived neurons in the cortical plate (red arrowheads), consistent with the ectopic expression of subpallial postmitotic signals in pallium-born neurons in the absence of Ngn2 function, as previously described4,5. All expression was analyzed by in situ hybridization. Scale bars represent 100 µm.
volume 12 | number 10 | OCTOBER 2009 nature neuroscience
articles
© 2009 Nature America, Inc. All rights reserved.
region of dorsal subpallial progenitors (Supplementary Fig. 1), as well as in other developing systems32,33. Taken together, these data indicate that SOX6 and SOX5 are necessary, but not sufficient, to repress the expression of each other in forebrain progenitors, strongly suggesting that there are combinatorial interactions with other regional patterning signals during telencephalic development. SOX6 and Ngn2 cooperatively control pallial identity The complementary and mutually exclusive expression of SOX6 and SOX5 in forebrain progenitor domains is highly reminiscent of the generally non-overlapping expression of the patterning transcription factors Ngn2 (pallial) and Mash1 (subpallial)4,40. Loss of Ngn2 causes ectopic expansion of Mash1 expression into pallial progenitors, activating downstream subpallial differentiation programs4,5. Because the patterns of SOX6 and Ngn2 expression in telencephalic progenitors are similar, we examined whether the loss of SOX6 function would result in a similar ventralization of the pallium. Indeed, loss of SOX6 caused a marked expansion of Mash1 expression into the pallial ventricular zone throughout corticogenesis (Fig. 3a and Supplementary Fig. 3). Olig2, a transcription factor that is also expressed by subpallial progenitors during corticogenesis, was also ectopically expressed in Sox6–/– pallial ventricular zone (Supplementary Fig. 3). This domain-parcellating function is specific for SOX6, as loss of SOX5 function did not cause a reciprocal ventral expansion of palliumspecific Ngn2 expression, and simultaneous loss of both SOX6 and SOX5 function largely replicated the phenotype of Sox6–/– mice (Supplementary Fig. 3). These data indicate that SOX6 functions centrally in the molecular segregation of the pallial from the subpallial progenitor domain. We next examined whether the partial ventralization of pallial progenitors in Sox6–/– mice is a result of the disruption of the expression of mostly pallium-restricted Pax6 or its direct downstream target Ngn2 (ref. 41). Pax6 (Supplementary Fig. 3) and Ngn2 (Fig. 3b) were still normally expressed in the Sox6–/– pallium, indicating that their expression was not centrally driven by SOX6. Because Ngn2 is known to normally repress Mash1 (ref. 4), and because this repression is lost in Sox6–/– pallial progenitors (where Ngn2 and Mash1 are abnormally coexpressed; Fig. 3a,b), we hypothesized that SOX6 maintains pallial identity either in and transcriptionally activated by the well-described Pax6-Ngn2-Mash1 pathway, mediating Ngn2 repression of Mash1, or in a previously undefined genetic cascade that does not require the Ngn2 pathway for transcriptional activation. To discriminate between these two possibilities, we examined the expression of SOX6 in the abnormally ventralized pallium of Ngn2–/–; Ngn1–/– mice4, in which the repression of Mash1 by Ngn2 (supplemented by additive repression by Ngn1) is lost4 (Fig. 3c). Loss of SOX6 expression in the Ngn2–/–; Ngn1–/– pallium would suggest that SOX6 is a downstream transcriptional target of Ngn signaling, acting in this canonical pathway. The data exclude this alternative, as SOX6 expression was maintained in Ngn2–/–; Ngn1–/– pallium (Fig. 3d). Just as Ngn2 was normally expressed in the pallial ventricular zone in the absence of SOX6, SOX6 was normally expressed in the absence of Ngn2. These data indicate that the cooperative convergence of both SOX6 and Ngn2 pathways is necessary to repress Mash1 and maintain the dorsal identity of pallial progenitors, although neither is sufficient on its own. In the absence of either of these critical regulators, pallial progenitors adopt a mixed dorsal-ventral identity, inappropriately coexpressing genes that are normally specific to one or the other developmental domain. Despite the partial ventralization of Sox6–/– pallial progenitors, projection neuron laminar distribution and subtype- and layer-specific molecular expression appeared to be largely normal (Supplementary Fig. 4). Similarly, pallial progenitor proliferation was not affected by loss of SOX6 function, as assessed by BrdU uptake and PH3 expression (data not shown). To determine whether the expansion of Mash1 expression into nature neuroscience volume 12 | number 10 | OCTOBER 2009
the Sox6–/– pallium is indicative of the initiation of subpallium-specific programs of gene expression, we carried out comparative microarray analysis between wild-type and Sox6–/– pallium during mid-corticogenesis at E13.5 and identified several normally subpallium-expressed genes that were ectopically expressed in the Sox6–/– pallium, including Dlx1, Dlx2, Dlx4, Gsh2, Isl1, Meis1 and Sox5 (Supplementary Table 1). These data indicate that SOX6 critically maintains pallial progenitor identity by repressing subpallial programs of gene expression, but redundant and/or compensatory controls (for example, Ngn2 and Ngn1) persist that are sufficient to ensure largely appropriate pallial corticogenesis42. We conclude that SOX6 acts cooperatively with Ngn2 to control the segregation of telencephalic progenitor domains during development. SOX6 controls cortical interneuron subtype differentiation SOX6 was also expressed in postmitotic interneurons as they reside in the subpallium and populate the neocortex (Fig. 1d and Supplementary Figs. 1 and 2). We therefore examined whether loss of SOX6 function would affect important sequential steps of interneuron differentiation: early postmitotic molecular identity, cortical laminar location and morphology, and interneuron molecular subtype differentiation. Our data indicate that SOX6 acts postmitotically at all three of these stages of cortical interneuron differentiation, controlling their appropriate development. Because the early molecular programs of immature postmitotic cortical interneurons in the subpallial mantle zones largely determine and predict their appropriate differentiation43, we first examined early cortical interneuron molecular identity. We found that Sox6–/– MGE and CGE mantle zones abnormally expressed the proneural transcription factors Mash1 and Ngn2 (Fig. 3a,b). SOX6 repression of the ectopic and persistent expression of Mash1 (normally progenitor and subpallium specific) and Ngn2 (normally progenitor and pallium specific) in subpallial mantle zones strongly suggests that SOX6 controls the temporal segregation of transcriptional programs between progenitors and postmitotic neurons. Because MGE and CGE Sox6–/– mantle zone cells ectopically expressed Ngn2, which normally represses subpallial and maintains pallial identity4, we hypothesized that these cells might abnormally initiate pallium-like gene expression. Consistent with this hypothesis, Sox6–/– subpallial mantle zone cells inappropriately expressed Vglut2, a vesicular glutamate transporter whose expression is normally restricted to pallium-born excitatory projection neurons (Fig. 4a). This indicates that at least a subpopulation of Sox6–/– subpallial immature neurons in the mantle zone are inappropriately acquiring pallial properties. To determine whether this abnormal coexpression of pallial/subpallial and progenitor/postmitotic molecular regulators (Mash1, Ngn2 and Vglut2) in Sox6–/– immature subpallial neurons affects their ability to broadly differentiate into GABAergic neurons, we examined whether they express GAD67 (also known as Gad1), an enzyme that is necessary for the synthesis of the inhibitory neurotransmitter GABA, a fundamental indicator of their identity. Using Gad67 in situ hybridization (data not shown) and Gad67-gfp (delta-neo) transgenic mice, in which green fluorescent protein (GFP) is expressed in most GABA-positive neurons28,44, we found that GAD67 was expressed in neurons leaving the subpallial mantle zones in Sox6–/– mice (Fig. 4b), suggesting appropriate GABAergic neuron specification. However, as cortical interneurons continued their migration tangentially in one of two streams toward the cortex (a superficial marginal zone stream or a deeper intermediate zone/subventricular zone (SVZ) stream), they failed to migrate properly in Sox6–/– cortex, as indicated by the consistently less advanced leading edge of the marginal zone stream compared with the leading edge of the intermediate zone/SVZ stream (Fig. 4b,c and Supplementary Fig. 5). There was no change in the number of interneurons in either migratory stream at E13.5 (Fig. 4d and Supplementary Fig. 5). From these data,
1241
a
b
E15.5
E13.5
articles c
Distance (µm)
Figure 4 Loss of SOX6 function results in Vglut2 GAD67-GFP P = 0.03 abnormal early cortical interneuron differentiation, WT Sox6 –/– WT Sox6 –/– 200 without a change in interneuron number. 160 (a) Although excitatory neuron–specific Vglut2 120 is not normally expressed in neurons born in the 80 subpallium, it was ectopically expressed in the 40 subpallial mantle zone in Sox6–/– mice at E15.5 0 WT Sox6 –/– (red arrowheads; analyzed by in situ hybridization). (b,c) GAD67-GFP–positive neurons were born 160 WT in both wild-type and Sox6–/– MGE (white Sox6 –/– 120 arrowheads, analyzed by immunocytochemistry). 80 However, as GABAergic cortical interneurons tangentially migrated into the cortex, the leading 40 edge of the marginal zone (MZ) migratory stream 0 MZ IZ/SVZ was consistently less advanced in the Sox6–/– Migratory route cortex compared with the wild-type cortex (white arrows; 60% reduction in distance, P = 0.03; c). Dotted lines (b) indicate lateral ventricle boundary. IZ, intermediate zone. (d) There was no difference between wild-type and Sox6–/– cortex in the number of migrating cortical interneurons in either the marginal zone or intermediate zone/SVZ migratory streams. Scale bars represent 150 µm (low magnification, a), 100 µm (high magnification in a and low magnification in b), and 50 µm (high magnification, b). Results are expressed as mean ± s.e.m.
we conclude that loss of SOX6 function perturbs the initial temporal segregation of progenitor-specific factors from postmitotic neurons, potentially causing their abnormal tangential migration, without affecting overall GABAergic neuron specification and abundance. To further investigate whether the molecular and migratory irregularities at early stages of Sox6–/– subpallial neuron differentiation are associated with abnormalities in subsequent stages of interneuron
a
b
GAD67-GFP P0 WT
cortical invasion, we examined the laminar location and morphology of these interneurons as they populated the cortex. In Gad67-gfp mice at postnatal day 0 (P0; Fig. 5a), just after the interneurons have begun their radial migration into the maturing cortex, and at P14 (Fig. 5b), as they have more fully adopted their mature phenotypes, Sox6–/– interneurons preferentially populated deeper neocortical layers (Fig. 5c,d), without any change in their total numbers. Although the large majority of Sox6–/– mice die perinatally, small numbers survive a few weeks postnatally, which allowed us to GAD67-GFP P14 analyze them at P14 (ref. 32). Sox6+/– mice WT Sox6 –/– survived to adulthood, with a small number having occasional seizure behavior. We also examined the morphology of GAD67-GFPpositive neurons at P0 and found that Sox6–/–
Sox6 –/–
MZ Bin 4 Bin 3 Bin 2
I
Bin 4
CP
II/III Bin 3
V
IV V
Bin 2
VI
VI Bin 1
Bin 1 WM
WM
c
Bin4 (superficial)
P0
P = 0.004 P = 0.05
Bin3
P = 0.04
Bin2 Bin1 (deep)
P = 0.01 0 0.1 0.2 0.3 0.4 Proportion of total GAD67-GFP+ neurons
d
WT Sox6 –/–
Bin4 (superficial)
P14
© 2009 Nature America, Inc. All rights reserved.
Number of GAD67-GFP + neurons
d
Bin3
P = 0.0003
Bin2 Bin1 (deep)
P = 0.001 0 0.1 0.2 0.3 0.4 Proportion of total GAD67-GFP+ neurons
1242
Figure 5 Loss of SOX6 function disrupts the normal laminar position and morphology of cortical interneurons. (a–d) Analysis of Gad67-gfp mice revealed that, although there were equal numbers of cortical interneurons at P0 (a) and P14 (b) in wild-type and Sox6–/– cortex, there was a redistribution of interneurons toward deeper cortical layers in Sox6–/– cortex compared with wild type (c,d). Quantification at P0 (c) revealed a proportional increase in interneuron density in the deepest bin, bin 1, by 13% (P = 0.01) and bin 2 by 5% (P = 0.04), and a proportional decrease in the more superficial bin 3 by 7% (P = 0.05) and bin 4 by 10% (P = 0.004). Quantification at P14 (d) revealed an increase in interneuron density in bin 1 by 10% (P = 0.001) and a decrease in the more superficially located bin 3 by 5% (P = 0.0003). Red lines (a,b) indicate subdivision into four bins for quantification (see Online Methods). Interneurons had abnormal tangential morphology in Sox6–/– cortex compared with the radially oriented interneurons in wild-type cortex (a, red arrowheads). (a,b) immunocytochemistry. CP, cortical plate; I–VI, cortical layers I–VI; WM, white matter. Dotted lines indicate pial surface. Scale bars represent 200 µm (low magnification, a), 50 µm (intermediate magnification, a), 25 µm (high magnification, a), 300 µm (low magnification, b) and 100 µm (high magnification, b). Results are expressed as mean ± s.e.m.
volume 12 | number 10 | OCTOBER 2009 nature neuroscience
articles b
SOX6 / PV
SOX6 / SST / Calret 100
I
d
P14 WT
Percentage of SOX6+ interneurons
P14
Percentage neuron density in Sox6 –/– compared with WT P = 0.0009
270
WT
240
II/III
Sox6 –/–
210 50
180 150
VI
P = 0.03 P < 0.0001 P = 0.001 P = 0.002
al re t+
Sox6 –/–
30
G
G
II/III
nl y O
V
SS
IV
Sox6 –/–
P+
0 +
I
al re t+
+
P+
PY
VI
C
60
VIP P14
WT
lC
Sox6 –/–
N
+
NPY P14
WT
Al
SST Calret
Sox6 –/–
+
WT
VI
SST / Calret P14
PY
Sox6 –/–
N
WT
90
AD 6 FP 7-
PV P14
G
+
WT
G
c
PV
AD 6 FP 7-
0
SS
WM
T+
120
+
SOX6 / VIP
A T + ll S ST an + d SS C al T+ re an t+ d C al re t–
SOX6 / NPY
nl y
IV V
O
SOX6 P14 WT
PV
a
© 2009 Nature America, Inc. All rights reserved.
VI
WM
Figure 6 SOX6 is necessary for cortical interneuron subtype development. (a,b) At P14, SOX6 was expressed in ~65% of all neocortical GAD67-GFP– positive interneurons (analyzed by immunocytochemistry), including essentially all parvalbumin (PV)-positive (86%), SST-positive (96%), SST and calretinin (Calret) double-positive (83%) (white arrowheads), and SST-positive and calretinin-negative (95%) interneurons. SOX6 was expressed in more than onethird of all NPY-positive interneurons (37%) (white arrowheads; open arrowheads indicate NPY-positive, SOX6-negative interneurons), in essentially no VIP-positive interneurons (3%; open arrowheads indicate VIP-positive, SOX6-negative interneurons) and in very few calretinin-positive interneurons (11%; open arrowheads indicate calretinin-positive, SOX6-negative interneurons). Red lines in a indicate approximate regions of magnification in b. (c,d) At P14, parvalbumin-positive cortical interneuron numbers (white arrowheads) were diminished in Sox6–/– compared with wild-type cortex (93% reduction, P < 0.0001), as were SST-positive cortical interneuron numbers (red neurons, white arrowheads; 70% reduction, P = 0.002). There was no change in the number of calretinin-positive interneurons (green neurons, open arrowheads), although the subset of SST and calretinin double-positive interneurons (yellow neurons, white arrows) was reduced in number in Sox6–/– compared with wild-type cortex (79% reduction, P = 0.03). The subset of SST-positive, calretininnegative interneurons was also reduced (70% reduction, P = 0.001). The number of NPY-positive cortical interneurons (white arrowheads) was increased in Sox6–/– compared with wild-type cortex (137% increase, P = 0.0009). There was no change in the number of VIP-positive interneurons (white arrowheads). The positions of the yellow boxes in the low-magnification panels (c) are representative of the neocortical position examined in the high-magnification panels. Quantification is represented as the percentage of neuron density in Sox6–/– compared with wild type. Dotted lines indicate pial surface (a,c). Scale bars represent 300 µm (low magnification, a,c), 100 µm (high magnification, a,c) and 50 µm (b). Results are expressed as mean ± s.e.m.
interneurons had abnormal tangential orientation, in contrast with the mostly radial orientation of wild-type interneurons (reflecting their transition from tangential to radial migration) (Fig. 5a). These laminar distribution and morphological abnormalities were confirmed by analysis of the broad (at P0) interneuron marker calbindin (Supplementary Fig. 6). These data indicate that SOX6 is necessary for the appropriate differentiation of cortical interneurons as they integrate into the neocortical circuitry, manifested by their inappropriately deep laminar location and abnormal morphology in the absence of SOX6. To determine whether SOX6 differentially affects the development of distinct cortical interneuron subtypes, we examined interneuron subpopulations using subtype-defining molecular markers2,11. By P14, the calcium-binding protein parvalbumin and the peptide hormone somatostatin (SST) are expressed by two non-overlapping subclasses of predominantly MGE-born interneurons, many of which are born early in corticogenesis. The peptide neurotransmitter neuropeptide Y (NPY) is expressed by later-born MGE and CGE-derived interneurons, many of which coexpress SST. The peptide hormone vasoactive intestinal peptide (VIP) is predominantly expressed by late-born, CGE-derived interneurons, although a subpopulation of VIP-positive interneurons coexpresses SST and might be of MGE origin. The calcium-binding protein calretinin is predominantly expressed by late-born CGE interneurons, although a subpopulation also coexpresses SST and might be of MGE origin2,6,7,45,46. We found that at P14, ~65% of all GAD67-GFP-positive interneurons expressed SOX6, including essentially all of the parvalbumin-positive (86%) and SST-positive (96%) interneurons (both the calretinin-positive nature neuroscience volume 12 | number 10 | OCTOBER 2009
(83%) and calretinin-negative (95%) subpopulations), and over onethird of the NPY-positive interneurons (37%). Essentially no VIP-positive interneurons (3%) and only a small minority of calretinin-positive interneurons (11%) expressed SOX6 (Fig. 6a,b). Loss of SOX6 function resulted in a marked reduction in the number of interneurons expressing parvalbumin (93% reduction, P < 0.0001) and SST (70% reduction, P = 0.002), including the small subpopulation of SST and calretinin double-positive interneurons (79% reduction, P = 0.03), and the SSTpositive and calretinin-negative interneurons (70% reduction, P = 0.001). Conversely, there was a corresponding marked increase in the number of NPY-positive interneurons (137% increase; P = 0.0009), and no change in the number of VIP- or calretinin-positive interneurons (Fig. 6c,d and Supplementary Fig. 6). Notably, as observed with the general interneuron marker GAD67, all interneuron subtypes that normally express SOX6 inappropriately redistributed to deeper cortical layers in Sox6–/– cortex (Supplementary Fig. 6). At P0, when SST expression is normally seen in lateral neocortex and piriform cortex, there were already markedly reduced numbers of Sox6–/– SST-positive interneurons (Supplementary Fig. 6), indicating that SOX6 function is necessary at early stages of cortical interneuron molecular differentiation. Together, these data indicate that loss of SOX6 function causes a decrease in the abundance of specific molecularly-defined subtypes of cortical interneurons, many of which are normally MGE-derived during early corticogenesis, without affecting overall cortical interneuron number. To investigate potential temporal control by SOX6 over cortical interneuron subtype differentiation, particularly given its preferential
1243
articles Figure 7 Loss of SOX6 function produces P14 NPY / IdU / CIdU P14 an increased number of early- and late-born Percentage of early- and late-born neurons WT Sox6 –/– P = 0.04 NPY-positive cortical interneurons. (a,b) Dual WT I birthdating of cortical interneurons using Sox6 –/– 200 IdU (E11.5) and CldU (E15.5) (analyzed by immunocytochemistry) revealed a decrease in the II/III number of parvalbumin- and SST-positive earlyP = 0.03 and late-born interneurons (early parvalbumin: 150 IV 83% decrease, P = 0.002; early SST: 65% P = 0.04 P = 0.002 P = 0.02 decrease, P = 0.009; late parvalbumin: 88% P = 0.009 V decrease, P = 0.02; late SST: 93% decrease, P 100 = 0.04; b) and a large increase in the number of NPY-positive early- (white arrowheads) and lateVI born (white arrows) interneurons (early NPY: 40% 50 increase, P = 0.03; late NPY: 90% increase, P = 0.04; b) in Sox6–/– cortex. Colocalization was strictly assessed as homogenous, strong nuclear WM 0 IdU or CldU label surrounded by cytoplasmic PV+ SST+ NPY+ PV+ SST+ NPY+ parvalbumin, SST or NPY labeling (NPY/IdU Early (IdU) Late (CIdU) colocalization provided as a representative example in a, white arrowhead). Quantification is represented as the percentage of neuron density in Sox6–/– cortex compared with wild type. Dotted lines in a indicate pial surface. Scale bars represent 100 µm (low magnification, a) and 10 µm (high magnification, a). Results are expressed as mean ± s.e.m.
© 2009 Nature America, Inc. All rights reserved.
a
effects on largely early-born parvalbumin- and SST-positive MGEderived interneurons2,6,7,45, we performed dual birthdating of interneuron subtypes born at E11.5 and E15.5 using the halogenated thymidine analogs iododeoxyuridine (IdU) and chlorodeoxyuridine (CldU), respectively47, for examination at P14 (Fig. 7). We found that in wild-type cortex NPY-positive interneurons were preferentially born during late versus early corticogenesis (about twice as many CldU and NPY double-positive neurons as there were IdU and NPY double-positive neurons, 180%, P = 0.02; Fig. 7a), confirming previous reports46. In addition, although the number of early- and late-born neurons were equivalent between wild-type and Sox6–/– cortices, far fewer Sox6–/– early- and late-born interneurons were positive for parvalbumin or SST (early parvalbumin: 83% decrease, P = 0.002; early SST: 65% decrease, P = 0.009; late parvalbumin: 88% decrease, P = 0.02; late SST: 93% decrease, P = 0.04), while an increased number of Sox6–/– early- and late-born cortical interneurons were NPY positive (early NPY: 40% increase, P = 0.03; late NPY: 90% increase, P = 0.04; Fig. 7b). These data strongly suggest that, when SOX6 function is absent, many early- and late-born cortical interneurons that would normally differentiate into parvalbumin- and/or SST-positive subtypes inappropriately express NPY, which is normally preferentially expressed by later-born subtypes. Taken together, these data suggest that, much like the function of SOX5 in corticofugal projection neurons22,26, SOX6 is necessary for appropriate cortical interneuron molecular subtype diversity, ensuring the appropriate temporal expression of subtype- and function-defining proteins. Given the pronounced effects of loss of SOX6 function on the largely MGE-derived parvalbumin- and SST-positive interneurons and the lack of an effect on overall cortical interneuron numbers, we next examined whether loss of SOX6 function affects the expression of LHX6, a transcription factor that is necessary for the appropriate development of MGE-derived cortical interneuron subtypes28,29. In Sox6–/– mice, LXH6 was expressed in MGE-born interneurons, but these neurons were disorganized as they segregated into migratory streams and populated the cortex at E13.5 (Fig. 8a), confirming the tangential migratory abnormalities that we observed in Sox6–/–; Gad67-gfp mice (Fig. 4b,c). These data indicate that SOX6 is required for appropriate subtype-specific differentiation from these early stages of interneuron development. At P14, after maturing interneurons had populated the cortex, although there was a modest drop in the abundance of LHX6-
1244
b
positive neurons in Sox6–/– cortex (35% reduction, P = 0.04), most LHX6-positive interneurons populated the cortex (Fig. 8b). LHX6 was not ectopically expressed in neurons born from abnormally ventralized Sox6–/– pallial progenitors (Fig. 8a) or in Sox6–/– CGE (data not shown), indicating that, as in the normal brain, LHX6-positive neurons in Sox6–/– cortex are MGE-derived. To determine whether the abnormally abundant NPY-positive interneurons arise from the Sox6–/– MGE population itself (population autonomous, rather than a result of changes outside of this population), we investigated whether there was an increase in the number of MGEderived LHX6-positive interneurons that express NPY in Sox6–/– cortex. In wild-type cortex, very few LHX6-positive neurons coexpressed NPY (1% ± 0.5% of LHX6-positive neurons), whereas the number that coexpressed NPY in Sox6–/– cortex markedly increased (23% ± 3% of LHX6-positive neurons; ~11.5-fold increase, P = 0.004; Fig. 8c,d). Similarly, there was a very large increase in the number of NPY-positive neurons that immunocytochemically colabeled with LHX6 in Sox6–/– cortex (22% ± 3% of NPY-positive neurons) as compared with wild type (4% ± 2% of NPY-positive neurons). In addition, the large majority of these Sox6–/– LHX6 and NPY double-positive neurons were found in deep cortical layers (81% ± 5%), further suggesting they were of earlyborn MGE origin (Fig. 8d). Taken together, these interneuron subtype analyses indicate that SOX6 functions in a population autonomous manner, controlling the appropriate molecular differentiation of MGEderived cortical interneuron subtypes. An additional (although not mutually exclusive) potential explanation for the increase in the number of Sox6–/– NPY-positive interneurons is that they might be born from abnormally ventralized Sox6–/– pallial progenitors. However, our data indicate that Sox6–/– NPY-positive neurons were not pallium derived; LHX6-positive interneurons in Sox6–/– cortex did not express TBR1 (Supplementary Fig. 7), a transcription factor that is broadly expressed by palliumderived pyramidal neurons through P14, and all of the NPY-positive neurons in Sox6–/– cortex expressed GAD67-GFP (Supplementary Fig. 7), which was not expressed by neurons born from partially ventralized Sox6–/– pallial progenitors (Fig. 4b). In sum, we found that SOX6 is largely mutually exclusively expressed and cross-repressively interacts with highly related SOX5 during telencephalic development, critically controlling pallial progenitor identity and cortical interneuron differentiation and diversity. We volume 12 | number 10 | OCTOBER 2009 nature neuroscience
articles a
b
c
Figure 8 SOX6 control over MGE-derived Lhx6 P14 P14 Lhx6 E13.5 cortical interneuron subtype differentiation is Sox6 –/– WT Sox6 –/– WT Density (neurons per mm2) population autonomous. (a) At E13.5, during I 60 P = 0.004 early stages of cortical interneuron differentiation, WT MGE-born interneurons in wild-type and Sox6–/– Sox6 –/– II/III telencephalon expressed Lhx6 (low magnification; 40 red arrowheads; in situ hybridization), but the IV development of these neurons was disrupted V in Sox6–/– mice. The migratory streams were disorganized compared with wild-type mice (high 20 VI magnification, red arrowheads), and the leading edge of the marginal zone migratory stream compared with the intermediate zone/SVZ stream WM 0 LHX6+ / NPY+ was consistently shorter in Sox6–/– compared with wild-type mice (red arrows). (b) By P14, a –/– P14 WT NPY / LHX6 P14 Sox6 NPY / LHX6 large subset of Lhx6-positive neurons present I in the cortex of wild-type mice had populated the maturing Sox6–/– cortex (red arrowheads; in situ hybridization). (c,d) At P14 in wild-type II/III cortex, the vast majority of LHX6-positive neurons IV (99% ± 0.5%) did not express NPY (d, white V arrowheads; immunocytochemistry), whereas LHX6-positive neurons substantially increased VI their coexpression of NPY in Sox6–/– cortex (d, white arrows) (~11.5-fold increase, P = 0.004; c), especially in deeper layers (81% ± 5% of WM colocalization in the two deepest bins; see Fig. 5b for bin placement). Quantification is represented as the density of neurons per mm2 (c). Dotted lines in b and d indicate pial surface. Scale bars represent 100 µm (low magnification in a, b, and d), 50 µm (high magnification, a) and 25 µm (high magnification, d). Results are expressed as mean ± s.e.m.
© 2009 Nature America, Inc. All rights reserved.
d
previously found that subtype differentiation in the complementary population of corticofugal projection neurons is analogously controlled by SOX5 (ref. 22). Taken together, these simultaneous, independent and functionally parallel controls critically underlie much of the tremendous neuronal diversity in the neocortex (Supplementary Fig. 8). DISCUSSION The cellular diversity of the CNS arises largely from the parsimonious use of a relatively small number of genes across distinct cell types, exemplified here by the multiple and distinct functions of SOX6 during neocortical development. We found that the highly related transcription factors SOX6 and SOX5 (which are coexpressed with largely overlapping functions in other organ systems32,33) are expressed and function in the telencephalon in a cross-repressive and complementary fashion. SOX6 functions cooperatively with previously described pallial/subpallial parcellation programs to control pallial progenitor identity, and it is critical for the subtype diversity of cortical interneurons, parallel to SOX5 function in pallium-derived corticofugal projection neurons22. In the developing telencephalon, the repressive action of SOX6 and Ngn2 (ref. 4) on Mash1 expression is critical for maintaining pallial progenitor identity. Our data indicate that the expression of SOX6 and Ngn2 are independent, that these two Mash1-repressive interactions are cooperative and that both are individually necessary, although neither of them is sufficient alone, for repressing subpallial identity. Although Pax6 directly activates Ngn2 expression in the telencephalon, spinal cord and retina41, it does not appear to act transcriptionally upstream or downstream of SOX6, as others have shown that microarray analysis of Pax6–/– pallium does not reveal a change in SOX6 expression48, and we found that the Sox6–/– pallium continued to express Pax6. This strongly suggests that there are at least two pathways that restrict Mash1 expression to the subpallium: a classic Pax6-Ngn2 pathway, and a cooperative pathway in which SOX6 is expressed independently of Pax6 and Ngn2. nature neuroscience volume 12 | number 10 | OCTOBER 2009
Despite dorsal ectopic expression of Mash1, SOX5 and other subpallial ventricular zone signals in Sox6–/– pallial progenitors, postmitotic projection neuron progeny appeared to develop normally. This contrasts with the apparently more severe ventralization of these neurons in Ngn2–/– mice (including ectopic expression of GAD67)4,5, suggesting that, although initial stages of ventralization occur in Sox6–/– pallium (Supplementary Table 1), dorsal identity regulators that persist in Sox6–/– pallial progenitors, including perhaps Ngn2 and Ngn1, are sufficient to override and mask Mash1 and other subpallial fate programs, as has been suggested previously42. Therefore, SOX6 likely functions in concert with additional pallial patterning regulators to control dorsal identity. Notably, SOX6 and SOX5 are coexpressed in a discrete region of the dorsal subpallial ventricular zone near the PSB. This region encompasses a proliferative source for the lateral cortical stream, which populates structures of the basal limbic system, including the amygdala and piriform cortex38,39. These paleopallial structures are of older evolutionary origin than the neocortex of the neopallium. SOX6 and SOX5 may have been evolutionarily selected to act cooperatively in this unique population of progenitors, as they do in chondrocytes and spinal cord oligodendrocytes. In contrast, the later evolution of the neocortex may have driven the separation of these transcription factors, contributing to the evolution of mammalian neocortical development49. Further molecular phylogenetic analysis might elucidate whether SOX6 and SOX5 cooperate during paleopallium development, and at what point SOX6 and SOX5 function diverged into discrete telencephalic progenitor and neuronal populations. SOX6 is also necessary for successive stages of cortical interneuron postmitotic differentiation. Immature neurons in Sox6–/– subpallial mantle zones had mixed progenitor/postmitotic and dorsal-ventral molecular identity, aberrantly expressing subpallial progenitor–restricted Mash1, pallial progenitor–restricted Ngn2 and pallial postmitotic–restricted Vglut2. As Sox6–/– cortical interneurons matured, they were broadly specified as GABAergic and populated the cortex in correct numbers,
1245
© 2009 Nature America, Inc. All rights reserved.
articles but finer molecular analysis revealed aberrant subtype differentiation, as exemplified by MGE-born cortical interneuron populations. In the absence of SOX6 function, there was a large increase in the abundance of NPY-positive interneurons at the expense of parvalbumin- and SST-positive interneurons, revealing abnormal subtype-defining neurotransmitter/molecular identity, one of multiple core contributing factors to the overall subtype identity and function of a neuron. Additional morphological and electrophysiological analyses might further examine whether these aberrant NPY-positive interneurons fully adopt functions that are normally associated with NPY expression. Three potential (and not mutually exclusive) processes might account for the loss of Sox6–/– cortical interneuron molecular subtype diversity, without an overall reduction of interneuron number. One possibility is that population autonomous subtype specification is primarily affected, such that MGE-derived interneurons that would normally differentiate into subtypes that express parvalbumin and/or SST abnormally differentiate and express NPY. Another possibility is that MGE-born interneurons that normally would have been positive for either or both of parvalbumin and SST might selectively not populate the cortex, and CGE progenitors might simultaneously increase their NPY-positive interneuron output. A third, similar possibility is that abnormally partially ventralized Sox6–/– pallial progenitors are a source of these new NPY-positive neurons, which populate the cortex in place of MGE-born interneurons. Our data very strongly favor the first interpretation of SOX6 control over population autonomous subtype differentiation. First, there was a very large increase in the number of MGE-derived LHX6-positive interneurons that expressed NPY concomitant with their loss of parvalbumin and SST expression. Second, our birthdating analysis revealed that, although the overall numbers of both early- and lateborn neurons were unaffected by the loss of SOX6 function, a large number of early-born interneurons, which tend to arise from the MGE rather than the CGE, did not express parvalbumin or SST, but instead inappropriately expressed NPY. Third, abnormal molecular identity in the Sox6–/– MGE mantle zone (Ngn2, Mash1, Vglut2), observed as soon as the interneurons were born, strongly suggests a population autonomous effect of SOX6 function very early in neuronal differentiation. Fourth, there was no evidence of a substantial increase in the number of GAD67-GFP–positive migrating interneurons originating from Sox6–/– CGE, or any from the pallium, that would be required to compensate for the hypothetical loss of MGE-born interneurons (predicted for the second and third possibilities listed above). Fifth, regarding the second possibility, that the NPY-positive neurons were all CGE-derived, it is neither likely nor supported by any of the data that Sox6–/– CGE progenitor populations would increase their neurogenic rate in the absence of SOX6, as SOX6 is not normally expressed in CGE ventricular zone progenitors. Finally, regarding the third possibility, that the NPY-positive neurons were pallium derived, in Sox6–/– cortex, all of the NPY-positive neurons expressed GAD67, which was not ectopically expressed in Sox6–/– pallium-born neurons, and they did not express TBR1, which is broadly expressed by palliumderived projection neurons, indicating that the NPY-positive neurons were not born from pallial progenitors. Taken together, these results reinforce previous findings on the population autonomous functions of SOX6 both in and outside of the nervous system32,33, indicating that SOX6 functions as a critical control over the appropriate molecular differentiation of MGE-derived cortical interneuron subtypes. Additional interneuron developmental deficits might be occurring in the absence of SOX6 function. Although all of the LHX6-positive neurons in Sox6–/– cortex expressed GAD67-GFP, indicating their broad differentiation into GABAergic interneurons, some did not express
1246
NPY, parvalbumin, SST or other major cortical interneuron subtype molecular markers. This suggests that some Sox6–/– MGE-derived interneurons might stall during later stages of subtype differentiation. In addition, these data do not exclude the possibility that SOX6 also functions in the population autonomous subtype differentiation of SOX6-positive CGE-derived cortical interneurons, or perhaps via additional non–population autonomous pathways. NKX2-1 is a transcription factor that acts upstream of LHX6 (ref. 30), and was recently shown to be critical for multiple stages of cortical interneuron development, including the temporal fate specification of cortical interneuron subtypes31,50. Loss of NKX2-1 function results in a reduction in the number of parvalbumin- and SST-positive cortical interneurons and a corresponding increase in the number of VIP- and calretinin-positive interneurons. Given our results, SOX6 might be functioning, at least partially, in the postmitotic downstream execution of NKX2-1 signaling, potentially interacting with LHX6 (refs. 28,29), thereby regulating the temporal pacing of MGE-derived cortical interneuron fate specification and differentiation (Supplementary Fig. 8). Additional gain- and loss-of-function analyses might reveal potential functional interactions between these transcription factors during interneuron subtype specification and differentiation. Much like pallium-born projection neuron subtypes, cortical interneuron subtype identity is largely determined by the time of birth. Fate-mapping experiments using H3-thymidine labeling and, more recently, genetic tools investigating subtype specification in MGE-born interneurons have shown that SST-positive interneurons, which are diminished in number in Sox6–/– cortex, are on average born at earlier stages of corticogenesis, whereas NPY-, VIP- and calretinin-positive interneurons, whose numbers are either increased or maintained in Sox6–/– cortex, are born later7,31,45,46. These data raise the hypothesis that, during cortical interneuron development, SOX6 participates in setting the pace for the proper timing of developmental transitions. In this model, loss of SOX6 might result in premature differentiation into neurons that are normally born at later developmental stages, at the expense of those born at earlier stages. Supporting this interpretation, our dual IdU/CldU birthdating analysis of the molecular differentiation of early- and lateborn neurons in Sox6–/– cortex revealed that early-born neurons aberrantly differentiated and expressed the later-born subtype-defining protein NPY. As the lineage relationships of particular cortical interneuron subtypes are further clarified, it will be possible to discern whether loss of SOX6 function alters temporal development in a lineage (for example, those that would normally be SST-positive neurons aberrantly differentiate into NPY-positive neurons born later from potentially the same lineage) and/or whether loss of SOX6 function results in inappropriate differentiation across lineages (for example, those that would normally be parvalbuminpositive neurons aberrantly differentiate into NPY-positive neurons born later from a potentially distinct lineage). We recently reported that the loss of SOX5 function in pallium-derived corticofugal projection neurons results in the premature adoption of subcerebral projection neuron features that are characteristic of later stages of cortical projection neuron development22. Therefore, it is possible that SOX6 and SOX5 both suppress coordinately regulated controls that promote premature transition into later stages of subtype differentiation. Consistent with this interpretation are the largely redundant roles of both SOX6 and SOX5 in chondroblasts during cartilage development and in oligodendroglial progenitors in the spinal cord in preventing the premature transition of these cell types to subsequent stages of development32,33. Given their analogous loss-of-function phenotypes, it is interesting to speculate that SOX6 and SOX5 separated in function during the evolution of the increasingly complex neuronal diversity of the telencephalon, and assumed complementary, but distinct, roles. volume 12 | number 10 | OCTOBER 2009 nature neuroscience
articles METHODS Methods and any associated references are available in the online version of the paper at http://www.nature.com/natureneuroscience/.
© 2009 Nature America, Inc. All rights reserved.
Note: Supplementary information is available on the Nature Neuroscience website. ACKNOWLEDGMENTS We thank K. Billmers, A. Palmer, L. Pasquina, K. Quinn, D. Schuback, E. Sievert, A. Wheeler and T. Yamamoto for superb technical assistance, G. Fishell, R. BatistaBrito, G. Miyoshi, P. Arlotta, B. Molyneaux, H. Padmanabhan, F. Guillemot, Q. Ma, C. Cepko and L. Goodrich for helpful discussions and input, U. Berger for technical assistance with in situ hybridization, C. Lois, R. Hevner, V. Lefebvre, F. Guillemot, V. Pachnis and Y. Yanagawa for generously sharing mice, antibodies and reagents, and current and past members of our laboratory for helpful suggestions. This work was partially supported by grants from the US National Institutes of Health (NS49553 and NS45523; additional infrastructure supported by NS41590), the Travis Roy Foundation, the Spastic Paraplegia Foundation, the Massachusetts Spinal Cord Injury research program, and the Harvard Stem Cell Institute to J.D.M., and by the Jane and Lee Seidman Fund for CNS Research, and the Emily and Robert Pearlstein Fund for Nervous System Repair. E.A. was partially supported by a US National Institutes of Health individual predoctoral National Research Service Award fellowship (F31 NS060421). D.J. was partially supported by fellowships from the Swiss National Science Foundation and the Holcim Foundation. R.M.F. was partially supported by a National Science Foundation Graduate Research Fellowship. AUTHOR CONTRIBUTIONS E.A. and J.D.M. designed the overall experimental directions and specific analyses, and wrote and edited the manuscript. E.A. also performed all of the experiments and data analysis. D.J. co-performed the microarray experiments and assisted with interneuron quantification, microarray data evaluation, experimental design and data analysis, and manuscript writing and editing. R.M.F. performed whole-mount in situ hybridization/immunocytochemistry and assisted with BrdU/PH3 pallial progenitor analysis, microarray data evaluation, interneuron quantification, and manuscript editing. J.D.M. also contributed to data analysis and biological interpretation. Published online at http://www.nature.com/natureneuroscience/. Reprints and permissions information is available online at http://www.nature.com/ reprintsandpermissions/. 1. Schuurmans, C. & Guillemot, F. Molecular mechanisms underlying cell fate specification in the developing telencephalon. Curr. Opin. Neurobiol. 12, 26–34 (2002). 2. Wonders, C.P. & Anderson, S.A. The origin and specification of cortical interneurons. Nat. Rev. Neurosci. 7, 687–696 (2006). 3. Molyneaux, B.J., Arlotta, P., Menezes, J.R. & Macklis, J.D. Neuronal subtype specification in the cerebral cortex. Nat. Rev. Neurosci. 8, 427–437 (2007). 4. Fode, C. et al. A role for neural determination genes in specifying the dorsoventral identity of telencephalic neurons. Genes Dev. 14, 67–80 (2000). 5. Parras, C.M. et al. Divergent functions of the proneural genes Mash1 and Ngn2 in the specification of neuronal subtype identity. Genes Dev. 16, 324–338 (2002). 6. Butt, S.J. et al. The temporal and spatial origins of cortical interneurons predict their physiological subtype. Neuron 48, 591–604 (2005). 7. Miyoshi, G., Butt, S.J., Takebayashi, H. & Fishell, G. Physiologically distinct temporal cohorts of cortical interneurons arise from telencephalic Olig2-expressing precursors. J. Neurosci. 27, 7786–7798 (2007). 8. Flames, N. et al. Delineation of multiple subpallial progenitor domains by the combinatorial expression of transcriptional codes. J. Neurosci. 27, 9682–9695 (2007). 9. Wonders, C.P. et al. A spatial bias for the origins of interneuron subgroups within the medial ganglionic eminence. Dev. Biol. 314, 127–136 (2008). 10. Fogarty, M. et al. Spatial genetic patterning of the embryonic neuroepithelium generates GABAergic interneuron diversity in the adult cortex. J. Neurosci. 27, 10935–10946 (2007). 11. Ascoli, G.A. et al. Petilla terminology: nomenclature of features of GABAergic interneurons of the cerebral cortex. Nat. Rev. Neurosci. 9, 557–568 (2008). 12. Flames, N. & Marin, O. Developmental mechanisms underlying the generation of cortical interneuron diversity. Neuron 46, 377–381 (2005). 13. Corbin, J.G., Nery, S. & Fishell, G. Telencephalic cells take a tangent: non-radial migration in the mammalian forebrain. Nat. Neurosci. 4 Suppl, 1177–1182 (2001). 14. Levitt, P., Eagleson, K.L. & Powell, E.M. Regulation of neocortical interneuron development and the implications for neurodevelopmental disorders. Trends Neurosci. 27, 400–406 (2004). 15. Armijo, J.A., Valdizan, E.M., De Las Cuevas, I. & Cuadrado, A. Rev. Neurol. Advances in the physiopathology of epileptogenesis: molecular aspects. 34, 409–429 (2002). 16. Rubenstein, J.L. & Merzenich, M.M. Model of autism: increased ratio of excitation/ inhibition in key neural systems. Genes Brain Behav. 2, 255–267 (2003).
nature neuroscience volume 12 | number 10 | OCTOBER 2009
17. Lewis, D.A. GABAergic local circuit neurons and prefrontal cortical dysfunction in schizophrenia. Brain Res. Brain Res. Rev. 31, 270–276 (2000). 18. Arlotta, P. et al. Neuronal subtype–specific genes that control corticospinal motor neuron development in vivo. Neuron 45, 207–221 (2005). 20. Chen, B., Schaevitz, L.R. & McConnell, S.K. Fezl regulates the differentiation and axon targeting of layer 5 subcortical projection neurons in cerebral cortex. Proc. Natl. Acad. Sci. USA 102, 17184–17189 (2005). 21. Chen, J.G., Rasin, M.R., Kwan, K.Y. & Sestan, N. Zfp312 is required for subcortical axonal projections and dendritic morphology of deep-layer pyramidal neurons of the cerebral cortex. Proc. Natl. Acad. Sci. USA 102, 17792–17797 (2005). 22. Lai, T. et al. SOX5 controls the sequential generation of distinct corticofugal neuron subtypes. Neuron 57, 232–247 (2008). 23. Alcamo, E.A. et al. Satb2 regulates callosal projection neuron identity in the developing cerebral cortex. Neuron 57, 364–377 (2008). 24. Britanova, O. et al. Satb2 is a postmitotic determinant for upper-layer neuron specification in the neocortex. Neuron 57, 378–392 (2008). 25. Joshi, P.S. et al. Bhlhb5 regulates the postmitotic acquisition of area identities in layers II–V of the developing neocortex. Neuron 60, 258–272 (2008). 26. Kwan, K.Y. et al. SOX5 postmitotically regulates migration, postmigratory differentiation and projections of subplate and deep-layer neocortical neurons. Proc. Natl. Acad. Sci. USA 105, 16021–16026 (2008). 27. Cobos, I. et al. Mice lacking Dlx1 show subtype-specific loss of interneurons, reduced inhibition and epilepsy. Nat. Neurosci. 8, 1059–1068 (2005). 28. Liodis, P. et al. Lhx6 activity is required for the normal migration and specification of cortical interneuron subtypes. J. Neurosci. 27, 3078–3089 (2007). 29. Zhao, Y. et al. Distinct molecular pathways for development of telencephalic interneuron subtypes revealed through analysis of Lhx6 mutants. J. Comp. Neurol. 510, 79–99 (2008). 30. Du, T., Xu, Q., Ocbina, P.J. & Anderson, S.A. NKX2.1 specifies cortical interneuron fate by activating Lhx6. Development 135, 1559–1567 (2008). 31. Butt, S.J. et al. The requirement of Nkx2-1 in the temporal specification of cortical interneuron subtypes. Neuron 59, 722–732 (2008). 32. Smits, P. et al. The transcription factors L-Sox5 and Sox6 are essential for cartilage formation. Dev. Cell 1, 277–290 (2001). 33. Stolt, C.C. et al. SoxD proteins influence multiple stages of oligodendrocyte development and modulate SoxE protein function. Dev. Cell 11, 697–709 (2006). 34. Wegner, M. From head to toes: the multiple facets of Sox proteins. Nucleic Acids Res. 27, 1409–1420 (1999). 35. Wegner, M. & Stolt, C.C. From stem cells to neurons and glia: a Soxist’s view of neural development. Trends Neurosci. 28, 583–588 (2005). 36. Connor, F., Wright, E., Denny, P., Koopman, P. & Ashworth, A. The Sry-related HMG box-containing gene Sox6 is expressed in the adult testis and developing nervous system of the mouse. Nucleic Acids Res. 23, 3365–3372 (1995). 37. Narahara, M., Yamada, A., Hamada-Kanazawa, M., Kawai, Y. & Miyake, M. cDNA cloning of the Sry-related gene Sox6 from rat with tissue-specific expression. Biol. Pharm. Bull. 25, 705–709 (2002). 38. Puelles, L. et al. Pallial and subpallial derivatives in the embryonic chick and mouse telencephalon, traced by the expression of the genes Dlx-2, Emx-1, Nkx-2.1, Pax-6 and Tbr-1. J. Comp. Neurol. 424, 409–438 (2000). 39. Carney, R.S. et al. Cell migration along the lateral cortical stream to the developing basal telencephalic limbic system. J. Neurosci. 26, 11562–11574 (2006). 40. Ma, Q., Sommer, L., Cserjesi, P. & Anderson, D.J. Mash1 and neurogenin1 expression patterns define complementary domains of neuroepithelium in the developing CNS and are correlated with regions expressing notch ligands. J. Neurosci. 17, 3644–3652 (1997). 41. Scardigli, R., Baumer, N., Gruss, P., Guillemot, F. & Le Roux, I. Direct and concentration-dependent regulation of the proneural gene Neurogenin2 by Pax6. Development 130, 3269–3281 (2003). 42. Britz, O. et al. A role for proneural genes in the maturation of cortical progenitor cells. Cereb. Cortex 16 Suppl 1, i138–i151 (2006). 43. Batista-Brito, R., Machold, R., Klein, C. & Fishell, G. Gene expression in cortical interneuron precursors is prescient of their mature function. Cereb. Cortex 18, 2306–2317 (2008). 44. Tamamaki, N. et al. Green fluorescent protein expression and colocalization with calretinin, parvalbumin and somatostatin in the GAD67-GFP knock-in mouse. J. Comp. Neurol. 467, 60–79 (2003). 45. Cavanagh, M.E. & Parnavelas, J.G. Development of somatostatin immunoreactive neurons in the rat occipital cortex: a combined immunocytochemical-autoradiographic study. J. Comp. Neurol. 268, 1–12 (1988). 46. Cavanagh, M.E. & Parnavelas, J.G. Development of neuropeptide Y (NPY) immunoreactive neurons in the rat occipital cortex: a combined immunohistochemicalautoradiographic study. J. Comp. Neurol. 297, 553–563 (1990). 47. Vega, C.J. & Peterson, D.A. Stem cell proliferative history in tissue revealed by temporal halogenated thymidine analog discrimination. Nat. Methods 2, 167–169 (2005). 48. Holm, P.C. et al. Loss- and gain-of-function analyses reveal targets of Pax6 in the developing mouse telencephalon. Mol. Cell. Neurosci. 34, 99–119 (2007). 49. Molnár, Z. & Butler, A.B. The corticostriatal junction: a crucial region for forebrain development and evolution. Bioessays 24, 530–541 (2002). 50. Nóbrega-Pereira, S. et al. Postmitotic Nkx2-1 controls the migration of telencephalic interneurons by direct repression of guidance receptors. Neuron 59, 733–745 (2008).
1247
articles ONLINE METHODS
© 2009 Nature America, Inc. All rights reserved.
Mice. Sox6+/– and Sox5+/– mice were the generous gift of V. Lefebvre (Cleveland Clinic)32 (Sox6 GeneID 20679; Sox5 GeneID 20678). The Gad67-gfp (delta-neo) mice were the generous gift of Y. Yanagawa (Gunma University)28,44. Ngn2+/–; Ngn1+/– mice were a generous gift from F. Guillemot (National Institute for Medical Research)4. The Sox6 and Sox5 transgenic mice were on a pure C57BL/6 background. The Ngn2; Ngn1 transgenic mice were on a pure CD1 background. The Sox6; Gad67-gfp transgenic crosses were a mix between C57BL/6 and Swiss Webster; controls always had the same degree of mixed background. The day of vaginal plug detection was designated as E0.5. The day of birth was designated as P0. All mouse studies were approved by the Massachusetts General Hospital Institutional Animal Care and Use Committee and were performed in accordance with institutional and federal guidelines. Immunocytochemistry and in situ hybridization. Brains were fixed and stained using standard methods18. For primary antibodies, we used rabbit antibody to SOX6 (1:500, Abcam), goat antibody to SOX5 (1:250, Santa Cruz Biotech), mouse antibody to BrdU (1:500, Becton Dickinson; detects IdU), rat antibody to BrdU (1:500, Accurate; detects CldU), mouse antibody to BrdU (1:750, Chemicon), rabbit antibody to PH3 (1:200, Upstate); mouse antibody to PH3 (1:400, Abcam), mouse antibody to PCNA (1:5,000, Sigma), rabbit antibody to TBR1 (1:1,500, gift of R. Hevner (University of Washington)), rabbit antibody to TBR1 (1:500, Abcam), rat antibody to CTIP2 (1:1,000, Abcam), mouse antibody to Reelin (1:500, Chemicon), rabbit antibody to GFP (1:500, Molecular Probes), mouse antibody to parvalbumin (1:500, Sigma), rat antibody to SST (1:100, Chemicon), mouse antibody to calbindin (1:500, Chemicon), rabbit antibody to calretinin (1:1,000, Chemicon), mouse antibody to calretinin (1:400, Chemicon), rabbit antibody to NPY (1:500, Immunostar), rabbit antibody to VIP (1:100, Immunostar), and rabbit antibody to LHX6 (1:1,000, gift of V. Pachnis (National Institute for Medical Research)). Appropriate secondary antibodies were from the Molecular Probes Alexa series. When double immunocytochemistry was performed with two primary antibodies raised in the same species (only in the case of SOX6 colocalization with NPY and VIP and NPY colocalization with TBR1 and LHX6), immunocytochemistry for each antibody was performed sequentially using different secondary antibodies. Tissue was fixed for 30 min in 4% paraformaldehyde (wt/vol) and rinsed in phosphate-buffered saline before application of the second primary antibody. In instances of minor cross-reactivity, nuclear versus cytoplasmic localization of fluorescence was used to distinguish between the two. Riboprobes were generated and nonradioactive in situ hybridization was performed as previously described18. The Sox6 cDNA clone was a gift from V. Lefebvre32. RT-PCR was used to generate the following cDNA clones: Ngn2 (NM_009718.2, BGEM), Mash1 (RP_050927_04_D07, Allen Brain Atlas), Olig2 (NM_016967.2, BGEM), Pax6 (RP_050927_01_H01, Allen Brain Atlas), Cux2 (ref. 19), PlexinD1 (ref. 19), Lhx6 (MTF#274, Gudmap), and Vglut2 (nucleotides 2477–2933 of NM_080853). Molecular and mitotic characterization of progenitors. For examination of pallial progenitor phenotype, we performed immunocytochemistry for SOX5 (n = 5 wild-type and n = 4 Sox6–/– at E13.5, n = 4 wild type and n = 4 Sox6–/– at P0; Fig. 2) and in situ hybridization for Mash1 (n = 3 wild type, n = 3 Sox6–/–, n = 1 Sox5–/–, n = 1 Sox6–/–; Sox5–/– at E13.5; n = 2 wild type and n = 1 Ngn2–/–; Ngn1–/– at E14.5; n = 1 wild type and n = 1 Sox6–/– at E17.5; Fig. 3 and Supplementary Fig. 3), Sox6 (n = 2 wild type and n = 1 Ngn2–/–; Ngn1–/– at E14.5; Fig. 3), Olig2 (n = 1 wild type and n = 1 Sox6–/– at E13.5; Supplementary Fig. 3) and Pax6 (n = 1 wild type and n = 1 Sox6–/– at E13.5; Supplementary Fig. 3). For examination of subpallial progenitor phenotype, we performed in situ hybridization for Sox6 (n = 3 wild type and n = 2 Sox6–/– at E13.5; n = 3 wild type and n = 3 Sox6–/– at P0; Fig. 2) and Ngn2 (n = 3 wild type, n = 2 Sox6–/–, n = 1 Sox5–/–, n = 1 Sox6–/–; Sox5–/– at E13.5; n = 1 wild type and n = 1 Sox6–/– at E17.5; Fig. 3 and Supplementary Fig. 3). For BrdU birthdating and PH3 quantification, timed pregnant females received a single intraperitoneal injection of BrdU (50 mg per kg of body weight) at E13.5 (pallial progenitor analysis) or E14.5 (subpallial progenitor analysis in Supplementary Fig. 2). Embryos were collected 1 h later and processed for BrdU immunocytochemistry19. For pallial progenitor quantification, we selected four anatomically matched cortical sections from each mouse (n = 3 wild type, n = 3 Sox6–/–), carried out BrdU and PH3 immunocytochemistry
nature neuroscience
(2 h of 2N HCl treatment preceded BrdU immunocytochemistry), obtained fluorescent images, and selected two anatomically matched areas (130 µm × 200 µm) from each hemisphere of each section (one medial and one lateral) for blinded quantification of BrdU positivity (defined a priori as having strong and homogenous nuclear labeling) and PH3 positivity by two independent investigators. Normal distribution was confirmed, and the unpaired, two-tailed t test was used for statistical analysis. Microarray analysis. Each embryo from four E13.5 litters (generated by mating male and female Sox6+/– mice) was placed in cold Hank’s buffered salt solution, the pallium was microdissected and immediately preserved in RNAlater (Ambion), and the remaining embryo tissue was subsequently genotyped. RNA extraction, quality assessment, and amplification followed previously reported methods18. Briefly, to ensure biological importance and reproducibility, biological replicate RNA samples from four wild-type and four Sox6–/– embryos were extracted using the StrataPrep Total RNA Mini Prep Kit (Stratagene), RNA was quantified using a NanoDrop (Thermo Fisher Scientific), and the quality was assessed with a Nanochip in a Bioanalyzer (Agilent Technologies). RNA was amplified via two rounds of in vitro transcription and biotinylated using a BioArray HighYield RNA Transcript Labeling Kit (Enzo), yielding approximately 20–50 µg of labeled cRNA for hybridization18, and the quality of the amplified RNA was assessed with a Nanochip in a Bioanalyzer before hybridization on Affymetrix 430.2 GeneChips. Homotypic (biological replicates) and heterotypic comparisons (wild type versus Sox6–/–) were performed using Rosetta Resolver software (Rosetta Inpharmatics). Differentially expressed genes with an absolute fold change of more than 1.8 and a P value of less than 0.005 were selected for further analysis. To rigorously ensure statistical significance of identified candidate genes, we normalized the data via three additional independent methods (RMA (Robust Multi-Array Analysis), GCRMA (Guanine Cytosine Robust Multi-Array Analysis) and MAS 5.0-Affymetrix) in Bioconductor and cross-referenced the significance of candidate genes with the Rosetta Resolver normalized dataset. Additional cross-referencing was performed with genes identified as significant with a significance analysis of microarrays approach using an absolute fold change of more than 1.8 and a d value of 0.772 for GCRMA-normalized data and a d value of 0.35 for RMA-normalized data. The biological relevance of candidate genes was assessed in an integrated gene analysis platform developed in our laboratory (D.J., E.A., J.D.M., unpublished data), using online in situ hybridization, gene ontology, protein function, and literature databases to individually assess expression and function of each gene. Statistically significant genes with normally segregated pallial or subpallial expression during development were identified (Supplementary Table 1). Mis-expression of SOX6 and SOX5 via electroporation. For control experiments, a vector containing IRES-egfp under the control of a constitutively active CMV/β-actin promoter was used19 (a generous gift of C. Lois (Picower Institute)). Sox6 and Sox5 (ref. 22) were cloned upstream of IRES-egfp for misexpression. We mixed 750 nl of purified DNA (1.0 µg µl–1) with 0.005% Fast Green (for visualization), injected it in utero into the lateral ventricle of CD1 embryos at E12.5 under ultrasound guidance (Vevo 770, VisualSonics), and electroporated into the subpallial (Sox6) or pallial (Sox5) ventricular zone, as described previously19,22. Embryos were analyzed at E16.5 (n = 3 subpallial control, n = 3 subpallial Sox6, n = 3 pallial control, n = 3 pallial Sox5, multiple independent litters were examined in each condition). Interneuron quantification. For quantification of the tangential distance between the leading edges of the marginal zone and intermediate zone/SVZ interneuron tangential migratory streams at E13.5, we selected anatomically matched sections (n = 3 Sox6+/+; Gad67-gfp+/–, n = 3 Sox6–/–; Gad67-gfp+/–; 10–12 hemispheres per mouse, spanning the rostro-caudal extent of the telencephalon), and carried out GFP immunocytochemistry. The distance was measured in micrometers between the position of the soma of the leading neuron of the marginal zone stream, radially projected to the pial surface, and the corresponding position of the soma of the leading neuron in the intermediate zone/SVZ stream, radially projected to the pial surface, determined by drawing imaginary lines radially from the marginal zone and intermediate zone/SVZ streams perpendicular to the surface of the brain (Supplementary Fig. 5).
doi:10.1038/nn.2387
© 2009 Nature America, Inc. All rights reserved.
articles For quantification of interneuron number at E13.5, anatomically matched sections were selected (n = 3 Sox6+/+; Gad67-gfp+/–, n = 3 Sox6–/–; Gad67-gfp+/–, 10–12 hemispheres per mouse spanning the rostro-caudal extent of the telencephalon), GFP immunocytochemistry was performed, and Grid Confocal (Improvision) images were obtained of a single plane of GFP-positive neurons. Digital boxes of fixed area (marginal zone stream, 85 µm × 175 µm; intermediate zone/SVZ stream, 115 µm × 175 µm) were superimposed at predetermined anatomical landmarks at the base, middle and leading edge of each migratory stream of each hemisphere, and GFP-positive neurons were quantified (Supplementary Fig. 5). For P0 quantification, anatomically matched sections were selected (n = 3 wild type, n = 3 Sox6–/–; 10–12 hemispheres per mouse spanning the rostrocaudal extent of the telencephalon), GFP immunocytochemistry was performed, digital boxes of fixed width were equally divided into four bins (see Fig. 5a) and superimposed on the cortices adjacent to the PSB of each hemisphere, and height was adjusted to extend from the bottom of layer VI to directly under the marginal zone (at P0, the marginal zone was still very interneuron dense and was therefore not included in the quantification). GFP-positive neurons were quantified, and density values were calculated on the basis of the area of the box in each image; densities are expressed as neurons per mm2. Laminar distributions were determined by dividing the proportion of neurons in each bin by the total number of neurons in all bins. For P14 quantification, anatomically matched sections were selected (n = 3 wild type, n = 3 Sox6–/–; 8–12 hemispheres per mouse spanning the rostrocaudal extent of the telencephalon), and SOX6, GFP, parvalbumin, SST, NPY (n = 6 wild type, n = 6 Sox6–/–), VIP (n = 4 wild type, n = 4 Sox6–/–), calretinin, calbindin and LHX6 immunocytochemistry was performed. Digital boxes of fixed width were equally divided into four bins (see Fig. 5b), superimposed on the cortices adjacent to the PSB of each hemisphere and height was adjusted to extend from the top of the white matter to the cortical surface. The percentages of wild-type interneurons that expressed SOX6 (Fig. 6b) were calculated from the total numbers in all four bins. Density values and laminar distributions in wild-type and Sox6–/– were calculated as above. For IdU and CldU interneuron birthdating, equimolar delivery of IdU (57.5 mg per kg) at E11.5 and CldU (42.5 mg per kg) at E15.5 was performed47. We
doi:10.1038/nn.2387
killed the mice at P14 by perfusion, genotyped them and prepared brains for immunocytochemistry. Anatomically matched sections were selected, and IdU, CldU, parvalbumin, SST and NPY immunocytochemistry was performed (IdU and parvalbumin, IdU and NPY: n = 2 wild type, n = 3 Sox6–/–; IdU and SST, CldU and parvalbumin, CldU and SST, CldU and NPY: n = 2 wild type, n = 2 Sox6–/–; 7–8 hemispheres per mouse spanning the rostro-caudal extent of the telencephalon). Digital boxes of fixed width were superimposed on the cortices adjacent to the PSB, and height was adjusted to extend from the top of the white matter to the cortical surface. Interneurons coexpressing SST, parvalbumin or NPY, and either IdU or CldU (defined a priori as having strong and homogenous nuclear labeling) were quantified (colabeling was defined as clear nuclear label surrounded by cytoplasmic parvalbumin, SST or NPY label; Fig. 7a) and density values were calculated on the basis of the area of the box in each image. All quantification was performed on 50-µm vibratome sections using wellestablished modified stereological methods, beginning at the rostral limit of the corpus callosum and continuing caudally, skipping four sections between samples, so that no cell could be counted twice in an adjacent section. We used strict a priori criteria, whereby the entire soma of a cell needed to be present to be counted, which was effectively accomplished by counting with high numerical aperture optics in the central approximately 30 µm of the thick 50-µm sections, avoiding cut neurons present in the top or bottom approximately 10 µm of each section. All quantifications were blinded, normal distribution was confirmed, and the unpaired, two-tailed t test was used for statistical analysis. Welch corrections were performed in the rare instances when the s.d. of the two groups varied significantly. All results are expressed as the mean ± s.e.m. Microscopy and image analysis. Tissue sections were viewed on a Nikon E1000 microscope equipped with an X-Cite 120 illuminator (EXFO), and images were collected and analyzed with Volocity image analysis software (Improvision, v4.0.1). Images were optimized for size, color, and contrast using Photoshop 7.0 (Adobe). Single plane fluorescence images for E13.5 interneuron quantification were obtained using the Volocity Grid Confocal microscopy system (Improvision). Images were collected at the approximate midpoint between the top and bottom planes of focus.
nature neuroscience
a r t ic l e s
Hippocampal development and neural stem cell maintenance require Sox2-dependent regulation of Shh © 2009 Nature America, Inc. All rights reserved.
Rebecca Favaro1, Menella Valotta1,3,4, Anna L M Ferri1,4, Elisa Latorre1, Jessica Mariani1, Claudio Giachino2, Cesare Lancini1, Valentina Tosetti1, Sergio Ottolenghi1, Verdon Taylor2 & Silvia K Nicolis1 Neural stem cells (NSCs) are controlled by diffusible factors. The transcription factor Sox2 is expressed by NSCs and Sox2 mutations in humans cause defects in the brain and, in particular, in the hippocampus. We deleted Sox2 in the mouse embryonic brain. At birth, the mice showed minor brain defects; shortly afterwards, however, NSCs and neurogenesis were completely lost in the hippocampus, leading to dentate gyrus hypoplasia. Deletion of Sox2 in adult mice also caused hippocampal neurogenesis loss. The hippocampal developmental defect resembles that caused by late sonic hedgehog (Shh) loss. In mutant mice, Shh and Wnt3a were absent from the hippocampal primordium. A SHH pharmacological agonist partially rescued the hippocampal defect. Chromatin immunoprecipitation identified Shh as a Sox2 target. Sox2-deleted NSCs did not express Shh in vitro and were rapidly lost. Their replication was partially rescued by the addition of SHH and was almost fully rescued by conditioned medium from normal cells. Thus, NSCs control their status, at least partly, through Sox2-dependent autocrine mechanisms. Sox2 encodes a transcription factor that is essential for the pluri potency of epiblast, embryonic stem (ES) cells and reprogrammed, induced pluripotent stem cells1–4. Sox2 is also expressed at early stages of CNS development and marks NSCs5,6 and precursors. In humans, rare SOX2 mutations cause anophtalmia, defective hippocampal development, cognitive defects and seizures7–9. In mice, hypomorphic Sox2 mutants, expressing 30% of normal Sox2 levels, show a loss of striatal and thalamic tissue, epilepsy and neurodegeneration10, which might result from decreased stem cell numbers and defective neuronal differentiation10,11. These observations raise the possibility that Sox2 is important for controlling self-renewal and multi/pluripotency in several stem cell types. To evaluate this hypothesis and to investigate the molecular mechanisms of Sox2 function, we developed nervous system–specific conditional Sox2 knockout mice, bypassing the early embryonic lethality of homozygous Sox2 mutants. We found that Sox2 was required for NSC maintenance in the hippocampus and in long-term in vitro neurosphere cultures. NSC maintenance in neonatal hippo campus and in neonatal neurosphere cultures required soluble factors, including SHH, which was controlled by Sox2 itself. RESULTS Sox2 deletion causes hippocampal defects with NSC loss We generated compound heterozygous mice carrying a beta-geo knock-in5,10 null mutation in Sox2 and a Sox2loxP allele (Fig. 1a), together with a nestin-cre transgene12, in which cre activity is driven by a neural nestin enhancer; the selection cassette in the original
flox-targeted allele had been removed in previous generations (Fig. 1a). Cre activity, which began at embryonic day 10.5 (E10.5), caused the complete loss of Sox2 in the CNS by E12.5 (Fig. 1b and data not shown). Homozygous Sox2-deleted mice were born in the expected ratio, but most died by 4 weeks of age. At birth (postnatal day 0, P0), we detected limited abnormalities in the brain, with the exception of a slightly reduced hippocampus, a moderate lateral ventricle enlargement and slight size reduction of the posterior ventrolateral cortex (Fig. 1c). Subsequently, however, the development of the hippocampus was markedly reduced, particularly caudally, resulting in an almost complete absence of the dentate gyrus at P7 (Fig. 1c). To understand the causes of this defect, we studied NSCs. During brain development, periventricular NSCs have radial glia morphology and express RC2 and BLBP13,14. Postnatally, radial glia are maintained in the hippocampus from P0 as GFAP and nestin double-positive cells and are the source of continuous neurogenesis in the dentate gyrus throughout life13,14. In Sox2-ablated embryos, neither the abundance of radial glia (RC2 and BLBP positive14) nor neurogenesis (BrdU incorporation) were substantially altered (BrdU-positive cells, 89 ± 6% of normal; Supplementary Fig. 1). In P0 hippocampus, the number of GFAP/nestin-positive cells in the dentate gyrus sub granular layer was only slightly decreased (Fig. 2a). Neurogenesis was almost normal at this stage (Fig. 2b). By P2, however, the number of GFAP/nestin-positive cells was strongly reduced and the cells were completely lost in the dentate gyrus by P7 (Fig. 2a). BrdU labeling showed a similar decrease; at P7, the small residual part of the dentate
1Department of Biotechnology and Biosciences, University of Milano-Bicocca, Milano, Italy. 2Department of Molecular Embryology, Max Planck Institute for Immunobiology, Freiburg im Breisgau, Germany. 3Present address: Axxam, Milano, Italy. 4These authors contributed equally to this work. Correspondence should be addressed to S.K.N. (
[email protected]).
Received 17 February; accepted 5 August; published online 6 September 2009; doi:10.1038/nn.2397
1248
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
a r t ic l e s a
1 Kbp
15.5 Kbp E
Sox2
H
S
S
E
Sox2 locus
Vector neo E
E
E
1
dta S
2
E
Sox2 loxP
neo 6.5 Kbp
FLP E
E 3
7.5 Kbp S
E 4
5
E
Sox2 loxP∆neo
6
Cre E
3
E
Sox2 loxP-null
6
E 10.5
E 11.5
E 11.5
E 12.5
E 12.5
E 10.5
E 10.5
E 11.5
E 11.5
E 12.5
E 12.5
WT
E 10.5
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1249
Sox2 null
Sox2 ablation causes early regional loss of Shh and Wnt3a The defective hippocampal development in Sox2-deleted mice quali tatively mimics that described for mutations of other transcription factors that are part of the Wnt signaling pathway (Lef1) or of Wnt3a itself15,16; moreover, it is similar to the effect of a conditional mutation of Shh or of its receptor, smoothened (Smo)17. We analyzed Wnt3a and Shh mRNA expression during development. By E14.5, Wnt3a expres sion in the cortical hem, which includes the hippocampal primor dium, was strongly reduced in Sox2-deleted mutants, particularly in posterior regions (Fig. 3). We observed this reduction at E17.5 as well, just before hippocampus formation (Fig. 3). We detected no Wnt3a expression in normal or Sox2 mutant hippocampus after birth (data not shown). Shh mRNA expression was strongly reduced at E14.5 in telencephalon and diencephalon, but not in midbrain (Fig. 3) and
spinal cord (data not shown). This defect was maintained until at least P0, when Shh mRNA was easily detectable in normal, but not mutant, telencephalon (Fig. 3) and hippocampus (data not shown). At birth, Shh mRNA was absent in the hippocampal hilus of Sox2 mutants (data not shown). In normal mice, SHH protein was clearly detectable at birth in both hippocampal and lateral ventricle wall neurogenic regions (Fig. 4). In the hippocampus, SHH was abundant in the hilus and marked cells with radial orientation in the dentate gyrus (Fig. 4a); in the lateral ventricle, cells that were positive for both SOX2 and SHH formed a continuous layer along the wall (Fig. 4a). In wild-type mice, confocal microscopy of both regions showed colocalization of SHH and SOX2 in individual cells, with SHH forming a rim around SOX2positive nuclei (Fig. 4a,b). In contrast, in mutant (Sox2-null) mice, SHH was absent in these regions (Fig. 4a,b). Coexpression of SOX2 and SHH was further confirmed in normal NSCs (neurospheres) grown in vitro, and Shh loss was similarly evident in neurospheres from Sox2-deleted brains (see below). The conditional knockout of Smo and Shh17 causes an arrest of postnatal hippocampal development, with partial loss of the dentate gyrus. This phenotype closely resembles that of our Sox2 knockout mouse, strongly suggesting that Sox2 controls hippocampal devel opment via regulation of Shh. To rescue the mutant Sox2-deleted phenotype (Figs. 1b and 2), we administered an SHH pharmaco logical agonist (SHH-Ag) to pregnant mothers (between E12.5 and P1), which was previously used to rescue embryonic defects of the Shh null mutant18. SHH-Ag–treated Sox2-deleted mutants were analyzed at P2 (TUNEL) or at P7 for GFAP/nestin-positive radial glia and for hippocampal BrdU incorporation; untreated Sox2-deleted mutants and normal SHH-Ag–treated P7 mice were also examined. SHH-Ag treatment (Fig. 5) greatly stimulated BrdU incorporation in the hippocampus of Sox2-deleted mutants, reaching ~40–50% that observed in wild-type controls (versus 3–4% in untreated Sox2-deleted mutants). In the dentate gyrus, we observed large numbers of nestin and GFAP double-positive radial glia in treated mutants, in contrast with the complete disappearance in untreated mutants (Fig. 5; see also Fig. 2a,b); in addition, the size of the dentate gyrus was increased, par ticularly in the dorsal blade, in treated mutants (Fig. 5). SHH-Ag also greatly decreased apoptosis at P2 in Sox2 mutants (Fig. 5). The substan tial rescue by SHH-Ag of the postnatal loss of nestin/GFAP-positive
P0
P2
P7
Adult
Adult
P0
P2
P7
Adult
Adult
WT
c
Sox2 null
© 2009 Nature America, Inc. All rights reserved.
b
S
E
Figure 1 Sox2 conditional null allele, SOX2 protein ablation by nestinCre in mutant embryonic brain and in vivo morphological defects of nestin-cre Sox2-deleted mutants (Sox2null). (a) Top to bottom, the Sox2 locus and targeting vector, the Sox2loxP allele obtained by homologous recombination, the Sox2loxP∆neo allele obtained by subsequent FLP recombinase–mediated excision of the neo cassette, and the Sox2loxP-null allele, resulting from Cre-mediated Sox2 excision. Filled triangles represent loxP sites (Cre recombinase substrate) and open vertical rectangles represent FRT sites (FLP recombinase substrate). Note that the Sox2loxP∆neo allele is identical to the wild-type locus, except for the insertion of the loxP and FRT sites. The dashed boxes under the Sox2 locus represent probes used for Southern analysis. E, EcoRI site; H, HindIII site; S, SalI site. (b) SOX2 immunohisto chemistry on wild-type (WT) and Sox2loxP∆neo/loxP∆neo; nestin-cre (Sox2null) embryonic brain at E10.5, 11.5 and 12.5. At E10.5, wild-type and mutant were still undistinguishable; SOX2 ablation was seen at E11.5, particularly ventrally, and was essentially complete by E12.5. Scale bar represents 200 µm. (c) Thionine staining of sections of P0, P2, P7 and adult hippocampus (left four columns, scale bar represents 500 µm) and adult forebrain (far right column, scale bar represents 200 µm). The mutant hippocampus, in particular the dentate gyrus, remained underdeveloped after P0 compared with wild type. We observed a smaller cortex and a prematurely interrupted corpus callosum in adult forebrain, along with the smaller hippocampus.
gyrus showed no BrdU incorporation (Fig. 2b). A transient increase in apoptotic cell death was also evident at P2 by TUNEL staining (Fig. 2b). There were no major changes in the lateral ventricle wall during this limited time window (Supplementary Fig. 1). The loss of neurogenesis and stem/precursor cells observed between P2 and P7 in the hippocampus suggests that Sox2 is important for the maintenance, but not in the genesis, of hippocampal NSCs.
a r t ic l e s Nestin GFAP
P0
Nestin GFAP
P2
Nestin GFAP
P7
Nestin GFAP
P0
Nestin GFAP
P2
Nestin GFAP
P7
Figure 2 Cellular defects of nestin-cre Sox2-deleted mutant brain. (a) Nestin/GFAP immunohistochemistry of hippocampus dentate gyrus, labeling postnatal NSCs13,14. Abundant nestin/GFAP-positive radial cells developed in normal brain between P0 and P7, but were lost in the mutant. Scale bars represent 50 µm. (b) Left, BrdU (top) and TUNEL (bottom) dentate gyrus labeling at P0, P2 and P7 (scale bars represent 100 µm). Right, quantification of BrdU-positive cells (top, wild-type set = 100%) or TUNEL-positive cells (bottom) (absolute numbers on y axes are total number of cells counted). Histograms represent mean ± s.d. values for n = 5 mice per genotype assayed for each stage.
Sox2 null
WT
a
BrdU
P0
BrdU
P2
BrdU
BrdU
P0
BrdU
P2
BrdU
Percentage of BrdU-positive cells % Cells 100 400 320 80 240 60 160 P7 40 80 20 P7
WT
Sox2 null
WT
b
0
Tunel
P0 Tunel
100 P0 Tunel
P2 Tunel
P7
Sox2 null
P7
60 20
NSCs and of neurogenesis in Sox2-mutant hippocampus suggests (together with the observed SHH expression defect in the mutant; Figs. 3 and 4) that non–cell-autonomous mechanisms may contribute to the observed phenotype.
E14.5
E14.5
E14.5
E14.5
P0
WT
Adult Sox2 deletion causes loss of hippocampal radial glia The loss of nestin/GFAP-positive radial glia and neurogenesis in post natal hippocampus might reflect subtle embryonic NSC damage that becomes evident only after birth and would thus be a developmental defect rather than an actual requirement for Sox2 in early postnatal NSC maintenance. To evaluate this issue, we deleted Sox2 postnatally in Sox2expressing cells of Sox2loxP∆neo/loxPneo mutants E12.5 (Fig. 6) by tamoxifen-dependent activation of creERT2 recombinase19 driven by the Sox2 tel encephalic-specific promoter and enhancer5. This transgene was highly expressed in the hippocampus, as shown by its efficient activa tion of a transgenic ROSA26RloxP-stop-loxP Shh Shh E12.5 lacZ reporter of Cre activity20 after tamoxifen administration (Supplementary Fig. 2). We administered tamoxifen for 8 d to 2-monthold Sox2loxPneo/loxPneo; Sox2-CreERT2 mice;
* Shh E14.5
E14.5
E14.5
P0
Sox2 null
E14.5
1250
Shh
Shh
Shh
E14.5
E14.5
E14.5
E14.5
E17.5
E17.5
E17.5
E17.5
WT
Figure 3 Sox2-deleted mutant brains have defective Shh and Wnt3a mRNA expression. Shh (top) and Wnt3a (bottom) in situ hybridization in E12.5–P0 brain sections. At E12.5, Shh expression in ventral forebrain was reduced in the mutant. At E14.5, expression was unmodified in midbrain (circle), but strongly reduced in diencephalon (arrowhead) and telencephalon (arrow), particularly anteriorly (arrow, right). At P0, no expression was detected in the mutant lateral ventricle (asterisk). At both E14.5 and E17.5, the Wnt3a signal (arrowhead) was severely decreased in the mutant cortical hem. Scale bars represent 500 µm.
*
Wnt3a
Wnt3a
Sox2 null
© 2009 Nature America, Inc. All rights reserved.
140 Tunel
P2
TUNEL-positive cells
P7 220 180
P2 Tunel
we used wild-type mice (treated or untreated with tamoxifen) and mice with the same geno type, but without tamoxifen administration, as controls (Fig. 6 and Supplementary Fig. 3). All mice were labeled with BrdU between days 8 and 10. We killed and analyzed the mice 4 d after the last tamoxifen administration (day 12) for SOX2 and for stem/precursor cell markers. P0 P2 P3 null WT Sox2 The numbers of SOX2/GFAP-positive cells with radial glia morphology and of GFAP/ nestin-positive cells at the base of the dentate gyrus (radial glia stem cells) in tamoxifen-treated hippocampi of Sox2loxPneo/loxPneo; Sox2creERT2 mice were reduced by over 40% relative to controls (Fig. 6a–c). Partial loss of SOX2-positive cells (Fig. 6a,b and Supplementary Fig. 3) was expected, as a result of the limited efficiency of induc ible Cre19. These data indicate that stem/precursor cells are lost in connection with Sox2 deletion; notably, a small proportion of GFAPpositive radial glia had empty SOX2-negative nuclei, suggesting that Sox2 deletion does not immediately lead to the loss of radial glia cells (Fig. 6a). Consistent with the loss of SOX2/GFAP- and nestin/GFAPpositive cells, the proliferation of precursors downstream to the stem
P0
Wnt3a
Wnt3a
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
a r t ic l e s Figure 4 SHH colocalizes with SOX2 in neural SHH SHH cells in postnatal neurogenic regions and its Sox2 Sox2 null mutants. expression is lost in Sox2 DG Hil Svz (a) SHH and SOX2 immunofluorescence on P0 neurogenic regions, dentate gyrus (left) and lateral ventricle (right). In normal dentate Hil gyrus, SHH (green) was expressed along radial SHH SHH DG Sox2 processes in dentate gyrus (DG) and in the Sox2 SHH SHH underlying hilus (Hil), where abundant SOX2Hil Svz Sox2 Sox2 positive nuclei (red) were seen. In the lateral DG ventricle, SHH was expressed in periventricular cells, together with SOX2. Expression was lost SHH SHH in Sox2null mutants at both locations. Inserts, DG Sox2 Sox2 higher magnification details, by confocal Hil microscopy, of lateral ventricle walls from the lower ventricle region (see connecting line), showing chains of SOX2/SHH double-positive cells lining the ventricular space (asterisk) in wild type, and loss of SHH and SOX2 in mutant. Scale bars represent 100 µm. (b) Colocalization of SHH (green) and SOX2 (red) in cells of wild-type dentate gyrus, lateral ventricle subventricular zone (Svz), and absence of SHH in Sox2null mutant cells (confocal microscopy; original magnification was 100x).
b
© 2009 Nature America, Inc. All rights reserved.
Sox2 null
Sox2 null
WT
WT
a
cell was also decreased, as seen by a reduction (40%) in the number of Ki67-positive (that is, cycling) cells and of recently labeled BrdU-posi tive cells. Doublecortin-positive cells, a more downstream precursor cell population that is partly nonproliferating21, and NeuN/BrdUpositive cells (early neurons) were decreased, but to a lesser extent (about 20%), suggesting that compensatory mechanisms may operate downstream to early precursors, as has been reported in Sox2 adult hypomorphic mutants10. These data indicate that Sox2 is required for NSC maintenance in the hippocampus during its early development (Figs. 1 and 2) and during adulthood. NSC exhaustion in vitro is rescued by conditioned medium To directly confirm that Sox2 is involved in NSC maintenance, we established neurosphere long-term cultures22 from in vivo Sox2deleted brains (Fig. 7). SOX2 protein is completely absent from these mutant cells11. We expanded large numbers of Sox2-deleted (or wild-type controls from the same litter) NSC populations briefly (1–2 passages) in basic fibroblast growth factor (bFGF) and epidermal growth factor (EGF)22. Subsequently, we plated (passage 0 in EGF; Fig. 7) aliquots of Sox2-deleted or wild-type cells in EGF alone (or in EGF and bFGF) and counted both the number of neurospheres WT + SHH-Ag Nestin GFAP
BrdU
Sox2 P7 Nestin GFAP
null
Sox2 null + SHH-Ag P7
P7 BrdU
P7
P7
P7
Percentage of BrdU-positive cells Cells %
Nestin GFAP
BrdU
P7
P7
P7
TUNEL-positive cells
100
375
Cells
80
300
300
60
225
225
40
150
150
20
75
75
WT + SHH-Ag Sox2 null Sox2 null + SHH-Ag
0
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
and total cells at each passage. Using P0 cultures (Fig. 7), we found little or no difference between mutant and wild-type cells during the initial expansion; however, by passage 2–4 in EGF, neurospheres and cell numbers were markedly decreased in mutant cultures (Fig. 7a). Subsequently, we continued to replate equal numbers of wild-type and Sox2-deleted cells; however, Sox2-deleted cultures were com pletely exhausted and died out by passages 5 or 6 (Fig. 7a). In EGF and bFGF, we obtained similar results, except that mutant cells would adhere to the plate at early passages (2–5) before becoming exhausted (data not shown). P7 cultures were similarly exhausted by passage 3 or 4 (Supplementary Fig. 4). E14.5 cultures were more variable, as some stopped growth around passage 4, whereas others continued to grow at a low rate (Supplementary Fig. 4). The observed exhaustion differs from that seen in a previous study23, which reported minor differences between wild-type and Sox2-mutant NSCs; however, these experiments (in FGF and EGF) did not extend the culture beyond pas sage 2, at which point little or no decay of proliferation was observable in our cultures (Fig. 7). The pronounced defect of NSC maintenance was accompanied by a strong decrease in neurosphere size (Fig. 7b) and the loss of lacZ staining (Fig. 7c), reflecting the activity of the beta-geo knock-in inserted into the Sox2 locus. We5 and others6 have previously shown that lacZ expression, reflecting the activity of the Sox2 gene, is retained in stem cells/early Figure 5 Stimulation of the Shh signaling pathway rescues hippocampal NSC and neurogenesis in Sox2 mutants. SHH-Ag was administered to Sox2null or wild-type controls. Top row, GFAP and nestin double immunofluorescence, showing hippocampal radial glia NSCs; these were lost in Sox2null hippocampi compared with wild types (see also Fig. 2a), but were substantially rescued by SHH-Ag (Sox2null + SHH-Ag; arrowheads point to GFAP and nestin double-positive radial glia, scale bars represent 50 µm). Middle row, BrdU immunofluorescence. Scale bars represent 100 µm. Bottom row, thionine staining of hippocampus sections. The dentate gyrus of Sox2null mutants, very poorly developed compared with wild types (see also Fig. 1), showed visible recovery of size in Sox2null SHH-Ag–treated mutants (arrowheads). Scale bars represent 500 µm. Left histogram, quantitation of BrdU-positive cells (wild type + SHH-Ag is set to 100%). Histogram bars represent mean ± s.d. values for five mice per genotype assayed for each stage and treatment (+SHH-Ag or untreated). BrdU labeling, essentially lost in Sox2null mutant (see also Fig. 2b), recovered to about 50% of wild-type levels in SHH-Ag–treated mutants. Total numbers of BrdU-positive cells for WT + SHH-Ag (right y axis) did not differ substantially from those found with untreated wildtype mice (see experiment in Fig. 2b). Right histogram, quantitation of TUNEL-positive cells at P2. SHH-Ag reduced the apoptosis seen in the mutant hippocampus at P2 (see also Fig. 2).
1251
a r t ic l e s a
WT
b
Sox2+/GFAP+ %
** **
100
loxP�neo/loxP�neo
c
GFAP Sox2
d
Nestin+/GFAP+
%
** **
100
100
60
60
20
20
20
e%
BrdU+
** **
f%
100
100
60 20 Versus Versus
P < 0.001 P = 0.002
Ki67+
** *
g
WT WT + Tam Sox2-creERT2 + Tam
** P < 0.01 * P < 0.05
P < 0.001 P < 0.001
Versus Versus
P < 0.001 P < 0.001
Nestin+
** **
%
60
Versus Versus
© 2009 Nature America, Inc. All rights reserved.
Sox2-creERT2; Sox2 GFAP Sox2
Versus Versus
P < 0.001 P < 0.001
DCX+ %
* *
h
BrdU+/NeuN+
%
100
100
60
60
60
20
20
Versus Versus
P < 0.001 P = 0.012
Versus Versus
** *
20 P = 0.041 P = 0.017
Versus Versus
P = 0.006 P = 0.019
precursors, but is lost in more differentiated progeny; thus, loss of lacZ staining in Sox2null neurospheres suggests a loss of NSC properties. We also observed increased numbers of beta-tubulin–expressing cells (that is, cells spontaneously differentiating into neurons) in mutant neuro spheres (Fig. 7d). At the same time, BrdU incorporation was decreased by at least 50%; furthermore, we detected increased apoptotic cell death in mutant cells (Fig. 7d), consistent with in vivo results (Fig. 2b). These data indicate that NSCs progressively lose their status, changing into more differentiated cell types. To evaluate the role of secreted molecules in wild-type versus mutant NSCs, we supplemented mutant cells with conditioned medium from wild-type NSC cultures. The conditioned medium efficiently rescued the proliferation defect of Sox2-deleted cells (Fig. 7c,e,f), as well as lacZ staining (Fig. 7c). In contrast, conditioned medium from Sox2-deleted cells had no effect on wild-type NSCs (data not shown). These experiments indicate that wild-type cells release molecule(s) into the medium that maintain the stem cell status and are not produced by mutant cells. Shh and Wnt3a are downregulated in vivo in Sox2null mutants (Figs. 3 and 4). In vitro, Shh (but not Wnt3a) was expressed in wildtype neurospheres, but was extremely downregulated in Sox2-deleted cultures (Fig. 7g). Transduction of Sox2-deleted cultures with a Sox2IRES-gfp–expressing lentivirus11 at passage 1 (in EGF), just before the decline of self-renewal (see Fig. 7a), induced Shh re-expression by 36 h (Fig. 7h) and rescued the formation of neurospheres, which continued to grow, in contrast with parallel cultures of nontransduced mutant cells (Fig. 7i). SHH is a NSC mitogen in brain and in NSC grown in vitro17,24–28. Addition of SHH (alone or with oxysterols, acting on the same pathway29; Fig. 7f) rescued the ability of mutant NSCs to con tinue to proliferate, albeit at substantially lower rate than cells treated with conditioned medium, without showing any effect on wild-type NSCs. Doubling times for wild-type cells were 33.8 ± 5.5 h, with no difference following treatment with conditioned medium or SHH (32 ± 1.8 and 33 ± 2.4 h, respectively); for mutant cells, doubling times in conditioned medium and SHH were 45.3 ± 3.8 and 60.7 ± 10 h, respectively. Bio30, a drug that generically stimulates the Wnt pathway, had no effect (data not shown). Other cytokine genes that are known to affect neural cells growth (Egf, Fgf2 and Fgf8) were expressed in neuro sphere cultures at normal levels (by RT-PCR), as were their receptors 1252
Figure 6 Sox2 deletion in adult brain leads to rapid loss of radial glia cells and of cell proliferation in the hippocampus dentate gyrus. Sox2 was deleted via tamoxifen treatment of Sox2loxP∆neo/loxP∆neo adult mice carrying a creERT2 transgene driven by the Sox2 telencephalic enhancer-promoter (Sox2-creERT2)11. Sox2-creERT2 induced efficient, tamoxifen-dependent activity in the expected expression domain11 (in the hippocampus, the cells at the base of the dentate gyrus), as verified by breeding with a ROSA26RloxP-stop-loxP lacZ reporter of Cre activity (Supplementary Fig. 2 and data not shown)20. (a) GFAP and SOX2 double immuno fluorescence on dentate gyrus of wild-type controls Sox2loxP∆neo/loxP∆neo without the cre-ERT2 transgene (left) and tamoxifen-treated Sox2-creERT2; Sox2loxP∆neo/loxP∆neo mutants (right) (confocal microscopy). The vast majority of GFAP-positive radial glia cells in wild types were positive for SOX2 (green nuclei); in mutant, together with GFAP/SOX2 double-positive cells (broad arrowheads), some cells were GFAP positive, but SOX2 negative (thin arrowheads, SOX2-negative nuclei surrounded by GFAP-positive cytoplasm; see also Supplementary Fig. 3). Scale bars represent 20 µm. (b–h) Quantitation of dentate gyrus cells that were positive, by immunofluorescence, for the proteins indicated above each histogram. P values (Student’s t test, two-tailed) are indicated below the histograms. Wild-type values were set to 100%. Histogram bars represent mean ± s.d. values for seven mice for each genotype (wild type, wild type + tamoxifen, Sox2loxP∆neo/loxP∆neo; Sox2-creERT2 + tamoxifen). Further controls (Supplementary Fig. 3 and data not shown), that is, Sox2loxP/+; Sox2-creERT2 and Sox2loxPneo/+; Sox2-creERT2 + tamoxifen, did not substantially differ from untreated wild type and wild type + tamoxifen. DCX, doublecortin.
and downstream effector molecules (for example, Egfr, Fgfr1, Notch1, Ctnnb1, Bmi-1, p21, p16, Hes5, Rbpj, etc; data not shown). The Shh gene is a direct target of Sox2 Shh expression was lost both in vivo in brain and in vitro in Sox2deleted NSC cultures (Figs. 4 and 7g). Shh expression is controlled by multiple functionally characterized regulatory regions, which are specifically active in different neural tube regions31. Several conserved (mouse versus human) potential SOX2-binding sites are present in three of the described regions (T3 and T4, active in telencephalon, and Die, active in the diencephalon) and 5 kb downstream of the gene (P) (Fig. 8a). In T3, one such site is centered on the regula tory region, whereas two further sites are 5 and 35 nucleotides from the described region31. Using electrophoresis mobility shift assays (EMSAs), we found that SOX2 bound to sites in T3, Die and P, but not in T4 (Fig. 8). To confirm that SOX2 is directly involved with these sites in vivo, we carried out chromatin immunoprecipitation (ChIP) experiments on embryonic brain cells. SOX2 bound three out of four of the tested regions (T3, Die and P) in vivo (Fig. 8b–d), whereas T4 sequences were not immunoprecipitated. Notably, we did not detect enrich ment in any of these sequences in SOX2 ChIPs that were carried out in parallel using Sox2null control brain chromatin (Fig. 8c), further demonstrating the SOX2 specificity of the immunoprecipitation. These data indicate that Shh is a direct target of SOX2. DISCUSSION Our results suggest that Sox2 is important during development for controlling NSC maintenance, at least in part via non–cell-autono mous autocrine mechanisms. In vivo, neural Sox2 ablation leads to a loss of SHH and Wnt3a, two signaling molecules that are active in NSC proliferation15–17,24–26, in some (but not all) of the embryonic/ neonatal neural domains that normally express them (Figs. 3 and 4), followed by a failure of postnatal development and NSC maintenance in the hippocampus (Fig. 2). Pharmacological activation of the Shh signaling pathway substantially rescued the maintenance of hippo campal stem cell development during early postnatal morphogenesis VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
a r t ic l e s
16
WT1 Null1
14
c
WT2 Null2
Sox2 null p4
WTp4
p8
p10
TUJ1
p12
f
WT + CM Null + CM
WT WT + S Null Null + S
10
12
WT + CM WT + S + O Null + CM Null + S + O
60
8
10 8
6
p2
p4
p8
4
p0
p2
h
Sox2 null
WT
M
p6
RT
+
–
Sox2 NT +LentiSox + – + –
35
38
40 30
WT
35 Sox2
NSC
38
40
null
NSC
LentiSox
WT –
Sox2 +
15
20
10
10
5
0
0
i
Shh Cycles 30
20
30
p4 null
WT
25
40
T
p0
null
–
30
50
W
4
Sox2 null
Percentage of TUNEL+ cells
Sox2
SHH Sox2
SHH Sox2
HPRT
NT
© 2009 Nature America, Inc. All rights reserved.
6
Percentage of BrdU+ cells
ul l
WT Null
14
p6
W T So x2 n
p4
So x2
e
p2
LentiSox-GFP
p0
TUNEL-DAPI
Sox2
4
nu ll
null
8 6
g
Sox2 null
WT
10 cond medium
Log10
12
Log10
d BRDU-TUJ1 DAPI
b
Sox2 null
a
Figure 7 Impaired maintenance of Sox2null NSCs in culture and rescue by extracellular factors. (a) Growth in EGF of neurosphere cultures from two wild-type and Sox2null P0 brains. Total cell numbers (Log 10) are reported over 12 passages (48 d). Data are representative of eight wild-type and eight mutant cultures. (b) Neurospheres from two wild-type and two Sox2null cultures at passage 4 (p4). (c) X-gal staining of two Sox2null cultures (p4) in normal medium (top) or in medium conditioned by wild-type cultures. Scale bars in b and c represent 100 µm. Blue color reflects the activity of Sox2β-geo, indicating an undifferentiated state 5,10. Most Sox2null spheres were partly or completely white; conditioned medium restored blue staining in most neurospheres and normal size. (d) BrdU (red), TuJ1 (green, β-tubulin) immunofluorescence and TUNEL (green) analysis of (passage 3) dissociated neurospheres and (histograms) percentage of BrdU- or TUNEL-positive cells in wild-type and Sox2null cells. Original magnification was 50x. (e,f) Growth of wild-type and Sox2null cultures in standard medium, medium conditioned by wild-type cells (CM) or in medium with SHH or SHH and oxysterols (S, S + O). We observed a response to CM in 6 out of 6 and to SHH in 4 out of 5 mutant cultures. (g) Top, RT-PCR of Shh mRNA from wild-type and Sox2null neurospheres. We detected no Shh mRNA in 2 out of 3 Sox2null lines and very low levels in another line. Bottom, SOX2 and SHH immunofluorescence in wild-type and Sox2null neurospheres. Original magnification was 200x. (h) Top, RT-PCR of Shh mRNA in Sox2null neurospheres either transduced with Sox2-gfp–expressing lentivirus (LentiSox) or not transduced (NT), and harvested after 36 h, versus wild-type control. Bottom, RT-PCR of Sox2 mRNA and Hprt (for normalization) in the same cultures. (i) Bright field and GFP fluorescence of Sox2-gfp lentivirus–transduced Sox2null (top) and nontransduced cultures (bottom) 1 week after transduction. Magnification as in b and c.
Sox2 is required for NSC maintenance The complete loss of hippocampal nestin/GFAP-positive stem/ progenitor cells starting at early stages of postnatal development (Fig. 2) might be viewed as a developmental defect resulting from damage to the embryonic NSCs. Do normally born adult stem cells still require Sox2 for their functions? Deletion of Sox2 in Sox2-expressing cells of adult mice caused a rapid loss of GFAP/nestin-positive stem/precursor cells and of cell proliferation in the dentate gyrus, indicating that Sox2 is still required in normally generated NSCs of the adult mouse
hippocampus (Fig. 6). The only partial Sox2 deletion obtained in adult hippocampal NSCs did not allow us to carry out a rescue experiment with SHH-Ag, as this would stimulate the nondeleted cells. Although SHH-Ag stimulates NSCs in the adult brain17, it is not known whether Shh is continuously required postnatally in hippocampus27. Thus, it remains possible that Sox2 controls hippocampal neurogenesis purely by cell-autonomous mechanisms at this stage. Our in vivo evidence for a role of Sox2 in NSC maintenance was further corroborated by our observation of a progressive and com plete loss of in vitro NSC renewal in Sox2-deleted neurosphere cul tures at P0 and P7 (Fig. 7 and Supplementary Fig. 4) and by the rescue of neurosphere formation obtained using Sox2 lentiviral transduction of mutant neurospheres (Fig. 7h,i).
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1253
(Fig. 5). In vitro experiments further suggest that it is NSCs them selves, or some of their early progeny, that control NSC maintenance via Sox2-dependent secretion of growth factors (Fig. 7).
a r t ic l e s a
Figure 8 Shh is a direct target of SOX2. (a) Sox2 consensus sites in flanking/ regulatory regions of the Shh gene. Black Chr. 5 P Shh T4 Die T3 29.22 Mb boxes, Shh exons; gray boxes, regions 28.78 Mb 3 2 1 tested by ChIP with SOX2 antibodies; vertical lines, Sox2 consensus. T3, Die and T4 overlapped with Shh regulatory elements active in telencephalon (T3, T4) or diencephalon (Die) 31. P is a region Ab Ab Ab downstream of Shh. (b) SOX2 ChIP, PCR Input Ab Input Ab Input Ab Input Ab Input IgG – Sox2 T Input IgG – Sox2 T Input IgG – Sox2 T of SOX2 antibody–precipitated E14.5 1 Sox2 Sox2 1 Sox2 1 Sox2 1 0.05 1 0.05 1 0.05 embryonic brain chromatin with primers P DIE P T3 DIE amplifying regions shown in a. Srr2 and Gfap (positive controls) are identified SOX2 T4 T3 targets11. Actin is a negative control 11. SRR2 Gfap Actin Input: input chromatin aliquots. ChIP with null null WT Sox2 WT Sox2 unrelated (antibody to SV40T, or pre-immune IgG) or no antibodies (−) were T4 negative controls. (c) Control SOX2 ChIP on Sox2null brain chromatin, compared with wild type. Ab, SOX2 antibody; In, input chromatin aliquots. (d) Quantitation of SOX2: – – + + – – + + – – + + 50 SOX2 ChIP experiments. Values are ratios P T3 Gfap 40 between PCR signal intensity of the SOX2 30 antibody–precipitated sample and that 20 of the three negative controls (average). T4 DIE SRR2 10 Results are average of three independent experiments. (e) EMSAs with nuclear 0 extracts from COS cells that were (+) –10 P T3 DIE T4 SRR2 Gfap Actin Probe: WT mut WT mut WT mut WT mut WT mut WT mut or were not (−) transfected with a Sox2-expressing vector 11. Probes contain the sites indicated in a (T3 only the two more downstream sites), as normal (WT) or mutated (mut) versions. Gfap and Srr2 are positive controls 11. P, Die and T3 were bound by SOX2 and T4 was not, consistent with our ChIP data.
c
Sox2 ChIP 14.5 dpc WT
b
Shh and Wnt3a expression depend on Sox2 The expression of Wnt3a and of Shh, two critical regulators of brain development, depends on Sox2 (Figs. 3, 4 and 7g,h). This regulation is context dependent. In fact, Shh expression was deficient in telen cephalon and diencephalon, but not in the midbrain and spinal cord of Sox2-deleted mutants; moreover, Shh and Wnt3a were expressed only in specific cell subsets of the Sox2-expression domain. Our ChIP results (Fig. 8) suggest that Sox2 might contribute to the activation of distant regulatory elements of the Shh locus. Whether Wnt3a is directly regulated by SOX2 remains to be defined, as regulatory regions have not been identified. The Notch pathway is strongly repressed in the eyes of mice by retina-specific Sox2 deletion32, but it was only moderately, if at all, decreased in the hippocampus of our mutants, as evaluated by the expression of the downstream effector Hes5 (Supplementary Fig. 5). Effector genes that mediate Sox2 function may thus differ in different neural tube regions and/or developmental stages (see below).
e
EMSA
ChIP signal intensity
© 2009 Nature America, Inc. All rights reserved.
d
Non–cell-autonomous effects of in vivo Sox2 ablation The almost complete postnatal loss of the dentate gyrus that we observed in Sox2-ablated brains (Figs. 1c and 2) closely resembles the effects of the Smo and Shh conditional knockout mice17. These knock outs have the same nestin-cre transgene that we used in this study and developed the expression defect with kinetics that are similar to those of Sox2 in our mutants. Thus, our data strongly suggest that the loss of Shh in the hippocampal primordium and the neonatal hilus of Sox2-deleted mutants (Figs. 3 and 4) may be sufficient to affect hippocampal development. This hypothesis is directly supported by the substantial rescue of stem/progenitor cells and neurogenesis (over tenfold increase in BrdU incorporation; Fig. 5) in Sox2-deleted P7 hippocampus following activation of the Shh pathway by a chemical
agonist; the marked stimulation of BrdU incorporation stands in con trast with the limited effect reported17 in wild-type mice (see also the similar BrdU incorporation in untreated versus treated wildtype mice in Figs. 2 and 5, respectively). Thus, these data indicate that limiting Shh activity in Sox2-deleted mutants contributes to the hippocampal NSC defect. In addition, the concomitant late loss in Sox2 mutants (from E14.5) of Wnt3a expression (Fig. 3) might justify the somewhat increased severity of the Sox2 deletion phenotype compared with that of the Smo and Shh knockout mice17. Indeed, total ablation of Wnt3a (nor mally expressed in the cortical hem by E10) by conventional knockout leads to the complete loss of the hippocampus15,16. In vivo, loss of stem cells and of neurogenesis was observed, in our Sox2 deletion model, mainly in postnatal hippocampus, but at much lower levels in other brain regions, such as the subventricular zone (Supplementary Fig. 1), indicating that the hippocampus is particularly sensitive to Sox2 deletion. The preferential hippocampal localization was consistent with its relatively high dependence on Wnt3a15,16 and Shh expression (at P15; see Fig. 5 in ref. 17), as compared with the subventricular zone, and with the increased responsiveness to SHH of brain NSCs at late stages of embryogenesis28 when most hippocampus development occurs. Moreover, the conditional deletion of Sox2 is complete only by E12.5 (Fig. 1), whereas Sox2 expression begins, in the nervous system, by E7.5 or earlier. It is possible that there is a restricted time window, during which the requirement for Sox2 in NSC is critical. Indeed, we found (unpublished data, A.L.M.F., R.F. and S.K.N.) that early Sox2 deletion (E9.5 instead of E12.5) with other cre-expressing transgenes causes marked neural cell loss in several fore-, mid- and hindbrain regions. This is consistent with the embryonic neural tissue loss that we observed previously in the germline Sox2 hypomorphic mutant10.
1254
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
a r t ic l e s
© 2009 Nature America, Inc. All rights reserved.
An additional possibility is that exhaustion of the proliferative ability of the stem cell in the subventricular zone, following Sox2 deletion, requires a critical number of cell divisions, leading to a late defect (as in other mutants33,34). It should be noted that exhaustion of stem cell renewal in the mutant requires several passages in vitro and multiple stem cell divisions (Fig. 7). Recently, another study reported conditional Sox2 deletion using a different nestin-cre transgene23. The phenotype observed in that study was more severe than that reported here, as the mice died just before birth, precluding the study of the hippocampus. Brain mor phology, however, was essentially normal; BrdU incorporation was substantially decreased in the ganglionic eminence, but not in the cortex. Overall, these data are consistent with ours, in the examined time window, indicating that late embryonic Sox2 deletion has only minor effects on brain development in general. Sox2-dependent NSC autocrine mechanisms in vitro In vitro, NSC cultures from Sox2-deleted forebrain progressively failed to grow and became exhausted, but exponential growth was almost fully rescued by medium conditioned by wild-type NSC cultures (Fig. 7e,f) together with full expression of a beta-geo reporter knocked-in at the Sox2 locus (Fig. 7c), reflecting NSC status5,10. Thus, in vitro, as in vivo, Sox2 has an important non–cell-autono mous role in controlling NSC growth through diffusible products. One such product was SHH, a direct Sox2 target (Fig. 8). Endogenous Hedgehog signaling is important for normal NSC growth in vitro25,26. Shh was widely expressed in wild-type neural cells in culture (Fig. 7g), as was Sox2 (ref. 11 and Fig. 7g), and was lost in Sox2null cells in vitro (Fig. 7g), consistent with our in vivo observations (Fig. 4). Addition of SHH to mutant NSCs in vitro rescued cell growth (Fig. 7f), although much less than conditioned medium, indicating that SHH is but one of the Sox2-dependent secreted factors. Indeed, an antibody to SHH slightly inhibited the effect of added conditioned medium on Sox2-deleted cells (as well as on wild-type cells; Supplementary Fig. 6). Furthermore, when Sox2 was re-expressed in mutant cells by lentiviral Sox2 transduction, just before the beginning of the growth decline, Shh expression was reactivated, together with neurosphere formation (Fig. 7h,i). These results indicate that the effect of Sox2 on NSCs growth and maintenance in vitro is partially mediated by SHH secretion, although other undefined factors are probably important. This conclusion complements our in vivo finding that SHH, in the hippocampus, was an important mediator of Sox2 function on NSC (Fig. 5). Neural cells are a component of the stem cell niche, which provides the necessary environment for NSC maintenance and expansion. The niche includes astrocytes, ependyma and non-neural cell types, such as endothelial cells35,36; indeed, it has been viewed as a specialized cell function separated from neural stem/precursor cells. The role of Sox2 in the control of Shh (and Wnt3a) in developing brain in vivo (Figs. 3 and 4) and in cells in vitro (Fig. 7), and evidence for direct SOX2 activity on the Shh gene (Fig. 8), suggest that there is a previ ously unknown self-regulatory loop for stem cell regulation in the neural niche during development.
of both tumorigenic mutations and experimental pharmacological treatments39,41, a role of Sox2 in cancer stem cell maintenance should be considered. Methods Methods and any associated references are available in the online version of the paper at http://www.nature.com/natureneuroscience/. Note: Supplementary information is available on the Nature Neuroscience website. Acknowledgments We thank G. Dehò and S. Zangrossi for the pcnBTn10 recAcat C-5507 E. coli C strain, S. Dymecki for the pLM-FLRT-3 vector and the FLPeR mouse strain, R. Klein through Jackson Laboratories for the nestin-cre mouse strain, P. Chambon for the creERT2 gene, Curis for the 1.2 SHH agonist18, S. Lugert for help with the Notch pathway expression studies, J. Cozzi and GenOway for help with ES cell targeting, C. Tiveron and Layline Genomics for blastocyst injections, C. Tiveron and L. Tatangelo for Sox2-creERT2 pronuclear injections, R. Caccia for help with in situ hybridization, E. Bagnaresi and A. Marinelli for help with constructs, A. Ronchi for advice on EMSAs and ChIP and D. Santoni for animal care. The monoclonal antibody to SHH developed by T.M. Jessell was obtained from the Developmental Studies Hybridoma Bank developed under the auspices of the US National Institute of Child Health and Human Development and maintained by the University of Iowa. This work was supported by Telethon (GGP05122), the European Economic Community (STEMBRIDGE), Fondazione Cariplo, Associazione Italiana Ricerca sul Cancro, Fondazione Banca del Monte di Lombardia, the Ministero Istruzione Università e Ricerca (Cofin) and FAR 2004-7 grants to S.K.N., and by the Max Planck Gesellschaft through V. Taylor. AUTHOR CONTRIBUTIONS R.F. and M.V. generated the Sox2 mutations in ES cells, and carried out animal breeding, cell cultures and RT-PCR. A.L.M.F. performed in situ hybridization and immunohistochemistry. E.L. carried out ChIP and EMSAs. J.M. performed the lentiviral rescue experiment. C.G. carried out the Notch signaling studies and initial work with in vitro cultures and for in vitro SOX2 deletion. C.L. participated in immunochemical studies. V. Tosetti participated in the in vitro culture work and immunocytochemistry. S.O. discussed the experiments and wrote the paper. V. Taylor participated in the in vivo rescue experiments, discussed data and supervised initial in vitro studies. S.K.N. devised the experiments, supervised the work and wrote the paper. Published online at http://www.nature.com/natureneuroscience/. Reprints and permissions information is available online at http://www.nature.com/ reprintsandpermissions/.
Perspectives It will be interesting to evaluate whether Sox2 contributes to ES cell pluripotency, in part, by non–cell-autonomous mechanisms. A role for Wnt3a (a Sox2 target in neural cells; Fig. 3) in reprogramming somatic cells to pluripotency was recently described37. Furthermore, Sox2 is expressed in many neural brain tumor cells, particularly in the stem cell component38–40. As the Shh pathway is an important target
1. Avilion, A.A. et al. Multipotent cell lineages in early mouse development depend on SOX2 function. Genes Dev. 17, 126–140 (2003). 2. Masui, S. et al. Pluripotency governed by Sox2 via regulation of Oct3/4 expression in mouse embryonic stem cells. Nat. Cell Biol. 9, 625–635 (2007). 3. Takahashi, K. & Yamanaka, S. Induction of pluripotent stem cells from mouse embryonic and adult fibroblast cultures by defined factors. Cell 126, 663–676 (2006). 4. Takahashi, K. et al. Induction of pluripotent stem cells from adult human fibroblasts by defined factors. Cell 131, 861–872 (2007). 5. Zappone, M.V., Galli, R., Catena, R. & Nicolis, S.K. Sox2 regulatory sequences direct expression of a beta-geo transgene to telencephalic neural stem cells and precursors of the mouse embryo, revealing regionalization of gene expression in CNS stem cells. Development 127, 2367–2382 (2000). 6. Suh, H. et al. In vivo fate analysis reveals the multipotent and self-renewal capacities of Sox2+ neural stem cells in the adult hippocampus. Cell Stem Cell 1, 515–528 (2007). 7. Fantes, J. et al. Mutations in SOX2 cause anophthalmia. Nat. Genet. 33, 461–463 (2003). 8. Kelberman, D. et al. Mutations within Sox2/SOX2 are associated with abnormalities in the hypothalamo-pituitary-gonadal axis in mice and humans. J. Clin. Invest. 116, 2442–2455 (2006). 9. Sisodiya, S.M. et al. Role of SOX2 mutations in human hippocampal malformations and epilepsy. Epilepsia 47, 534–542 (2006). 10. Ferri, A.L., Cavallaro, M., Braida, D., Di Cristofano, A. & Nicolis, S.K. Sox2 deficiency causes neurodegeneration and impaired neurogenesis in the adult mouse brain. Development 131, 3805–3819 (2004). 11. Cavallaro, M., Mariani, J., Lancini, C., Latorre, E. & Nicolis, S.K. Impaired generation of mature neurons by neural stem cells from hypomorphic Sox2 mutants. Development 135, 541–557 (2008). 12. Medina, D.L. et al. TrkB regulates neocortex formation through the Shc/PLCγmediated control of neuronal migration. EMBO J. 23, 3803–3814 (2004).
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1255
© 2009 Nature America, Inc. All rights reserved.
a r t ic l e s 13. Merkle, F.T., Tramontin, A.D., Garcia-Verdugo, J.M. & Alvarez-Buylla, A. Radial glia give rise to adult neural stem cells in the subventricular zone. Proc. Natl. Acad. Sci. USA 101, 17528–17532 (2004). 14. Doetsch, F. The glial identity of neural stem cells. Nat. Neurosci. 6, 1127–1134 (2003). 15. Roelink, H. Hippocampus formation: an intriguing collaboration. Curr. Biol. 10, R279–R281 (2000). 16. Lee, S.M., Tole, S., Grove, E. & McMahon, A. A local Wnt3a signal is required for development of the mammalian hippocampus. Development 127, 457–467 (2000). 17. Machold, R. et al. Sonic Hedgehog is required for progenitor cell maintenance in the telencephalic stem cell niches. Neuron 39, 937–950 (2003). 18. Frank-Kamenetsky, M. et al. Small-molecule modulators of Hedgehog signaling: identification and characterization of Smoothened agonists and antagonists. J. Biol. 1, 10 (2002). 19. Weber, P., Metzger, D. & Chambon, P. Temporally controlled targeted somatic mutagenesis in the mouse brain. Eur. J. Neurosci. 14, 1777–1783 (2001). 20. Akagi, K. et al. Cre-mediated somatic site-specific recombination in mice. Nucleic Acids Res. 25, 1766–1773 (1997). 21. Kempermann, G., Jessberger, S., Steiner, B. & Kronenberg, G. Milestones of neuronal development in the adult hippocampus. Trends Neurosci. 27, 447–452 (2004). 22. Gritti, A. et al. Multipotential stem cells from the adult mouse brain proliferate and self-renew in response to basic fibroblast growth factor. J. Neurosci. 16, 1091–1100 (1996). 23. Miyagi, S. et al. Consequence of the loss of Sox2 in the developing brain of the mouse. FEBS Lett. 582, 2811–2815 (2008). 24. Lai, K., Kaspar, B.K., Gage, F.H. & Schaffer, D.V. Sonic hedgehog regulates adult neural progenitor proliferation in vitro and in vivo. Nat. Neurosci. 6, 21–27 (2003). 25. Palma, V. & Ruiz i Altaba, A. Hedgehog-GLI signaling regulates the behavior of cells with stem cell properties in the developing neocortex. Development 131, 337–345 (2004). 26. Palma, V. et al. Sonic hedgehog controls stem cells behavior in the postnatal and adult brain. Development 132, 335–344 (2005). 27. Balordi, F. & Fishell, G. Hedgehog signaling in the subventricular zone is required both for the maintenance of stem cells and the migration of newborn neurons. J. Neurosci. 27, 5936–5947 (2007).
28. Ahn, S. & Joyner, A.L. In vivo analysis of quiescent adult neural stem cells responding to Sonic hedgehog. Nature 437, 894–897 (2005). 29. Wang, Y., McMahon, A.P. & Allen, B.L. Shifting paradigms in Hedgehog signaling. Curr. Opin. Cell Biol. 19, 159–165 (2007). 30. Sato, N., Meijer, L., Skaltsounis, L., Greengard, P. & Brivanlou, A.H. Maintenance of pluripotency in human and mouse embryonic stem cells through activation of Wnt signaling by a pharmacological GSK-3-specific inhibitor. Nat. Med. 10, 55–63 (2004). 31. Jeong, Y., El Jaick, K., Roessler, E., Muenke, M. & Epstein, D.J. A functional screen for sonic hedgehog regulatory elements across a 1 Mb interval identifies long-range ventral forebrain enhancers. Development 133, 761–772 (2006). 32. Taranova, O.V. et al. SOX2 is a dose-dependent regulator of retinal neural progenitor competence. Genes Dev. 20, 1187–1202 (2006). 33. Fasano, C.A. et al. shRNA knockdown of Bmi-1 reveals a critical role for p21-Rb pathway in NSC self-renewal during development. Cell Stem Cell 1, 87–99 (2007). 34. Shi, Y. et al. Expression and function of orphan nuclear receptor TLX in adult neural stem cells. Nature 427, 78–83 (2004). 35. Riquelme, P.A., Drapeau, E. & Doetsch, F. Brain micro-ecologies: neural stem cell niches in the adult mammalian brain. Phil. Trans. R. Soc. Lond. B 363, 123–137 (2008). 36. Gilbertson, R.J. & Rich, J.N. Making a tumour’s bed: glioblastoma stem cells and the vascular niche. Nat. Rev. Cancer 7, 733–736 (2007). 37. Marson, A. et al. Wnt signaling promotes reprogramming of somatic cells to pluripotency. Cell Stem Cell 3, 132–135 (2008). 38. Lee, J. et al. Tumor stem cells derived from glioblastomas cultured in bFGF and EGF more closely mirror the phenotype and genotype of primary tumors than do serum-cultured cell lines. Cancer Cell 9, 391 (2006). 39. Clement, V., Sanchez, P., de Tribolet, N., Radovanovic, I. & Altaba, A. HEDGEHOGGLI1 signaling regulates human glioma growth, cancer stem cell self-renewal, and tumorigenicity. Curr. Biol. 17, 165–172 (2007). 40. Schmitz, M. et al. Identification of SOX2 as a novel glioma-associated antigen and potential target for T cell–based immunotherapy. Br. J. Cancer 96, 1293–1301 (2007). 41. Romer, J.T. et al. Suppression of the Shh pathway using a small molecule inhibitor eliminates medulloblastoma in Ptc1(+/−)p53(−/−) mice. Cancer Cell 6, 229 (2004).
1256
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
© 2009 Nature America, Inc. All rights reserved.
ONLINE METHODS Generation of Sox2loxP and Sox2loxP∆neo alleles and mice. The targeting vector (Fig. 1a) was generated with Sox2 genomic fragments from a 129/SvJ mouse genomic library, cloned into the pLM-FLRT-3 vector (gift from S. Dymecki, Harvard Medical School). The vector (Fig. 1a) includes the 5′ homology arm (10,942 bp, a HindIII fragment, position 59,108–70,060 in BAC, Genbank acces sion number AL606746), the 5′ loxP site, the HindIII-SalI fragment containing the Sox2 gene (position 70,060–76,650), the 3′ loxP site, a neomycin-resistance cas sette flanked by FRT sites, substrate of FLP recombinase42, the 3′ homology arm, a 2.5 kb SalI fragment (position 76,650–79,180 in AL606746), and a dyphteria toxin–encoding (DTA) cassette (a 2-kb SalI-XhoI PGK-DTA fragment). To allow stable replication in E. coli, we changed the vector backbone to pBR322 carrying a SalI-AatII deletion. To improve stability, we grew the construct in a pcnB mutant E.coli (pcnBTn10 recAcat C-5507 E. coli C; a gift of S. Zangrossi and G. Dehò, University of Milano) that reduces the replication efficiency of ColE1 origins. The vector was linearized with KpnI for transfection. Gene targeting was carried out in CJ7 ES cells. G418-resistant clones were analyzed by PCR (primers 1–4; Fig. 1a) and Southern blotting; EcoRI-digested genomic DNA was probed with a 2.6-kb SalI-EcoRI fragment (Fig. 1a). In the wild-type Sox2 locus, this hybridized to a 15.5-kb fragment and to a 7.5-kb frag ment in the Sox2loxP allele. The Sox2 internal probe (Fig. 1a) hybridized to a 15.5-kb fragment in the wild type or to a 6.5-kb in the Sox2loxP allele. Chimeras were bred to C57BL/6J or B6D2F1 females to obtain mice carry ing the Sox2loxP allele, which were bred to FLPeR (FLP recombinase–expressing) mice42 to remove the neomycin-resistance cassette. nestin-cre mice12 were crossed with mice carrying the Sox2βgeo null allele1,5 to obtain double heterozygotes; experimental mice were obtained crossing Sox2loxP∆neo/loxP∆neo with Sox2βgeo/+; nestin-cre mice. Experimental procedures involving mice were approved by the Italian Ministry of Health. Immunohistochemistry and in situ hybridization of brain sections. Embryos or dissected brains were fixed overnight at 4 °C in 4% paraformaldehyde (PFA, wt/vol) in phosphate-buffered saline (PBS), cryoprotected with 30% sucrose in PBS and cryostat sectioned onto slides (SuperFrost Plus). For immunohistochemistry, sections were incubated overnight at 4 °C with primary antibody diluted in 0.1% BSA (wt/vol) in PBS, extensively washed in PBS containing 0.1% BSA, incubated for 1 h at 20–25 °C with a secondary anti body conjugated with a fluorochrome (Molecular Probes), washed in PBS and mounted in FluorSave reagent (Calbiochem 345789). For primary antibodies, we used mouse antibody to SHH (1:10, Developmental Studies Hybridoma Bank, 5E1), rabbit antibody to SHH (1:100, Chemicon), rabbit antibody to SOX2 (1:500, AB5603, Chemicon), mouse antibody to SOX2 (1:50, R&D MAB2018)11, mouse antibody to nestin (1:500, MAB353, Chemicon), rabbit antibody to GFAP (1:400, 18-0063, Zymed). Jagged-1 immunohistochemistry was carried out as described previously43. For SHH/SOX2 immunohistochemistry, antigen unmasking was carried out by boiling sections in 0.01 M citric acid and 0.01 M sodium citrate for 3 min in a microwave oven before blocking. All images were collected on a Zeiss Axioplan 2 microscope and processed with Adobe Photoshop 7.0 software (Adobe Systems). We used five mice of each genotype and stage for the experi ments shown in Figures 2–4. For in situ hybridization, embryos were fixed in 4% PFA overnight, cryopre served in 30% sucrose, frozen in OCT compound and sectioned onto slides. In situ hybridization was performed as described previously44, on fixed tissue with digoxigenin-labeled single-stranded RNA probes at 65 °C, followed by incubation with BM Purple AP substrates (Roche). We used antisense RNA probes to Shh45, Wnt3a16 and Hes5 (ref. 43). TUNEL and BrdU labeling. Cryostat sections were washed for 10 min in PBS containing 0.1% Triton X-100 (vol/vol), fixed for 5 min in 4% PFA and washed again in PBS. The enzymatic reaction was then performed at 37 °C according to the manufacturer’s protocol (G3250, Promega). NSCs, dissociated and attached to slides, were treated the same way. Quantitative immunocytochemical data (for sections) represent mean ± s.d. for cell counts of the hippocampus in consecu tive sections through its entire length, every 150 µm. For P0, P2 and P7 labeling, BrdU was injected at 100 µg per g of body weight 2 h before the mice were killed. Proliferating cells were revealed by BrdU immunohistochemistry on frozen sec tions (or dissociated cells attached to slides). Sections (or cells) were denaturated
doi:10.1038/nn.2397
with 2N HCl in H2O at 37 °C for 1 h (30 min for cells), neutralized with 0.1 M borate buffer (pH 8.5) for 10 min, blocked in 1% BSA in PBS (with 0.2% Triton X-100 for cells) for 1 h at room temperature (20–25 °C), and probed with a Harlan rat monoclonal antibody to BrdU (1:500 in 0.1% BSA in PBS) overnight at 4 °C. All images were collected on a Zeiss Axioplan 2 microscope and processed with Adobe Photoshop 7.0 software. Quantitative immunocytochemical data represent mean ± s.d. for cell counts of the hippocampus, obtained by reacting and counting 1 every 4 sections through the entire hippocampus, every 150 µm. We assayed five mice for each genotype and stage. In vivo rescuing and adult deletion. For the in vivo rescuing experiment, Shh agonist #1.2 (a gift from Curis)17,18 was administered to pregnant mothers by oral gavage at a concentration of 1.5 mg ml−1 in 0.5% methylcellulose (wt/vol)/0.2% Tween 80 (vol/vol), 100 µl per 10 g of body weight, as described previously18, on E12.5, 14.5, 17.5 and P1. Newborns were then analyzed at P2 or P7 (as for Fig. 2). For adult Sox2 deletion, we generated mice carrying a transgene with the 5.7-kb Sox2 5′ telencephalic enhancer/promoter5 driving creERT2 (ref. 19; a gift from P. Chambon (Institut de Génétique et de Biologie Moléculaire et Cellulaire)). Six transgenic lines were obtained and each had the expected tamoxifen-dependent Cre activity5, as assayed by breeding the mice to a transgenic mouse with a loxPlacZ reporter of Cre activity20; the most efficient line was used (deletion efficiency was about 50% in embryonic and adult neural precursors in vivo). Adult mice (2.5 months) were treated with tamoxifen (20 mg ml−1 in ethanol/ corn oil 1:10, 0.1 mg per g of body weight) by intraperitoneal injection, one injection per d for 8 d (day 1–8), treated with BrdU on days 8, 9 and 10, perfused 4 d later with 4% PFA, and cryostat sectioned (20-µm sections). We analyzed 5–7 sections that were representative of the whole hippocampus length (1 every 8 for P2–P7; 1 every 10 for adults) for GFAP, nestin, SOX2, BrdU (as for Fig. 2, see above), doublecortin21 (Santa Cruz goat SC8066, 1:200), Ki67 (NovoCastra rabbit polyclonal, 1:500) and NeuN (Chemicon mouse MAB377, 1:500) immuno fluorescence. Statistical analysis was performed by Student’s t test (two-tailed), comparing experimental groups of mice of the same genotype. In all histograms, values are shown as mean ± s.d. from a number of independent samples (n indicated). NSC culture. P0 brain cells for NSC (neurosphere) cultures from wild-type and Sox2-deleted (Sox2loxP/β-geo; nestin-cre) mice were obtained as described previ ously21,41, plated at 20,000 cells per ml in 25-ml flasks and cultured to expand their number in complete medium (2% B27 (vol/vol) in DMEM F12 with Glutamax43), supplemented with 10 ng ml−1 of EGF, 10 ng ml−1 of bFGF with 0.2% heparin, vol/vol), and passaged (by dissociation and dilution) two or three times every 5 d. After sufficient cell numbers had been obtained, cells dissociated from neuro spheres (10,000 cells per 0.5 ml in triplicate, in 24-well plates) were passaged into complete medium with EGF, but not bFGF; this initial passage is taken to represent passage 0 (Fig. 7a,e,f). In some experiments, SHH (200 ng ml−1, R&D Systems, 464-SH), oxysterols (22(S)-and 20α-hydroxycholesterol, Sigma H5884 and H6378, 0.2 µM each) or conditioned medium from normal cells were added (Fig. 7c,e,f). No substantial difference was found between treatments with SHH and oxysterols and SHH only (Fig. 7f and data not shown). Conditioned medium was the supernatant obtained by growing wild-type cells in complete medium (with EGF) for 2 d, followed by removal of the cells by centrifugation. Complete removal of the wild-type cells was assessed by routine microscopy. Note that cells grown from mutant cell cultures treated with conditioned medium were all X-gal positive, confirming that they originated from Sox2β-geo/loxP (mutant), but not (potentially contaminating) Sox2+/+ (wild type), cells (Fig. 7c). In some experiments, conditioned medium from mutant cells was added to wild-type cell cultures, but no effect was observed (data not shown). Sox2-GFP lentivirus transduction. Sox2null neurosphere cells grown for three passages in EGF and bFGF were passaged once in EGF, dissociated, plated in EGF at 40,000 cells per well in 4-well chambered slides and transduced 4 h later with a Sox2-gfp–expressing lentivirus as described previously11. Virus was removed by medium change at 24 h; nontransduced controls were treated equivalently (without virus). For RT-PCR (Fig. 7h), cells were harvested 36 h after infection and lysed in Tryzol, RNA was extracted from two transduced (or nontransduced) wells, DNAase treated (RQ1 DNAase, Promega) and reverse transcribed with
nature NEUROSCIENCE
© 2009 Nature America, Inc. All rights reserved.
r andom hexamer primers using an Applied Biosystems cDNA reverse transcrip tion kit (with a reverse transcriptase–negative control). About 5 µl. of a 1:10 dilution (adjusted following normalization by Hprt expression) were used for the RT-PCR shown in Figure 7h. In parallel, for the cultures shown in Figure 7i, 20,000 cells per well were plated in suspension (to allow for subsequent neuro sphere formation) in a 24-well plate and transduced as described above (4 h after plating, changing medium after 24 h). Small amounts of medium were added every 3 d; cells were passaged first after 1 week and re-passaged twice before freezing (nontransduced controls were exhausted at passage 1). Images showing neurosphere formation by transduced cells were taken 7 days after transduction. For RT-PCR, we used two primers for Shh, Shh forward (GGA AAG AGG CGG CAC CCC AAA AAG) and Shh reverse (CTC ATC CCA GCC CTC GGT CAC TCG). Annealing was carried out at 65 °C for 45 cycles and we obtained a 278-bp reaction product. RT-PCR for Sox2 and HPRT was as described previously11. ChIP and EMSA. ChIP was carried out as described previously46. Briefly, BDF1 E14.5 brains were fixed in 1% PFA in DMEM cell culture medium (with 10% fetal calf serum (vol/vol) and penicillin-streptomycin) for 15 min at 37 °C, then crosslinking was blocked treating with 125 mM glycine for 15 min at 37 °C. Tissue was digested with 10 mg ml−1 collagenase at room temperature (20–25 °C), centrifuged at 800g and harvested in ice-cold RIPA Lyses Buffer (0.1% SDS, 1% DOC, 150 mM NaCl, 2 mM EDTA, 1% NP-40 (vol/vol), protease inhibitor cocktail; Complite Roche) at a final concentration of about 104 cells per ml. Keeping samples ice cold, we passed the cells through a syringe needle (U-100 insulin syringe, four times) and sonicated them to shear chromatin into DNA fragments of about 1,000 base pairs (six intervals of 25 s, 2-min ‘rest on ice’ between intervals, power setting 3, Branson 150 cell disruptor). Cell debris were removed by a 15-min centrifugation at 9,000g. Prior to immunoprecipitation, 1 µg µl−1 of yeast tRNA was added as a blocking agent to the sonicated samples. Sepharose nProt A 4fast flow (Amersham #17.5280.01) beads were pre-blocked in RIPA buffer with low molecular weight salmon sperm DNA, and the antibody of interest was added. We used 10 mg of antibody-coated beads to immuno precipitate chromatin from 106 cells (antibody to SOX2, R&D MAB2018; IgG, Santa Cruz sc-2027; SV40 T-Ag, Santa Cruz sc-147). Immunoprecipitation was carried out for 2 h at 4 °C and beads were then washed three times in an equal volume of ice-cold washing buffer (100 mM Tris-HCl (pH 8.0), 500 mM LiCl, 1% NP-40). DNA-protein complexes were eluted by ice-cold elution buffer (1% SDS, 50 mM NaHCO3) and treated with 200 mM NaCl, 100 ng µl−1 RNAse A and 200 ng µl−1 proteinase K at 52 °C for 3 h. DNA was purified by phe nol-chloroform-isoamylic alcohol (25:24:1, pH 8) equal volume extraction and ethanol precipitated. DNA was resuspended in 10 mM Tris-HCl (pH 7.5) and 1 mM EDTA and used as template for PCR. Zero (background) values (Fig. 8e)
nature NEUROSCIENCE
were measured as the average of the values (PCR band intensities) obtained with the three negative controls in Figure 8c (no antibody or unrelated antibody, antibody to SV40T antigen; pre-immune IgG), as described previously47. We used five primer sets for PCR, P_fw: CCAGGTACATCTTTGATTGACATTCAGC and P_rev: TGTTTCTGAACTAAGTTGGTGTTGCGTT, die_f (AAA ATA AAC CCC AGC CAG ACG CAA CC) and die_r (TTC ATC TGA TCC CCT GCT TTT AGC), T3_f (GGA AAT GGC ACT GAG AGT AAG AAC) and T3_r (TTT CCA AAT CAG CAG AGT GGC TCC G), T4_f (CTT TAA TTT TGC GTT ATT TCC AGC C) and T4_r (TCC GCT TAA ATC TTA GAG AGC G), and actin_f (GGT CAG AAG GAC TCC TAT GT) and actin_r (ATG AGG TAG TCT GTC AGG TC). The Gfap and Srr2 primer sequences were described previously11. EMSAs were carried out as described11. Nuclear extracts were obtained from COS cells transfected with Sox2-IRES-GFP11 expression vector or backbone alone. For probes, we used T3s (AAG AAC AAA GAG CTG TTC GGA GCA AGC AGC ACA CTT), T3r (GTG AAC AAA GTG TGC TGC TTG CTC CGA ACA GCT CTT TGT TCT T), T3smut (AAG CCC CCC GAG CTG TTC GGA GCA AGC AGC ACA CCC CGC CC), T3rmut (GTG GGC GGG GTG TGC TGC TTG CTC CGA ACA GCT CGG), Dies (CAA CCT GCC TTT GTT CCC TAA GCT GCT T), Dier (AAG CAG CTT AGG GAA CAA AGG CAG GTT G), Die_mutr (CAA CCT GCC CCC GCC CCC TAA GCT GCT T), Die_muts (AAG CAG CTT AGG GGG CGG GGG CAG GTT G), T4s (TCT CTA CAG AAC AAA GTG GGC TTT ACC T), T4r (AGG TAA AGC CCA CTT TGT TCT GTA GAG A), T4_mutr (TCT CTA CAG CCC CCC GTG GGC TTT ACC T), T4_muts (AGG TAA AGC CCA CGG GGG GCT GTA GAG A), Pr (AGG GAG GGG GGC ATT GTG TAC AAG CCC TG), Ps (CAG GGC TTG TAC ACA ATG CCC CCC TCC CT), P_mutr (AGG GAG GGG GGC CCC GCG TAC AAG CCC TG) and P_muts (CAG GGC TTG TAC GCG GGG CCC CCC TCC CT). 42. Farley, F.W., Soriano, P., Steffen, L.S. & Dymecki, S.M. Widespread recombinase expression using FLPeR (flipper) mice. Genesis 28, 106–110 (2000). 43. Nyfeler, Y. et al. Jagged1 signals in the postnatal subventricular zone are required for neural stem cell self-renewal. EMBO J. 24, 3504–3515 (2005). 44. Conlon, R.A. & Herrmann, B.G. Detection of messenger RNA by in situ hybridization to postimplantation embryo whole mounts. Methods Enzymol. 225, 373–383 (1993). 45. Echelard, Y. et al. Sonic hedgehog, a member of a family of putative signaling molecules, is implicated in the regulation of CNS polarity. Cell 75, 1417–1430 (1993). 46. Weinmann, A.S. & Farnham, P.J. Identification of unknown target genes of human transcription factors using chromatin immunoprecipitation. Methods 26, 37–47 (2002). 47. Szutorisz, H. et al. Formation of an active tissue-specific chromatin domain initiated by epigenetic marking at the embryonic stem cell stage. Mol. Cell. Biol. 25, 1804– 1820 (2005).
doi:10.1038/nn.2397
a r t ic l e s
Presynaptic CaV2 calcium channel traffic requires CALF-1 and the 2 subunit UNC-36
© 2009 Nature America, Inc. All rights reserved.
Yasunori Saheki & Cornelia I Bargmann Presynaptic voltage-gated calcium channels provide calcium for synaptic vesicle exocytosis. We show here that a green fluorescent protein–tagged 1 subunit of the Caenorhabditis elegans CaV2 channel, UNC-2, is localized to presynaptic active zones of sensory and motor neurons. Synaptic localization of CaV2 requires the 2 subunit UNC-36 and CALF-1 (Calcium Channel Localization Factor-1), a neuronal transmembrane protein that localizes to the endoplasmic reticulum. In calf-1 mutants, UNC-2 is retained in the endoplasmic reticulum, but other active-zone components and synaptic vesicles are delivered to synapses. Acute induction of calf-1 mobilizes preexisting UNC-2 for delivery to synapses, consistent with a direct trafficking role. The 2 subunit UNC-36 is likewise required for exit of UNC-2 from endoplasmic reticulum but has additional functions. Genetic and cell biological interactions suggest that CALF-1 couples intracellular traffic to functional maturation of CaV2 presynaptic calcium channels. Neuronal voltage-gated calcium channels (VGCCs) are central regulators of synaptic vesicle exocytosis, dendritic integration and calcium-dependent gene regulation1. A VGCC consists of one pore-forming α1 subunit that defines intrinsic channel properties, and auxiliary α2δ, β and sometimes γ subunits that modify channel kinetics and channel density2. Specific genes in the CaV gene family encode physiologically distinct α1 subunits. Mammalian L-type channels with CaV1 α1 subunits are mainly required for gene regulation and dendritic integration; N-, P/Q- and R-type channels with CaV2 α1 subunits are required for neurotransmitter release and dendritic calcium transients; and T-type channels with CaV3 α1 subunits are required for repetitive firing1. Disruptions of VGCC function are implicated in human epilepsy, migraine, autism-spectrum disorders and bipolar disease, underlining the importance of these channels in the regulation of neuronal excitability and function3. CaV1, CaV2 and CaV3-related genes are found in invertebrates as well as vertebrates4. The fruitfly Drosophila melanogaster and the nematode C. elegans each have one predicted CaV2 α1 subunit, encoded by the cacophony and unc-2 genes, respectively. Fly cacophony mutants are inviable, with defects in calcium-dependent neurotransmitter release at the neuromuscular junction suggesting the loss of the presynaptic calcium current5,6. A green fluorescent protein (GFP)-tagged Cacophony protein is localized to presynaptic active zones, consistent with a role at synapses7. C. elegans unc-2 mutants are uncoordinated, with defects in evoked neurotransmitter release at the neuromuscular junction8–10. These phenotypes suggest a conserved role for CaV2 channels as presynaptic regulators of synaptic transmission. The surface expression and localization of presynaptic VGCCs can be affected by channel subunit composition and by other proteins. The α2δ auxiliary subunit increases channel activity and plasma membrane expression of mammalian CaV2 α1 subunits11 and increases
synaptic expression and activity of the Drosophila Cacophony CaV2 protein12,13. The β auxiliary subunit increases plasma membrane expression of many mammalian VGCC classes14,15. Other proteins that regulate presynaptic VGCC localization in vivo include the Drosophila active zone protein Bruchpilot16, the Drosophila eighttransmembrane-domain protein Fuseless17 and the vertebrate extracellular matrix protein laminin-β2 (ref. 18). Many more candidate regulators of presynaptic VGCCs have been studied in cultured cells, including scaffolding proteins such as CASK, Mint and Veli19 and the dynein light chain protein Tctex1 (ref. 20). To complement studies of VGCCs in cultured cells, and to explore CaV2 channel traffic in vivo, we here analyze neuronal calcium channel localization and function in C. elegans. The C. elegans genome encodes three predicted VGCC α1 subunits: egl-19 (CaV1), unc-2 (CaV2) and cca-1 (CaV3)4. UNC-2 is a candidate presynaptic voltage-gated calcium channel on the basis of its sequence similarity to CaV2 channels, neuronal expression and synaptic transmission defects when mutated. We show that a functional GFP-tagged UNC-2 is concentrated at presynaptic active zones of sensory neurons and motor neurons. UNC-2 localization requires the α2δ subunit UNC-36 and a newly described endoplasmic reticulum protein, CALF-1 (Calcium Channel Localization Factor-1). CALF-1 and UNC-36 have partly overlapping activities in the traffic and functional maturation of UNC-2 channels. RESULTS The CaV2 1 subunit UNC-2 localizes to presynaptic zones A full-length, GFP-tagged unc-2 cDNA rescued the uncoordinated movement of unc-2 mutants when expressed from a pan-neuronal promoter (see Online Methods). To examine its subcellular localization, GFPUNC-2 was expressed under cell type–specific
Howard Hughes Medical Institute and Laboratory of Neural Circuits and Behavior, The Rockefeller University, New York, New York, USA. Correspondence should be addressed to C.I.B. (
[email protected]). Received 24 April; accepted 9 July; published online 30 August 2009; doi:10.1038/nn.2383
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1257
a r t ic l e s
a
AWC
b
Synaptic vesicles
AWC
* Merge
RAB-3::mCherry
e
Active zone
*
AWC
GFP::UNC-2
Presynaptic regions
*
Axon
Merge
ELKS-1::mCherry
Dorsal cord/DD5-DD6
AWB
Dendrite
f
*
GFP::UNC-2
h
g
*
* GFP::UNC-2
d
c
Ventral cord/VD5
i
GFP::UNC-2
j Axon
VD
DD
RAB-3::mCherry
RAB-3::mCherry
Axon
Dendrite
© 2009 Nature America, Inc. All rights reserved.
Merge
k
DA9
Merge
l
GFP::UNC-2
Presynaptic region
Axon
* RAB-3::mCherry
Ventral cord
n
m *
*
Dorsal cord
Dendrite
Dendrite Merge
DA9
Figure 1 GFP-tagged UNC-2 localizes to presynaptic puncta in sensory neurons and motor neurons. (a–c) Representative images of GFPUNC-2 (green) and RAB-3mCherry (red) in the AWC cell body (white arrowhead) and axon (yellow arrowheads). The more dorsal cell body is AWB (asterisk). (d–f) Representative images of GFPUNC-2 and ELKS-1mCherry. (g) Schematic of AWC processes, with synapses in red. (h,i) Representative images of GFPUNC-2 and RAB-3mCherry in GABAergic motor neurons: DD axons (dorsal nerve cord) and VD axons (ventral nerve cord). (j) Schematic of VD and DD processes. VD has a presynaptic region in its ventral process and DD has a presynaptic region in its dorsal process. (k–m) Representative images of GFPUNC-2 and RAB-3mCherry in the synaptic region of DA9 cholinergic neurons. The central autofluorescent region is the intestine (asterisk). (n) Schematic of DA9 processes. In all figures, the AWC promoter is from odr-3, and is also expressed weakly in AWB, ASH, AWA and ADF sensory neurons; the VD and DD promoter is from unc-25; the DA9 promoter is from itr-1; and all data are taken from adult worms, unless otherwise noted. Head is to the left and dorsal is up in all images unless otherwise noted. Scale bar, 10 µm.
promoters together with the fluorescent synaptic-vesicle marker RAB-3mCherry (hereafter, RAB-3)21. When expressed in AWC olfactory neurons, GFPUNC-2 localized to axonal puncta that overlapped with RAB-3, consistent with presynaptic localization (Fig. 1a–g). GFPUNC-2 was also present in the cell body but was excluded from the dendrite, cilia and nucleus. When expressed in VD and DD GABAergic motor neurons, GFPUNC-2 localized with RAB-3 in the ventral and dorsal nerve cords (Fig. 1h–j). When expressed in the DA9 cholinergic motor neuron, GFPUNC-2 localized with RAB-3 in the dorsal presynaptic region of the axon (Fig. 1k–n). In each case, the GFPUNC-2 protein was present in the cell body but largely excluded from dendrites and asynaptic regions of axons. Presynaptic calcium channels function at active zones, the plasma membrane sites of synaptic vesicle secretion. GFPUNC-2 puncta were often more focal than RAB-3 puncta (Fig. 1), suggesting that UNC-2 might localize to active zones. In agreement with this idea, GFPUNC-2 in AWC axons colocalized closely with the active zone markers ELKS-1mCherry and SYD-2mCherry (Fig. 1d–f, Supplementary Fig. 1a–c). To define genes required for UNC-2 localization to synapses, we first examined candidate mutants using GFPUNC-2 and RAB-3 expressed in AWC. Synaptic vesicle clustering and active zone structure in C. elegans are regulated by SAD-1 kinase, SYD-2 (liprin-α), SYD-1 and RPM-1 (Highwire)22–25. In all of these mutants, GFPUNC-2 puncta associated with RAB-3 in AWC axons, although there were defects in the spacing, size and number of clusters (Supplementary Fig. 2a–d). Several other candidate genes had no obvious effect on
GFPUNC-2 localization in AWC: elks-1, the C. elegans homolog of Drosophila bruchpilot16; nrx-1, the sole C. elegans neurexin homolog; lin-2 (CASK), lin-7 (Veli) and lin-10 (MINT), PDZ proteins of the tripartite complex; or the synaptic exocytosis and endocytosis mutants unc-13, unc-10 (RIM), dpy-23 (AP2), unc-101 (AP1), unc-11 (AP180) and unc-31(CAPS) (Supplementary Fig. 2e–o). A mutation in the KIF1A kinesin gene unc-104 caused RAB-3 to disappear from AWC axons, consistent with the known requirement for KIF1A in synaptic vesicle traffic26, but GFPUNC-2 puncta were present and apparently normal in unc-104 mutants (Supplementary Fig. 2p). A partial loss of function mutation in the KIF5 kinesin heavy chain gene unc-116 affected RAB-3 localization but maintained GFPUNC-2 colocalization with RAB-3 (Supplementary Fig. 2q). The absence of obvious GFPUNC-2 phenotypes in these mutants does not exclude subtle functions, redundant functions or functions in other classes of neurons. unc-36 α2δ mutants have uncoordinated phenotypes and other neuronal phenotypes similar to those of unc-2 mutants10,27,28. GFPUNC-2 was barely detectable in the AWC axons of a null unc-36 mutant but was still detectable in the cell body, demonstrating a requirement for UNC-36 in the sorting, folding or localization of UNC-2 in vivo (Supplementary Fig. 2r). RAB-3 puncta were normal, suggesting that synaptic vesicle clustering was unaffected.
1258
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
UNC-2 synaptic puncta are reduced in calf-1 mutants A genetic screen for mutants with altered GFPUNC-2 expression in AWC yielded three mutants with few axonal GFPUNC-2 puncta but apparently normal RAB-3 puncta (see Online Methods).
a r t ic l e s
AIB
Control
unc-2(Ij1)
1
0
calf-1(ky867)
1
** GFP::UNC-2
1
RAB-3:: mCherry 100
o
** GFP::UNC-2 RAB-3:: mCherry
**
**
1
0.2
lf-
0.4
400 350 300 250 200 150 100 50 0
ca
Merge
0.6
m
Dorsal cord
l
Merge
1
18 16 14 12 10 8 6 4 2 0
Axon RAB-3:: mCherry
)
RAB-3::mCherry
400
60 40 20
+ –
0 –20
+
–40
–
Average ∆F/F (%) 10–60 s
80
Response
300
200
100
0
–60
Odor
lf-
1
12/37*
ca
) Ij1
–100
2(
37 10 20 30 40 50 60
l
34 10 20 30 40 50 60 0
18/34**
c-
0
3/35
–100
un
0
35 10 20 30 40 50 60
tro
–80
–
on
+
C
© 2009 Nature America, Inc. All rights reserved.
n
RAB-3::mCherry
l
0.8
** Axon GFP::UNC-2
RAB-3:: mCherry
1.2
5 0
Axon
Dorsal cord
10
tro
GFP::UNC-2
GFP::UNC-2
1.4
15
Ij1
Ventral cord/VD5
Cell body
20
2(
j
k
calf-1(ky867)
Axon
25
on
i
Dorsal cord/DD5-DD6
** GFP::UNC-2
Merge
ELKS-1::mCherry
GFP::UNC-2
calf-1(ky867)
*
*
Control calf-1
c-
*
f
**
C
e
Active zone
Merge
h 30 AWC
AWC
un
d
AWC
*
RAB-3::mCherry
2 1.8 1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 0
Body bends per 2 min
*
* GFP::UNC-2 calf-1(ky867)
g
c
Number of clusters
Synaptic vesicles
Puncta number per 50 µm
b
Normalized total fluorescence intensity
AWC
Normalized total fluorescence intensity per 50 µm
a calf-1(ky867)
Figure 2 Presynaptic GFPUNC-2 puncta are lost in calf-1(ky867) mutants. (a–c) Representative images of GFPUNC-2 and RAB-3mCherry in AWC neuron of a calf-1(ky867) mutant. (d–f) Representative images of GFPUNC-2 and ELKS-1mCherry in a calf-1(ky867) mutant. White arrowhead, AWC cell body; yellow arrowheads, AWC synapses; asterisk, AWB cell body. Compare Figure 1a–f. (g,h) Quantification of GFPUNC-2 and RAB3mCherry in AWC: normalized total fluorescence intensity (g) and number of fluorescent clusters (h). (i,j) Representative images of GFPUNC-2 and RAB-3mCherry in VD and DD neurons in a calf-1(ky867) mutant. Compare Figure 1h,i. (k,l) Quantification of GFPUNC-2 and RAB-3mCherry in 50 µm covering DD5 and DD6 axons in the dorsal nerve cord: normalized total fluorescence intensity (k) and number of fluorescent puncta (l). Scale bar, 10 µm. (m) Quantification of swimming behavior in M9 buffer. All error bars, s.e.m. In g,h,k–m, results different from wild-type controls by unpaired t-test, **P < 0.01. (n) Calcium signals in AIB interneurons upon removal of the attractive odor isoamyl alcohol, which is sensed by AWC. Heat maps of individual recordings are shown for wild-type (n = 35), unc-2 (n = 34) and calf-1 (n = 37) adults. Odor was removed at t = 10 s. (o) Average AIB response to odor removal for traces shown in n. Lines mark median response. Neurons with an average ∆F/F ≤ 0 (where F is fluorescence) were scored as failures; both mutants differed from wild type in the fraction of failures, *P < 0.05 or **P < 0.01 by chi-squared test.
Two mutants had an uncoordinated phenotype and did not complement the α2δ subunit mutant unc-36(e251), suggesting that they are mutant for unc-36. The third mutant affected a new gene, here named calf-1(ky867) for CAlcium channel Localization Factor-1. In calf-1(ky867) mutants, GFPUNC-2 was nearly undetectable in AWC axons, but the synaptic vesicle marker RAB-3 (Fig. 2a–c) and the active zone markers ELKS-1 (Fig. 2d–f) and SYD-2 (Supplementary Fig. 1d–f) appeared normal. The total fluorescence intensity of axonal GFPUNC-2 as well as the number of puncta per axon were greatly reduced, whereas GFPUNC-2 fluorescence in the cell body was slightly increased (Fig. 2g,h); minimal effects were detected upon similar quantification of RAB-3 (Fig. 2g,h). These results suggest that calf-1 mutants have a selective defect in presynaptic calcium channel localization.
calf-1 also affected GFPUNC-2 localization in motor neurons. The dorsal nerve cord of calf-1 mutants had reduced GFPUNC-2 fluore scence and few GFPUNC-2 puncta but near-normal RAB-3 puncta, suggesting a loss of UNC-2 from DD synapses (Fig. 2i–l). In the ventral nerve cord, some GFPUNC-2 puncta were visible, but these puncta did not localize with RAB-3 at VD synapses (Fig. 2j). Consistent with a defect in motor neurons, calf-1 mutants had a distorted sinusoidal posture and moved very slowly on agar surfaces or in liquid, like unc-2 mutants (Fig. 2m and data not shown). In calf-1 mutants, the GFPUNC-2 signal in AWC cell bodies overlapped with the endoplasmic reticulum marker CP450mCherry29, suggesting that GFPUNC-2 was retained in the endoplasmic reticulum (Supplementary Fig. 3a–c). Support for this conclusion came from examining GFPUNC-2 in ventral cord processes of VD and
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1259
a r t ic l e s Figure 3 calf-1 encodes a type I transmembrane protein. (a) A predicted topology of CALF-1, with a transmembrane domain near its N terminus and basic and proline-rich regions in the predicted cytosolic region. The ky867 allele has a termination codon after the transmembrane domain. (b) Phylogenetic tree of CALF-1 in nematodes. (c) Alignment of predicted nematode CALF-1 proteins. Invariant amino acids are in red.
a
H2N
© 2009 Nature America, Inc. All rights reserved.
Litomosoides sigmodontis CALF-1 Onchocerca volvulus
Transmembrane domain ky 867 Q42Stop
CALF-1
Brugia malayi CALF-1 Ascaris suum CALF-1 Cytosol Meloidogyne incognita CALF-1 C. elegans CALF-1
c DD neurons. In wild-type worms, GFPUNC-2 elegans in ventral cord processes was largely sepa- C. Ascaris suum Litomosoides sigmodontis rated from the endoplasmic reticulum marker Onchocerca volvulus Meloidogyne incognita CP450mCherry, but in calf-1 mutants, Brugia malayi GFPUNC-2 and CP450mCherry overlapped extensively (Supplementary Fig. 3d–i). These results suggest that GFPUNC-2 preferentially accumulates in endoplasmic reticulum or related components in calf-1 mutants. C. elegans suum We examined the functional consequences of Ascaris Litomosoides sigmodontis Onchocerca volvulus unc-2 and calf-1 mutations by calcium imaging Meloidogyne incognita of AWC sensory neurons and AIB interneurons, Brugia malayi which are postsynaptic targets of AWC. AWC calcium levels reported by the fluorescent indicator G-CaMP fall slightly upon addition of the attractive odor isoamyl alcohol and rise upon odor removal; similar signals are subsequently observed in AIB interneurons30. In unc-2 and calf-1 mutants, the AWC sensory responses to odors were of normal magnitude (Supplementary Fig. 4), but postsynaptic AIB responses to odor removal were reduced (Fig. 2n and Supplementary Fig. 4). Approximately half of the unc-2 and one third of calf-1 AIB neurons did not respond to odor removal, although those AIB neurons that did respond had similar response magnitudes to those seen in the wild type (Fig. 2o). These results suggest that unc-2 and calf-1 affect neuronal signaling between AWC and AIB neurons in similar ways, reducing, but not eliminating, synaptic communication (see Discussion). calf-1 encodes a neuronal transmembrane protein Genetic mapping and transgenic rescue identified calf-1 as the predicted gene B0250.2 (Fig. 3, Online Methods and Supplementary Fig. 3j). Sequencing of calf-1(ky867) DNA revealed a C to T mutation resulting in an early stop codon in the B0250.2 open reading frame. The calf-1 mutant is fully recessive, and a calf-1 gene with the ky867 mutation did not affect GFPUNC-2 localization or locomotion when injected into wild-type worms (data not shown). These results suggest that calf-1(ky867) is a loss-of-function allele of B0250.2. calf-1 encodes a predicted type I transmembrane protein with a hydrophobic membrane-spanning region, a highly basic region and a proline-rich region (Fig. 3a). The region C-terminal to the transmembrane domain is predicted to be cytosolic. We identified homologs of calf-1 in other nematode species (Fig. 3b) but not in other organisms. Among nematodes, conservation of CALF-1 was highest in the predicted transmembrane domain and the adjacent basic region (Fig. 3c). The uncoordinated phenotype and defective GFPUNC-2 localization in calf-1 mutants were rescued by a 0.5-kb calf-1 cDNA expressed under 0.8 kb of calf-1 upstream sequence (Fig. 4a,b). When the same 0.8 kb promoter was used to drive expression of GFP, fluorescence was detected in many or all neurons but not in other tissues (Fig. 4c,d). Coexpression with an odr-1mCherry transgene confirmed that the calf-1 promoter drove expression in AWC neurons (Fig. 4e–g). 1260
b
177 amino acids
Proline-rich region
COOH Transmembrane domain
Basic region
Basic motif ky 867 Q42Stop Proline-rich region
C terminus
Expression of the calf-1 cDNA under the control of the pan euronal tag-168 promoter rescued the GFPUNC-2 localization n phenotypes and locomotory behavior of calf-1 mutants, but expression from the muscle-specific myo-3 promoter did not (Fig. 4a,b). Expression of calf-1 under the AWC-selective odr-3 promoter rescued GFPUNC-2 localization in AWC neurons, and expression of calf-1 in VD and DD motor neurons using the unc-25 promoter rescued dorsal GFPUNC-2 localization in DD neurons (Fig. 4h,i). These results suggest that calf-1 acts cell autonomously to localize UNC-2. A calf-1 cDNA that was tagged with GFP fully rescued the locomotion defects in calf-1 mutants (see Online Methods). When expressed in AWC neurons, CALF-1GFP was localized exclusively to the cell body and not to axons or synapses (Fig. 4j–l). The CALF-1GFP signal overlapped extensively with mCherry-labeled endoplasmic reti culum markers CP450, cb5 and RAMP4 (Fig. 4m–o, Supplementary Fig. 5a–f) but not with the Golgi marker ManIImCherry (Fig. 4p–r). When expressed in VD and DD motor neurons, CALF-1GFP was present in cell bodies and in a few puncta in ventral processes; these puncta did not overlap with RAB-3 or the Golgi marker but did overlap with an endoplasmic reticulum marker (Supplementary Fig. 5g–r). These results suggest that CALF-1 is a neuron-specific endoplasmic reticulum protein. CALF-1 functions in the endoplasmic reticulum To identify sequences necessary for CALF-1 function, we tested mutant proteins for their ability to rescue either GFPUNC-2 clusters in AWC or coordinated locomotion. Deletion of the CALF-1 transmembrane domain or replacement with the integrin PAT-3 transmembrane domain eliminated its activity (Fig. 5a). CALF-1 was active after individual deletion of three other regions: the basic region (deletion I), the proline-rich region (deletion II) or the Cterminal region (deletion III). Simultaneous deletion of all three regions inactivated CALF-1, but inclusion of either the basic region or the combination of the proline-rich and C-terminal regions was sufficient for rescue (deletion I-VI). Any of the three regions of CALF-1 that promote its function—the transmembrane domain, the basic region and the combined proline-rich and C-terminal region—was sufficient to cause endoplasmic reticulum VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
a r t ic l e s
j
AWC
d
Ventral cord
m
p
AWC cell body
Synapse
lf1
l
*
*
RAB-3::mCherry
CALF-1::GFP AWC cell body
Dorsal cord: GFP::UNC-2
k
*
calf-1::GFP
lf1 un
AWC axon: GFP::UNC-2
c
© 2009 Nature America, Inc. All rights reserved.
tro l
od lf1; ca
AWC axon: GFP-UNC-2
::c a
C on
tro lfl 1; lf- ca cal f lf1; 1: -1 t a :c ca g lf- -16 alf1; 1 8 m ::c yo al -3 f-1 ::c al f-1
tro l
0
ca
ca tro l ca f-1; l lf- ca ca 1; l lfta 1: f-1 ca g- :c a 1 lf1; 68 lf-1 m ::c yo al -3 f-1 ::c al f-1
C on
5
c25
0
0
10
ca
5
15
C on
50
20
**
16 14 12 10 8 6 4 2 0
lf1;
100
10
25
VD/DD rescue
ca
150
15
Puncta per 50 µm
200
20
**
30
lf r-3 -1 ::c al f-1
250
**
i 18
AWC rescue
ca
300
25
Number of clusters
350
h 35
**
ca
Body bends per 2 min
** **
400
Number of clusters
b 30
450
C on
a
n
ER
Merge
o
calf-1::GFP
e
Head
calf-1::GFP
f
g odr-1::mCherry
CP450::mCherry
CALF-1::GFP
q
Golgi
Merge
r
Merge CALF-1::GFP
ManII::mCherry
Merge
Figure 4 CALF-1 acts cell-autonomously in neurons and localizes to endoplasmic reticulum. (a,b) Rescue of calf-1(ky867) mutants with calf-1 cDNA under the endogenous calf-1 promoter, pan-neuronal tag-168 promoter or muscle-specific myo-3 promoter. (a) Swimming behavior in M9 buffer. (b) GFPUNC-2 clusters in AWC axons. **P < 0.01, strains different from calf-1(ky867) control by Bonferroni t-test. (c) Expression of calf-1GFP, 0.8 kb of promoter sequence. Scale bar, 100 µm. (d) Boxed region from c. Motor neurons in the ventral nerve cord express calf-1GFP. Arrowheads, cell bodies. (e–g) AWC expresses odr-1mCherry and calf-1GFP. Scale bar, 10 µm. (h,i) Cell-specific rescue of calf-1(ky867) mutants. (h) GFPUNC-2 clusters in AWC axons, odr-3calf-1 rescue. (i) GFPUNC-2 clusters in DD (dorsal cord), unc-25calf-1 rescue. **P < 0.01, strains different from calf-1(ky867) controls by unpaired t-test. All error bars, s.e.m. (j–l) Representative images of CALF-1GFP and RAB-3mCherry in AWC neurons. CALF-1GFP is not visible at RAB-3-positive synapses. Arrowhead, AWC cell body; asterisk, AWB cell body. Scale bar, 10 µm. (m–r) Localization of CALF-1 in AWC cell body. (m–o) CALF-1GFP and the endoplasmic reticulum (ER) marker CP450mCherry in AWC. (p–r) CALF-1GFP and the Golgi marker ManIImCherry in AWC. Scale bar, 5 µm.
retention of a GFP-tagged protein in intestinal cells (Fig. 5b and Online Methods). Embedded in the basic region and C-terminal region of CALF-1 are multiple Arg-X-Arg motifs (Fig. 3c), which can function as endoplasmic reticulum retention motifs in other transmembrane proteins31. A C-terminal truncation of CALF-1 that removed Arg-Gln-Arg and ArgLys-Arg motifs (RKR deletion) was competent for rescue, but a larger C-terminal deletion that eliminated Arg-Gln-Arg, Arg-Lys-Arg, Arg-Ala-Arg and Arg-Leu-Arg motifs (RLRE deletion) inactivated CALF-1 (Supplementary Fig. 6a–c). Notably, a small fusion protein consisting of the CALF-1 transmembrane domain and a 16-amino-acid, arginine-rich, endoplasmic reticulum retention motif from the G-protein coupled α2 adrenergic receptor32 partly rescued calf-1 mutations (Supplementary Fig. 6a,b); similar fusions of the CALF-1 transmembrane domain to Lys-Asp-Glu-Lys (KDEL) or Lys-Lys-Tyr-Leu (KKYL) endoplasmic reticulum retention motifs were inactive. These results suggest that arginine-rich endoplasmic reticulum retention motifs contribute to calf-1 activity. unc-36 affects UNC-2 maturation and function To gain further insight into the relationships between unc-2, calf-1 and unc-36, we examined various phenotypes and genetic interactions in
these mutants. Both canonical null unc-36 mutations and new unc-36 mutations from our screen caused defects in GFPUNC-2 localization that resembled those of calf-1 mutants: GFPUNC-2 was absent from AWC axons and dorsal VD and DD processes, but synaptic RAB-3 localization appeared normal (Fig. 6a–d). In the ventral nerve cord, GFPUNC-2 in VD and DD neurons rarely overlapped with the synaptic vesicle marker RAB-3 (Fig. 6e) but overlapped extensively with the endoplasmic reticulum marker CP450 (Fig. 6f). These observations suggest that unc-36 mutations cause GFPUNC-2 to accumulate in the endoplasmic reticulum. In agreement with this hypothesis, unc-36 mutants had more accumulation of GFPUNC-2 in the AWC cell body and perinuclear region than did wild type, although the effect was less marked than in calf-1 mutants (Supplementary Fig. 7a–d). A biologically active, GFP-tagged UNC-36 protein was localized both to the plasma membrane and to internal membranes of neurons, suggesting that it could function either in the endoplasmic reticulum with CALF-1, in the synapse with UNC-2 or at both locations (Fig. 6g). In AWC neurons, UNC-36GFP was largely perinuclear and overlapped with the endoplasmic reticulum marker CP450 (Supplementary Fig. 7e–g); unlike GFPUNC-2, it was
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1261
a r t ic l e s
© 2009 Nature America, Inc. All rights reserved.
Figure 5 Structure-function analysis of CALF-1. (a) Schematic representation of CALF-1 and mutants tested for rescue. For AWC axon clusters: ++, >20 GFPUNC-2 clusters per worm; −, <3 GFPUNC-2 clusters. Compare Figure 2h. For locomotion: ++, >250 body bends per 2 min; −, <100 body bends per 2 min. Compare Figure 2m. (b) Distribution of endoplasmic reticulum (ER) retention motifs in CALF-1. Top, proteins tested by expression in intestinal epithelial cells. Bottom, plasma membrane and endoplasmic reticulum localization of representative fusion proteins in intestinal cells in L4 larva. Arrowheads mark regions of strong CALF-1 expression. Scale bar, 10 µm.
a
ky867 Q>Stop Transmembrane Basic region Proline-rich domain region
C terminus 20 aa Full length
UNC-2 localization Locomotion ++ ++
Deletion I (18–93)
++
++
Deletion II (94–143)
++
++
Deletion III (144–177)
++
++
Deletion IV (94–177)
++
++
Deletion V (18–93, 144–177)
–
–
Deletion VI (18–143)
–
–
Transmembrane deletion
–
–
Transmembrane replacement
–
–
not concentrated at AWC synapses. Perinuclear b ER UNC-36GFP localization in AWC was unchanged Surface in calf-1 mutants; similarly, CALF-1GFP localizaER tion in AWC, VD and DD was unchanged in unc-2 ER and unc-36 mutants (Supplementary Fig. 7h–v). ER Genetic interactions were consistent with ER related functions of calf-1, unc-36 and unc-2. All Surface three mutants and all double mutants were slowER moving but not paralyzed, with similar phenotypes Surface (Fig. 6h). Overexpression of untagged UNC-2 from a pan-neuronal promoter significantly improved CALF-1 TM the calf-1 locomotion phenotype (Fig. 6h). This ER retention ER retention ER retention PAT-3 TM result suggests that the locomotion defect in calf-1 motif motif motif is related to reduced unc-2 activity and supports a ER PAT-3::CALF-1 (Cytosolic region)::GFP primary role for calf-1 as a cofactor for unc-2. Overexpression of calf-1 from a pan-neuronal promoter rescued synaptic GFPUNC-2 puncta in unc-36 mutants, an effect that was weak in AWC and CP450::mCherry Merge robust in VD and DD (Fig. 6i–k). However, calf-1 Surface PAT-3::CALF-1(C terminus)::GFP overexpression did not rescue the locomotion defects of unc-36 mutants (Fig. 6h), suggesting that unc-36 mutants are defective in locomotion even when some UNC-2 is delivered to synapses. unc-36 overexpresMerge CP450::mCherry sion did not rescue GFPUNC-2 localization or locomotion defects in calf-1 mutants (Fig. 6h–j). calf-1, unc-2 and unc-36 also function together in a calcium- to calf-1 mutants (Fig. 7a–c). The adult rescue of calf-1 mutants argues dependent developmental pathway that generates asymmetric gene- for a role of CALF-1 in ongoing delivery of UNC-2 to synapses and expression patterns in the left and right AWC neurons33. Left–right against an essential role in synaptic development. The rapid action of calf-1 after heat shock made it possible to examasymmetry is disrupted in about 50% of unc-2 mutants, an effect that is enhanced in unc-2 egl-19 (CaV1) double mutants28. As unc-36 ine effects of calf-1 on the dynamic behavior of UNC-2 protein. We mutations affect both CaV1 and CaV2 channels, unc-36 has a stronger designed a pulse-chase protocol to test the mobilization hypothesis defect than unc-2 (ref. 28). calf-1 mutants had defects in AWC gene directly, using hscalf-1 and a fluorescently labeled pool of UNC-2 expression that closely resembled those of unc-2 null mutants, and protein (Fig. 7d). UNC-2 was tagged at its N terminus with the photo an unc-2 calf-1 double mutant was similar to the single mutants convertible protein Dendra2, which irreversibly changes from green (Supplementary Table 1). These results suggest that the calf-1 muta- to red emission upon UV irradiation35. Dendra2UNC-2 protein tion specifically affects unc-2 function in AWC and not the genetically behaved similarly to GFPUNC-2, both before and after photoconseparable activities of unc-36 and egl-19 in the same cell. version: it had a synaptic location in wild-type worms but accumulated in cell bodies of calf-1 mutants (data not shown). In the pulse-chase CALF-1 acts acutely to deliver UNC-2 to synapses experiment, a pool of Dendra2UNC-2 protein in the cell bodies of The genetic analysis of calf-1 suggests that UNC-2 accumulates in the tail was photoconverted to red in calf-1; hscalf-1 worms raised endoplasmic reticulum until CALF-1 allows its exit, but they do not at low temperatures. After photoconversion, these worms were subdemonstrate a direct mobilization of UNC-2. To examine the acute jected to a heat shock to induce calf-1 expression (Fig. 7d,e). In 7 of 9 effects of calf-1, we used the heat-shock promoter hsp16.2 (ref. 34) to worms subjected to heat shock, red Dendra2UNC-2 was mobilized drive expression of calf-1 under temperature control. A 3-h heat pulse from the cell bodies to puncta within the nerve ring, where many tail in adult worms was sufficient to rescue synaptic GFPUNC-2 locali- neurons form synapses (Fig. 7f). We found no red Dendra2UNC-2 zation in AWC neurons, and it also restored coordinated locomotion puncta in the nerve ring in the absence of heat shock (Fig. 7f, 1262
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
a r t ic l e s a
unc-36(e251)
AWC
b
Synaptic vesicles
*
*
e
Dorsal cord
unc-36(e251)
*
RAB-3::mCherry
GFP::UNC-2
d
c
GFP::UNC-2
Merge Ventral cord
f
Ventral cord
GFP::UNC-2
GFP::UNC-2
RAB-3::mCherry
RAB-3::mCherry
Merge
Merge
h
g
Body bends per 2 min
*
400 350
CP450::mCherry
Merge NS NS
NS NS
** **
300 250 200 150 100
in vivo calcium imaging, a relatively lowresolution method, so they do not provide * detailed information on synaptic mechanisms. However, the general conclusions are UNC-36::GFP consistent with electrophysiological studies at the C. elegans neuromuscular junction i j showing that unc-2 mutants have reduced 30 20 synaptic release but retain residual synaptic ** 18 25 16 function (ref. 9 and J. Madison and J. Kaplan, Dorsal cord unc-36(e251); tag-168::calf-1 14 k 20 personal communication). The axonal NCA 12 15 10 channels, which are related to sodium leak 8 GFP::UNC-2 10 6 channels, are attractive candidates for a sec** 4 5 ond presynaptic activity, since these mutants 2 0 0 have variable failures in presynaptic calcium RAB-3::mCherry signals that are reminiscent of the variable failures in unc-2 mutants36. EGL-19 CaV1 channels are also candidates; although EGL-19 Merge is expressed mainly in the cell body (Y.S., unpublished data), inhibitory interactions between EGL-19 and UNC-2 may allow AWC axon: GFP::UNC-2 Dorsal cord: GFP::UNC-2 EGL-19-dependent compensation in unc-2 n = 9 worms). These results demonstrate that UNC-2 within the cell mutants28. Homeostatic compensation may also upregulate post body, most likely the endoplasmic reticulum, is acutely mobilized by synaptic glutamate receptors to potentiate the AIB response37. CALF-1. In agreement with this conclusion, the heat-shock protocol Efficient exit of a GFP-tagged UNC-2 from the endoplasmic reticulum significantly reduced the amount of red Dendra2UNC-2 in the cell requires both the α2δ subunit UNC-36 and the endoplasmic reticulum body while increasing the amount at synapses (Fig. 7g). protein CALF-1. Proteins that promote the surface expression of channels can be divided into two categories: auxiliary subunits and regulators of DISCUSSION biogenesis38. Auxiliary proteins such as TARPs (for glutamate receptors) Calcium channels in C. elegans, like their homologs in other animals, and MinK (for potassium channels) first associate with the channel in the have distinctive functions and subcellular locations. We found that endoplasmic reticulum, but remain associated at the plasma membrane the UNC-2 CaV2 protein is highly enriched in presynaptic puncta, where they modify channel properties. Their function may be channel where it may provide calcium for exocytosis8–10. Signaling from AWC modulation primarily and channel traffic secondarily. Our results suggest neurons to postsynaptic AIB neurons is reduced but not eliminated that the α2δ subunit UNC-36 acts as an auxiliary subunit that regulates in unc-2 mutants, indicating that UNC-2 cannot be the only source both UNC-2 exit from the endoplasmic reticulum and UNC-2 function of presynaptic calcium in AWC. These observations were made by (Supplementary Fig. 7w). 50
C on l u t c- calf f-1; nc- ca rol 36 -1 u 36 lf (e ; ta nc- (e -1 25 g 36 25 1) -16 (e 1) ; t 8 25 ag ::u 1 -1 nc ) 68 -3 ::c 6 al f-1 ca
un
C on ca tr l u c- calf f-1; nc- ca ol 36 -1 u 36 lf(e ; ta nc (e 1 25 g -3 25 1) -16 6(e 1) ;t 8 2 ag ::u 51 -1 nc ) 68 -3 ::c 6 al f-1
Puncta per 50 µm
Number of clusters
C un on c- tro 2( l lj1 un un c- un c- cal ) 2( c 36 f-1 lj1 -2 ( ca ); u (lj1 e25 ) 1 l ca f-1; nc-3 ; ca ) un c lf-1 un 6( lf-1 e c c- al ; t -3 2 36 f-1 ag 6 51 - (e ) ( ; un e25 tag 168 25 1 ca c-2 1); -16 ::ca ) lf- (lj1 ta 8:: lfun c- ca 1; t ); t g-1 unc 1 un 36 lf-1 ag ag 68 -3 c- (e2 ; t -16 -16 ::ca 6 36 5 ag 8 8 lf (e 1); -1 ::u ::u -1 25 ta 68 nc nc 1) g- ::u -2 -2 ; t 16 n lin ag 8 c- e -1 ::u 2 l 1 68 nc in ::u -2 e 2 nc lin -2 e lin 1 e 2
0
un
© 2009 Nature America, Inc. All rights reserved.
ER
Figure 6 CALF-1 and UNC-36 have related trafficking functions. (a–c) GFPUNC-2 and RAB-3mCherry in AWC neuron of a unc-36(e251) mutant. White arrowhead, AWC cell body; yellow arrowheads, AWC synapses; asterisk, AWB cell body. (d) GFPUNC-2 and RAB-3mCherry in DD neurons of a unc-36(e251) mutant, dorsal nerve cord. (e) GFPUNC-2 and RAB-3mCherry in VD and DD neurons in a unc-36(e251) mutant, ventral nerve cord. (f) GFPUNC-2 and the endoplasmic reticulum (ER) marker CP450mCherry in VD and DD neurons in a unc-36(e251) mutant, ventral nerve cord. Compare Figure 1a–i. (g) UNC-36GFP expressed under the unc-36 promoter, in the head of an adult; image analogous to a–c. Asterisks, diffuse localization of UNC-36GFP in axons at nerve ring. Scale bar, 10 µm. (h) Swimming behavior in M9 buffer. (i) GFPUNC-2 clusters in AWC axons. (j) GFPUNC-2 clusters in 50 µm of dorsal cord covering DD5 and DD6 axons. In h–j, **P < 0.01, different from relevant single-mutant strains by unpaired t-test or Bonferroni t-test, as appropriate; error bars, s.e.m. (k) GFPUNC-2 and RAB-3mCherry in DD neurons, dorsal nerve cord, of unc-36(e251) mutant overexpressing tag-168calf-1.
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1263
a r t ic l e s 30
15 10 5
UV photoconversion
50
g
Heat shock (+)
Heat shock (–)
Heat shock (+)
Heat shock (–)
*
(–)
(+)
Heat shock calf-1; hsp16.2::calf-1
Photoconverted Non-photoconverted
f
± 1 h heat shock 3 h recovery
Dendra2::UNC-2
1264
100
0
(–) (+) Heat shock calf-1; hsp16.2::calf-1
*
In other species also, α 2δ subunits are implicated in CaV2 traffic. Drosophila straightjacket α 2δ mutants have a major defect in synaptic transmission and a minor decrease (25%–40%) in CaV2 channel abundance at the synapse 12,13. Similarly, mammalian α 2δ subunits can affect both CaV2 traffic and function; mammalian α 2δ 1 and α 2δ 2 are the primary targets of the antiepilepsy drug gabapentin, which reduces surface expression of CaV2 channels in cultured neurons and heterologous cells11,39. The precise trafficking step affected by straightjacket and mammalian α2δ subunits has not been defined, but our results indicate that one effect of UNC-36 on UNC-2 occurs during exit from the endoplasmic reticulum. Mammalian and fly α2δ mutants appear to have a milder trafficking defect than unc-36, perhaps because of less redundancy among α 2δ genes: Drosophila has three predicted genes encoding α2δ subunits, and mammals have four, but C. elegans has only two (unc-36 and the uncharacterized gene tag-180). However, the interpretation of our experiments and those in Drosophila is limited by the fact that localization was only examined using overexpressed GFP-tagged CaV2 proteins, which could have different sorting requirements from those of native CaV2 channels. CALF-1 appears to function primarily in UNC-2 biogenesis, not as an auxiliary subunit (Supplementary Fig. 7w). CALF-1 resides in the endoplasmic reticulum, and CALF-1GFP is not detectable at synapses, whereas synaptic GFPUNC-2 is easily detected. Thus if CALF-1 remains associated with UNC-2 at the cell surface, that association is transient or substoichiometric. calf-1 acts rapidly and cell-autonomously to affect UNC-2 localization, apparently by ongoing regulation of UNC-2 exit from the endoplasmic reticulum. This activity is consistent with a role as cargo-specific chaperone, or ‘outfitter’, for UNC-2 (ref. 40). Among the overlapping functions of cargo-specific chaperones are protein folding activities, prevention of aggregation and retrotranslocation of transmembrane proteins from the endoplasmic reticulum into the cytosol, and recruitment of COPII vesicle proteins for endoplasmic reticulum exit41–43. CALF-1 contains multiple Arg-X-Arg motifs, sequences that were first identified for their ability to retain unfolded or partially assembled potassium channels in the endoplasmic reticulum31,32. Cis-acting Arg-X-Arg motifs regulate sorting of Kir and Kv potassium channels, cystic fibrosis–associated CFTR channels,
150
1
e d calf-1; Dendra2::UNC-2; hsp16.2::calf-1 UV photoconversion
Nerve ring puncta
200
lf-
0
ca
GFP::UNC-2 in AWC Swimming behavior
Body bends per 2 min
Recovery, 2 h
20
**
250
1
Heat shock (–) Heat shock (+) 3h 3h
300
lf-
calf-1; hsp16.2::calf-1
c
**
25
ca
b Number of clusters in AWC
a
* Dendra2::UNC-2
H S he tr Remaining perinuclear signal (%) at an H shosge S he tr ck ne at an (– (+) s ) H sho ge S he tr ck ne at an (+ (–) sh sg ) oc en k e( (+ + ) )
© 2009 Nature America, Inc. All rights reserved.
Figure 7 Acute CALF-1 expression transports UNC-2 from the cell body to the synapse. (a–c) Heat shock induction of calf-1 rescues adult calf-1(ky867) mutants. (a) Schematic illustrating heat shock experiment. (b) GFPUNC-2 puncta in AWC axons of heat-shocked and non-heat-shocked calf-1; hsp16.2calf-1 worms. (c) Swimming behavior in M9 buffer. (d) Schematic illustration of Dendra2UNC-2 pulse-chase experiment. (e) Photoconverted Dendra2UNC-2 in tail neurons; converted region circled in purple. Arrowheads, non-photoconverted cells. The central autofluorescence is from the intestine (asterisk). (f) Head region of heat-shocked and non-heat-shocked worms. Arrowheads, trafficked Dendra2UNC-2 puncta at the nerve ring. Asterisks, pharyngeal autofluorescence. (g) Photoconverted Dendra2UNC-2 in the cell body. **P < 0.01, results different from no-heat-shock controls by unpaired t-test. All error bars, s.e.m. Scale bar, 10 µm.
70 60 50 40 30 20 10 0
**
NMDA-type glutamate receptors, cardiac sodium channels and GABAB receptors43. Unlike other endoplasmic reticulum retention motifs such as KDEL, Arg-X-Arg motifs can stimulate endoplasmic reticulum exit in some contexts, particularly when multimerized or when bound by 14-3-3 proteins43. CALF-1 contains many Arg-X-Arg motifs but acts in trans to UNC-2 and not in cis. In its small size, transmembrane structure and proposed function, it resembles the Arg-X-Arg– containing invariant chain that regulates MHC class II traffic through internal membranes44. We suggest that CALF-1 acts at an assembly or endoplasmic reticulum exit checkpoint for UNC-2, perhaps by recruiting coat proteins or releasing UNC-2 from endoplasmic reticulum retention factors (Supplementary Fig. 7w). CALF-1-dependent endoplasmic reticulum exit normally occurs after UNC-2 and UNC-36 interact but can occur under other circumstances when UNC-2 or CALF-1 is overexpressed. Conserved CALF-1 homologs are only recognizable in nematodes, but more distantly related proteins in other species could have analogous activities. For example, the poorly understood γ-subunits of mammalian calcium channels have multiple Arg-X-Arg motifs and multiple prolines in their C-terminal cytoplasmic domains, like CALF-1. Defining the conserved, species-specific and cell type–specific components of presynaptic CaV2 localization is a challenge for further experiments. Methods Methods and any associated references are available in the online version of the paper at http://www.nature.com/natureneuroscience/. Note: Supplementary information is available on the Nature Neuroscience website. Acknowledgments We thank S. Chalasani, T. Maniar and P. McGrath for their insights and advice, A. Bendesky, E. Feinberg, G. Lee, B. Lesch, M. Tsunozaki and L. Winzenread for comments on the manuscript, L. Looger for G-CaMP2.2b, K. Shen for mCherryrab-3, and the Caenorhabditis Genetic Center (CGC) and the National Bioresource Project for strains. This work was supported by the Howard Hughes Medical Institute (C.I.B.), and Y.S. was supported by the Nakajima Foundation. We dedicate this paper to Masanori Obayashi.
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
a r t ic l e s
1. Catterall, W.A. Structure and regulation of voltage-gated Ca2+ channels. Annu. Rev. Cell Dev. Biol. 16, 521–555 (2000). 2. Arikkath, J. & Campbell, K.P. Auxiliary subunits: essential components of the voltage-gated calcium channel complex. Curr. Opin. Neurobiol. 13, 298–307 (2003). 3. Bidaud, I., Mezghrani, A., Swayne, L.A., Monteil, A. & Lory, P. Voltage-gated calcium channels in genetic diseases. Biochim. Biophys. Acta 1763, 1169–1174 (2006). 4. Jeziorski, M.C., Greenberg, R.M. & Anderson, P.A. The molecular biology of invertebrate voltage-gated Ca2+ channels. J. Exp. Biol. 203, 841–856 (2000). 5. Smith, L.A. et al. A Drosophila calcium channel α1 subunit gene maps to a genetic locus associated with behavioral and visual defects. J. Neurosci. 16, 7868–7879 (1996). 6. Brooks, I.M., Felling, R., Kawasaki, F. & Ordway, R.W. Genetic analysis of a synaptic calcium channel in Drosophila: intragenic modifiers of a temperature-sensitive paralytic mutant of cacophony. Genetics 164, 163–171 (2003). 7. Kawasaki, F., Zou, B., Xu, X. & Ordway, R.W. Active zone localization of presynaptic calcium channels encoded by the cacophony locus of Drosophila. J. Neurosci. 24, 282–285 (2004). 8. Schafer, W.R. & Kenyon, C.J. A calcium-channel homologue required for adaptation to dopamine and serotonin in Caenorhabditis elegans. Nature 375, 73–78 (1995). 9. Richmond, J.E., Weimer, R.M. & Jorgensen, E.M. An open form of syntaxin bypasses the requirement for UNC-13 in vesicle priming. Nature 412, 338–341 (2001). 10. Mathews, E.A. et al. Critical residues of the Caenorhabditis elegans unc-2 voltagegated calcium channel that affect behavioral and physiological properties. J. Neurosci. 23, 6537–6545 (2003). 11. Canti, C. et al. The metal-ion-dependent adhesion site in the Von Willebrand factorA domain of α2δ subunits is key to trafficking voltage-gated Ca2+ channels. Proc. Natl. Acad. Sci. USA 102, 11230–11235 (2005). 12. Dickman, D.K., Kurshan, P.T. & Schwarz, T.L. Mutations in a Drosophila α2δ voltagegated calcium channel subunit reveal a crucial synaptic function. J. Neurosci. 28, 31–38 (2008). 13. Ly, C.V., Yao, C.-K., Verstreken, P., Ohyama, T. & Bellen, H.J. straightjacket is required for the synaptic stabilization of cacophony, a voltage-gated calcium channel α1 subunit. J. Cell Biol. 181, 157–170 (2008). 14. Bichet, D. et al. The I–II loop of the Ca2+ channel α1 subunit contains an endoplasmic reticulum retention signal antagonized by the β subunit. Neuron 25, 177–190 (2000). 15. Viard, P. et al. PI3K promotes voltage-dependent calcium channel trafficking to the plasma membrane. Nat. Neurosci. 7, 939–946 (2004). 16. Kittel, R.J. et al. Bruchpilot promotes active zone assembly, Ca2+ channel clustering, and vesicle release. Science 312, 1051–1054 (2006). 17. Long, A.A. et al. Presynaptic calcium channel localization and calcium-dependent synaptic vesicle exocytosis regulated by the Fuseless protein. J. Neurosci. 28, 3668–3682 (2008). 18. Nishimune, H., Sanes, J.R. & Carlson, S.S. A synaptic laminin-calcium channel interaction organizes active zones in motor nerve terminals. Nature 432, 580–587 (2004). 19. Butz, S., Okamoto, M. & Sudhof, T.C. A tripartite protein complex with the potential to couple synaptic vesicle exocytosis to cell adhesion in brain. Cell 94, 773–782 (1998). 20. Lai, M. et al. A tctex1-Ca2+ channel complex for selective surface expression of Ca2+ channels in neurons. Nat. Neurosci. 8, 435–442 (2005).
21. Patel, M.R. et al. Hierarchical assembly of presynaptic components in defined C. elegans synapses. Nat. Neurosci. 9, 1488–1498 (2006). 22. Crump, J.G., Zhen, M., Jin, Y. & Bargmann, C.I. The SAD-1 kinase regulates presynaptic vesicle clustering and axon termination. Neuron 29, 115–129 (2001). 23. Zhen, M., Huang, X., Bamber, B. & Jin, Y. Regulation of presynaptic terminal organization by C. elegans RPM-1, a putative guanine nucleotide exchanger with a RING-H2 finger domain. Neuron 26, 331–343 (2000). 24. Hallam, S.J., Goncharov, A., McEwen, J., Baran, R. & Jin, Y. SYD-1, a presynaptic protein with PDZ, C2 and rhoGAP-like domains, specifies axon identity in C. elegans. Nat. Neurosci. 5, 1137–1146 (2002). 25. Zhen, M. & Jin, Y. The liprin protein SYD-2 regulates the differentiation of presynaptic termini in C. elegans. Nature 401, 371–375 (1999). 26. Hall, D.H. & Hedgecock, E.M. Kinesin-related gene unc-104 is required for axonal transport of synaptic vesicles in C. elegans. Cell 65, 837–847 (1991). 27. Frokjaer-Jensen, C. et al. Effects of voltage-gated calcium channel subunit genes on calcium influx in cultured C. elegans mechanosensory neurons. J. Neurobiol. 66, 1125–1139 (2006). 28. Bauer Huang, S.L. et al. Left-right olfactory asymmetry results from antagonistic functions of voltage-activated calcium channels and the Raw repeat protein OLRN-1 in C. elegans. Neural Dev 2, 24 (2007). 29. Rolls, M.M., Hall, D.H., Victor, M., Stelzer, E.H. & Rapoport, T.A. Targeting of rough endoplasmic reticulum membrane proteins and ribosomes in invertebrate neurons. Mol. Biol. Cell 13, 1778–1791 (2002). 30. Chalasani, S.H. et al. Dissecting a circuit for olfactory behaviour in Caenorhabditis elegans. Nature 450, 63–70 (2007). 31. Zerangue, N., Schwappach, B., Jan, Y.N. & Jan, L.Y. A new ER trafficking signal regulates the subunit stoichiometry of plasma membrane K(ATP) channels. Neuron 22, 537–548 (1999). 32. Schwappach, B., Zerangue, N., Jan, Y.N. & Jan, L.Y. Molecular basis for K(ATP) assembly: transmembrane interactions mediate association of a K+ channel with an ABC transporter. Neuron 26, 155–167 (2000). 33. Troemel, E.R., Sagasti, A. & Bargmann, C.I. Lateral signaling mediated by axon contact and calcium entry regulates asymmetric odorant receptor expression in C. elegans. Cell 99, 387–398 (1999). 34. Mello, C. & Fire, A. DNA transformation. Methods Cell Biol. 48, 451–482 (1995). 35. Chudakov, D.M., Lukyanov, S. & Lukyanov, K.A. Tracking intracellular protein movements using photoswitchable fluorescent proteins PS-CFP2 and Dendra2. Nat. Protoc. 2, 2024–2032 (2007). 36. Yeh, E. et al. A putative cation channel, NCA-1, and a novel protein, UNC-80, transmit neuronal activity in C. elegans. PLoS Biol. 6, e55 (2008). 37. Grunwald, M.E., Mellem, J.E., Strutz, N., Maricq, A.V. & Kaplan, J.M. Clathrinmediated endocytosis is required for compensatory regulation of GLR-1 glutamate receptors after activity blockade. Proc. Natl. Acad. Sci. USA 101, 3190–3195 (2004). 38. Schwappach, B. An overview of trafficking and assembly of neurotransmitter receptors and ion channels. Mol. Membr. Biol. 25, 270–278 (2008). 39. Hendrich, J. et al. Pharmacological disruption of calcium channel trafficking by the α2δ ligand gabapentin. Proc. Natl. Acad. Sci. USA 105, 3628–3633 (2008). 40. Herrmann, J.M., Malkus, P. & Schekman, R. Out of the ER–outfitters, escorts and guides. Trends Cell Biol. 9, 5–7 (1999). 41. Fromme, J.C., Orci, L. & Schekman, R. Coordination of COPII vesicle trafficking by Sec23. Trends Cell Biol. 18, 330–336 (2008). 42. Kota, J. & Ljungdahl, P.O. Specialized membrane-localized chaperones prevent aggregation of polytopic proteins in the ER. J. Cell Biol. 168, 79–88 (2005). 43. Michelsen, K., Yuan, H. & Schwappach, B. Hide and run. Arginine-based endoplasmic-reticulum-sorting motifs in the assembly of heteromultimeric membrane proteins. EMBO Rep. 6, 717–722 (2005). 44. Schutze, M.P., Peterson, P.A. & Jackson, M.R. An N-terminal double-arginine motif maintains type II membrane proteins in the endoplasmic reticulum. EMBO J. 13, 1696–1705 (1994).
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1265
AUTHOR CONTRIBUTIONS Y.S. and C.I.B. designed the project, and Y.S. conducted the experiments. Y.S. and C.I.B. wrote the paper.
© 2009 Nature America, Inc. All rights reserved.
Published online at http://www.nature.com/natureneuroscience/. Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/.
© 2009 Nature America, Inc. All rights reserved.
ONLINE METHODS Strains. Wild-type worms were Bristol variety N2. Strains were maintained using standard methods at 21–23 °C. Some strains were provided by the Caenorhabditis Genetics Center and the National Bioresource Project. Mutants used were calf-1 (ky867), unc-2(lj1), unc-36(e251), syd-2(ju37), sad-1(ky289), rpm-1(js410), syd-1 (ju82), elks-1(tm1233), nrx-1(ds1), lin-2(n1610), lin-7(n308cs), lin-10(e1439), unc-13(e51), unc-10(e102), dpy-23(e840), unc-101(m1), unc-11(ky280), unc-31 (e928), unc-104(e1265) and unc-116(e2310). Germline transformation was carried out as described34. odr-3GFPunc-2, tag-168Dendra2unc-2 and unc-25GFPunc-2 were injected at 100 ng µl–1. For rescue, overexpression and structure-function analysis, all calf-1 plasmids were injected at 20 ng µl–1 except hsp16.2calf-1 at 10 ng µl–1. Relatively high levels of GFPUNC-2 and Dendra2UNC-2 were needed for reliable visualization, and these overexpressed, tagged proteins might distort endogenous traffic. However, GFPUNC-2 and Dendra2UNC-2 were able to rescue unc-2dependent locomotion, and they were reliably trafficked to synapses in heatshocked calf-1 worms carrying hsp16.2calf-1, indicating that the proteins in transgenic worms can interact effectively with the trafficking machinery. For localization experiments, calf-1GFP fusion plasmids were injected at 10 ng µl–1. unc-36GFP fusion plasmids were injected at 50 ng µl–1. All cell compartment markers, including odr-3mCherryrab-3 and odr-3CP450mCherry, were injected at 0.5 to 5 ng µl–1. ofm-1GFP, ofm-1DsRed, elt-2mCherry, odr-1mCherry, odr-1DsRed, odr-1GFP and flp-17mCherry were used as coinjection marker and injected at 6–20 ng µl–1. A complete list of transgenes and strains is included in Supplementary Methods. Isolation and characterization of calf-1(ky867). A strain expressing GFPUNC-2 in AWC (kyIs442) was mutagenized with ethylmethane sulfonate according to standard protocols45. We cloned 209 F1 worms into separate plates and screened 30 to 50 F2 progeny from individual F1 parents visually under a compound microscope. The mutants were chosen on the basis of the loss of GFPUNC-2 puncta from the AWC axon as observed with a Plan-Apochromat ×63 objective on a Zeiss Axioplan2 microscope. Mapping and cloning of calf-1. calf-1(ky867) was mapped to the far right end of LGV using single nucleotide polymorphisms in the CB4856 strain46. A genomic fragment containing only the B0250.2 reading frame with 0.8 kb of 5′ sequence and 1.2 kb of 3′ sequence was generated by PCR and injected at 1 ng µl–1 into calf-1(ky867) mutants. The PCR fragment rescued both uncoordinated movements and GFPUNC-2 localization in AWC axons. To identify the calf-1 mutation, the calf-1 genomic coding region in ky867 was amplified by PCR and PCR products were sequenced. Molecular biology. Standard molecular biology techniques were used. Details of plasmid construction and primers are in Supplementary Methods. Fluorescence microscopy and quantification. Worms were mounted on 4% agarose pads containing 400 µM tetramisole. We examined multiple transgenic lines of each transgene for fluorescent expression and localization patterns. Widefield fluorescence images (Fig. 1a–f,h,i,k–m, Fig. 2a–f,i,j, Fig. 4c,d, Fig. 6a–f,k, Fig. 7e,f, Supplementary Fig. 1a–f, Supplementary Fig. 2a–r and Supplementary Fig. 3a–i) were obtained on a Zeiss Axioplan2 imaging system. Confocal images (Fig. 4e–g,j–r, Fig. 5b bottom panels, Fig. 6g, Supplementary Fig. 5a–r and Supplementary Fig. 7a–c,e–v) were obtained on Zeiss LSM 510 META laser scanning confocal imaging system. To quantify fluorescence intensities and number of fluorescent clusters, we captured images under consistent detector settings with a Hamamatsu Photonics C2400 CCD camera on a Zeiss Axioplan2 Imaging System with a ×63 Plan-Apochromat objective and MetaMorph software. ImageJ (US National Institutes of Health) was used to quantify fluorescence in AWC axons and cell bodies and DD dorsal axons. Images of AWC nerve rings and cell bodies were projected into a single plane by maximum projection; for DD, we chose a single image of best focus for the quantification. Background intensity was subtracted and fluorescent clusters containing signals above an arbitrary threshold were measured for the total fluorescence intensity and the number of fluorescent
nature NEUROSCIENCE
clusters. We used the same thresholds were used for all images in each quantification. Normalized fluorescence intensity was obtained by dividing individual values with mean total fluorescence intensity of wild-type control worms. For the perinuclear region of AWC, a single image of best focus was chosen for the quantification and maximum fluorescence intensity was measured after background subtraction (Supplementary Fig. 7d). We scored 6–10 worms for each experiment. Calcium imaging. We performed calcium imaging as described30. For AWC imaging, the strain CX10536, expressing the calcium indicator G-CaMP2.2b47 in one of the two AWC neurons, AWCON, under the str-2 promoter, was crossed to unc-2(lj1) and calf-1(ky867) to generate the strains CX11391 and CX11386. For AIB imaging, the strain CX7469, expressing G-CaMP1.0 in AIB neurons30, was crossed with unc-2(lj1) and calf-1(ky867) to generate CX11394 and CX11383. Worms were washed in buffer without food for ~20 min before imaging, a protocol designed to mimic the washes before chemotaxis assays, and imaging was conducted in a polydimethylsiloxane chamber in which a worm’s nose was exposed to a stream of buffer that could be switched between odor-containing and odor-free solutions using an electronically gated valve. The standard stimulus protocol consisted of a 5-min step pulse of the 10−4 dilution of odor in S-basal (without cholesterol) followed by odor removal. G-CaMP fluorescence intensity was measured for 10 s before and 50 s after the onset or offset of the odor stimulus; the same worms were imaged for odor onset and offset. The 100% values were set by taking the average response from 1–4 s in the trace. Controls and mutants were interleaved during imaging. Subcellular localization in neurons and intestinal epithelial cells. For endoplasmic reticulum markers, we obtained cDNAs of C. elegans homologs of mammalian cytochrome P450, RAMP4 and cytochrome b5 (abbreviated as CP450 and cb5), using primers flanking the open reading frame C49C8.4, F59F4.2 and C31E10.7 (ref. 29). For the Golgi marker, a cDNA fragment of C. elegans αmannosidase II (ManII; the first 82 amino acids, including signal sequence and transmembrane-anchor domain) was amplified from F58H1.1 (refs. 29,48). cDNAs were fused to mCherry at their C termini. We tested individual regions of CALF-1 for endoplasmic reticulum localization by expressing GFP-tagged CALF-1 mutants in intestinal epithelial cells, which are larger than neurons and easier to examine for subcellular protein localization. Full-length GFP-tagged CALF-1 colocalized with the endoplasmic reticulum marker CP450mCherry in intestinal epithelial cells. Heat shock experiments. Experiments with hsp16.2calf-1 were performed on young adult hermaphrodites. A 30 °C heat shock was given for 3 h. The plates were then incubated at 20 °C for 2 h for recovery before scoring GFPUNC-2 localization and swimming behavior. Photoconversion experiments. Dendra2UNC-2 was expressed under tag-168 pan-neuronal promoter in a calf-1(ky867); unc-2(lj1) double-mutant background; transgenic worms carrying hsp16.2calf-1 were crossed into the Dendra2UNC-2–expressing worms. Before photoconversion, we mounted L4 larvae expressing Dendra2UNC-2 on an agar pad. The tail regions of individual larvae were illuminated with ultraviolet light with a ×63 PlanApochromat objective to photoconvert the Dendra2 locally. Worms were moved to agar plates with E. coli food, and a 30 °C heat shock was given for 1 h. The plates were then incubated at 20 °C for 3 h before scoring Dendra2UNC-2 localization in the head region. The maximum fluorescence intensity of the photoconverted red Dendra2UNC-2 was quantified at the perinuclear region of a single image of best focus after background fluorescence intensity subtraction. Individual worms were mounted on agar pads in the same orientation before and after heat shock to allow the comparison of the same tail neurons. Swimming assay. We transferred individual young adult worms into a drop of M9 buffer (22 mM KH2PO4, 22 mM Na2HPO4, 85 mM NaCl, 1 mM MgSO4) on top of an agar plate and, after a 30-s recovery period, counted body bends for 2 min.
doi:10.1038/nn.2383
45. Anderson, P. Mutagenesis. Methods Cell Biol. 48, 31–58 (1995). 46. Wicks, S.R., Yeh, R.T., Gish, W.R., Waterston, R.H. & Plasterk, R.H. Rapid gene mapping in Caenorhabditis elegans using a high density polymorphism map. Nat. Genet. 28, 160–164 (2001). 47. Tallini, Y.N. et al. Imaging cellular signals in the heart in vivo: cardiac expression of the high-signal Ca2+ indicator GCaMP2. Proc. Natl. Acad. Sci. USA 103, 4753–4758 (2006). 48. Chen, C.C. et al. RAB-10 is required for endocytic recycling in the Caenorhabditis elegans intestine. Mol. Biol. Cell 17, 1286–1297 (2006).
© 2009 Nature America, Inc. All rights reserved.
Statistical analysis. For fluorescence images and swimming assay, statistical analysis was performed using Student’s unpaired t-test, Bonferroni t-test or Dunnett’s test as appropriate. For AIB calcium imaging experiments, responses with an average value of zero were counted as failures, and the fraction of failures was compared for each genotype by chi-squared test (a nonparametric method appropriate for non-normally distributed data). Consistent results were obtained in two independent blocks of experiments.
doi:10.1038/nn.2383
nature NEUROSCIENCE
a r t ic l e s
EFHC1 interacts with microtubules to regulate cell division and cortical development
© 2009 Nature America, Inc. All rights reserved.
Laurence de Nijs1, Christine Léon1, Laurent Nguyen1, Joseph J LoTurco2, Antonio V Delgado-Escueta3, Thierry Grisar1 & Bernard Lakaye1 Mutations in the EFHC1 gene are linked to juvenile myoclonic epilepsy (JME), one of the most frequent forms of idiopathic generalized epilepsies. JME is associated with subtle alterations of cortical and subcortical architecture, but the underlying pathological mechanism remains unknown. We found that EFHC1 is a microtubule-associated protein involved in the regulation of cell division. In vitro, EFHC1 loss of function disrupted mitotic spindle organization, impaired M phase progression, induced microtubule bundling and increased apoptosis. EFHC1 impairment in the rat developing neocortex by ex vivo and in utero electroporation caused a marked disruption of radial migration. We found that this effect was a result of cortical progenitors failing to exit the cell cycle and defects in the radial glia scaffold organization and in the locomotion of postmitotic neurons. Therefore, we propose that EFHC1 is a regulator of cell division and neuronal migration during cortical development and that disruption of its functions leads to JME. JME is the most common form of idiopathic generalized epilepsy, accounting for 10–30% of all epilepsies on the basis of hospital reports1. JME symptoms appear at the onset of adolescence and include myoclonic jerks, tonic-clonic seizures and occasionally absence seizures2. Postmortem analysis of individuals with JME reveals cortical and subcortical dystopic neurons and subtle abnormalities of intracortical architecture, referred to as microdysgenesis3. Studies with quantitative, high-resolution, voxelbased magnetic resonance imaging have found that there is an increase in the thickness of cortical gray matter in some individuals with JME4. Genetic inheritance of JME is complex and 15 chromosomal loci are linked to the disease5. In particular, mutations have been identified in GABRA1 (ref. 6) and CLCN2 (ref. 7), two genes encoding ion channels, and in EFHC1 (ref. 8), a gene encoding a protein containing three DM10 domains of unknown function and a single EF-hand motif, a Ca2+-binding domain. To date, EFHC1 mutations are the most frequent cause of JME, accounting for 9% of the cases5. Molecular analyses have identified several missense mutations in the coding sequence of EFHC1 that cosegregate with the phenotype of JME8–12. Overexpression of EFHC1 induces neuronal apoptosis in vitro by increasing R-type voltage-dependent Ca2+ channel currents, which are reduced by mutated EFHC1 (ref. 8). Therefore, the reduction of apoptosis by EFHC1 mutations may increase neuron density, with precarious Ca2+ homeostasis leading to seizure susceptibility. On the other hand, the mouse ortholog of EFHC1 is an axonemal protein expressed in tissues that possess motile cilia and flagella, such as lung, testis or ependyma13,14, suggesting that EFHC1 is involved in ciliary functions15. In a previous study, we found that EFHC1 associates with the centrosome and the mitotic spindle in cultured cells through its N-terminal region16. Moreover, expression of EFHC1 transcript and
protein in mouse brain is highest during embryogenesis, when cell division and migration are prominent8,17. These results strongly suggest that EFHC1 is involved in cell division and brain development. The development of the cerebral cortex is a very complex pro cess that follows strictly regulated and tightly linked sequences of proliferation, cell cycle exit, cell migration to specific cell layers and neuronal differentiation. Projection neurons are generated in the proliferative ventricular zone/subventricular zone (SVZ), where cortical progenitor cells undergo rapid proliferative divisions to expand the progenitor pool before they exit the cell cycle. Newly born neurons leave the ventricular zone/SVZ and migrate through the intermediate zone toward the cortical plate to reach their final position, where they actively extend axonal and dendritic branches18. Notably, these current steps imply cell shape remodeling, which largely depends on dynamic rearrangements of the cytoskeleton. Several microtubule-associated proteins (MAPs) and proteins localized at the centrosomes are direct regulators of the microtubule dynamics underlying these morphological changes and are essential for cortical development19–21. In this study, we found that EFHC1 is a MAP that directly interacts with α-tubulin through a new type of microtubule-binding domain (MTBD) located at its N terminus. We found that EFHC1 loss of function induced mitotic spindle defects, disruption of M phase progression, microtubule bundling and increased apoptosis. Furthermore, EFHC1 impairment in the rat developing neocortex by ex vivo and in utero electroporation disrupted radial migration of projection neurons by affecting the mitosis and the cell cycle exit of cortical progenitors, the radial glia scaffold organization and the locomotion of postmitotic neurons. Thus, we identified EFHC1 as a new MAP that appears to be essential for normal assembly and function of
1GIGA-Neurosciences,
University of Liège, Liège, Belgium. 2Department of Physiology and Neurobiology, University of Connecticut, Storrs, Connecticut, USA. 3David Geffen School of Medicine at University of California Los Angeles, Epilepsy Genetics/Genomics Laboratories, Veterans Affairs Greater Los Angeles Healthcare System, West Los Angeles, California, USA. Correspondence should be addressed to T.G. (
[email protected]). Received 28 May; accepted 4 August; published online 6 September 2009; doi:10.1038/nn.2390
1266
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
a r t ic l e s
1–92
DM10 (1)
TAG
NH
1–92
COOH
TAG
NH
1–60
TAG
NH
1–45
TAG
NH
1–40
TAG
NH 1–35
COOH
TAG
NH 1–30
COOH
EFH
239–359
416–520 578–606 TAG-hDM101
COOH
TAG-hN92
COOH
TAG-hN60
COOH
DM10 (1)
Merge
TAG-hN45
COOH
NH
α-tubulin/DNA
EGFP
COOH TAG-hEFHC1
TAG-hN40 TAG-hN35 TAG-hN30 DM10 (2)
DM10 (3)
EFH
COOH TAG-hDM110
mitotic spindle and for neuronal migration. This mechanism would imply that EFHC1 is important in cerebral cortex development. RESULTS EFHC1 is a MAP We have previously shown that the N-terminal portion (residues 1 to 92) of EFHC1 is necessary for its colocalization with the centrosome and the mitotic spindle16. To determine more precisely which region of EFHC1 mediates this association, we expressed various enhanced green fluorescent protein (EGFP)-tagged truncated EFHC1 proteins in HEK293 cells (Fig. 1a) and carried out immunocytochemistry with an antibody to α-tubulin. We found that EGFP-hN45, but not EGFPhN30, colocalized with the mitotic spindle (Fig. 1b), suggesting that the first 45-amino-acid region contains a motif that is required for the localization of EFHC1 with the mitotic spindle. This colocalization prompted us to investigate whether EFHC1 interacts with α- and/or γ-tubulin. Using immunoprecipitation procedures on HEK293 cell lysates, we found that EFHC1 and α-tubulin coprecipitated mutually (Fig. 2a,b) and that EFHC1 did not interact with γ-tubulin (Fig. 2c). To determine whether EFHC1 interacts directly with α-tubulin, we used in vitro cosedimentation assays using pure prepolymerized microtubules and different purified GST-tagged truncated EFHC1 proteins (Fig. 1a). We found that GST-hEFHC1 and GST-hN45, but not GST-hN30 and GST, cosedimented with microtubules, indicating that there was a direct interaction between
a
EFHC1 is required for mitotic spindle organization In addition to its association with α-tubulin, we found that overexpression of EGFP-hN92, EGFP-hN60 or EGFP-hN45 in HEK293 cells resulted in severe mitotic spindle defects, including m onopolar spindle and chromosome alignment failure during metaphase (Fig. 3a). We quantified their occurrence in cultures overexpressing different EGFP-tagged EFHC1 proteins (Fig. 1a) and found that the percentage of mitotic defects was significantly higher (P < 0.001) for EGFP-hN92, EGFP-hN60 and EGFP-hN45 (73.7 ± 5.5%, 75 ± 1.7% and 92.1 ± 1.7%, respectively) compared with EGFP-hEFHC1 (23.7 ± 4.1%) or EGFP (22.7 ± 5.2%; Fig. 3b).
b
IP: c-Myc c-Myc– hEFHC1 L IP
Control
L IP
L IP
WB
α-tubulin β-actin
c
IP: α-tubulin
55 42
L IP
GST-hEFHC1 +MT –MT S P S P
GST-hEFHC1
GST-hN45 +MT S P
–MT S P
α-tubulin
GST-hN30 +MT S P
–MT S P
55
kDa γ-tubulin β-actin
48 42
c-Myc– hEFHC1
70
e
GST +MT S P
–MT S P kDa 100
Tubulin
55
GST-hN45 GST-hN30 GST
35 33 30
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
L IP
30
70
WB
d
L IP kDa 100
EGFPhEFHC1
EGFP c-Myc– hEFHC1
IP: c-Myc c-Myc– hEFHC1 Control
EGFPhEFHC1 EGFP kDa
WB: c-Myc
Figure 2 EFHC1 is a MAP. (a,c) Immunoprecipitations with antibody to c-Myc in HEK293 cell lysates transfected with c-Myc–hEFHC1 or empty pcDNA5 vector (control). Western blots (WB) of immunoprecipitates (IP) and cell lysates (L) were probed with antibodies to α-tubulin (a, upper panel), γ-tubulin (c, upper panel), c-Myc (a,c, lower panel) or β-actin (a,c, upper panel). (b) Immunoprecipitations with antibody to α-tubulin in HEK293 cell lysates expressing EGFP-hEFHC1 or EGFP. Western blot on immunoprecipitates and cell lysates were probed with antibodies to GFP (upper panel) and α-tubulin (lower panel). (d) Microtubule (MT) cosedimentation assays of GST-hEFHC1, GST-hN45, GST-hN30 or GST with (+MT) or without (−MT) pure prepolymerized microtubules. Supernatants (S) and pellets (P) were analyzed on a 10% SDS-PAGE and visualized by Coomassie blue staining. (e) Affinity of EFHC1 for microtubules was determined by plotting the percentage of GST-hEFHC1 bound to microtubules versus the tubulin dimer concentration.
EFHC1 and α-tubulin through the first 45amino-acid region (Fig. 2d). This association was not affected by the presence of Ca2+ (data not shown). We therefore estimated the affinity of EFHC1 for microtubules and found an equilibrium dissociation constant of about 1.5 µM, indicating that EFHC1 binds to microtubules with a relatively high affinity (Fig. 2e). Finally, we carried out in vitro tubulin polymerization assays with GST-hEFHC1. We did not detect any substantial change in either the polymerization rate or the nucleation phase in the presence of GST-hEFHC1 (Supplementary Fig. 1). This indicates that EFHC1 is not required for the polymerization of microtubules in vitro. It is noteworthy that the region involved in the binding to α-tubulin (that is, the first 45 amino acids) showed no obvious homology with the MTBDs of other known MAPs. Thus, EFHC1 would appear to be a unique MAP, unrelated to conventional MAPs.
WB
93–198
NH
DM10 (3)
WB: c-Myc
1–92 TAG
DM10 (2)
WB: EGFP
DM10 (1)
EGFP-hN30
1–92
EGFP-hN45
NH
TAG
© 2009 Nature America, Inc. All rights reserved.
Figure 1 EFHC1 associates with the mitotic spindle through a sequence located in the first 45-amino-acid region. (a) Schematic representation of the different tagged forms of EFHC1 used in this study. TAG represents EGFP, c-Myc or GST. (b) HEK293 cells expressing EGFP-hN30 or EGFP-hN45 (green) stained for α-tubulin (red) and DNA (blue). Arrows indicate association with the mitotic spindle. Scale bar represents 20 µm.
b
TAG
EFHC1 bound to MTs (%)
a
100 75 50 25 0
0 1 2 3 4 5 6 7 8 9 [Tubulin dimer] (µM)
1267
a r t ic l e s α-tubulin/DNA
Merge
b Abnormal spindles (%)
EGFP
100 90 80 70 60 50 40 30 20 10 0
*** *** *** ***
RT-PCR kDa
EGFPhEFHC1
***
Mitotic index (%)
15
Merge
Control shRNA
α-tubulin/DNA
400 200
GAPDH 42 EFHC1
f EGFP
bp
100
β-actin
***
10 5
EG
FP EG -h FP E EG FH FP C1 C on -h N t 45 hE rol FH sh hE C RN FH 1 s A C hR 1 sh NA rE RN FH A C + 1
0
** *
60 50
EGFP EGFP-hEFHC1 EGFP-hN45 Control shRNA hEFHC1 shRNA
** *
40 30
EGFP-hEFHC1 by 90%, whereas β-actin levels remained unchanged (Fig. 3d). Similar knockdown of endogenous EFHC1 mRNA * * * * * * * levels was observed, as demonstrated by semiquantitative reverse transcription PCR (RT-PCR) (Fig. 3d). We transfected HEK293 cells with these shRNAs and observed a significant increase (P < 0.001) in the number of disorganized mitotic spindles with the presence of monopolar spindles and misaligned metaphase compared with controls (49.3 ± 3.1% versus 18.3 ± 2.9%; Fig. 3b,e). Moreover, expression of rEFHC1, an shRNA-resistant form, rescued the mitotic spindle defects, demonstrating the specificity of the shRNA effect (Fig. 3b,e). Finally, we found that the mitotic index was significantly higher (P < 0.001) in cultures expressing EGFP-hN45 (13.9 ± 1.1%) and hEFHC1 shRNA (12.1 ± 1.2%) than in control cultures (EGFP, 9.7 ± 0.9%; control shRNA, 9.3 ± 0.7%; Fig. 3f). In contrast with control cells that were distributed across all mitotic phases, EGFP-hN45 and hEFHC1 shRNA–expressing cells were mostly arrested at prometaphase, with a concomitant decrease in the number of cells in anaphase/ telophase and cytokinesis (Fig. 3g), resulting in subsequent M phase delay or arrest. Collectively, these results obtained by independent approaches (RNAi and dominant-negative protein) strongly suggest that EFHC1 is involved in mitotic microtubules organization and M phase progression. is
e
es
in
ok
ph
ha
se
/te
lo
ap et m
ap
ha
d ne lig -a is
as
se
se ha
se
ap
ha ap
et M
et
ap m ro rp la po ni U
Bi
po
la
rp
ro
m
et
Pr
op
ha
ha
se
se
0
yt
** * * **
** * * *
An
10
C
20
Figure 3 EFHC1 is necessary for mitotic spindle organization and M phase progression. (a) HEK293 cells overexpressing EGFP-hN45 or EGFP (green) stained for α-tubulin (red) and DNA (blue). A normal metaphase cell is shown in the EGFP panels. The EGFP-hN45 panels show metaphase cells with misaligned chromosomes (top) or monopolar spindle (bottom). (b) Quantification of abnormal spindles. (c) HEK293 cells overexpressing EGFP-hN45 (green) stained for γ-tubulin (red) and DNA (blue). Insets show a higher magnification of centrosomes. (d) Western blot analysis with antibody to GFP of stably transfected HEK Flp-In T-Rex 293 cell extracts overexpressing EGFP-hEFHC1 72 h after transfection with control shRNA or two hEFHC1 shRNAs. β-actin was used for normalization (left). Semiquantitative RT-PCR 72 h after administration of two hEFHC1 shRNAs in HEK293 cells is shown on the right. GAPDH was used for normalization (right). (e) HEK293 cells transfected with control shRNA, hEFHC1 shRNA and hEFHC1 shRNA + rEFHC1 (green) stained for α-tubulin (red) and DNA (blue). hEFHC1 shRNA panels show monopolar spindle (top) or metaphase cells with misaligned chromosomes (bottom). (f) Quantification of mitotic index. (g) Quantification of mitotic stages on HEK293 cells. Quantifications were performed 48 h after transfections. Data are means from three independent experiments. Error bars show s.e.m. * P < 0.05, ** P < 0.01 and *** P < 0.001. Scale bars represent 20 µm.
** *
** *
M
Mitotic cells (%)
hEFHC1 shRNA
g
hEFHC1 shRNA + rEFHC1
© 2009 Nature America, Inc. All rights reserved.
e
WB
on t hE rol FH shR C C1 NA on tr sh hE ol s RN F H h R A# C NA 1 1 sh R N A# 2
Merge
C
γ-tubulin/DNA
C
EGFP
EGFP-hN45
c
on
d
t hE rol FH shR C C1 NA on tr sh hE ol s RN F H h R A# C NA 1 1 sh R N A# 2
EG
EG
FP -h FP E EG F H FP C1 EG -hN FP 30 EG -h N FP 45 EG -hN EG F P 6 0 FP -hN EG -hD 92 F P M1 C -hD 01 on M hE trol 110 FH sh R hE C1 NA FH sh C RN 1 + sh A rE R FH N C A 1
EGFP-hN45
EGFP
a
Notably, truncated forms that did not associate with the mitotic spindle (EGFP-hN30 and EGFP-hDM110, ref. 16) did not promote spindle defects, suggesting that binding to microtubules is necessary for this phenotype. EGFP-hDM101, which contains both the N-terminal region and the first DM10 domain, failed to induce mitotic spindle defects. This suggests that at least the first DM10 domain is required for proper EFHC1 function at the mitotic spindle. Moreover, we observed that monopolar spindles showed an abnormal spherical distribution of chromosomes around a pair of closely and centrally located centrosomes, indicating that the centrosome was duplicated, but that there was an insufficient separation to form a bipolar spindle (Fig. 3c). In the next experiments, we only used the EGFP-hN45 protein, as it produced, in our opinion, the most important phenotype. Using an antibody to the C terminus of EFHC1, we observed that EGFP-hN45 saturated the microtubule’s EFHC1 binding sites and, as such, acted as a dominant-negative protein (Supplementary Fig. 2). We next carried out RNA interference (RNAi)-mediated gene silencing in HEK293 cells. We validated the knockdown efficiency of two small hairpin RNAs (shRNAs) directed to distinct sites of the EFHC1 transcript via western blot. Both of the EFHC1 shRNAs, but not a nontargeting control shRNA, silenced the expression of
EFHC1 impairment induces microtubule bundling and apoptosis During interphase, EGFP-hN45 and hEFHC1 shRNA expression markedly induced microtubule bundling around the cell periphery (Fig. 4a). These bundled microtubules were extremely stable and
1268
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
a r t ic l e s
© 2009 Nature America, Inc. All rights reserved.
b
EGFP
EGFP-hEFHC1
EGFP-hN45
c
d 0.25
0.15 0.10 0.05 0.00
***
***
*
5 0
f +
25
10
A 350 nm
***
20 15
GST GST-hEFHC1 GST-hN45
0.20
***
EG FP E -h GF EG EF P C FP HC on - 1 hE tro hN FH l s 45 hE C1 hRN FH s A C hRN 1 + sh A rE R FH N C A 1
e
Apoptosis (%)
Figure 4 EFHC1 loss of function induces microtubule bundling and apoptosis. (a) HEK293 cells expressing EGFP, EGFP-hN45, control shRNA or hEFHC1 shRNA (green) stained for α-tubulin (red) and DNA (blue). EGFP and control shRNA panels show normal microtubule network in interphase. EGFP-hN45 and hEFHC1 shRNA panels show microtubule bundling around the cell periphery during interphase. (b) HEK293 cells expressing EGFP, EGFPhEFHC1 or EGFP-hN45 (green) stained for α-tubulin (red) and DNA (blue). Cells were treated with nocodazole (10 µM) for 3 h before fixation. (c) Quantification of microtubule bundling. (d) Microtubule bundling assay in vitro with GST, GST-hEFHC1 or GST-hN45. (e) Quantification of apoptosis by TUNEL assays. (f) Analysis of cell cycle phase distribution of apoptotic EGFP and EGFP-hN45 cells. Quantifications were performed 48 h after transfections. Data are means from three independent experiments; error bars show s.e.m. * P < 0.05 and *** P < 0.001. Scale bars represent 20 µm.
did not depolymerize in the presence of nocodazole (Fig. 4b). Microtubule bundling was significantly higher (P < 0.001) in cultures expressing EGFP-hN45 (28.4 ± 1.9%) and hEFHC1 shRNA (21.5 ± 1.7%) than in those expressing EGFP-hEFHC1 (10.1 ± 1.8%), EGFP (6.5 ± 1.4%) or control shRNA (8.9 ± 0.8%; Fig. 4c). Moreover, using in vitro microtubule bundling assays, we found that GST-hN45, but not GST-hEFHC1 or GST, induced microtubule bundling, indicating that the MTBD of EFHC1 is capable of promoting microtubule bundling formation in vivo and in vitro when expressed alone (Fig. 4d). Moreover, in cultures with impaired EFHC1 function, we observed frequent occurrences of nuclear abnormalities such as fragmented and picnotic nuclei, as seen in apoptotic cells (Fig. 4b). Therefore, we carried out TUNEL assays and found a significant increase (P < 0.001) of apoptosis in cells overexpressing EGFP-hEFHC1 (10.1 ± 1.1%), EGFP-hN45 (21.3 ± 1.9%) and hEFHC1 shRNA (21.4 ± 0.9%) compared with the EGFP control (5.5 ± 1.3%; Fig. 4e). Finally, to determine the cell cycle phase distribution of apoptotic cells expressing EGFP or EGFP-hN45, we sorted apoptotic EGFP-positive cells by flow cytometry and carried out cell cycle analysis (Fig. 4f). We observed a significant enhancement (P < 0.05) of EGFP-hN45 cells in S (35.1 ± 0.2%) and G2/M phases (12.2 ± 0.5%), with a concomitant decrease of cells in the G0/G1 phases (52.6 ± 0.3%) compared with EGFP controls (G0/G1, 61.5 ± 1.3%; S, 31.1 ± 0.8%; G2/M, 7.4 ± 0.6%), indicating that mitotic spindle defects induced apoptosis.
35 30 25 20 15 10 5 0
Apoptotic EGFP cells (%)
MT bundling (%)
EGFP
DMSO Nocodazole DMSO Nocodazole DMSO Nocodazole
α-tubulin DNA
hEFHC1 shRNA
EG FP E -h GF EG EF P H C FP C1 on -h hE tro N4 FH l s 5 h hE C1 RN F H sh A C RN 1 + sh A rE R FH N C A 1
Merge
α-tubulin DNA
EGFP
Control EGFP EGFP-hN45 shRNA
Merge
a
0
10 15 20 Time (min)
5
25
30
70 60 50
EGFP EGFP-hN45
*
40
*
30 20
*
10 0
G0/G1
S
G2/M
EFHC1 is required for radial migration Based on previous studies reporting a higher expression of EFHC1 during mouse brain development compared with adult8,17, and on the importance of EFHC1 in cell division, we further investigated the role of EFHC1 on cerebral cortex development by modulating the expression of rEFHC1 into ventricular zone/SVZ cells of the rat developing neocortex using ex vivo electroporation. Immunoblots and RT-PCR on extracts from microdissected embryonic day 17 (E17) cortices indicated that rEFHC1 protein and mRNA were expressed in this structure (Fig. 5a). We carried out immunostaining on E17 brains and found that rEFHC1 was expressed throughout the different cortical regions as a cytoplasmic protein and in the choroid plexus, as described previously14 (Fig. 5b). We validated the effectiveness of our rEFHC1 shRNA vector by western blot using extracts
from microdissected ventricular zone/SVZ after ex vivo electroporation. In comparison with control shRNA, rEFHC1 shRNA silenced EFHC1 expression by 70%, whereas EGFP levels remained unchanged (Fig. 5c). To determine whether EFHC1 regulates the radial migration of projection neurons, we examined the position of EGFPpositive electroporated cells in the ventricular zone/SVZ, intermediate zone or cortical plate 1 or 4 d after ex vivo electroporation of various plasmids at E17. On day 1, cells were mostly located in the ventricular zone/SVZ, with no obvious difference in the position of cells in the different transfected conditions (Fig. 5d). By day 4, the expression of EGFP-hN45 and rEFHC1 shRNA induced a marked disruption of radial migration compared with control shRNA (Fig. 5d). In the presence of rEFHC1 shRNA, most cells were localized in the ventricular zone/SVZ (65.3 ± 3.7% versus 44.7 ± 3.2%) and intermediate zone (28.9 ± 3.5% versus 33.9 ± 4.5%), whereas few cells reached the cortical plate (5.8 ± 2.3% versus 21.4 ± 2.9%; Fig. 5d,e). RNAi-mediated alteration of radial migration was rescued by the expression of EGFP-hEFHC1, although EGFPhEFHC1 expression alone had no significant effect (P < 0.001; Fig. 5d,e). When electroporated with EGFP-hN45, cells migrated out of the ventricular zone/SVZ (44.4 ± 2.7%), but failed to enter the cortical plate (9.3 ± 2.6%), and thus accumulated in the intermediate zone (46.3 ± 3.1%; Fig. 5d,e). These results were supported by experiments conducted in vivo using in utero electroporation (Fig. 5f). The impairment of EFHC1 function appears to have a direct effect on radial redistribution to the cortical plate. Furthermore, immunohistochemistry using antibodies to brain lipid-binding protein (BLBP) revealed a disruption of radial glial processes extension, with concomitant irregular accumulation of cells strongly stained for BLBP in the ventricular zone/SVZ (Fig. 5g). Finally, we observed that EFHC1 loss of function affected the morphology of ventricular zone/SVZ cells, with many cells presenting an enlarged multipolar morphology and lacking or with only a very short leading process, which was sometimes not radially oriented (Fig. 5h). Moreover, we found many round-shaped cells with condensed chromosomes (Fig. 5h) and a decreased number of EGFP-positive cells in the ventricular zone/SVZ of cortices expressing rEFHC1 shRNA and EGFP-hN45 (Fig. 5d). We carried out
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1269
a r t ic l e s
EFHC1 is essential for cell cycle of cortical progenitors Because cortical progenitor cell division occurs before and is tightly coupled to neuronal migration, we decided to study the influence of
b
β-actin
d
Control shRNA
kDa 70
RT-PCR
42 GAPDH rEFHC1
rEFHC1 shRNA
EFHC1 /DNA
EFHC1
c
bp
rEFHC1
400 200
EGFP
EGFP-hN45
rEFHC1 shRNA + EGFP-hEFHC1
EGFP-hEFHC1
*** *
rEFHC1 shRNA + EGFP-hEFHC1
*** ***
*
1
C
FH
h
EGFP
DNA
EGFP/DNA
EGFP-hN45
rEFHC1 shRNA
Control shRNA
EGFP-hEFHC1
rEFHC1 shRNA
VZ/SVZ
***
hR
rEFHC1 shRNA + EGFP-hEFHC1
30
N A rE EG RN FH FP A EG C -h FP 1 s N4 h 5 EG -hE RN FP FH A + -h C1 EF H C 1
+
ls tro on
EGFP-hN45
rE
C
rEFHC1 shRNA
IZ
kDa 70
sh
EGFP cells (%)
1d
100 90 80 70 60 50 40 30 20 10 0
4d EGFP/BLBP BLBP
Figure 5 Essential role of EFHC1 in radial neuronal migration. (a) Western blot with antibody to EFHC1 and RT-PCR analysis showing rEFHC1 expression in E17 rat cortices. (b) EFHC1 protein expression (red) in the choroid plexus (chp) and in the neocortex (ncx) of E17 rat brain with predominant cytoplasmic staining (DNA in blue). CP, cortical plate; IZ, intermediate zone; VZ, ventricular zone. (c) Brain slices obtained after ex vivo electroporation of E17 embryos with 40 *** control shRNA or rEFHC1 shRNA were cultured for 2 d. The electroporated ventricular zone/ 35 SVZ were microdissected (left; GFP in green, DNA in blue) and analyzed by western blot with 30 *** 25 antibody to EFHC1 (right). EGFP was used for normalization. (d) Distribution of EGFP-positive 20 cells in different cortical regions (ventricular zone/SVZ, intermediate zone and cortical plate) 15 1 and 4 d after ex vivo electroporation of rat brains at E17 with control shRNA, rEFHC1 10 shRNA, EGFP-hN45, rEFHC1 shRNA + EGFP-hEFHC1 or EGFP-hEFHC1 (green). 5 (e) Quantification of EGFP-positive cells in different cortical regions (ventricular zone/SVZ, 0 intermediate zone and cortical plate) after ex vivo electroporation. (f) Distribution of EGFPpositive cells in different cortices regions (ventricular zone/SVZ, intermediate zone and cortical plate) of rat embryonic brains electroporated in utero at E14 with control shRNA, rEFHC1 shRNA and rEFHC1 shRNA + EGFP-hEFHC1 (green, DNA in blue) and analyzed at E18. (g) Immunolabeling for BLBP (red, gray) in cortices electroporated with control shRNA, rEFHC1 shRNA, EGFP-hN45, rEFHC1 shRNA + EGFP-hEFHC1 or EGFP-hEFHC1 (green). (h) High magnification of ventricular zone/SVZ after electroporation of control shRNA, rEFHC1 shRNA or EGFP-hN45 (green, DNA in blue). Arrowheads indicate small round-shaped cells, arrows indicate normal bipolar cells (control shRNA) or multipolar cells (rEFHC1 shRNA and EGFP-hN45), and red arrowheads indicate cells with shortened and nonradially oriented leading process. (i) Quantification of apoptosis by TUNEL assays on EGFP-positive transfected ventricular zone/SVZ cells. Quantifications were performed 4 d after ex vivo electroporation. Data are means from eight independent experiments. Error bars show s.e.m. * P < 0.05, ** P < 0.01 and *** P < 0.001. Scale bars represent 500 µm (b, left and middle panels), 200 µm (d,f,g) and 50 µm (h,b right panel).
1270
FP rE F EG HC hN4 5 FP 1 s -h hR EG EF NA FP HC + -h 1 EF H C 1
R
EG
sh
1
C
rE FH
on
tro
ls
hR
N
N A
A
Apoptosis (%)
i
C
© 2009 Nature America, Inc. All rights reserved.
Control shRNA
Control shRNA
f
e CP
g
EGFP/DNA
EGFP/DNA
WB rEFHC1
EFHC1
C
a
EFHC1 on mitosis, proliferation and cell cycle exit of cortical progenitors by ex vivo electroporation of E17 rat brains. We observed that both EGFP-hEFHC1 and EGFP-hN45 associated in situ with the mitotic spindle, although EGFP-hN45 and rEFHC1 shRNA induced mitotic spindle defects similar to those observed in vitro (Fig. 6a). To assess whether theses abnormalities influence mitotic progression, we carried out immunohistochemistry with antibody to phosphohistone H3 (PH3), a mitotic marker. The mitotic index was significantly on tro rE l sh FH R N C A 1 sh R N A
TUNEL assays and found a significantly increased (P < 0.001) number of apoptotic cells in brain slices with impaired EFHC1 function (control shRNA, 12.7 ± 1.6%; rEFHC1 shRNA, 33.3 ± 2.2%; EGFP-hN45, 24.5 ± 1.5%; Fig. 5i).
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
a
EGFP-hN45
rEFHC1 shRNA
EGFP
EGFPhEFHC1
b
α-tubulin DNA
Mitotic index (%)
15
e
Control shRNA
rEFHC1 shRNA
EGFP-hN45
g
Control shRNA
increased (P < 0.001) by impairment of EFHC1 function (rEFHC1 shRNA, 9.5 ± 0.9%; EGFP-hN45, 8.3 ± 1.1%) compared with control shRNA (4.8 ± 0.8%; Fig. 6b). This accumulation of mitotic cells strongly suggests an M phase arrest/delay of progenitor cells, as we observed in vitro. Moreover, although most of the mitotic chromosomes of cells electroporated with control shRNA were normally localized along the ventricular lumen, many EFHC1-deficient mitotic chromosomes were found in the region above the ventricular surface (Fig. 6c). Quantifications revealed a significant increase (P < 0.001) of these ectopic PH3 cells in brain slices with EFHC1 loss of function (control shRNA, 8.7 ± 0.9%; rEFHC1 shRNA, 45.6 ± 1.1%; EGFPhN45, 43.2 ± 1.2%; Fig. 6d). We also explored whether EFHC1 controls the rate at which cortical progenitors exit the cell cycle. For this purpose, we exposed brain slices to a short BrdU pulse (1 h) 24 h after ex vivo electro poration and we fixed the slices 24 h later. Triple immunolabeling with antibodies to GFP, BrdU and Ki67 (Fig. 6e) allowed us to establish the cell cycle exit index. We found that it was significantly decreased (P < 0.001) in brain slices expressing rEFHC1 shRNA (32.6 ± 2.7%) and EGFP-hN45 (34.7 ± 2.3%) compared with control shRNA (44.6 ± 1.9%; Fig. 6f). Finally, we determined the percentage of transfected cells expressing Sox2, a progenitor cell marker. The fraction of ventricular zone/SVZ cells expressing Sox2 was significantly enhanced (P < 0.001) by EFHC1 inhibition (control shRNA, 18.6 ± 1.7%; rEFHC1 shRNA, 49.2 ± 1.2%; EGFP-hN45, 37.1 ± 1.6%; Fig. 6g,h). Altogether, these data suggest that the impairment of EFHC1 function in the developing nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
rEFHC1 shRNA
EGFP-hN45
Ectopic mitotic cells (%)
f Cell cycle exit index (%)
EGFP-hN45
h
EGFP+/Sox2+ cells (%)
rEFHC1 shRNA
EGFP/PH3
Control shRNA
***
5
Merge
d
c
***
10
0
EGFP/BrdU/Ki67
Figure 6 EFHC1 is crucial for mitosis and cell cycle exit of cortical progenitors. (a) Ventricular zone/SVZ of rat embryo brains electroporated ex vivo at E17 with EGFP-hEFHC1, EGFPhN45 and rEFHC1 shRNA (green), cultured as slices for 1 d and stained for α-tubulin (red) and DNA (blue). Association of EGFP-hEFHC1 and EGFP-hN45 with the mitotic spindle is indicated by arrows. The EGFP-hEFHC1 panels show a normal mitotic cell in metaphase, and the EGFP-hN45 and rEFHC1 shRNA panels show a monopolar spindle (left) and misaligned chromosomes in metaphase (right, arrowheads). (b–h) Ventricular zone/SVZ of E17 rat brains electroporated ex vivo with control shRNA, rEFHC1 shRNA or EGFP-hN45 (green) and cultured as slices for 2 d. (b) Quantification of mitotic index. (c) Immunostaining for PH3 (red), a mitotic marker. Ectopic mitotic cells are indicated by arrows (yellow). (d) Quantification of ectopic mitotic cells. (e) Immunolabeling for BrdU (red) and Ki67 (blue). Electroporated cells that exited the cell cycle were EGFP and BrdU double-positive and Ki67 negative (yellow, arrows), whereas cells that were still in the cell cycle are positive for EGFP, BrdU and Ki67 (white, arrowheads). (f) Quantification of cell cycle exit index. (g) Immunolabeling for Sox2, a marker of progenitor cells. EGFP and Sox2 double-positive cells are indicated by arrows (yellow). (h) Quantification of EGFP and Sox2 double-positive cells. Quantification was carried out 2 d after ex vivo electroporation. Data are means from eight independent experiments. Error bars show s.e.m. *** P < 0.001. Scale bars represent 20 µm (a) and 200 µm (c,e,g).
EGFP/Sox2
© 2009 Nature America, Inc. All rights reserved.
a r t ic l e s
Control shRNA
50
rEFHC1 shRNA
***
EGFPhN45
***
40 30 20 10 0
Control shRNA
rEFHC1 shRNA
EGFPhN45
***
***
rEFHC1 shRNA
EGFPhN45
50 40 30 20 10 0
Control shRNA
60
***
50
***
40 30 20 10 0
Control shRNA
rEFHC1 shRNA
EGFPhN45
neocortex interferes with both cell division and cell cycle exit of cortical progenitors, leading to their accumulation. EFHC1 is required for locomotion of postmitotic neurons To investigate whether EFHC1-deficient postmitotic neurons were able to migrate, we acutely electroporated E19 brain slices in the intermediate zone with plasmids encoding control shRNA or rEFHC1 shRNA following the ex vivo electroporation of a plasmid encoding a red fluorescent protein (RFP) to label ventricular zone/SVZ progenitor cells. This method allowed us to distinguish between cells that were progenitors (RFP positive) from those that were postmitotic neurons (EGFP positive and RFP negative) at the time of electroporation. We analyzed the cortical scattering of EGFP-positive and RFP-negative cells 2 d after electroporation using an arbitrary scale divided into ten bins (Fig. 7a). Compared with control shRNA, expression of rEFHC1 shRNA impaired cortical redistribution of postmitotic neurons in the cortical plate, as the majority of cells were located in the upper intermediate zone (bin 6, 27.2 ± 1.6% versus 18.8 ± 0.9%; bin 7, 22.7 ± 3.6% versus 17.7 ± 1.1%; Fig. 7b) with a concomitant decrease of cells in the cortical plate (bin 9, 1.7 ± 1.2% versus 7.1 ± 1.1%; bin 10, 0.7 ± 0.6% versus 3.3 ± 0.5%; Fig. 7b). To confirm these results, we carried out ex vivo electroporation at E17 of plasmids encoding EGFP or EGFP-hN45 under the control of the NeuroD promoter22,23. Cortical scattering of EGFP-positive cells 1271
a r t ic l e s Control shRNA/ RFP
b
10
10
10
9
9
9
8
8
8
7
7
6
6
5
5
4
4
3
3
2
2
1
1
c
EGFP
Control shRNA rEFHC1 shRNA
** **
7
*
6
***
5 4 3 2 1
EGFP-hN45
0
d 10
10
10
9
9
9
8
8
8
7
7
6
6
5
5
4
4
3
3
2
2
1
1
5 10 15 20 25 + – EGFP /RFP cells (%)
*
30
EGFP EGFP-hN45
Figure 7 EFHC1 is implicated in the locomotion of postmitotic neurons. (a) E19 rat brain slices electroporated ex vivo with RFP plasmid (ventricular zone/SVZ progenitors in red) followed by focal electroporation of control shRNA or rEFHC1 shRNA plasmids in the intermediate zone (green). Postmitotic neurons are EGFP positive and RFP negative. (c) E17 rat brain slices electroporated ex vivo with NeuroD EGFP or NeuroDEGFP–hN45 plasmids (green), expressing protein only in postmitotic neurons. (b,d) Quantification of cortical scattering of postmitotic neurons in the above conditions using an arbitrary scale divided into ten bins. Quantification was carried out 2 d (b) or 4 d (d) after electroporation. Data are means from seven independent experiments. Error bars show s.e.m. * P < 0.05, ** P < 0.01 and *** P < 0.001. Scale bars represent 200 µm.
1272
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
Cortical scattering
DISCUSSION In a previous study, we found that EFHC1, a protein that is mutated in JME, a common form of idiopathic generalized epilepsies, colocalizes with the centrosome and the mitotic spindle in cultured cells16. Here, we found that EFHC1 is a new MAP interacting directly with micro tubules through a region containing the first 45 N-terminal amino acids. These results are consistent with those of other reports demonstrating the association of EFHC1 with microtubule-based structures, such as motile cilia and flagella13,14. The first 45-amino-acid region is well conserved among EFHC1 orthologs; however, it shows no clear homology with the MTBDs described in other MAPs, suggesting that EFHC1 possesses a different type of MTBD. Thus, EFHC1 would appear to be a unique MAP, unrelated to conventional MAPs. Using RNAi and overexpression of a dominant-negative construct (EGFP-hN45), we found that EFHC1 is important in the control of mitotic spindle formation in HEK293 cells. Moreover, EFHC1 impairment caused microtubule bundling that was resistant to the depolymerization induced by nocodazole, as has been observed with other MAPs, such as ASAP24, hCLASP/Orbit25 and mNAV1 (ref. 26). These results further support the hypothesis that EFHC1, a centrosomal MAP, is important for microtubule scaffold organization. Furthermore, we found that EFHC1 loss of function led to disruption of mitotic progression, as shown by the accumulation of cells arrested at prometaphase and the increase of the mitotic index. These abnormalities are more likely to be a consequence of mitotic spindle defects, as it is the case for LIS1 (ref. 27). Finally, the increased apoptosis induced by EFHC1 impairment cells seems to be the consequence of mitotic spindle defects, as shown by the increased number
of apoptotic cells in the S and G2/M phases of the cell cycle. These observations seem to be consistent, as perturbations in centrosome and microtubule organization that result in aberrant spindle formation have been previously associated with cell death24,28. In the developing neocortex, we analyzed the consequences of acute EFHC1 deficiency on radial migration of projection neurons. Our results indicate that EFHC1 deficiency gave rise to a disruption of radial migration by affecting different steps of neurogenesis: division and cell cycle exit of cortical progenitors, organization of radial glia scaffolding and locomotion of postmitotic neurons in the intermediate zone. In EFHC1-deficient cortical progenitors, we observed mitotic spindle defects and disruption of mitotic progression, as shown by the accumulation of mitotic cells with ectopic mitotic chromosome position, suggesting a disorganization of interkinetic nuclear oscillation. Indeed, the division of cortical progenitors is precisely regulated by a unique process in which appropriate positioning of mitotic chromosomes by proper assembly of the mitotic spindle appears to have a pivotal role for mitotic progression29. The position of the nucleus is known to correlate with the phase of the cell cycle, in which the nuclei of mitotic cells normally descend to the apical ventricular surface by interkinetic nuclear migration30. As mitotic spindles control the position and the alignment of mitotic chromosomes, the mispositioning and the accumulation of the mitotic nuclei in EFHC1-deficient progenitor cells is probably a direct reflection of mitotic microtubule disorganization31. On the other hand, it is now well established that the developing neocortex contains both types of progenitors: radial progenitors dividing at the apical surface (ventricular zone) and basal progenitors dividing at a basal position (SVZ)32,33. Therefore, ectopic mitotic cells observed after EFHC1 impairment could also correspond to an accumulation of dividing basal progenitors. We also observed a significant reduction (P < 0.001) of the number of cells that exited the cell cycle, and this resulted in the accumulation of cycling cortical progenitor cells. Cortical progenitors exhibit two different modes of division that correlate with cell fate determination. Symmetrical cell divisions produce two daughter progenitor cells, whereas asymmetrical cell divisions result in an apical progenitor cell and a basal postmitotic neuron. EFHC1 may participate in fate determination of cortical progenitors by regulating centrosome position and/or mitotic spindle orientation, as has been suggested by recent studies34,35. Without the completion of M phase and the segregation of the two sets of chromosomes, many cellular events that are essential for fate determination may not occur properly. This may result in a block of the transition out of the cortical progenitor stage, leading to the accumulation of progenitors. Because cortical progenitor cell division occurs before and is tightly coupled to neuronal migration, the defects in cortical progenitor mitosis and cell cycle exit, together with apoptosis, may have a strong effect on subsequent neuronal migration process20,31. Furthermore, the nearly complete depletion of the radial glia scaffold after EFHC1 loss of function probably contributes to the
Cortical scattering
© 2009 Nature America, Inc. All rights reserved.
rEFHC1 shRNA/ RFP
Cortical scattering
Cortical scattering
a
** *
7
**
6 5 4 3
*
2 1
0
5 10 15 20 25 30 35 40 EGFP+ cells (%)
was analyzed 4 d after electroporation (Fig. 7c). Compared with brains expressing EGFP, only a subset of cells had reached the lower cortical plate in brains expressing EGFP-hN45 (bin 7, 6.2 ± 0.6% versus 12.1 ± 1.6%; bin 8, 3.1 ± 0.7% versus 5.6 ± 2.1%; bin 9, 1.7 ± 0.4% versus 4.2 ± 1.6%; bin 10, 0.02 ± 0.02% versus 3.5 ± 1.3%; Fig. 7d). Because postmitotic neurons were not connected to the pial surface, migration occurred by locomotion in these experiments. These results suggest that EFHC1 is involved in the locomotion of postmitotic neurons in the intermediate zone.
© 2009 Nature America, Inc. All rights reserved.
a r t ic l e s isruption of radial migration. Actually, radial glial cells are neuronal d progenitors during cortical development36 and are essential for both modes of radial migration: locomotion37 and somal translocation21. Molecular abnormalities affecting the development of radial glial cells lead to abnormal neuronal migration38,39. Whether the disruption of radial glia scaffolding is a consequence of defects in mitosis and cell cycle exit of progenitors or a direct influence of EFHC1 on radial processes extension remains unclear. Finally, we found that EFHC1 is critical for the locomotion of postmitotic neurons in the intermediate zone, which contribute to neuronal migration defects. Migration by locomotion involves forward extension of the migratory process and somal translocation to keep up21. Recent evidence has suggested a role for LIS1, dynein and NDEL1 in coupling the nucleus and the centrosomes in migratory neurons to allow somal translocation during locomotion40,41. As EFHC1 is a MAP associated to the centrosome, it could be involved in such a mechanism. All of the phenotypes that we observed after EFHC1 loss of function in the rat neocortex closely resembled defects produced by knockdown of other MAPs, including LIS1, NDEL1 and DCLK. Indeed, these MAPs also localize to the centrosomes and affect both neurogenesis and neuronal migration20,31,42–44. Therefore, EFHC1 could be implicated in the same cellular pathways as these proteins for regulating progenitor cell mitosis and radial migration. A recent study found that Efhc1−/− adult mice have an almost normal outward appearance, but nonetheless have a reduced seizure threshold and some spontaneous myoclonus, indicating that EFHC1 defects are probably involved in producing cortical hyperexcitability45. However, no detailed developmental studies or specific physiological role were proposed for Efhc1, as the study found that the ciliary defect is not related to the epileptic phenotype and does not create abnormal distribution of GABAergic interneurons in the forebrain15. The absence of gross anatomical defects in Efhc1−/− brains could be surprising if one considers our data. We found that our RNAi-mediated silencing of rEFHC1 was specific through the rescue of rEFHC1 shRNA phenotype with full-length hEFHC1. These differences somewhat recall what has been observed with doublecortin (DCX), another well-known MAP. Indeed, DCX null mutations in mice neither disrupt neuronal migration nor cause DCX syndrome46. However, in utero RNAi of DCX in rats can be used to model DCX syndrome47,48. These apparent paradoxes are probably the results of differences between genetic deletion and RNAi methodologies. First, the time of disruption in the RNAi technique corresponds to the transfection time and is then temporally discrete. Second, only a subset of cells are transfected in otherwise normal tissue. Third, a substantial amount of endogenous protein and mRNA is still present in targeted cells, as is unlikely that any RNAi is 100% effective47. These differences could explain a lack of compensatory mechanism with RNAi methodology that may take place with genetic deletion, especially if paralog genes are present and functionally compensate for each other, as it could be the case with EFHC2 (ref. 49). Moreover, one has to consider that some effects can be species specific48. Finally, it should be kept in mind that, because only missense mutations are reported in EFHC1, our models (RNAi and dominant-negative protein) may not totally reflect what happens in JME. In summary, we have found for the first time, to the best of our knowledge, that a gene responsible for idiopathic generalized epilepsy encodes a new MAP that is crucial for the regulation of cell division and neuronal migration during development. Accordingly, our findings suggest that microdysgenesis found in individuals suffering from JME3,4 partly results from radial neuronal migration defects and that
these defects could lead to abnormal epileptogenic circuitry during cortical maturation at adolescence onset.
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1273
Methods Methods and any associated references are available in the online version of the paper at http://www.nature.com/natureneuroscience/. Note: Supplementary information is available on the Nature Neuroscience website. Acknowledgments We thank B. Coumans for skillful technical assistance, J. Bai for help with in utero electroporation processing and S. Ormenese from the GIGA-Imaging and Flow Cytometry platform for support with flow cytometry. We are also grateful to A. Adamantidis for critical reading of the manuscript. We thank F. Polleux for the NeuroD-IRES-GFP plasmid. This work was supported by grants from the F.R.S.-FNRS (Fonds de la Recherche Scientifique Médicale 3.4565.03 to T.G. and B.L.) and the Léon Fredericq Foundation (to L.d.N.). L.N. and B.L. are research associates at the F.R.S.-FNRS. AUTHOR CONTRIBUTIONS L.d.N. performed all of the experiments (except for immunoprecipitations), data analysis and wrote the manuscript with help from and editing by all the co-authors. C.L. conducted immunoprecipitations experiments. L.N. helped with the ex vivo and focal electroporation processing. J.J.L. trained L.d.N. for in utero electroporation studies and gave advice as to the interpretation of data. A.V.D.-E. provided plasmids and strong scientific support. T.G. and B.L. were the project leaders. They directed the follow-up of all experiments and supervised the data analysis. Published online at http://www.nature.com/natureneuroscience/. Reprints and permissions information is available online at http://www.nature.com/ reprintsandpermissions/. 1. Delgado-Escueta, A.V. et al. Mapping and positional cloning of common idiopathic generalized epilepsies: juvenile myoclonus epilepsy and childhood absence epilepsy. Adv. Neurol. 79, 351–374 (1999). 2. Delgado-Escueta, A.V. & Enrile-Bacsal, F. Juvenile myoclonic epilepsy of Janz. Neurology 34, 285–294 (1984). 3. Meencke, H.J. & Janz, D. Neuropathological findings in primary generalized epilepsy: a study of eight cases. Epilepsia 25, 8–21 (1984). 4. Woermann, F.G., Free, S.L., Koepp, M.J., Sisodiya, S.M. & Duncan, J.S. Abnormal cerebral structure in juvenile myoclonic epilepsy demonstrated with voxel-based analysis of MRI. Brain 122, 2101–2108 (1999). 5. Delgado-Escueta, A.V. Advances in genetics of juvenile myoclonic epilepsies. Epilepsy Curr. 7, 61–67 (2007). 6. Cossette, P. et al. Mutation of GABRA1 in an autosomal dominant form of juvenile myoclonic epilepsy. Nat. Genet. 31, 184–189 (2002). 7. Haug, K. et al. Mutations in CLCN2 encoding a voltage-gated chloride channel are associated with idiopathic generalized epilepsies. Nat. Genet. 33, 527–532 (2003). 8. Suzuki, T. et al. Mutations in EFHC1 cause juvenile myoclonic epilepsy. Nat. Genet. 36, 842–849 (2004). 9. Annesi, F. et al. Mutational analysis of EFHC1 gene in Italian families with juvenile myoclonic epilepsy. Epilepsia 48, 1686–1690 (2007). 10. Ma, S. et al. Mutations in the GABRA1 and EFHC1 genes are rare in familial juvenile myoclonic epilepsy. Epilepsy Res. 71, 129–134 (2006). 11. Stogmann, E. et al. Idiopathic generalized epilepsy phenotypes associated with different EFHC1 mutations. Neurology 67, 2029–2031 (2006). 12. Medina, M.T. et al. Novel mutations in Myoclonin1/EFHC1 in sporadic and familial juvenile myoclonic epilepsy. Neurology 70, 2137–2144 (2008). 13. Ikeda, T. et al. The mouse ortholog of EFHC1 implicated in juvenile myoclonic epilepsy is an axonemal protein widely conserved among organisms with motile cilia and flagella. FEBS Lett. 579, 819–822 (2005). 14. Suzuki, T. et al. Sequential expression of Efhc1/myoclonin1 in choroid plexus and ependymal cell cilia. Biochem. Biophys. Res. Commun. 367, 226–233 (2008). 15. King, S.M. Axonemal protofilament ribbons, DM10 domains and the link to juvenile myoclonic epilepsy. Cell Motil. Cytoskeleton 63, 245–253 (2006). 16. de Nijs, L. et al. EFHC1, a protein mutated in juvenile myoclonic epilepsy, associates with the mitotic spindle through its N-terminus. Exp. Cell Res. 312, 2872–2879 (2006). 17. Grisar, T. et al. Some genetic and biochemical aspects of myoclonus. Neurophysiol. Clin. 36, 271–279 (2006). 18. Gupta, A., Tsai, L.H. & Wynshaw-Boris, A. Life is a journey: a genetic look at neocortical development. Nat. Rev. Genet. 3, 342–355 (2002). 19. LoTurco, J.J. & Bai, J. The multipolar stage and disruptions in neuronal migration. Trends Neurosci. 29, 407–413 (2006). 20. Shu, T. et al. Doublecortin-like kinase controls neurogenesis by regulating mitotic spindles and M phase progression. Neuron 49, 25–39 (2006).
© 2009 Nature America, Inc. All rights reserved.
a r t ic l e s 21. Nadarajah, B. & Parnavelas, J.G. Modes of neuronal migration in the developing cerebral cortex. Nat. Rev. Neurosci. 3, 423–432 (2002). 22. Hand, R. et al. Phosphorylation of Neurogenin2 specifies the migration properties and the dendritic morphology of pyramidal neurons in the neocortex. Neuron 48, 45–62 (2005). 23. Heng, J.I. et al. Neurogenin 2 controls cortical neuron migration through regulation of Rnd2. Nature 455, 114–118 (2008). 24. Saffin, J.M. et al. ASAP, a human microtubule-associated protein required for bipolar spindle assembly and cytokinesis. Proc. Natl. Acad. Sci. USA 102, 11302–11307 (2005). 25. Aonuma, M. et al. Microtubule bundle formation and cell death induced by the human CLASP/Orbit N-terminal fragment. Cell Struct. Funct. 30, 7–13 (2005). 26. Martínez-López, M.J. et al. Mouse neuron navigator 1, a novel microtubuleassociated protein involved in neuronal migration. Mol. Cell. Neurosci. 28, 599–612 (2005). 27. Faulkner, N.E. et al. A role for the lissencephaly gene LIS1 in mitosis and cytoplasmic dynein function. Nat. Cell Biol. 2, 784–791 (2000). 28. Wakefield, J.G., Stephens, D.J. & Tavare, J.M. A role for glycogen synthase kinase3 in mitotic spindle dynamics and chromosome alignment. J. Cell Sci. 116, 637–646 (2003). 29. Ohnuma, S. & Harris, W.A. Neurogenesis and the cell cycle. Neuron 40, 199–208 (2003). 30. Noctor, S.C., Martinez-Cerdeno, V., Ivic, L. & Kriegstein, A.R. Cortical neurons arise in symmetric and asymmetric division zones and migrate through specific phases. Nat. Neurosci. 7, 136–144 (2004). 31. Feng, Y. & Walsh, C.A. Mitotic spindle regulation by Nde1 controls cerebral cortical size. Neuron 44, 279–293 (2004). 32. Pontious, A., Kowalczyk, T., Englund, C. & Hevner, R.F. Role of intermediate progenitor cells in cerebral cortex development. Dev. Neurosci. 30, 24–32 (2008). 33. Kowalczyk, T. et al. Intermediate neuronal progenitors (basal progenitors) produce pyramidal-projection neurons for all layers of cerebral cortex. Cereb. Cortex published online, doi:10.1093/cercor/bhn260 (23 January 2009). 34. Roegiers, F. & Jan, Y.N. Asymmetric cell division. Curr. Opin. Cell Biol. 16, 195–205 (2004).
35. Yamashita, Y.M., Jones, D.L. & Fuller, M.T. Orientation of asymmetric stem cell division by the APC tumor suppressor and centrosome. Science 301, 1547–1550 (2003). 36. Noctor, S.C., Flint, A.C., Weissman, T.A., Dammerman, R.S. & Kriegstein, A.R. Neurons derived from radial glial cells establish radial units in neocortex. Nature 409, 714–720 (2001). 37. Rakic, P. Mode of cell migration to the superficial layers of fetal monkey neocortex. J. Comp. Neurol. 145, 61–83 (1972). 38. Halfter, W., Dong, S., Yip, Y.P., Willem, M. & Mayer, U. A critical function of the pial basement membrane in cortical histogenesis. J. Neurosci. 22, 6029–6040 (2002). 39. Hartfuss, E. et al. Reelin signaling directly affects radial glia morphology and biochemical maturation. Development 130, 4597–4609 (2003). 40. Tanaka, T. et al. Lis1 and doublecortin function with dynein to mediate coupling of the nucleus to the centrosome in neuronal migration. J. Cell Biol. 165, 709–721 (2004). 41. Tsai, L.H. & Gleeson, J.G. Nucleokinesis in neuronal migration. Neuron 46, 383–388 (2005). 42. Yingling, J. et al. Neuroepithelial stem cell proliferation requires LIS1 for precise spindle orientation and symmetric division. Cell 132, 474–486 (2008). 43. Tsai, J.W., Chen, Y., Kriegstein, A.R. & Vallee, R.B. LIS1 RNA interference blocks neural stem cell division, morphogenesis, and motility at multiple stages. J. Cell Biol. 170, 935–945 (2005). 44. Shu, T. et al. Ndel1 operates in a common pathway with LIS1 and cytoplasmic dynein to regulate cortical neuronal positioning. Neuron 44, 263–277 (2004). 45. Suzuki, T. et al. Efhc1 deficiency causes spontaneous myoclonus and increased seizure susceptibility. Hum. Mol. Genet. 18, 1099–1109 (2009). 46. Corbo, J.C. et al. Doublecortin is required in mice for lamination of the hippocampus but not the neocortex. J. Neurosci. 22, 7548–7557 (2002). 47. Bai, J. et al. RNAi reveals doublecortin is required for radial migration in rat neocortex. Nat. Neurosci. 6, 1277–1283 (2003). 48. Ramos, R.L., Bai, J. & LoTurco, J.J. Heterotopia formation in rat, but not mouse, neocortex after RNA interference knockdown of DCX. Cereb. Cortex 16, 1323–1331 (2006). 49. Young-Pearse, T.L. et al. A critical function for beta-amyloid precursor protein in neuronal migration revealed by in utero RNA interference. J. Neurosci. 27, 14459–14469 (2007).
1274
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
ONLINE METHODS
© 2009 Nature America, Inc. All rights reserved.
Plasmid constructions. For expression in eukaryotic cells, various EGFP and c-Myc–tagged truncated EFHC1 constructs, as well as the rEFHC1 coding sequence, were cloned into the pcDNA5/FRT/TO vector (Invitrogen)16. For bacterial expression of GST-tagged proteins, the coding sequences of hEFHC1, hN45 and hN30 were cloned into the pGEX-5X-1 vector (GE Healthcare). For RNAi in HEK293 cells, we tested eight shRNAs and selected two representatives. The targeting sequences were 5′-CTA CCG ACA TGA TCA CTA TCT-3′ for hEFHC1 shRNA#1, a predesigned GFP-SureSilencing shRNA plasmid (SuperArray), and 5′-GTT GTT AAA CCA TAC TCT AC-3′ for hEFHC1 shRNA#2, an shRNA constructed from synthesized oligonucleotides (Eurogentec) and cloned into the RNAi-Ready pSIREN vector (BD Biosciences). For ex vivo, in utero and focal electroporations, pCAGGS-EGFP, pCAGGSRFP, mU6pro (gift from J. LoTurco, University of Connecticut) and NeuroDIRES-EGFP plasmids (gift from F. Polleux, University of North Carolina) were used. Expression plasmids were constructed by cloning EGFP-hEFHC1 and EGFP-hN45 into pCAGGS and NeuroD-EGFP vectors. Plasmids coding for shRNAs specific for rEFHC1 were constructed from synthesized oligonucleotides (Eurogentec) and cloned into the mU6pro vector. We tested eight shRNAs and selected one representative. The targeting sequence was 5′-GAC ATT GAA ATC CAC CAC AAA-3′. All constructs were verified by sequencing. Cell culture and transfection. HEK293 cells were grown in DMEM (Gibco) supplemented with 10% fetal bovine serum (vol/vol), Greiner Bio-one in a humidified incubator at 37 °C under 5% CO2 atmosphere. Stably transfected HEK Flp-In T-Rex 293 cells expressing EGFP-hEFHC1 were cultured and induced as described previously16. Transient transfections were performed using TransIT Transfection Reagent (Mirus). Animals. Time-pregnant female Wistar rats were housed under standard conditions and were treated according to the guidelines of the Belgian Ministry of Agriculture in agreement with European community Laboratory Animal Care and Use regulations. RNA isolation and semiquantitative RT-PCR. Total RNAs were extracted from HEK293 cells using Instapure (Eurogentec) and reverse transcribed using MMLV and random hexamer primers (Promega). EFHC1 and GAPDH cDNA were simultaneously amplified with specific primers: hEFHC1 forward 5′-CAT CCC AGA AGG ACA AAG ACC G-3′ (exon 8) and reverse 5′-GTA TCA AGG ATG ATG AAC CGG-3′ (exon 9), rEFHC1 forward 5′-CGC CAG CGA CTA GCC AAG AAT GAT GTG-3′ (exon 5) and reverse 5′- CTG CTT CAG TTG GTC AAA GTC TGA TGG-3′ (exon 6), and GAPDH forward 5′-ACC ACA GTC CAT GCC ATC AC-3′ and reverse 5′-TCC ACC ACC CTG TTG CTG TA-3′. For analysis, 10 µl of PCR reaction were loaded onto a 1% agarose (wt/vol) gel. Band intensities were quantified using the ImageMaster 1D Elite software (GE Healthcare). Tissue processing. Embryonic brains were dissected in 0.1 M phosphate-buffered saline (PBS, pH 7.4) and fixed in 4% paraformaldehyde (wt/vol) for 1 h at 4 °C. Cultured brain slices were fixed in 4% paraformaldehyde for 30 min at 4 °C. Fixed samples were cryoprotected overnight in 20% sucrose (wt/vol) in PBS at 4 °C, embedded in Optical cutting temperature compound (VWR International) and sectioned coronally onto slides (SuperFrost Plus, VWR International, 12 µm) with a cryostat (CM3050S, Leica). Immunostaining procedures. For immunostaining, we used the following primary antibodies: rabbit polyclonal antibody to EFHC1 (gift from T. Ikeda13, University of Tokyo, 1:50), mouse monoclonal antibody to α-tubulin (clone B-5-1-2, Sigma, 1:2,000), mouse monoclonal antibody to γ-tubulin (clone GTU-88, Sigma, 1:500), rabbit polyclonal antibody to GFP (Molecular Probe, 1:2,000), mouse monoclonal antibody to βIII-tubulin (Tuj1, Covance, 1:1,000), rabbit polyclonal antibody to BLBP (Chemicon, 1:500), rabbit monoclonal antibody to PH3 (clone JY325, Upstate, 1:500), rat monoclonal antibody to BrdU (clone BU1/75, AbD serotec, 1:200) or mouse monoclonal antibody to Ki67 (clone B56, BD Pharmigen, 1:50). For secondary antibodies, we used RRX-, FITC- or Cy5-conjugated antibody to rabbit, mouse or rat IgG (Jackson ImmunoResearch, 1:500). For immunocytochemistry, cells grown on glass coverslips coated with polyornithine (0.1 mg ml−1, Sigma) and laminin (5 µg ml−1, MP Biomedicals) were
doi:10.1038/nn2390
fixed in methanol for 6 min at −20 °C or in 4% paraformaldehyde for 10 min at 20–25 °C, followed by permeabilization in 0.1% Triton X-100 (vol/vol). Blocking was carried out in PBS containing 1.5% nonfat dried milk (wt/vol) for 1 h at 37 °C. Primary antibodies were incubated overnight at 4 °C. After washing, the incubation with secondary antibodies was done for 1 h at 20–25 °C. Coverslips were mounted in Vectashield Hard Set Mounting Medium with DAPI (Vector Laboratories) and observed using a Nikon TE 2000-U microscope coupled to a digital camera DXM1200F. Images were processed using ImageJ software (US National Institutes of Health). For immunohistochemistry, cryostat sections were washed three times in PBS with 0.1% Triton X-100 and blocked at 20–25 °C for 1 h in PBS with 0.1% Triton X-100 supplemented with 10% normal goat serum (vol/vol Vector Laboratories). For BrdU immunostaining, cryosections were pretreated with 2 N HCl for 30 min at 37 °C followed by neutralization in Na2B4O7 for 15 min at 37 °C before blocking. Primary antibodies were incubated overnight at 4 °C. After washing, slides were incubated for 1 h at 20–25 °C with the appropriate secondary antibodies. Slides were mounted in Vectashield Hard Set Mounting Medium with DAPI (Vector Laboratories). Images were acquired using an Olympus Fluoview FV1000 confocal system equipped with an Olympus IX81 inverted microscope (Olympus) and processed using ImageJ software. Detection of apoptosis. Apoptotic cells were detected by TUNEL assay using the In Situ Cell Death Detection kit, TMR red (Roche Applied Science). Flow cytometry. Apoptotic cells were stained with annexin V–phycoerythrin and 7-amino actinomycin (7-AAD) using an Annexin V-PE Apoptosis Detection kit I (BD Biosciences). EGFP-positive cells in early apoptosis (Annexin V–phycoerythrin positive and 7-AAD negative) were sorted by flow cytometry using a FACSVantage cytometer (BD Biosciences). Cell nuclei were isolated and stained with propidium iodide using the CycleTEST PLUS DNA Reagent kit (BD Biosciences). Cell cycle was analyzed by flow cytometry using a FACSVantage cytometer. Immunoprecipitation. Cells were washed in PBS and homogenized in lysis buffer (50 mM Tris-HCl (pH 7.5), 150 mM NaCl and 1% Triton X-100), supplemented with phosphatase inhibitors (10 mM NaF and 1 mM Na3VO4) and protease inhibitors (complete, Roche). Lysates were clarified by centrifugation at 16,000 g for 20 min at 4 °C. -tubulin was immunoprecipitated using antibody to -tubulin (clone DM1A, Sigma) and Immunoprecipitation Kit, Protein G (Roche Applied Science) and c-Myc–tagged proteins were immunoprecipitated using antibody to Myc coupled to agarose beads (Alpha Diagnostics). Immunoprecipitates were washed five times in lysis buffer and subjected to western blot analysis. Western blotting. Proteins were separated on 10% SDS-polyacrylamide gels and transferred onto PVDF membranes using an electroblot apparatus (Bio-Rad). Membranes were blocked in TTBS (50 mM Tris-HCl (pH 7.4), 150 mM NaCl, 0.1% Tween 20, vol/vol) containing 10% nonfat dried milk for 1 h and subsequently probed overnight at 4 °C with primary antibodies. For primary antibodies, we used mouse monoclonal antibody to α-tubulin (clone B-5-1-2, Sigma, 1:2,000), mouse monoclonal antibody to γ-tubulin (clone GTU-88, Sigma, 1:1,000), mouse monoclonal antibody to β-actin (clone AC-15, Sigma, 1:3,000), rabbit polyclonal antibody to GFP (Invitrogen, 1:5,000), mouse monoclonal antibody to c-Myc (Santa Cruz Biotechnology, 1:1,000) and rabbit polyclonal antibody to EFHC1 (gift from T. Ikeda13, 1:200). The filters were washed three times in TTBS and incubated for 1 h at 20–25 °C with secondary antibodies. For secondary antibodies, we used horseradish peroxidase–conjugated goat antibody to mouse (1:3,000) or rabbit IgG (1:5,000) or Cy5-conjugated donkey antibody to rabbit IgG (1:5,000, Jackson ImmunoResearch). Staining was detected by enhanced chemiluminescence (SuperSignal, Pierce) or by fluorescence using a Typhoon apparatus (GE Healthcare). Band intensities were quantified using the ImageMaster 1D Elite software. Production and purification of GST-tagged proteins. pGEX-5X-1 plasmid was used to express GST-tagged proteins in Escherichia coli BL21. Different proteins were affinity purified on a GSTrap HP column (GE Healthcare). In vitro microtubule binding assays. Microtubules were polymerized in vitro and stabilized with taxol using the Microtubules/Tubulin Biochem
nature NEUROSCIENCE
kit (Cytoskeleton). Microtubule binding assays were performed with purified GST-tagged proteins as described previously24. In vitro tubulin polymerization assays. Tubulin polymerization assays were performed using the Tubulin Polymerization Assay kit (Cytoskeleton).
© 2009 Nature America, Inc. All rights reserved.
In vitro microtubule bundling assays. Microtubules were polymerized in vitro and stabilized with taxol using the Microtubules/Tubulin Biochem kit (Cytoskeleton). Purified GST-tagged proteins (1 µg µl−1) were incubated at 20–25 °C with polymerized microtubules (5 µM) in BRB80 buffer (80 mM potassium PIPES, 1 mM MgCl2 and 1 mM EGTA). Absorbance was measured at 350 nm over 30 min. Ex vivo and in utero electroporation. Ex vivo electroporation was performed as described previously50 with minor modifications. Briefly, the heads of rat embryos (17 d of gestation) were cut off and placed in L15 medium (Invitrogen) supplemented with 1 M glucose, 1 M NaHCO3 and 1% penicillin/streptomycin (vol/vol). We injected 1–3 µl of plasmid solution in PBS with Fast Green (2 mg ml−1, Sigma) into the lateral ventricles using a pulled glass micropipette and a microinjector (Femtojet, Eppendorf). The following vector combinations were used: mU6pro control or rEFHC1 shRNAs (1.5 µg µl−1) with pCAGGS-EGFP (0.5 µg µl−1), rEFHC1 shRNA (1.5 µg µl−1) with pCAGGS-EGFP-hEFHC1 (2.5 µg µl−1), pCAGGS-EGFP-hN45 or EGFP-hEFHC1 (2.5 µg µl−1), NeuroDEGFP or NeuroD-EGFP-hN45 (2.5 µg µl−1). Electroporation experiments were carried out by placing heads between tweezer-type electrodes (BTX). Square electric pulses (55 V, 50 ms) were passed five times at 1-s intervals using an electroporator (ECM 830, BTX). Following electroporation, brains were dissected and transferred into liquid 3% low-melting agarose (38 °C, BioRad) and incubated on ice for 1 h. Embedded brains were cut coronally (300 µm) with a vibratome (VT1000S, Leica). All these steps were performed at 4 °C. Slices were transferred onto sterilized culture plate inserts (0.4-µm pore size, Millicell-CM, Millipore) and cultured in semidry conditions in a humidified incubator at 37 °C under 5% CO2 atmosphere in wells containing Neurobasal medium (Invitrogen) supplemented with 1% B27 (vol/vol), 1% N2 (vol/vol), 1% glutamine (vol/vol) and 1% penicillin/streptomycin (Invitrogen). Slices were cultured for up to 4 d, with half the culture medium renewed every day. In utero electroporation was performed as described previously47. Focal electroporation. Brains from E19 rat embryos were subjected to ex vivo electroporation with pCAGGS-RFP plasmid (0.5 µg µl−1). After cutting, brain slices were transferred onto 1% low-melting agarose placed onto a petridish square platinum plate electrode (Nepagene). We injected mU6pro control or rEFHC1 shRNAs plasmids (1.5 µg µl−1) along with pCAGGS-GFP (0.5 µg µl−1) in PBS solution with Fast Green (2 mg ml−1, Sigma) into the intermediate zone of brain slices using a pulled glass micropipette and a microinjector (Femtojet, Eppendorf). Electroporations were performed by placing a cover square platinum plate electrode (Nepagene) onto the microinjected region. Square electric pulses
(100 V, 5 ms) were passed five times at 1-s intervals using an electroporator (ECM 830, BTX). Micropipettes and electrodes were guided using a micromanipulator (WPI) under a stereomicroscope (Stemi DV4, Carl Zeiss). All these steps were performed at 4 °C. Following electroporation, slices were transferred onto sterilized nuclepore track-etched membrane (Whatman) and cultured for 2 d in semidry conditions in a humidified incubator at 37 °C in a 5% CO2 atmosphere in wells containing Neurobasal medium supplemented with 1% B27, 1% N2, 1% glutamine and 1% penicillin/streptomycin. Quantitative analysis. For quantification of abnormal spindles, mitotic index, mitotic stages and apoptosis on HEK293, cells were counted 48 h after transfection in three independent experiments (500 cells were randomly counted in each experiment). For quantification of neuronal migration in brain slices 4 d after ex vivo electroporation, different subregions (ventricular zone/SVZ, intermediate zone and cortical plate) of the cerebral cortex were identified on the basis of cell density visualized with DAPI nuclear staining and neuronal marker (β-III-tubulin) expression staining. For each experimental condition, the number of EGFP-positive cells in the ventricular zone/SVZ, intermediate zone and cortical plate was counted in two adjacent slices from eight independent embryos. For quantification of cortical scattering in brain slices 2 d after focal electroporation and 4 d after ex vivo electroporation of NeuroD plasmids, cortices were divided into ten bins using an arbitrary scale. For each experimental condition, the number of EGFP-positive and RFP-negative or EGFP-positive cells in each bin was counted in slices from seven independent embryos. For quantification of the mitotic index, brain slices were stained with antibody to PH3 48 h after electroporation. The number of EGFP-positive and EGFPand PH3 double-positive cells in the ventricular zone/SVZ were counted in two adjacent slices from eight independent embryos. The mitotic index corresponds to the ratio of EGFP and PH3 double-positive cells (mitotic cells) to the total EGFP-positive population in the ventricular zone/SVZ. For quantification of the cell cycle exit index, brain slices were incubated with 10 µM BrdU for 1 h 24 h after electroporation. Slices were fixed and stained with antibodies to Ki67 and BrdU 24 h after the BrdU pulse. EGFP and BrdU doublepositive and EGFP, BrdU and Ki67 triple-positive cells in the ventricular zone/SVZ cells were counted in two adjacent slices from eight independent embryos. The cell cycle exit index corresponds to the percentage of EGFP-positive cells that exited the cell cycle (EGFP and BrdU double-positive, Ki67 negative) to total EGFP-positive cells that incorporated BrdU (EGFP and BrdU double-positive). Statistical analysis. Results are shown as mean ± s.e.m. Statistical analysis was performed using one-way ANOVA followed by Dunnett’s post hoc test for multiple comparisons or by a Student’s t test for double comparisons. Differences between groups were considered significant for P < 0.05. 50. Nguyen, L. et al. p27kip1 independently promotes neuronal differentiation and migration in the cerebral cortex. Genes Dev. 20, 1511–1524 (2006).doi:10.1038/
nn.2390doi:10.1038/nn.2390
nature NEUROSCIENCE
doi:10.1038/nn2390
a r t ic l e s
Epac2 induces synapse remodeling and depression and its disease-associated forms alter spines
© 2009 Nature America, Inc. All rights reserved.
Kevin M Woolfrey1,5, Deepak P Srivastava1,5, Huzefa Photowala1, Megumi Yamashita2, Maria V Barbolina3, Michael E Cahill1, Zhong Xie1, Kelly A Jones1, Lawrence A Quilliam4, Murali Prakriya2 & Peter Penzes1 Dynamic remodeling of spiny synapses is crucial for cortical circuit development, refinement and plasticity, whereas abnormal morphogenesis is associated with neuropsychiatric disorders. We found that activation of Epac2, a PKA-independent cAMP target and Rap guanine-nucleotide exchange factor (GEF), in cultured rat cortical neurons induced spine shrinkage, increased spine motility, removed synaptic GluR2/3-containing AMPA receptors and depressed excitatory transmission, whereas its inhibition promoted spine enlargement and stabilization. Epac2 was required for dopamine D1-like receptor–dependent spine shrinkage and GluR2 removal from spines. Epac2 interaction with neuroligin promoted its membrane recruitment and enhanced its GEF activity. Rare missense mutations in the EPAC2 (also known as RAPGEF4) gene, previously found in individuals with autism, affected basal and neuroligin-stimulated GEF activity, dendritic Rap signaling, synaptic protein distribution and spine morphology. Thus, we identify a previously unknown mechanism that promotes dynamic remodeling and depression of spiny synapses, disruption of which may contribute to some aspects of disease. Remodeling of central neural circuits depends on the bidirectional control of synapse stability, structure and strength. Synapse stabilization, enlargement and potentiation contribute to the establishment of long-lasting synaptic connections. On the other hand, recent imaging studies have suggested that a fraction of spines become thin and small, and display increased motility and turnover1. Such spines have reduced AMPA receptor (AMPAR) content and form weaker synapses2; spine shrinkage is associated with depressed glutamatergic transmission3,4. Synaptic dynamic remodeling therefore contri butes to neural circuit development and to the experience-dependent refinement and plasticity of brain circuits during critical periods5 and throughout life1. Conversely, abnormal synapse remodeling underlies many neuropathologies6. However, the mechanisms that actively promote coordinated spine shrinkage, increased motility and turnover, and synaptic depression, without leading to synapse elimination, collectively referred to here as destabilization, are not well understood7. Epac2 (exchange protein directly activated by cAMP, cAMP-GEFII and RapGEF4) is a signaling protein previously detected in forebrain postsynaptic densities (PSD)8,9. However, its signaling functions in central spiny synapses are not well understood. Epac2 is a GEF for Rap, a Ras-like small GTPase that, in its active form, promotes formation of thin spines and AMPAR endocytosis10–12, and is required for long-term depression (LTD), depotentiation10,13, long-term potentiation (LTP) and spatial memory storage14. Two genes, with complementary tissue distributions, encode Epac proteins, EPAC1 (also known as RAPGEF3) and EPAC2. Epac2 is highly enriched in
the brain and adrenal glands, whereas Epac1 is expressed in most non-neural tissues and is far less abundant in the adult brain15. In addition to other domains, Epac2 contains a Rap-GEF domain and two cAMP-binding domains, only one of which seems to be functional (Fig. 1a). Binding of cAMP enhances Epac’s GEF activity toward Rap16. In vitro and in non-neuronal cells, both Epac1 and 2 activate Rap1 and 2 (ref. 16). Epac proteins therefore represent a previously unknown class of PKA-independent cAMP targets15,16, linking cAMP signaling to the regulation of small GTPase function in neurons. A screen for candidate genes in the 2q21-33 autism susceptibility region identified rare non-synonymous variants in the EPAC2 gene17. These missense mutations segregated with autistic family members and were not present in a large number of unafflicted control individuals. However, it is unknown whether these mutations affect protein function or neuronal phenotypes. Neuroligins are postsynaptic adhesion molecules that bind to presynaptic neurexins. Neuroligins 3 and 4 (NLGN3 and NLGN4), as well as neurexin1 (NRXN1), have been genetically associated with autism18. Neuroligins regulate synapse morphology19 and the balance between excitatory and inhibitory synapses20,21. However, little is known about postsynaptic signaling by neuroligins. We examined the functions and regulation of Epac2 in spines. Our data indicate that Epac2 promotes spine shrinkage and enhanced turnover, resulting in synapse structural destabilization without synapse elimination. Epac2 activation also functionally depresses synapses as a result of removal of GluR2/3-containing AMPARs. Furthermore, two disease-associated Epac2 mutations altered
1Departments
of Physiology and 2Molecular Pharmacology and Biological Chemistry, Northwestern University Feinberg School of Medicine, Chicago, Illinois, USA. of Biopharmaceutical Sciences, University of Illinois, Chicago, Illinois, USA. 4Department of Biochemistry and Molecular Biology, Indiana University School of Medicine and Walther Cancer Institute, Indianapolis, Indiana, USA. 5These authors contributed equally to this work. Correspondence should be addressed to P.P. (
[email protected]). 3Department
Received 22 April; accepted 14 July; published online 6 September 2009; doi:10.1038/nn.2386
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1275
a r t ic l e s d
Epac2 RA
Rap-GEF
c
EPAC2
220 114 80 49 32 26
Epac2
EPAC1
Epac2 NR1
Epac2 PSD-95
Epac2
Epac2
Epac2
Bassoon
NR1
PSD-95
g
Epac2
ro
l
GluR2/3
nt Co
ro nt
CP
+8
γS
DP
TP G
G
f
l
Epac2 Bassoon
T
e
40
T
20 30 Ct value
CP
10
Epac2 GluR2/3
F
RPL-19
DN
0.4 0.3 0.2 0.1 0 0
Rap-GTP
p-CREB
Ral-GDS
Actin
Total Rap1
h
+8-CPT Control p–B-Raf
Control p–B-Raf
p–B-Raf
p–B-Raf
j
100 Avg. p–B-Raf intensity (a.u)
© 2009 Nature America, Inc. All rights reserved.
REM
+8
Rn
b
cAMP
+B
DEP
cAMP
Co
a
60 80 40 20
i
RNAi p–B-Raf
+8-CPT RNAi p–B-Raf
p–B-Raf
p–B-Raf
* Control +8-CPT RNAi RNAi +8-CPT
0
Figure 1 Epac2 is present in synapses in cultured cortical pyramidal neurons. (a) Domain structure of Epac2. (b) Quantitative PCR analysis of Epac1 and Epac2 mRNA in cortical neurons (28 DIV) revealed the relative enrichment of Epac2 (Rn, normalized reporter; Ct, threshold cycle). (c) Western blot detection of Epac2 in rat forebrain homogenate. (d) Localization of Epac2 and GluR2/3 in cultured cortical pyramidal neurons (28 DIV). White arrowheads indicate colocalization and green arrowheads indicate noncolocalized Epac2 puncta. (e) Double immunofluorescence with antibodies to the synaptic proteins bassoon, NR1 and PSD-95. (f) Epac2 activation by 8-CPT (50 µM, 1 h) in cortical neurons. 8-CPT increased endogenous Rap activation (1.57 ± 0.11–fold increase compared with control, n = 4, P < 0.001). (g) Specificity of 8-CPT for Epac2 in neurons. We measured the effect of 8-CPT or BDNF on CREB phosphorylation (n = 3). (h) Effect of incubation with 8-CPT (50 µM, 1 h) on the phosphorylation of the Rap target B-Raf in situ in pyramidal neuronal dendrites. (i) Effect of incubation with 8-CPT (50 µM, 1 h) on B-Raf phosphorylation in dendrites of neurons expressing Epac2 RNAi. (j) Quantification of B-Raf fluorescence intensities in h and i (n = 9–12 cells per condition, 3 experiments). * indicates P < 0.001. Error bars represent s.e.m. Scale bars represent 15 µm (d) and 5 µm (e, h, i and inset in d).
(Supplementary Figs. 1 and 2). Small puncta of Epac2 signal were also detected in axons and colocalized with tau (Supplementary Fig. 2). Thus, Epac2 may participate in protein complexes with postsynaptic proteins. Epac2 co-immunoprecipitated with the PSD scaffolding protein PSD-95 from rat forebrain homogenates (Supplementary Fig. 1), indicating that they participate in the same postsynaptic protein complexes.
RESULTS Epac2 participates in postsynaptic protein complexes Although several studies have reported an enrichment of Epac2 over Epac1 in brain15,22, some uncertainty still persists over this issue. To compare the relative abundances of Epac1 and 2 in cortical pyramidal neurons with largely stable synapses approaching maturity (28 d in vitro, DIV), we performed quantitative PCR analysis of mRNA. We found 32-fold more Epac2 mRNA than Epac1 mRNA in these neurons (Fig. 1b). Proteomic studies have detected Epac2 in forebrain postsynaptic densities8,9. To investigate its synaptic localization, we immunostained cultured cortical neurons using an antibody that detects a single protein band of ~110 kDa in rat cerebral cortex (Fig. 1c). In pyramidal neurons (28 DIV), we detected Epac2 in punctate structures along dendrites (Fig. 1d) and in the soma, suggesting that Epac2 has a functional role in dendrites. Epac2 colocalized with the synaptic markers bassoon, GluR2/3, PSD-95 and NR1, indicating that it is enriched in excitatory synapses (Fig. 1d,e). A substantial amount of Epac2 was present in synapses and spines, in addition to other subcellular compartments
Specific pharmacological activation of Epac2 in neurons Epac2 is one of two known PKA-independent cAMP targets; binding of cAMP to its C-terminal cAMP-binding domain enhances its GEF activity in vitro and in non-neuronal cells16. The cAMP analog 8-(4-chloro-phenylthio)-2′-O-methyladenosine-3′,5′-cyclic monophosphate (8-CPT) specifically activates Epac, but not PKA 16, and has been extensively used to study Epac function (Supplementary Discussion and Supplementary Fig. 3). The 8-CPT concentrations that we used were similar to those used in other cell types and to the concentration required for half-maximal activation of Epac216,23,24. Incubation of cultured cortical neurons with 8-CPT induced Rap activation (P < 0.001; Fig. 1f). Because mature cortical neurons expressed small amounts of Epac1 relative to Epac2 (Fig. 1b)15, the effects of 8-CPT on these neurons are mainly a result of Epac2 activation. Incubation with 8-CPT did not cause CREB phosphorylation (Fig. 1g), a known PKA-dependent target of cAMP16, whereas treatment with brain-derived neurotrophic factor (BDNF), a known activator of CREB, phosphorylated CREB. Incubation with 8-CPT enhanced Epac2 dendritic clustering (Supplementary Fig. 3). 8-CPT may also activate Rap1 signaling by the direct phosphorylation of Rap1 by PKA or through C3G or PDZ-GEF1 (ref. 25), but see the Supplementary Results and Supplementary Discussion.
1276
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
protein function, synaptic protein distribution and spine morphology, suggesting potential contributions to disease states.
a r t ic l e s b
RNAi
2
RNAi + rescue RNAi +8-CPT
+8-CPT
Epac2
d
*
1.5
Epac2∆GEF
0.8
Control +8-CPT RNAi RNAi + 8-CPT RNAi + rescue
0.4
f
e Control
8-CPT
0.5 0
0.4 0.3
*
0.3
*
0.2 0.1 0
0.2 0.1 0
nt 8- rol CP T Co nt r o 8CP l T
+8-CPT
GFP + Epac2∆GEF
Epac2∆GEF
*
1.0
g Extensions and retractions
© 2009 Nature America, Inc. All rights reserved.
GFP + Epac2∆GEF
Epac2
Epac2 Epac2 + 8-CPT Epac2-∆GEF Epac2-∆GEF + 8-CPT
Co
GFP + Epac2
Total motility
+8-CPT
2
GFP + Epac2
*
*
0
Spine area (µm )
c
Rescue Epac2
* 1.2
Control – 60’
– 50’
– 40’
0’
10’
20’
– 30’
– 20’
– 10’
30’
40’
50’
0’
+8-cpt
h
Percentage change in spine area
Control
Spine area (µm )
a
60’ +8-CPT
Figure 2 Epac2 activation induces dendritic spine shrinkage, reduces presynaptic contact and enhances spine 100 motility and turnover. (a) Effect of incubation with 8-CPT (50 µM, 1 h) in absence or presence of Epac2 RNAi or ** 90 rescue RNAi on spine morphology. (b) Quantification of average spine areas in a (area: control, 0.92 ± 0.04 µm2; 8-CPT, 0.68 ± 0.04; Epac2 RNAi, 1.10 ± 0.05; Epac2 RNAi + 8-CPT, 1.04 ± 0.07; Epac2 RNAi + rescue, 0.87 ± 80 0.03; * indicates P < 0.001, n = 102–252 spines, 5–10 cells per condition, 3 experiments; see also Supplementary –60 0 60 Fig. 7). (c) Epac2 lacking the GEF domain (Epac2-∆GEF) prevented 8-CPT–induced spine shrinkage. Time (min) 2 (d) Quantification of c (area: Epac2, 0.88 ± 0.05 µm ; Epac2 + 8-CPT, 0.72 ± 0.03; Epac2∆GEF, 1.24 ± 0.06; Epac2-∆GEF + 8-CPT, 1.17 ± 0.05; n = 169–274 spines, 5–9 cells per condition, 3 experiments). All neurons were analyzed at 28 DIV. (e) Time-lapse imaging of spine dynamics in GFP-expressing cortical pyramidal neurons (25 DIV) pretreated with or without 8-CPT (50 µM). Spine dynamics were visualized at the beginning, middle and end of 80-min imaging sessions (red, retracting; green, transient; blue, newly extended). (f) Quantification of total spine motility, expressed as the fraction of spines undergoing extension, retraction or head morphing, and the fraction of spines undergoing extensions or retractions are shown (normalized total motility: control, 0.21 ± 0.02; 8-CPT, 0.34 ± 0.01; n = 1,218 spines, 5 cells per condition). (g) Example of time-lapse imaging of an individual spine before and after 8-CPT (50 µM, 1 h) incubation; spine shrank following 8-CPT treatment. (h) Quantification of g (−60 min, 100%; 0 min, 97.9 ± 3.0%; 60 min, 87.5 ± 2.4%; ** indicates P < 0.01, n = 84 spines, 3 experiments). Error bars represent s.e.m. Scale bars represent 5 µm (a,c,e) and 2.5 µm (g).
Epac2 activates Rap and causes spine shrinkage To determine whether Epac2 activation stimulates Rap signaling in dendrites in situ, we examined the effect of 8-CPT on the phosphorylation of a known Rap target, B-Raf25,26. 8-CPT incubation significantly increased the amount of dendritic B-Raf that was phosphorylated (P < 0.001; Fig. 1h–j). This effect was not Ras dependent, as it was not blocked by the Ras inhibitor FTase II (Supplementary Fig. 4). To confirm that 8-CPT signals through Epac2, we used RNA interference (RNAi) to knockdown endogenous Epac2. We tested the specificity of the RNAi for Epac2 in HEK293 cells and in neurons (Supplementary Figs. 5 and 6). Epac2 knockdown prevented B-Raf phosphorylation in response to 8-CPT treatment (Fig. 1i–j), demonstrating Epac2 dependence and specificity. To determine whether activation of endogenous Epac2 caused structural modifications in spines, we incubated mature green fluorescent protein (GFP)-expressing cultured pyramidal neurons (28 DIV) with 8-CPT (50 µM, 1 h). This treatment induced shrinkage of existing spines, as seen by a reduction in average spine area (P < 0.001 Fig. 2a,b), breadth and breadth/length ratios, without affecting spine linear density (Supplementary Fig. 7). Incubation with 8-CPT for 24 h did not affect spine density. To determine the requirement for Epac2 in basal and 8-CPT– induced spine morphology, we knocked down endogenous Epac2.
Neurons were transfected with Epac2-specific shRNA that coexpressed GFP; in RNAi-expressing neurons, 8-CPT was incapable of inducing spine shrinkage, demonstrating that normal levels of Epac2 expression are required for 8-CPT–dependent spine shrinkage (Fig. 2a,b). Epac2 knockdown caused an increase in the average basal spine area (P < 0.001). This effect was rescued by overexpression of an Epac2 rescue mutant, in which three silent point mutations were introduced to render it insensitive to RNAi, demonstrating the specificity of the knockdown (Fig. 2a,b). RNAi knockdown of Epac2 did not affect spine density (Supplementary Fig. 6). The effects of Epac2 activation on spine morphology were specifically dependent on Epac2 Rap-GEF activity, as 8-CPT did not induce spine shrinkage in neurons expressing an Epac2 mutant protein lacking the catalytic portion of the Rap-GEF domain (Epac2∆GEF; Fig. 2c,d). These neurons had significantly larger spines than those overexpressing wild-type Epac2 (P < 0.001). The effects of Epac2 activation were occluded by overexpression of dominantnegative Rap1 and Rap2 (Rap1-DN and Rap2-DN; Supplementary Fig. 7). As expected, Epac2 overexpression did not alter spine area (Supplementary Fig. 6), as the default state of Epac2 is autoinhibitory and requires activation to activate Rap16. To determine whether 8-CPT affects the morphology of presynaptic terminals, we visualized presynaptic active zones with an antibody
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1277
a
b
Control
In
pu t G lu R1 G lu R M 2/3 yc
IP
+8-CPT
Epac2 GluR1
GFP GluR2/3
GFP
GluR2/3
GFP GluR2/3
GFP
GluR2/3
GFP
GluR2/3
GFP
GluR2/3
RNAi GluR2/3 GFP GluR2/3
IgG +8-CPT
RNAi
GFP GluR2/3
c
4 GluR2/3 shaft intensity (a.u.)
GluR2/3 spine intensity (a.u.)
3
**
2 1 0
l tro
on
C
d
2 1 0
Ai N T R P 2 c2 8-C + c a pa + Ep E PT
C 8-
*
3
Ai
N
R
l tro
on
C
e
PT
-C
+8
Ep
2 ac
Ai N T R P 2 8-C c pa + Ai
N
R
E
Control
+ 8-CPT
GluR2/3 Bassoon
GluR2/3 Bassoon
GluR1
Figure 3 Epac2 interacts with GluR2/3-containing AMPAR and removes them from spines. (a) Co-immunoprecipitation (IP) of Epac2 with GluR2/3, but not with GluR1, from cortical neurons (28 DIV). Antibody to Myc was used as a control. (b) Effects of 8-CPT (50 µM, 1 h) and Epac2 RNAi knockdown on GluR2/3 content in spine heads. GluR2/3 clusters were visualized in spines outlined by GFP (arrowhead, clusters in spines; open arrowhead, shafts). (c) Quantification of the effects in b on GluR2/3 signal intensity in spines (top) and shaft (bottom) (GluR2/3 immunofluorescence: control, 2.94 ± 0.27 a.u.; 8-CPT, 1.78 ± 0.13; Epac2 RNAi, 2.81 ± 0.37; Epac2 RNAi + 8-CPT, 2.61 ± 0.24; * indicates P < 0.001 and ** indicates P < 0.01, n = 10–14 cells, 3 experiments). (d) GluR1 and NR1 cluster intensity was not affected (GluR1 immunofluorescence: control, 1.29 ± 0.12 a.u.; 8-CPT, 1.08 ± 0.11; NR1 immunofluorescence: control, 0.95 ± 0.04 a.u.; 8-CPT, 1.05 ± 0.08; n = 8–14 cells per condition). (e) Effect of 8-CPT on GluR2/3 colocalization with bassoon (arrowhead, clusters on spines; open arrowhead, shafts) The percentage of GluR2/3 puncta expressing bassoon was 0.89 ± 0.02 for controls and 0.80 ± 0.03 for 8-CPT–treated cells (P < 0.05, n = 17 cells per condition). Error bars represent s.e.m. Scale bars represent 5 µm.
Control
+ 8-CPT
to bassoon. Epac2 activation significantly reduced the extent of presynaptic overlap GluR2/3 GluR2/3 Control + 8-CPT with spines, as revealed by quantification of the intensity of bassoon immunofluorescence overlapping with individual spines Bassoon Bassoon (P < 0.001; Supplementary Fig. 7), suggesting that pre-postsynaptic apposition was reduced and synapses were weaker. These effects on spine/bassoon overlap were and after 8-CPT perfusion further confirmed Epac2-dependent spine occluded by Epac2 RNAi (Supplementary Fig. 7). Incubation with shrinkage (P < 0.01; Fig. 2g,h). These data indicate that Epac2 acti8-CPT also caused a reduction in bassoon-immunoreactive cluster vation causes structural destabilization of spines resulting in overall size (Supplementary Fig. 7), indicating a potential presynaptic effect. increased motility. These data provide evidence that Epac2 activation induces shrinkage of spines and reduction of presynaptic contacts and that these effects Epac2 activation promotes AMPAR removal from spines As spine morphology, stability and function are coordinated2, Epac2 require Epac2’s Rap-GEF activity. may also regulate glutamate receptor function in synapses. In cortical Epac2 activation enhances spine motility neurons, Epac2 co-immunoprecipitated with GluR2/3 AMPAR sub In vivo studies have revealed that smaller and thinner spines, units, but not with GluR1 (Fig. 3a). We observed substantial colocalimorphologically resembling those induced by Epac2 activation, are zation between Epac2 and GluR2/3 in spines and dendrites (Fig. 1d, very dynamic and undergo rapid remodeling, whereas large spines quantified in Supplementary Fig. 1). This interaction places Epac2 are stable1. To test whether Epac2 activation affects spine motility in proximity to a subset of AMPARs, potentially allowing for their and turnover, we performed time-lapse imaging of GFP-expressing rapid regulation. To determine whether GluR2/3 content in spines was altered by cortical neurons (Fig. 2e–g). As indicated by color-coded and overlaid images taken at three equally separated time points during the Epac2 activation, we measured the integrated intensity of GluR2/3 imaging session, spines in control neurons were largely unchanged immunofluorescence signals in individual spines (Fig. 3b,c). over the imaging period (80 min), undergoing only limited Following 8-CPT treatment, the amount of GluR2/3 was significantly reduced in spines (P < 0.01) and increased in shaft clusters (P < 0.001; remodeling (morphing). In 8-CPT–treated neurons, extensive spine morphing, motility Fig. 3b,c). These effects were specifically dependent on Epac2; RNAiand turnover were detectable (Fig. 2e); spines retracted, formed mediated knockdown prevented 8-CPT–induced GluR2/3 removal or were transient (Supplementary Fig. 8). Quantification of total from spines (Fig. 3c). RNAi expression did not alter the amount of spine motility demonstrated a ~62% increase in motility following GluR2/3 present in spines, suggesting that Epac2 activation is only Epac2 activation (P < 0.001; Fig. 2f). 8-CPT did not increase spine involved in GluR2/3 removal from synapses. GluR1 or NR1 cluster motility in Epac2-RNAi–expressing neurons, demonstrating that intensities were not affected by 8-CPT (Fig. 3d). Incubation with Epac2 is required for 8-CPT–dependent increased spine motility 8-CPT also resulted in a significant reduction in GluR2/3-bassoon (Supplementary Fig. 8). Time-lapse imaging of the same spine before overlap (P < 0.05; Fig. 3e) and in a trend toward reduced numbers NR1
© 2009 Nature America, Inc. All rights reserved.
a r t ic l e s
1278
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
a mEPSC
Control
+8-CPT
amplitude (pA)
20
**
10
Control 0
mEPSC frequency (events per s)
RNAi
RNAi + 8-CPT 10 pA
20 pA
10 ms
1s
10
80
Control +8-CPT
20 0
10
20
30
40
50
60
RNAi
40
RNAi + 8-CPT
20
>60
0
10
20
30
40
50
>60
mEPSC amplitude (pA)
mEPSC amplitude (pA) Pretreatment
+8-CPT 10 pA
20 pA
10 ms
1s
Percentage control mEPSC amplitude (pA)
b
*
*
0
100
40
RNAi RNAi + 8-CPT
80 60
+8-CPT
20
100 Cumulative probability
Cumulative probability
100
0 Pretreatment
80 60 40
**
50
100
Pretreatment +8-CPT
Percentage control mEPSC frequency (events per s)
Figure 4 Epac2 activation depresses AMPARmediated synaptic transmission. (a) Effect of 8-CPT on AMPAR-mediated mEPSC amplitudes and frequency in pyramidal neurons (28 DIV). Synaptic currents were recorded in single cells pretreated with vehicle or 8-CPT (50 µM, 1 h). Traces show representative recordings. Bar graphs show quantification of mean amplitudes (control, 13.85 ± 2.12 pA; 8-CPT, 9.19 ± 0.53; ** indicates P < 0.05) and frequency (control, 13.40 ± 2.00 events per s; 8-CPT, 5.55 ± 1.06; * indicates P < 0.01) of AMPAR-mediated mEPSCs (n = 9–11 cells per condition). Epac2 RNAi blocked the 8-CPT–induced decrease of the AMPAR-mediated mEPSC amplitude (RNAi + vehicle, 12.54 ± 0.98 pA; RNAi + 8-CPT, 12.74 ± 1.80), but not of the mEPSC frequency (RNAi + vehicle, 12.98 ± 1.09 events per s; RNAi + 8-CPT, 6.70 ± 0.67; n = 5 cells per condition). Cumulative probability plots show a shift of mEPSC amplitudes toward smaller values in response to 8-CPT treatment (left), whereas we found no difference in mEPSC amplitude distribution in Epac2 RNAi expressing cells (right). (b) Synaptic currents were recorded in single cells before (gray) and 10–15 min after (black) perfusion with 8-CPT (50 µM). This resulted in a rapid reduction of mean amplitude of AMPARmediated mEPSCs (post 8-CPT treatment, −16.51 ± 3.65% relative to control mean pA, ** indicates P < 0.05) but not frequency (post 8-CPT, +13.84 ± 12.99% relative to control events per s). Insets show mEPSC rise and decay time (n = 3). Error bars represent s.e.m.
Cumulative probability
© 2009 Nature America, Inc. All rights reserved.
a r t ic l e s
+8-CPT 100
20 of puncta with colocalized GluR2/3 and bas50 soon (Supplementary Fig. 7). Thus, in addi0 10 20 30 40 >50 0 tion to destabilizing spines, Epac2 activation mEPSC amplitude (pA) also removes GluR2/3-containing AMPARs from synapses, potentially leading to weaker synaptic connections. in spines. Activation of Epac2 also caused a reduction in mEPSC frequency, suggesting that it has potential presynaptic effects. Epac2 activation depresses excitatory transmission To determine the functional outcome of this reduction in GluR2/3 Epac2 mediates dopamine-dependent synaptic remodeling content in spines, we examined the effects of 8-CPT on AMPAR- We next examined the upstream mechanisms regulating Epac2 in dependent miniature excitatory postsynaptic currents (mEPSCs) cortical pyramidal neurons. Epac2 is one of the few PKA-independent (Fig. 4). Incubation of neurons with 8-CPT (50 µM, 1 h) resulted targets of cAMP. In neurons, cAMP levels are elevated by dopamine in a robust reduction in the mean amplitudes (34%) and frequency activation of the D1/D5 G protein–coupled receptor (DAR-D1/D5); (59%) of AMPAR-mediated mEPSCs (P < 0.05; Fig. 4a), resulting in a in non-neuronal cells, DAR-D1/D5 activation enhances Epac activity. shift in the distribution of mEPSC amplitudes toward smaller values. Therefore, we tested whether Rap1 activity in neurons was enhanced RNAi knockdown of Epac2 in postsynaptic neurons prevented 8-CPT by activation of DAR-D1/D5. Incubation of neurons with the D1/D5from reducing the basal mean mEPSC amplitude (Fig. 4a), confirm- specific agonist SKF-38393 (20 µM, 30 min) induced an enhanceing the Epac2 specificity of the 8-CPT effect. RNAi expression did not ment of Rap1 activity that was comparable to that caused by 8-CPT increase the mean amplitude or frequency of AMPAR mEPSCs, con- (P < 0.05; Fig. 5a). To determine whether DAR-D1/D5 activation sistent with our immunostaining data. Notably, the presence of RNAi stimulated Rap1 activity in dendrites, we examined the effect of SKFin the postsynaptic cell did not affect the 8-CPT–dependent reduction 38393 on B-Raf phosphorylation (Fig. 5b). Incubation of neurons in mEPSC frequency (P < 0.01), suggesting that these effects were with SKF-38393 significantly increased B-Raf phosphorylation in situ caused by activation of presynaptic Epac2, dissociating the pre- and in dendrites (P = 0.001). postsynaptic actions of Epac2. DAR-D1/D5 signaling has not yet been investigated with regards Perfusion of neurons with 8-CPT also resulted in a rapid (~10 min), to spine plasticity. Incubation of neurons with SKF-38393 (20 µM, significant reduction of the mean amplitude of AMPAR mEPSCs 30 min) caused a reduction in spine areas (P < 0.001; Fig. 5c,d). In (P < 0.05; Fig. 4b), without significantly affecting mean frequency the presence of Epac2 RNAi, the effect of SKF-38393 on dendritic (P > 0.05). For all manipulations, the rise and decay time of mEP- spines was occluded (Fig. 5c,d); treatment with SKF-38393 did SCs were not affected. These results indicate that postsynaptic Epac2 not affect linear density (Supplementary Fig. 6). Furthermore, activation rapidly and robustly depresses AMPAR-mediated synaptic SKF-38393 caused a significant reduction in the amount of surface transmission by reducing GluR2/3-containing AMPAR content GluR2 on spines (P < 0.001; Fig. 5c,e). This effect required Epac2,
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1279
a
l
tro
n Co
93
83
-3
F SK
b +SKF-38393 p–B-Raf Map2
p–B-Raf Map2
p–B-Raf
p–B-Raf
Rap-GTP
Ral-GDS
Total Rap1
c
d
Control
GFP GluR2-n
Spine area (µm2)
Figure 5 Dopamine D1/D5-like receptors modulate Rap activity, spine morphology and GluR2 surface expression. (a) Rap activation by SKF-38393 (20 µM, 30 min) in cortical neurons. SKF-38393 increased Rap activation compared with controls (1.62 ± 08–fold increase, P < 0.05, n = 4 experiments). (b) Effect of SKF-38393 (20 µM, 30 min) on B-Raf phosphorylation in situ in dendrites (B-Raf immunofluorescence: control, 68.52 ± 4.6 a.u.; SKF-38393, 141.55 ± 15.2; P < 0.001, n = 6–9 cells per condition, 2–3 experiments). (c) Effect of SKF-38393 (20 µM, 30 min) on spine morphology and surface GluR2 in spines in cortical neurons (28 DIV). Epac2 knockdown prevented SKF-38393–dependent spine remodeling and AMPAR removal. GluR2 was detected using an antibody to its extracellular N terminus (GluR2-n) in nonpermeabilized cells (arrowhead, clusters in spines; open arrowhead, shafts). (d) Quantification of spine areas in c (area: control, 0.74 ± 0.02 µm2; SKF-38393, 0.59 ± 0.01; Epac2 RNAi, 0.86 ± 0.03; Epac RNAi + SKF-38393, 0.90 ± 0.04; * indicates P < 0.001, n = 308–426 spines from 9–13 cells per condition, 4 experiments). (e) Quantification of surface GluR2 (GluR2-n) clusters in c (GluR2-n immunofluorescence: control, 60.6 ± 3.53 a.u.; SKF-38393, 36.9 ± 1.65; Epac2 RNAi, 67.7 ± 4.48; Epac2 RNAi + SKF-38393, 59.6 ± 3.94; n = 9–13 cells per condition, 4 experiments). Error bars represent s.e.m. Scale bars represent 5 µm.
GluR2-n
Control + SKF38393
GFP GluR2-n
GluR2-n
as it did not occur when Epac2 was knocked down. Together, these data indicate that DAR-D1/D5 activation induces Rap1 activation, as well as Epac2-dependent spine shrinkage and GluR2 removal from spines.
GluR2-n
*
0.9 0.6 0.3 0 Control +SKF-38393 RNAi RNAi + SKF-38393
e
RNAi + SKF38393
GFP GluR2-n
*
GluR2-n
RNAi
GFP GluR2-n
1.2
GluR2-n intensity in spines (a.u.)
© 2009 Nature America, Inc. All rights reserved.
a r t ic l e s
80
*
60 40 20 0
8-CPT had more puncta with colocalized Epac2 and NL3 (Fig. 6f), an effect that was driven by an increased number of Epac2-immuno reactive puncta (Supplementary Fig. 3). The number of NL3-positive puncta was not affected (Fig. 6f and Supplementary Fig. 9). Because Epac2 translocation to the plasma membrane is associated with its activation28, we reasoned that NL3 may activate Epac2. Indeed, coexpression of Epac2 with NL3 robustly enhanced its Rap-GEF activity (P < 0.001; Fig. 6g,h). Collectively these data suggest that Epac2 and NL1/3 form protein complexes in neurons, NL3 is capable of recruiting Epac2 to the plasma-membrane, Epac2’s proximity to NL3 is modulated in parallel with Epac2 activation and NL3 enhances Epac2’s Rap-GEF activity.
Epac2 complexes with neuroligins Neuroligins are a class of synaptic adhesion molecules that modulate synapse morphology and excitatory/inhibitory synapse balance20. NL3 and 4 have also been genetically associated with autism27. We reasoned that Epac2 may participate in protein complexes with neuroligins. Using co-immunoprecipitation in cortical neurons, we found that NL3 and 1 strongly and specifically interacted with Epac2 (Fig. 6a). Epac2 also interacted with NL2, albeit more weakly than with NL3 or 1. Epac2 did not interact with N-cadherin, another synaptic adhesion molecule. Reverse co-immunoprecipitation confirmed the interaction of Epac2 with NL1 and 3 (Fig. 6b). Epac2 immunoprecipitated NL1 and 3, and to a lesser extent NL2, from rat cortex (Fig. 6c). This interaction was enhanced by 8-CPT (Supplementary Fig. 9). Hemagglutinin (HA)-tagged Epac2 immunoprecipitated NL3 or PSD-95 when these proteins were overexpressed in HEK293 cells, suggesting that they are members of a complex (Supplementary Fig. 9). Consistent with this, Epac2 colocalized with NL1 and 3 in a large fraction of spine-like structures (Fig. 6d and Supplementary Fig. 1). We hypothesized that neuroligins could potentially recruit Epac2 to the plasma membrane. To test this, we used COS7 cells. When expressed alone, GFP-Epac2 was diffusely distributed in the cytosol, whereas HA-NL3 was found at the plasma membrane (Fig. 6e). In contrast, when coexpressed with NL3, a fraction of Epac2 was recruited to the plasma membrane (Fig. 6e). Similarly, overexpression of NL3 enhanced Epac2 localization to the membrane in dendrites (Supplementary Fig. 9). We next sought to determine whether Epac2 activation altered Epac2 colocalization with NL3 in neurons. Neurons treated with
Disease-associated mutants of Epac2 affect Rap signaling Four rare coding mutations in Epac2 (M165T, V646F, G706R and T809S) have been genetically associated with autism17 (Supplementary Fig. 10). These variants strictly segregate with a utistic family members and are not present in unafflicted individuals. To test whether these mutations affected protein function and synapse morphology, we generated point mutants of the Epac2 protein that corresponded to these variants (Supplementary Fig. 10). We first examined whether the Epac2 mutations affected its Rap-GEF activity by measuring Rap1-GTP in HEK293 cells t ransfected with either mutant or wild-type Epac2. The Epac2V646F mutant impaired the Rap-GEF activity of Epac2 (P = 0.01; Fig. 7a,b). Because NL3 enhanced Epac2 activity, we examined the effects the Epac2 mutants on NL3-dependent stimulation of Epac2’s GEF activity (Fig. 7c,d). Coexpression of NL3 with Epac2-V646F resulted in reduced Rap activation, similar to the effect of this mutation on basal Rap-GEF activity. Conversely, coexpression of NL3 with Epac2-T809S increased its Rap-GEF activity (P < 0.001). We next examined the effects of Epac2 mutations on dendritic B-Raf phosphorylation (Fig. 7e,f). Consistent with our biochemical data, expression
1280
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
a r t ic l e s
NL3
Epac2
NL3
NL2
NL1
NL1
NL1
NL2
NL2
Epac2
NL3
Epac2
PSD95
IgG
IgG
IgG
N-cad
NL3
Epac2
Epac2 Ep ac 2 Ep ac NL 2 + 3
g
Co nt ro l NL 3
+8-CPT
γS
f
DP
GFP-Epac2 + HA-NL3
NL1
TP
HA-NL3
Epac2 NL3
G
GFP-Epac2
Epac2 NL1
Rat cortex
N-Cad
e
d
IP
G
yc M
In pu t N L1
N L3
1 ac Ep
2 ac
ut
M yc
Ep
In p
c
IP
In pu t Ep ac 2 M yc
b
IP
N L2
a
Rap-GTP Ral-GDS Epac2 NL3
Epac2 NL3
Total Rap1
NL3
NL3
4
*
2
2 ac 2 N + L3 Ep
tro on C
L3
0 ac
Epac2
l
Epac2
*
6
Ep
h
N
NL3
Fold Rap activation
© 2009 Nature America, Inc. All rights reserved.
Epac2
Figure 6 Epac2 interacts with neuroligins. (a) Co-immunoprecipitation of NL1–3 with Epac2, but not with Epac1, from cortical neurons (28 DIV). (b) Reverse co-immunoprecipitation of NL1–3 with Epac2 from cortical neurons (28 DIV). (c) Co-immunoprecipitation of Epac2 with NL1–3 from rat forebrain. We used antibody to Myc as a control. All co-immunoprecipitation experiments were performed three times and the western blots show typical results. (d) Immunofluorescence colocalization of Epac2 and NL1 and 3 on dendrites of cortical pyramidal neurons. (e) NL3 affected Epac2 localization. Epitope-tagged Epac2 and NL3 were expressed individually or together in COS7 cells and visualized by immunostaining (open arrowhead, cytosolic expression; arrowhead, membrane expression). (f) Treatment with 8-CPT (50 µM) increased Epac2/NL3 colocalization along dendrites of mature cortical neurons (control, 7.69 ± 0.21 puncta with colocalized Epac2 and NL3; 8-CPT, 11.53 ± 0.58; P < 0.001; n = 7–8 cells per condition, 2–3 experiments). (g) Coexpression of Epac2 with NL3 in HEK293 cells enhanced Epac2’s Rap-GEF activity. (h) Quantification of Rap-GTP in g (fold increase in Rap activity relative to control: NL3, 1.08 ± 0.23; Epac2, 2.61 ± 0.19; NL3 + Epac2, 6.45 ± 0.08; * indicates P < 0.001, n = 3 experiments). Error bars represent s.e.m. Scale bars represent 5 µm (d,f) and 15 µm (e).
of Epac2-V646F reduced dendritic phospho–B-Raf immunofluorescence, indicative of reduced Rap signaling in dendrites. Conversely, Epac2-T809S increased dendritic phospho–B-Raf immunofluorescence, indicative of increased Rap signaling in dendrites (P < 0.001). The M165T and G706R mutations did not affect Epac2 function. Disease-associated mutants of Epac2 alter spine morphology Because they affected protein function, we hypothesized that some of the Epac2 mutations may affect spine morphology. Overexpression of two of the mutants in neurons (28 DIV) affected spine morphology (Fig. 8a,b). Epac2-V646F caused a significant increase in average spine area (P < 0.001; Fig. 8b). Epac2-T809S increased the spine average linear density (P < 0.001). To gain insight into the potential mechanisms underlying the spine morphological changes, we examined the effects of Epac2 mutations on the average intensity and number of PSD-95 immuno fluorescent puncta in dendrites (Fig. 8c,d). Overexpression of PSD-95 has previously been shown to induce synaptogenesis and increased spine number 7 . Expression of Epac2-V646F increased the average PSD-95 immunofluorescence intensity in individual puncta, consistent with its effect on spine size (P < 0.001). Conversely, Epac2-T809S increased the number of PSD-95 clusters along dendrites, consistent with its effect on spine nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
density (P < 0.001). These data indicate that the Epac2-V646F and Epac2-T809S mutants affect Epac2 protein function and dendritic spine morphology. DISCUSSION Our results identify a previously unknown mechanism that promotes the dynamic remodeling of spiny synapses on cortical pyramidal neurons, actively destabilizing spines by promoting their shrinkage and increasing their motility, as well as removing GluR2/3-containing AMPAR and depressing glutamatergic transmission (Supplementary Fig. 11). Similar effects were caused by the stimulation of DAR-D1/D5s; these effects were prevented by post synaptic knockdown of Epac2. Notably, two disease-associated Epac2 mutations affected protein function and dendritic Rap signaling and their expression in pyramidal neurons affected spine morphology and PSD-95 clustering. Because synapse dynamic remodeling is important in the maturation, refinement and plasticity of brain circuits, and may malfunction in some disorders, our results suggest that Epac2 is involved in normal and pathological brain plasticity. Epac2 activation enhances spine dynamic remodeling, consisting of spine shrinkage, turnover, head morphing and reduced overlap with presynaptic release sites (Supplementary Fig. 11). Conversely, Epac2 inactivation causes spine enlargement and stabilization. 1281
c
Ral-GDS
Epac2
Total Rap1
*
Control Epac2 Epac2-M165T Epac2-V646F Epac2-G706R Epac2-T809S
NL3
d
p-B-Raf
p-B-Raf
p-B-Raf HA
p-B-Raf
p-B-Raf HA
16
8 4
p-B-Raf
p-B-Raf HA
p-B-Raf HA
**
0 Control Epac2 + NL3 Epac2-M165T + NL3 Epac2-V646F + NL3 Epac2-G706R + NL3 Epac2-T809S + NL3
f
80 Average p–B-Raf intensity (a.u)
Epac2 -M165T Epac2 -V646F
p-B-Raf
**
12
Epac2 -WT
Control
e
2
0
Rap-GTP
Total Rap1
Epac2
3
1
Ral-GDS
Fold Rap activation
Fold Rap activation
Rap-GTP
b
+ NL3 G TP γS G DP Co nt ro Ep l ac 2 Ep ac 2 pa - M1 6 c2 -V 5TE Ep 64 6F ac 2G 70 Ep 6R ac 2T8 09 S
G TP γS G DP Ep ac 2 Ep ac 2M 16 Ep 5T ac 2V6 Ep 46 ac F 2G 70 Ep 6R ac 2T8 09 S
a
Epac2 -G706R
Figure 7 Disease-associated missense mutations affect Epac2 function. (a) Epac2 mutations affected Rap-GEF activity in HEK293 cells transfected with either wild-type (Epac2-WT) or mutant Epac2. (b) Quantification of Rap1-GTP in a. Rap1 activation by Epac2-V646F was reduced (fold increase in Rap activity relative to control: Epac2-WT, 2.67 ± 0.40; Epac2-M165T, 2.52 ± 0.50; Epac2-V646F, 0.55 ± 0.17; Epac2G706R, 2.62 ± 0.45; Epac2-T809S, 1.80 ± 0.43; * indicates P < 0.01, n = 3 experiments). (c) Effect of Epac2’s missense mutations on NL3-dependent stimulation of its GEF activity. (d) Quantification of Rap1-GTP in c. NL3enhanced Rap activation was reduced in cells transfected with Epac2-V646F and increased in cells transfected with Epac2-T809S (fold increase in Rap activity relative to control: Epac2-WT, 5.98 ± 0.24; Epac2-M165T, 5.56 ± 0.76; Epac2-V646F, 3.26 ± 0.38; Epac2-G706R, 4.93 ± 0.50; Epac2-T809S, 15.21 ± 2.32; ** indicates P < 0.001, n = 3 experiments). (e) Effect of Epac2 mutations on dendritic B-Raf phosphorylation (p–B-Raf) in cortical pyramidal neurons (28 DIV). (f) Quantification of dendritic B-Raf fluorescence intensity in e. Epac2-V646F reduced and Epac2-T809S increased dendritic p–B-Raf immunofluorescence (control, 232.2 ± 11.4 a.u.; Epac2-WT, 272.2 ± 17.3; Epac2M165T, 265.3 ± 34.2; Epac2-V646F, 145.7 ± 10.5; Epac2-G706R, 255.6 ± 32.0; Epac2T809S, 528.1 ± 49.5; n = 6–8 cells per condition from 3 experiments). Error bars represent s.e.m. Scale bar represents 5 µm.
**
60 40 20
**
0 Control Epac2 Epac2-M165T Epac2-V646F Epac2-G706R Epac2-T809S
However, Epac2 does not promote synapse elimination. Spine shrinkage occurs during p-B-Raf various forms of brain plasticity, including hippocampal LTD3,4. Small, thin and highly dynamic spines are important for the development, refinement and experience-dependent plasticity of cortical neuronal circuits1,5. Although filopodia in young neurons are highly motile, spines with mature morphologies are more stable. Even in the mature cortex, however, there is a steady state of spine morphing and turnover, modulated by physiological stimuli such as experience1,2. Epac2 may promote the dynamic remodeling of such otherwise stable spines. Few molecules, on activation, are known to promote spine shrinkage and increased motility without elimination7 (Supplementary Discussion). Epac2 activation caused a reduction in the amount of AMPARs in spines and reduced the amplitude and frequency of AMPAR-mediated mEPSCs, indicating that Epac2 also depresses glutamatergic transmission. Epac2 specifically regulates GluR2/3, probably as a result of the participation of Epac2, along with PSD-95, in protein complexes with GluR2/3. In these complexes, Epac2 can respond to stimuli and activate Rap, which in turn diffuses to nearby GluR2/3 receptors and triggers their internalization. Our data suggest that Epac2 activation selectively removes GluR2/3 subunit–containing AMPARs from synapses; the remaining synaptic AMPARs may consist of a larger fraction of GluR1/ GluR2-containing receptors and fewer GluR2/GluR3-containing receptors. As GluR2/3 is removed from spines, the total amount of functional AMPARs is reduced, leading to reduced AMPAR mEPSC amplitudes. Epac2-dependent changes in synaptic GluR2/3 content may contribute to several types of plasticity29,30. A range of effects of 8-CPT incubation have been reported in different neuronal preparations. Short-term and transient presynaptic potentiation has been reported
in invertebrate neuromuscular junctions31,32, the calyx of Held, and in young hippocampal and cortical neurons33 (Supplementary Discussion). Postsynaptically, from 8-CPT responsiveness, recent studies found PACAP-, protein synthesis– and ERK-dependent LTD34, and facilitation of βAR-, HFS-, protein synthesis– and ERK-dependent LTP, without affecting LTP induction35. One potential explanation for these diverging effects is that transient destabilization could make synapses more receptive to subsequent activity-dependent potentiating or depressing stimuli, leading to LTP or LTD, respectively12. Modulation of synapses by Epac proteins may also affect cognitive functions and behavior. Epac and PKA are jointly required for hippocampal memory retrieval36, and Epac activation with 8-CPT rescued psychiatric disease–related deficits in sensory motor gating and memory caused by overactivation of Gαs signaling37. cAMP signaling is important in synaptic plasticity 38, learning, memory, psychiatric disease and drug addiction. Most previous work on cAMP signaling in pyramidal neurons has focused on its actions through PKA. However, several studies have reported that postsynaptic cAMP-dependent, but PKA-independent, mechanisms induce LTD, depress basal synaptic transmission and reverse potentiation39. In dendrites, cAMP is produced on activation of Gs-coupled receptors, such as D1-like receptors. Little is known about the mechanisms of regulation of spine morphology by dopamine signaling. We found that DAR-D1/D5 activation stimulated dendritic Rap signaling, causing Epac2-dependent spine shrinkage and GluR2 removal. In cortical pyramidal neurons, DAR-D1/D5s
1282
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
Epac2 -T809S
© 2009 Nature America, Inc. All rights reserved.
a r t ic l e s
p-B-Raf HA
GFP
GFP + Epac2-V646F
GFP
b
GFP + HA
GFP + Epac2-WT
0.6 0.3 0
GFP + Epac2-G706R GFP + HA
GFP + Epac2-M165T GFP + HA
*
0.9
Spine density (spines per 10 µm)
GFP + HA
Spine area (µm2)
a
GFP + Epac2-T809S GFP + HA
*
9 6 3 0 GFP Epac2 Epac2-M165T
PSD-95
PSD-95 + HA
PSD-95
PSD-95 + HA
PSD-95 cluster intensity (a.u.)
4
Epac2 -WT Epac2 -M165T
Epac2-V646F Epac2-G706R Epac2-T809S
d 3
*
2 1 0
PSD-95 + HA
Epac2 -V646F
PSD-95
PSD-95
PSD-95 + HA
PSD-95 clusters per 10 µm
c
Epac2 -G706R
Figure 8 Epac2 missense mutants affect spine morphology. (a) Coexpression of HA-tagged Epac2 or its disease-associated mutants with GFP in cortical pyramidal neurons (28 DIV). Epac2V646F and Epac2-T809S altered dendritic spine morphology. (b) Quantification of the effects on spine area and number in a (area: GFP, 0.67 ± 0.02 µm2; Epac2-WT, 0.66 ± 0.02; Epac2M165T, 0.70 ± 0.02; Epac2-V646F, 0.82 ± 0.02; Epac2-G706R, 0.67 ± 0.02; Epac2-T809S, 0.66 ± 0.02; * indicates P < 0.001; number of spines per 10 µm: GFP, 5.98 ± 0.37; Epac2-WT, 7.00 ± 0.32; Epac2-M165T, 6.07 ± 0.34; Epac2-V646F, 6.66 ± 0.52; Epac2-G706R, 5.60 ± 0.40; Epac2-T809S, 8.23 ± 0.31; n = 406–592 spines from 9 cells per condition, 4 experiments). (c) Effects of expression of HA-Epac2 mutations on the average intensity and number of endogenous PSD-95 immunofluorescent puncta in dendrites. (d) Quantification of average immunofluorescence intensity and linear density of PSD-95 puncta in c. Epac2-V646F increased PSD-95 average intensity of individual puncta (Epac2-WT, 170.5 ± 7.07 a.u.; Epac2-M165T, 152.4 ± 9.16; Epac2-V646F, 291.0 ± 26.0; Epac2-G706R, 203.7 ± 8.24; Epac2-T809S, 205.3 ± 7.46). Epac2-T809S increased the number of PSD-95 clusters (Epac2-WT, 8.26 ± 0.76; Epac2-M165T, 7.80 ± 0.29; Epac2-V646F, 10.05 ± 0.93; Epac2-G706R, 8.34 ± 0.98; Epac2-T809S, 14.5 ± 1.01; n = 4–8 cells, 2–3 experiments). Error bars represent s.e.m. Scale bars represent 5 µm.
*
16 12 8 4 0
PSD-95
PSD-95 + HA
Epac2 Epac2-M165T Epac2-V646F Epac2-G706R Epac2-T809S
control plasticity bidirectionally, inducing both LTD and LTP 40,41. Dopamine also facilitates LTD in rat prefrontal cortex42, and application of cAMP under specific conditions depresses synaptic transmission38,43. Epac activation rescues Gαs signaling overactivation-induced deficits in animals 37. Other studies have reported dopamine- and D1-dependent enhancement of synaptic transmission44. The specific conditions under which dopamine signaling promotes potentiation or depression are not completely understood. Our data indicate that Epac2 mediates neuromodulation by DAR-D1/D5 and link dopamine signaling with synapse structural remodeling. Epac2 participated in protein complexes with NL1 and 3, and these proteins showed extensive colocalization in dendrites. Our data suggest that NL3 recruits Epac2 to the plasma membrane and enhances its Rap-GEF activity. Enhanced Epac2 activity promoted spine shrinkage and increased spine dynamics. Neuroligins are synaptic adhesion molecules that have previously been shown to promote synapse formation and maturation19–21. Such seemingly opposite functions are also simultaneously performed by another class of synaptic adhesion molecules, ephrinB and EphB, which promote filopodia motility and motility-dependent synaptogenesis45. As suggested for EphB45, neuroligin-Epac2-Rap signaling may promote local sampling of the presynaptic environment through increased spine motility and may also provide transcellular interaction between the dynamic spines and newly contacted presynaptic terminals. Neuroligins and their ligands neurexins are associated with autism spectrum disorders (ASD). Chromosomal regions of NLGN3, NLGN4 and NRXN1 have been implicated in ASD27,46, but rare mutations in these genes do not account for the strong association of their loci with ASD. Mutations in NLGN3 and NLGN4 that
have been detected in autistic individuals encode proteins with altered function 46. Interaction with neuroligins places Epac2 in functional proximity to proteins that have previously been implicated in ASD, suggesting that it participates in the same synaptic signaling network. A screen of autistic individuals identified four rare nonsynonymous variants in the EPAC2 gene17. We found that two of these mutations affected protein function, signaling and synapse remodeling. Specifically, Epac2-V646F impaired the basal GEF activity and Rap-dependent signaling in dendrites, leading to larger spines and larger PSD-95 clusters (Supplementary Fig. 11). These effects are similar to those caused by Epac2 knockdown or Epac2-∆GEF and can be explained by reduced GEF activity. On the other hand, the Epac2-T809S was more responsive to NL3-dependent enhancement of its GEF activity, leading to increased Rap signaling in dendrites (Supplementary Fig. 11). Its neuronal expression increased spine density, probably as a result of an increased responsiveness to NL3dependent enhancement of its GEF activity and increased clustering of PSD-95, a molecule with established synaptogenic properties. Although the other two mutations did not affect any of the tested parameters, this does not exclude potential effects on other neuronal properties or interaction with other unknown genetic or environmental factors. Aberrant synaptic connectivity is thought to occur in autism and comorbid diseases, including fragile-X, and increased cortical dendritic spine density has been reported in some individuals with autism6. Understanding Epac2 function in spines may therefore shed light on normal and disease-associated spine plasticity.
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1283
Epac2 -T809S
© 2009 Nature America, Inc. All rights reserved.
a r t ic l e s
a r t ic l e s Methods Methods and any associated references are available in the online version of the paper at http://www.nature.com/natureneuroscience/.
1. Holtmaat, A.J. et al. Transient and persistent dendritic spines in the neocortex in vivo. Neuron 45, 279–291 (2005). 2. Kasai, H., Matsuzaki, M., Noguchi, J., Yasumatsu, N. & Nakahara, H. Structurestability-function relationships of dendritic spines. Trends Neurosci. 26, 360–368 (2003). 3. Nägerl, U.V., Eberhorn, N., Cambridge, S.B. & Bonhoeffer, T. Bidirectional activitydependent morphological plasticity in hippocampal neurons. Neuron 44, 759–767 (2004). 4. Zhou, Q., Homma, K.J. & Poo, M.M. Shrinkage of dendritic spines associated with long-term depression of hippocampal synapses. Neuron 44, 749–757 (2004). 5. Oray, S., Majewska, A. & Sur, M. Dendritic spine dynamics are regulated by monocular deprivation and extracellular matrix degradation. Neuron 44, 1021–1030 (2004). 6. Pickett, J. & London, E. The neuropathology of autism: a review. J. Neuropathol. Exp. Neurol. 64, 925–935 (2005). 7. Tada, T. & Sheng, M. Molecular mechanisms of dendritic spine morphogenesis. Curr. Opin. Neurobiol. 16, 95–101 (2006). 8. Jordan, B.A. et al. Identification and verification of novel rodent postsynaptic density proteins. Mol. Cell. Proteomics 3, 857–871 (2004). 9. Peng, J. et al. Semiquantitative proteomic analysis of rat forebrain postsynaptic density fractions by mass spectrometry. J. Biol. Chem. 279, 21003–21011 (2004). 10. Zhu, J.J., Qin, Y., Zhao, M., Van Aelst, L. & Malinow, R. Ras and Rap control AMPA receptor trafficking during synaptic plasticity. Cell 110, 443–455 (2002). 11. Xie, Z., Huganir, R.L. & Penzes, P. Activity-dependent dendritic spine structural plasticity is regulated by small GTPase Rap1 and its target AF-6. Neuron 48, 605–618 (2005). 12. Srivastava, D.P. et al. Rapid enhancement of two-step wiring plasticity by estrogen and NMDA receptor activity. Proc. Natl. Acad. Sci. USA 105, 14650–14655 (2008). 13. Zhu, Y. et al. Rap2-JNK removes synaptic AMPA receptors during depotentiation. Neuron 46, 905–916 (2005). 14. Morozov, A. et al. Rap1 couples cAMP signaling to a distinct pool of p42/44MAPK regulating excitability, synaptic plasticity, learning and memory. Neuron 39, 309–325 (2003). 15. Kawasaki, H. et al. A family of cAMP-binding proteins that directly activate Rap1. Science 282, 2275–2279 (1998). 16. Bos, J.L. Epac proteins: multi-purpose cAMP targets. Trends Biochem. Sci. 31, 680–686 (2006). 17. Bacchelli, E. et al. Screening of nine candidate genes for autism on chromosome 2q reveals rare non-synonymous variants in the cAMP-GEFII gene. Mol. Psychiatry 8, 916–924 (2003).
18. Abrahams, B.S. & Geschwind, D.H. Advances in autism genetics: on the threshold of a new neurobiology. Nat. Rev. Genet. 9, 341–355 (2008). 19. Chih, B., Afridi, S.K., Clark, L. & Scheiffele, P. Disorder-associated mutations lead to functional inactivation of neuroligins. Hum. Mol. Genet. 13, 1471–1477 (2004). 20. Chih, B., Engelman, H. & Scheiffele, P. Control of excitatory and inhibitory synapse formation by neuroligins. Science 307, 1324–1328 (2005). 21. Craig, A.M. & Kang, Y. Neurexin-neuroligin signaling in synapse development. Curr. Opin. Neurobiol. 17, 43–52 (2007). 22. Ulucan, C. et al. Developmental changes in gene expression of Epac and its upregulation in myocardial hypertrophy. Am. J. Physiol. Heart Circ. Physiol. 293, H1662–H1672 (2007). 23. Enserink, J.M. et al. A novel Epac-specific cAMP analogue demonstrates independent regulation of Rap1 and ERK. Nat. Cell Biol. 4, 901–906 (2002). 24. Kang, G. et al. Epac-selective cAMP analog 8-pCPT-2′-O-Me-cAMP as a stimulus for Ca2+-induced Ca2+ release and exocytosis in pancreatic beta-cells. J. Biol. Chem. 278, 8279–8285 (2003). 25. York, R.D. et al. Rap1 mediates sustained MAP kinase activation induced by nerve growth factor. Nature 392, 622–626 (1998). 26. Vossler, M.R. et al. cAMP activates MAP kinase and Elk-1 through a B-Raf– and Rap1-dependent pathway. Cell 89, 73–82 (1997). 27. Jamain, S. et al. Mutations of the X-linked genes encoding neuroligins NLGN3 and NLGN4 are associated with autism. Nat. Genet. 34, 27–29 (2003). 28. Li, Y. et al. The RAP1 guanine nucleotide exchange factor Epac2 couples cyclic AMP and Ras signals at the plasma membrane. J. Biol. Chem. 281, 2506–2514 (2006). 29. Lüthi, A. et al. Hippocampal LTD expression involves a pool of AMPARs regulated by the NSF-GluR2 interaction. Neuron 24, 389–399 (1999). 30. Xia, J., Chung, H.J., Wihler, C., Huganir, R.L. & Linden, D.J. Cerebellar long-term depression requires PKC-regulated interactions between GluR2/3 and PDZ domain– containing proteins. Neuron 28, 499–510 (2000). 31. Cheung, U., Atwood, H.L. & Zucker, R.S. Presynaptic effectors contributing to cAMP-induced synaptic potentiation in Drosophila. J. Neurobiol. 66, 273–280 (2006). 32. Zhong, N. & Zucker, R.S. cAMP acts on exchange protein activated by cAMP/cAMPregulated guanine nucleotide exchange protein to regulate transmitter release at the crayfish neuromuscular junction. J. Neurosci. 25, 208–214 (2005). 33. Gekel, I. & Neher, E. Application of an Epac activator enhances neurotransmitter release at excitatory central synapses. J. Neurosci. 28, 7991–8002 (2008). 34. Ster, J. et al. Epac mediates PACAP-dependent long-term depression in the hippocampus. J. Physiol. (Lond.) 587, 101–113 (2009). 35. Gelinas, J.N. et al. Activation of exchange protein activated by cyclic-AMP enhances long-lasting synaptic potentiation in the hippocampus. Learn. Mem. 15, 403–411 (2008). 36. Ouyang, M., Zhang, L., Zhu, J.J., Schwede, F. & Thomas, S.A. Epac signaling is required for hippocampus-dependent memory retrieval. Proc. Natl. Acad. Sci. USA 105, 11993–11997 (2008). 37. Kelly, M.P. et al. Developmental etiology for neuroanatomical and cognitive deficits in mice overexpressing Galphas, a G-protein subunit genetically linked to schizophrenia. Mol. Psychiatry 14, 398–415 (2009). 38. Frey, U., Huang, Y.Y. & Kandel, E.R. Effects of cAMP simulate a late stage of LTP in hippocampal CA1 neurons. Science 260, 1661–1664 (1993). 39. Otmakhov, N. & Lisman, J.E. Postsynaptic application of a cAMP analogue reverses long-term potentiation in hippocampal CA1 pyramidal neurons. J. Neurophysiol. 87, 3018–3032 (2002). 40. Chen, Z. et al. Roles of dopamine receptors in long-term depression: enhancement via D1 receptors and inhibition via D2 receptors. Receptors Channels 4, 1–8 (1996). 41. Huang, Y.Y., Simpson, E., Kellendonk, C. & Kandel, E.R. Genetic evidence for the bidirectional modulation of synaptic plasticity in the prefrontal cortex by D1 receptors. Proc. Natl. Acad. Sci. USA 101, 3236–3241 (2004). 42. Otani, S., Blond, O., Desce, J.M. & Crepel, F. Dopamine facilitates long-term depression of glutamatergic transmission in rat prefrontal cortex. Neuroscience 85, 669–676 (1998). 43. Gereau, R.W. IV. & Conn, P.J. Potentiation of cAMP responses by metabotropic glutamate receptors depresses excitatory synaptic transmission by a kinaseindependent mechanism. Neuron 12, 1121–1129 (1994). 44. Smith, W.B., Starck, S.R., Roberts, R.W. & Schuman, E.M. Dopaminergic stimulation of local protein synthesis enhances surface expression of GluR1 and synaptic transmission in hippocampal neurons. Neuron 45, 765–779 (2005). 45. Kayser, M.S., Nolt, M.J. & Dalva, M.B. EphB receptors couple dendritic filopodia motility to synapse formation. Neuron 59, 56–69 (2008). 46. Südhof, T.C. Neuroligins and neurexins link synaptic function to cognitive disease. Nature 455, 903–911 (2008).
1284
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
Note: Supplementary information is available on the Nature Neuroscience website.
© 2009 Nature America, Inc. All rights reserved.
Acknowledgments We thank R.L. Huganir (Johns Hopkins University) for antibodies to AMPAR and NMDAR subunits, J. Bos (Utrecht University) and P. Stork (Vollum Institute) for plasmids, A. El-Husseini (University of British Columbia) and P. Scheiffele (University of Basel) for antibodies to neuroligin and plasmid constructs, and G. Borisy and S.-I. Kojima (Northwestern University) for the pGSuper plasmid. We thank A.K. Srivastava (J.C. Self Research Institute of Human Genetics) and G. Swanson (Northwestern University) for thoughtful discussion. This work was supported by the National Alliance for Autism Research, the National Alliance for Research on Schizophrenia and Depression, the Alzheimer’s Association, grants from the US National Institutes of Health (MH 071316 to P.P., NS057499 to M.P. and CA108647 to L.A.Q.), a pre-doctoral American Heart Association fellowship to K.M.W. and a post-doctoral American Heart Association fellowship to D.P.S. AUTHOR CONTRIBUTIONS K.M.W. and D.P.S. designed and performed the experiments; H.P., M.V.B., M.E.C., Z.X. and K.A.J. carried out experiments; M.Y. and M.P. performed the mEPSC experiments and assisted in data analysis; L.A.Q. contributed reagents and provided technical expertise; and K.M.W., D.P.S. and P.P. wrote the manuscript. P.P. directed the project. Published online at http://www.nature.com/natureneuroscience/. Reprints and permissions information is available online at http://www.nature.com/ reprintsandpermissions/.
ONLINE METHODS Reagents. Plasmids expressing Rap1, Rap2-CA, Rap-DN and Epac2 were generated by us or were generous gifts of J. Bos (Utrecht University) and P. Stork (Vollum Institute). Antibodies recognizing AMPA and NMDA receptor subunits were generous gifts from R. Huganir (Johns Hopkins University). Antibodies to NL2 and 3 were from A. El-Husseini (University of British Columbia). We purchased Epac2 monoclonal and polyclonal antibodies, NL-1 polyclonal antibody (Santa Cruz), PSD-95 monoclonal antibody, GluR1 monoclonal antibody, GluR2 monoclonal antibody (Millipore), and NR1 monoclonal antibody (BD Biosciences). 8-CPT (Tocris) and BDNF (Millipore) were also purchased. NL1 antibody (Santa Cruz) was generated to be specific for the extracellular domain of NL1 (amino acids 645–689).
© 2009 Nature America, Inc. All rights reserved.
Neuronal cultures. Dissociated cultures of primary cortical neurons were prepared from embryonic day 18 (E18) Sprague-Dawley rat embryos and cultured as described previously12. Neurons were transfected at DIV 24–28 with plasmids using Lipofectamine 2000 for 2 d. Neurons were then fixed and immuno stained as described1. HEK293 cells were cultured and transfected as described1. Transfection of newly harvested cortical neurons (E18) was accomplished using a Rat Neuron Nucleofector kit (Amaxa Biosystems). Agonist treatments. HEK293 cells overexpressing Epac2 were activated by 10 µM 8-CPT for 1 h. Endogenous Epac2 was activated by incubating neurons with 50 µM 8-CPT for 1 h. Endogenous D1/D5 dopamine receptors were activated by incubation with 20–100 µM SKF-38393 for 30 min. Quantitative PCR analysis. Total mRNA was purified from 2–5 × 106 cultured cortical neurons, 28 DIV, using the SV Total RNA Isolation system (Promega). cDNA was synthesized using qScript cDNA SuperMix (Quanta). mRNA purification and cDNA synthesis experiments were repeated three times. Real-time PCR was carried out with the 7500Fast ABI Prizm (Applied Biosystems) according to the manufacturer’s instructions. SYBR Green was used for quantitative PCR as a double-stranded DNA-specific fluorophore. The efficiency of amplification of genes of interest EPAC1 and EPAC2, as well as for a housekeeping gene RPL19, was 95%. Relative quantification of gene expression for EPAC1 and EPAC2 was measured by normalization against endogenous RPL19 using the ∆CT method. Fold changes were quantified as 2 − (∆CT sample − ∆CT control). The primer sequences that we used for real-time PCR are listed in Supplementary Table 1. Epac2 point mutation generation. Epac2 missense mutants were generated using an HA-tagged Epac2 construct. Four separate clones were generated, each expressing a single mutation corresponding to an individual mutant (Supplementary Fig. 10). The sequences used to generate these mutations are shown in Supplementary Table 2. Point mutations were generated with the QuikChange II XL site-directed mutagenesis kit (Stratagene) following the manufacturer’s instructions. Immunofluorescent labeling. Cells were fixed in either 4% formaldehyde (vol/vol)/4% sucrose(wt/vol) in phosphate-buffered saline (PBS) or at 4 °C in methanol pre-chilled to −20 °C for 20 min. Fixed neurons were permeablized and blocked simultaneously in PBS containing 2% (v/v) normal goat serum and 0.1% (v/v) Triton X-100 for 1 h at 24 °C. Primary antibodies were added in PBS containing 2% normal goat serum overnight at 4 °C, followed by washes in PBS. Secondary antibodies were incubated for 1 h at 24 °C in 2% normal goat serum in PBS. Following washes in PBS, coverslips were mounted using ProLong antifade reagent (Invitrogen). To label surface GluR2, we incubated live cells with an antibody to the extracellular N-terminal of GluR2 (GluR2-n) at 4 °C for 30 min. Cells were fixed for 20 min in 4% formaldehyde/4% sucrose in PBS and were then processed for immunostaining. In the green/purple color scheme, colocalization is indicated by white overlap. These images were taken in the linear range. Quantitative immunofluorescence. Changes in protein clustering and synaptic localization were quantified using immunofluorescence on fixed neurons and visualized with antibodies to various synaptic proteins. Healthy pyramidal neurons were imaged using a confocal microscope (Zeiss LSM5 Pascal). Z stacks of images were taken using the 63× objective. Experiments were carried out blind to condition and on sister cultures. Two-dimensional maximum projection images were reconstructed using Metamorph software (Universal Imaging). Cultures that were directly compared were stained simultaneously and imaged
doi:10.1038/nn.2386
with the same acquisition parameters. For each condition, 7–15 neurons from 3–5 separate experiments and 100 µm of apical dendrite from each neuron were analyzed. For fluorescence intensity measurements, the background corresponding to areas without cells were subtracted to generate a background-subtracted image. Images were then thresholded equally to include clusters with intensity at least twofold above the adjacent dendrite. Regions along dendrites were outlined using the ‘Perimeters’ utility and the linear density (number per 100 µm of dendrite length), area and total gray value (total immunofluorescence intensity) of each AMPAR cluster were measured automatically. To determine the degree of colocalization between two channels, we thresholded each channel to select distinct puncta as described above. A 100-µm dendritic region was selected and puncta were counted; puncta smaller than 0.08 µm2 were excluded from analysis. Regions representing the measured puncta in one channel were generated using Metamorph and overlaid on the other channel. Puncta were counted as being colocalized if the average intensity in the overlaid region exceeded threshold. Thresholds were set individually for each antibody and held constant across treatment condition. Visualization and quantification of spine morphogenesis. To outline spines, we transfected neurons with GFP, fixed them and visualized them using an antibody to GFP. Images were acquired and processed as explained above. Only spines on secondary and tertiary dendrites were measured to reduce variability. Spines were manually outlined so that each spine had a closed perimeter. Spine parameters (length, width, area and density) were measured in Metamorph. Co-immunoprecipitations. Cells (HEK293) were harvested in RIPA buffer (150 mM NaCl, 10 mM Tris-HCl (pH 7.2), 5 mM EDTA, 0.1% SDS (wt/vol), 1% Triton X-100 (vol/vol), 1% deoxycholate (wt/vol), and inhibitors). Precleared lysates were incubated with 3–5 µl of antibody for 2–4 h; 60 µl of protein-A Sepharose was added for 1 h at 4 °C, after which samples were washed with 1 ml of RIPA buffer. Co-immunoprecipitation from cortical pyramidal neurons was carried out essentially as described above. Following treatments, cells were lysed in RIPA buffer and sonicated and cell debris was removed by centrifugation. Samples were then incubated with antibodies for 3 h, 60 µl of protein Sepharose-A was added and the cells were washed three to four times with 0.5 ml of RIPA buffer. Uncropped western blots from these experiments are shown in Supplementary Figure 12. Rap activation assay. Activation of endogenous Rap1 in neurons or HEK293 cells was measured using GST-Ral-GDS affinity resin (Upstate)12. Intensities of bands were quantified by densitometry using ImageJ (rsb.info.nih.gov/ij/). Treated conditions in each experiment were normalized to an untreated control. Uncropped western blots from these experiments are shown in Supplementary Figures 12 and 13. RNAi. Several gene-specific inserts were designed using the BLOCK-iT software (Invitrogen) to encode 21-nucleotide sequences derived from the target transcript, separated by spacer loops of nine nucleotides, followed by the reverse sequence of the target sequence, and subcloned into the pGsuper vector47, which expresses small interfering RNA and EGFP simultaneously, allowing for the identification of transfected cells and outlining neuronal morphology. Sequences corresponding to nucleotides 244–265, 937–958 and 1,236–1,257 of rat Epac2 cDNA were targeted. Each pGsuper-RNAi construct was then tested for its ability to knockdown the expression of exogenous GFP-Epac2 in HEK293 cells. Control plasmid or RNAi targeting an unrelated protein (other RNAi) did not affect the expression of GFP-Epac2, but several plasmids containing small interfering RNA inserts corresponding to Epac2 lowered the expression of exogenous GFP-Epac2 to various degrees. The RNAi construct corresponding to nucleotides 1,236–1,257 effectively knocked down Epac2 and was designated RNAi. To further establish the specificity of the RNAi, we generated an HA-Epac2 construct containing two silent point mutations on the RNAi target sequence (referred to as rescue Epac2). Point mutations were generated with the QuikChange II XL site-directed mutagenesis kit (Stratagene) following the manufacturer’s instructions. Time-lapse imaging. Neurons on coverslips were pre-incubated in artificial cerebrospinal fluid (ACSF), after which they were transferred to a stage chamber and imaged at 37 °C. To examine spine motility, we pretreated cells with or without drugs for up to 40 min before imaging sessions; this was to allow 8-CPT time to fully enter the cell. Images were acquired using a Zeiss LSM 5
nature NEUROSCIENCE
© 2009 Nature America, Inc. All rights reserved.
Pascal confocal microscope. Dendrites were captured through a 63× objective with 2× averaging. Healthy neurons with overall pyramidal morphologies expressing GFP or GFP-tagged protein were identified and imaged. To minimize photodamage, we reduced the laser power to 0.5% and took images at 10-min intervals. Following imaging sessions, 20× images of the neurons were taken to ensure that no photodamage had occurred. Any neurons exhibiting signs of distress were omitted from quantification. Neurons were imaged for 60–80 min. At each time point, Z stacks of images were collected, which were later collapsed into two-dimensional projections in Metamorph. To evaluate spine morphing and motility, we took images at the beginning, middle and end of the imaging sessions and color-coded and overlaid them in Photoshop (Adobe). At least 100 µm of dendrite per cell was analyzed. The total spine motility fraction was defined as the total number of motility events, that is, extension, retraction or head morphing, normalized to spine number48. This method measures the frequency of events, without considering their magnitude, pools all types of events and is a general estimate of overall motility. A protrusive event was defined as the appearance of a new, transient protrusion from a spine head or dendritic shaft. A retraction event was defined as the disappearance of an existing or transient protrusion located on a spine head or dendritic shaft. The sum of protrusion and extension events was divided by the total number of spines in the region of dendrite that was quantified. Motility events were counted by an experimenter blind to condition. To assess changes in individual spine morphology in response to drug treatment, cells were imaged for 60 min before 8-CPT perfusion and imaged for a further 60 min. Area measurements were made 60 min before 8-CPT treatment, immediately preceding treatment and 60 min following treatment. Each time point was normalized to 60 min prior to treatment. To determine whether spine shrinkage was a result of photodamage, we imaged neurons for 60 min before perfusion with vehicle and then imaged them for a further 60 min. Electrophysiology. Pyramidal neurons were identified by their morphology. The external solution contained ACSF (150 mM NaCl, 4.5 mM KCl,
nature NEUROSCIENCE
10 mM HEPES, 10 mM glucose and 2 mM CaCl 2), 1 mM MgCl2 and 200 µM D(–)-2-amino-5-phosphonovaleric acid (pH 7.3, 320–330 mOsm per L). AMPAR-mediated mEPSCs were measured from whole-cell voltage patchclamp recordings with a gap-free model using Igor Pro software data a cquisition system. Signals were low-pass filtered at 1 kHz, and digitized (sampled) at 10 kHz and were amplified with an Axopatch 200B patchclamp amplifier (Axon Instruments). EPSCs were recorded at a holding potential of −60 mV in the presence of tetrodotoxin (1 µM), picrotoxin (100 µM) and strychnine (1 µM) at 24 °C. Patch pipettes were pulled from borosilicate glass and had a resistance of approximately 3–5 MΩ. The internal pipette solution contained 140 mM cesium asparagine, 2 mM MgCl2, 1 mM CaCl2, 10 mM HEPES, 1.1 mM EGTA, 2 mM ATP-Mg2, 0.3 mM GTP-Na2 (pH 7.2 adjusted with CsOH, 300–320 mOsm per L). Signals were analyzed using Clampfit 9.2 (Axon Instruments) and Mini Analysis Program 6.0.3 (Synaptosoft). Statistical analysis. For quantitative immunofluorescence and spine morphology experiments, differences among condition means were identified by Student’s unpaired t tests and ANOVAs performed in Excel (Microsoft) and SPSS (SPSS). Tukey-b post hoc analysis was used for multiple comparisons. In time-lapse experiments, differences in means were determined by Student’s unpaired t tests or one-sample t tests. For Rap assays, differences among conditions were detected by ANOVA followed by Tukey-b post hoc analysis or by Bonferroni-corrected t tests. For mEPSC amplitude and frequency analysis, differences between conditions were determined by one-way or repeated-measure ANOVA. 47. Kojima, S., Vignjevic, D. & Borisy, G.G. Improved silencing vector coexpressing GFP and small hairpin RNA. Biotechniques 36, 74–79 (2004). 48. Dunaevsky, A., Tashiro, A., Majewska, A., Mason, C. & Yuste, R. Developmental regulation of spine motility in the mammalian central nervous system. Proc. Natl. Acad. Sci. USA 96, 13438–13443 (1999).
doi:10.1038/nn.2386
a r t ic l e s
Neuron-glia communication via EphA4/ephrin-A3 modulates LTP through glial glutamate transport
© 2009 Nature America, Inc. All rights reserved.
Alessandro Filosa1,9, Sónia Paixão1,9, Silke D Honsek2, Maria A Carmona3, Lore Becker4,5, Berend Feddersen5, Louise Gaitanos1, York Rudhard6,8, Ralf Schoepfer6, Thomas Klopstock4,5, Klas Kullander7, Christine R Rose2, Elena B Pasquale3 & Rüdiger Klein1 Astrocytes are critical participants in synapse development and function, but their role in synaptic plasticity is unclear. Eph receptors and their ephrin ligands have been suggested to regulate neuron-glia interactions, and EphA4-mediated ephrin reverse signaling is required for synaptic plasticity in the hippocampus. Here we show that long-term potentiation (LTP) at the CA3–CA1 synapse is modulated by EphA4 in the postsynaptic CA1 cell and by ephrin-A3, a ligand of EphA4 that is found in astrocytes. Lack of EphA4 increased the abundance of glial glutamate transporters, and ephrin-A3 modulated transporter currents in astrocytes. Pharmacological inhibition of glial glutamate transporters rescued the LTP defects in EphA4 (Epha4) and ephrin-A3 (Efna3) mutant mice. Transgenic overexpression of ephrin-A3 in astrocytes reduces glutamate transporter levels and produces focal dendritic swellings possibly caused by glutamate excitotoxicity. These results suggest that EphA4/ephrin-A3 signaling is a critical mechanism for astrocytes to regulate synaptic function and plasticity. Interactions between neurons and astrocytes are crucial in both synapse or spine development and synaptic transmission1. Astrocytes release substances such as the matrix-associated protein thrombospondin to regulate synaptogenesis, and several other factors, including the neurotransmitter D-serine, to regulate synaptic transmission2,3. At excitatory synapses, astrocytes can sense synaptic activity by detecting glutamate released from presynaptic terminals and respond to this stimulus with the release of gliotransmitters that, in turn, modulate the activity of the neurons2,3. Glutamate released into the synaptic cleft is cleared by a set of high-affinity transporters found on neurons and astrocytes. The glial transporters are responsible for clearing most glutamate in the hippocampus4. Rapid removal from the extracellular milieu restrains spill-over of glutamate to nearby synapses and protects cells from glutamate excitotoxicity4,5. Glutamate uptake by astrocytes is dynamic and increases during neuronal activity, including long-term potentiation5–7. However, the molecular mechanisms that regulate glutamate transport in astrocytes are poorly understood, and it is unclear to what extent astrocytes contribute to long-term synaptic plasticity. In mouse hippocampus and cerebral cortex, signaling by Eph receptor tyrosine kinases and their cell surface–associated ephrin ligands has been implicated in synapse and spine formation 8,9. B-type Eph receptors—which interact with transmembrane ephrin-Bs—regulate synapse/spine development, at least in part by trans-synaptic interaction with ephrin-Bs expressed in axon terminals10. In contrast, the A-type Eph receptor, EphA4, which has the potential to interact with both A-type and B-type ephrins, has been suggested to interact with
ephrin-A3 expressed on astrocytic processes. Activation of EphA4 forward signaling reduces spine length, whereas inhibition of EphA4 signaling increases spine length11. Hence, astrocytes use the Eph-ephrin system to shape spine morphology and possibly synaptic function. Eph–ephrin signaling also promotes certain forms of hippocampal synaptic plasticity independently of morphological changes9. At the CA3–CA1 synapse (between cornu ammonis area 3 and area 1 neurons), both EphB2 and EphA4 are required for LTP; however, unlike the mechanism that promotes spine remodeling, both Eph receptors act in a kinase-independent fashion12–14. EphB2 may either act postsynaptically, by interacting in cis with NMDA receptors15, or presynaptically, interacting in trans with postsynaptic ephrin-Bs9,16. Unlike EphB2, EphA4 does not seem to interact with NMDA receptors15, and the mechanism by which it promotes LTP is unknown. Here we show that only postsynaptic dendritic, but not axonal, EphA4 is required for certain forms of LTP. Loss of ephrin-A3 affects the same forms of LTP and raises glutamate transporter currents in astrocytes. Loss of neuronal EphA4 increases, whereas transgenic overexpression of ephrin-A3 in astrocytes decreases, glial glutamate transporter abundance. The LTP deficiency observed in both Epha4 and Efna3 mutants is rescued by blocking glial glutamate transporters. These results suggest that interactions between dendritic EphA4 and ephrin-A3 control glial glutamate transport, which regulate synaptic glutamate concentration and postsynaptic depolarization and ultimately modulate the expression of LTP at excitatory synapses.
1Department of Molecular Neurobiology, Max Planck Institute of Neurobiology, Martinsried, Germany. 2Institute for Neurobiology, Heinrich Heine University Düsseldorf, Düsseldorf, Germany. 3Burnham Institute for Medical Research, La Jolla, California, USA. 4German Mouse Clinic, Institute of Experimental Genetics, Helmholtz Center, German Research Center for Environmental Health (GmbH), Munich, Germany. 5Friedrich Baur Institute, Department of Neurology, University of Munich, Munich, Germany. 6Laboratory for Molecular Pharmacology, University College London, London, UK. 7Department of Neuroscience, Uppsala University, Uppsala, Sweden. 8Present address: Evotec AG, Hamburg, Germany. 9These authors contributed equally to this work. Correspondence should be addressed to R.K. (
[email protected]) or S.P. (
[email protected]).
Received 2 July; accepted 5 August; published online 6 September 2009; doi:10.1038/nn.2394
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1285
P21
f
P31
CA3 DG
g
h
i
CA1 CA3
1 mV 200
Epha4lx/– CA1-cre; Epha4lx/–
175 150 125 100 75
k
10 ms
0
TBS 10 20 30 40 50 60 70 80 Time (min)
Slope (% of baseline)
Slope (% of baseline)
© 2009 Nature America, Inc. All rights reserved.
j
CA1-cre; Epha4lx/–
Epha4lx/–
CA3-cre; Epha4lx/– 1 mV
DG
10 ms
200
Epha4lx/– CA3-cre; Epha4lx/–
175 150 125 100 75
0
TBS 10 20 30 40 50 60 70 80 Time (min)
RESULTS EphA4 is required for LTP in postsynaptic CA1 cells To remove EphA4 from subregions of the hippocampus, we used a conditional allele of EphA4 (Epha4loxP, hereafter referred to as Epha4lx; K.K., unpublished results). PGK-cre;Epha4lx/lx mice, in which EphA4 is ubiquitously deleted, showed phenotypes previously described in Epha4 null mutants (Supplementary Fig. 1)17. Although EphA4 expression in control Epha4lx/lx mice is reduced to 15–20% of that in Epha4+/+ mice (Supplementary Fig. 1a), this reduction did not cause phenotypic alterations (Supplementary Fig. 1). We also did not find alterations in CA3–CA1 LTP in control Epha4lx/– mice, which only had one intact EphA4 allele, compared to Epha4lx/+ mice (Supplementary Fig. 1j,k). To generate CA1 pyramidal cell–specific Epha4 knockout mice, we used the R4Ag11CamK2a-cre mouse line (hereafter referred to as CA1-cre)18 which shows full activity in CA1 at postnatal day Figure 2 Ephrin-A3 is required for TBS-induced LTP. (a) fEPSPs slopes at various stimulus intensities (FV, fiber volley) and representative traces (Efna3+/+, black; Efna3–/–, gray; n = 12 slices, 6 mice per group; analysis of covariance, P = 0.1). (b) PPF at various ISIs and representative traces at 40 ms ISI (Efna3+/+, black; Efna3–/–, gray; n = 10 slices, 5 mice per group, two-way repeated measures analysis of variance: between genotypes, F1,90 = 0.03, P = 0.9). (c–f) Scatter plots showing LTP induced by stimulation of presynaptic CA3 neurons with three TBSs (c) or a single tetanus (d–f). (c) Efna3–/– mice show a strong deficit in TBS-induced LTP compared to Efna3+/+ controls (Efna3+/+, 140.7 ± 6.1%, n = 14 slices, 10 mice; Efna3–/–, 114.8 ± 5.0% n = 12 slices, 10 mice, at 55–60 min after stimulus, t-test, P = 0.004). (d) Efna3–/– mice show normal tetanus-induced LTP (Efna3+/+, 138.6 ± 6.0%, n = 13 slices, 10 mice; Efna3–/–, 137.9 ± 7.7% n = 12 slices, 10 mice; at 55–60 min after stimulus, t-test, P = 0.9). (e) CA1-cre;Epha4lx/– mice show normal tetanus-induced LTP (142.8 ± 6.0% in mutants, n = 11 slices, 9 mice; versus 151.0 ± 11.0% in controls, n = 12 slices, 9 mice; at 55–60 min after stimulus, t-test, P = 0.5). (f) CA3-cre;Epha4lx/– mice have normal tetanus-induced LTP (133.8 ± 7.9% in mutants, n = 10 slices, 8 mice; versus 134.0 ± 5.4% in controls, n = 9 slices, 8 mice; at 55–60 min after stimulus, t-test, P = 1.0). Insets: representative traces from controls (black) and mutants (gray) recorded before (stars) and 55–60 min after (triangles) LTP induction. Stimulation artifacts were removed. Error bars, s.e.m.
1286
(P) 31 (Fig. 1a–c). To generate CA3 pyramidal cell–specific EphA4 knockout mice, we used a new Grik4-cre knock-in mouse line (hereafter referred to as CA3-cre) that expresses Cre from the endogenous Grik4 locus (Supplementary Fig. 2). The CA3-cre line recombined in nearly all cells of CA3 and in a smaller fraction of dentate gyrus cells, starting from the first postnatal week (Fig. 1d–f). The recombination efficiencies in the CA1-cre and CA3-cre lines, quantified by in situ hybridizations (Fig. 1g–i), were approximately 80% for both lines (CA1/CA3 ratio in CA1-cre;Epha4lx/–, 20% ± 12%, compared to 91% ± 4% in Epha4lx/–; CA3/CA1 ratio in CA3-cre;Epha4lx/–, 24% ± 4%, compared to 116% ± 7% in Epha4lx/–; n = 3–5 mice; t-test, P < 0.0001). The recombination efficiency of the CA3-cre in the dentate gyrus was approximately 36% (data not shown). CA1-cre;Epha4lx/– and CA3-cre;Epha4lx/– mice showed no gross developmental abnormalities. Before analyzing synaptic plasticity, we analyzed potential morphological alterations and basic synaptic parameters. Biolistic labeling with DiI in hippocampal slices
a
c
b
1.25 1.05 0.85 0.65 0.45
+/+
Efna3 Efna3–/–
0.25
0.05 0.05 0.15 0.25 0.35 0.45 FV amplitude (mV)
1 mV 10 ms
+/+
Efna3 Efna3–/–
225 200
10 ms
175 150 125 100 75
TBS
0
e 250 225 200 175 150 125 100 75 0
d
10 20 30 40 50 60 70 80 Time (min)
Epha4lx/– CA1-cre;Epha4lx/–
180
Efna3+/+ Efna3–/–
160 140
1 mV 40 ms
120 100
0
Tetanus 10 20 30 40 50 60 70 80 Time (min)
80 120 160 ISI (ms) +/+
Efna3 Efna3–/–
225 200
10 ms
175 150 125 100 75
0
f 10 ms
40
1 mV
e
CA1
250 225 200 175 150 125 100 75
Tetanus 10 20 30 40 50 60 70 80 Time (min) Epha4lx/– CA3-cre;Epha4lx/–
1 mV
d
Slope 2/slope 1 (%)
P5
CA3 DG
Figure 1 EphA4 is required for LTP in postsynaptic CA1 cells. (a–f) Staining for -galactosidase activity in hippocampal sections of the indicated ages of CA1-cre;Rosa26lx/+ (a–c) and CA3-cre;Rosa26lx/+ (d–f) mice (see Online Methods). DG, dentate gyrus. (g–i) In situ hybridizations for Epha4 mRNA in adult hippocampus from mice of the indicated genotypes. Small square, regions used for densitometric measurements for the calculation of the recombination efficiency: red, CA1; blue, CA3; yellow, background. Scale bars, 500 µm. (j,k) Scatter plots showing LTP, represented as a percentage of the baseline, induced by stimulation of presynaptic CA3 neurons with three TBSs. Insets: representative traces from controls (black) and mutants (gray) recorded before (stars) and 55–60 min after (triangles) LTP induction. For clarity, the stimulation artifact was removed. (j) CA1-cre;Epha4lx/– mice showed a marked deficit in early CA3–CA1 LTP compared to littermate Epha4lx/– controls (123.0 ± 4.4% in mutants, n = 12 slices, 7 mice; versus 144.8 ± 7.6% in controls, n = 14 slices, 7 mice; at 55–60 min after stimulus, t-test, P = 0.02). (k) CA3-cre;Epha4lx/– mice show normal TBS-induced LTP (142.9 ± 10.0% in mutants, n = 10 slices, 8 mice; versus 136.9 ± 9.7% in controls, n = 10 slices, 8 mice; at 55–60 min after stimulus, t-test, P = 0.7). Error bars, s.e.m.
Slope (% of baseline)
P31
Slope (% of baseline)
c
1 mV
P19
1 mV
b
fEPSP slope (mV ms–1)
P15 CA1
Slope (% of baseline)
a
Slope (% of baseline)
CA3-Cre
CA1-Cre
a r t ic l e s
10 ms
0
Tetanus 10 20 30 40 50 60 70 80 Time (min)
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
a r t ic l e s b
Epha4
c
Hip +/+ –/–
Cx +/+ –/–
GLAST
100
kDa
100
EphA4
50
GFAP
50
Tubulin
d
Tubulin
50
e
Epha4
kDa
100
EphA4
50
GFAP
Epha4+/+ –/– Epha4
*
GLAST
GLT1
GFAP EAAC1
50
GLT1
50
Tubulin
1.8 1.5 1.2 0.9 0.6 0.3 0
Epha4+/+ Epha4EGFP/EGFP
GLAST
GLT1
GFAP EAAC1
Epha4 –
i CA3 region
lx/–
+
CA1-cre;Epha4lx/–, CA3 region
Epha4lx/– +
CA1-cre – GLAST Tubulin
–
+
Epha4lx/– CA1-cre;Epha4lx/–, CA1 region
+
kDa 100
GLAST
50
Tubulin
1.8 1.5 1.2 0.9 0.6 0.3 0
***
GLAST
did not reveal alterations in the morphology of dendritic spines in CA1-cre;Epha4lx/– mice, compared to the Epha4lx/– controls, possibly owing to the incomplete recombination efficiencies (data not shown). Extracellular recordings in acute hippocampal slices of the CA3–CA1 pathway in both conditional mutants did not reveal significant deficiencies in the slopes of field excitatory postsynaptic potentials (fEPSPs) nor in paired-pulse facilitation (PPF) at various interstimulus intervals (ISIs) (Supplementary Fig. 3). We tested AMPA and NMDA receptor functions, as well as CaMKIIα protein abundance, in Epha4 null mice and found them to be unaffected (Supplementary Fig. 4k,l). To investigate a requirement for either postsynaptic (CA1) or presynaptic (CA3) EphA4 in LTP, we used theta-burst stimulation (TBS). After recording a stable baseline, three TBSs were given to fibers of the CA3 presynaptic neurons and LTP was recorded from
a
GLT1
Figure 3 Upregulation of GLAST and GLT-1 protein levels in Epha4 mutants. (a,b) Protein lysates from cerebral cortex (Cx) and hippocampus (Hip) derived from Epha4–/– and littermate Epha4+/+ controls compared by western blot analysis for their content of GLAST, GLT-1, GFAP, EAAC1, EphA4 and tubulin. (c) Quantification of glutamate transporters and GFAP in hippocampi of Epha4–/– and littermate Epha4+/+ controls. Expression was normalized to tubulin in densitometric measurements. Mean values of intensities of the indicated proteins are shown relative to values for Epha4+/+ protein lysates. GLAST and GLT1 levels were 50% and 25% higher, respectively, in Epha4–/– mice (GLAST **P = 0.009, n = 10 mice per group; GLT1 *P = 0.03, n = 15 mice; GFAP P = 0.2, n = 10 mice; EAAC1 P = 0.3, n = 3 mice; t-test) than in Epha4+/+ controls. (d,e) Western blot analysis of hippocampal protein lysates from Epha4EGFP/EGFP and littermate Epha4+/+ controls for their content of GLAST, GLT-1, GFAP, EphA4 and tubulin. (f) Quantification of protein, calculated as in c (n = 7–9 mice per group, P > 0.3, t-test). (g,h) Western blot analysis of protein lysates from dissected CA1 and CA3 regions derived from CA1-cre;Epha4lx/– mice or Epha4lx/– controls for their content of GLAST and tubulin. (i) GLAST abundance in CA1, but not CA3, was increased (GLAST in CA1, n = 7 mice, ***P = 0.0009; GLAST in CA3, n = 6 mice, P = 0.7; GLT1 in CA1, n = 13 mice, P = 0.07; GLT1 in CA3, n = 6 mice, P = 0.5; and GFAP, n = 6 mice, P = 0.9, t-test). Error bars, s.e.m.
CA1 neurons for up to 60 min after stimulation. Comparison of CA1cre;Epha4lx/– mice with littermate Epha4lx/– controls revealed a marked reduction in LTP (Fig. 1j) comparable to the situation when EphA4 was removed postnatally from all pyramidal neurons of the forebrain (Camk2a-cre;Epha4lx/– mice; Supplementary Fig. 3g,h). In contrast, TBS-induced LTP was not altered in CA3-cre;Epha4lx/– mice (Fig. 1k). These results demonstrate that EphA4 functions postsynaptically to regulate LTP at the CA3–CA1 synapse. Ephrin-A3 is required for TBS-induced LTP Dendritic EphA4 has been shown to respond to ephrin-A3 expressed in astrocytes11,19, and immunoelectron microscopic studies have suggested that EphA4 is expressed mainly perisynaptically20. We therefore asked whether ephrin-A3 function was critical for LTP.
b
c
Figure 4 Astrocytic glutamate transporter Efna3 +/+ Single pulse TBS (10 bursts of 4 pulses at 100 Hz) currents. (a) Mean inward currents in response Efna3 –/– 150 * Burst number to a single pulse stimulation of Schaffer ** 1 2 3 4 5 6 7 8 9 10 10 pA +/+ –/– collaterals in Efna3 (black) and Efna3 mice 100 20 ms (gray). The slow, persistent component did not differ between Efna3+/+ and Efna3–/– mice. 50 Inset: mean fiber volleys in Efna3+/+ (black) and 0.1 mV 4 ms Efna3–/– (gray) mice (the peaks of individual 50 pA 0 traces were aligned in time before calculating Single pulse Average of 200 ms the average). (b) Mean inward currents in 10 bursts response to TBS in Efna3+/+ (black) and Efna3–/– mice (gray). Traces are averages from all stimulations performed. Stimulation artifacts in a and b were replaced by a vertical line indicating the start of the stimulation. (c) Mean glutamate transporter currents amplitudes (TC), divided by the amplitude of the corresponding fiber volleys (FV) to normalize for the numbers of fibers activated. Transporter currents in Efna3–/– mice were significantly higher than in Efna3+/+ controls in response to a single pulse stimulation (117.7 ± 11.1 pA mV–1 in mutants, n = 42 recordings, 20 cells; 95.7 ± 8.9 pA mV–1 in Efna3+/+ controls, 34 recordings, 17 cells; **P = 0.01, t-test) and in response to TBS (124.8 ± 18.8 pA mV–1 in mutants, n = 20 measurements, 15 cells; 83.5 ± 8.5 pA mV–1 in Efna3+/+ controls, n = 18 measurements, 10 cells; *P = 0.03, t-test). Error bars, s.e.m. TC/FV (pA mV–1)
© 2009 Nature America, Inc. All rights reserved.
kDa
h
CA1 region
50
**
Tubulin
50
CA1-cre –
1.8 1.5 1.2 0.9 0.6 0.3 0
f
Epha4 +/+ G/G +/+ G/G
GLAST
100
g
GLT1
50
+/+ G/G +/+ G/G
kDa 100
Hip +/+ –/–
Relative expression
kDa
Epha4
Relative expression
Cx +/+ –/–
Relative expression
a
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1287
a r t ic l e s b γ-DGG 40 pA 500 ms
Control
25
15 10 5 0
d 1.8
1.5
1.5
0.3
Slope (% of baseline)
+/
Epha4 CA1-cre; Epha4lx/–
175 150 125
TBS
100 75 100 nM TFB-TBOA 0 10 20 30 40 50 60 70 80 Time (min)
4 ha
1 mV
1 mV
© 2009 Nature America, Inc. All rights reserved.
lx/–
0.3
g
10 ms
200
1 mV 20 ms
0.6 0
Ef na Ef 3 na 3
Ep
–
ha C 4 lx Ep A1 /– ha -cr 4 lx/ e;
0
0.9
–
0
f
1 mV 20 ms
0.6
1.2
h C a4 lx Ep A3 /– ha -cr 4 lx/ e;
0.3
1 mV 20 ms
1.5
Ep
*
Slope (% of baseline)
0.6
0.9
+
0.9
*
1.2
e 1.8
–/ –
1.2
Ep
1.8
Slope 4/slope 1
Slope 4/slope 1
c
Ep
ha
4
+/
–/ –
+
γ-DGG
*
20
Slope 4/slope 1
Control
(1– (DGG/control)) ×100
Epha4 –/–
Epha4 +/+
a
Figure 5 Glutamate levels, postsynaptic responses to high frequency stimulation and pharmacological rescue of LTP. (a) Representative traces of mEPSCs in Epha4+/+ and Epha4–/– mice in absence and presence of 1 mM γ-DGG. (b) Inhibitory effect of γ-DGG on mean mEPSCs amplitude in Epha4–/– mice and controls (18.6% ± 2.4% versus 11.6% ± 2.3%, *P = 0.04, t-test; n = 13 mice per group). (c–e) fEPSP slope 4/slope 1 ratios during a train of four stimuli at 100 Hz (0.67 ± 0.07 in CA1cre;Epha4lx/–, n = 12 slices, 7 mice; 1.07 ± 0.16 in Epha4lx/–, n = 14 slices, 7 mice; P = 0.03; 1.06 ± 0.09 in Efna3–/–, n = 14 slices, 9 mice; 1.51 ± 0.2 in Efna3+/+, n = 13 slices, 9 mice; P = 0.05; 0.98 ± 0.14 in CA3-cre;Epha4lx/–, n = 9 slices, 6 mice; 1.12 ± 0.13 in Epha4lx/–, n = 10 slices, 6 mice; P = 0.5, t-test). Insets: traces from one representative train in controls (black) and mutants (gray). Stimulation artifacts were removed. (f,g) TBS-induced LTP in presence of TFB-TBOA (black bar; applied 8 min before TBS and washed out 2 min after) in CA1-cre;Epha4lx/– (f) and Efna3–/– (g) mice and respective controls (153.5 ± 9.1% in CA1-cre;Epha4lx/–, n = 9 slices, 7 mice; 145.2 ± 6.7% in Epha4lx/–, n = 10 slices, 7 mice, P = 0.6; 152.6 ± 7.6% in Efna3–/–, n = 11 slices, 9 mice; 149.9 ± 10.1% in Efna3+/+, n = 10 slices, 9 mice; P = 0.8, t-test). Insets: representative traces from controls (black) and mutants (gray) recorded before (stars) and 55–60 min after (triangles) LTP induction. Stimulation artifacts were removed. Error bars, s.e.m.
10 ms
Efna3 +/+ Efna3 –/–
200 175 150 125
TBS
100 75
100 nM TFB-TBOA 0 10 20 30 40 50 60 70 80 Time (min)
We used an Efna3 null mutant mouse, which has alterations in spine morphology19 but does not show significant defects in basal synaptic transmission parameters such as PPF, input-output ratios, and amplitude and frequency of miniature excitatory postsynaptic currents (mEPSCs), suggesting that the observed spine alterations are not critical for basic synaptic functions (Fig. 2a,b and Supplementary Fig. 4). As with Epha4 mutants, LTP induced by a TBS protocol was strongly reduced in Efna3–/– slices (Fig. 2c). Notably, this LTP defect in Efna3–/– slices was specific for TBS and was not observed after tetanic stimulation (Fig. 2d). Therefore, the requirement of ephrin-A3 for LTP is dependent on a stimulus protocol that is physiologically closer to what happens in the hippocampus during learning and memory21, which suggests a direct, possibly signaling function of ephrin-A3 in LTP. The requirement for TBS- but not tetanus-induced LTP was also observed in EphA4 mutant mice (Fig. 2e,f and Supplementary Fig. 3h). These findings suggest that EphA4 interacts with ephrin-A3 to regulate TBS-induced LTP at the CA3–CA1 synapse.
by GFP12, suggesting that the EphA4 ectodomain was sufficient to maintain normal GLAST and GLT-1 levels (Fig. 3d–f). GLAST upregulation was also seen in dissected CA1 regions, but not CA3 regions, of CA1-cre;Epha4lx/– mice compared to that in littermate controls (Fig. 3g–i). Similar changes in GLAST and GLT-1 were observed in Efna3–/– mice19. These results suggest that the EphA4 ectodomain, presumably by interacting with ephrin-A3, restricts the expression of glial glutamate transporters. These changes were not secondarily caused by an increase in astrocyte numbers in Epha4–/– mice, and the regulation appeared to happen at the post-transcriptional level, as the amounts of Slc1a3 and Slc1a2 mRNAs (encoding GLAST and GLT-1, respectively) in Epha4–/– and Efna3–/– mice were similar to those in control mice (Supplementary Fig. 5). The upregulation of glial glutamate transporters may result in increased glutamate transporter currents in mutant astrocytes. We performed whole-cell patch-clamp recordings from astrocytes in the stratum radiatum of Efna3–/– hippocampal slices. Transporter currents were evoked by endogenous release of glutamate from presynaptic terminals upon Schaffer collateral stimulation. The mean peak amplitude of transporter currents normalized to the respective fiber volley in response to single-pulse stimulation was significantly higher in Efna3–/– than in littermate controls (Fig. 4 and Supplementary Fig. 6). Similar results were obtained with a TBS protocol (Fig. 4b,c and Supplementary Fig. 6), which had produced a much reduced LTP in Efna3−/− mice (Fig. 2). These results indicate that glutamate uptake by astrocytes in response to a single burst or a TBS protocol was significantly elevated in Efna3−/− mice.
Glial glutamate transporter regulation by EphA4/ephrin-A3 Next, we reasoned that the EphA4 ectodomain may activate ephrin-A3 reverse signaling in astrocytes, thereby modulating astrocytic functions that impact on LTP. As glial glutamate transporters are known to regulate synaptic transmission by clearing glutamate from the synaptic cleft5,22, we investigated the expression of the glial glutamate/aspartate transporter (GLAST or EAAT1) and glutamate transporter subtype-1 (GLT1/EAAT2) in our mutant mice. A marked upregulation of GLAST protein abundance and a modest increase in GLT-1 were observed in Epha4–/– brains compared to littermate controls (Fig. 3a–c). Amounts of glial fibrillary acidic protein (GFAP) and the neuronal excitatory amino acid carrier-1 (EAAC-1/EAAT3) were not altered (Fig. 3a–c). Notably, the upregulation of GLAST and GLT-1 was rescued in Epha4EGFP/EGFP mice, in which the cytoplasmic domain is replaced
Glutamate concentration near synapses and effect on LTP We next asked whether elevation of glutamate transporter abundance in astrocytes would decrease synaptic glutamate. To estimate synaptic glutamate concentrations, we performed whole-cell patch-clamp recordings from CA1 hippocampal neurons in acute slices in the presence of the low-affinity competitive AMPA receptor antagonist γ-D-glutamylglycine (γ-DGG). γ-DGG binds non-NMDA receptors with low affinity and with rapid dissociation kinetics comparable to the dissociation kinetics of synaptically released glutamate23,24. Thus, the degree of inhibition by γ-DGG is sensitive to the concentration of glutamate. If less glutamate is present, the inhibition by γ-DGG is stronger. We found comparable amplitudes and frequencies of mEPSCs in slices derived from Epha4–/– mice and Epha4+/+ littermate
1288
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
–/ –
Ep h
a4
e 1
ty p kDa
18
c
Tg
b
W ild
a
116
pTyr
116
EphA4
116
EphA4
50
Tubulin
IP EphA4 str. I-m
TCL
d
e
HA
GFP
i
g
GFAP
j
Merge
k Relative GLAST expression
h
f
GLAST
HA
m
l
n
GLT1
p
HA
q
o
GLAST
GFP
s
Merge
0.8 0.6 0.4 0.2 0 0.2 0.4 0.6 0.8 1.0 Relative HA expression
1.0 0.8 0.6 0.4 0.2 0
Merge
r
1.0
0
Merge
Relative GLT1 expression
Figure 6 Ephrin-A3 overexpression in astrocytes reduces glutamate transporters. (a,b) Antihemagglutinin immunohistochemistry in Tg181 showing scattered distribution of transgenic protein throughout the brain and hippocampus. str. l-m, stratum lacunosum-moleculare. (c) Transgenic ephrin-A3 induces higher endogenous phosphorylation of EphA4. EphA4 was immunoprecipitated (IP) from wild-type, Tg181 and control Epha4–/– hippocampal lysates and blotted with phosphotyrosine (pTyr) and EphA4 antibodies. Total cell lysates (TCL) of the same fractions show EphA4 and tubulin. (d–g) Specific expression of transgenic ephrinA3 in glial cells. Immunofluorescence confocal images of hippocampal sections from adult Tg181 crossed to a GFAP-GFP line26. Triple labeling for the indicated proteins shows colocalization of hemagglutinin (HA) with GFP but not with the cytoplasmic marker GFAP (g), indicating that the transgenic protein is expressed in fine processes of astrocytes. (h–j) Immunofluorescence singleplane confocal images showing double labeling for GLAST and HA in the stratum lacunosummoleculare from Tg181 mice. (k) Scatter plot showing the negative correlation between GLAST and HA relative pixel intensities (correlation factor –0.46, n = 198 regions, 3 mice, t-test, P < 0.0001). Red lines, linear regression. (l–n) Singleplane confocal images of double labeling for GLT1 and HA in the stratum lacunosum-moleculare from Tg181 mice. (o) Scatter plot showing the negative correlation between GLT1 and HA relative pixel intensities (correlation factor −0.58, n = 280 regions, 3 mice; t-test; P < 0.0001). (p–r) Immunofluorescence single-plane confocal images showing double-labeling for GLAST and GFP in GFAP-GFP mice. Expression of GLAST (red) is not affected in locations where the expression of transgenic GFP (green) is high. (s) Scatter plot showing lack of correlation between GLAST/GFP relative pixel intensities (correlation factor –0.02, n = 121 regions, 2 mice; t-test, P > 0.1). Scale bars: c, 1 mm; d, 300 µm; e–n, 10 µm.
Relative GLAST expression
© 2009 Nature America, Inc. All rights reserved.
a r t ic l e s
0 0.2 0.4 0.6 0.8 1.0 Relative HA expression
1.0 0.8 0.6 0.4 0.2 0
0 0.2 0.4 0.6 0.8 1.0 Relative GFP expression
controls (Fig. 5a and Supplementary Fig. 4a–d). γ-DGG reduced the amplitude of mEPSCs in Epha4–/– slices more than in Epha4+/+ controls (Fig. 5b), indicating that glutamate concentrations near synapses in Epha4–/– slices are reduced. These results suggest that the clearance of glutamate is more efficient in the EphA4 mutants, possibly because of the upregulation of glial glutamate transporters. To test whether the changes in glutamate concentration resulted in insufficient postsynaptic depolarization, we analyzed fEPSPs during high-frequency stimulation. fEPSP slopes at the end of the first train of each TBS were significantly reduced in CA1-cre;Epha4lx/– and Efna3–/–, but not CA3-cre;Epha4lx/– mice, compared to those in their respective controls (Fig. 5c–e). Next, we investigated the effects of blocking glutamate reuptake on LTP. We used (2S, 3S)-3-{3-[4-(trifluorome thyl)benzoylamino)benzyloxy}aspartate (TFB-TBOA) an untransportable inhibitor that primarily inhibits GLT-1 and GLAST but also substantially targets the neuronal transporter EAAC1 (ref. 25). If the LTP impairments were due to glutamate transporter upregulation, TFB-TBOA should rescue the LTP defects. Indeed, the application of TBS in the presence of TFB-TBOA induced similar amounts of synaptic potentiation in CA1-cre;Epha4lx/– and Efna3–/– mice compared to their respective controls (Fig. 5f,g). Interestingly, the presence of
TFB-TBOA also allowed normal postsynaptic depolarization in CA1cre;Epha4lx/– and Efna3–/– slices during the application of TBS (slope 4/ slope 1: 1.23 ± 0.16 in CA1-cre;Epha4lx/–, n = 9 slices, 7 mice versus 1.30 ± 0.08 in Epha4lx/–, n = 13 slices, 7 mice, t-test, P = 0.7; 1.24 ± 0.19 in Efna3–/–, n = 12 slices, 8 mice versus 1.34 ± 0.12 in Efna3+/+, n = 11 slices, 8 mice, t-test, P = 0.7). These results suggest that the impaired LTP observed in CA1-cre;Epha4lx/– and Efna3–/– mice is largely due to the reduced levels of glutamate near synapses, caused by increased glutamate transport into astrocytes.
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1289
Overexpression of ephrin-A3 in astrocytes To obtain further support for regulation of glial glutamate transport by ephrin-A3, we generated transgenic lines overexpressing ephrin-A3 (fused to a hemagglutinin epitope tag) in glial cells using the GFAP promoter26 (Supplementary Fig. 7a). Two lines (Tg181 and Tg7047) produced viable offspring and were further analyzed. Based on quantitative reverse transcription (qRT)-PCR analysis, Tg181 hippocampi contained 4.5 times the endogenous ephrin-A3 abundance (Supplementary Fig. 7e). Anti-hemagglutinin immunostaining of P30–P60 brain sections revealed the typical scattered distribution expected for glial cells (Fig. 6a,b and Supplementary
a r t ic l e s a
b
c Wild type
d
Tg181
e
f
Wild type
Wild type
Phase Phase Phase Phase Phase 1 3a 3b 3c 4
6 4 2 0
1
H2O 100 µM glutamate
8
pe
0
**
10
18
0.02
12
ty
0.04
Wild type Tg181
Tg
0.06
i Mean number of myocloni
Tg181
8 7 6 5 4 3 2 1 0
ild
0.08
Number of mice
0.10
Wild type **
Tg181
h9
W
© 2009 Nature America, Inc. All rights reserved.
Swelling frequency
g 0.12
Tg181
Fig. 7b,c). Endogenous EphA4 phosphorylation was higher in hippo campal lysates from Tg181 mice than in those from littermate controls (Fig. 6c), indicating that the transgenic protein was functional and that EphA4 efficiently interacted with glial ephrin-A3. We confirmed ephrin-A3 overexpression in the fine processes of astrocytes by using a GFAP-GFP indicator line26 (Fig. 6d–g). To investigate the effects of transgenic ephrin-A3 on endogenous glial glutamate transporter abundance, we co-stained hippocampal sections from Tg181 mice with hemagglutinin and GLAST or GLT1 antibodies (Fig. 6). We selected areas of high, medium and low hemagglutinin immunoreactivities and compared average pixel intensities in the same areas to those produced by immunostaining for glial transporters. An inverse correlation was seen between hemagglutinin and GLAST/GLT1 expression (Fig. 6k,o); similar reductions of glial glutamate transporter abundance were observed in line Tg7047, which expressed lower amounts of transgenic ephrin-A3 than Tg181 (Supplementary Fig. 7), suggesting that modest overexpression of ephrin-A3 is sufficient to reduce glial glutamate transporter expression. To verify that the transgenic protein was not targeted to a subpopulation of astrocytes that express less of the glutamate transporters, we co-stained hippocampal sections from the GFAP-GFP indicator line with GFP and GLAST or GLT1 antibodies. We found that the expression of glutamate transporters did not correlate with the expression of GFP in astrocytes (Fig. 6p–s and Supplementary Fig. 7). These results indicate that ephrin-A3 in astrocytes is sufficient to suppress glial glutamate transporter expression. Reduction of glial glutamate transporters has previously been shown to increase synaptic glutamate concentrations and to cause epilepsy and neurodegeneration27–29. Excessive glutamate concentrations cause excitotoxicity that is reflected in dendritic beading and spine loss30. To assess the degree of dendritic beading in ephrin-A3 transgenic mice, we performed biolistic labeling of neurons with DiI or intercrossed Tg181 with a transgenic line (GFPM) which expresses GFP in small neuronal subsets 31. CA1 neurons in P29–P31 hippocampal slices derived from Tg181 mice, but not from wild-type littermates, showed dendritic beading (Fig. 7a–d) with variable penetrance possibly owing to variable expression of the transgene (38.8 ± 3.0% of swelling coverage, n = 16 dendrites, 6 mice, in Tg181; versus 21.7 ± 5.7%, n = 10 dendrites, 6 mice, in 1290
Figure 7 Ephrin-A3 overexpression in astrocytes increases susceptibility to excitotoxicity and seizures. (a,b) Transgenic ephrin-A3 causes dendritic beading. Confocal images of single CA1 pyramidal neurons from wild-type (a) and Tg181 (b) mice crossed to GFP-M mice31. (c,d) Confocal stack of stretches of dendrite of CA1 neurons from wild-type (c) and Tg181(d) mice labeled with DiI; arrowheads, focal swellings; arrow, thin dendritic stretch separating the swellings. (e,f) Confocal stacks of CA1 pyramidal dendrites in organotypic slices from wild-type (e) and Tg181 (f) mice crossed to GFP-M mice treated with 100 µM glutamate. (g) Bar graph of swelling frequency (number of swellings per dendrite complexity index) in control (H 2O) and treated slices. In Tg181 mice swelling frequency was 5.7-fold greater than in wild-type controls (0.088 ± 0.018 in Tg181, n = 6 slices, 3 mice; 0.015 ± 0.007 in wild-type littermate mice, n = 7 slices, 3 mice; t-test, **P = 0.002). (h) PTZ-induced epileptic seizures, depicted as maximal phase reached after intraperitoneal injection of PTZ, were more severe in Tg181 mice than in wild-type mice (n = 14 mice, P = 0.01, Mann-Whitney U-test). Phase classification was performed as described 44. (i) Mean number of whole-body myocloni in phase 3 after injection of PTZ (3.4 ± 0.8 in wild-type mice, n = 12 mice; 8.3 ± 1.4 in Tg181 mice, n = 12; t-test, **P = 0.009). Scale bars: a,b, 50 µm; c,d, 5 µm; e,f, 10 µm. Error bars, s.e.m.
littermate controls, t-test, P = 0.003). Neurons from line Tg181 also showed significant spine loss (1.37 ± 0.06 spines µm–1 in Tg181, n = 16 dendrites, 6 mice; versus 1.58 ± 0.07 spines µm–1 in wild-type control mice, n = 10 dendrites, 6 mice; t-test, P = 0.02), probably as an indirect consequence of beading32. Dendritic beading can also be induced in hippocampal slice cultures by application of agonists of ionotropic glutamate receptors30. In cultured slices derived from line Tg181, neurons did not contain dendritic beads. However, they were more sensitive to glutamate bath application than control slices (Fig. 7e–g). Deficiency in glial glutamate transporters results in increased sensitivity to pentylenetetrazole (PTZ)-induced seizures28,29. Consistent with these reports, Tg181 mice showed increased seizure severity and more whole-body myocloni after application of PTZ (45 mg per kilogram body weight) than controls (Fig. 7h,i). These results suggest that overexpression of ephrin-A3 in astrocytes downregulated glutamate transporter levels, which caused glutamate excitotoxicity and exacerbated PTZ-induced seizures. DISCUSSION The results presented here show a new mechanism by which astrocytes modulate neuronal plasticity in the CA1 region of the hippo campus. Astrocytes receive a signal from dendritic EphA4 receptors by means of ephrin-A3, which prevents them from upregulating glial glutamate transporter expression to unphysiologically high levels. Dendritic EphA4 and ephrin-A3 in astrocytes thereby control glutamate concentrations near synapses and promote LTP. In the absence of either dendritic EphA4 or ephrin-A3, the abundance of glutamate transporters is increased (see also ref. 19) and glutamate is more efficiently removed during high-frequency stimulation. As a consequence, peri- and extrasynaptic glutamate receptors33,34 may not be sufficiently activated, resulting in insufficient depolarization of the postsynapse and partial impairment of LTP. These findings are consistent with previous observations in the cerebellum, where synaptic plasticity is controlled by the neuronal glutamate transporter EAAT435,36. They also concur with previous observations that glutamate transporter levels are regulated by neuronal activity7. Induction of LTP in area CA1 has been shown to increase both neuronal37 and glial6 glutamate uptake. It is therefore possible that neuronal activity regulates EphA4–ephrin-A3 signaling to control glial glutamate transport and in this way fine-tune synaptic transmission. VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
© 2009 Nature America, Inc. All rights reserved.
a r t ic l e s We cannot exclude the possibility that the observed changes in LTP are exacerbated by morphological changes. Previous studies in the hypothalamus have shown that reduced astrocyte coverage enhanced extracellular glutamate concentrations and activation of metabotropic glutamate receptors24. It is also possible that loss of EphA4 and ephrinA3 changes the motility of astrocytic processes, as previously suggested based on stimulation with Fc fusion proteins38,39. However, we found that pharmacological inhibition of glutamate reuptake rescued the LTP defects, arguing that the observed LTP defects in EphA4 and ephrin-A3 mutant mice were caused by deficiencies in acute signaling rather than by morphological alterations. The fact that LTP defects were only observed after TBS, not tetanus, could be explained by the intrinsic capacity of transporters to clear glutamate. TBS consists of short trains of high-frequency stimulations with interburst intervals of 200 ms, which lead to modest glutamate release and to glutamate concentrations that are sensitive to transporter abundance40. We showed that transporter currents are higher in Efna3–/– than in control mice, also under TBS conditions, suggesting that these modest glutamate elevations are cleared more efficiently than normal and thereby reduce the degree of postsynaptic depolarization during LTP induction. Using a stronger stimulation protocol, such as tetanus, which leads to much higher glutamate release, may overwhelm the capacity of the transporters. In that case, the increase in glutamate transporters in mutant mice may be as ineffective as in control mice at buffering extracellular glutamate. Alternatively, tetanus stimulation may influence transporter currents by mechanisms independent of Eph–ephrin signaling. EphA4 is expressed both pre- and postsynaptically20, but here we have discovered that only the postsynaptic fraction of EphA4 plays a role in LTP. Therefore, the previously suggested model of axonal EphA4 activating ephrin-B reverse signaling in spines is probably incorrect12. To be able to activate ephrin-Bs, EphA4 would have to interact with ephrin-Bs in cis, which we consider unlikely. Instead, the results in this study (see also ref. 19) suggest that EphA4 interacts with astrocytic ephrin-A3. The requirement for postsynaptic, but not presynaptic, EphA4 could be explained by the fact that glial coverage at hippocampal synapses is asymmetrically distributed, with threefold more glial contact with spines compared to presynaptic boutons41. The role of EphA4 in LTP is independent of forward signaling12 and therefore distinct from its role in spine morphogenesis11. It is likely that the EphA4 ectodomain activates ephrin-A3 reverse signaling in astrocytes; however, the mechanism that regulates glial glutamate transporter levels downstream of ephrin-A3 is currently unknown. As ephrin-A3 mRNA is also detectable in pyramidal neurons (data not shown), we cannot formally rule out a function of ephrin-A3 in hippocampal neurons. What are the physiological and pathophysiological implications of our observations? Hippocampal LTP is a critical component of the cellular mechanisms underlying certain aspects of learning and memory42. Efna3–/– mice were shown to have impairments in certain behavioral tests requiring the hippocampus19, suggesting that the regulation of glial glutamate transport and synaptic plasticity may be critical for certain forms of hippocampal learning. Moreover, the phenotypes described in the ephrin-A3 transgenic lines suggest that mechanisms promoting ephrin-A3 reverse signaling may contribute to lowering glutamate transporter expression, which results in glutamate excitotoxicity27–29. Dysfunction of glial glutamate transporters is implicated in the pathology of various neurological and neurodegenerative diseases, such as epilepsy and amyotrophic lateral sclerosis (ALS)3,22. Notably, the human VAPB protein, a diffusible ligand for Eph receptors and potential antagonist for ephrins,
is associated with familial ALS43. Mutations of VAPB might enhance EphA4–ephrin-A3 signaling, thereby causing the downregulation of glial glutamate transporters observed in ALS.
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1291
Methods Methods and any associated references are available in the online version of the paper at http://www.nature.com/natureneuroscience/. Note: Supplementary information is available on the Nature Neuroscience website. Acknowledgments We thank M. Bösl and the transgenic core facility for generating transgenic mice; E. Kandel (Columbia University) and F. Kirchhoff (Max Planck Institute of Experimental Medicine, Göttingen) for transgenic mice; M. Klein and O. Gökce for technical help; K. Deininger, C. Erlacher, V. Staiger, V. Stein and M. Traut for scientific input and suggestions; M. Korte, I. Kadow, V. Stein, J. Egea and R. Fonseca for critical comments on the manuscript. S.P. was supported by a postdoctoral fellowship from Fundação para a Ciência e Tecnologia of Portugal, co-funded by Programa Operacional Ciência e Inovação 2010 and Fundo Social Europeu. M.A.C. was supported by a fellowship from Fundación Española para la Ciencia y la Tecnología. This work was in part supported by grants from the European Union (Endotrack), the Deutsche Forschungsgemeinschaft (SPP1172) and the Max-Planck Society (all to R.K.), the Wellcome Trust and the Biotechnology and Biological Sciences Research Council, UK (R.S.), the German National Genome Research Network (NGF N grant 01GR0430) (T.K.), and US National Institutes of Health grant HD025938 (E.B.P.). AUTHOR CONTRIBUTIONS A.F. designed, performed, analyzed most of the electrophysiology experiments and co-wrote the manuscript. S.P. designed, performed, analyzed the biochemical and quantitative anatomical studies and co-wrote the manuscript. S.D.H. and C.R.R. designed, performed and analyzed the astrocyte patch clamp recordings. M.A.C. and E.B.P. provided the Efna3−/− model, gave advice and aided in the interpretation of data. L.B., B.F. and T.K. performed and analyzed the induced seizure experiments. L.G. performed biochemical studies. Y.R. and R.S. provided the CA3-Cre mouse. K.K. provided Epha4lx/+ ES cells. R.K. supervised the project, designed experiments and co-wrote the manuscript. The two first authors, who contributed equally, are listed in alphabetical order. Published online at http://www.nature.com/natureneuroscience/. Reprints and permissions information is available online at http://www.nature.com/ reprintsandpermissions/. 1. Allen, N.J. & Barres, B.A. Signaling between glia and neurons: focus on synaptic plasticity. Curr. Opin. Neurobiol. 15, 542–548 (2005). 2. Barres, B.A. The mystery and magic of glia: a perspective on their roles in health and disease. Neuron 60, 430–440 (2008). 3. Halassa, M.M., Fellin, T. & Haydon, P.G. The tripartite synapse: roles for gliotransmission in health and disease. Trends Mol. Med. 13, 54–63 (2007). 4. Bergles, D.E. & Jahr, C.E. Glial contribution to glutamate uptake at Schaffer collateral-commissural synapses in the hippocampus. J. Neurosci. 18, 7709–7716 (1998). 5. Tzingounis, A.V. & Wadiche, J.I. Glutamate transporters: confining runaway excitation by shaping synaptic transmission. Nat. Rev. Neurosci. 8, 935–947 (2007). 6. Pita-Almenar, J.D., Collado, M.S., Colbert, C.M. & Eskin, A. Different mechanisms exist for the plasticity of glutamate reuptake during early long-term potentiation (LTP) and late LTP. J. Neurosci. 26, 10461–10471 (2006). 7. Genoud, C. et al. Plasticity of astrocytic coverage and glutamate transporter expression in adult mouse cortex. PLoS Biol. 4, e343 (2006). 8. Essmann, C.L. et al. Serine phosphorylation of ephrinB2 regulates trafficking of synaptic AMPA receptors. Nat. Neurosci. 11, 1035–1043 (2008). 9. Klein, R. Bidirectional modulation of synaptic functions by Eph/ephrin signaling. Nat. Neurosci. 12, 15–20 (2009). 10. Kayser, M.S., Nolt, M.J. & Dalva, M.B. EphB receptors couple dendritic filopodia motility to synapse formation. Neuron 59, 56–69 (2008). 11. Murai, K.K., Nguyen, L.N., Irie, F., Yamaguchi, Y. & Pasquale, E.B. Control of hippocampal dendritic spine morphology through ephrin-A3/EphA4 signaling. Nat. Neurosci. 6, 153–160 (2003). 12. Grunwald, I.C. et al. Hippocampal plasticity requires postsynaptic ephrinBs. Nat. Neurosci. 7, 33–40 (2004). 13. Grunwald, I.C. et al. Kinase-independent requirement of EphB2 receptors in hippocampal synaptic plasticity. Neuron 32, 1027–1040 (2001). 14. Henderson, J.T. et al. The receptor tyrosine kinase EphB2 regulates NMDAdependent synaptic function. Neuron 32, 1041–1056 (2001). 15. Dalva, M.B. et al. EphB receptors interact with NMDA receptors and regulate excitatory synapse formation. Cell 103, 945–956 (2000).
© 2009 Nature America, Inc. All rights reserved.
a r t ic l e s 16. Bouzioukh, F. et al. Tyrosine phosphorylation sites in ephrinB2 are required for hippocampal long-term potentiation but not long-term depression. J. Neurosci. 27, 11279–11288 (2007). 17. Kullander, K. et al. Kinase-dependent and kinase-independent functions of EphA4 receptors in major axon tract formation in vivo. Neuron 29, 73–84 (2001). 18. Tsien, J.Z. et al. Subregion- and cell type-restricted gene knockout in mouse brain. Cell 87, 1317–1326 (1996). 19. Carmona, M.A., Murai, K.K., Wang, L., Roberts, A.J. & Pasquale, E.B. Glial ephrinA3 regulates hippocampal dendritic spine morphology and glutamate transport. Proc. Natl. Acad. Sci. USA 106, 12524–12529 (2009). 20. Tremblay, M.E. et al. Localization of EphA4 in axon terminals and dendritic spines of adult rat hippocampus. J. Comp. Neurol. 501, 691–702 (2007). 21. Albensi, B.C., Oliver, D.R., Toupin, J. & Odero, G. Electrical stimulation protocols for hippocampal synaptic plasticity and neuronal hyper-excitability: are they effective or relevant? Exp. Neurol. 204, 1–13 (2007). 22. Beart, P.M. & O’Shea, R.D. Transporters for L-glutamate: an update on their molecular pharmacology and pathological involvement. Br. J. Pharmacol. 150, 5–17 (2007). 23. Liu, G., Choi, S. & Tsien, R.W. Variability of neurotransmitter concentration and nonsaturation of postsynaptic AMPA receptors at synapses in hippocampal cultures and slices. Neuron 22, 395–409 (1999). 24. Oliet, S.H., Piet, R. & Poulain, D.A. Control of glutamate clearance and synaptic efficacy by glial coverage of neurons. Science 292, 923–926 (2001). 25. Tsukada, S., Iino, M., Takayasu, Y., Shimamoto, K. & Ozawa, S. Effects of a novel glutamate transporter blocker, (2S, 3S)-3-[3-[4-(trifluoromethyl)benzoylamino]benz yloxy]aspartate (TFB-TBOA), on activities of hippocampal neurons. Neuropharmacology 48, 479–491 (2005). 26. Nolte, C. et al. GFAP promoter-controlled EGFP-expressing transgenic mice: a tool to visualize astrocytes and astrogliosis in living brain tissue. Glia 33, 72–86 (2001). 27. Rothstein, J.D. et al. Knockout of glutamate transporters reveals a major role for astroglial transport in excitotoxicity and clearance of glutamate. Neuron 16, 675–686 (1996). 28. Tanaka, K. et al. Epilepsy and exacerbation of brain injury in mice lacking the glutamate transporter GLT-1. Science 276, 1699–1702 (1997). 29. Watanabe, T. et al. Amygdala-kindled and pentylenetetrazole-induced seizures in glutamate transporter GLAST-deficient mice. Brain Res. 845, 92–96 (1999). 30. Greenwood, S.M. & Connolly, C.N. Dendritic and mitochondrial changes during glutamate excitotoxicity. Neuropharmacology 53, 891–898 (2007).
31. Feng, G. et al. Imaging neuronal subsets in transgenic mice expressing multiple spectral variants of GFP. Neuron 28, 41–51 (2000). 32. Hasbani, M.J., Schlief, M.L., Fisher, D.A. & Goldberg, M.P. Dendritic spines lost during glutamate receptor activation reemerge at original sites of synaptic contact. J. Neurosci. 21, 2393–2403 (2001). 33. Lu, Y.M. et al. Mice lacking metabotropic glutamate receptor 5 show impaired learning and reduced CA1 long-term potentiation (LTP) but normal CA3 LTP. J. Neurosci. 17, 5196–5205 (1997). 34. Mulholland, P.J. et al. Glutamate transporters regulate extrasynaptic NMDA receptor modulation of Kv2.1 potassium channels. J. Neurosci. 28, 8801–8809 (2008). 35. Wadiche, J.I. & Jahr, C.E. Patterned expression of Purkinje cell glutamate transporters controls synaptic plasticity. Nat. Neurosci. 8, 1329–1334 (2005). 36. Nikkuni, O., Takayasu, Y., Iino, M., Tanaka, K. & Ozawa, S. Facilitated activation of metabotropic glutamate receptors in cerebellar Purkinje cells in glutamate transporter EAAT4-deficient mice. Neurosci. Res. 59, 296–303 (2007). 37. Levenson, J. et al. Long-term potentiation and contextual fear conditioning increase neuronal glutamate uptake. Nat. Neurosci. 5, 155–161 (2002). 38. Nishida, H. & Okabe, S. Direct astrocytic contacts regulate local maturation of dendritic spines. J. Neurosci. 27, 331–340 (2007). 39. Nestor, M.W., Mok, L.P., Tulapurkar, M.E. & Thompson, S.M. Plasticity of neuronglial interactions mediated by astrocytic EphARs. J. Neurosci. 27, 12817–12828 (2007). 40. Diamond, J.S. & Jahr, C.E. Synaptically released glutamate does not overwhelm transporters on hippocampal astrocytes during high-frequency stimulation. J. Neurophysiol. 83, 2835–2843 (2000). 41. Lehre, K.P. & Rusakov, D.A. Asymmetry of glia near central synapses favors presynaptically directed glutamate escape. Biophys. J. 83, 125–134 (2002). 42. Morris, R.G. et al. Elements of a neurobiological theory of the hippocampus: the role of activity-dependent synaptic plasticity in memory. Philos. Trans. R. Soc. Lond. B 358, 773–786 (2003). 43. Tsuda, H. et al. The amyotrophic lateral sclerosis 8 protein VAPB is cleaved, secreted, and acts as a ligand for Eph receptors. Cell 133, 963–977 (2008). 44. Weiergraber, M. et al. Altered seizure susceptibility in mice lacking the Ca(v)2.3 E-type Ca2+ channel. Epilepsia 47, 839–850 (2006).
1292
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
© 2009 Nature America, Inc. All rights reserved.
ONLINE METHODS Mice. To generate the CA3-cre line, a yeast artificial chromosome containing the kainate receptor 1 (Grik4) gene was modified to carry a targeting cassette in the 3′ untranslated region of Grik4. The targeting cassette consisted of an IRES element, EGFP-cre and frt-flanked selection markers neo and his3, which were removed by crossing CA3-cre mice with Flp deleter mice. Cre recombination was investigated by crossing CA1-cre and CA3-cre lines to the lacZ reporter line Rosa26. Efna3 knockout mice were crossed with Flp deleter mice to remove the neo selection gene. To generate ephrin-A3 transgenic lines, the human Efna3 cDNA (nucleotides 137–787, RefSeq DNA NM_004952.4) was cloned in frame with a hemagglutinin tag at its N terminus, under the human glia-specific promoter GFAP. The ApaLI–DraIII fragment containing the Efna3 transgene was isolated by electroelution and purified with Elutip-D minicolumns (Schleicher & Schüll). Oocytes from FVB mice were microinjected using conventional technology. Founder lines were mated with C57BL/6 mice. Heterozygous offspring were screened for hemagglutinin expression. Mice were genotyped by PCR using GFAP forward (5′-CAGAGCAGGTTGGAGAGG AG-3′) and Efna3 reverse primers (5′-GCGGGCAGTAAATATCCAGA-3′). Tg181 and Tg7047 were crossed to the transgenic indicator mice Thy1-GFP (line GFPM) and GFAP-GFP. Heterozygous mice were used for all experiments. All mutant lines were maintained in a heterozygous state on a C57BL/6, 129 × C57BL/6 or FVB × C57BL/6 background. Experimental procedures and animal care were carried out in accordance with the European Community Council Directive of 24 November 1986 (86/609/EEC) and approved by the ethics committee in charge of animal experimentation at Helmholtz-Zentrum München and the Regierung von Oberbayern (number 209.1/211-2531-70/07). In situ hybridization and densitometric analysis. In situ hybridization was carried according to standard procedures. A fragment extending from nucleotides 880–1551 of Epha4 cDNA (GenBank accession number NM_007936) was used as riboprobe. For quantification of Cre recombination efficiency, densitometric measurements of mRNA expression was performed using ImageJ software. Three equally sized areas were selected in the pyramidal layers of CA1 and CA3 in every digital picture of Epha4 in situ hybridizations. The signal intensities were measured and the values were corrected for background, determined as the signal intensity in the stratum radiatum of CA1. The relative abundance of Epha4 mRNA was expressed as CA3/CA1 or CA1/CA3 ratios. The recombination efficiency was calculated as a ratio of CA1/CA3 values in CA1-cre;Epha4lx/– versus Epha4lx/– slices for the CA1-cre;Epha4lx/– mice and as a ratio of CA3/CA1 values in CA3-cre;Epha4lx/– versus Epha4lx/– slices for the CA3-cre;Epha4lx/– mice. Slice electrophysiology. For extracellular field recordings, mice (P40–P80) were decapitated under diethylether anesthesia. Hippocampi were dissected in cold artificial cerebrospinal fluid (ACSF: 124 mM NaCl, 3 mM KCl, 1.25 mM KH2PO4, 2.5 mM CaCl2, 2 mM MgSO4, 26 mM NaHCO3, 10 mM glucose, saturated with 95% O2 and 5% CO2). Slices (400 µm thick) were prepared using a custom-made tissue slicer and maintained in ACSF at 24–26 °C for 1.5–2 h before recording in ACSF at 33 °C. Input–output relations were measured at 25 °C. Synaptic responses were evoked by stimulating Schaffer collaterals with 0.2 ms pulses using monopolar tungsten electrodes. fEPSPs were recorded in the stratum radiatum of the CA1 region using glass microelectrodes filled with 3 M NaCl (5−15 MΩ). For baseline recordings, slices were stimulated at 0.1 Hz for 20 min at stimulation intensities of 40%–50% of the highest measured fEPSP size. LTP was induced by applying a tetanus (100 Hz for 1 s) or a TBS consisting of three bursts (10 s interval), each composed of ten trains (5 Hz) with four pulses (100 Hz). PPF was tested by applying two pulses with ISIs ranging from 10 ms to 160 ms. An Axoclamp 2B amplifier (Axon Instruments) was used for the experiments. Data were sampled at 5 kHz and analyzed using a program written in LabView (National Instruments). Whole-cell patch-clamp recordings were performed in 400-µm-thick acute hippocampal slices obtained from P12–P28 mice using a vibratome. The recording chamber was perfused with ACSF containing 100 µM picrotoxin (PTX) and 200 nM tetrodotoxin (TTX). Recordings were performed at 33 °C. Borosilicate patch pipettes (resistance 2–7 MΩ) were filled with internal solution containing 150 mM cesium gluconate, 10 mM HEPES, 2 mM Mg-ATP, 0.2 mM EGTA, 8 mM NaCl, 290 milliosmoles per kilogram, pH 7.2. Spontaneous mEPSCs were recorded in whole-cell configuration at −70 mV. mEPSCs were recorded first in ACSF containing PTX and TTX and then in ACSF containing PTX, TTX and 1 mM γ- D-glutamylglycine (γ-DGG). AMPA and NMDA currents in CA1 pyramidal cells
doi:10.1038/nn.2394
were evoked by stimulating Schaffer collaterals with a monopolar tungsten electrode. AMPA currents were recorded in voltage-clamp mode at a membrane potential of −70 mV and peak amplitudes of the currents were measured. NMDA currents were recorded at +40 mV and measured 70 ms after stimulus to avoid the initial mixed AMPA/NMDA component. A Digidata 1440A digitizer and a Multiclamp 700B (Axon Instruments) were used for the experiments. Data were collected and analyzed using pCLAMP 10 Software (Axon Instruments). Astrocyte patch clamp recordings were performed at 24–26 °C in 300-µm-thick slices of P18–P24 mice with an EPC10 Double amplifier (HEKA Electronics). Pipettes (resistance 2–4 MΩ) were filled with intracellular solution containing (in mM): 120 KCH3SO3, 32 KCl, 10 HEPES, 4 NaCl, 4 Mg-ATP and 0.4 Na-GTP, pH 7.30. During experiments, slices were perfused with ACSF containing (in mM): 125 NaCl, 2.5 KCl, 1.25 NaH2PO4, 26 NaHCO3, 2 CaCl2, 1 MgCl2 and 20 glucose, bubbled with 95% O2 and 5% CO2, pH 7.4. Bicuculline (20 µM) and 2.5 mM kynurenic acid were added to the ACSF to suppress activation of ionotropic glutamate as well as GABAA receptors. Astrocytes were held in voltage-clamp mode at a membrane potential of −85 mV. Recordings were discarded when the series resistance exceeded 17 MΩ. Astrocytes were identified based on their membrane potential close to –85 mV, their high capacitance, low membrane resistance and passive current–voltage relationship upon application of depolarizing current pulses. Schaffer collaterals were stimulated with a tungsten concentric bipolar electrode (pulse duration, 0.1 ms). Currents were corrected for the slow current insensitive to the glutamate transporter blockers L-trans-pyrrolidine-2,4-dicarboxylic acid (PDC, 300 µM) and dihydrokainic acid (DHK, 300 µM), likely reflecting accumulation of extracellular potassium45. Extracellular field potentials for fiber volley detection were recorded with ACSF-filled glass microelectrodes at a maximal distance of 50 µm from the recorded astrocyte. The stimulation intensity was adjusted such that fiber volley amplitudes were between −0.15 and −0.4 mV. Evoked glial transporter currents were normalized by dividing their amplitude by the amplitude of the corresponding fiber volley. For each stimulation intensity, 5–10 (single pulse experiments) or 3 (TBS) recordings were averaged. Data were collected at 10 KHz and were analyzed with IgorPro (Wavemetrics). All measurements were carried out and analyzed by an investigator blind to the genotype. Western blot analysis and immunoprecipitation. Tissues were homogenized in ice-cold buffer (50 mM Tris-HCl, pH 7.4, 150 mM NaCl, 1% Triton X-100) supplemented with protease inhibitor cocktail tablets and phosphatase inhibitor cocktail tablets (Roche). The lysates were separated by SDS-PAGE and probed by immunoblotting with the following primary antibodies: EphA4 mouse monoclonal (anti-Sek, Becton Dickinson Biosciences), 1:1,000; EphA4 rabbit polyclonal serum 1383 (raised against an intracellular peptide46), 1:1,000; EphA4 globular domain rabbit polyclonal serum 1078 (ref. 46), 1:1,000; GFAP rabbit polyclonal (DAKO), 1:2,000; GFAP mouse monoclonal (Sigma), 1:1,000; GLAST guinea pig polyclonal (Chemicon), 1:5,000; GLT1 guinea pig polyclonal (Chemicon), 1:2,000; EAAC1 mouse monoclonal (Chemicon), 1:500; hemagglutinin rat monoclonal clone 3F10 (Sigma), 1:2,000; phosphotyrosine (pTyr, hybridoma clone 4G10), 1:10,000; tubulin mouse monoclonal clone DM 1A (Sigma), 1:50,000. Horseradish peroxidase (HRP)-conjugated secondary antibodies (Amersham Biosciences), 1:5,000. Detection was performed with luminol (Santa Cruz). For immunoprecipitations, hippocampal lysates were incubated with Protein A–Sepharose beads (Amersham) preconjugated with EphA4 serum 1383. Western blots were quantified using ImageJ software. Optical density values were normalized to tubulin signal. Glutamate transporter proteins tend to form homomultimers47. For GLAST a more prominent band of 120 kDa was quantified, probably corresponding to the dimeric form. For GLT1 a band of 70 kDa was quantified, corresponding to the monomeric form. Immunofluorescence and immunohistochemistry. Mice were perfused with 4% paraformaldehyde in phosphate-buffered saline (PBS). Brains were postfixed overnight and further processed or cryoprotected in 30% sucrose before freezing. For immunofluorescence, vibratome free-floating sections (50 µm thick) were permeabilized with 0.5% TritonX-100 in PBS, blocked in 0.1% TritonX-100, 5% BSA and 5% donkey serum in PBS for 1 h at 24–26 °C and incubated with primary antibodies overnight at 4 °C in blocking solution: GLAST rabbit polyclonal (Frontier Science), 1:5,000; GLT1 guinea pig polyclonal (Chemicon), 1:5,000; hemagglutinin rat monoclonal clone 3F10 (Sigma), 1:1,000; GFAP rabbit
nature NEUROSCIENCE
© 2009 Nature America, Inc. All rights reserved.
polyclonal (DAKO), 1:1,000; GFP rabbit polyclonal (RDI), 1:1,000; GFP mouse monoclonal antibody (Molecular Probes), 1:2,000. The sections were washed three times in PBS with 0.1% Triton X-100 for 30 min and incubated for 1 h at 24–26 °C with donkey Cy2, Cy3 and Cy5-conjugated secondary antibodies (Jackson ImmunoResearch), diluted 1:800 in blocking solution, mounted using aqueous mounting medium with antifading reagent (Biømeda). The Vectastain ABC kit (Vector Laboratories) was used for immunohistochemistry, the staining procedure was adapted according to provider’s instructions. Assessment of dendrite morphology. To visualize hippocampal neurons in vivo, P29–P31 mice carrying the Thy-GFP transgene were perfused, hippocampi were dissected, and 200-µm-thick, coronal vibratome sections were collected. Alternatively, 400-µm hippocampal slices were prepared using a custom-made tissue slicer and labeled with DiI using a biolistic method. Fluorescence images from apical dendrites were acquired in the stratum radiatum of the CA1 region with a TCS SP2 confocal microscope (Leica). Quantification of dendritic spines and swellings were done on the thinnest dendrites (0.3–0.5 µm diameter) in single confocal z-sections. Swellings were defined as enlargements of the mean dendritic stalk width. All quantitative measurements were performed using MetaMorph software (Molecular Devices). For analysis of spine morphology in cultured slices, two stacks of pictures per slice in the stratum radiatum of the CA1 region were acquired. To automatically delineate the dendrites, the stack projections were quantified with an automatic counting-grid macro implemented in the MetaMorph software (dendrite complexity index)48. All image analyses were done by an investigator blind to the genotype. Organotypic hippocampal slice cultures. Hippocampal slices (300 µm thick) were prepared from P6 progeny of Tg181 mice crossed to GFP-M mice and maintained using the Muller technique49. Slices were kept in culture for 5 d before the experiments. Glutamate exposure. Slice cultures were exposed to glutamate at concentrations of 0–120 µM glutamate for 1 h at 35 °C and then immediately fixed with 2% paraformaldehyde, 15% glucose. Glutamate was dissolved in water and applied to the culture medium. PTZ treatment. Mice were injected intraperitoneally with 45 mg per kilogram body weight PTZ (Sigma), diluted in saline. After injection, mice were carefully monitored and videotaped.
nature NEUROSCIENCE
Quantitative RT-PCR analysis. Total RNA was isolated from hippocampi of P60 mice using Rneasy Mini Kit (Qiagen) according to the manufacturer’s instructions. Total RNA (100 ng) was reverse transcribed and amplified using the One-Step RT-PCR system LightCycler RNA Master SYBR Green I (Roche Applied Science) in triplicates, using the LightCycler (Roche Applied Science). QuantiTect Primer Assays (Qiagen) were used for Slc1a3 (Mm_Slc1a3_1_SG), Slc1a2 (Mm_Slca2_1_SG), GFAP (Mm_Gfap_1_SG) and Gapdh (Mm_Gapd_ 2_SG). For Efna3, primers were designed common to the mouse and human sequence RT-A3exon2/3.F (5′-GTTCTCCGAGAAGTTCCAGC-3′) and RT-A3exon2/3.R (5′-CAGTGCAGGTTGTGAGTGGG-3′). The threshold cycle Ct was determined manually. Taking into account that primers used had the same amplification efficiency, experiments were analyzed by relative quantification using the 2–∆∆Ct method50. Statistical analysis. Statistical significance was determined using two-tailed Student’s t-tests in Microsoft Excel, two-way analysis of variance (ANOVA) with repetitions using the software Statistics (Blackwell Scientific Publications), analysis of covariance (ANCOVA) from the VasserStats statistical computations website (http://faculty.vassar.edu/lowry/VassarStats.html) and two-sample KolmogorovSmirnov tests using the Statistics to Use website (http://www.physics.csbsju. edu/stats/KS-test.html; T.W. Kirkman, 1996). The Mann-Whitney U-test was performed using SPSS software (release 11.0; SPSS Inc.). All values in the text and in the figure legends indicate mean ± s.e.m. All error bars in the graphs represent s.e.m. 45. Luscher, C., Malenka, R.C. & Nicoll, R.A. Monitoring glutamate release during LTP with glial transporter currents. Neuron 21, 435–441 (1998). 46. Egea, J. et al. Regulation of EphA 4 kinase activity is required for a subset of axon guidance decisions suggesting a key role for receptor clustering in Eph function. Neuron 47, 515–528 (2005). 47. Haugeto, O. et al. Brain glutamate transporter proteins form homomultimers. J. Biol. Chem. 271, 27715–27722 (1996). 48. Kramer, E.R. et al. Absence of Ret signaling in mice causes progressive and late degeneration of the nigrostriatal system. PLoS Biol. 5, e39 (2007). 49. Stoppini, L., Buchs, P.A. & Muller, D. A simple method for organotypic cultures of nervous tissue. J. Neurosci. Methods 37, 173–182 (1991). 50. Livak, K.J. & Schmittgen, T.D. Analysis of relative gene expression data using realtime quantitative PCR and the 2(-delta delta C(T)) method. Methods 25, 402–408 (2001).
doi:10.1038/nn.2394
a r t ic l e s
Nicotine activates the chemosensory cation channel TRPA1
© 2009 Nature America, Inc. All rights reserved.
Karel Talavera1, Maarten Gees1, Yuji Karashima1, Víctor M Meseguer2, Jeroen A J Vanoirbeek3, Nils Damann1,5, Wouter Everaerts1,4, Melissa Benoit1, Annelies Janssens1, Rudi Vennekens1, Félix Viana2, Benoit Nemery3, Bernd Nilius1 & Thomas Voets1 Topical application of nicotine, as used in nicotine replacement therapies, causes irritation of the mucosa and skin. This reaction has been attributed to activation of nicotinic acetylcholine receptors (nAChRs) in chemosensory neurons. In contrast with this view, we found that the chemosensory cation channel transient receptor potential A1 (TRPA1) is crucially involved in nicotineinduced irritation. We found that micromolar concentrations of nicotine activated heterologously expressed mouse and human TRPA1. Nicotine acted in a membrane-delimited manner, stabilizing the open state(s) and destabilizing the closed state(s) of the channel. In the presence of the general nAChR blocker hexamethonium, nociceptive neurons showed nicotine-induced responses that were strongly reduced in TRPA1-deficient mice. Finally, TRPA1 mediated the mouse airway constriction reflex to nasal instillation of nicotine. The identification of TRPA1 as a nicotine target suggests that existing models of nicotine-induced irritation should be revised and may facilitate the development of smoking cessation therapies with less adverse effects. Nicotine is a powerful psychoactive drug inducing an addiction that kills about five million people per year as consequence of the noxious effects of tobacco smoke1. In addition, nicotine elicits taste and smell sensations and, at increasing concentrations, it produces strong burning, stinging and pain2. Notably, all variants of smokingcessation therapies that are based on nicotine replacement produce local irritation side effects1,3,4, which has been suggested to reduce treatment compliance and efficacy5. It is currently assumed that the irritant effects of nicotine are exclusively mediated by nAChRs expressed in nerve fibers that convey painful stimuli from the skin and mucosa6–8. However, some observations are not completely consistent with nAChRs being the sole targets of nicotine. For example, nAChRs quickly desensitize under the high local concentrations of nicotine used in replacement therapies (up to 60 mM), but the irritating effects of nicotine are long lasting. The TRP superfamily of cation channels is important in chemical nociception9–11, but no TRP channel has been shown to be directly activated by nicotine. However, nicotine seems to modulate the activity of the vanilloid receptor TRPV1, sensitizing it to capsaicin stimulation12 and desensitizing it after nicotine-induced activation of nAChRs in sensory neurons13. We found that nicotine actually inhibited hTRPV1 (Supplementary Fig. 1), indicating that TRPV1 is not directly involved in the irritation caused by nicotine. The emerging role of the ankyrin-rich channel TRPA1 as a broadly tuned chemosensor14 prompted us to test whether this channel contributes to the irritating effects of nicotine. TRPA1 is expressed in a subset of polymodal nociceptive neurons15, determines behavioral
responses to noxious cold16 and to multiple irritant substances such as mustard oil (allyl isothiocyanate)17,18 and the unsaturated aldehydes contained in cigarette smoke19, and is involved in inflammatory pain20,21. We found that TRPA1 is activated by nicotine and is crucial for the airway constriction reflex caused by application of this compound to the nasal mucosa. RESULTS Nicotine activates TRPA1 Extracellular application of nicotine to mTRPA1-expressing Chinese hamster ovary (CHO) cells induced a reversible increase of currents (Fig. 1a–c), which was not observed in control cells (Supplementary Fig. 2). Stimulation with the nicotine analog anabasine induced a very similar mTRPA1 current activation (Supplementary Fig. 3). Cell-attached patch-clamp recordings (Supplementary Fig. 4) and Fura-2–based measurements of the intracellular Ca2+ concentration (Supplementary Fig. 5) further confirmed that nicotine activated mTRPA1. In addition, nicotine activated hTRPA1 that we transfected into CHO cells (Supplementary Fig. 2). We consistently found that the magnitude of the nicotine-activated TRPA1 current declined above ~300–1,000 µM, particularly at positive potentials (Fig. 1b–d). Notably, the inhibitory effect reversed much faster than the activating effect, which explains the current rebound observed on washout of high doses of nicotine. This differential kinetics allowed us to dissect the activating and inhibitory effects of nicotine (Supplementary Fig. 6), yielding a maximal relative current increase of 4.0 ± 0.3 and 2.7 ± 0.2 at −75 and +50 mV,
1Laboratory for Ion Channel Research, Department of Molecular Cell Biology, KU Leuven, Leuven, Belgium. 2Instituto de Neurociencias de Alicante, Universidad Miguel Hernández–Consejo Superior de Investigaciones Científicas, San Juan de Alicante, Spain. 3Laboratory of Lung Toxicology, KU Leuven, Leuven, Belgium. 4Laboratory of Experimental Urology, Department of Surgery, KU Leuven, Leuven, Belgium. 5Present address: Molecular Pharmacology, Grünenthal GmbH, Aachen, Germany. Correspondence should be addressed to K.T. (
[email protected]).
Received 18 May; accepted 6 July; published online 13 September 2009; doi:10.1038/nn.2379
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1293
a r t ic l e s
Current (nA)
a
Nicotine 1 µM
0.5
–0.5 +50 mV –75 mV
Current (nA)
2
Current (nA)
3
4
0
Nicotine
Nicotine + menthol
b
g (pS)
*
40 50 mV
0
Outward
Inward
Menthol 1 mM 10
0
C In-out N
0
1294
M +75 mV −75 mV
Nicotine 1 mM
–20 1
50
0
50
0
50
100
(nA)
0 NMDG
–2 NMDG+
2
4
Time (min) 6
8
(mV)
–100
–50
100
Nicotine 1 mM 1
2 (nA) 1
0
0
–1
NMDG+
–2
Time (min) 0
1
6
2
3
4
–2 (mV)
–3 –100
e 10
+50 mV –75 mV
5
–1
4
–50
100
(µM)
103
4
102
3 2
101
1 1
10 100 1,000 10,000 Nicotine (µM)
100
EC50
IC50
10 pA
c
–10
0
membrane. Nicotine caused a significant reduction of the single channel conductance at positive potentials, from 114 ± 7 pS in controls to 80 ± 6 pS in the presence of 1 mM nicotine (n = 5, P = 0.004). Taken together, these data indicate that nicotine interacts with TRPA1 in a membrane-delimited manner. Activation of TRPA1 by electrophiles, such as mustard oil, cinnamaldehyde and acrolein, involves covalent modification of N-terminal cysteine residues27,28. Thus, we tested nicotine’s effect on an mTRPA1 mutant in which the critical cysteine residue C622 is mutated to serine. Consistent with previous work27,28, this mutant was unresponsive to mustard oil (20 µM), but was activated by 1 mM nicotine (Supplementary Fig. 7). These results indicate that nicotine-induced
120 80
–50
+
–1
0.1 Control
(mV)
–100
2
0
a
–1.0
Nicotine 1 mM Nicotine 30 µM
1
0
d
Time (min)
1
2
–2
c
(nA) 0.0
0.0
0
b
NMDG+ 0.5
–0.5
Relative current
r espectively, and the concentrations for half-maximal activation (EC50; Fig. 1e) and half-maximal inhibition (IC50; Fig. 1e). As the response to nicotine occurred with a notable delay in wholecell (Fig. 1a–c) and cell-attached (Supplementary Fig. 4) patch-clamp experiments, as well as in intracellular Ca2+-imaging experiments (Supplementary Fig. 5), we predicted that nicotine acts indirectly on the channel via a mechanism involving other membrane receptors and/or intracellular signaling pathways. To investigate this possibility, we tested the action of nicotine application to the intracellular side of the membrane in cell-free inside-out patches in the absence of extra- and intracellular Ca2+. Under this condition, 1 mM nicotine was still able to activate single mTRPA1 channels. These channels had a conductance of 112 ± 8 pS at −75 mV (n = 5) and were reversibly blocked by 1 mM menthol (Fig. 2), consistent with previous studies22–26. Notably, nicotine-induced activation was substantially faster in inside-out patches than in whole-cell recordings. This indicates that the delay in TRPA1 activation on extracellular application of nicotine reflects, at least partly, the rate of nicotine diffusion across the plasma
Current (pA)
© 2009 Nature America, Inc. All rights reserved.
Figure 1 Activation of TRPA1 by nicotine. (a–c) Time course of the effects of nicotine on the amplitude of mTRPA1 currents measured at −75 and +50 mV. The horizontal lines indicate the periods of extracellular application of nicotine at the indicated concentrations. The colored data points correspond to the current traces shown to the right. (d) Dosedependent modulation of mTRPA1 currents by nicotine. Data points represent the relative change of the current amplitude with respect to the values in the control condition. The continuous lines represent the fit with a bimodal function comprising stimulatory and inhibitory components (see Online Methods). Error bars represent s.e.m. (e) EC50 and IC50 of nicotine on mTRPA1 currents at −75 mV (solid columns) and +50 mV (empty columns). Error bars represent fitting errors.
2
3 4 Time (min)
5
6
Figure 2 Nicotine activates TRPA1 in cell-free inside-out patches. (a) Examples of single mTRPA1 channel currents elicited by a voltage ramp from −100 to +100 mV in the control condition (left) or with the addition of 1 mM nicotine (center) or 1 mM nicotine plus 1 mM menthol (right). Note that in the upper panels the single channel conductance at positive potentials was larger in the control condition than it was in the presence of nicotine (solid line versus dashed line). Menthol application produced strong current inhibition, showing the typical flickering pattern of block23. (b) Average single channel conductance for the inward (black) and outward (gray) currents in control (solid) and in the presence of 1 mM nicotine (empty, n = 5). Error bars represent s.e.m. The asterisk indicates statistically significant difference from control (P = 0.004). (c) Time course of the amplitudes of currents at −75 and +75 mV. The arrow indicates the moment of patch excision. The labels C, N and M mark the time points corresponding to the traces shown in a: C is the control, N is nicotine addition, and M is the addition of both nicotine and menthol.
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
a r t ic l e s
–5
0 –100
0
0
2
3 4 Time (min)
10 5 1 0
1 5 [Mec] (mM)
–2
NMDG+ NMDG+ 1
2 3 4 Time (min)
e
1.5 1.0 0.5 0.1 1 [Mec] (mM)
5
1
MO 20 µM
10
100
–4 0
0
(mV) 0
0
CHO
0
(nA)
–100
2
5
d
CHO
Hex 3 mM
5
Trigeminal neurons 3 2 1 0
0
1 5 [Mec] (mM)
(nA)
0
5
–1
Trigeminal neurons 3 2 1
O
O
0
M
(dRatio/dt )MAX (min–1)
50
ex
ex H
ex
+ H
ex H
O
0
O
O
0.5
0
H
M
O
ex
1
1.0
M
5
1.5
ex
10
i
CHO (dRatio/dt )MAX (min–1)
15
3.0
ex
h
CHO
H
2.5
M
1.5 2.0 Time (min)
H
g
1.0
M
0.5
+
0
(mV)
H
+50 mV –75 mV
–5
(nA) 5 0 –5 –10 –100 –50
+
0
M
Normalized current
100
1
c
Current (nA)
+50 mV –75 mV
(mV)
Current (nA)
10 (nA)
(dRatio/dt)MAX (min–1)
Current (nA)
Mec 5 mM
5 0 –5
MO 20 µM
4
0
–10 –10
f
b
MO 20 µM
5
(dRatio/dt)MAX (min–1)
a
Normalized current
© 2009 Nature America, Inc. All rights reserved.
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
Current (nA)
activation does not involve modification of cysteine 622, which is fully in accordance with the nonelectrophilic character of nicotine. Activation of TRPA1 by other nonelectrophilic chemicals, such as menthol, clotrimazole and nifedipine, or by cold temperatures involves a shift of the voltage dependence of channel activation toward negative voltages23,25,26. To examine whether nicotine acts in a similar manner, we applied a voltage-step protocol that allowed us to assess the voltage dependence of activation from the measurement of peak tail currents. In these experiments, Ca2+ was omitted from the extracellular solution and 2 mM EDTA was added to avoid any interference of Ca2+ ions with the effect of nicotine on TRPA1 activation29,30. In the absence of nicotine, clear tail currents were only observed at prepulse potentials greater than +50 mV (Supplementary Fig. 8). In the presence of 300 µM nicotine, tail currents were already obvious at prepulse potentials above 0 mV. We fitted average tail currents in
Current (nA)
0 –10 –20
c
b
MO 100 µM 10
+50 mV –75 mV 0
1
2
3 4 Time (min)
5
d
MO 100 µM 10
Nicotine 100 µM
0 –10 –20
0
1
2
3 4 Time (min)
5
6
Nicotine 100 µM
10
MO 100 µM
0 –10 –20
6
Current (nA)
a Current (nA)
Figure 3 Cross desensitization of TRPA1 activation by nicotine and mustard oil. (a,b) Examples of the effect of 100 µM mustard oil (MO) on the amplitude of mTRPA1 currents at −75 and +50 mV without (a) or with (b) pre-application of 100 µM nicotine. (c) Nicotine did not affect mTRPA1 currents after stimulation with 100 µM mustard oil. (d) Maximal average current amplitudes at −75 and +50 mV elicited by mustard oil without (n = 8) or with (n = 7) pre-application of nicotine and by nicotine without (n = 7) or with (n = 5) pre-application of mustard oil. Error bars represent s.e.m.
0
1
2
3 4 Time (min)
5
6
+50 mV –75 mV
10 0 –10 –20
MO
MO Nicotine after Nicotine after nicotine MO
the presence of 300 µM nicotine with a Boltzmann function (see Online Methods), which yielded a voltage for half maximal activation (Vact) of 162 ± 4 mV and a slope factor (sact) of 36.9 ± 1.5 mV. In the absence of nicotine, tail currents were far from saturation at the most depolarizing prepulse potentials (+200 mV), which precluded the accurate independent estimation of Vact or sact. However, assuming a constant sact, the nicotine-induced changes in tail currents can be accounted for by a 51-mV shift of the activation curve toward negative voltages (Supplementary Fig. 8). In principle, such a leftward shift of the activation curve can be a result of a stabilization of the open state, a destabilization of the closed state or both. We observed that nicotine decreased the rate of whole-cell current deactivation at negative voltages and increased the rate of activation and very positive potentials (Supplementary Fig. 8). This is consistent with an increased mean open time and a reduced mean close time in the presence of nicotine (Supplementary Fig. 4). Oral application of mustard oil is known to prevent later stimulation with other irritant chemicals including nicotine, a phenomenon known as cross-desensitization31. Thus, we tested whether nicotineinduced activation of TRPA1 influences the effect of mustard oil and vice versa. Indeed, pre-activation with 100 µM nicotine significantly blunted (P < 0.01) a subsequent response to 100 µM mustard oil (Fig. 3). Conversely, pre-activation of mTRPA1 with mustard oil fully Figure 4 TRPA1 activation is prevented by the nAChR inhibitor mecamylamine, but is unaffected by hexamethonium. The colored data points correspond to the current traces shown in the inserts. (a,b) Effect of mustard oil on mTRPA1 currents, without (a) or with (b) pre-application of mecamylamine (Mec) in CHO cells. (c) Maximal mustard oil–induced currents (relative to the amplitude in control) at −75 mV in the absence (n = 7) and in the presence of 1 mM (n = 7) or 5 mM (n = 5) mecamylamine in CHO cells. (d,e) Maximal amplitude of the first time derivative of the intracellular Ca2+ signal elicited by 20 µM mustard oil in the absence and in the presence of mecamylamine in mTRPA1-expressing CHO cells (n = 22–35, d) or wild-type mouse trigeminal neurons (n = 12–18, e). (f) Amplitude of mTRPA1 currents at −75 and +50 mV during extracellular application of hexamethonium (Hex) and mustard oil in CHO cells. (g) Maximal amplitude of mTRPA1 currents (relative to the amplitude in control) at −75 mV in the presence of hexamethonium (n = 8), hexamethonium plus mustard oil (n = 8), and mustard oil (n = 7). (h,i) Maximal amplitude of the first time derivative of the intracellular Ca2+ signal elicited by 20 µM mustard oil in the absence or presence of 3 mM hexamethonium in mTRPA1-expressing CHO cells (n = 14 in mustard oil and n = 41 in mustard oil plus hexamethonium, h) or wild-type mouse trigeminal neurons (n = 28 in mustard oil and n = 35 in mustard oil plus hexamethonium, i). Error bars represent s.e.m.
1295
a r t ic l e s
0.8
b
a
2.0
Penh Wild type Trpa1 KO
1.5 Nic
1.0
1.5 HBSS 1.0
0.5 0
Penh
0.5
–5
c
0
5
10 15 20 25 30 (min)
Penh
1.5
0
d6 5
MO
1.0
4
–5
0
5
10 15 20 25 30 (min)
Penh Wild type Trpa1 KO
3 2
0.5
1 0
0 –5
1296
0
5
10 15 20 25 30 (min)
12.5
25
50
100
Methacholine (mg ml−1)
Nic 100 µM
2.5
3
6 9 Time (min)
K+
0.8
12
0
Caps 1 µM
Trpa1 KO
d
1.0
Ratio
K+
Nic 1 mM
1.5
MO 100 µM
0.5
3
6 9 12 Time (min)
15
18
Caps 1 µM
Wild type
2.0
2.0 Ratio
Nic 1 mM MO 100 µM
1.2
0 0
MO 100 µM
1.6
Nic 1 mM
1.2
Hex 3 mM
Nic 1 mM
0.8 0.4 0
0 0
e
Caps 1 µM
Wild type
0.4
0
c
2.0 1.6
3
6 9 Time (min)
Nic(+) MO(–) Nic(–) MO(+)
12
15
Nic(+) MO(+)
f Cumulative probability
Inhibition of TRPA1 by the nAChR blocker mecamylamine The mechanisms of nicotine-induced irritation have been extensively studied in several animal models and in humans. Notably, from the effects of mecamylamine, a general inhibitor of nAChRs, several studies have suggested that nAChRs are involved in nicotine-induced irritation2,7,32. However, reduction of the irritation caused by application of high concentrations of nicotine requires prolonged pre-application of mecamylamine at concentrations that are much higher than those needed to fully inhibit nAChRs in vitro (1–5 mM versus 30–100 µM, respectively)7,32,33. Because mecamylamine has structural similarities with camphor, a known inhibitor of TRPA1 (ref. 34), we hypothesized that it may also inhibit TRPA1, which would account, at least in part, for its inhibitory effect on the irritation caused by nicotine at high concentrations. We found that mecamylamine prevented the mustard oil–induced activation of mTRPA1 current with almost total inhibition at 5 mM (Fig. 4a–c). These experiments were repeated and the results confirmed in intact mTRPA1-expressing CHO cells (Fig. 4d) and cultured mouse trigeminal ganglion neurons (Fig. 4e). In addition, mecamylamine inhibited mTRPA1 currents pre-activated
b
0.4
Responding TGN (%)
abolished the response to nicotine. These data indicate that the sensory cross-desensitization between mustard oil and nicotine occurs, at least partly, at the level of the chemosensor TRPA1.
Caps 1 µM MO 100 µM K+
Wild type
1.2
Ratio
Ratio
a
60
Ba sa l Ve hi cl e
© 2009 Nature America, Inc. All rights reserved.
Figure 5 Nicotine activates TRPA1 in mouse trigeminal ganglion neurons. (a,b) Intracellular Ca2+-imaging experiments showing the effects of extracellular application of 100 µM (a) or 1 mM (b) nicotine (Nic) to trigeminal ganglion neurons isolated from wild-type mice. Caps, capsaicin. (c) Examples of nicotine’s effects on trigeminal ganglion neurons isolated from Trpa1 knockout (KO) mice. (d) Examples of nicotine’s effects on trigeminal ganglion neurons isolated from wild-type mice in the presence of the nAChR blocker hexamethonium. In a–d, blue represents cells sensitive to nicotine, but not to mustard oil (100 µM), dark gray represents cells that responded to both nicotine and mustard oil, orange represents mustard oil–sensitive cells that did not respond to nicotine and green represents cells that did not respond to nicotine or mustard oil. (e) Stack bar plot representing the proportions of neurons responding to mustard oil and nicotine (100 µM, 1 mM and 1 mM in the presence of 3 mM hexamethonium) in wild-type and Trpa1 knockout mice. The regions are color coded as in a–d. (f) Cumulative probability plot of the time to the maximal response elicited by nicotine in wild-type mustard oil–sensitive cells, wild-type cells in the presence of hexamethonium, wild-type mustard oil–insensitive cells and Trpa1 knockout cells. We took the duration of nicotine application (2 min) as the cutoff time for determining the time to peak.
40
20
0 WT
KO
WT
KO
Nic 100 µM Nic 1 mM
WT
KO
Nic 1 mM in Hex 3 mM
0
5
10 15 Time (min)
20
1.0 0.8 0.6 0.4 0.2 0
WT MO sensitive WT in Hex 3 mM WT MO insensitive Trpa1 KO 0 0.5 1.0 1.5 2.0 Time to peak nicotine 1 mM (min)
by mustard oil and nicotine (Supplementary Fig. 9) and the human TRPA1 ortholog (Supplementary Fig. 10). Consequently, we tested whether TRPA1 is modulated by hexa methonium, a mecamylamine-unrelated nAChR inhibitor that has been widely used to assess the role of these receptors in the irritating effects of cigarette smoke8. In contrast with mecamylamine, hexamethonium did not substantially affect basal TRPA1 currents (Fig. 4f,g), nor did it prevent TRPA1 activation by mustard oil in mTRPA1-expressing CHO cells (Fig. 4f–h) and trigeminal ganglion neurons (Fig. 4i). Hence, hexamethonium is a better tool than mecamylamine for selectively inhibiting nAChRs without interfering with mTRPA1. TRPA1-mediated response to the nicotine in sensory neurons We next tested whether TRPA1 contributes to the sensory responses to nicotine in mouse trigeminal ganglion neurons. In contrast with earlier in vitro studies13,35,36, we considered the relatively high EC50 and slow onset of nicotine effects on TRPA1 (see above) and tested the effect of long (2 min) applications of nicotine at 100 µM and 1 mM. Exposure to 100 µM nicotine caused a significant increase (P < 0.01) in the intracellular Ca2+ concentration in ~10% (29 out of 298) of wild-type trigeminal ganglion neurons (Fig. 5). All of the nicotine-responsive cells were sensitive to capsaicin and 90% were Figure 6 TRPA1 mediates the airway constriction reflex triggered by nasal instillation of nicotine and mustard oil. (a–c) The effects of nasal instillation of 10 µl of 60 mM nicotine (a), the vehicle (Hanks’ balanced salt solution; b) or 50 mM mustard oil (c) on the Penh determined with unrestrained whole-body plethysmography in wild-type and Trpa1 knockout mice (n = 4–7). The arrows mark the moment of instillation (time = 0 min). (d) Average Penh obtained in the basal condition and after 1 min administration of aerosols containing the vehicle alone (HBSS) or the bronchoconstrictor methacholine at various concentrations (n = 5 for each group). Error bars represent s.e.m.
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
a r t ic l e s a
2 Menthol 100 µM
TRPA1 mediates the nasal irritation induced by nicotine To test whether TRPA1 contributes to the known irritant effects of nicotine in vivo, we compared the airway constriction reflexes of wildtype and Trpa1 knockout mice to stimulation of the nasal mucosa37. To monitor the respiratory function before and after nasal instillation of test solutions, we used unrestrained whole-body plethysmography and used the increase in enhanced pause (Penh) as a measure of airway constriction38. Penh significantly increased (P = 0.027) above control levels after application of nicotine in wild-type, but not in Trpa1 knockout mice (P = 1; Fig. 6a). Instillation of vehicle alone had no effect on Penh (Fig. 6b), whereas instillation of mustard oil selectively increased Penh in wild-type mice (Fig. 6c). Trpa1 knockout mice showed normal responses to increasing concentrations of aerosolized methacholine, a muscarinic receptor agonist causing contraction of airway smooth muscle cells38 (Fig. 6d). Menthol is popularly used as an additive in nicotine-containing products to produce cooling, soothing and analgesic effects, which are thought to counteract the irritation caused by tobacco smoke 39 and nicotine40. It was therefore interesting to determine the effects nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
(mV) 0
NMDG
1
2
b 1.5
3
+
+50 mV –75 mV
Nicotine 100 µM
–2
Ratio
100
0
–1
4 Time (min)
5
6
Menthol 100 µM
1.0 0.5 Nicotine 1 mM 0
0
2
4
6
8
Time (min)
c 1.5 Instillation Penh
© 2009 Nature America, Inc. All rights reserved.
0
–100
0
also activated by 100 µM mustard oil, indicating that they were nociceptive neurons. In contrast, a significantly lower proportion of Trpa1 knockout neurons (4%, 11 out of 262, P = 0.011) responded to 100 µM nicotine. Application of 1 mM nicotine activated 23% (88 out of 383) of wild-type trigeminal ganglion neurons; about 72% of these neurons also responded to 100 µM mustard oil (Fig. 5b,e). Again, the fraction of nicotine-sensitive cells was significantly reduced in Trpa1 knockout mice (12%, 40 out of 320, P = 0.00034; Fig. 5c,e). Nicotine was able to trigger strong responses in wild-type neurons in the presence of the nAChR blocker hexamethonium (15%, 48 out of 311; Fig. 5d), which, as shown above, did not affect TRPA1. These responses were restricted to the mustard oil–sensitive neuronal population (Fig. 5e). Notably, only 4 out of 257 Trpa1 knockout neurons responded to 1 mM nicotine in the presence of hexamethonium (<1.6%, significantly lower than in wild type, P = 10−8; Fig. 5e), which indicates that the vast majority (presumably >93%) of the responses in wild-type mice are mediated by nAChRs and/or TRPA1. Notably, the responses to nicotine in TRPA1-expressing (mustard oil sensitive) neurons were characterized by a broad distribution of the time to maximal increase (time to peak) and recovery occurred only after washout of nicotine (Fig. 5a,b,d,f). In contrast, in TRPA1negative neurons, the large majority of the responses (88% in nicotine 1 mM) reached the maximum before 1 min and all of them decayed in the presence of nicotine (desensitized) within 2 min (Fig. 5a,b,c,f). In addition, whole-cell patch-clamp recordings in mouse trigeminal neurons revealed the presence of TRPA1-like and nAChR-like responses to extracellular nicotine application (Supplementary Fig. 2).
(nA)
–2
1
Current (nA)
Figure 7 Menthol inhibits nicotine-induced activation of TRPA1. (a) Time course of the amplitude of mTRPA1 currents at −75 and +50 mV during extracellular application of nicotine and nicotine plus menthol in CHO cells. The colored data points correspond to the current traces shown in the inset. Washout of menthol caused a strong current rebound, as has been previously reported23,24. (b) Ratiometric intracellular Ca2+-imaging experiment showing that menthol inhibited the response elicited by nicotine in mTRPA1-induced CHO cells (n = 7). The thick trace represents the mean and the dashed traces represent the mean ± s.e.m. (c) Menthol inhibited nicotine-induced airway constriction reflex. The effect of nasal instillation of 10 µl of 60 mM nicotine and 10 mM menthol on wild-type mice (n = 5) is shown. The arrow marks the moment of instillation (time = 0 min). The data for nicotine alone is the same as that shown in Figure 6a. Error bars represent s.e.m.
Nicotine Nicotine + menthol
1.0
0.5
0 –5
0
5
10
15 Time (min)
20
25
30
of this compound on the TRPA1-mediated responses to nicotine. Extracellular application of 100 µM menthol caused a significant reduction in mTRPA1 currents prestimulated with 100 µM nicotine (60 ± 7% at −75 mV, P = 10−4; n = 6; Fig. 7a) and reversed the increase in intracellular Ca2+ concentration induced by 1 mM nicotine in mTRPA1-expressing CHO cells (Fig. 7b). Notably, nasal instillation of a menthol (10 mM) and nicotine (60 mM) mixture triggered a smaller Penh response than the instillation of nicotine alone (Fig. 7c), suggesting that menthol reduces the airway constriction reflex triggered by nicotine. The Penh response to instillation of menthol alone (10 mM) was similar to that obtained with vehicle (Supplementary Fig. 11). This result is consistent with a previous report that menthol causes oral trigeminal stimulation in rats only at concentrations above 32 mM (ref. 41). DISCUSSION It has been generally thought that the irritating effect of nicotine is exclusively mediated by nAChRs expressed in nociceptive nerves fibers6–8,32. In contrast, we found that nicotine caused irritation via activation of TRPA1, a cation channel involved in the transduction of noxious chemical stimuli. First, we found that nicotine stimulates heterologously expressed TRPA1, acting as a gating modifier. Second, we identified TRPA1-mediated nicotine responses in a subset of nociceptive neurons. These responses could be clearly distinguished from those mediated by nAChRs on the basis of their dose dependence, kinetics and pharmacological profile. Finally, we found that TRPA1 1297
© 2009 Nature America, Inc. All rights reserved.
a r t ic l e s was necessary for the nicotine-induced airway constriction reflex in mice. To the best of our knowledge, this is the first demonstration of the in vitro and in vivo activation of a TRP channel by nicotine. We found that nicotine actually had a bimodal action on TRPA1, with activation and inhibition occurring at low and high concentrations, respectively. As the inhibitory effect reversed much faster than the activating effect, a prominent current rebound was observed on washout of higher doses of nicotine. The effect of nicotine is reminiscent of that of menthol and menthol-related compounds, for which a bimodal action on the mouse TRPA1 clone has also been reported23,24. Using a mutagenesis approach based on the differences in menthol sensitivity of various TRPA1 orthologs, a previous study identified residues in the putative S5 segment as determinants for the stimulatory effect24. The study found that menthol’s ability to inhibit mouse TRPA1, but not human TRPA1, is determined by specific amino acids distributed all across the putative pore domains. Our data suggest that nicotine interacts with the pore, as application of 1 mM nicotine induced a ~30% reduction of the single-channel conductance at positive potentials. Moreover, the fast relief of inhibition on nicotine washout is consistent with a binding site in the pore, which is quickly accessible from the extracellular solution. A bimodal effect has also been reported for 2-aminoethoxydiphenyl borate on TRPV3 (ref. 42), with an abrupt, but transient, increase in current on washout of high doses of the agonist. It is well known that prior oral stimulation with the epitomic TRPA1agonist mustard oil reduces the sensitivity to other irritant chemicals, such as nicotine31. Our data suggest that this cross-desensitization arises at the level of the common chemoreceptor, TRPA1. Indeed, we found that prior stimulation of TRPA1 by mustard oil fully abolished a later response to nicotine. Conversely, prior stimulation with nicotine led to a reduction of the mustard oil response. In this respect, it should be noted that nicotine and mustard oil have similar EC50 values for activating inward TRPA1 currents (17 and 11 µM, respectively), but that mustard oil is a much more effective activator, producing up to a 30-fold current increase, as compared with the maximal fivefold increase in inward current caused by nicotine (Fig. 3). Activation of TRPA1 by electrophiles, such as mustard oil, cinnamaldehyde and acrolein, occurs through covalent modification of cysteine residues on the channel27,28. We were able to exclude the possibility that nicotine acts via a similar mechanism, as we found that the mustard oil–insensitive mutant C622S (ref. 28) can be readily activated by nicotine, which is consistent with the nonelectrophilic nature of nicotine. Thus, we consider it most likely that nicotine acts on TRPA1 through a noncovalent interaction with the channel, similar to the actions of other nonelectrophilic agonists, such as icilin15, menthol23,24, clotrimazole25 and nifedipine26. Indeed, as has been reported for the latter three compounds, nicotine caused a negative shift in the voltage dependence of channel activation. In addition, we found that nicotine reduced the rate of whole-cell current deactivation and accelerated activation. These observations are fully consistent with our observations of a negative shift of the voltage dependence of channel activation and an increase in the mean open time and reduction of the mean closed time of single TRPA1 channels. Thus, nicotine causes both stabilization of the open conformation and destabilization of the closed conformation of TRPA1. Our experiments in mouse trigeminal ganglion neurons strongly support the role of TRPA1 as mediator of nicotine-induced irritation. First, nicotine activated a subset of neurons that largely overlaps with the TRPA1-expressing (mustard oil sensitive) population. Second, the proportion of nicotine-sensitive neurons was strongly reduced in Trpa1 knockout mice. Third, nicotine elicited robust
responses in the presence of the specific nAChRs inhibitor hexa methonium. Fourth, nicotine responses were virtually abolished in Trpa1 knockout mice neurons in the presence of hexamethonium. Finally, the responses to nicotine in TRPA1-expressing (mustard oil sensitive) neurons were kinetically different from the responses in TRPA1-negative (mustard oil insensitive) neurons. Thus, we conclude that nicotine evokes two distinct types of responses in wild-type trigeminal ganglion neurons: rapid and quickly desensitizing responses mediated by nAChRs and slower, more sustained responses mediated by TRPA1. Our results prompt a re-evaluation of the mechanisms underlying the irritating effects of nicotine by introducing TRPA1 as a previously unknown ionotropic nicotine receptor that is distinct from nAChRs. On the one hand, the mucosal nicotine concentrations attained during tobacco smoke exposure are submicromolar43 and are therefore lower than those necessary to activate TRPA1. Thus, the reported acute irritant effects of nicotine delivered in this way are probably mediated by nAChRs8. However, many studies aimed at understanding the irritating effects of nicotine on mucosal irritation in vivo have used millimolar concentrations of nicotine, as high as 600 mM7,32,44. Moreover, in many instances the inhibition of nicotineinduced responses by mecamylamine has been taken as evidence for the involvement of nAChRs. Our current data challenge this view, as we found that TRPA1 was not only activated by nicotine, but was also inhibited by mecamylamine. To test the relevance of nicotine-induced activation of TRPA1 in vivo, we studied the irritating effects of nicotine on wild-type and Trpa1 knockout mice. We simulated the application of nicotine nasal sprays, which generally contain nicotine at concentrations around 60 mM, by studying the well-known airway constriction reflex to noxious stimulation of the nasal mucosa37,45. Notably, this form of nicotine replacement therapy is the most effective one, but has the highest treatment dropout rate as a result of mucosal irritation 5. Nasal instillation of nicotine provoked an increase of Penh, a surrogate measure of airway constriction, in wild-type, but not in Trpa1 knockout, mice. The normal response of Trpa1 knockout mice to the bronchoconstrictor methacholine indicates that the lack of nicotineand mustard oil–induced response in these mice is not caused by an unspecific contractile dysfunction of the airway smooth muscle. Altogether, these results indicate that TRPA1 mediates the irritating effect of nasal nicotine instillation. Notably, it is apparent from our data that nAChRs did not contribute to the airway constriction reflex, which contrasts with the observation of nAChR-like responses in primary cultured trigeminal ganglion neurons. This disagreement could be explained by the transient character of nAChR activation, which may have hindered the contribution of nAChRs to the airway constriction reflex in our in vivo experimental model. Alternatively, the sensory neurons that respond to nicotine via activation of nAChRs may not be involved in the long-lasting airway reflex following nicotine stimulation of the nasal mucosa. We found that menthol, a known blocker of mouse TRPA1, induced an inhibition of mouse TRPA1 currents that were pre-activated by nicotine and reduced the Penh response on intranasal instillation of nicotine in mice. Caution should be taken, however, when trying to extrapolate the effects of menthol to humans, as this compound does not inhibit human TRPA1 (ref. 24). The well-known soothing effect of menthol in tobacco-containing products39 may well be related to activation of TRPM8, which has been shown to produce analgesia46. Nevertheless, our results indicate that inhibition of TRPA1 represents an interesting approach for developing smoking cessation therapies with less adverse effects.
1298
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
a r t ic l e s
1. Hatsukami, D.K., Stead, L.F. & Gupta, P.C. Tobacco addiction. Lancet 371, 2027–2038 (2008). 2. Thuerauf, N. et al. The influence of mecamylamine on trigeminal and olfactory chemoreception of nicotine. Neuropsychopharmacology 31, 450–461 (2006). 3. Stead, L.F., Perera, R., Bullen, C., Mant, D. & Lancaster, T. Nicotine replacement therapy for smoking cessation. Cochrane Database Syst. Rev. CD000146 (2008). 4. Nides, M. Update on pharmacologic options for smoking cessation treatment. Am. J. Med. 121, S20–S31 (2008). 5. Hajek, P. et al. Randomized comparative trial of nicotine polacrilex, a transdermal patch, nasal spray and an inhaler. Arch. Intern. Med. 159, 2033–2038 (1999). 6. Dussor, G.O. et al. Potentiation of evoked calcitonin gene–related peptide release from oral mucosa: a potential basis for the pro-inflammatory effects of nicotine. Eur. J. Neurosci. 18, 2515–2526 (2003). 7. Simons, C.T., Sudo, S., Sudo, M. & Carstens, E. Mecamylamine reduces nicotine cross-desensitization of trigeminal caudalis neuronal responses to oral chemical irritation. Brain Res. 991, 249–253 (2003). 8. Lee, L.Y. & Gu, Q. Cough sensors. IV. Nicotinic membrane receptors on cough sensors. Handb. Exp. Pharmacol. 187, 77–98 (2009). 9. Talavera, K., Nilius, B. & Voets, T. Neuronal TRP channels: thermometers, pathfinders and lifesavers. Trends Neurosci. 31, 287–295 (2008). 10. Damann, N., Voets, T. & Nilius, B. TRPs in our senses. Curr. Biol. 18, R880–R889 (2008). 11. Venkatachalam, K. & Montell, C. TRP channels. Annu. Rev. Biochem. 76, 387–417 (2007). 12. Liu, L. et al. Nicotine inhibits voltage-dependent sodium channels and sensitizes vanilloid receptors. J. Neurophysiol. 91, 1482–1491 (2004).
13. Fucile, S., Sucapane, A. & Eusebi, F. Ca2+ permeability of nicotinic acetylcholine receptors from rat dorsal root ganglion neurones. J. Physiol. (Lond.) 565, 219–228 (2005). 14. Bessac, B.F. & Jordt, S.E. Breathtaking TRP channels: TRPA1 and TRPV1 in airway chemosensation and reflex control. Physiology (Bethesda) 23, 360–370 (2008). 15. Story, G.M. et al. ANKTM1, a TRP-like channel expressed in nociceptive neurons, is activated by cold temperatures. Cell 112, 819–829 (2003). 16. Karashima, Y. et al. TRPA1 acts as a cold sensor in vitro and in vivo. Proc. Natl. Acad. Sci. USA 106, 1273–1278 (2009). 17. Jordt, S.E. et al. Mustard oils and cannabinoids excite sensory nerve fibres through the TRP channel ANKTM1. Nature 427, 260–265 (2004). 18. Bandell, M. et al. Noxious cold ion channel TRPA1 is activated by pungent compounds and bradykinin. Neuron 41, 849–857 (2004). 19. Andrè, E. et al. Cigarette smoke–induced neurogenic inflammation is mediated by alpha,beta-unsaturated aldehydes and the TRPA1 receptor in rodents. J. Clin. Invest. 118, 2574–2582 (2008). 20. Bautista, D.M. et al. TRPA1 mediates the inflammatory actions of environmental irritants and proalgesic agents. Cell 124, 1269–1282 (2006). 21. Kwan, K.Y. et al. TRPA1 contributes to cold, mechanical, and chemical nociception but is not essential for hair-cell transduction. Neuron 50, 277–289 (2006). 22. MacPherson, L.J. et al. More than cool: promiscuous relationships of menthol and other sensory compounds. Mol. Cell. Neurosci. 32, 335–343 (2006). 23. Karashima, Y. et al. Bimodal action of menthol on the transient receptor potential channel TRPA1. J. Neurosci. 27, 9874–9884 (2007). 24. Xiao, B. et al. Identification of transmembrane domain 5 as a critical molecular determinant of menthol sensitivity in mammalian TRPA1 channels. J. Neurosci. 28, 9640–9651 (2008). 25. Meseguer, V. et al. Transient receptor potential channels in sensory neurons are targets of the antimycotic agent clotrimazole. J. Neurosci. 28, 576–586 (2008). 26. Fajardo, O., Meseguer, V., Belmonte, C. & Viana, F. TRPA1 channels: novel targets of 1,4-dihydropyridines. Channels (Austin) 2, 429–438 (2008). 27. Hinman, A., Chuang, H.H., Bautista, D.M. & Julius, D. TRP channel activation by reversible covalent modification. Proc. Natl. Acad. Sci. USA 103, 19564–19568 (2006). 28. Macpherson, L.J. et al. Noxious compounds activate TRPA1 ion channels through covalent modification of cysteines. Nature 445, 541–545 (2007). 29. Zurborg, S., Yurgionas, B., Jira, J.A., Caspani, O. & Heppenstall, P.A. Direct activation of the ion channel TRPA1 by Ca2+. Nat. Neurosci. 10, 277–279 (2007). 30. Doerner, J.F., Gisselmann, G., Hatt, H. & Wetzel, C.H. Transient receptor potential channel A1 is directly gated by calcium ions. J. Biol. Chem. 282, 13180–13189 (2007). 31. Carstens, E., Kuenzler, N. & Handwerker, H.O. Activation of neurons in rat trigeminal subnucleus caudalis by different irritant chemicals applied to oral or ocular mucosa. J. Neurophysiol. 80, 465–492 (1998). 32. Simons, C.T., Boucher, Y., Carstens, M.I. & Carstens, E. Nicotine suppression of gustatory responses of neurons in the nucleus of the solitary tract. J. Neurophysiol. 96, 1877–1886 (2006). 33. Papke, R.L., Sanberg, P.R. & Shytle, R.D. Analysis of mecamylamine stereoisomers on human nicotinic receptor subtypes. J. Pharmacol. Exp. Ther. 297, 646–656 (2001). 34. Xu, H., Blair, N.T. & Clapham, D.E. Camphor activates and strongly desensitizes the transient receptor potential vanilloid subtype 1 channel in a vanilloidindependent mechanism. J. Neurosci. 25, 8924–8937 (2005). 35. Liu, L. & Simon, S.A. Capsaicin and nicotine both activate a subset of rat trigeminal ganglion neurons. Am. J. Physiol. 270, C1807–C1814 (1996). 36. Haberberger, R.V. et al. Nicotinic acetylcholine receptor subtypes in nociceptive dorsal root ganglion neurons of the adult rat. Auton. Neurosci. 113, 32–42 (2004). 37. Widdicombe, J. Reflexes from the lungs and airways: historical perspective. J. Appl. Physiol. 101, 628–634 (2006). 38. Vanoirbeek, J.A. et al. Respiratory response to toluene diisocyanate depends on prior frequency and concentration of dermal sensitization in mice. Toxicol. Sci. 80, 310–321 (2004). 39. Kreslake, J.M., Wayne, G.F. & Connolly, G.N. The menthol smoker: tobacco industry research on consumer sensory perception of menthol cigarettes and its role in smoking behavior. Nicotine Tob. Res. 10, 705–715 (2008). 40. Dessirier, J.M., O’Mahony, M. & Carstens, E. Oral irritant properties of menthol: sensitizing and desensitizing effects of repeated application and cross-desensitization to nicotine. Physiol. Behav. 73, 25–36 (2001). 41. Zanotto, K.L., Merrill, A.W., Carstens, M.I. & Carstens, E. Neurons in superficial trigeminal subnucleus caudalis responsive to oral cooling, menthol and other irritant stimuli. J. Neurophysiol. 97, 966–978 (2007). 42. Chung, M.K., Lee, H., Mizuno, A., Suzuki, M. & Caterina, M.J. 2-aminoethoxydiphenyl borate activates and sensitizes the heat-gated ion channel TRPV3. J. Neurosci. 24, 5177–5182 (2004). 43. Dhar, P. Measuring tobacco smoke exposure: quantifying nicotine/cotinine concentration in biological samples by colorimetry, chromatography and immunoassay methods. J. Pharm. Biomed. Anal. 35, 155–168 (2004). 44. Carstens, E., Albin, K.C., Simons, C.T. & Carstens, M.I. Time course of selfdesensitization of oral irritation by nicotine and capsaicin. Chem. Senses 32, 811–816 (2007). 45. Togias, A. Mechanisms of nose-lung interaction. Allergy 54 Suppl 57: 94–105 (1999). 46. Dhaka, A. et al. TRPM8 is required for cold sensation in mice. Neuron 54, 371–378 (2007). 47. Rask, L. et al. Myrosinase: gene family evolution and herbivore defense in Brassicaceae. Plant Mol. Biol. 42, 93–113 (2000). 48. Baldwin, I. & Ohnmeiss, T. Alkaloidal responses to damage in Nicotiana native to North America. J. Chem. Ecol. 19, 1143–1153 (1993).
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1299
Our findings are also relevant to the ecological role and industrial uses of nicotine and its analog anabasine, which are known to be strong repellents of herbivores. Notably, Brassicaceae and Nicotiana plants increase their production of either isothiocyanates47 or nicotine and anabasine48, respectively, following herbivore attack. This suggests that the promiscuous character of TRPA1 chemo-activation underlies a unified mechanism for the detection of a wide range of noxious compounds that function as botanical defensive traits. Methods Methods and any associated references are available in the online version of the paper at http://www.nature.com/natureneuroscience/.
© 2009 Nature America, Inc. All rights reserved.
Note: Supplementary information is available on the Nature Neuroscience website. Acknowledgments We are grateful to K.Y. Kwan for providing us with the Trpa1 knockout mice, M.R. Sepúlveda for helpful discussions, and V. De Vooght and P. Hoet for help in some plethysmography experiments. The expert technical assistance of J. Prenen is greatly acknowledged. The mTRPA1 CHO cell line was kindly provided by A. Patapoutian. K.T. and J.A.J.V. were supported by a postdoctoral mandate from KU Leuven and are currently postdoctoral fellows of the Research Foundation– Flanders (Fonds voor Wetenschappelijk Onderzoek, FWO). M.G. and W.E. are doctoral FWO fellows. V.M.M was supported by Spanish CONSOLIDERINGENIO 2010 CSD2007-00023. This work was supported by grants from Inter-university Attraction Poles Programme (Belgian Science Policy, P6/28), FWO (G.0172.03 and G.0565.07), the Research Council of the KU Leuven (GOA 2004/07) and the Flemish Government (Excellentiefinanciering, EF/95/010). AUTHOR CONTRIBUTIONS K.T. carried out patch-clamp and Ca2+-imaging recordings, plethysmography experiments, analyzed the data, wrote the paper and supervised the project. M.G. and Y.K. performed patch-clamp and Ca2+-imaging recordings. V.M.M. carried out patch-clamp and Ca2+-imaging recordings in neurons. J.A.J.V. performed plethysmography experiments. N.D. carried out Ca2+-imaging recordings in neurons and edited the paper. W.E. performed Ca2+-imaging and mouse experiments and edited the paper. M.B. carried out mouse genotyping. A.J. performed the molecular biology work. R.V. supervised mouse genotyping and edited the paper. F.V. edited the paper and supervised the project. B. Nemery edited the paper and supervised the plethysmography experiments. B. Nilius edited the paper and supervised the project. T.V. analyzed the data, wrote the paper and supervised the project. Published online at http://www.nature.com/natureneuroscience/. Reprints and permissions information is available online at http://www.nature.com/ reprintsandpermissions/.
ONLINE METHODS
© 2009 Nature America, Inc. All rights reserved.
Cells and animals. We used a tetracycline-regulated system for inducible expression of mTRPA1 in CHO cells15. Human TRPA1 was heterologously expressed in CHO cells by transient transfection using TransIT 293 reagents (Mirus). Trigeminal ganglion neurons from adult (postnatal weeks 8–12) C57Bl/6J (wild type) and Trpa1 knockout mice16 were cultured as described previously23. Experiments were carried out in accordance with the European Union Community Council guidelines and were approved by the Animal Experiments Ethics Committee of KU Leuven. Patch-clamp experiments. Patch-clamp recordings were performed as described previously49. Before current recordings, cells were rinsed with Krebs solution containing 150 mM NaCl, 6 mM KCl, 1 mM MgCl2, 1.5 mM CaCl2, 10 mM glucose and 10 mM HEPES and titrated to pH 7.4 with NaOH. Bath solutions were perfused by gravity via a multi-barreled pipette tip with a single outlet of 0.8-mm inner diameter. This system allows the full exchange of the medium bathing the recorded cell in less than 2–4 s. Currents were routinely elicited by 400-ms-long voltage ramps from −100 to +100 mV at a stimulation frequency of 0.5 Hz. The holding potential was 0 mV. The extracellular solution contained 150 mM NaCl, 2 mM CaCl2, 1 mM MgCl2 and 10 mM HEPES and was titrated to pH 7.4 with NaOH. To monitor the quality of the recordings, we regularly tested the effect of substituting all extracellular cations with NMDG+, which is largely impermeable through TRPA1 and is therefore expected to reduce the inward current to negligible background levels. The intracellular solution contained 156 mM CsCl, 1 mM MgCl2, 10 mM HEPES and 10 mM BAPTA to minimize possible activation by intracellular Ca2+ (refs. 29,30) and was titrated to pH 7.2 with NaOH. The patch-clamp data were analyzed using WinASCD (G. Droogmans; ftp://ftp. cc.kuleuven.ac.be/pub/droogmans/winascd.zip) and Origin 7.0 (OriginLab). The bimodal dose-response curves for the effects of nicotine on TRPA1 currents (Fig. 1d) were fit by a function of the form: H IC50I Max × [ Nic ]H S Irel = 1 + × HS H H [Nic ]HS + EC50 [Nic ] I + IC500I
where Irel is the steady-state amplitude of the current recorded in the presence of nicotine at concentration [Nic] normalized to the value in control. Max is the maximal relative current increase expected in the absence of inhibitory effect of nicotine, EC50 and IC50 are the effective concentrations for current stimulation
nature NEUROSCIENCE
and inhibition, respectively, and HS and HI are the corresponding Hill coefficients. Max, EC50 and HS were fixed to the corresponding values obtained from the fit of the dose-response curves for the stimulatory effect of nicotine (see Supplementary Fig. 6). Intracellular Ca2+-imaging experiments. Cells were incubated with 2 µM Fura-2 acetoxymethyl ester for 30 min at 37 °C. Intracellular Ca2+ concentration was monitored via the ratio of fluorescence measured on alternating illumination at 357 and 380 nm using an MT-10 illumination system and cellM software (Olympus). As a control, we used extracellular Krebs solution (see above). To identify neurons in trigeminal ganglion cultures, we applied a Krebs-based solution in which the KCl concentration was increased to 45 mM by iso-osmotic substitution of NaCl. The system for exchange of extracellular solutions was similar to that used for patch-clamp experiments (see above). Whole-body plethysmography. The ventilatory function of mice was monitored using unrestrained whole-body plethysmography, as described previously38. Drugs were delivered by nasal instillation (without anesthesia) of 10 µl of test solutions to restrict the area of stimulation to the upper airways50. The increase in Penh was used as an indicator of bronchoconstriction. Penh is a dimensionless parameter that represents a proportion of maximal expiratory to maximal inspiratory box pressure signals in relation to the timing of expiration and is calculated as expiratory time peak expiratory flow relaxation time − 1 × peak inspiratory flow . All chemicals were purchased from Sigma-Aldrich. Statistics. Data are presented as mean ± s.e.m. Significance between groups was tested using the unpaired or paired Student’s t tests or the χ2 test as appropriate.
49. Talavera, K., Janssens, A., Klugbauer, N., Droogmans, G. & Nilius, B. Extracellular Ca2+ modulates the effects of protons on gating and conduction properties of the T-type Ca2+ channel α1G (CaV3.1). J. Gen. Physiol. 121, 511–528 (2003). 50. Southam, D.S., Dolovich, M., O’Byrne, P.M. & Inman, M.D. Distribution of intranasal instillations in mice: effects of volume, time, body position and anesthesia. Am. J. Physiol. Lung Cell. Mol. Physiol. 282, L833–L839 (2002).
doi:10.1038/nn.2379
a r t ic l e s
Suppression of hippocampal TRPM7 protein prevents delayed neuronal death in brain ischemia
© 2009 Nature America, Inc. All rights reserved.
Hong-Shuo Sun1, Michael F Jackson2,3, Loren J Martin4, Karen Jansen5, Lucy Teves1, Hong Cui1, Shigeki Kiyonaka6, Yasuo Mori6, Michael Jones1, Joan P Forder1, Todd E Golde5, Beverley A Orser2,4,7, John F MacDonald2,3 & Michael Tymianski1,2,4,8 Cardiac arrest victims may experience transient brain hypoperfusion leading to delayed death of hippocampal CA1 neurons and cognitive impairment. We prevented this in adult rats by inhibiting the expression of transient receptor potential melastatin 7 (TRPM7), a transient receptor potential channel that is essential for embryonic development, is necessary for cell survival and trace ion homeostasis in vitro, and whose global deletion in mice is lethal. TRPM7 was suppressed in CA1 neurons by intrahippocampal injections of viral vectors bearing shRNA specific for TRPM7. This had no ill effect on animal survival, neuronal and dendritic morphology, neuronal excitability, or synaptic plasticity, as exemplified by robust long-term potentiation (LTP). However, TRPM7 suppression made neurons resistant to ischemic death after brain ischemia and preserved neuronal morphology and function. Also, it prevented ischemia-induced deficits in LTP and preserved performance in fear-associated and spatial-navigational memory tasks. Thus, regional suppression of TRPM7 is feasible, well tolerated and inhibits delayed neuronal death in vivo. Hypoxic-ischemic injuries to the mammalian brain elicit a delayed neuronal death (DND), whose mechanism is uncertain, that characterizes neurological disorders such as strokes, Alzheimer’s, Huntington’s and Parkinson’s disease, and may mirror ischemic cell death in other tissues1. In survivors of cardiac arrest, transient global ischemia leads to DND of CA1 hippocampal neurons, impaired cognition and defects in memory functions within days2,3. These same events are recapitulated in rodents that are exposed to experimental global cerebral ischemia, which leads to DND in the hippocampus, striatum and cortex4, and to deficits of learning and memory5,6. Previous research has implicated excitotoxicity as a possible mechanism of DND in the hippocampus7, especially via activation of AMPA/kainate glutamate receptors4. However, there is a growing awareness that non-excitotoxic mechanisms also contribute to DND8. For example, in cultured neurons exposed to prolonged oxygen-glucose deprivation (OGD), treating excitotoxicity is insufficient to prevent death. This is a result, in part, of the simultaneous activation of TRPM7 channels by OGD9. These members of the TRP superfamily10 comprise broadly expressed, nonselective cation channels that may affect trace ion and magnesium homeostasis11. When activated in hypoxic cultured neurons, TRPM7 channels elicit death independently of excitotoxicity9. Inhibiting these channels inhibits anoxic cell death irrespective of excitotoxicity, suggesting that TRPM7-mediated death processes may modulate or act upstream from excitotoxicity9. To date, the suppression of TRPM7 as
a means of studying DND mechanisms in vivo has not been feasible. There are currently no selective pharmacological inhibitors of this protein. Moreover, although TRPM7 inhibition by siRNA is tolerated by cultured neuronal cells9, its deletion in DT-40B cell lines affects their survivability12,13 and its deletion in T cells elicits dysregulated synthesis of growth factors needed for thymopoesis14. In addition, TRPM7 is essential for embryonic development and its global deletion in mice is lethal before day 7.5 of embryogenesis14. Because of this lack of viable knockout animals, concerns about adverse effects of TRPM7 deletion on cell viability and a lack of selective pharmacological blockers, the role of TRPM7 in ischemic brain damage has never been investigated. Nonetheless, TRPM7 channels are strong candidates for mediating non-excitotoxic ischemic brain injury. Ischemia elicits large reductions in extracellular divalents, acidosis and oxidative stress 15–17, all of which are conditions that potentiate TRPM7 channels. Although they conduct only a few pA of inward current under physiological pH, extracellular calcium concentration ([Ca2+]e), [Mg2+]e and low oxidative stress12,18–20, TRPM7 currents increase markedly when extracellular divalents are reduced20, as is the case in ischemia16. Here, we directly investigated the hypothesis that, despite the lethality of global TRPM7 deletion, these channels are important in the pathways mediating DND after ischemic injury in vivo and that a regional suppression of TRPM7 leads to neuronal preservation.
1Toronto
Western Hospital Research Institute, Toronto, Ontario, Canada. 2Department of Physiology, University of Toronto, Toronto, Ontario, Canada. 3Robarts Research Institute, University of Western Ontario, London, Ontario, Canada. 4Institute of Medical Science, University of Toronto, Toronto, Ontario, Canada. 5Department of Neuroscience, Mayo Clinic, Jacksonville, Florida, USA. 6Laboratory of Molecular Biology, Department of Synthetic Chemistry and Biological Chemistry, Graduate School of Engineering, Kyoto University, Kyoto, Japan. 7Department of Anesthesia, Sunnybrook Health Sciences Centre, Toronto, Ontario, Canada. 8Department of Surgery, University of Toronto, Toronto, Ontario, Canada. Correspondence should be addressed to M.T. (
[email protected]). Received 8 June; accepted 11 August; published online 6 September 2009; doi:10.1038/nn.2395
1300
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
a r t ic l e s a
b
Before LDM
c β-actin
d
EGFP
After LDM
shSCR
shTRPM7
TRPM2
e
TRPM3
Actin
sh
sh
EG FP
N o
Merge
TRPM7
GFP
NeuN
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1301
shSCR
EGFP
RESULTS TRPM7 suppression using rAAV vectors To suppress TRPM7 in neurons, we generated a small interfering RNA (siRNA) hairpin sequence (shRNA) corresponding to coding regions 5,152–5,172 relative to the first nucleotide of the start codon of murine TRPM7 (GenBank accession number AY032951) 9 and packaged it in a recombinant serotype 1 adeno-associated virus (rAAVshTRPM7) that included enhanced green fluorescent protein (EGFP; Online Methods, Supplementary Notes and Supplementary Fig. 1). Controls were packaged with scrambled siRNA (rAAVshSCR) or with EGFP alone (empty vector, rAAV EGFP). We first validated TRPM7 suppression by the rAAVs in cultured rat cortical neurons using separately validated antibodies (Supplementary Notes and Supplementary Fig. 2). Cultures were infected at 7 d in vitro (DIV) and studied at 14 DIV. The vectors were highly neurotropic, infecting neurons in vitro with near 100% efficiency, as gauged by EGFP expression (data not shown). Infection with the rAAVshTRPM7, but not with the control rAAVs, suppressed TRPM7 immunostaining without affecting TRPM2 expression, a related TRP channel (Supplementary Fig. 1). We examined TRPM2 because it and TRPM7 are unique among TRP family members in that both are stimulated by intracellular free radicals and both have been implicated in cell death9,21. Infecting the neurons with rAAVshTRPM7 yielded results that were consistent with those previously obtainable by siRNA knockdown of TRPM7 (ref. 9), which rendered the cells more resistant to OGD (Supplementary Fig. 1, Online Methods and Supplementary Notes).
Suppression of TRPM7 in CA1 neurons of adult rats We next examined whether rAAVshTRPM7 infection in vivo could suppress TRPM7 in a sufficient number of cells to allow us to study cerebral ischemia. Stereotactic microinjections (7.6 × 109 genomes per injection) into the right hippocampus 10 d prior to evaluation (see Online Methods and Supplementary Fig. 3) produced widespread infection (Fig. 1a and Supplementary Figs. 3–5). Suppression of TRPM7 was confirmed by reverse-transcription PCR, western blots, immunostaining and electrophysiology. To avoid contamination from uninfected or non-neuronal cells, we separated neurons for RT-PCR analysis from frozen sections by laser-dissection microcapture (Fig. 1b). TRPM7 transcript levels were suppressed in rAAVshTRPM7infected cells, whereas other TRPM channel mRNA levels (TRPM2, TRPM3 and TRPM6) were unaffected (Fig. 1c). Next, we immuno stained frozen sections from rats injected with the various rAAV vectors for TRPM7 (Fig. 1d). The hippocampi were dissected from sister sections in the same sample and used for immunoblots with a separately validated antibody (Supplementary Notes and Supplementary Fig. 2). We found that TRPM7 protein expression was suppressed in samples from rats treated with rAAVshTRPM7, but not with rAAVshSCR or rAAVEGFP (Fig. 1e). These samples contained proteins pooled from both neuronal (rAAV infected) and non-neuronal (uninfected) hippocampal cells, and may overestimate the relative amounts of TRPM7 protein remaining in the neurons. Consistent with our in vitro results (Supplementary Fig. 1), the rAAV vectors had no apparent effect on neuronal morphology. However, treatment with rAAVshTRPM7 attenuated TRPM7 immunostaining in hippocampal CA1 neurons (Fig. 1f). Induction of RNAi in mammalian cells by expression of doublestranded RNA can activate innate antiviral (interferon) response pathways that may elicit off-target gene expression22,23. Therefore, we confirmed that the rAAV vectors did not activate interferon target genes (Supplementary Notes and Supplementary Fig. 2). We then addressed the concern that TRPM7 may be needed for the survival of certain mammalian cells12,13,24 and that its suppression in vivo could reduce neuronal viability and complicate the interpretation of functional and stroke experiments involving DND. We counted the EGFP-positive CA1 pyramidal neurons in six coronal planes at 7, 10 and 14 d after rAAV infection (Supplementary Figs. 3–5). Rats infected with rAAVshTRPM7 had similar numbers of EGFP-positive CA1 pyramidal neurons as those infected with control vectors, suggesting that suppressing TRPM7 in vivo does not affect basal neuronal viability (Fig. 2a and Supplementary Figs. 3–5). To evaluate the downregulation of TRPM7 function by the rAAVs, we carried out whole-cell patch-clamp recordings from acutely
shTRPM7
© 2009 Nature America, Inc. All rights reserved.
f
TR PM 7
TRPM7
SC R
TRPM7
vi ru EG s F sh P sh SC TR R PM 7
TRPM6
Figure 1 Suppression of TRPM7 expression in adult rat hippocampal neurons. All rats were stereotaxically injected with 7.6 × 10 9 genomes of rAAV vectors 10 d before analysis. (a) Representative EGFP fluorescence in a hippocampus injected with rAAVEGFP. Arrowhead indicates the needle tract. Scale bar represents 500 µm. (b) Representative frozen section of the CA1 sector before and after laser dissection microcapture (LDM) of EGFP-positive pyramidal neurons (yellow circles) infected with rAAVshTRPM7. Scale bar represents 25 µm. (c) RT-PCR of TRPM7, TRPM2, TRPM3 and TRPM6 from 60 CA1 pyramidal neurons removed by LDM as in b (n = 3 experiments). (d) EGFP fluorescence (green) and TRPM7 immunostaining (red) of representative frozen sections of rAAV-microinjected hippocampi used for immunoblots. Scale bar represents 75 µm. (e) Immunoblots of TRPM7 from hippocampal CA1 sectors infected with the indicated rAAV vectors (n = 3 experiments). (f) Immunostaining of hippocampal CA1 regions from rats infected with the indicated rAAV vectors. Scale bars represent 50 µm. Merged images show GFP and TRPM7 staining (n = 3 experiments).
a r t ic l e s a
7d
10 d
14 d
600 400 200
EG sh shS FP TR C PMR 7
EG sh shS FP TR C PMR 7
EG sh shS FP TR C PMR 7
0
b
c
–60 mV
600
2+
1.3 mM Ca
Current (pA)
500 0.1 mM Ca2+
500 pA
400
200 100
0.5 s
0
sh sh SC TR R PM 7
shTRPM7
shTRPM7
shSCR 1,500
pA
500
500 mV
–1,000
pA
1,000
1,000
–100 –50 –500
1,500
mV 50
100
–100
1.3 mM Ca2+ 0.1 mM Ca2+
–50 –500
50
100
–1,000
isolated hippocampal neurons from 4–5-week-old rats that were given intrahippocampal injections of rAAVshTRPM7 or rAAVshSCR at 3 weeks of age. In acutely isolated hippocampal neurons and in neurons grown in dissociated cell cultures, lowering the concentration of
a
extracellular Ca2+ and/or Mg2+ transiently evokes an inward cation current25,26. This current is at least partly mediated by TRPM7, as determined previously using the shRNA/siRNA sequence to TRPM7 that we used here. When delivered into cultured hippocampal neurons using adenoviral vectors, this shRNA reduces both TRPM7 mRNA levels and low [Ca2+]e-evoked currents20. The siRNA sequence, when transfected into cultured cortical neurons by conventional means, inhibits anoxia-evoked TRPM7-dependent currents9. In the acutely dissociated neurons taken from rAAVSCR-infected rats, we evoked a current using a 5-s application of extracellular solution (ECS) containing 0.1 mM Ca2+ (Fig. 2b). The amplitude of the low [Ca2+]e-evoked current was attenuated in neurons infected with rAAVshTRPM7 (Fig. 2b,c), which is consistent with TRPM7 inhibition. Furthermore, TRPM7-mediated currents have an outwardly rectifying I-V curve that is lost when extracellular divalent
b
×3.5 ×25
×25
×3.5
Figure 3 TRPM7 suppression in vivo imparts Stroke rats 72 h after 4VO Sham surgery rats 72 h after 4VO resilience to DND. (a) Representative coronal EGFP shSCR shTRPM7 EGFP shSCR shTRPM7 images of neurons and dendrites derived from hippocampi of rats infected with the indicated rAAV vector 7 d before a 15-min 4VO. Brains were cryostat sectioned (25 µm) and imaged 72 h after the ischemic insult. Images are representative of 6 rats per group. (b) Representative images taken from rats infected with the indicated rAAV vectors and undergoing sham surgery under otherwise identical conditions to those described in a. Scale bars in a and b represent 200, 50, 20 Contralateral (no virus) Ipsilateral (shTRPM7) and 10 µm for images taken with ×3.5, ×25, ×63 and ×126 power objectives, respectively (×126 power was obtained with a ×63 objective and a ×2 digital gain). (c) Representative TUNEL counts EGFP counts Contralateral coronal sections of TUNEL-staining in the CA1 4VO Sham Ipsilateral 1.2 sectors of hippocampi of a rat injected with 600 1.0 500 rAAVshTRPM7 7 d before 4VO. The hippocampus 0.8 * 400 ipsilateral (right) and contralateral (left) to 0.6 300 0.4 the injection is shown, stained 3 d after 4VO 200 0.2 100 ** (representative of 6 rats). (d) Counts (see Online 0 0 ** Methods) of TUNEL-stained CA1 neurons as in c from the ipsilateral (rAAV-microinjected) and contralateral (uninjected) hippocampus (n = 6 rats per group). * indicates difference from all other groups (ANOVA, P < 0.002; Fisher LSD test, shTRPM7 injected versus contralateral, P < 0.001). (e) Counts (mean ± s.e.m.) of EGFP-expressing CA1 neurons from the 4VO experiment represented in a (4VO, n = 6 rats per group; sham, 3 rats per group). ** indicates difference from shTRPM7 and sham groups (ANOVA, P < 0.01; Fisher LSD test, EGFP versus shTRPM7, P < 0.001; shSCR versus shTRPM7, P < 0.001; EGFP versus shSCR, P = 0.378). ×63
c
e
1302
E s G sh hS FP TR CR PM EG 7 s sh hS FP TR CR PM 7
7
R
PM
TR
sh
sh SC
EG
FP
CA1 cell counts (×1,000)
EGFP cell counts
d
×126
© 2009 Nature America, Inc. All rights reserved.
shSCR
d
*
300
Figure 2 Suppression of TRPM7 is well tolerated in vivo. (a) Counts of EGFP-positive CA1 neurons in hippocampi infected with the indicated rAAV vectors at the indicated days post-microinjection (n = 6 rats per group, mean ± s.e.m.). (b) Representative traces of the response of acutely isolated CA1 neurons (dark gray, rAAV shSCR-infected neuron; light gray, rAAVshTRPM7-infected neuron) to a 5-s application of ECS containing 0.1 mM Ca2+. The top and middle traces show the timing of the applied voltage ramps (±100 mV, 500 ms) and of solution exchange, respectively. (c) Summary of peak currents from each of the treatment groups (shSCR, n = 10; shTRPM7, n = 7). Currents evoked by 0.1 mM Ca2+ were smaller in shTRPM7-infected than in shSCR-infected neurons (* indicates P < 0.05, two-tailed Student’s t test for unpaired samples, mean ± s.e.m). (d) Current-voltage relations derived from the voltage ramps in b applied to neurons from rats infected with shSCR (left) or shTRPM7 (right) rAAVs in ECS containing 1.3 mM (black) or 0.1 mM (red) Ca 2+. Consistent with the suppression of TRPM7 expression, current rectification was almost entirely abolished in neurons expressing shSCR, but not in those expressing shTRPM7, when the extracellular Ca 2+ concentration was reduced to 0.1 mM.
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
a r t ic l e s a ×3.5
shSCR
shTRPM7
b
shSCR
Figure 4 Persistent resilience of TRPM7-deficient hippocampi to ischemia 7 d post 4VO. Rats were microinjected with the rAAV shSCR and rAAVshTRPM7 in the left and right hippocampus, respectively, 7 d before 4VO. (a) Five coronal hippocampal sections taken from a representative rat 7 d after 4VO. Arrows indicate injection sites of rAAV shSCR (blue) and rAAVshTRPM7 (red). (b) Representative higher-magnification images of neurons and dendrites from the same rat. Scale bars in a and b represent 200, 50, 20 and 10 µm for images taken with ×3.5, ×25, ×63 and ×126 power objectives, respectively.
shTRPM7
×25
×63
and a paired-pulse procedure (Supplementary Fig. 8). Thus, TRPM7 suppression in the CA1 in vivo is compatible with the continued survival and functionality of the affected neurons.
are reduced20. Consistent with this, application of ECS containing 0.1 mM Ca2+ almost entirely abolished rectification in neurons infected with rAAVshSCR, but not with the rAAVshTRPM7, providing further evidence that TRPM7 function was downregulated (Fig. 2d). To further evaluate neuronal viability and function after rAAV infection, we carried out electrophysiological recordings from brain slices derived from adult rats infected 10 d previously with the vectors. After the recordings, we immunostained slices to confirm TRPM7 suppression (Supplementary Fig. 6). The loss of TRPM7 had no effect on short-term plasticity that was elicited immediately after high-frequency stimulation (HFS, data not shown), LTP in the CA1 region (Supplementary Fig. 6), input/output EGFP a relationships (Supplementary Fig. 7) or short-term plasticity of glutamatergic transmission, as tested using field recordings
TRPM7 suppression inhibits ischemic DND in CA1 neurons We next examined the role of TRPM7 in DND in the hippocampus following transient global cerebral ischemia. Global cerebral ischemia produces CA1 neuronal death within 24–72 h27,28 and causes longstanding memory deficits in rats5,6 and humans2,3. DND after global cerebral ischemia is partially responsive to treatment using several strategies, including anti-excitotoxic or antioxidant agents7,29. However, studies of cultured neurons that undergo hypoxia have suggested that TRPM7 activity potentiates oxidative pathways and promotes cell death even when excitotoxicity is blocked9. To test whether TRPM7 participates in DND, we stereotaxically microinjected the rAAVEGFP, rAAVshSCR or rAAVshTRPM7 vectors into the right hippocampus as described above (n = 6 rats per group). After 7 d, the rats were subjected to a transient (15 min) episode of forebrain ischemia TRPM7
Merge
NeuN
×63
c
shTRPM7 + 4VO
10 mV 100 ms
shTRPM7 + 4VO 20 15 10 5 0 100 200 300 400 500 Current injected (pA)
Instantaneous freq. (Hz)
Control
Number of action potentials
b
60
Control
50 40 30 20 10 1 2 3 4 Action potential number
+500 pA –200 pA
d
e
shTRPM7 + 4VO
Control 4 mV 25 ms
15
0 IPSP amplitude (mV)
Figure 5 Persistence of function in surviving TRPM7-deficient CA1 neurons 30 d after ischemia. Recordings were made from brain slices obtained 30 d after 4VO from rats previously microinjected with rAAVshTRPM7 and from age-matched (nonischemic) controls. (a) Direct EGFP, TRPM7 and NeuN immunofluorescence from the CA1 sector of a hippocampal slice taken from a representative rAAVshTRPM7-infected rat and stained post recording, confirming neuronal preservation and lack of TRPM7 staining at 30 d. Scale bar represents 20 µm. (b) Superimposed representative traces illustrating the responsiveness of neurons to the indicated series of current steps. (c) Similar response of the two neuronal populations (n = 8 per group) to increasing depolarizing current intensity (+100 to +500 pA) with respect to the number of action potentials fired (left) and spikefrequency accommodation (right). (d) Excitatory postsynaptic potential (EPSP, left) and IPSP (right) amplitude response of the two neuronal populations (n = 8 per group) to the indicated stimulus intensity. * indicates difference from controls (P < 0.05, two-way ANOVA with Bonferroni post-test). Inset: representative EPSP-IPSP traces. Error bars are s.e.m. (e) Representative traces illustrating unaltered paired-pulse facilitation of EPSCs.
EPSP amplitude (mV)
© 2009 Nature America, Inc. All rights reserved.
×126
10 5 0
*
–5
*
*
*
Control
shTRPM7 + 4VO
–10 –15
300
350 400 450 500 Stimulus intensity (µA)
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
300
350 400 450 500 Stimulus intensity (µA)
1303
a r t ic l e s
fEPSP slope (%)
a
300
Ischemia only
b
TRPM7 shRNA
Ischemia only Ischemia + shTRPM7
250
rAAV shSCR + 4VO
150 100
f
Fear conditioning
** ***
*
*=p<0.05 **=p<0.001 ***=p<0.0001
60 50 40 30
***
GFP
80 70
Mean % freezing
shTRPM7
+ 4VO
200
50 –20 –10 0 10 20 30 40 50 60 Time (min)
c
e rAAV
10
11
20
20
10
Sham rAAV shSCR 4VO + 4VO
d
rAAV shTRPM7 + 4VO
MWM probe trial 45 40 35
*
*
Goal Opposite
* Adjacent left Adjacent right
30 25 20
TRPM7
Percentage time spent in quadrant
© 2009 Nature America, Inc. All rights reserved.
0
NeuN
10
15 10 5 0
Sham
rAAV shSCR + 4VO
4VO
rAAV shTRPM7 + 4VO
using the 4-vessel occlusion technique (4VO; see Online Methods)27 and then allowed to recover for 72 h. We subjected sham controls (3 rats per group) to the same protocol, but without ischemia. Brain sections derived from the rats were examined to assess the survival and morphology of hippocampal neurons (Fig. 3). Ischemic rats treated with the rAAVEGFP or rAAVshSCR vectors showed an extensive, progressive loss of EGFP expression in the hippocampus (Fig. 3a and Supplementary Fig. 9). The EGFP loss is consistent with a turnover of existing EGFP (t1/2 of ~24 h)30 and inhibition of protein synthesis in global ischemia7. The latter is attributable to phosphorylation of the alpha subunit of eukaryotic initiation factor 2, as demonstrated in our rats (Supplementary Fig. 9), and leads to failure of translation initiation31. The remaining EGFP-labeled neurons showed dendritic beading, compaction and reduced dendrite density in all of the CA1 areas that we examined (Fig. 3a). In contrast, the hippocampi of ischemic rats infected with rAAVshTRPM7 robustly expressed EGFP, indicating that the neurons were viable and that the dendritic structures were preserved (Fig. 3a). We used a blinded observer to count the number of CA1 cells expressing EGFP, which reflect living infected neurons, and found significant preservation of EGFP-expressing CA1 neurons in the rAAVshTRPM7-treated rats, but not in those treated with control rAAVs (P < 0.01; Fig. 3e). CA1 neuronal numbers and dendritic structures were preserved in sham controls (no ischemia) throughout the observation period (Fig. 3b,e). EGFP expression is only a marker for neurons infected by the neurotropic rAAVs. However, ischemia causes astrocyte and microglial activation and inflammatory cell migration, which peak in the affected area around 3–4 d post-ischemia32. To determine whether uninfected cells were also protected, we repeated these experiments 1304
Figure 6 TRPM7 deficiency prevents loss of memory functions in rats subjected to global ischemia. (a,b) LTP was preserved in TRPM7-deficient hippocampi derived from rats subjected to ischemia. A summary of the LTP experiments in slices obtained ~30 d after 4VO in rats previously microinjected with rAAVshTRPM7 (filled triangles, n = 9) or in postischemic, uninjected age-matched controls (open circles, n = 8) is shown in a. Synaptic responses recorded during the last 5 min after LTP induction were averaged and compared using Student’s t test (unpaired, two-tailed). The differences between the two groups were significant (P < 0.05). Within each treatment group, the distribution of normalized fEPSP slope values, recorded 60 min after HFS, was plotted for each slice (b). LTP at t = 60 min, defined as an increase in normalized fEPSP of greater than two s.e.m. above baseline, was induced in 9 out of 9 slices from rAAVshTRPM7-treated rats, but in only 4 out of 8 slices in ischemiaonly rats. (c,d) Rats that underwent bilateral hippocampal injections of rAAVs and control rats were subjected to behavioral assays designed to test hippocampal-dependent memory. The results of constitutional fear conditioning 14 d after 4VO are shown in c. The numbers of rats per group are indicated in the bars. Probe trial performance in the MWM 11 d post-ischemia is shown in d. Rats underwent sham surgery, 4VO or 4VO 7d after microinjection of the indicated vector (n = 20 rats per group). * indicates P < 0.05. (e) Whole-brain EGFP fluorescence of representative rats that were tested in the MWM in d, at 30 d post 4VO. (f) Representative immunofluorescence sections of the CA1 sector of the shTRPM7-treated rat in e. Scale bar represents 25 µm. Error bars in a, c and d represent s.e.m.
(n = 6 ischemia rats per rAAV vector group) and processed the rat brains for terminal deoxynucleotidyl transferase dUTP nick-end labeling (TUNEL)33. TUNEL stains cells with DNA fragmentation, irrespective of cell type or whether they were infected by the vectors. Blinded cell counts revealed that hippocampi treated with the rAAVshTRPM7 vector had about 30% fewer TUNEL-positive cells as compared with the contralateral side or with rAAVEGFP- and rAAVshSCR-treated rats (Fig. 3c,d). Unlike our counts of EGFP-stained neurons, these TUNEL counts reflect the staining of fragmented DNA in uninfected neurons, activated astrocytes and microglia/macrophages, which accounted for up to 40% of TUNEL-stained cells 72 h after 4VO (Supplementary Fig. 10). These results indicate that global ischemia also affected cells that were not infected by the vectors and illustrate that selectively protecting hippocampal CA1 cells using a neurotropic vector–based approach does not prevent all TUNEL-detectable DNA fragmentation in the post-ischemic hippocampus. Because both hippocampi are rendered equally ischemic in the 4VO model, we next compared ischemic neuronal survival after microinjection of rAAVshTRPM7 and rAAVshSCR into the right and left hippocampus, respectively (Fig. 4a). We microinjected rats with the vectors 7 d before administering a 15-min 4VO ischemic insult (n = 6). Brain slices derived from the rats were examined 7 d later and we found that neuronal and dendritic morphology were preserved in the hippocampi injected with rAAVshTRPM7, but not in the rAAVshSCRtreated hippocampi (Fig. 4a,b). We counted the surviving (EGFP expressing) CA1 neurons and found that the rAAVshTRPM7-treated ischemic hippocampi retained neurons in numbers that were comparable (allowing for variability in counts between non-concurrent experiments) to those present in nonischemic infected brains (586.5 ± 134.84 and 14.67 ± 7.98 for rAAVshTRPM7 and rAAVshSCR rats, respectively; Student’s t test, P < 0.002). TRPM7 suppression enhances recovery post ischemia To examine the effect of TRPM7 suppression on the functionality of surviving cells, we examined the electrophysiological properties of CA1 neurons in brain slices taken from post-ischemic rats infected with rAAVshTRPM7 7 d before 4VO. The extensive hippocampal VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
© 2009 Nature America, Inc. All rights reserved.
a r t ic l e s damage sustained by post-ischemic rats treated with rAAVshSCR or rAAVEGFP prevented us from obtaining stable recordings. Therefore, the data from rAAVshTRPM7-treated rats were compared to those from uninfected nonischemic controls. Whole-cell currents and field recordings were difficult to obtain in rAAVshTRPM7-treated slices taken 7–14 d post ischemia, suggesting that the surviving cells were still fragile or metabolically compromised. At 30 d post-ischemia, however, we easily obtained whole-cell current- and voltage-clamp recordings (Supplementary Notes) and a robust EGFP fluorescence in the rAAVshTRPM7-treated CA1 neurons indicated that shRNA construct expression persisted and demonstrated the long-term survivability of the neurons (Fig. 5a). TRPM7-deficient neurons surviving the ischemic insult had a normal capacity to fire action potentials and had normal spike-frequency adaptation (Fig. 5b,c). Surviving TRPM7-deficient neurons also maintained dendritic function, as suggested by the preservation of excitatory and inhibitory synaptic current and our observation of spontaneous excitatory and inhibitory postsynaptic currents (Fig. 5d and Supplementary Notes). We observed a depression of inhibitory postsynaptic potential (IPSP) amplitudes in surviving TRPM7-deficient neurons (Fig. 5d). The mechanism underlying such depression remains unclear and could reflect potential modification of GABAA receptor expression or function, reduced excitatory drive onto interneurons, or, alternatively, an injury to interneurons. The neurons showed unaltered paired-pulse facilitation of excitatory postsynaptic currents (EPSCs; pulse 2/pulse 1 ratios were 1.41 ± 0.08 and 1.51 ± 0.10 in neurons of control and rAAVshTRPM7 + 4VO rats, respectively; n = 8 per group), suggesting that presynaptic function is normal (Fig. 5e). Thus, with the exception of evoked IPSPs, intrinsic membrane properties and synaptic responses in post-ischemic TRPM7-deficient neurons were largely indistinguishable from those of nonischemic controls, suggesting that these cells were functionally preserved (see Supplementary Table 1 and Supplementary Notes). Certain forms of memory require intact hippocampal neuronal networks34,35. We next determined whether the functional neuronal preservation that we observed during TRPM7 suppression (Fig. 5) would translate into a capacity to induce LTP in the hippocampal CA1 area by HFS. LTP induction by HFS has been the primary model used to study the cellular and molecular basis of memory35,36. For these experiments, we bilaterally injected rAAVshTRPM7 into the rats’ hippocampus. Controls underwent sham injection (no rAAV infection) and all of the rats were subjected to ischemia by 4VO 7 d later. We measured LTP, paired-pulse responses and input-output functions in brain slices taken from rats in each group 30 d after the 4VO procedure. We found no changes in paired-pulse responses or input-output functions in these recordings (data not shown). However, rats that had undergone TRPM7 suppression in the hippocampus before ischemia had a greater capacity for LTP as compared with rats that experienced ischemia alone (Fig. 6a,b). Synaptic response recording during the last 5 min after LTP induction confirmed a significant difference between the TRPM7-deficient and the ischemia-only groups (P < 0.05). Consistent with the human situation2,3, 4VO in rats leads to behavioral consequences, mainly in the performance of learning and memory tasks5,6. To determine whether these were preserved postischemia by TRPM7 suppression, we subjected the rats to two behavioral assays that test hippocampal-dependent memory: a contextual fear conditioning procedure37 and the fixed platform version of the Morris water maze (MWM)38. We carried out bilateral hippocampal injections of rAAVshTRPM7 or rAAVshSCR, followed 7 d later by 4VO. Additional controls underwent 4VO without rAAV infection or sham surgery (no ischemia or rAAV infection). To study the rats’ ability to
learn and remember an association between an adverse experience and environmental cues, we evaluated contextual fear conditioning at 14 d post-ischemia (Online Methods). All of the groups had similar freezing scores before conditioning (data not shown). The freezing scores of the rAAVshTRPM7 + 4VO group were similar to those of sham-operated (non-ischemic) rats 24 h after conditioning (P > 0.05; Fig. 6c), and those scores were higher than those of the rAAVshSCR + 4VO rats (P < 0.05) and 4VO only controls (P < 0.001; Fig. 6c). Thus, 4VO impaired the acquisition of contextual fear memory, whereas TRPM7 suppression preserved fear memory performance post-ischemia. The MWM test of spatial navigation measures the ability of the rat to learn and remember a location defined by its position relative to distal extramaze cues. We subjected the rats to MWM learning acquisition trials for 5 consecutive days (7–11 d post-ischemia; Online Methods) 7 d after 4VO. A probe trial that measured the ability to locate the quadrant that previously contained the hidden platform was carried out 4 h after the final acquisition trial on day 5. The rate of task acquisition was initially slowed in all of the groups subjected to ischemia, as evidenced by the increased time to find the hidden platform (Supplementary Fig. 11). There was no evidence of impaired motor performance, as indicated by unaffected swim speeds on any of the days of testing or the probe trial (Supplementary Fig. 11). However, the impairment in spatial memory demonstrated during the probe trial was only observed in uninfected post-ischemic rats and in post-ischemic rats infected with rAAVshSCR. Post-ischemic rats deficient in TRPM7 performed similarly to sham surgery rats in remembering the quadrant that previously contained the hidden platform (Fig. 6d), indicating that their capacity to learn the task was preserved, as was their memory performance. Examination of the brains revealed that GFP fluorescence persisted (Fig. 6e,f) and that TRMP7 was suppressed (as shown by immunoreactivity), indicating continued construct expression in the tested rats (Fig. 6f, compare with Fig. 1f).
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1305
DISCUSSION Collectively, our data show that TRPM7 suppression in hippocampal CA1 neurons in vivo is well tolerated, imparts resilience to ischemic damage, and preserves neuronal function and performance for hippocampus-dependent learning tasks after ischemic brain injury. Although our prior work has implicated TRPM7 in the anoxic death of cultured neurons9, other studies have suggested that over- or underexpressing TRPM7 may have adverse consequences. For example, its overexpression leads to the death of HEK-293 cells12, its disruption affects the survival of cultured DT-40B cells12,13 and its deletion in T cells elicits dysregulated synthesis of growth factors that are necessary for thymopoesis14. TRPM7 is also essential for embryonic development and its global deletion in mice is lethal14. To the best of our knowledge, our results are the first demonstration that suppressing TRPM7 in adult mammals is feasible and that this markedly reduces delayed neuronal death after ischemia. Thus, there is a possibility that TRPM7 can be selectively targeted to prevent ischemic brain damage. Recently, we demonstrated that TRPM7 channels in cultured CA1 neurons may act as extracellular detectors of divalent cation levels20. Tissue ischemia in vivo causes a reduction in extracellular Ca2+ and Mg2+ concentrations16,17 and such conditions enhance TRPM7 currents20. Thus, ischemia may promote ionic circumstances that are compatible with TRPM7 overactivity. In cultures, such overactivity leads to free and total intracellular calcium overload, which results in nitric oxide production, oxidative stress and cell death. These
© 2009 Nature America, Inc. All rights reserved.
a r t ic l e s mechanisms are similar to those previously attributed to excitotoxicity, but are now known to occur even if excitotoxicity and other known Ca2+ influx pathways are pharmacologically blocked9. Moreover, excitotoxicity may be self-limiting, as glutamate receptors desensitize and anoxic cell membranes depolarize, and this leads to reductions in spontaneous activity and neurotransmitter release9. Consequently, TRPM7 proteins may govern cell death pathways that can be triggered independently of excitotoxicity, but converge on similar downstream injurious pathways. If so, then anti-excitotoxic therapy may be insufficient for treating ischemic brain damage in stroke or other conditions involving DND39,40. Our report supports an increasing body of literature that describes non-excitotoxic mechanisms of neuronal death, including those involving TRP channels9,21, acid-sensing channels41 and hemichannels42. Our finding that TRPM7 channels mediate DND in brain ischemia identifies these proteins and their downstream signaling pathways as targets for future research. However, our use of siRNA to regionally inhibit TRPM7 is limited by several uncertainties. First, the functional role of TRPM7 channels in CA1 neurons is still poorly understood. Although we have demonstrated that they may act as divalent cation sensors in cultured hippocampal neurons20, the physiological purpose of this is incompletely understood, as is the effect of these observations on neuronal function. The rAAV approach only enhanced the resilience of a subset of cells (infected neurons) and this may need to be improved to treat other cell types affected by brain ischemia. The siRNA approach was insufficient to fully elucidate the functional role of TRPM7 channels in neurons, and such future research will require, among others, the development of specific pharmacological inhibitors for ion channel studies. A further uncertainty of the use of siRNA is the inability of this approach to completely block a protein’s expression or function. This raises the possibility that, although complete global deletion of TRPM7 is lethal14, a partial suppression may be well tolerated and would not require a regional strategy. However, in the absence of specific pharmacological inhibitors of TRPM7, this possibility remains to be tested. Further experiments are warranted to determine whether the effect of TRPM7 suppression lasts beyond the times that we examined and whether the effects that we observed involve the preservation of original cells, enhanced neurogenesis, plasticity or other means by which TRPM7 suppression might improve outcome. Lastly, the chronic nature of TRPM7 suppression by RNAi raises the possibility that its partial deletion causes sublethal stress that induces unknown signals that induce an ischemia-resistant state. Although TRPM7 is thought to be important for cellular trace ion and Mg2+ homeostasis under certain circumstances11,13, its deletion in thymocytes disrupts thymopoiesis, but does not affect acute uptake of Mg2+ or the maintenance of total cellular Mg2+ (ref. 14). It is possible that diverse cell types have different dependencies on TRPM7 or different redundant mechanisms that compensate for this protein’s functions. Nevertheless, our results provide a rationale for developing pharmacologic means of targeting TRPM7 channels to further study their physiological and pathological roles in vitro and in vivo and to inhibit ischemic damage. The knowledge that these channels are broadly expressed in vertebrate tissues should not dissuade such research. Many commonly used drugs, such as calcium channel blockers for the treatment of hypertension, also target widely expressed proteins, but are now in commonly used in humans. Instead, the awareness of the broad expression of TRPM7 channels in and outside the CNS raises the possibility that they might also participate in other neurodegenerative disorders, as well as in ischemic mechanisms of other tissues in which they are found.
Methods Methods and any associated references are available in the online version of the paper at http://www.nature.com/natureneuroscience/.
1306
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
Note: Supplementary information is available on the Nature Neuroscience website. Acknowledgments We thank M. Aarts and W. Czerwinski for technical information, J.C. Roder for advice on MWM testing, and A. Fleig, C. Montell, A. Scharenberg and E. Lo for a critical review of the manuscript. This work was supported by grants from the Canadian Institute of Health Research (MOP68939 and MOP89720 to M.T. and MOP15514 to J.F.M.), the US National Institutes of Health (NS048956 to M.T.), the Canadian Stroke Networks (M.T., J.F.M. and M.F.J.) and the Krembil Seed fund (M.T.). H.-S.S. is a recipient of Postdoctoral Fellowship Awards from the Heart and Stroke Foundation of Canada Focus on Stroke Training Initiative Program. AUTHOR CONTRIBUTIONS H.-S.S. carried out the stereotactic rAAV infections, sectioning, immunochemistry, imaging, cell counts, laser dissection microcapture and PCR. K.J. and T.E.G. designed and manufactured the rAAV vectors, M.F.J. and J.F.M. performed the electrophysiology experiments, L.T. carried out the 4VO and histology procedures, Y.M. generated the antibodies to TRPM7, S.K. and H.C. performed the immunoblots, M.J. carried out the TUNEL cell counts, and J.P.F., M.J., H.C. and H.-S.S. performed the OGD experiments. L.J.M. and B.A.O. carried out the neurobehavioral evaluations. M.T. and H.-S.S. wrote the paper. All authors discussed the results and commented on the manuscript. Published online at http://www.nature.com/natureneuroscience/. Reprints and permissions information is available online at http://www.nature.com/ reprintsandpermissions/. 1. Hausenloy, D.J. & Scorrano, L. Targeting cell death. Clin. Pharmacol. Ther. 82, 370–373 (2007). 2. Volpe, B.T. & Petito, C.K. Dementia with bilateral medial temporal lobe ischemia. Neurology 35, 1793–1797 (1985). 3. Petito, C.K., Feldmann, E., Pulsinelli, W.A. & Plum, F. Delayed hippocampal damage in humans following cardiorespiratory arrest. Neurology 37, 1281–1286 (1987). 4. Bennett, M.V. et al. The GluR2 hypothesis: Ca2+-permeable AMPA receptors in delayed neurodegeneration. Cold Spring Harb. Symp. Quant. Biol. 61, 373–384 (1996). 5. Volpe, B.T., Pulsinelli, W.A., Tribuna, J. & Davis, H.P. Behavioral performance of rats following transient forebrain ischemia. Stroke 15, 558–562 (1984). 6. Block, F. Global ischemia and behavioral deficits. Prog. Neurobiol. 58, 279–295 (1999). 7. Lipton, P. Ischemic cell death in brain neurons. Physiol. Rev. 79, 1431–1568 (1999). 8. Besancon, E., Guo, S., Lok, J., Tymianski, M. & Lo, E.H. Beyond NMDA and AMPA glutamate receptors: emerging mechanisms for ionic imbalance and cell death in stroke. Trends Pharmacol. Sci. 29, 268–275 (2008). 9. Aarts, M. et al. A key role for TRPM7 channels in anoxic neuronal death. Cell 115, 863–877 (2003). 10. Montell, C., Birnbaumer, L. & Flockerzi, V. The TRP channels, a remarkably functional family. Cell 108, 595–598 (2002). 11. Monteilh-Zoller, M.K. et al. TRPM7 provides an ion channel mechanism for cellular entry of trace metal ions. J. Gen. Physiol. 121, 49–60 (2003). 12. Nadler, M.J. et al. LTRPC7 is a Mg.ATP-regulated divalent cation channel required for cell viability. Nature 411, 590–595 (2001). 13. Schmitz, C. et al. Regulation of vertebrate cellular Mg2+ homeostasis by TRPM7. Cell 114, 191–200 (2003). 14. Jin, J. et al. Deletion of Trpm7 disrupts embryonic development and thymopoiesis without altering Mg2+ homeostasis. Science 322, 756–760 (2008). 15. Siesjo, B.K., Katsura, K. & Tibor, K. Acidosis related brain damage. in Advances in Neurology: Cellular and Molecular Mechanisms of Ischemic Brain Damage (eds. Siesjo, B.K. & Wieloch, T.) (Raven Press, New York, 1994). 16. Silver, I.A. & Erecinska, M. Intracellular and exctracellular changes of [Ca2+] in hypoxia and ischemia in rat brain in vivo. J. Gen. Physiol. 95, 837–866 (1990). 17. Lin, M.C. et al. Microdialysis analyzer and flame atomic absorption spectrometry in the determination of blood glucose, lactate and magnesium in gerbils subjected to cerebral ischemia/reperfusion. J. Am. Coll. Nutr. 23, 556S–560S (2004). 18. Runnels, L.W., Yue, L. & Clapham, D.E. TRP-PLIK, a bifunctional protein with kinase and ion channel activities. Science 291, 1043–1047 (2001). 19. Kozak, J.A., Kerschbaum, H.H. & Cahalan, M.D. Distinct properties of CRAC and MIC channels in RBL cells. J. Gen. Physiol. 120, 221–235 (2002). 20. Wei, W.L. et al. TRPM7 channels in hippocampal neurons detect levels of extracellular divalent cations. Proc. Natl. Acad. Sci. USA 104, 16323–16328 (2007). 21. Kaneko, S. et al. A critical role of TRPM2 in neuronal cell death by hydrogen peroxide. J. Pharmacol. Sci. 101, 66–76 (2006). 22. Bridge, A.J., Pebernard, S., Ducraux, A., Nicoulaz, A.L. & Iggo, R. Induction of an interferon response by RNAi vectors in mammalian cells. Nat. Genet. 34, 263–264 (2003).
a r t ic l e s 33. Gavrieli, Y., Sherman, Y. & Ben-Sasson, S.A. Indentification of programmed cell death in situ via specific labeling of nuclear DNA fragmentation. J. Cell Biol. 119, 493–501 (1992). 34. Ji, J. & Maren, S. Hippocampal involvement in contextual modulation of fear extinction. Hippocampus 17, 749–758 (2007). 35. Whitlock, J.R., Heynen, A.J., Shuler, M.G. & Bear, M.F. Learning induces long-term potentiation in the hippocampus. Science 313, 1093–1097 (2006). 36. Lisman, J., Schulman, H. & Cline, H. The molecular basis of CaMKII function in synaptic and behavioral memory. Nat. Rev. Neurosci. 3, 175–190 (2002). 37. Fanselow, M.S. Conditioned and unconditional components of post-shock freezing. Pavlov. J. Biol. Sci. 15, 177–182 (1980). 38. Morris, R. Developments of a water-maze procedure for studying spatial learning in the rat. J. Neurosci. Methods 11, 47–60 (1984). 39. Morris, G.F. et al. Failure of the competitive N-methyl-D-aspartate antagonist Selfotel (CGS 19755) in the treatment of severe head injury: results of two phase III clinical trials. The selfotel investigators. J. Neurosurg. 91, 737–743 (1999). 40. Davis, S.M., Albers, G.W., Diener, H.C., Lees, K.R. & Norris, J. Termination of acute stroke studies involving selfotel treatment. ASSIST steering committed. Lancet 349, 32 (1997). 41. Xiong, Z.G. et al. Neuroprotection in ischemia: blocking calcium-permeable acidsensing ion channels. Cell 118, 687–698 (2004). 42. Thompson, R.J., Zhou, N. & MacVicar, B.A. Ischemia opens neuronal gap junction hemichannels. Science 312, 924–927 (2006).
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1307
© 2009 Nature America, Inc. All rights reserved.
23. Sledz, C.A., Holko, M., de Veer, M.J., Silverman, R.H. & Williams, B.R. Activation of the interferon system by short-interfering RNAs. Nat. Cell Biol. 5, 834–839 (2003). 24. Schmitz, C. et al. The channel kinases TRPM6 and TRPM7 are functionally nonredundant. J. Biol. Chem. 280, 37763–37771 (2005). 25. Xiong, Z.G., Chu, X.P. & MacDonald, J.F. Effect of lamotrigine on the Ca2+-sensing cation current in cultured hippocampal neurons. J. Neurophysiol. 86, 2520–2526 (2001). 26. Xiong, Z., Lu, W. & MacDonald, J.F. Extracellular calcium sensed by a novel cation channel in hippocampal neurons. Proc. Natl. Acad. Sci. USA 94, 7012–7017 (1997). 27. Pulsinelli, W.A. & Brierly, J.B. A new model of bilateral hemispheric ischemia in the unanesthetized rat. Stroke 10, 267–272 (1979). 28. Pulsinelli, W.A., Brierly, J.B. & Plum, F. Temporal profile of neuronal damage in a model of transient forebrain ischemia. Ann. Neurol. 11, 491–498 (1982). 29. Saito, A. et al. Oxidative stress and neuronal death/survival signaling in cerebral ischemia. Mol. Neurobiol. 31, 105–116 (2005). 30. Corish, P. & Tyler-Smith, C. Attenuation of green fluorescent protein half-life in mammalian cells. Protein Eng. 12, 1035–1040 (1999). 31. Kumar, R. et al. Brain ischemia and reperfusion activates the eukaryotic initiation factor 2α kinase, PERK. J. Neurochem. 77, 1418–1421 (2001). 32. Stoll, G., Jander, S. & Schroeter, M. Inflammation and glial responses in ischemic brain lesions. Prog. Neurobiol. 56, 149–171 (1998).
ONLINE METHODS All experiments were carried out in compliance with the relevant laws and guidelines set by the Canadian Council for Animal Care and with the approval of the University Health Network Animal Care Committee.
© 2009 Nature America, Inc. All rights reserved.
Expression constructs. An siRNA sequence targeted to TRPM7, corresponding to coding regions 5,152–5,172 relative to the first nucleotide of the start codon of murine TRPM7 (AAG AGT GCA TGA CTG GTG AAT, GenBank accession number AY032951)9, or a mismatch negative control siRNA sequence (ACT ACC GTT GTA TAG GTG) was inserted into the pSilencer 3.0-H1 expression vector (Ambion). The H1 promoter-sense-loop-antisense-termination region was excised and subcloned into the pAdTrack adeno-associated viral shuttle vector to produce a vector having EGFP driven by a separate CMV promoter. We previously studied this 21-bp siRNA sequence, but only as delivered using a conventional siRNA transfection reagent or as an shRNA packaged in an adenoviral vector. These approaches suppressed TRPM7 mRNA and activity in both cortical9 and hippocampal cultures20, but adenoviral vectors only infected a small proportion of exposed neurons. Here, we used the same vector, but packaged in an AAV vector. Further details are provided in the Supplementary Notes. AAVs. HEK293 cells were transfected with serotype 1 adeno-associated viruses (AAV1) and helper plasmids using standard CaPO4 transfection. Cells were harvested 60 h after transfection and AAV1 vectors were purified from the cell lysate by ultracentrifugation through an iodixanol density gradient followed by Q column purification, and then concentrated and dialyzed using phosphate-buffered saline (PBS). Vectors were titered using real-time PCR (ABI Prism 7700) and all vector stocks diluted to 3.8 × 1012 genomes per ml43. Further details are provided in the Supplementary Notes. Viral vector administration in vivo. Under ketamine/xylazine anesthesia (100 mg and 50 mg per kg of body weight, respectively), we made a midline skin incision between the bregma and interaural line. The stereotaxic coordinates for the hippocampus were 3.3 mm posterior to bregma (anterior-posterior), 2.0 mm lateral to the midline (medial-lateral) and 2.6 mm below the dura (dorsalventral)44. rAAV vectors (3.8 × 1012 genomes per ml) were microinjected through a 1-mm craniotomy in a volume of 2 µl of rAAV stock plus 1 µl of 20% mannitol (vol/vol; Sigma) in PBS45 using a 10-µl Hamilton syringe with a 26 gauge needle. The infusion rate was 0.2 µl min−1 for 15 min and the needle stayed in place for another 10 min after the injection45. The needle was then slowly withdrawn and bone wax was applied to the craniotomy. Rats were allowed to recover for 7 to 14 d. Immunohistochemistry of hippocampal sections and cell cultures. Immunostaining was performed as described previously46. In brief, hippo campal sections were blocked using 3% normal goat or rabbit serum (vol/vol), 0.3% Triton X-100 (vol/vol) and 1% BSA (vol/vol) in PBS at 20–22 °C for 90 min. Neuronal cell cultures on coverslips treated with rAAVEGFP, rAAVshSCR and rAAVshTRPM7 were fixed and cryo-protected in 2.5% sucrose (vol/vol) with 4% paraformaldehyde (wt/vol) in PBS at 20–22 °C for 20 min. The coverslips were then blocked using 1% BSA, 3% goat serum and 0.3% Triton X-100 in PBS solution for 90 min. Samples were then double or triple labeled with goat antibody to TRPM7 (#ab729, 1:50, Abcam) or rabbit antibody to TRPM2 (#ab11168, 1:50, Abcam) and mouse antibody to NeuN (#MAB377, 1:100, Chemicon) overnight at 4 °C in a rocker. NeuN is a neuron-specific marker47. The sections were subsequently washed in PBS and blocked briefly with the blocking solution. Subsequently, the sections were incubated with the affinity-purified second antibody, rabbit antibody to goat Alexa 568 (1:100, Molecular Probes), goat antibody to rabbit Alexa568 (1:100, Molecular Probes) and goat antibody to mouse Alexa 350 (1:100, Molecular Probes) for 1 h at 20–22 °C. Slides were finally coverslipped and mounted using ProLong Gold antifade reagent with and without DAPI (Invitrogen and Molecular Probes). Hippocampal slice preparation for electrophysiology. Wistar rats, infected with the rAAV vectors by stereotaxic injection, were anesthetized with isoflurane and decapitated. The brains were rapidly removed and submerged in chilled (4 °C) artificial cerebrospinal fluid (aCSF) comprising 124 mM NaCl, 3 mM KCl, 2.6 mM CaCl2, 1.3 mM MgCl2, 26 mM NaHCO3, 1.25 mM NaH2PO4 and 10 mM glucose (300–310 mOsmol, pH 7.4). Both hippocampi were isolated, mounted
nature NEUROSCIENCE
on a block of agar and sliced (300–400 µm) using a vibratome (VT1000E, Leica). After at least 1 h in a submerged holding chamber at 20–22 °C, they were transferred to a recording chamber on an Olympus BX51WI microscope equipped with differential interference contrast and epifluorescence optics and were continuously superfused with aCSF at a rate of 3 ml min−1. EGFP fluorescence, a marker of successful infection by rAAV vectors, was confirmed in all slices before recording. Field recordings were conducted at 31 ± 2 °C and whole-cell voltage- and current-clamp recordings were made at 20–22 °C. Further details are provided in Supplementary Notes. Field recordings and LTP. Field EPSPs (fEPSPs) were evoked every 20 s (0.05 Hz) by electrical stimulation (100-µs duration) delivered to the Schaffer-collateral pathway using a concentric bipolar stimulating electrode (25-µm exposed tip) and recorded using glass microelectrodes (3–5 MΩ, filled with aCSF) positioned in the stratum radiatum of the CA1 area. The input-output relationship was determined in each slice by varying the stimulus intensity (100–1,000 µA) and recording the corresponding fEPSP. Using a stimulus intensity that evoked a 30–40% of maximum fEPSP, paired-pulse responses were recorded by delivering two electrical stimuli in rapid succession at intervals (interpulse interval) ranging from 10–1,000 ms. Following this assessment, the stability of synaptic responses, evoked using the same stimulus intensity as used for the paired-pulse protocol, was monitored for 20 min, at which time LTP was induced by three 1-s, 100-Hz trains of stimuli delivered 20 s apart. All field data are expressed as mean ± s.e.m. Statistical difference between means was determined by ANOVA. Further details are provided in the Supplementary Notes. Recordings from acutely isolated hippocampal neurons. The preparation of acutely isolated neurons were performed as previously described48. Briefly, Wistar rats, infected by stereotaxic injection at 3 weeks of age with the rAAV vectors as described, were anesthetized with halothane and decapitated at 4–5 weeks of age. Hippocampi were rapidly removed and placed in cold, oxygenated ECS, cut with a razor blade into 0.5–1 mm slices and allowed to recover at 20–22 °C for 45 min. Slices were digested for 30 min at 20–22 °C in ECS containing 2.5 mg ml−1 papaya latex (Sigma) and then maintained, until needed, in oxygenated ECS for periods of up to 8–10 h. GFP fluorescence, a marker of successful infection by rAAV vectors, was confirmed in all slices before recording. Hippocampal CA1 pyramidal neurons were isolated from the CA1 region using fine forceps to mechanically abrade the CA1 region of GFP-positive slices. Details of the tight-seal whole-cell recordings are provided in the Supplementary Notes. Global cerebral ischemia (4VO). Male Wistar rats (250–300 g, Charles River) underwent hippocampal microinjections of rAAV vectors 7 d prior. On the day of ischemia, they were anesthetized with isoflurane (3% induction and 1–1.5% maintenance, mixed with oxygen) and breathed spontaneously. Body temperature was controlled and maintained at 37 ± 0.5 °C by a heating lamp. A 15-min transient global ischemia was induced by 4VO (occluding both common carotid arteries and vertebral arteries)27. During the carotid artery ligation, the following parameters were assessed to assure the successful occlusion: completely flat bi-temporal electroencephalogram, dilated pupils, absence of corneal reflex and steady body temperature readings. Rats were allowed to recover for 3, 7 or 30 d before any further immunohistochemistry, histological, functional or behavioral evaluations. Contextual fear memory. Fear conditioning was performed as previously described49. In brief, the conditioning chamber consisted of a Perspex arena with a light mounted in the lid (350 × 200 × 193 mm, Technical and Scientific Equipment). The floor consisted of stainless steel bars (4 mm in diameter, 5 mm apart) that were connected to a computer, which controlled the duration, timing and intensity of the shock. On day 1, single subjects were allowed to explore the chamber for 180 s. They then received three unsignaled foot shocks (2-s duration, 0.7-mA intensity) at 60-s intervals. On day 2, 24 h after the conditioning session, the rats were returned to the same chamber and the freezing response was assessed immediately and then every 8 s for 8 min. The freezing response was defined as the lack of any movement except that required for breathing. The contextual fear conditioning scores were analyzed with a one-way ANOVA, and post hoc analysis consisted of the Tukey-Kramer method for unequal sample sizes with the P value set at 0.05.
doi:10.1038/nn.2395
© 2009 Nature America, Inc. All rights reserved.
MWM. The MWM consisted of a circular pool (2.0 m in diameter, 0.8 m high) constructed of white fiberglass. The water was maintained at 22 ± 2 °C and was made opaque by the addition of a white nontoxic paint. During testing in the water maze, a platform (15 cm in diameter) was located 1.5 cm below the water surface in one of four quadrant locations, approximately 35 cm from the sidewalls. The pool was surrounded by many external extra-maze cues. A video camera was mounted in the ceiling above the pool and was connected to a video-recorder and tracking device (HVS 2020 water maze software), which permitted on- and off-line automated tracking of the path taken by the rats. The rats were subjected to four learning acquisition trials per session 7 d after a 15-min 4VO insult for five consecutive days (7–11 d post-ischemia), during which they were trained to locate the hidden escape platform, which remained in a fixed location throughout the testing. Trials lasted a maximum of 60 s and if a rat failed to find the platform in this time frame, it was gently guided to the platform by the experimenter. The rats were allowed to remain on the platform for 10 s to increase the difficulty of the task, with a 1-min rest period between trials. The rats were then tested in a probe trial 4 h after the final session on day 5 (11 d post-ischemia). For the probe trial, the platform was removed from the pool and the rat was released from the quadrant opposite where the platform had been initially located. The length of the probe trial was 60 s, after which the rat was taken out of the pool. The proportion of time the rat spent searching for the platform in the training quadrant, that is, the previous location of the platform, was recorded and used as a measure of spatial memory retention. Swim speed was also recorded to determine whether differences in motivation or motor impairment were caused by 4VO or the rAAV injection. Multivariate analysis of variance with one level repeated (day of testing)
doi:10.1038/nn.2395
and one level not repeated (treatment) was used to analyze the acquisition and the swim speed data over the days of training. The probe trial data were analyzed using a two-way ANOVA, and post hoc analysis consisted of the Tukey-Kramer method with the P set at 0.05. Statistics. Data are presented as mean ± s.e.m. Unless otherwise indicated, group data were compared using one-way ANOVA and the Fisher LSD test (SigmaStat 3.0, SPSS). All experiments and analyses were performed by observers blinded to the treatment groups. 43. Lawlor, P.A. et al. Novel rat Alzheimer’s disease models based on AAV-mediated gene transfer to selectively increase hippocampal Aβ levels. Mol. Neurodegener. 2, 11 (2007). 44. Paxinos, G. & Watson, C. The Rat Brain in Stereotaxic Coordinates (Academic Press, San Diego, 1998). 45. Mastakov, M.Y., Baer, K., Xu, R., Fitzsimons, H. & During, M.J. Combined injection of rAAV with mannitol enhances gene expression in the rat brain. Mol. Ther. 3, 225–232 (2001). 46. Sun, H.S., Feng, Z.P., Miki, T., Seino, S. & French, R.J. Enhanced neuronal damage after ischemic insults in mice lacking Kir6.2-containing ATP-sensitive K+ channels. J. Neurophysiol. 95, 2590–2601 (2006). 47. Mullen, R.J., Buck, C.R. & Smith, A.M. NeuN, a neuronal specific nuclear protein in vertebrates. Development 116, 201–211 (1992). 48. Wang, L.Y. & MacDonald, J.F. Modulation by magnesium of the affinity of NMDA receptors for glycine in murine hippocampal neurones. J. Physiol. (Lond.) 486, 83–95 (1995). 49. Cheng, V.Y. et al. α5GABAA receptors mediate the amnestic, but not sedativehypnotic, effects of the general anesthetic etomidate. J. Neurosci. 26, 3713–3720 (2006).
nature NEUROSCIENCE
a r t ic l e s
Approach sensitivity in the retina processed by a multifunctional neural circuit
© 2009 Nature America, Inc. All rights reserved.
Thomas A Münch1,3,4, Rava Azeredo da Silveira2,4, Sandra Siegert1, Tim James Viney1, Gautam B Awatramani1,3 & Botond Roska1 The detection of approaching objects, such as looming predators, is necessary for survival. Which neurons and circuits mediate this function? We combined genetic labeling of cell types, two-photon microscopy, electrophysiology and theoretical modeling to address this question. We identify an approach-sensitive ganglion cell type in the mouse retina, resolve elements of its afferent neural circuit, and describe how these confer approach sensitivity on the ganglion cell. The circuit’s essential building block is a rapid inhibitory pathway: it selectively suppresses responses to non-approaching objects. This rapid inhibitory pathway, which includes AII amacrine cells connected to bipolar cells through electrical synapses, was previously described in the context of night-time vision. In the daytime conditions of our experiments, the same pathway conveys signals in the reverse direction. The dual use of a neural pathway in different physiological conditions illustrates the efficiency with which several functions can be accommodated in a single circuit. In animals1–5 and humans6,7, approaching motion, such as that of looming objects, elicits well-documented behaviors, such as startle and protective motor responses. The relevance of approaching motion to survival, the requirement for rapid action upon the detection of an approach event and the stereotypical nature of motor responses all suggest the existence of dedicated neural hardware for the detection of approaching motion. Neurons have indeed been identified in locust8–10 and in pigeon11 that respond selectively to approaching motion stimuli. These are found in higher visual areas and seem to achieve a sophisticated whole-field computation that predicts the collision time of the looming object. Here we describe an approach-sensitive neuron in the mammalian retina, namely a mouse ganglion cell type. We elucidate elements of its afferent circuit and show how these allow the approach-sensitive behavior of the ganglion cell. An important component of the circuit is a rapid inhibitory pathway that relies upon an electrical synapse. Finally, we summarize the mechanism of approach sensitivity in a computational model of a ‘composite receptive field’. Experiments in frog4,5 have suggested that approach detection may occur in the retina, but the mechanism there4,5 seems to be different from the one we discuss here. RESULTS A ganglion cell type is sensitive to approaching motion We recorded from ganglion cells, the output neurons of the retina, in transgenic mice (PvalbCre × Thy1Stp-EYFP) in which a few ganglion cell types, and no other cell classes, were brightly labeled with enhanced yellow fluorescent protein (EYFP)12,13 (Fig. 1a). We identified seven
labeled ganglion cell types and denoted them PV-1 to PV-7 in order by depth of their dendritic arborizations within the inner plexiform layer (IPL). With a two-photon microscope, we distinguished these genetically labeled ganglion cells on the basis of their morphology and targeted them for recording. Here we discuss one OFF ganglion cell type, which we refer to as the PV-5 cell. The dendrites of this neuron arborized in the IPL at 80.6 ± 0.8% (n = 44; Fig. 1b top panel and Supplementary Fig. 1) relative to two strata marked by an antibody to choline acetyltransferase (ChAT)14,15. The dendritic tree of each PV-5 cell extended over a large area (diameter 350 ± 30 µm, n = 20; Fig. 1b, bottom panel), covering ~10° of the visual field. Unless otherwise noted, experiments were performed with PV-5 cells in isolated and light-adapted wholemount retinas. We presented PV-5 cells with a set of stimuli that mimicked approaching motion (an expanding bar), lateral motion (a drifting bar) and receding motion (a shrinking bar) of a dark object within the dendritic field (Fig. 1c). All stimuli began with the presentation of a black bar at the center of the dendritic field. After a 2-s pause during which the image was held fixed, the two edges of the bar moved at different velocities to the left or to the right, drawn randomly from a set of velocities (Fig. 1c). Spiking responses were evoked preferentially by expanding bars compared to either drifting or shrinking bars (n = 7), at all velocities of the bar edges tested (Fig. 1c,d). The absent (or greatly reduced) response to non-approaching motion was not due to an overall change in the responsiveness of the cell during the course of the experiment, as the response to the onset of the black bar was similar for all stimuli (Fig. 1c). The peak spiking rate increased monotonically with edge velocity, while the spike count
1Neural
Circuit Laboratories, Friedrich Miescher Institute for Biomedical Research, Basel, Switzerland. 2Department of Physics and Department of Cognitive Studies, École Normale Supérieure, Paris, France. 3Present addresses: Laboratory for Retinal Circuits and Optogenetics, Centre for Integrative Neuroscience, Eberhard-Karls University Tübingen, Tübingen, Germany (T.A.M.); Department of Anatomy and Neurobiology, Dalhousie University, Halifax, Nova Scotia, Canada (G.B.A.). 4These authors contributed equally to this work. Correspondence should be addressed to B.R. (
[email protected]). Received 24 November 2008; accepted 28 July 2009; published online 6 September 2009; doi:10.1038/nn.2389
1308
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
a r t ic l e s a
b
d
Figure 1 PV-5 ganglion cells are sensitive to Approaching motion Lateral motion approaching motion. (a) Wholemount retina INL from a PvalbCre × Thy1Stp-EYFP mouse. Bright spots, ganglion cell somas; bright radial lines, PV-5 ChAT Model PV-5 400 µm s–1 axons extending toward the optic disc (central ChAT dark spot). (b) Side (top panel) and top (bottom GCL 100 Hz panel) projections of a confocal image stack 1s of a PV-5 cell, acquired after fixation. Green, neurobiotin-filled cell; red, ChAT; blue, DAPI; 200 µm s–1 INL, inner nuclear layer; GCL, ganglion cell layer. (c,d) Spiking responses of a PV-5 cell to 500 µm 50 µm different motion stimuli. A 60-µm black bar 100 µm s–1 was turned on within a 400-µm-diameter mask in the center of the dendritic field (“Stimulus onset”). After a 2-s pause, the left and the right Stimulus Motion 50 µm s–1 edges of the bar began to move at velocities onset drawn at random from a number of values 400 (“Motion”). (c) Overview of the responses over Approaching Lateral Motion Motion the parameter space. Radii of dotted circles and 200 Stimulus onset Stimulus onset gray disks are proportional to the average spike count after the onset and during the motion of 0 Spike count Peak spiking rate the bar, respectively. Scale radius is shown at lower right. The quadrant that corresponds to 1 Approaching –200 30 0 motion (n = 11) approaching motion is shaded in light gray. Lateral Receding Spike count (d) Responses to expanding (“Approaching –400 Lateral motion (n = 7) motion”) and drifting (“Lateral motion”) –400 –200 0 200 400 0 –1 Velocity of left edge (µm s ) bars for different edge velocities. Ticks, –1 50 100 200 400 µm s individual spikes; black traces, average spiking Edge velocity (µm s–1) rates; dotted gray traces, model outputs (Supplementary Fig. 5). Black arrowheads, times at which the moving edges exit the dendritic field of the cell. (e) Spike count and peak spiking rate during motion as a function of edge velocity, for expanding (“Approaching motion”) and drifting (“Lateral motion”) bars. Error bars, s.e.m. Velocity of right edge (µm s–1)
Normalized response
e
during motion was roughly invariant (Fig. 1e). In summary, within the array of stimuli we used, PV-5 cells favored approaching motion over lateral and shrinking motion. An approaching dark object results in a retinal image with both overall dimming and moving contrast edges. To dissect responses of PV-5 cells to these two components, we designed stimuli that distinguished overall dimming from the motion of contrast edges. First, we presented PV-5 cells with spatially uniform dimming stimuli with light levels matched to those of the approaching motion stimuli. As expected from an OFF cell, these purely dimming stimuli elicited significant responses in PV-5 cells (data not shown; n = 4). Next, we presented PV-5 cells with stimuli in which a bar expanded concomitantly with an overall brightening that maintained the total light intensity constant (Fig. 2a). Notably, PV-5 cells also responded significantly to this stimulus (Fig. 2b, n = 6). In summary, PV-5 cells responded not only to dimming, but also to the rapid expansion of a negativecontrast boundary within their dendritic field, even if it occurred in the absence of any dimming (Fig. 2).
a
b
Stimulus
Space
Dendritic field width
Mean intensity
© 2009 Nature America, Inc. All rights reserved.
c
1
2
2
3
Experimental response traces 1 100 Hz n*** =4 1s 0
200 µm s–1
n*** =3
100 µm s–1
n*** =6
4 5
3
Time 1
400 µm s–1
4
5
50 µm s–1
Approach sensitivity relies upon a composite receptive field We hypothesized that inhibitory activity was responsible for suppressing the response of PV-5 cells to non-approaching motion. To test this hypothesis, we recorded excitatory and inhibitory inputs during stimulation with different spatiotemporal patterns. When stimulated for 2 s with a dark spot (400-µm diameter, n = 30) against a gray background, PV-5 cells responded with a burst of spikes at the onset of this OFF (dimming) stimulus (Fig. 3a, top trace). Excitatory synaptic currents (‘excitation’; see Online Methods), which follow glutamatergic inputs from bipolar cells16–18, were also activated only at the onset of the stimulus (Fig. 3a, bottom trace; n = 38). By contrast, inhibitory synaptic currents (‘inhibition’; see Online Methods), which follow inputs from amacrine cells16–18, were evoked at the offset of the stimulus, when the dark spot was removed: effectively an ON (brightening) stimulus (Fig. 3a, middle trace; n = 38). Hence, PV-5 cells were excited at light decrements and inhibited at light increments. In the case of this simple stimulus, excitation and inhibition did not interact because they occurred at different times.
c Model response traces 1s
n ** =4 100 200 ms
Motion Stimulus onset
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
Motion Stimulus onset
Figure 2 PV-5 ganglion cells respond to approaching motion even in the absence of dimming. (a) Illustration of the constant-luminance stimulus, which consisted of an expanding dark bar together with progressive overall brightening that maintained the overall light intensity constant. (b) PV-5 cell spiking rate in response to the stimulus illustrated in a. Insets, cumulative distributions of the interspike intervals during baseline activity (dashed lines) and during the first 500 ms of expansion (solid lines). **P < 0.01, ***P < 0.001 (one-sided Kolmogorov-Smirnov test; n, number of stimulus repetitions). (c) Model output in response to the stimulus illustrated in a.
1309
a r t ic l e s a
b
d
pr es
Normalized spike count F en ir ta st Control t P ion APB re ola ve rit rs y al APB
Figure 3 Response of PV-5 ganglion cells to NS *** lateral motion is suppressed by an ON inhibitory Stimulus *** 1 signal. (a) Spiking, excitation and inhibition 0 1 2 responses of a PV-5 ganglion cell to a 400-µmSpiking 0 diameter black disk presented for 2 s (black bar at bottom). (b–d) Spiking responses to a polarity-reversing black-and-gray checkerboard Inhibition (checker size, 50 µm). (b) Stimulus frames. Control (c) Top: responses in control conditions; Excitation bottom: responses under 80 µM APB. Ticks, 500 pA 1s individual spikes; black traces, average spiking 200 Hz rate; bottom row, stimulus frame sequence. APB 1s (Thin gray lines between stimulus frames do not represent a stimulus; they are graphical 0 1 2 1 2 1 2 0 separators only.) (d) Comparison of normalized Motion onset spike count after the first presentation of the Bar width *** checkerboard and after polarity reversals, in NS 50 µm *** 100 µm control conditions and with APB. Error bars, 150 s.e.m.; ***P < 0.001; NS, not significant Contrast 100 50 Hz PV-5 control 174% (t-test). (e) Top: spiking rate responses of a 0.5 s 86% PV-5 APB PV-5 cell to the onset of lateral motion in 50 Speed control conditions and with APB (80 µM). 0 250 µm s–1 Bottom: spiking rate response of a PV-1 cell PV-5 PV-5 PV-1 500 µm s–1 to the onset of lateral motion under control Control APB Control PV-1 control 750 µm s–1 conditions. (f) Mean spiking rate responses 1,000 µm s–1 during the first 200 ms after onset of lateral motion, in PV-5 cells under control conditions and with APB (80 µM) and in PV-1 cells under control conditions. Dashed lines connect data points originating from the same cell stimulated with the same stimulus. Thick horizontal lines, mean responses of the cell populations. Thin vertical lines, s.e.m.; ***P < 0.001; NS, not significant (Mann-Whitney rank test).
c
Mean spike rate (Hz)
f
checkerboard resulted in a net darkening, and consequently the ganglion cell fired a burst of spikes (Fig. 3c). At each reversal, half of the checkers became brighter and the other half became darker, such that there was no change in the mean light intensity across the
The situation is different when OFF and ON stimuli appear simultaneously in different regions within the dendritic field. We presented PV-5 cells with a polarity-reversing checkerboard against a gray background (Fig. 3b). The initial presentation of the black
a
Control
b
CPP/NBQX
Control CPP/NBQX 20 mV
0.25 s
Max 20% of max
0 mV –20 mV
Duration
–40 mV
d
–60 mV 500 pA 1s
Approaching motion
Control APB
Spike rate
Duration (ms)
400
10 10 1,000 Control
0 0
400 800 1,200 Control
e
Control CPP/NBQX + Strychnine
100 pA 0.5 s
50 Hz
Excitation
h
NS NS
Time to peak (ms)
g
Lateral motion
Amplitude (pA) 1,000
Control CPP/NBQX + APB
–80 mV
f
CPP/NBQX
c
20 mV
Amplitude
© 2009 Nature America, Inc. All rights reserved.
e
*** 200
n=3
n=3
n=3
n=3
100 0
Inhibition Excitation Control Control CPP/ Control – NBQX CPP/ NBQX
i
Inhibition Control CPP/NBQX Control – CPP/NBQX
100 pA Inhibition
50 pA 0
1310
200 Time (ms)
400
0
200 Time (ms)
400
0
100 200
300
400 Time (ms)
Figure 4 PV-5 ganglion cells receive a rapid inhibitory input required to suppress responses to lateral motion. (a) Synaptic currents at various holding potentials, in response to a 400-µm black disk presented for 2 s. Left, control; right, CPP/NBQX. (b) Inhibitory currents from a at a finer temporal resolution. Peak amplitudes are normalized. (c) Left, trace of inhibitory current, displayed to illustrate the definitions of ‘amplitude’ and ‘duration’. Middle and right, amplitude and duration of the inhibitory currents in control conditions and with CPP/NBQX. (d,e) Effects of APB (d, 80 µM) and strychnine (e) on the rapid inhibitory current isolated by CPP/NBQX. (f,g) Responses to lateral (f) and to approaching (g) motion in control conditions and with APB (10 µM). The stimulus, presented within a 300-µm-diameter mask, was a blackand-gray grating (100-µm bar width) with edges moving at 500 µm s−1. Dashed box in f, duration of the burst of spikes; arrows, motion onset. (h) Time to peak of the different components of the excitatory and inhibitory inputs in PV-5 cells presented with the same stimulus as in f. ‘Time to peak’ is the delay, counting from stimulus onset, before the response reaches 67% of its peak. ‘Control – CPP/NBQX’ is the CPP/NBQXblocked component. Error bars, s.e.m.; ***P < 0.001; NS, P ≥ 0.05. (i) Inhibitory input in response to the same stimulus as in f, showing the relative timing of the CPP/NBQXresistant component (light gray) and the CPP/ NBQX-blocked component (dark gray). Dashed box as in f.
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
a r t ic l e s a
b
PV-6 Figure 5 PV-6 OFF ganglion cells respond to PV-6 control Spiking PV-6 CPP/NBQX lateral motion. (a) Spiking activity, as well PV-5 CPP/NBQX as excitatory and inhibitory inputs, in a PV-6 20 mV ganglion cell, in response to a 400-µm0.25 s diameter black disk presented for 2 s. (b) Inhibitory currents in PV-6 cells, in control PV-6 PV-5 PV-6 PV-5 PV-6 Inhibition conditions and with CPP/NBQX. For comparison, *** *** 1.0 the CPP/NBQX-resistant component of the n=3 n=3 inhibitory input to PV-5 cells is reproduced from 10 Hz 200 1s Figure 4b. Peak amplitudes are normalized to n=3 0.5 100 emphasize the relative timing of the responses. Excitation n=3 200 pA (c–e) Responses to a black-and-gray grating 0 (100-µm bar width), with edges moving at 1s Approaching motion Inh Exc 0 500 µm s−1, within a 300-µm-diameter mask. Lateral motion (c) Time to peak of excitatory and inhibitory currents in PV-6 cells during lateral motion. (d) Spiking rates in PV-5 and PV-6 cells, in response to lateral motion and to approaching motion. Arrows, motion onset. (e) Ratio of peak spiking rate responses to lateral motion and approaching motion, in PV-5 and in PV-6 cells. Error bars, s.e.m.; ***P < 0.001.
c
e
PV-5 cell dendritic field. These reversals elicited very little spiking in PV-5 cells under control conditions (Fig. 3c, top panel; Fig. 3d; n = 5). By contrast, when the ON pathway was blocked with 80 µM APB (L-(+)-2-amino-4-phosphonobutyric acid)19, PV-5 cells responded at each polarity reversal (Fig. 3c, bottom panel; Fig. 3d). Hence, in control conditions ON inhibition suppressed the spiking response (see also Supplementary Fig. 2). ON inhibition influenced not only the spiking activity but also the magnitude of the excitatory input (Supplementary Fig. 2b, n = 5). Our interpretation of this finding is that ON inhibition acts both on PV-5 cells directly and on OFF bipolar cell terminals. These results suggest a composite receptive field, made up of many ‘push-pull’ (OFF excitation, ON inhibition) subunits20,21 that can interact when their outputs are pooled by the ganglion cell. The composite nature of the receptive field could explain the sensitivity of PV-5 cells to approaching stimuli. Approaching motion of a dark object would result in an OFF stimulus and, consequently, excitation without inhibition. By comparison, receding motion would result in an ON stimulus and, consequently, inhibition without excitation. Lateral motion of a dark object would result in a leading OFF stimulus and a trailing ON stimulus and, consequently, both excitation and inhibition, which interact because they occur concomitantly. We used drifting gratings, displayed within a mask of 300-µm dia meter, to investigate the putative inhibitory suppression of responses to lateral motion (Fig. 3e,f). When we blocked ON inhibition with APB, a vigorous spiking response appeared at the onset of motion that was not observed in control conditions (Fig. 3e, top panel, n = 5). This result indicates that ON inhibition serves to suppress responses of PV-5 cells to lateral motion. The suppression was robust: we observed it for a variety of spatial frequencies, grating contrasts and drifting
velocities (Fig. 3f). It was instructive to compare the behavior of PV-5 cells to that of the similarly large (360 ± 70 µm, n = 6) and transient PV-1 cells. PV-1 cells stratified at −40 ± 4% (Supplementary Fig. 1b) and received fast ON excitation and slow ON inhibition (data not shown). Hence, their receptive field was not composed of pushpull subunits. PV-1 cells did respond to the onset of lateral motion (Fig. 3e bottom panel and Fig. 3f, n = 3). This comparison supports the claim that the presence of push-pull subunits in the composite receptive field of PV-5 cells is responsible for the suppression of responses to undesired stimuli. Rapid inhibition is involved in approach sensitivity A key aspect of the approach-sensitivity mechanism not discussed thus far pertains to dynamics: suppression of the response to nonapproaching stimuli occurs only if ON inhibition acts sufficiently rapidly to cancel OFF excitation. We analyzed the temporal structure of synaptic inputs under the effect of blockers of ionotropic glutamate receptors (CPP/NBQX: 10 µM (±)-3-(2-Carboxypiperazin-4-yl) propyl-1-phosphonic acid, blocking NMDA receptors; plus 10 µM 6-nitro-2,3-dioxo-1,4-dihydrobenzo[f]quinoxaline-7-sulfonamide, blocking AMPA and kainate receptors). CPP/NBQX abolished the excitatory input (Fig. 4a; n = 29). However, a rapid inhibitory signal remained (Fig. 4a–c; n = 24 of 29 cells), with comparable amplitude to that recorded in the control experiment (Fig. 4c: 84% ± 47% of control; n = 24, P = 0.38, paired t-test). This CPP/NBQX-resistant, rapid inhibition was blocked by APB (10 µM, n = 4, or 80 µM, n = 4; Fig. 4d) and also by strychnine (10 µM, n = 3; Fig. 4e), a glycinergic receptor antagonist. Curare (50 µM, n = 3), a nicotinic acetylcholine receptor antagonist, and SR-95531 (5 µM; n = 3), a GABAA receptor antagonist, had no effect on the rapid inhibition (data not shown).
Figure 6 The rapid inhibitory pathway is mediated by an electrical synapse. Unless noted, all traces –/– Control CPP/NBQX Cx36 1.0 on this figure are from PV-5 cells in Cx36−/− Wild type *** *** 20 mV background. (a) Synaptic currents at various n=3 n=6 0 mV 200 holding potentials in response to a 400-µm black –20 mV 200 0.5 disk presented for 2 s in control conditions and –40 mV * n=3 * 100 –60 mV * 100 with CPP/NBQX. (b) Effect of CPP/NBQX on the –80 mV n = 6 magnitude of the inhibition. (c) Time to peak of 100 pA 0 0 0 1s inhibitory (Inh) and excitatory (Exc) currents. Inh Exc 250 500 750 1,000 –1 Edge velocitiy (mm s ) Stimulus, presented within a 300-µm-diameter mask, was a black-and-gray grating (100-µm bar width), with edges moving at 500 µm s−1. (d) Ratio of peak spiking rate responses to lateral motion and approaching motion, for different edge velocities, in Cx36−/− and wild-type mice. The P-value at 750 µm s−1 was 0.0502, only slightly above significance level 0.05. Error bars, s.e.m.; *P < 0.05, ***P < 0.001. tro l
P BQ P/ X
C
N
on
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
d
Lateral / approaching
c
Time to peak (ms)
b
Peak current (pA)
a
C
© 2009 Nature America, Inc. All rights reserved.
Time to peak (ms)
Lateral / approaching
d
1311
a r t ic l e s Electrical stimulation of amacrine cell
b
–60
Amacrine
Amacrine
n=3
0 mV 20 pA 0.5 s Control Strychnine
PV-5
Peak current (pA)
a
PV-5
c
d
ON
Live-2P
40 PV-5
30 20 10 0
n=3 Control Strychnine
e
OFF
Live-2P
Live-2P
f
***
g
h
INL
PV-5
The pharmacological actions of all blockers used in our experiments were reversible (data not shown). These experiments suggest that a three-synapse pathway (Supplementary Fig. 3a) carries the rapid inhibitory component: cones to ON bipolar cells, ON bipolar cells to amacrine cells through a conduit unimpaired by CPP/NBQX and amacrine cells to PV-5 ganglion cells through glycine receptors. By contrast, the excitatory pathway seems to be conventional: cones to OFF bipolar cells, and OFF bipolar cells to the PV-5 cell through ionotropic glutamate receptors. Inhibition must act rapidly enough to prevent spiking in PV-5 cells presented with lateral motion. We compared the timing of spiking responses, excitatory inputs, and inhibitory inputs of PV-5 cells stimulated with moving black-and-gray gratings (100-µm-wide bars at a speed of 500 µm s−1), both in control conditions and when the ON pathway was blocked (Fig. 4f,g, 10 µM APB; n = 3). We found, first, that during lateral motion the excitatory and the inhibitory
currents occurred with similar time courses in control conditions (Fig. 4f,h). Second, when inhibition was blocked with APB, a burst of spikes appeared at the onset of lateral motion. Notably, the timing of this burst of spikes was comparable to that of the rapid component of inhibition, which was blocked by APB (Fig. 4f). We also observed that the presence of APB enhanced the excitatory input (Fig. 4f), an effect already noted in the context of the checkerboard stimuli (Supplementary Fig. 2b). By contrast, during approaching motion, no inhibition was detected in control conditions and APB had no effect on the spiking and excitatory responses (Fig. 4g, 10 µM APB; P = 0.65 for spiking and P = 0.7 for excitation, t-test; n = 3). These results are consistent with the notion that rapid, ON-inhibitory signal blocks spiking at the onset of lateral motion. PV-5 cells receive both CPP/NBQX-blocked and CPP/NBQXresistant inhibitory inputs, but the CPP/NBQX-resistant component occurs over a significantly shorter time scale than the blocked component (Fig. 4h,i). Within the time window defined by the burst of spikes that follows the onset of lateral motion (recorded in the presence of APB), 80 ± 6% of the inhibitory signal (n = 4, quantified as the integral of the inhibitory current over time) was CPP/NBQXresistant. This finding suggests that the CPP/NBQX-resistant inhibitory pathway dominates in the suppression of PV-5 cell responses to the onset of lateral motion. The relevance of rapid inhibition to the PV-5 cell physiology was also apparent when we compared this cell to another EYFP-labeled OFF cell in the PvalbCre × Thy1Stp-EYFP mouse line, which we called the PV-6 cell. Similarly to the PV-5 cell, the PV-6 cell had a wide dendritic field (diameter 400 ± 40 µm, stratification at 125 ± 8%, n = 6) and received OFF excitation and ON inhibition (Fig. 5a, n = 6). However, the inhibitory input to PV-6 cells was slower than that observed for PV-5 cells, and it was entirely blocked by CPP/NBQX (Fig. 5b; n = 3). The PV-6 cell circuit was therefore devoid of a rapid inhibitory pathway. Excitation was faster than inhibition in PV-6 cells (Fig. 5c) and, unlike PV-5 cells, these cells responded to both approaching and lateral motion (Fig. 5d,e). The contrast between the physiology of PV-5 cells and PV-6 cells further supports the relevance of the CPP/NBQX-resistant inhibitory component in the mechanism of approach sensitivity. In the rapid inhibitory pathway, the excitatory synapse between the ON cone bipolar cell and the amacrine cell is neither glutamatergic
1312
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
GCL
Fixed
i 3 2
Fixed
j
Level 1 Rod bp ON cone bp Neurobiotin AII AC
Dab1
Merged
OPL
INL
© 2009 Nature America, Inc. All rights reserved.
Fixed
Figure 7 PV-5 cells receive an inhibitory input from AII amacrine cells. (a) Schematic of a double-patch experiment (side view). The experiment was performed on wholemount retina. (b) Left, inhibitory currents in a PV-5 cell (black, control; gray, strychnine) evoked by stimulating the double-patched amacrine cell. Right, peak inhibitory currents in PV-5 cells after amacrine cell stimulation. Error bars, s.e.m.; ***P < 0.001. (c–e) Two-photon (2P) images of a presynaptic amacrine cell in a double patch. Red arrows, dendrite of the connected PV-5 cell. (Images overexposed to show dim signal are in Supplementary Fig. 4.) (c) Side projection. (d) Top projection of proximal IPL. (e) Top projection of distal IPL. (f) Maximum intensity projection of a confocal stack of a recorded amacrine–PV-5 cell pair. (g) Confocal section at higher magnification. Magenta, pixels with neurobiotin staining only; green, pixels with neurobiotin and EYFP staining. (h) Magnified side view from f. Cyan, DAPI. (i) Schematic representation of cell bodies in the inner nuclear layer (INL). Rod bp, rod bipolar cells; ON cone bp, ON cone bipolar cell; AII AC, AII amacrine cell. Levels refer to the depths of confocal sections marked in j–l. (j) A recorded and neurobiotin-filled (magenta) amacrine cell (level 1) in a double-patch experiment (arrow). Dab1 staining, cyan. (k) Neurobiotin-labeled cells (magenta), in level 2, are positive for Gγ13 (cyan) and negative for PKCα (yellow). (l) No neurobiotin-labeled cells (magenta) are present in level 3, where most cells are positive for Gγ13 (cyan) and PKCα (yellow). The secondary antibody to mouse also labels blood vessels (asterisks). Scale bars, 10 µm.
1 IPL
k
Level 2
Neurobiotin
l
Gγ13
PKCα
Gγ13
PKCα
Merged
Level 3
Neurobiotin
Merged
* *
a r t ic l e s a
b
c
© 2009 Nature America, Inc. All rights reserved.
nor cholinergic, but it may be electrical. Electrical synapses consist of connexin proteins, of which connexin36 (Cx36) is one of the most abundant in the mouse retina22–24. In mice lacking Cx36 (Cx36−/−) CPP/NBQX blocked the excitatory and inhibitory inputs to PV-5 cells (Fig. 6a,b). Inhibition was slower than excitation (Fig. 6c), and the ability of the PV-5 cell to discriminate between approaching and lateral motion in the Cx36−/− background was inferior to that in wild type for three of four velocities tested (Fig. 6d, n = 3). These observations, together with the pharmacological observations described above, indicate that rapid inhibition transits between ON cone bipolar cells and amacrine cells through an electrical synapse (Supplementary Fig. 3a) that requires functional Cx36. The approach-sensitive neural circuit is multifunctional The circuit element responsible for the rapid inhibitory input to PV-5 cells—namely, an ON cone bipolar cell connected by a Cx36-containing electrical synapse to a glycinergic amacrine cell—is reminiscent of the rod pathway circuitry22–25. During scotopic (night-time) vision, the glycinergic AII amacrine cells26, after activation by rod bipolar cells, transmit signals through Cx36-containing electrical synapses to ON cone bipolar cells22–24,27. An attractive hypothesis is that the rapid inhibitory pathway afferent to PV-5 cells makes use of the same electrical synapse as the one associated with the rod circuit, but with a reversed information flow15,28–31 (Supplementary Fig. 3b). As a direct test of this hypothesis, we performed dual-patch recordings of amacrine cells and PV-5 ganglion cells in flatmount retinas (Fig. 7a). We targeted amacrine cells in the most proximal sublayer of the inner nuclear layer (INL), where we estimated that roughly onethird of the cells were AII amacrine cells. This estimation was based on antibody staining for Dab1, a selective AII amacrine cell marker32. We recorded from 16 amacrine–PV-5 cell pairs. In six pairs, depolarizing voltage steps applied to the amacrine cell elicited inhibitory currents in the PV-5 ganglion cell (Fig. 7b). The rapid component of these inhibitory currents survived the application of CPP/NBQX (n = 3), reflecting direct input from the stimulated cell (data not shown). Strychnine abolished the inhibitory currents in PV-5 cells that were otherwise elicited by amacrine cell stimulation in control conditions (Fig. 7b; n = 3). From this observation, we inferred that the inhibitory transmission used glycine receptors. To determine the identity of the six amacrine cells that evoked inhibitory currents in PV-5 cells when they were electrically stimulated, we visualized the morphology of three of these cells under the two-photon microscope after the recordings (Fig. 7c–e and Supplementary Fig. 4), and all six cells by confocal fluorescence microscopy post hoc in fixed retinas (Fig. 7f–h). The narrow-field nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
Peak current (pA)
Velocity of right edge (µm s–1)
Velocity of right edge (µm s–1)
Figure 8 The functional properties of AII Stimulus onset Motion Stimulus onset Motion amacrine cells are consistent with the rapid 400 400 inhibitory signal in PV-5 ganglion cells. *** Approaching Lateral Approaching Lateral NS (a,b) Motion-response map of an AII amacrine n=4 –40 200 200 n=4 cell in control conditions (a) and with CPP/ NBQX (b). Map is analogous to that in 0 0 Figure 1c. The recorded cell was clamped to –20 Lateral Receding Lateral Receding −60 mV. The radii of the disks are proportional –200 –200 to the peak magnitudes of inward currents n=4 0 evoked by stimulus motion. The radii of Control CPP/ APB –400 –400 the dotted circles are proportional to the NBQX –400 –200 0 200 400 –400 –200 0 200 400 reduction of the excitatory currents after the –1 –1 -40 0 Velocity of left edge (µm s ) Velocity of left edge (µm s ) initial presentations of the black bar. The pA AII amacrine cell: control AII amacrine cell: CPP/NBQX quadrant that corresponds to approaching motion is shaded in light gray. (c) Average peak magnitudes of excitatory currents in AII amacrine cells, in the lateral and receding quadrants of the motion-response map, under different pharmacological conditions (CPP/NBQX and 10 µM APB). Error bars, s.e.m.; ***P < 0.001; NS, P ≥ 0.05.
(cell lateral diameter 26 ± 2 µm; n = 6), vertically oriented morpho logy, with small boutons in the ON layer and larger lobules in the OFF layer, suggested that the recorded cells were AII amacrine cells (Fig. 7c–h). For verification, we double-stained three of the six amacrine cells with anti-Dab1 (Figs. 7i,j). In all three cases, the recorded cells were positive for Dab1. Moreover, tracer-coupled cells in the vicinity of the recorded amacrine cells, in the same nuclear layer, were also Dab1 positive. This result suggested coupling between neighboring AII amacrine cells (Fig. 7j). Tracer-coupled somas were also detected in more distal nuclear layers within the INL, and, therefore, we tested in the three other pair-recorded amacrine cells (that evoked inhibition in PV-5 cells) whether the distally coupled cells might be ON cone bipolar cells. ON cone bipolar cells can be identified through labeling with an antibody recognizing the marker Gγ13 (ref. 33) and can be differentiated from rod bipolar cells because they do not stain for the marker PKCα (ref. 34). In all three samples we examined, the tracer-coupled cells in the distal INL were Gγ13-positive and PKCαnegative (Fig. 7k,l). We therefore concluded that the cells in the distal INL tracer-coupled with AII amacrine cells were indeed ON cone bipolar cells. We completed the analysis of the approach-sensitive circuit by measuring the response of AII amacrine cells presented with the same stimulus array used for probing PV-5 cells (Fig. 1c). AII amacrine cells were recorded under voltage clamp conditions (at −60 mV). After the recordings, AII amacrine cells were identified by their morphology using two-photon laser microscopy and staining with the Dab1 antibody (data not shown). As expected, light increments evoked excitatory currents in AII amacrine cells, but light decrement did not (traces not shown). The response map of AII amacrine cells (n = 4) in control conditions and under CPP/NBQX treatment showed a pattern complementary to the response map of PV-5 cells (Fig. 8a,b). Furthermore, during lateral motion and receding motion, CPP/NBQX blocked the slow component of the excitatory input but did not affect its rapid component (data not shown). APB, however, abolished the response of AII amacrine cells to both stimulus onset and lateral motion (Fig. 8c). These results indicate that AII amacrine cells provide the rapid component of the inhibition, as well as a portion, or possibly the entirety, of the slow inhibitory input to PV-5 cells. Computational model of approach sensitivity We incorporated the various elements of the proposed composite receptive field of the PV-5 cell into a computational model (Supplementary Fig. 5). The model PV-5 cell sums over a large region covered by many push-pull subunits (Supplementary Figs. 5a,b) that excite the PV-5 cell in response to local OFF inputs and inhibit it 1313
a r t ic l e s
© 2009 Nature America, Inc. All rights reserved.
in response to local ON inputs. The two processes—excitation and inhibition—occur with similar dynamics (Supplementary Fig. 5c). As a result, inhibition prevents responses to undesired stimuli (such as the laterally moving object in Supplementary Fig. 5b). As a key element, signals from subunits are rectified before being summed by the PV-5 cell (Supplementary Fig. 5c). Because of this concave nonlinearity, strong local signals are favored over weak diffuse ones. Thus, the model PV-5 cell responds to the expanding edges of an approaching object even if the visual field undergoes slow brightening so as to prevent overall dimming (such as in Fig. 2a). The computational model reproduces the data (Figs. 1d and 2c) and closely follows experimental traces for an array of input patterns and velocities. DISCUSSION Retinal analysis of motion In the retina, visual information is broken up and transmitted in parallel channels to central brain regions by different types of ganglion cells. An appreciable fraction of retinal circuitry is devoted to the analysis of different categories of motion. In addition to the approach-sensitive ganglion cell type described here, eight types of direction-selective ganglion cells (four ON-OFF types35, three ON types35 and one OFF type13) report either the direction of lateral object motion or the direction of global image drift. Still other ganglion cell types respond to differential object motion relative to global background motion36. In all three cases of motion sensitivity—direction selectivity13,35, differential-motion sensitivity36 and approach sensitivity—the ganglion cells respond most vigorously to a so-called preferred stimulus, whereas their responses to so-called null stimuli are suppressed. The preferred stimuli are object motion in a given direction, differential object-background motion and approaching motion, respectively, whereas null stimuli are motion in the opposite direction, coherent object-background motion and receding or lateral motion, respectively. But ganglion cells are broadly tuned, and ‘sensitivity’ does not mean ‘exclusivity’: motion-sensitive cells do not respond to their preferred stimulus alone. For example, an OFF directionselective, differential motion-sensitive, or approach-sensitive cell will respond vigorously to a dark flash, like any other OFF ganglion cell. The essence of motion sensitivity lies in the suppression of responses to null stimuli; that is, in what the motion-sensitive cell does not respond to. The approaching motion of an object was mimicked in our experiments by an expanding bar projected onto a fixed retina. In natural situations, an expanding retinal image can result from either object motion or observer motion. The PV-5 cell may distinguish these two situations. When an observer moves toward a static scene, one expects the retinal image to comprise many concomitantly expanding dark and bright areas (along with lateral drift). According to our model, these lead to both excitation and inhibition, which cancel each other, and, hence, we expect a weak or negligible response. Thus, PV-5 cells may be more responsive to true object motion than to apparent motion in the visual field due to observer motion.
PV-5 cells may fulfill an ‘alert function’37 by signaling approaching motion, such as that of a falcon aiming at a mouse, to higher brain centers. Looming-sensitive cells in locust brain have spatial and functional properties that are different from PV-5 cells. The receptive fields of looming-sensitive cells are larger than those of PV-5 cells8–11. Their response is also more sophisticated in that they can respond to object approach independent of the object contrast and in that they encode the time of collision (between the approaching object and the observer) in the temporal structure of the spiking rate. In PV-5 cells, we found a simple correlation between the velocity of an approaching object and the spiking rate (Fig. 1e). PV-5 cells may serve as elementary building blocks in motion-sensing streams: their outputs may be pooled by downstream detectors, such as looming-sensitive cells, that cover a wider field. Furthermore, by combining information from several approach-sensitive cells, higher-order cells may achieve more involved spatiotemporal computations. One kind of putative spatiotemporal computation that would be useful relates to the resolution of visual ambiguities. Any procedure that aims at inferring a three-dimensional trajectory from its twodimensional (retinal) projection faces ambiguity. In the case of PV-5 cells, approaching dark objects can be confused with receding bright objects (for example, a bright, narrowing gap between two large, approaching objects). Ambiguities may also arise when object motion occurs near the edge of a cell’s receptive field—a special case of the well-known ‘aperture problem’38. For example, a large object that enters the receptive field laterally may be confused with an approaching object. All these are fundamental ambiguities pertaining to the stimulus, which cannot be resolved by the PV-5 cell. Instances of the aperture problem may be solved by the use of a receptive field covering the entire visual scene. More often, the resolution of visual ambiguities necessitates the integration of information coming from several ‘elementary’ cells such as approach-sensitive ones. So, here again PV-5 cells come into the picture as candidate building blocks that can feed useful information to higher-order cells.
Scales and ambiguities in approach sensitivity We have referred to the retinal PV-5 ganglion cell as an ‘approachsensitive’ cell and thereby differentiated it from the ‘looming-sensitive’ cells observed in higher brain centers of locust8 and pigeon11. Besides their location in the visual stream, approach- and looming-sensitive cells differ by the scales of their receptive fields and in the nature of the information they encode. They may also differ in the way in which they handle visual ambiguities.
Implementation of approach sensitivity The PV-5 cell receives excitatory and inhibitory inputs from small subunits, namely bipolar (excitatory) and bipolar–amacrine (inhibitory) circuits. The excitatory subunits are of the OFF type, and the inhibitory subunits are of the ON type. This push-pull20,21 structure, taken together with the finding that excitatory and inhibitory subunit responses follow similar time courses, ensures that concomitant bright and dark inputs of comparable intensity ‘cancel’ each other and, hence, trigger no appreciable PV-5 cell response. This explains the insensitivity to lateral motion: trailing (ON) edges inhibit the excitatory effect of leading (OFF) edges. As for the sensitivity to approaching motion, the key point is that the PV-5 cell receptive field is large and ‘composite’: it sums over many ON-OFF subunits, and this integration is nonlinear. The responses of individual subunits are thresholded by a concave nonlinearity, and the outcomes of this operation are then summed linearly into a global ganglion cell input. When the PV-5 cell is presented with an approaching dark object, the moving image on the retina is that of an expanding dark area. In a sense, there is no ‘trailing’ edge, but everywhere an OFF ‘leading’ edge that stimulates excitatory subunits and, in turn, the PV-5 cell. The thresholding nonlinearity ensures that the PV-5 cell is activated even in the absence of any dimming— for example, if a compensating, uniform brightening is applied concomitantly to the approach motion (Fig. 2). At the expanding edge, excitatory OFF subunits are recruited and respond vigorously to edge motion. By contrast, the slow overall brightening of the receptive field
1314
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
© 2009 Nature America, Inc. All rights reserved.
a r t ic l e s elicits moderate responses diffusely throughout the ON inhibitory subunits. In a linear system, this global inhibition would cancel out the local excitation. In the nonlinear system, localized but strong inputs have greater impact than weak inputs summed over a wide region. PV-5 cells make use of rapid inhibition mediated by an electrical synapse. Given the overall slow time courses of retinal responses, it is surprising that inhibition mediated by chemical synapses only is not sufficiently fast to cancel excitation. One explanation for the inhibitory delay may be the rapid inhibition of narrow-field amacrine cells by wide-field amacrine cells. In salamander retina, this has been shown to result in a ~100-ms time difference between excitatory and inhibitory inputs39. Indeed, in all inhibitory pathways examined in our work, which use only chemical synapses between bipolar and amacrine cells—namely, the Cx36−/− PV-5 cell pathway, the CPP/NBQX-sensitive pathway of the PV-5 cell, and the PV-6 cell pathway—the inhibitory inputs to the ganglion cell were delayed by ~100 ms with respect to the excitatory ones. By contrast, AII amacrine cells do not seem to receive inhibitory inputs from rapid widefield amacrine cells40, and, what is more, they receive rapid excitatory inputs through their electrical synapses with bipolar cells. As such, AII amacrine cells combine several properties favorable to their role as part of a spatially narrow, inhibitory pathway used to convey rapid, transient information. Finally, we suggested the presence of both presynaptic inhibition (acting on OFF bipolar terminals) and postsynaptic inhibition (acting on PV-5 cells) in the PV-5 cell circuit. Postsynaptic inhibition is necessary for approach sensitivity for two reasons: first, because of the considerable scale difference between the size of the subunits— bipolar cells and the AII amacrine cell on the one hand and the PV-5 cell on the other hand—and, second, because of the thresholding that occurs in the bipolar cell–to–ganglion cell information transfer. If inhibition acted only presynaptically, then an inhibitory signal that occurs simultaneously to but in a different location from an excitatory signal would have no effect on the behavior of the PV-5 cell. For example, inhibition caused by the trailing edge of a laterally moving dark object and excitation caused by its leading edge would not couple in the PV-5 ganglion cell. The mechanism of approach sensitivity thus highlights different computational roles for presynaptic inhibition and postsynaptic inhibition.
Methods Methods and any associated references are available in the online version of the paper at http://www.nature.com/natureneuroscience/. Note: Supplementary information is available on the Nature Neuroscience website. Acknowledgments We are grateful to S. Arber (Friedrich Miescher Institute), D. Paul (Harvard Medical School) and J. Sanes (Harvard University) for providing mouse lines and Robert Margolskee (Mount Sinai School of Medicine) for providing the Gγ13 antibody. We are grateful for the technical assistance of S. Djaffer, B. Gross Scherf and Y. Shimada. We thank members of the Roska lab, P. Lagali, P. Caroni, R. Friedrich and A. Lüthi for comments on the manuscript. The study was supported by Friedrich Miescher Institute funds, a US Office of Naval Research Naval International Cooperative Opportunities in Science and Technology program grant, a Marie Curie Excellence Grant, a Human Frontier Science Program Young Investigator grant, a National Centers of Competence in Research in Genetics grant and a European Union HEALTH-F2-223156 grant to B.R., a Marie Curie Postdoctoral Fellowship to T.A.M., the Centre National de la Recherche Scientifique through the Unité Mixte de Recherche 8550 to R.A.d.S. AUTHOR CONTRIBUTIONS T.A.M. performed electrophysiological experiments, designed experiments and model, and wrote manuscript; R.A.S. designed experiments and model and wrote manuscript; S.S. performed immunohistochemistry; T.J.V. performed electrophysiological experiments, G.B.A. performed and designed electrophysiological experiments; and B.R. designed experiments and model and wrote manuscript. Published online at http://www.nature.com/natureneuroscience/. Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/.
Multifunctionality in small neural circuits A key property of the approach-sensitive circuit is the rapid inhibition mediated by the AII amacrine cell: it arrives in time to cancel excitation. In this scenario, the AII amacrine cell fulfills a function very different from the one it is well known for—namely, amplifying rod signals by conveying them to ON and OFF cone pathways 41. In the present work, we suggest that the AII amacrine cell is also relevant to photopic (daytime) vision, a conclusion that supports recent results in guinea pig15 and mouse42. Our direct demonstration of AII amacrine cell to PV-5 cell connectivity with double-patch experiments and, hence, of the involvement of the AII amacrine cell in the approach-sensitive circuit establishes a functional role of AII amacrine cells in photopic conditions. Notably, neural signals flow along the same circuit module—ON cone bipolar cell through an electrical synapse to AII amacrine cell—under photopic and scotopic conditions, but the direction of the flow is reversed in photopic conditions as compared to scotopic conditions (Supplementary Fig. 3b). It thus seems that the nervous system can use the same circuit for entirely different functional purposes under different physiological conditions—an illustration of the efficiency with which biological function can be packed into neural circuits.
1. Schiff, W., Caviness, J.A. & Gibson, J.J. Persistent fear responses in rhesus monkeys to the optical stimulus of “looming”. Science 136, 982–983 (1962). 2. King, S.M. & Cowey, A. Defensive responses to looming visual stimuli in monkeys with unilateral striate cortex ablation. Neuropsychologia 30, 1017–1024 (1992). 3. Waldeck, R.F. & Gruberg, E.R. Studies on the optic chiasm of the leopard frog. I. Selective loss of visually elicited avoidance behavior after optic chiasm hemisection. Brain Behav. Evol. 46, 84–94 (1995). 4. King, J.G. Jr., Lettvin, J.Y. & Gruberg, E.D. Selective, unilateral, reversible loss of behavioral responses to looming stimuli after injection of tetrodotoxin of cadmium chloride into the frog optic nerve. Brain Res. 841, 20–26 (1999). 5. Ishikane, H., Gangi, M., Honda, S. & Tachibana, M. Synchronized retinal oscillations encode essential information for escape behavior in frogs. Nat. Neurosci. 8, 1087–1095 (2005). 6. Ball, W. & Tronick, E. Infant responses to impending collision: optical and real. Science 171, 818–820 (1971). 7. King, S.M., Dykeman, C., Redgrave, P. & Dean, P. Use of a distracting task to obtain defensive head movements to looming visual stimuli by human adults in a laboratory setting. Perception 21, 245–259 (1992). 8. Hatsopoulos, N., Gabbiani, F. & Laurent, G. Elementary computation of object approach by wide-field visual neuron. Science 270, 1000–1003 (1995). 9. Gabbiani, F., Krapp, H.G. & Laurent, G. Computation of object approach by a widefield, motion-sensitive neuron. J. Neurosci. 19, 1122–1141 (1999). 10. Gabbiani, F., Cohen, I. & Laurent, G. Time-dependent activation of feed-forward inhibition in a looming-sensitive neuron. J. Neurophysiol. 94, 2150–2161 (2005). 11. Sun, H. & Frost, B.J. Computation of different optical variables of looming objects in pigeon nucleus rotundus neurons. Nat. Neurosci. 1, 296–303 (1998). 12. Huberman, A.D. et al. Architecture and activity-mediated refinement of axonal projections from a mosaic of genetically identified retinal ganglion cells. Neuron 59, 425–438 (2008). 13. Kim, I.J., Zhang, Y., Yamagata, M., Meister, M. & Sanes, J.R. Molecular identification of a retinal cell type that responds to upward motion. Nature 452, 478–482 (2008). 14. Haverkamp, S. & Wässle, H. Immunocytochemical analysis of the mouse retina. J. Comp. Neurol. 424, 1–23 (2000). 15. Manookin, M.B., Beaudoin, D.L., Ernst, Z.R., Flagel, L.J. & Demb, J.B. Disinhibition combines with excitation to extend the operating range of the OFF visual pathway in daylight. J. Neurosci. 28, 4136–4150 (2008). 16. Roska, B. & Werblin, F. Vertical interactions across ten parallel, stacked representations in the mammalian retina. Nature 410, 583–587 (2001). 17. Fried, S.I., Münch, T.A. & Werblin, F.S. Mechanisms and circuitry underlying directional selectivity in the retina. Nature 420, 411–414 (2002). 18. Roska, B., Molnar, A. & Werblin, F.S. Parallel processing in retinal ganglion cells: how integration of space-time patterns of excitation and inhibition form the spiking output. J. Neurophysiol. 95, 3810–3822 (2006).
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1315
© 2009 Nature America, Inc. All rights reserved.
a r t ic l e s 19. Slaughter, M.M. & Miller, R.F. 2-Amino-4-phosphonobutyric acid: a new pharmacological tool for retina research. Science 211, 182–185 (1981). 20. Belgum, J.H., Dvorak, D.R., McReynolds, J.S. & Miyachi, E. Push-pull effect of surround illumination on excitatory and inhibitory inputs to mudpuppy retinal ganglion cells. J. Physiol. (Lond.) 388, 233–243 (1987). 21. McGuire, B.A., Stevens, J.K. & Sterling, P. Microcircuitry of beta ganglion cells in cat retina. J. Neurosci. 6, 907–918 (1986). 22. Mills, S.L., O’Brien, J.J., Li, W., O’Brien, J. & Massey, S.C. Rod pathways in the mammalian retina use connexin 36. J. Comp. Neurol. 436, 336–350 (2001). 23. Feigenspan, A., Teubner, B., Willecke, K. & Weiler, R. Expression of neuronal connexin36 in AII amacrine cells of the mammalian retina. J. Neurosci. 21, 230–239 (2001). 24. Massey, S.C. et al. Multiple neuronal connexins in the mammalian retina. Cell Commun. Adhes. 10, 425–430 (2003). 25. Bloomfield, S.A. & Dacheux, R.F. Rod vision: pathways and processing in the mammalian retina. Prog. Retin. Eye Res. 20, 351–384 (2001). 26. Pourcho, R.G. & Goebel, D.J. A combined Golgi and autoradiographic study of (3H)glycine-accumulating amacrine cells in the cat retina. J. Comp. Neurol. 233, 473–480 (1985). 27. Veruki, M.L. & Hartveit, E. Electrical synapses mediate signal transmission in the rod pathway of the mammalian retina. J. Neurosci. 22, 10558–10566 (2002). 28. Geraghty, R.J., Krummenacher, C., Cohen, G.H., Eisenberg, R.J. & Spear, P.G. Entry of alphaherpesviruses mediated by poliovirus receptor-related protein 1 and poliovirus receptor. Science 280, 1618–1620 (1998). 29. Cohen, E.D. & Miller, R.F. The network-selective actions of quinoxalines on the neurocircuitry operations of the rabbit retina. Brain Res. 831, 206–228 (1999). 30. Murphy, G.J. & Rieke, F. Signals and noise in an inhibitory interneuron diverge to control activity in nearby retinal ganglion cells. Nat. Neurosci. 11, 318–326 (2008).
31. Xin, D. & Bloomfield, S.A. Comparison of the responses of AII amacrine cells in the dark- and light-adapted rabbit retina. Vis. Neurosci. 16, 653–665 (1999). 32. Rice, D.S. & Curran, T. Disabled-1 is expressed in type AII amacrine cells in the mouse retina. J. Comp. Neurol. 424, 327–338 (2000). 33. Huang, L. et al. G protein subunit G gamma 13 is coexpressed with G alpha o, G beta 3, and G beta 4 in retinal ON bipolar cells. J. Comp. Neurol. 455, 1–10 (2003). 34. Greferath, U., Grunert, U. & Wassle, H. Rod bipolar cells in the mammalian retina show protein kinase C-like immunoreactivity. J. Comp. Neurol. 301, 433–442 (1990). 35. Oyster, C.W. The analysis of image motion by the rabbit retina. J. Physiol. (Lond.) 199, 613–635 (1968). 36. Olveczky, B.P., Baccus, S.A. & Meister, M. Segregation of object and background motion in the retina. Nature 423, 401–408 (2003). 37. Franconeri, S.L. & Simons, D.J. Moving and looming stimuli capture attention. Percept. Psychophys. 65, 999–1010 (2003). 38. Bradley, D.C. & Goyal, M.S. Velocity computation in the primate visual system. Nat. Rev. Neurosci. 9, 686–695 (2008). 39. Roska, B., Nemeth, E. & Werblin, F.S. Response to change is facilitated by a threeneuron disinhibitory pathway in the tiger salamander retina. J. Neurosci. 18, 3451–3459 (1998). 40. Volgyi, B., Xin, D. & Bloomfield, S.A. Feedback inhibition in the inner plexiform layer underlies the surround-mediated responses of AII amacrine cells in the mammalian retina. J. Physiol. (Lond.) 539, 603–614 (2002). 41. Wässle, H. Parallel processing in the mammalian retina. Nat. Rev. Neurosci. 5, 747–757 (2004). 42. Pang, J.J. et al. Relative contributions of rod and cone bipolar cell inputs to AII amacrine cell light responses in the mouse retina. J. Physiol. (Lond.) 580, 397–410 (2007).
1316
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
ONLINE METHODS
© 2009 Nature America, Inc. All rights reserved.
Animals. Mice used in our experiments included PvalbCre × Thy1Stp-EYFP, and mice in which the Cx36−/− alleles were crossed into the PvalbCre × Thy1Stp-EYFP background so that PV-5 cells were labeled in a homozygous Cx36−/− background. In PvalbCre mice43, Cre recombinase is expressed under the control of the parvalbumin locus. In Thy1Stp-EYFP mice44, EYFP is expressed from a Thy1 promoter in those cells in which the transcriptional stop sequence has been removed by Cre recombinase45. Cx36−/− mice46 are homozygous knockouts for the electrical synapse protein connexin36. All animal procedures were performed in accordance with standard ethical guidelines (European Communities Guidelines on the Care and Use of Laboratory Animals, 86/609/EEC) and were approved by the Veterinary Department of the Canton of Basel-Stadt. Preparation of retinas. Light-adapted mice were killed by cervical dislocation and decapitation. Eyes were enucleated. The retinas were isolated and the pigment epithelium removed under ambient light in Ringer’s medium (in mM: 110 NaCl, 2.5 KCl, 1 CaCl2, 1.6 MgCl2, 10 D-glucose, 22 NaHCO3, bubbled with 5% CO2/95% O2, pH 7.4), mounted ganglion cell side up on a filter (MFmembrane, Millipore) that had a 2–3 mm rectangular aperture in the center, and superfused in Ringer’s medium at 35–36 °C in the microscope chamber for the duration of the experiment. In this retinal preparation, light responses could be measured for more than 8 h. Electrophysiology and pharmacology. Spike and current recordings were made with loose cell-attached patch and with whole-cell voltage clamp, respectively, using an Axon Multiclamp 700B amplifier and borosilicate glass electrodes (BF100-50-10, Sutter Instruments) pulled to 7–9 MΩ, and filled with (in mM) 112.5 CsCH3SO3, 1 MgSO4, 7.8 × 10−3 CaCl2, 0.5 BAPTA, 10 HEPES, 4 ATP-Na2, 0.5 GTP-Na3, 5 lidocaine N-ethyl bromide (QX314-Br), 7.5 neurobiotin chloride, pH 7.2. In some experiments, either Alexa Fluor 488 (amacrine cell patch) or Lucifer yellow (PV-5 cell patch) was added to the intracellular solution listed above. In the patch electrode for amacrine cells, QX314-Br was substituted by 5 CsCl, and BAPTA by 0.1 EGTA; CsCH3SO3 was adjusted to 113.7 mM. Excitatory and inhibitory synaptic currents (“excitation” and “inhibition,” respectively) were separated by voltage clamping the cell to the equilibrium potential of chloride (−60 mV) and unselective cation channels (0–20 mV), respectively. Data were analyzed offline with Mathematica (Wolfram Research). Spiking rate traces were obtained by convolving spike trains with a gaussian with s.d. σ = 30 ms, except on Figure 4f,g, where 25-ms flat binning windows were used. In pharmacological experiments, agents were bath-applied at the following concentrations: 10 µM CPP, 10 µM NBQX, 10 µM or 80 µM APB, 10 µM strychnine, 50 µM curare (tubocurarine chloride), 5 µM SR-95531. All chemicals were obtained from Sigma, with the exception of APB (Calbiochem), ATP (Labforce) and neurobiotin (Vector Laboratories). Two-photon microscopy. See Supplementary Figure 6. Labeled cells in PvalbCre × Thy1Stp-EYFP (PV) retinas. Before recording from PV retinas, we obtained an image stack with the two-photon microscope. In the retina, seven ganglion cell types were brightly labeled with EYFP. (Detailed morphological and physiological characterization of all seven classes will be reported elsewhere.) We named these cell classes PV-1 to PV-7, in the order in which their dendritic trees terminate in the IPL, with PV-1 cells arborizing closest to the ganglion cell layer and PV-7 cells closest to the INL. Note that in PV retinas not all members of the seven ganglion cell types are labeled. This is either due to the kinetics of the Cre-loxP reaction or due to the mosaic nature of Thy1 promoter activity. PV-5 cell targeting. To target PV-5 cells, we first analyzed the two-photon stack. We used two criteria to identify PV-5 cells. The first was cell body size. There were three labeled cell types with large cell bodies (around 20-µm diameter): PV-1, PV-5 and PV-6 cells. To distinguish the three cell types, we followed the course of the dendrites of each of them. We had a marker identifying strata in the IPL because ON-OFF directionally selective cells (PV-2 cells) were also labeled in the PV retina, and these cells arborize in two distinct bands in the IPL. (The same bands were labeled by the ChAT antibodies). These PV-2 bands were bright in the two-photon stacks. The dendrites of one of the cell types with large cell bodies terminated proximally to the proximal PV-2 band. These were the PV-1 cells. The
doi:10.1038/nn.2389
dendrites of the second cell type, PV-6, crossed both PV-2 bands. The dendrites of the third cell type with large cell bodies, the PV-5 cells, terminated between the two PV-2 bands. We filled all recorded cells with neurobiotin and confirmed their size and stratification relative to the ChAT strata (which were the same as the PV-2 strata) post hoc (see ‘Immunohistochemistry’ and ‘Confocal analysis’). In every case, the two-photon stratification analysis (relative to the PV-2 strata) was consistent with the confocal post hoc stratification analysis. In approximately half the double patch experiments (n = 16), we filled the PV-5 cells with Lucifer yellow in addition to neurobiotin. Finally, at the end of every experiment we confirmed the physiological identity of the recorded cell by showing that it received OFF excitation and ON inhibition and that the fast component of the inhibition was not blocked by CPP/NBQX. AII amacrine cell targeting. We randomly targeted cell bodies in the most proximal row of the INL, where AII amacrine cell bodies are located. On the basis of Dab1 staining (Dab1 is a marker for AII amacrine cells), one-third of all cell bodies in the most proximal row were AII amacrine cells. Therefore, the probability of hitting an AII amacrine cell in this row was 0.33. In the initial recordings, we determined that the cell was an AII amacrine cell only after the retina was fixed, on the basis of Dab1 and streptavidin double staining (neurobiotin was included in the pipette; see below under ‘Immunohistochemistry’ and ‘Confocal analysis’) and the characteristic morphology of AII amacrine cells, which includes larger lobules in the OFF strata, smaller terminals in the ON strata and small cell size. In subsequent sets of experiments, we also filled the cells with Alexa Fluor 488, which allowed us to visualize the randomly targeted cells under the two-photon microscope (after recordings). Statistical analysis. Population data are always reported as mean ± s.e.m. The specific statistical tests used to determine significant differences are reported with the results, and include the t-test, the Mann-Whitney rank test and the Kolmogorov-Smirnov test. Significance in all figures is denoted by * for P < 0.05, ** for P < 0.01, *** for P < 0.001 and NS for P ≥ 0.05. Light stimulation. See Supplementary Figure 6. Immunohistochemistry. After the experiments, retinas were fixed for 30 min in 4% (wt/vol) paraformaldehyde in PBS (137 mM NaCl, 2.7 mM KCl, 4.3 mM Na2HPO4, 1.47 mM KH2PO4, pH 7.4) and washed with PBS for at least 1 d at 4 °C. To aid penetration of the antibodies, we froze and thawed the retina three times after cryoprotecting it in 30% (wt/vol) sucrose. All other procedures were carried out at 22–23 °C. After washing the retina in PBS, we blocked it for 1 h in 10% (vol/vol) normal donkey serum (NDS; Chemicon), 1% (wt/vol) bovine serum albumin (BSA), and 0.5% (vol/vol) Triton X-100 in PBS. Primary antibodies were incubated for 7–14 d in 3% (vol/vol) NDS, 1% (wt/vol) BSA, 0.02% (wt/vol) sodium azide and 0.5% (vol/vol) Triton X-100 in PBS. Secondary antibodies incubated for 2 h in 3% (vol/vol) NDS, 1% (wt/vol) BSA, and 0.5% (vol/vol) Triton X-100 in PBS together with streptavidin–Alexa Fluor 555 (Molecular Probes, 1:200) and DAPI (4′,6-diamidine-2-phenylindole dihydrochloride, Roche Diagnostics, 10 µg ml−1). Streptavidin binds to neurobiotin and therefore labels neurobiotin-filled cells. DAPI binds to DNA and therefore labels nuclei. After a final wash in PBS, we embedded the retinas in ProLong Gold antifade (Molecular Probes). The following set of primary and secondary antibody combinations were used in experiments in which we recorded from only PV ganglion cells: (i) Primary: goat anti–choline acetyltransferase (ChAT, Chemicon, 1:100). Secondary: donkey anti–goat antibodies (immunoglobulin G (IgG) (H+L), Molecular Probes, 1:200, conjugated with Alexa Fluor 633). (ii) Primary: rabbit anti–green fluorescent protein (GFP; Molecular Probes, 1:200). This primary antibody binds not only to GFP but also to EYFP. We used antibody staining for EYFP because fixation decreases EYFP fluorescence. Secondary: donkey anti–rabbit antibodies (Molecular Probes, 1:200, conjugated with Alexa Fluor 488). The following set of primary and secondary antibody combinations were used to identify amacrine cells as AII amacrine cells (note that neurobiotin was always included in the recording pipettes): (i) Primary: rabbit anti–disabled-1 (Dab1, Chemicon, 1:1,000), a marker for AII amacrine cells. Secondary: donkey anti– rabbit antibodies (IgG (H+L), Jackson, 1:200, conjugated with Cy5). (ii) Primary: sheep anti–GFP (Biogenesis, 1:200). Secondary: donkey anti–sheep antibodies (IgG (H+L), Molecular Probes, 1:200, conjugated with Alexa Fluor 488). This
nature NEUROSCIENCE
the depth of the retina (Supplementary Fig. 1a). The depth of the peak of the neurobiotin staining in the IPL relative to the depth of the two IPL peaks of the ChAT staining was used to determine the stratification level of the recorded cell. The depths of the ChAT peaks in the IPL were defined as 0% (proximal ChAT band) and 100% (distal ChAT band). The boundaries of the IPL (relative to the ChAT strata) were defined by determining, first, the peak DAPI fluorescence in the ganglion cell layer and the INL, and, second, the position toward the IPL where the fluorescence intensity of DAPI fell to two-thirds of its peak value. Other confocal stacks were acquired using a ×63 oil immersion lens, NA 1.4 (Fig. 7g,h,j–l), or ×40 oil immersion lens, NA 1.3 (Fig. 7f). Computational model. See Supplementary Figure 6.
Confocal analysis. We analyzed the stained retinas with a Zeiss LSM 510 META confocal microscope. Overall morphologies of the recorded ganglion cells were assessed by using a ×40 oil immersion lens, numerical aperture (NA) 1.3 (Fig. 1b bottom panel). The stratification level of neurobiotin-filled ganglion cells (Fig. 1b top panel and Supplementary Fig. 1) was determined using image stacks acquired with a ×100 oil immersion lens, NA 1.4, at the periphery of the dendritic tree. We plotted first the intensity profiles of the ChAT and neurobiotin staining along
43. Hippenmeyer, S. et al. A developmental switch in the response of DRG neurons to ETS transcription factor signaling. PLoS Biol. 3, e159 (2005). 44. Buffelli, M. et al. Genetic evidence that relative synaptic efficacy biases the outcome of synaptic competition. Nature 424, 430–434 (2003). 45. Metzger, D. & Feil, R. Engineering the mouse genome by site-specific recombination. Curr. Opin. Biotechnol. 10, 470–476 (1999). 46. Deans, M.R., Gibson, J.R., Sellitto, C., Connors, B.W. & Paul, D.L. Synchronous activity of inhibitory networks in neocortex requires electrical synapses containing connexin36. Neuron 31, 477–485 (2001).
© 2009 Nature America, Inc. All rights reserved.
antibody was used only after paired recordings; it was omitted for retinas in which we had recorded only from amacrine cells. The following set of primary and secondary antibody combinations were used to identify cells in the distal INL as ON cone bipolar cells after paired recordings from amacrine and PV-5 cells (both neurobiotin filled). (i) Primary: rabbit anti– Gγ13 (R. Margolskee, 1:300). Gγ13 is a marker for ON bipolar cells. Secondary: donkey anti–rabbit antibodies (IgG (H+L), Jackson, 1:200, conjugated with Cy5). (ii) Primary: mouse anti–Protein Kinase C (PKC; BD Bioscience, 1:200). PKC is a marker for rod bipolar cells. Secondary: donkey anti–mouse antibodies (IgG (H+L), Jackson Labs, 1:200, conjugated with DyLight 405). (iii) Primary: sheep anti-GFP (Biogenesis, 1:200). Secondary: donkey anti–sheep (IgG (H+L) Molecular Probes, 1:200, conjugated with Alexa Fluor 488).
nature NEUROSCIENCE
doi:10.1038/nn.2389
a r t ic l e s
Coding of stimulus sequences by population responses in visual cortex
© 2009 Nature America, Inc. All rights reserved.
Andrea Benucci1,2, Dario L Ringach3 & Matteo Carandini1,2 Neuronal populations in sensory cortex represent time-changing sensory input through a spatiotemporal code. What are the rules that govern this code? We measured membrane potentials and spikes from neuronal populations in cat visual cortex (V1) using voltage-sensitive dyes and electrode arrays. We first characterized the population response to a single orientation. As response amplitude grew, the population tuning width remained constant for membrane potential responses and became progressively sharper for spike responses. We then asked how these single-orientation responses combine to code for successive orientations. We found that they combined through simple linear summation. Linearity, however, was violated after stimulus offset, when responses exhibited an unexplained persistence. As a result of linearity, the interactions between responses to successive stimuli were minimal. Our results indicate that higher cortical areas may reconstruct the stimulus sequence from V1 population responses through a simple instantaneous decoder. Therefore, spatial and temporal codes in area V1 operate largely independently. The computations performed by sensory cortex involve large neuronal populations1,2 whose activity must evolve over time to code an everchanging sensory input. We sought to understand the rules that govern this dynamic code. Such rules would describe the computations performed by the underlying circuits. We also sought to understand the strategies required to decode these responses, as these strategies must be followed by downstream neurons that interpret V1 population activity. In some neural systems, the spatial and temporal aspects of the neural code are inextricably linked3. For example, the population activity that codes for a sequence of two odors in the insect olfactory system differs not only from the representation of either odor alone, but also from a mixture of the two4. Similarly, in the mammalian motor cortex, the temporal evolution of the population activity that codes for a movement describes a complex trajectory in state space5. An analogous situation may occur in rodent somatosensory cortex, where responses to sequences of whiskers differ substantially from responses to individual whiskers6. We asked whether such a linkage of spatial and temporal coding is present in visual cortex and specifically in the representation of stimulus orientation by the primary visual area (V1). Orientation selectivity in V1 is an example of cortical elaboration of thalamic sensory afferents, providing a test bed for theories of cortical function7,8. Studies of the dynamics of V1 population activity have concentrated on membrane potential responses to individual stimuli and have found a dissociation of spatial and temporal coding. Using voltage-sensitive dye imaging, these studies found that the orientation selectivity of the population remained constant over the course of the response9,10. This invariance is consistent with
tuning width measurements made from the membrane potential of single neurons; this tuning width typically remains constant during the response11. Do the spike responses to individual stimuli show a similar dissociation of spatial and temporal coding? The dynamics of V1 spike responses to oriented stimuli have been studied extensively in single neurons. Over the course of the response, the orientation bandwidth of neurons has been found to sharpen12–16, broaden17 or remain largely constant8,18. This variety may stem, in part, from differences in experimental procedures or in data analysis. Given the variety of results observed even within individual studies, however, it remains unknown whether the spike response to a single orientation of V1 populations shows any dynamics such as sharpening of selectivity over time. More generally, it is not known whether the population response to individual orientations can predict population responses to stimulus sequences. The interactions between responses to subsequent orientations could be complex. Indeed, the spike responses of individual V1 neurons to pairs of orientations have effects ranging from suppression to a repulsion of tuning curves 19–22. We don’t know the degree to which these interactions may affect the overall population responses. Similarly, it is not known how population activity evolves once the stimulus drive has been removed. Some V1 neurons give prolonged responses at stimulus offset23, perhaps providing support to perceptual phenomena of visual persistence 24. These offset responses may reflect intracortical interactions that reverberate activity, possibly causing attractors in the dynamics of cortical activation25,26. Attractors would explain why spontaneous activity
1University College London Institute of Ophthalmology, University College London, London, UK. 2Smith-Kettlewell Eye Research Institute, San Francisco, California, USA. 3Departments of Neurobiology and Psychology, Jules Stein Eye Institute, University of California Los Angeles, Los Angeles, California, USA. Correspondence should be addressed to A.B. (
[email protected]).
Received 9 April; accepted 19 August; published online 13 September 2009; doi:10.1038/nn.2398
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1317
a
45
60
75
90
105
180 ms
–5
× 10
8 4 0 −4
b 45 90
0 135
c Fluorescence (dF F –1)
Figure 1 Population responses to an oriented stimulus. (a) Average maps of VSD fluorescence triggered on the appearance of a 90° grating, at various delays from grating onset. Scale bar indicates 1 mm. (b) Combining maps obtained with multiple stimulus orientations yielded a map of orientation preference. Hue indicates the preferred orientation, and brightness indicates tuning strength. (c) Summary of the membrane potential responses of the population. The responses of pixels with similar orientation preference were averaged and the result was plotted as a function of preferred orientation (relative to stimulus orientation). The mean across orientations was removed. Black curves indicate best-fitting Gaussian functions and error bars indicate ± s.e.m. (across six stimulus orientations). (d) Two superimposed population responses separated by a 20-ms interval. (e) Summary of the spike responses of the population. The same methods as described in a–c were applied to the firing rate measured with the electrode array. (f) Two superimposed population responses separated by a 20-ms interval. Data in a–d are from experiment 56-4-1; data in e and f are from experiment 75-5-16.
Fluorescence (dF F –1)
4
d
Membrane potential
–5
× 10
45 ms
60
75
90
105
100%
50 ms 70 ms
2 0
50%
–2 in V1 favors patterns that resemble those 0% elicited by oriented stimuli25,27. Perhaps, –90 0 90 –90 0 90 once the responses attain one of these patterns, they linger there even after the e f Spikes 38 ms 100% initial drive is removed. 58 ms 49 57 66 74 82 0.4 Finally, we don’t know how stimulus 0.2 sequences coded into population responses 50% should be decoded by later stages of the visual 0 system. Simple rules have been proposed for the –0.2 0% decoding of stimulus orientation from V1 popu–90 0 90 1,2,28–31 –90 0 90 lations . These rules, however, have rarely Prefered orientation Prefered orientation been tested on actual population responses and, relative to stimulus (deg) relative to stimulus (deg) in particular, they have never been applied to dynamically changing population responses. Can the time-evolving profile of the membrane potentials following a given orientation (Fig. 1a). Activation generally emerged ~45 ms after stimulus onset population activity of area V1 be decoded using simple rules? and assumed the characteristic patchy structure of orientation maps36. RESULTS As expected37, stimuli with lower contrast elicited slower responses We monitored the dynamics of population activity in visual cortex, (Supplementary Fig. 2). Responses to orthogonal stimuli tended to measuring both spikes and membrane potentials in identical stimulus be complementary so the single-orientation maps could be combined conditions. To measure a population’s spikes, we recorded from 10 × into a map of orientation preference36 (Fig. 1b). 10 electrode arrays32. To measure a population’s membrane potentials, To investigate the selectivity of these responses, we expressed them we used voltage-sensitive dye (VSD) imaging33. VSD imaging reveals as a function of preferred orientation (Fig. 1c,d). Having assigned the membrane potential responses in superficial layers34 with better a preferred orientation to each pixel (Fig. 1b), we summarized the than 10-ms temporal resolution9,10,33. In V1, its signals largely reflect activity of the population (Fig. 1a) by averaging the responses of pixels with similar orientation preference9 (Fig. 1c). For this analysis, the responses of complex cells10. Seeking to activate multiple subpopulations of neurons in rapid we averaged responses to stimuli of different orientations, expressed sequence, we used an orientation-noise stimulus. In this stimulus, relative to the difference between stimulus orientation and preferred gratings of random orientation (each flashed for 32 ms in random orientation. The shape of these responses remained constant; once phase) are presented in sequence8,14. To study how cortical responses responses at different times were normalized by their amplitude, their transition between resting and stimulated states and vice versa, we profiles became very similar (Fig. 1d). We applied a similar analysis to the spike responses and obtained randomly interleaved a fraction (~30%) of blank frames, uniform gray screens that had the same mean luminance as the gratings. This fairly different results (Fig. 1e,f). We identified the preferred orientation orientation-noise stimulus elicited lively spike responses35 and strong of each site in the electrode array and averaged together the responses VSD signals (Supplementary Fig. 1). of sites with similar preference (Fig. 1e). Unlike the membrane potential responses, the spike responses showed a dependence Population responses to an oriented stimulus of tuning width on time, with population activity being more The orientation-noise stimuli yielded detailed maps of orienta- narrowly distributed at the peak of the response than at earlier or tion preference (Fig. 1a,b). We measured the average activation later times (Fig. 1f). Firing rate (z score)
© 2009 Nature America, Inc. All rights reserved.
a r t ic l e s
1318
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
a r t ic l e s
Membrane potential
80 60 40 20
b Tuning width (deg)
Spikes
100
0
50
100
25
Spikes
40
d Tuning width (deg)
100 80 60 40 20
60
80
100
60 80 Time (ms)
100
35 30 25 20 40
50 100 150 Time (ms)
f
Membrane potential
Firing rate
0.15
Untuned
0.3 0.15 0
0.05 –0.05
90
Tuned 1.0
−90
0.5
0
0
−0.4 0
100 Time (ms)
200
−0.5 0
Amplitude (z score)
0
0.4
100 200 Time (ms)
To compare the population responses measured from spikes and membrane potentials, we fitted them with circular Gaussian functions38. These provided excellent fits (Fig. 1c,e) and yield two parameters: amplitude and tuning width (Fig. 2a,b). For membrane potential, the amplitude peaked 74 ± 2 ms after stimulus onset (n = 8; Fig. 2a). For spikes, the amplitude peaked earlier, at 50 ± 2 ms (n = 6; Fig. 2a). Tuning width (half-width at half-height) for membrane potential remained fairly constant during the course of the responses and averaged 31.2 ± 0.3° (Fig. 2b). For spikes, the tuning width decreased by 27%, from close to 30° to almost 20°, before broadening again (Fig. 2b). The sharpening observed in population responses measured from spikes was also seen in individual neurons (Fig. 2c,d). We studied the spike responses of 32 well-isolated single units in the population. Similar to the multiunit spike activity, these singleunit responses peaked 50 ± 2 ms after stimulus onset (Fig. 2c) and decreased in tuning width from close to 30° to almost 20° before broadening again (Fig. 2d). This clear sequence of sharpening Figure 3 Predicting the membrane potential responses of the population to the full stimulus sequence. (a) The stimulus expressed as an image, with ones (dots) indicating the sequence of stimulus orientations as a function of time and zeros elsewhere (gray). For graphical purposes, the orientation axis was duplicated to cover the full 360° range. Thus, each grating appears as two dots separated by 180°. Only 2 s of stimulation are shown here (typical stimuli lasted 30 s). Dots have been shifted in time by 74 ms to compensate for the delay of the membrane potential signal. (b) Elemental population response to a single orientation (from Fig. 2e). The asterisk denotes convolution. (c) Population responses predicted by convolving the stimulus with the elemental response. (d) The measured population responses. (e) Relationship between model predictions and measured responses. Dots are mean values taken across n = 8 hemispheres. Gray regions indicate ± 1 s.d. Data in a–d are from experiment 56-4-1.
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
and broadening during the course of the response was seen in all neurons, with no exceptions. The difference in tuning width between membrane potential and spikes is consistent with intracellular measures in single neurons and is largely a result of the spike threshold38. The difference in timing is less expected. It is probably a result of two factors. First, neurons tend to fire during the rising phase of the underlying membrane potential39. Second, an early untuned depolarization8,9,11 may help the potential to reach spike threshold in the initial portion of the responses. The properties of this depolarization are described below. The population response to a single-oriented stimulus could be decomposed into the sum of two components, one untuned and one tuned8,9,11 (Fig. 2e,f). The untuned component was the mean activity across preferred orientations; it varied only in time. The residual tuned component varied both in time and in preferred orientation. Consistent with earlier reports8,9,11, the untuned component led the tuned component in time. Responses to a sequence of orientations Having characterized the population response to single-oriented stimuli, we asked whether a sequential application of this elemental population response can predict the responses to the full stimulus sequence. We first considered the membrane potential responses and found that we could predict them on the basis of simple summation (Fig. 3). We expressed the stimulus as a function of orientation and time (Fig. 3a) and predicted responses by repeatedly summing the
a
c
d
b
Stimulus
360 270
90
180
0
*
90
–90
0
0
0.2
Predicted membrane potential
360
–0.5 0 0.5 z score
270 180
90
e
0
Prediction (z score)
0
Stimulus orientation (deg)
0
Preferred orientation (deg)
Amplitude (%)
30
150
c
© 2009 Nature America, Inc. All rights reserved.
Membrane potential
20
0
e
35
Figure 2 Properties of single-orientation responses. (a) Time course of response amplitude (from Gaussian fits in Fig. 1c,e) for membrane potential (dark gray) and spikes (light gray). Shaded regions indicate ± s.e.m. (n = 8 hemispheres for membrane potential and 6 hemispheres for spikes). Black dots indicate data points above background level. (b) Time course of the tuning width (half-width at half-height). Only values corresponding to amplitudes above background level (black dots in a) are shown. (c) Time course of response amplitude for isolated single units (n = 31). (d) Time course of tuning width for isolated single units for time points above background level (black dots in c). (e,f) Population responses for membrane potential (e) and spikes (f) were decomposed into two additive terms: a tuned component (which varies both in orientation and in time, shown on the bottom) and a baseline component (which varies only in time, shown by the black curve on top). To account for variability, we expressed responses as z scores (mean across repeats divided by s.d. across repeats). Data in e are from experiment 56-4-1, data in f are from experiment 75-5-16.
Preferred orientation (deg)
Amplitude (%)
a
Measured membrane potential
360
270 180
90 0
0
1 Time (s)
2
2 1 0 −1 −2
n=8 −2 −1 0 1 2 Response (z score)
1319
a r t ic l e s
Preferred orientation (deg) Preferred orientation (deg)
d
270
for high and low values of the response (Fig. 4e). Thus, the model performance was improved by including a compressive nonlinearity, yielding a correlation between predicted and actual responses of 0.58 ± 0.06 (s.d., P < 0.005, n = 6; Fig. 4e). These results indicate that population responses to the stimulus sequence are simply the sum of the consecutive responses to the individual elements in the sequence. On the other hand, there is a noticeable scatter in the relationship between predicted and actual population responses, both for membrane potential (Fig. 3e) and for spikes (Fig. 4e). Are the deviations systematic? To address this question, we focused on two key stimulus conditions: the transitions from orientation to orientation and the transitions from stimulus to blank.
90
180 90 0
−90
0
270 180
e
90 0 Measured firing rate
360 270 180 90 0
1 Time (s)
0
0.2
1.0
Predicted firing rate
360
=
0
*
0
2
–0.5 1.0 0.5 0 −0.5 n=6 −0.5 0 0.5 1.0 Response (z score)
Figure 4 Predicting the spike responses of the population to the full stimulus sequence. Data are presented as in Figure 3, with population responses measured from spikes rather than from membrane potentials. Data in a–d are from experiment 75-5-16.
Effect of recent history of stimulation We asked whether the summation model could explain how the population’s activity depends on the orientations seen in the immediate past (Fig. 5). We focused on four ‘orientation jumps’, in which consecutive gratings differed in orientation by 0° (no change), 90°, +45° and −45°. The membrane potential responses of the population faithfully represented the end points of the orientation jumps, but tended to interpolate through intermediate orientation preferences during the transitions (Fig. 5a,e). For 0° changes, the membrane potential response showed a single prolonged activation (Fig. 5a); there was no clear demarcation between the responses to the two stimuli even though the transition between stimuli involved a change in spatial phase in 75% of the cases. For a 90° jump, conversely, the population response showed a distinct transition (Fig. 5a), involving a sequence of two separate subpopulations with orthogonal orientation preferences. The responses to ±45° jumps showed a mixture of these behaviors (Fig. 5a); responses showed two distinct peaks, but these peaks were joined by a response passing through the intermediate orientation preferences. The profiles of the membrane potential responses during the transition showed a marked attraction
elemental population response (Fig. 3b) appropriately shifted in time and orientation (a two-dimensional convolution). The result was a linear prediction of the responses (Fig. 3c). Aside from occasional discrepancies, the predicted responses resembled the actual responses (Fig. 3d). Both predicted and actual responses showed peaks of activity that shifted to coincide with the appropriate stimulus orientation. Distributions of predicted versus actual responses clustered along a curve that deviated from linearity only for large responses (Fig. 3e). We were able to improve on the predictions of the model by fitting a mildly compressive nonlinear function to these distributions and applying it after the summation40. The resulting model explained 98 ± 1% of the variance of the responses (s.d., n = 8). The correlation between predicted and actual responses was high, at 0.69 ± 0.06 (s.d., P < 0.005, n = 8). We obtained similar results for the spike Membrane potential responses (Fig. 4). The summation model a b Predicted Measured (Fig. 4a,b) captured the transient nature of 90 these responses and the sequence of peak 0 activations (Fig. 4c,d). Just as for membrane –90 potential, the major deviations occurred Figure 5 Predicting the interactions between population responses to successive orientations. (a) Average membrane potential responses of the population to two successive stimuli. Left, schematic describing the stimulus. From top to bottom, successive stimuli differing in orientation by 0°, 90°, −45° and +45°. (b) Predictions of the summation model for the data presented in a. (c,d) Data for the spike responses of the population are presented as in a and b. (e) Attraction of population response profiles during ±45° jumps in orientation. Profiles are measured at the peak time of the second response (60.5 ± 1.7 ms from the onset of the second stimulus). Curves are model predictions. The average of five hemispheres is shown. (f) Data for the spike responses of the population are presented as in e. Data in a and b are from experiment 69-1-5, data in c and d are from experiment 75-5-16.
1320
0
0
Firing rate
c
100 200 Time (ms)
100 Time (ms)
f 0.8 0
0.8 n=6
−0.8
d
Measured
0
e Response (z score)
© 2009 Nature America, Inc. All rights reserved.
z score –1 0 1
z score
c
b
Stimulus
360
Prediction (z score)
Stimulus orientation (deg)
a
0
Predicted
200 −45 deg +45 deg Prediction n=5
−0.4 −90 −45 0 45 90 Preferred orientation (deg)
−90 −45 0 45 90 Preferred orientation (deg)
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
Orientation (deg)
Contrast (%)
a r t ic l e s Figure 6 Unexplained persistence of population responses after stimulus offset. (a) When the visual stimulus is removed (stimulus-to-blank conditions, as shown on top by the contrast change), membrane potential activity in the population persisted for well over 50 ms. Gray symbols indicate that the stimulus could have any orientation or be blank before and after the stimulus-blank sequence. (b) Data for spike responses are presented as in a. (c,d) The prediction of the summation model was substantially shorter for both spike responses and membrane potential. (e) The persistence was tuned for orientation. A separable model obtained through singular-value decomposition of the response shown in a yielded a very similar profile. (f) Data for spike responses are presented as in e. (g,h) Temporal dynamics of responses (red) and predictions (black). Shaded areas indicate ± s.d. (n = 6 hemispheres for g and 8 hemispheres for h). (i,j) Difference between the predicted and actual time courses shown in g and h, averaged over hemispheres. Shaded areas indicate ± s.d. (n as indicated). Data in a, c and e are from experiment 68-3-5, data in b, d and f are from experiment 75-5-16.
50 0
a
90
b
Membrane potential
Firing rate
0 −90
c
d
e
f
Predicted
Separable model
g
z score
0.4
0
h
0.4 z score
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1321
Amplitude (%)
between the responses to the first and second stimuli; the peak of the population response showed an average shift of 20.7 ± 2.6° (s.e.m., n = 6) toward the preceding stimulus orientation (Fig. 5e). Similar results were obtained when examining smaller orientation jumps (23° or 30°; data not shown). Similar, but less marked effects, were seen in the spike responses of the population (Fig. 5c,f), which more faithfully followed the transitions between orientations. Although the responses to 0° changes showed a single prolonged activation that was similar to that seen in membrane potential (Fig. 5c), responses to jumps by 90°, 45° and –45° showed clearly distinct peaks, with little response in neurons with intermediate orientation preferences (Fig. 5c). Just as for membrane potential, however, the population responses showed a marked attraction toward the orientation of the previous stimulus, with a shift of 6.3 ± 2.8° (s.e.m., n = 5) toward the preceding stimulus orientation (Fig. 5f). All of these effects were explained accurately by the simple summation model. From the population responses predicted by the model, we computed average responses to orientation jumps (Fig. 5b,d). They closely resembled the data, explaining 91 ± 3% (s.d., n = 8) of the variance for membrane potential (Fig. 5b) and 94.3 ± 2% (s.d., n = 6) of the variance for spikes (Fig. 5d). Specifically, the model accurately predicted the attraction in population responses that followed a ± 45° jump, both in membrane potentials and in spikes (Fig. 5e,f). Therefore, even when the population responses appear to be traveling waves that involve neurons with orientation preference intermediate between two successive orientations, these responses
Persistence of population responses We then asked how the population response evolves once the stimulus drive has been removed. We measured the population activity that follows stimulus offset and tried to explain it with the simple summation model (Fig. 6). The membrane potential responses showed a persistence that could not be explained by the summation model. We averaged all of the cases in which a stimulus was followed by a blank (Fig. 6a) and compared them with the model prediction (Fig. 6c). The measured responses clearly outlasted the prediction (Fig. 6g). They persisted 48 ± 3 ms (n = 8) longer than predicted, with the peak difference occurring 72 ± 24 ms after stimulus offset (Fig. 6i). Membrane potential responses, therefore, persisted well beyond what would be expected from the elemental response to a given orientation. We observed a similar persistence in the population responses measured from spikes. The spike responses to a stimulus followed by a blank (Fig. 6b) had a longer tail than the prediction of the summation model (Fig. 6d). The responses during the blank stimulus persisted 30 ± 4 ms (n = 6) longer than predicted (Fig. 6h) and the peak difference occurred 64 ± 31 ms after stimulus offset (Fig. 6j). The inadequacy of the model in predicting the offset responses is not a result of erroneous choice of model parameters. The model involves convolution with a filter, the average population response to a single orientation. This filter is biased toward stimulus-to-stimulus transitions (~70% of the data). To assess the effect of this bias, we computed the filter using only stimulus-to-blank transitions. By definition, the resulting model did a perfect job of predicting the offset responses. However, it did markedly worse at describing the average orientation-to-orientation transitions; the explained variance dropped from 91% to 75%. Therefore, the summation model could not explain population responses to both kinds of transition, from stimulus to stimulus and from stimulus to blank. Moreover, the failure of the summation model in predicting persistence is not simply a result of an inadequacy in dealing with blank stimuli; the model encountered no difficulties in predicting the responses to stimuli following a blank. We saw no consistent difference between model predictions and actual responses in onset latency, peak amplitude and width of the response. Thus, cortical patterns of activity produced by a stimulus are similar whether they follow another response or a period of spontaneous activity. Two observations suggest that the origin of response persistence is cortical. First, the persistence is tuned for orientation, which involves
100
100
75
75
50
50
25
25
0
Measured Predicted
0
i Difference (%)
© 2009 Nature America, Inc. All rights reserved.
−0.2
are more usefully described as the sum of two elemental waves, the successive responses to the individual orientations.
n=8
50 25
j
n=6
10 5
0
0
–25 0
100 Time (ms)
200
0
100 200 Time (ms)
a r t ic l e s Preferred orientation relative to stimulus (deg)
a
Probability
0
0 −90
0
0.1 0.2 Time (s)
0.5
1
–10 0 10 20 Response (spikes per s) Average firing rate
360
40 20 0 −20
270 180 90
Figure 7 Decoding stimulus orientation from the population responses. (a) The average spike response to an individual stimulus orientation. The dotted lines indicate the interval when this response was significantly higher than baseline. (b) Probability that a spike response was evoked by a specific orientation (computed in the time interval indicated in a). (c) Firing rates of the population in response to a stimulus sequence averaged over ten trials. (d) Firing rates for an individual trial. (e) Stimulus likelihood estimated by the instantaneous Bayesian decoder from the single-trial responses. (f) The stimulus sequence (shifted by 50 ms to compensate for response delay). (g) Confusion matrix comparing the estimated orientation to the actual stimulus orientation. (h) The average stimulus likelihood when a stimulus was followed by a blank. Data in a–h are from experiment 79-12-16.
0
Decoding population responses Finally, we asked how the spike responses of the population should be decoded to reconstruct the stimulus sequence. We have seen that, at any instant, population responses depend not only on the present stimulus, but also on the recent history of stimulation. This dependence was explained by the linear summation of elemental responses to
individual orientations. In principle, therefore, to estimate the present stimulus orientation, a decoder should know the rules of summation, the shape of the elemental response and the recent history of stimulation. Could a decoder neglect these aspects and still interpret the population responses instantaneously and accurately? We implemented a simple Bayesian decoder1,28,31 that operates on instantaneous population firing rates, with no knowledge of previous stimuli and responses or of the summation rule (Fig. 7a,b). This decoder quantizes time in 8-ms intervals. For each interval, it measures the likelihood that the present stimulus has orientation θ given a population response. This likelihood is the product of the probabilities of observing each of the responses in the bins of orientation preference, conditioned on θ. We obtained these probabilities from the averaged population response to a single orientation (Fig. 7a) by measuring the mean and variance across stimulus presentations (Fig. 7b). Such a simple decoder accurately predicted the sequence of stimulus orientations (Fig. 7c–g). To be realistic, the decoder must operate on population responses measured in single trials (Fig. 7d), which can be considerably noisier than responses averaged over trials. The decoder predicts a distribution of stimulus likelihoods (Fig. 7e). The maxima of this distribution are the decoded orientations. They matched the actual stimulus orientations very closely (Fig. 7f), with a circular correlation of r = 0.68 in the example dataset (Fig. 7g) and of r = 0.67 ± 0.03 (s.e.m., P < 10−6) across five experiments. Because of response persistence, the decoder continued to predict the presence of the previous orientation during blank intervals (Fig. 7h). To study this ‘perceived persistence’, we averaged all of the conditions in which a stimulus was followed by a blank. Consistent with the persistence of population responses at stimulus offset (Fig. 6b), the distribution of stimulus likelihoods during a blank interval was not flat, but instead resembled the distribution obtained during the preceding stimulus (Fig. 7h). In other words, during the blank intervals, the decoder indicates that the stimulus has the orientation of the preceding stimulus. These perceived orientations simply reflect the fact that the population responses during a blank interval maintain a trace of the previous stimulus. We performed this formal analysis on the spike responses, as these constitute the output of area V1 and are read by subsequent visual areas. An informal analysis indicated that it would be equally easy to decode the population responses measured from membrane potential (Supplementary Fig. 1 and Supplementary Video 1). We conclude that it is not necessary for a decoder of population activity in area V1 to know the relationship between temporal and spatial encoding. This relationship affected the population responses in two ways. First, the elemental population responses to a single orientation measured from spikes showed sharpening over time. Second, the population responses retained a trace of previous stimuli, interacting with each other through summation. These properties of the
1322
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
d
Preferred orientation (deg)
Single trial firing rate 360 270
90 0 Stimulus likelihood
360 270
0.4 0.2 0
180 90
Probability
Orientation (deg)
e
f
60 20 −20
180
Spikes per s
0
Orientation (deg)
Stimulus 360 270 180 90 0 0
180 135 90 45 0 0 45 90 135 180 Stimulus orientation (deg)
h
Preferred orientation relative to stimulus (deg)
Predicted orientation (deg)
g
1 Time (s)
2
90
1.0
0
0.5
−90
0 0
Probability
© 2009 Nature America, Inc. All rights reserved.
Spikes per s –5 5 15
Spikes per s
Preferred orientation (deg)
c
b
90
100 200 Time (ms)
the same subpopulation of neurons that were initially activated by the stimulus (Fig. 6e,f). Indeed, the averaged response to a stimulus followed by a blank was separable, both for membrane potential responses (Fig. 6a,e) and for spike responses (Fig. 6b,f). Second, the persistence is most pronounced in the superficial layers of cortex. In a control experiment using multiprobe electrodes spanning the depth of cortex, we computed field potentials and the associated profile of current source density. We compared this profile with the predictions of the summation model and found persistence in the superficial layers, but not in deeper layers (Supplementary Fig. 3). Together, these results suggest that response persistence is not inherited from the thalamus. Its causes may lie in the circuitry of cortex.
a r t ic l e s
DISCUSSION Using electrophysiology and imaging, we examined how large populations of neurons in area V1 code for stimulus orientation. We observed how population activity changes over time to reflect a rapidly changing sensory input and found that the dynamics of this activity are summarized by a set of simple rules. We first investigated the population responses to a single orientation. When measuring membrane potential, we found that the width of the population profile remained constant during the course of the responses. This result is consistent with earlier measurements obtained in populations with periodic stimuli9,10 and in single neurons with random stimuli11. When measuring spike responses, however, we found that activity showed a pronounced sharpening during the course of the response. This sharpening was present not only in the overall population profile, but also in all of the individual neurons in the sample, with no exceptions. The uniformity of sharpening across neurons is surprising to us, given the variety of observations reported in previous singleneuron studies8,12–18. This variety may be explained by multiple factors: differences in stimulus attributes such as size (large stimuli like ours lead to clearer sharpening)41, differences in cell sample, including layers and location in the orientation preference map42 and differences in data analysis, for example, the way responses are normalized16 and whether one’s measure of orientation selectivity discounts an untuned response8,9,11,14. We then investigated the interactions between oriented stimuli presented in succession and found such interactions to be prevalent in the population responses, both in membrane potential and in spike activity. Population activity was attracted toward the orientation of the preceding stimulus, and such attraction corresponded to repulsion of tuning curves of individual neurons. These results are consistent with studies in single neurons, where suppression near the conditioning stimulus19 can accompany an enhancement at other orientations, resulting in the repulsion of tuning curves away from the conditioning stimulus20–22. With the exception of one study21, previous reports of interactions between subsequent orientations in single neurons did not ask whether such interactions would be explained by simple summation. Our data support this interpretation. Specifically, summation could predict the population responses to changes in orientation with high accuracy. These responses can appear as traveling waves, peaking at a range of intermediate orientations between the stimulus orientations. However, the summation model reveals that they really are the sum of two elemental waves. The fundamental linearity that we observed in population responses may be surprising given the nonlinearities seen in individual V1 neurons. Our summation model is linear; population responses are a weighted sum of past stimulus orientations. The static nonlinearity at the output of the sum does not alter this fundamental linearity40. In comparison, complex cells, which constitute the bulk of the signal in our imaging experiments 10, are markedly nonlinear in their integration of spatiotemporal inputs43,44. Even simple cells, whose spatial summation properties are more linear43, have strong temporal nonlinearities45. Perhaps the orientationnoise stimulus places the cortex in a linear regime (possibly because contrast is mostly constant and fairly high). Indeed, as a first
approximation, the spike responses of individual neurons to this stimulus can be predicted by the summation model35. However, linear summation cannot be a complete description of V1 population responses. For instance, linear summation would not explain population responses to stimuli of different contrast. Linearity would predict that responses scale proportionally with contrast, whereas V1 responses vary nonlinearly with contrast, both in amplitude and in time course37,43. Accounting for these nonlinearities would probably require extending the model to include gain control mechanisms that are present at all stages of the early visual system43. Indeed, we found a previously unknown deviation from linearity: activity persisted once the stimulus drive was removed. This prolonged activation could be seen as a prolongation of the response to the stimulus (persistence) or a response to the disappearance of the stimulus (off response). We favor the first description for two reasons. First, the lag between an off response and the preceding on response would resemble the stimulus duration (32 ms), whereas the prolonged activation peaked at a later time. Second, an off response is already included in the average stimulus-triggered population responses: the stimulus disappears (on average, half of the pixels change gray level) every time there is a change in orientation. We argue that this neural persistence is probably a result of cortical mechanisms, possibly implicating the strong recurrent excitation that characterizes cortical circuits46. Recurrent circuits in area V1 may favor states that signal a single orientation7,8,25–27,47. Such states might act as attractors25,26. The stimulus before the blank might drive the network in one of these attractors; when the drive is removed, activity persists because it is transiently trapped in the attractor. Attractors are probably weak compared with the feedforward drive or they would lead to marked hallucinations47. However, their influence might be strong at low contrast or during a blank stimulus, when the feed forward drive is weaker or absent. Perceptually, the neural persistence that we have observed may be a correlate of phenomena of visual persistence, which is in turn related to iconic memory24. Such correlates have previously been found in higher visual areas48–50. Smaller, but similar, effects have been seen in area V1 (ref. 23). The dynamics of these responses, however, have been studied only in a few neurons and have been characterized qualitatively. Our results indicate that persistence is robustly present in the population responses of area V1 and that this persistence is not a result of the off responses that are generally expected from a receptive field. Finally, we found that the time-evolving spike responses of a V1 population can be read out through a simple instantaneous decoder. Multiple methods have been proposed for decoding stimulus orientation from V1 populations1,2,28–31. These methods have rarely been tested on actual population responses and, in particular, they have not been used to reconstruct the attributes of stimulus sequences from dynamically changing population responses. We implemented a simple decoder that operates instantaneously and found that it did an excellent job of predicting the stimulus orientation. The good performance of the decoder is surprising for a number of reasons. First, it is typically assumed that a decoder needs to take into account the large amount of variability that is shared among neurons2; our decoder ignores this covariance and yet it performs very well. Second, our decoder ignores a number of factors that would appear to be crucial: the rules of interaction between subsequent elemental responses, the shape and duration of the elemental response, and the recent history of stimulation. We suggest that the linearity of summation that we found reduces the importance of these factors. Indeed, these factors are essential for decoding the population
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1323
© 2009 Nature America, Inc. All rights reserved.
coding mechanisms, however, can be safely ignored by a subsequent decoder. Indeed, a decoder with no knowledge of any linkage between temporal and spatial encoding could successfully predict the sequence of input stimuli.
a r t ic l e s activity in systems in which the spatial and temporal aspects of the neural code are inextricably linked3. In conclusion, we determined a set of rules that govern the dynamical code by which populations of V1 neurons represent the attributes of a rapidly changing stimulus. Such rules ultimately describe the computations performed by the underlying circuits. We have also found a simple example of the strategy that downstream neurons could follow to decode the population responses. By elucidating both aspects, coding and decoding, these findings may ultimately help to relate neural activity to perception.
1. Pouget, A., Dayan, P. & Zemel, R.S. Inference and computation with population codes. Annu. Rev. Neurosci. 26, 381–410 (2003). 2. Averbeck, B.B., Latham, P.E. & Pouget, A. Neural correlations, population coding and computation. Nat. Rev. Neurosci. 7, 358–366 (2006). 3. Buonomano, D.V. & Maass, W. State-dependent computations: spatiotemporal processing in cortical networks. Nat. Rev. Neurosci. 10, 113–125 (2009). 4. Broome, B.M., Jayaraman, V. & Laurent, G. Encoding and decoding of overlapping odor sequences. Neuron 51, 467–482 (2006). 5. Yu, B.M. et al. Mixture of trajectory models for neural decoding of goal-directed movements. J. Neurophysiol. 97, 3763–3780 (2007). 6. Civillico, E.F. & Contreras, D. Integration of evoked responses in supragranular cortex studied with optical recordings in vivo. J. Neurophysiol. 96, 336–351 (2006). 7. Ferster, D. & Miller, K.D. Neural mechanisms of orientation selectivity in the visual cortex. Annu. Rev. Neurosci. 23, 441–471 (2000). 8. Shapley, R., Hawken, M. & Ringach, D.L. Dynamics of orientation selectivity in the primary visual cortex and the importance of cortical inhibition. Neuron 38, 689–699 (2003). 9. Sharon, D. & Grinvald, A. Dynamics and constancy in cortical spatiotemporal patterns of orientation processing. Science 295, 512–515 (2002). 10. Benucci, A., Frazor, R.A. & Carandini, M. Standing waves and traveling waves distinguish two circuits in visual cortex. Neuron 55, 103–117 (2007). 11. Gillespie, D.C., Lampl, I., Anderson, J.S. & Ferster, D. Dynamics of the orientationtuned membrane potential response in cat primary visual cortex. Nat. Neurosci. 4, 1014–1019 (2001). 12. Shevelev, I.A., Sharaev, G.A., Lazareva, N.A., Novikova, R.V. & Tikhomirov, A.S. Dynamics of orientation tuning in the cat striate cortex neurons. Neuroscience 56, 865–876 (1993). 13. Volgushev, M., Vidyasagar, T.R. & Pei, X. Dynamics of the orientation tuning of postsynaptic potentials in the cat visual cortex. Vis. Neurosci. 12, 621–628 (1995). 14. Ringach, D.L., Hawken, M.J. & Shapley, R. Dynamics of orientation tuning in macaque primary visual cortex. Nature 387, 281–284 (1997). 15. Chen, G., Dan, Y. & Li, C.Y. Stimulation of non-classical receptive field enhances orientation selectivity in the cat. J. Physiol. (Lond.) 564, 233–243 (2005). 16. Ringach, D.L., Hawken, M.J. & Shapley, R. Dynamics of orientation tuning in macaque V1: the role of global and tuned suppression. J. Neurophysiol. 90, 342–352 (2003). 17. Celebrini, S., Thorpe, S., Trotter, Y. & Imbert, M. Dynamics of orientation coding in area V1 of the awake primate. Vis. Neurosci. 10, 811–825 (1993).
18. Mazer, J.A., Vinje, W.E., McDermott, J., Schiller, P.H. & Gallant, J.L. Spatial frequency and orientation tuning dynamics in area V1. Proc. Natl. Acad. Sci. USA 99, 1645–1650 (2002). 19. Nelson, S.B. Temporal interactions in the cat visual system. I. Orientation-selective suppression in visual cortex. J. Neurosci. 11, 344–356 (1991). 20. Müller, J.R., Metha, A.B., Krauskopf, J. & Lennie, P. Rapid adaptation in visual cortex to the structure of images. Science 285, 1405–1408 (1999). 21. Felsen, G. et al. Dynamic modification of cortical orientation tuning mediated by recurrent connections. Neuron 36, 945–954 (2002). 22. Dragoi, V., Sharma, J., Miller, E.K. & Sur, M. Dynamics of neuronal sensitivity in visual cortex and local feature discrimination. Nat. Neurosci. 5, 883–891 (2002). 23. Duysens, J., Orban, G.A., Cremieux, J. & Maes, H. Visual cortical correlates of visible persistence. Vision Res. 25, 171–178 (1985). 24. Coltheart, M. Iconic memory and visible persistence. Percept. Psychophys. 27, 183–228 (1980). 25. Goldberg, J.A., Rokni, U. & Sompolinsky, H. Patterns of ongoing activity and the functional architecture of the primary visual cortex. Neuron 42, 489–500 (2004). 26. Ben-Yishai, R., Lev Bar Or, R. & Sompolinsky, H. Theory of orientation tuning in the visual cortex. Proc. Natl. Acad. Sci. USA 92, 3844–3848 (1995). 27. Kenet, T., Bibitchkov, D., Tsodyks, M., Grinvald, A. & Arieli, A. Spontaneously emerging cortical representations of visual attributes. Nature 425, 954–956 (2003). 28. Oram, M.W., Foldiak, P., Perrett, D.I. & Sengpiel, F. The ‘ideal homunculus’: decoding neural population signals. Trends Neurosci. 21, 259–265 (1998). 29. Salinas, E. & Abbott, L.F. Vector reconstruction from firing rates. J. Comput. Neurosci. 1, 89–107 (1994). 30. Seung, H.S. & Sompolinsky, H. Simple models for reading neuronal population codes. Proc. Natl. Acad. Sci. USA 90, 10749–10753 (1993). 31. Geisler, W.S. & Albrecht, D.G. Bayesian analysis of identification performance in monkey visual cortex: nonlinear mechanisms and stimulus certainty. Vision Res. 35, 2723–2730 (1995). 32. Normann, R.A., Maynard, E.M., Rousche, P.J. & Warren, D.J. A neural interface for a cortical vision prosthesis. Vision Res. 39, 2577–2587 (1999). 33. Grinvald, A. & Hildesheim, R. VSDI: a new era in functional imaging of cortical dynamics. Nat. Rev. Neurosci. 5, 874–885 (2004). 34. Petersen, C.C., Grinvald, A. & Sakmann, B. Spatiotemporal dynamics of sensory responses in layer 2/3 of rat barrel cortex measured in vivo by voltage-sensitive dye imaging combined with whole-cell voltage recordings and neuron reconstructions. J. Neurosci. 23, 1298–1309 (2003). 35. Ringach, D.L. & Malone, B.J. The operating point of the cortex: neurons as large deviation detectors. J. Neurosci. 27, 7673–7683 (2007). 36. Hübener, M. & Bonhoeffer, T. Optical imaging of functional architecture in cat primary visual cortex. in The Cat Primary Visual Cortex (eds Payne, B.R. & Peters, A.) 1–137 (Academic Press, New York, 2002). 37. Dean, A.F. & Tolhurst, D.J. Factors influencing the temporal phase of response to bar and grating stimuli for simple cells in the cat striate cortex. Exp. Brain Res. 62, 143–151 (1986). 38. Carandini, M. & Ferster, D. Membrane potential and firing rate in cat primary visual cortex. J. Neurosci. 20, 470–484 (2000). 39. Azouz, R. & Gray, C.M. Dynamic spike threshold reveals a mechanism for synaptic coincidence detection in cortical neurons in vivo. Proc. Natl. Acad. Sci. USA 97, 8110–8115 (2000). 40. Chichilnisky, E.J. A simple white noise analysis of neuronal light responses. Network 12, 199–213 (2001). 41. Xing, D., Shapley, R.M., Hawken, M.J. & Ringach, D.L. Effect of stimulus size on the dynamics of orientation selectivity in macaque V1. J. Neurophysiol. 94, 799–812 (2005). 42. Schummers, J. et al. Dynamics of orientation tuning in cat V1 neurons depend on the location within layers and orientation maps. Front. Neurosci. 1, 145–159 (2007). 43. Carandini, M. et al. Do we know what the early visual system does? J. Neurosci. 25, 10577–10597 (2005). 44. Touryan, J., Lau, B. & Dan, Y. Isolation of relevant visual features from random stimuli for cortical complex cells. J. Neurosci. 22, 10811–10818 (2002). 45. Tolhurst, D.J., Walker, N.S., Thompson, I.D. & Dean, A.F. Nonlinearities of temporal summation in neurones in area 17 of the cat. Exp. Brain Res. 38, 431–435 (1980). 46. McCormick, D.A. et al. Persistent cortical activity: mechanisms of generation and effects on neuronal excitability. Cereb. Cortex 13, 1219–1231 (2003). 47. Carandini, M. & Ringach, D.L. Predictions of a recurrent model of orientation selectivity. Vision Res. 37, 3061–3071 (1997). 48. Kovács, G., Vogels, R. & Orban, G.A. Cortical correlate of pattern backward masking. Proc. Natl. Acad. Sci. USA 92, 5587–5591 (1995). 49. Rolls, E.T. & Tovee, M.J. Processing speed in the cerebral cortex and the neurophysiology of visual masking. Proc. Biol. Sci. 257, 9–15 (1994). 50. Keysers, C., Xiao, D.K., Földiák, P. & Perrett, D.I. Out of sight but not out of mind: the neurophysiology of iconic memory in the superior temporal sulcus. Cogn. Neuropsychol. 22, 316–332 (2005).
1324
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
Methods Methods and any associated references are available in the online version of the paper at http://www.nature.com/natureneuroscience/.
© 2009 Nature America, Inc. All rights reserved.
Note: Supplementary information is available on the Nature Neuroscience website. Acknowledgments We thank I. Nauhaus, R.A. Frazor, L. Busse and S. Katzner for help with data acquisition. We thank W.T. Newsome, W.S. Geisler and G. Felsen for helpful discussions. This work was supported by a Scholar Award from the McKnight Endowment Fund for Neuroscience (M.C.) and by US National Institutes of Health grants EY017396 (M.C.) and EY018322 (D.L.R.). M.C. holds the GlaxoSmithKline/ Fight for Sight Chair in Visual Neuroscience. AUTHOR CONTRIBUTIONS A.B. and M.C. carried out the experiments, A.B. analyzed the data, and all of the authors contributed to the intellectual development of the project and to the writing of the manuscript. Published online at http://www.nature.com/natureneuroscience/. Reprints and permissions information is available online at http://www.nature.com/ reprintsandpermissions/.
ONLINE METHODS
© 2009 Nature America, Inc. All rights reserved.
Physiology. We used seven cats (eight hemispheres) for our VSD experiments and another four cats (six hemispheres) for the electrical recordings. Young adult cats (2–4 kg) were anesthetized first with ketamine (22 mg per kg of body weight) and xylazine (1.1 mg per kg) and then with sodium penthotal (0.5–2 mg per kg per h) and fentanyl (typically 10 µg per kg per h), supplemented with inhalation of N2O (typically 70% N2O and 30% O2). A 1-cm craniotomy was performed over area V1 (usually area 18, occasionally area 17). The eyes were treated with topical atropine and phenylephrine and protected with contact lenses. A neuromuscular blocker was given to prevent eye movements (pancuronium bromide, 0.15 mg per kg per h, intravenous). The cat was artificially respirated and received periodic doses of an antibiotic (cephazolin, 20 mg per kg, twice daily), an antiedematic steroid (dexamethasone, 0.4 mg per kg daily) and an anticholinergic agent (atropine sulfate, 0.05 mg per kg, i/m, daily). Fluid balance was maintained by intravenous infusion. The level of anesthesia was monitored through the electroencephalogram. Additional physiological parameters that were monitored included temperature, heart rate, end-tidal CO2 and lung pressure. Experiments typically lasted 48–96 h. These procedures were approved by the Institutional Animal Care and Use Committee of the Smith-Kettlewell Eye Research Institute. Stimuli. The stimulus consisted of full-field stationary gratings flashed in random sequence for 32 ms each14. The gratings had one of four spatial phases and one of six to twelve orientations. Their contrast was typically 50% and the spatial frequency was the optimal one, as assessed by a preliminary experiment (typically, 0.2 cycles per degree). Sequences were broken into 4–8 segments lasting 6 s each. Randomly interleaved with the gratings were blank frames (also lasting 32 ms), which occurred with a probability of about 30%. An additional 6-s control segment consisted entirely of blanks. Segments were presented in random order and each block of segments was generally presented ten times. Stimuli were viewed monocularly with the eye contralateral to the hemisphere being imaged.
In VSD imaging, signals of such low spatial frequency are dominated by noise. We removed this component from population responses because it reflects the activation of the entire region regardless of preferred orientation. Event-related analysis. To compute the average response to a generic oriented stimulus, we used-event related analysis and averaged the responses in a 200-ms window around the time of occurrence of a given oriented grating. This procedure was applied each time a grating was presented and the data sorted and averaged according to the stimulus orientation. Conditional average responses (second-order interactions) were computed with the same algorithm. Event-related analysis was performed on those conditions in which a stimulus was preceded by an identical stimulus or a stimulus with 45°, 90° or −45° orientation difference. Predicted population responses. As illustrated in Figure 3a,b, the linear predicted responses Lθ (t) are computed by convolving the average response to a stimulus Fθ (t) with the sequence of stimuli S(φ,t). Lq (t ) =
Untuned response component. The untuned component of the membrane potential and spike responses is the mean activity across preferred orientations.
doi:10.1038/nn.2398
S (j , t ) Fq −j (t − T ) dT
Percentage of variance. The percentage of variance for orientation jumps was computed as 2 ∑ Rt ,q − Pt ,q t ,q 1 − 2 ∑ Rt ,q − Rt ,q t ,q t ,q S1→ S 2
(
)
(
in our previous work10. We stained the cortex with the VSD RH-1692 and imaged
Array recordings. We implanted a 10 × 10 electrode array (0.4-mm separation and 1.5-mm electrode length) in the same patch of cortex. To minimize cortical damage, we inserted the arrays at high speeds (around 8 m s−1) using a pneumatic insertion device. Insertion depths were about 0.8–1 mm. As a result of the curvature of the cortex, the depth of penetration varied across sites. The array and surrounding tissue were covered in 1.5% agar to improve stability. Well-tuned multiunit activity was typically recorded from most of the 96 electrodes. Many electrodes also contributed well-isolated single-unit recordings. Traces were acquired at 12 kHz and firing rates were obtained by low-pass filtering the spike trains with a cutoff at 25 Hz.
∞
Here, the stimulus S(φ,t) is 1 if the orientation of the stimulus presented at time t was equal to φ and 0 otherwise, and θ indicates the preferred orientation of a subpopulation. Finally, the predicted responses Rθ (t) are computed by passing the linear prediction through a static nonlinearity f to account for the discrepancy between the linear model and prediction at high and low values of the responses: Rθ (t) = f (Lθ (t)). The static nonlinearity is fitted to the data so as to be optimal40.
Imaging. Methods for VSD imaging were developed previously9,33 and described its fluorescence in 15–30 mm2 of V1. The dye was circulated in a chamber over the cortex for 3 h and washed out with saline. We acquired images with a CMOS digital camera (1M60 Dalsa), as part of the Imager 3001 setup (Optical Imaging). Images were acquired at a frame rate of 110 Hz, with a spatial resolution of 0.028 mm per pixel. Additional spatial filtering was performed offline (bandpass, 0.2–2.2 cycles mm−1), except when measuring overall activation (Fig. 2a). Frame acquisition was synchronized with the respirator. Illumination from a 100-W halogen light was delivered through two optic fibers or via epi-illumination. Excitation and emission filters were bandpass, at 630 ± 10 nm and 665 ± 10 nm. Population responses were measured by grouping pixels in 24 bins according to their preferred orientation and averaging their responses to obtain a single data point for each preferred orientation bin. To analyze the VSD responses to orientation noise, we subtracted the response to the blank 6-s segment from the response to each 6-s segment; this subtraction removed artifacts resulting from respiration, which was synchronized with sequence onset. The resulting responses were then bandpass filtered between 1 and 25 Hz. We computed z scores by averaging the response across repeats and dividing responses by s.d. computed across repeats.
p
∫0 dj ∫0
)
where R and P are responses and prediction for a given orientation jump. The
∑
sums, t ,q , and the average, t ,q , are computed for time t between 50–150 ms after stimulus onset, an interval during which the response amplitude is significantly above the baseline noise. The average S1→ S 2 is done over all stimulus transitions in eight hemispheres. Bayesian decoder. To decode population responses, we computed firing rate over 8-ms time bins, and for each of these bins we estimated the probability that the stimulus had a certain orientation φ. Given a population response R = (r1, r2, …, rn) and a set of orientations {φ 1, φ 2, …, φ n} with equal probability of presentation p(φ), the probability of stimulus φ j is given by p (r1 j j ) p (r2 j j )... p (rn j j ) p (j j r1 , r2 ,..., rn ) = n ∑ p (r1 ji ) p (r2 ji )... p (rn ji ) i =1
Here, we make the simplifying assumption that p (ri j ) and p (rj j ) are independent. In reality they are not, but in separate analyses (data not shown) we computed the covariance matrix, included it in the model and found that doing so did not improve the quality of the decoding. For each response ri (i = 1, …n; n = 24), we computed the distribution p (ri j j ) as − 1 p (ri j j ) = e s 2p
(r (j j ) − ri )2 2s 2
nature NEUROSCIENCE
the average elemental response along the orientation axis. Only the time average of the elemental response (in a time interval centered on the peak time; Fig. 7a) was used to construct the tuning curves.
© 2009 Nature America, Inc. All rights reserved.
where r (j j ) is the mean across stimulus presentations of the response to the stimulus φ j and σ is the associated s.d. We obtained the n = 24 tuning curves used to derive the distributions p (ri j j ) (Fig. 7a,b) by doing 24 circular shifts of
nature NEUROSCIENCE
doi:10.1038/nn.2398
a r t ic l e s
Fragmentation of grid cell maps in a multicompartment environment
© 2009 Nature America, Inc. All rights reserved.
Dori Derdikman1, Jonathan R Whitlock1, Albert Tsao1, Marianne Fyhn1,2, Torkel Hafting1,2, May-Britt Moser1 & Edvard I Moser1 To determine whether entorhinal spatial representations are continuous or fragmented, we recorded neural activity in grid cells while rats ran through a stack of interconnected, zig-zagged compartments of equal shape and orientation (a hairpin maze). The distribution of spatial firing fields was markedly similar across all compartments in which running occurred in the same direction, implying that the grid representation was fragmented into repeating submaps. Activity at neighboring positions was least correlated at the transitions between different arms, indicating that the map split regularly at the turning points. We saw similar discontinuities among place cells in the hippocampus. No fragmentation was observed when the rats followed similar trajectories in the absence of internal walls, implying that stereotypic behavior alone cannot explain the compartmentalization. These results indicate that spatial environments are represented in entorhinal cortex and hippocampus as a mosaic of discrete submaps that correspond to the geometric structure of the space. Place cells in the hippocampus fire whenever animals pass through specific locations in the environment1–3. Neighboring cells have different firing fields (place fields) such that, as an ensemble, they form a distributed representation of the animal’s space4. All places in the local environment are represented at every level of the hippocampus4,5, suggesting that place cells are part of a Tolmanian cognitive map of the environment and the events that the animal experiences in it2. The hippocampal map contains a manifold of representations. Only a subset of the cells is active in each distinct environment4,6; when the animal is moved to a different environment, a new combination of cells is recruited through a process referred to as remapping 7–9. On top of each location signal, hippocampal cell ensembles convey information about nonspatial components of the experience, such as characteristic textures or odors, the passage of time, and motivational factors10–16. This multiplicity of the place cell representation indicates that the hippocampus is a storage site for individual place-related memories, but questions the idea of an intrahippocampal origin for the location signal itself17–19. The firing properties of several other cell types suggest that space is represented outside of the hippocampus. Head direction cells, for example, have been found in a number of brain structures20, first in the dorsal presubiculum21,22 and most recently in the medial ento rhinal cortex (MEC)23. Grid cells have been described in the MEC24,25 at the interface between head direction cells and place cells. Grid cells have multiple regularly spaced firing fields that, for each cell, form a hexagonal grid spanning the entire open space available to the animal. The directional preferences of different head direction cells26–29 and the spatial relationship between the firing fields of different grid
cells30 remain constant across environments, suggesting that these cell types, unlike place cells, are part of a general metric representation of space31. Since their discovery, grid cells have primarily been recorded in environments with no internal boundaries, such as the open field. Realistic environments, however, are more complex and often consist of multiple nested subenvironments. The manner in which partitioned environments are mapped by grid cells has not been established. Is the continuous periodic firing structure retained or is the representation fragmented into multiple maps in accordance with the boundaries of the local subenvironments? To distinguish between these possibilities, we constructed a multicompartment environment consisting of a zig-zagged stack of interconnected maze segments with equal shape and orientation (the hairpin maze). Grid cells and place cells were recorded while the rat ran through the maze. We found that the representation broke into discontinuous segments in both MEC and hippocampus. The fragmentation occurred abruptly near the point of turning from one compartment to the next, implying that grid cells and place cells formed discrete submaps for each part of the environment. RESULTS The rats were first trained to run randomly in an open-field, 1.5- × 1.5-m box. When the rats had covered the entire arena repeatedly across trials, they were trained to run in a multicompartment hairpin maze constructed from Perspex walls inserted into the open-field box. The rats ran westbound and eastbound in an alternating manner. Daily sessions consisted of a 20-min trial in the open field, two 20-min trials in the hairpin maze and a second 20-min trial in the open field (Fig. 1a).
1Kavli
Institute for Systems Neuroscience and Centre for the Biology of Memory, MTFS, Norwegian University of Science and Technology, Trondheim, Norway. address: Department of Physiology, University of California San Francisco, San Francisco, California, USA. Correspondence should be addressed to E.I.M. (
[email protected]).
2Present
Received 2 June; accepted 3 August; published online 13 September 2009; doi:10.1038/nn.2396
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1325
a r t ic l e s a
b
10 Hz
20 Hz
29 Hz
12 Hz
9 Hz
26 Hz
23 Hz
5 Hz
14 Hz
17 Hz
12 Hz
17 Hz
© 2009 Nature America, Inc. All rights reserved.
c
d
Figure 1 Grid fields repeat across arms with similar running directions. (a) Experimental protocol. The rat ran for 20 min in the open field, followed by two 20-min runs in the hairpin maze and another 20-min run in the open field. Between runs, the rat rested in its home cage or on a towel in a flower pot next to the maze. Horizontal bar indicates cue card. Right is east, left is west, top is north and bottom is south. (b–d) Three different cells are shown, one in each row (b, layer II/III; c, layer III; d, layer III). Alternating columns show trajectories with individual spike locations and color-coded rate maps. Trajectories are black and spike locations red. Walls are marked in green. The color code in the rate maps is from blue (silent) to red (peak rate), with the color scale maximum indicated beneath the rate map. The left pair of columns shows the first trial in the open field, the next two pairs show eastbound and westbound trajectories (arrows indicate running direction), respectively, in the hairpin maze, the right pair shows the second trial in the open field. Although clear grid fields were apparent in the open field, the grid broke up in the hairpin maze. Note the repetitive firing pattern across arms with similar running directions.
Grid maps repeat across alleys of the hairpin maze To characterize the responses of single cells in the hairpin maze, we divided the path into eastbound and westbound trajectories, which separated opposite running directions in individual arms of the maze. Grid cells had sharply defined firing fields in each arm, but the fields of different arms were not arranged as a two-dimensional master grid overlaid across the internal walls of the maze (Fig. 1b–d). Of the 105 cells, 81 had gridness scores below the criterion for grid cells; the average score in the hairpin maze was −0.14 ± 0.02, which was significantly lower than that in the open field (P < 0.001, z = 12.1, Wilcoxon rank sum test; scores are negative because of 90° rotational symmetry in the hairpin maze). The correlation with a perfect grid template was also lower in the hairpin maze than in the open field (r = 0.04 ± 0.16, P < 0.001, z = 10.9).
Although the two-dimensional periodicity of the grid was lost, the positions of the firing fields were highly correlated across arms with similar running directions (all even-numbered arms or all oddnumbered arms). Each field was a similar distance from the north or south walls in different alleys, with occasional exceptions in the end arms (Figs. 2 and 3). Arms with opposite running directions (even versus odd numbered) generally had fields at different locations. When multiple grid cells were recorded simultaneously (six grid cells; Fig. 2), we constructed a population vector and visualized spatial correlations between pairs of arms in a correlation matrix. The matrices showed a clear checkboard pattern (Fig. 2c,d), reflecting strong correlations between arms with similar running directions and weak correlations between arms with opposite running directions. A similar pattern was seen when population vectors were constructed for all cells recorded on different occasions from the same rat (29 grid cells; Fig. 3a,b) or for the entire sample of 105 grid cells from all rats (Fig. 3c,d). In the complete sample, correlations between firing fields on arms with similar running directions (median r = 0.58; Fig. 3e) were significantly higher than for arms with opposite running directions (median r = –0.02, z = 11.6, P < 0.001, Wilcoxon two-sided rank sum test; Fig. 3f) and differed significantly from a shuffled distribution (z = 12.3, P < 0.001; Fig. 3g). Limiting the analyses to the subset of cells with no head-direction modulation (n = 73) did not change the pattern of correlations (median r = 0.67 versus 0.09). Firing locations on alternating arms were not mirror images of one another (z = 11.6, P < 0.001; Fig. 3h). There was no
1326
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
Recording electrodes were placed in layers II, III or V of MEC (Supplementary Fig. 1). A total of 105 well-separated grid cells were recorded in 16 rats. The average gridness score of these cells was 0.57 ± 0.03 (mean ± s.e.m., gridness measures the 60° rotational symmetry of the spatial autocorrelation maps23; all cells had positive gridness scores). The average spatial correlation with a perfect grid template was 0.47 ± 0.02. The mean distance between two adjacent grid fields (estimated from the autocorrelation map) was 51.5 ± 1.6 cm. We recorded head direction in 94 grid cells; 21 of these showed weakto-moderate directional modulation (half of the spikes were in an arc of 135° or less, see Online Methods).
a r t ic l e s a
Figure 2 Population analysis for a single-cell ensemble. (a) A sample of six simultaneously recorded grid cells (one cell per column) during westbound and eastbound trajectories (top and bottom row, respectively). All six cells showed 20 Hz 20 Hz 20 Hz 20 Hz 20 Hz 20 Hz repetition of spatial firing patterns across alternating arms with similar running directions in both the westbound and the eastbound directions. All maps were clipped to the same rate range of 0–20 Hz. (b) A population vector 20 Hz 20 Hz 20 Hz 20 Hz 20 Hz 20 Hz was constructed from the concatenated rates in r each arm. A correlation between every pair of 1.0 such vectors (arms 3 and 7 in the schematic) 2 2 was calculated. For each pair, the rates in every 0.5 15-cm bin were multiplied with the rates in 4 4 the corresponding bin in the other arm (black 0 6 6 horizontal lines in the schematic). (c,d) Matrix showing the correlations between each pair –0.5 8 8 of population vectors. Arms are numbered from 1 at the west side to 10 at the east side. –1.0 10 10 Correlation (r) is color coded as indicated in 2 4 6 8 10 2 4 6 8 10 the scale bar. The correlation between the Arm number Arm number 3 7 population vectors constructed for arm 3 versus arm 7 in the schematic in b is outlined in black. In c, the correlation matrix is constructed for eastbound trajectories and is constructed for westbound trajectories in d (arrows). The checkboard pattern in both matrices indicates a high correlation between arms in which the rat was running in the same direction, but not between arms with opposite directions.
c
d
Arm number
repeated as the rat ran down individual alleys. We defined population vectors for concurrently recorded grid cells for each 3-cm segment of the maze arm. Autocorrelation of the population vectors showed clearly repetitive ensemble activity in many cases (Supplementary Fig. 2). Linear spatial autocorrelations for individual units always showed at least one peak in addition to the central peak in the autocorrelation function. In 97 out of 105 recordings, there was also a third peak, suggesting that a clear majority of the cells had periodic fields in the alleys. The occasional lack of repetition may reflect strong
c orrelation between arms on eastbound runs and arms with the same running direction on westbound runs (median r = 0.08). Taken together, these results indicate that the grid representation was discontinuous between the different compartments of the hairpin maze and that highly similar sequences of firing fields were unfolded in arms with similar running directions. To determine whether the periodic firing structure from the open field was retained in each compartment of the hairpin maze, we asked whether sequences of firing among simultaneously recorded cells were
Arm number
a
b
c
2
2
2
2
4
4
4
4
6
6
6
6
8
8
8
8
10
10 2
e Frequency
© 2009 Nature America, Inc. All rights reserved.
Arm number
b
25
4 6 8 Arm number
Same direction
10
f
25
20
20
15
15
10
10
5
5
0 –1.0
0 –1.0
–0.5 0 0.5 Correlation (r )
1.0
10 2
4 6 8 Arm number
Different direction
g
10
4 6 8 Arm number
Shuffled
0
–0.5
25
2
Reflected
5
0 1.0 –1.0
0 1.0 –1.0
–0.5 0 0.5 Correlation (r )
30
–1.0
West versus east
15
10
10
i
10
20
15
20
4 6 8 Arm number
25
20
30
–0.5 0 0.5 Correlation (r )
h
10
1.0
0.5
10 2
40
r
d
10 5 –0.5 0 0.5 Correlation (r )
0 1.0 –1.0
–0.5 0 0.5 Correlation (r )
1.0
Figure 3 Population analysis for all trials and all rats. (a,b) Correlation matrix for population vectors constructed from all of the cells recorded from one rat. The construction is similar to the one in Figure 2. Both westbound (a) and eastbound (b) trajectories are shown. (c,d) Population vector constructed from all of the cells recorded from all of the rats. Both westbound (c) and eastbound (d) trajectories are shown. Checkboard patterns demonstrate high correlation between all pairs of arms with similar running directions (as in Fig. 2) for the whole population. (e) Frequency distribution showing the mean correlations between arms where the rats ran in the same direction (every second arm, excluding arms 1 and 10) for all cells. The correlations are skewed toward 1. (f) Frequency distribution showing the mean correlations between arms in which the rat ran in opposite directions (excluding arms 1 and 10). The correlations between arms with opposite running directions were skewed. (g) Data are presented as in e, but the rates of the maze arms are shuffled randomly. (h) Data are presented as in f, but with the rate maps of the odd-numbered arms reflected in the north-south direction. (i) There was a lack of correlation between eastbound and westbound trajectories.
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1327
a r t ic l e s Running direction
r
b
Running direction
0.8 0.6 0.4 0.2
Diagonal correlation value (r )
a
–0.2
0.7 0.6 0.5 0.4 0.3
0
120 cm
0.8
–100
–50 0 50 Distance from turnpoint (cm)
c 250 200
Figure 4 Representations were reset near the turning points. (a) Spatial correlation matrix 150 based on population vectors of firing rates at the turning point and in 6-cm bins on both sides of 100 the turning point. Directionally tuned grid cells were not included in the population vectors. 50 Every pair of positions was correlated. Black arrows indicate running direction. The crossing 0 of the white horizontal and vertical lines marks 0.4 0.5 0.6 0.7 0.8 the turning point. The values around the central Correlation value around turnpoint diagonal are high because bins near each other have similar rates. The diagonal elements that are one bin away from the central diagonal are marked in the plot by two diagonal white lines; these values are the correlations between adjacent bins in the population vector. The white arrow points at the minimal value in this band, at the turn. (b) Correlation between adjacent position bins taken from the diagonal section marked in a. Grey shading indicates s.e.m. Note the sharp drop in spatial correlation before the turning point. (c) Distribution of correlations in 1,000 Monte-Carlo runs in which the trigger point on the trajectory was randomly shifted for each cell. The correlation value for the turning point in the observed data is indicated by a vertical red line. Note that all shuffled values are higher than the observed value. Frequency
© 2009 Nature America, Inc. All rights reserved.
100
rate differences between peaks of the grid, nonperfect alignment of the grid axes with the direction of the alley31 or, in the case of the population vectors, difference in the periodicity between the cells in the same session. For each unit, the interpeak distance, estimated from the central peak to the next peak in the linear autocorrelation function, correlated with the spacing of the grid in the open field (r = 0.46; P < 0.001; Supplementary Fig. 2). Resetting occurs at turning points If the rat has separate representations for each arm of the maze, we would expect that there should be a strong correlation in rate between adjacent positions when the rat is inside an arm, whereas the correlation between adjacent positions should drop when the rat is turning from one arm to the next32. Turning points were defined as the positions with maximum curvature at the distal ends of the alleys, a normalized rate population vector was constructed for the turning point and for the positions before and after it, and the population vectors for all possible position pairs were correlated with each other. Directionally modulated grid cells23 were not included. The analysis showed that the mean value of correlation between adjacent pixels (6 cm apart) was lower at the turning point than at any other position in the maze (Fig. 4). The estimated value (r = 0.44) was significantly below the chance level estimated by a shuffling Monte-Carlo procedure in which the trigger point was randomly shifted for each cell (the value was lower than the minimal value of 1,000 Monte-Carlo runs; Monte-Carlo mean = 0.62; Fig. 4c). There was no significant drop in the correlation around the turning point when the spikes were shifted 2 s or 20 s forward in time (P > 0.9 in both 1328
cases, Monte Carlo). The correlation between adjacent pixels was low for south turns (r = 0.51 compared with a Monte-Carlo mean of r = 0.66; the recorded value was smaller than the ninth smallest Monte-Carlo result out of 1,000 runs) and for north turns (r = 0.41 compared with a Monte-Carlo mean of r = 0.59; the recorded value was smaller than the third smallest Monte-Carlo result out of 1,000 runs; Supplementary Fig. 3). A similar drop in correlations was observed when values at the turning point were calculated separately for each rat with more than four cells (mean ± s.e.m., r = 0.48 ± 0.05). Taken together, these analyses indicate that there is a clear discontinuity in the grid representation at places at which the rat started curving into a new compartment.
Representations anchor to the preceding turning point Having found that the representation is reset when the rat turns into a new arm, we asked whether the representation in the alley is anchored to the preceding or upcoming turning point. We recorded 25 cells from five rats as they were performing a short-cut version of the hairpin task in which two arms were truncated. Inspection of individual rate maps showed that the firing locations on the short arms depended on the distance from the previous turn in most, but not all, of the cases (Fig. 5a–d). The extent to which representations in the alleys were aligned to preceding and upcoming turning points was quantified by a spatial crosscorrelation procedure in which the rate distribution in the truncated arm was compared with the rate distribution in an equal number of position bins in the adjacent nontruncated arms with the same running direction (reference arms). The truncated arm was shifted relative to each of the reference arms in steps of 10 cm. At shift = 0, the reference vectors were anchored to the wall behind the rat (Fig. 5e), and at shift = 50 cm, the reference vectors were aligned to the wall ahead of it (Fig. 5e). The correlations decreased significantly as the reference vectors moved away from the wall behind the rat (Kruskal-Wallis test, F5,144 = 45.01, P < 0.001; Fig. 5e,f). Post hoc pair-wise comparisons confirmed that the vectors correlated more significantly (P < 0.05) with smaller relative shifts (0–10 cm) than with larger shifts (30–50 cm). An exception to this pattern was seen at the end of the short alley, where the representation correlated more with reference vectors aligned to the upcoming than to the preceding wall (Fig. 5f). Consistent with this, curve-fit analyses showed that more variance was explained with a quadratic function than with a linear function (31.4 versus 17.0%; Fig. 5e). Collectively, these observations suggest that a self motion–based mechanism33 is involved in determining firing location in the alley, but also suggest that there is some alignment to the upcoming wall as the rat approaches the next turning point. Repetitive firing in place cells To determine whether the discontinuity of the entorhinal representation is accompanied by a similar break-up in the hippocampal VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
a r t ic l e s c
b
6 Hz
6 Hz
d
25 Hz
e Correlation (r )
a
13 Hz
© 2009 Nature America, Inc. All rights reserved.
5 Hz
6 Hz
25 Hz
10 Hz
Distance from start of short arm (cm)
f
0.6 0.5 0.4 0.3 0.2 0.1 0 –0.1
0
10
20 30 40 Shift (cm)
50
0 20
1.0 0.5
40
0
60
–0.5
80 100
r
0
50 100 150 Distance from start of long arm (cm)
–1.0
Figure 5 Shortcut experiments suggest a path-integration mechanism. (a–d) Rate maps of four cells (one cell per panel) recorded in a shortcut version of the hairpin maze (top row, westbound paths; bottom row, eastbound paths). Note the alignment to the preceding wall in a and b; the firing field in the south-to-north direction on the truncated arm on westbound runs in a, for example, had the same distance from the turning point behind the rat as the corresponding fields on the nontruncated arms. Likewise, the cell in c showed alignment to the preceding turning point in the north, but the turning point and the firing fields were shifted relative to the other arms. The cell in d deviated from the general pattern in that the fields were aligned to the outer wall (south) regardless of whether the rat was approaching it or running away from it. (e) Spatial crosscorrelation of rate vectors constructed from all cells for the truncated arm versus rate vectors for a similar number of bins in the corresponding reference arms. The truncated arm is offset relative to the reference arms at steps of 10 cm from 0 (alignment to preceding wall, left schematic) to 50 cm (alignment to approaching wall, right schematic). Spatial correlations are provided for each offset (mean ± s.e.m.). Arrows in the schematic indicate running direction. Red and blue lines indicate position bins that are compared between the truncated arm and the reference arms (red, rat runs toward the inserted barrier; blue, rat runs away from the inserted barrier; left, shift = 0; right, shift = 50 cm). Note that the best correlation was achieved with shift = 0, suggesting that the representation is preferentially anchored to the turning point behind the rat. (f) Correlation between activity in bins of the short arm with the mean values for bins in the long arms, shown as a function of the distance from the beginning of the arm (0 = the side from which the rat is running from). During most of the trajectory, the alignment was to the preceding wall (continuous line), although the alignment tended to shift to the upcoming wall (stippled line) toward the end. A similar behavior has been reported for hippocampal place cells 43,44.
Realignment is imposed by geometry of the environment We then sought to identify the factors that induce resetting of the grid map. We first asked whether a less repetitive pattern would develop if the rat could see the entire test box when it was running in the hairpin maze (Supplementary Fig. 5). Three rats were trained with transparent internal walls and the firing patterns of 18 cells were recorded in this condition and compared with those of 87 cells from
13 rats trained with opaque walls. The firing fields on arms with similar running directions were significantly correlated irrespective of the type of wall (transparent, z = 3.72; opaque, z = 8.10; P < 0.001 in both cases; Supplementary Fig. 5), although the correlation was numerically lower with the transparent walls than with opaque walls (medians of 0.45 versus 0.59, z = 1.81, P = 0.07, Wilcoxon rank sum test). The latter difference mostly reflected the low correlation values of one rat (median r = 0.16; Supplementary Fig. 5). There was no indication of a master grid spanning the alleys of the maze in any of the rats, even if the whole environment was visible, suggesting that visual access to the distal environment does not prevent grid realignment between the compartments. Next, we asked whether the resetting was caused by the constraints on the rat’s behavior imposed by the walls of the hairpin task. Three rats were tested in a ‘virtual hairpin’ task in which the rats ran in stereotypic laps in the absence of internal walls (see ref. 15 for a conceptually similar task). Using the same open field as was used in the pretraining stage, we trained rats to run in a north-south zig-zag pattern resembling the stereotypic paths of the hairpin maze (Fig. 7a and Supplementary Fig. 6) to collect food morsels that were delivered, one at a time, at the north and south walls and at increasing distances from the west or east corner. In all of the rats, the firing fields formed a regular two-dimensional hexagonal array, much like they did under random foraging conditions in the same open field (Fig. 7a). In general, the grid cells maintained their spatial phase as the open field was converted to a virtual hairpin task (spatial correlation: 0.49 ± 0.03, P < 0.001, n = 28 cells; Fig. 7b). The spatial phase was not maintained in the regular hairpin condition (spatial correlation, 0.004 ± 0.015). The correlation between the rate map for the recorded data on one hand and a perfect grid template on the other was lower in the virtual hairpin condition than in the open field, but it was
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1329
representation, we recorded from 111 place cells (four rats) in CA3 (Fig. 6) and 47 place cells (three rats) in CA1 (Supplementary Fig. 4). We excluded 38 place cells from analysis because they responded only in the open field and 23 cells because they fired only at the end arms in the hairpin maze. The remaining 97 cells had place fields along the alleys. As with grid cells, we correlated the rate maps for arms with similar running directions with each other. The correlations were skewed toward 1 in both CA3 (median r = 0.72; Fig. 6e) and CA1 (median r = 0.70; Supplementary Fig. 4). Correlations between arms with opposite running directions were low (CA3: median r = 0.14, z = 4.85, P < 0.001; CA1: median r = –0.02, z = 4.52, P < 0.001; Wilcoxon rank sum test for same direction versus different direction; Fig. 6f and Supplementary Fig. 4). In one experiment, cells were recorded simultaneously from MEC (5 cells) and CA1 (16 cells). We constructed population vectors for each maze arm in this experiment. A checkboard-like correlation pattern was observed not only in MEC (Fig. 6g), but also in CA1 (median correlations for MEC: 0.77 and –0.07 for similar and different directions, respectively; medians for CA1: 0.81 and 0.26; Fig. 6h). Thus, the fragmentation of the grid representation was accompanied by a similar discontinuity in the place cell representation, suggesting that there is strong coherence between spatial maps in MEC and the hippocampus.
a r t ic l e s
Frequency
b
43 Hz
Different direction 6 5 4 3 2 1 0 –1.0 –0.5 0 0.5 Correlation (r)
1.0
c
38 Hz
38 Hz
Arm number
g 2
2
4
4
6
6
8
8 2
4
6
8 10
h
significantly higher than 0 (r = 0.20 ± 0.04; d z = 3.57, P < 0.001, Wilcoxon rank sum test; Fig. 7c) and was significantly higher than in the regular hairpin, where the correlation was r = 0.04 ± 0.03 (z = 2.67, P = 0.008, Wilcoxon rank sum test). To identify the reason for the 2 Hz slight drop in the correlation with a grid template in the virtual hairpin task, we calculated grid template correlations in the open field when spikes were randomly drawn from the open field distribution exclusively along superimposed trajectories from the virtual hairpin and hairpin tasks. Under these restricted sampling conditions, we observed a similar drop to that seen in the open field (Fig. 7 and Supplementary Fig. 7), suggesting that the grid-template correlation is vulnerable to the homogeneity of the path, such that, in cases in which the path ‘misses’ a grid point, it has a moderate effect on the correlation with the grid template, whereas the spatial correlation measure is not affected as much. Taken together, our observations in the virtual hairpin task indicate that the physical structure of the environment (open versus compartmentalized) exerts a strong influence on whether and where resetting takes place; sharp turns and constraints on locomotor direction alone are not sufficient to generate discontinuity in the grid map. Finally, we asked whether the repetitive firing pattern in the hairpin task was influenced by the food protocol used to motivate running in the maze (Supplementary Fig. 5). Food was given either at the south end of every alley (n = 52 cells in six rats) or only at the end of the last arm (n = 53 cells in 14 rats). In both cases, the spatial correlations between arms with similar running directions were high (r = 0.54 for food at each arm and r = 0.62 for food at end; Supplementary Fig. 5), suggesting that the spatial map was less likely to be reset by motivational and behavioral factors than by changes in the local geometry of the environment.
29 Hz
r
1.0 0.5
10
10
29 Hz
© 2009 Nature America, Inc. All rights reserved.
43 Hz
f
Same direction 14 12 10 8 6 4 2 0 –1.0 –0.5 0 0.5 1.0 Correlation (r )
Frequency
e
a
Arm number
Figure 6 Firing pattern of hippocampal place cells in the hairpin maze. (a–d) Representative firing fields of CA3 place cells in the hairpin maze. Symbols are listed in Figure 1. (e) Frequency distributions showing correlations of arms with similar running directions. There is a clear skew toward high correlation between alternating arms. (f) Frequency distributions showing low correlations between arms with opposite running directions. The few cases with high correlation can be explained by fields responding at the north or south ends of the arms, such as in c. (g,h) Simultaneous recording of five grid cells (g) and 16 place cells from CA1 (h). Population vector correlations are presented as in Figure 3c,d. Scale bar indicates correlation value. The clear checkboard pattern indicates that the representations repeat across arms with similar running directions in both MEC and hippocampus.
2
4
6
8 10
0 –0.5
2
2
4
4
6
6
8
8
10
10
2
4 6 8 10 Arm number
–
2 4 6 8 Arm number
–1.0
10
2 Hz
patterns were observed in place cells, suggesting that entorhinal and hippocampal representations reset coherently.
DISCUSSION Our main finding was that grid cells form a discrete spatial representation for each subenvironment when an open environment is divided into multiple compartments. There was no master grid spanning across the alleys of the hairpin maze; instead the representation reset sharply each time the rat turned from one compartment to the next. In each alley, firing fields were usually spaced periodically, suggesting that each compartment had its own grid map. Firing fields were aligned primarily to the preceding turning point, resulting in a highly repetitive firing pattern across arms of the maze. Similar repetitive
Fragmentation of the spatial representation Our results suggest that spatial maps of grid cells and place cells form a mosaic of connected submaps. Each time the rat turns from one compartment to another in the hairpin maze, the spatial represent ation is reset and a new ensemble sequence is unfolded, despite the continuity of the maze space and the rat’s motion through it. The fragmentation of the grid map in the hairpin maze is reminiscent of the firing patterns recorded from cells in the deep layers of more intermediate-to-ventral parts of the entorhinal cortex of rats running on ш- or U-shaped tracks14. These cells were found to fire at spatially equivalent positions on the rat’s trajectory, for example, before each of the left turns on the way out of the left and middle alleys in the ш maze. The ‘path equivalence’ of entorhinal cells in that study resembles the repetitive firing fields that we observed in the hairpin maze and can, in retrospect, be interpreted as a resetting of grid representations at points at which the rat moves into a new subcompartment. Repetitive firing sequences were not only seen in MEC, but also in the hippocampus. The fact that place cells showed similar stereotypic sequences implies that the entire entorhinal-hippocampal representation is reset at the turning points. The discontinuous nature of the representation is consistent with the firing properties of place cells in rats that are running on linear tracks. Decades of investigation have demonstrated that, with standard training protocols, place cells form uncorrelated maps for alternating running directions on a linear track34. A similar lack of correlation can be observed in grid cells35. It has remained unclear whether these directionally specific representations are discrete, with a break at each end of the track, or part of a continuous map with loops at the turning points36. Our data from the hairpin maze are consistent with the former possibility. The reduced crosscorrelations at the turning points and the repetition of the firing maps across the alleys suggest that spatial maps may be composed of multiple discontinuous segments even when the space and movement through the space are continuous.
1330
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
a r t ic l e s OF1
HP
VH
OF2
18 Hz
18 Hz
25 Hz
20 Hz
10 Hz
15 Hz
10 Hz
4 Hz
10 Hz
10 Hz
15 Hz
4 Hz
b
c 0.5
Correlation with grid template
Figure 7 Preserved two-dimensional grid representations in a virtual hairpin 0.7 maze. Rats were trained to alternate between successive reward positions at 0.4 0.6 the north and south walls of an open field without constraining side walls. 0.5 (a) Rate maps for three representative grid cells (one per row). Alternating 0.3 columns show trajectories with individual spike locations and color-coded 0.4 rate maps (as in Fig. 1). The left pair of columns shows the first trial in the 0.2 0.3 open field (OF1), the second pair shows a trial in the regular hairpin maze 0.2 (HP), the third pair is a recording from the virtual hairpin task (VH) and the 0.1 0.1 fourth pair, to the right, shows another trial in the open field (OF2). Note that continuous two-dimensional grid fields were apparent in the virtual hairpin 0 0 HP VH OF2 HP’ VH’ OF1 HP VH OF2 HP’ VH’ task and the open field and that the firing fields under these conditions were Training condition Training condition highly correlated. In contrast, the representation in the regular hairpin maze consisted of multiple identical segments, corresponding to the individual alleys of the maze. The firing locations in this condition do not match those of the OF and VH tasks. (b) Spatial correlation between rate maps of the first open field, the hairpin task, the virtual hairpin and the second open field. Error bars are s.e.m. (c) Average correlation with a perfect grid for each of the trial types. In b and c, we also simulated the firing rates in the hairpin condition (HP’, non-filled bars) and in the virtual hairpin condition (VH’, non-filled bars) using the paths from those conditions and the rate map from the second open-field trial. The correlation with a perfect grid (c) dropped in the simulated HP’ and VH’ conditions relative to the real OF2, suggesting that the reduction of those measures in the real VH task is caused by incomplete sampling when the path is bidirectional only. Correlation with OF1
© 2009 Nature America, Inc. All rights reserved.
a
Geometric and idiothetic determinants The entorhinal-hippocampal map was reset whenever the rat turned into a new alley of the maze. The fact that such discontinuities were absent in the virtual hairpin task suggests that the realignments were imposed by the physical structure of the task. Constraints on locomotor direction were not by themselves sufficient to generate discontinuity; breaks in the grid map were only seen when the box was divided into visibly and tactilely distinct compartments. Individual landmarks such as cue card and position of the experimenter were probably less important, considering that the spatial structure of the firing repeated across alleys despite the changing position relative to individual cues. The implied sensitivity to geometric shape is consistent with the marked effect of the shape of the recording box on the
firing locations of place cells37 and grid cells38 in open field tasks. The influence of geometry is also consistent with behavioral studies pointing to arena shape as a major determinant of the rat’s behavior39 and with studies suggesting that this influence depends on the hippocampus40–42. Together with those studies, our observations suggest that compartment geometry is a major determinant of discontinuities in the entorhinal-hippocampal spatial representation. Although compartment geometry exerts a strong influence on resetting in the grid cell population, our observations do not exclude a role for path integration in determining the firing locations after the reset has taken place. Place cells43,44 and grid cells25 are strongly controlled by internal movement signals (idiothetic cues). Path integration on the basis of such signals provides a metric of the brain’s representation of space, but has the drawback that the mechanism accumulates error and needs to reset itself from time to time to register with external landmarks33,45–48. Our experiments in the shortcut version of the hairpin maze suggest that, for the first few seconds after the rat turns into a new alley, the position of firing, both in place cells and grid cells, is influenced by how far the rat has run into the arm. Later, as the rat approaches the next turning point, the representation becomes more similar to those at the ends of the long alleys, raising the possibility that the rat switches from a behavior-determined (path integration based) representation to an external cue–based representation as it
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1331
Grid representations in naturalistic environments with multiple proximal landmarks and confined spaces may be more similar to the multiple discontinuous maps of the hairpin maze than the continuous grid map of the open field. If fragmentation is the rule rather than the exception, only smaller spaces would be covered by continuous hexagonal arrays in the rat’s natural habitat. The mosaic structure of the entorhinal-hippocampal representation may extend to episodic memories, which are also fragmented, despite the continuity of each individual sequence stored in the system.
a r t ic l e s traverses the alley, in much the same way that place cell representations switch on tracks with variable starting positions43,44. However, the resetting as such is probably more dependent on nonlocomotor factors such as the physical structure of the environment. The recent description of border cells in MEC49,50 suggests a possible role for such cells in resetting of the path integrator, as subgroups of these cells may fire just at the time when the realignment should take place.
1. O’Keefe, J. & Dostrovsky, J. The hippocampus as a spatial map. Preliminary evidence from unit activity in the freely-moving rat. Brain Res. 34, 171–175 (1971). 2. O’Keefe, J. & Nadel, L. The Hippocampus as a Cognitive Map (Oxford University Press, New York, 1978). 3. Muller, R.U., Kubie, J.L. & Ranck, J.B. Jr. Spatial firing patterns of hippocampal complex-spike cells in a fixed environment. J. Neurosci. 7, 1935–1950 (1987). 4. Wilson, M.A. & McNaughton, B.L. Dynamics of the hippocampal ensemble code for space. Science 261, 1055–1058 (1993). 5. Kjelstrup, K.B. et al. Finite scale of spatial representation in the hippocampus. Science 321, 140–143 (2008). 6. Leutgeb, S., Leutgeb, J.K., Treves, A., Moser, M.B. & Moser, E.I. Distinct ensemble codes in hippocampal areas CA3 and CA1. Science 305, 1295–1298 (2004). 7. Bostock, E., Muller, R.U. & Kubie, J.L. Experience-dependent modifications of hippocampal place cell firing. Hippocampus 1, 193–205 (1991). 8. Colgin, L.L., Moser, E.I. & Moser, M.B. Understanding memory through hippocampal remapping. Trends Neurosci. 31, 469–477 (2008). 9. Muller, R.U. & Kubie, J.L. The effects of changes in the environment on the spatial firing of hippocampal complex-spike cells. J. Neurosci. 7, 1951–1968 (1987). 10. Wood, E.R., Dudchenko, P.A. & Eichenbaum, H. The global record of memory in hippocampal neuronal activity. Nature 397, 613–616 (1999). 11. Pastalkova, E., Itskov, V., Amarasingham, A. & Buzsaki, G. Internally generated cell assembly sequences in the rat hippocampus. Science 321, 1322–1327 (2008). 12. Hampson, R.E., Heyser, C.J. & Deadwyler, S.A. Hippocampal cell firing correlates of delayed-match-to-sample performance in the rat. Behav. Neurosci. 107, 715–739 (1993). 13. Wood, E.R., Dudchenko, P.A., Robitsek, R.J. & Eichenbaum, H. Hippocampal neurons encode information about different types of memory episodes occurring in the same location. Neuron 27, 623–633 (2000). 14. Frank, L.M., Brown, E.N. & Wilson, M. Trajectory encoding in the hippocampus and entorhinal cortex. Neuron 27, 169–178 (2000). 15. Markus, E.J. et al. Interactions between location and task affect the spatial and directional firing of hippocampal neurons. J. Neurosci. 15, 7079–7094 (1995). 16. Young, B.J., Fox, G.D. & Eichenbaum, H. Correlates of hippocampal complex-spike cell activity in rats performing a nonspatial radial maze task. J. Neurosci. 14, 6553–6563 (1994). 17. Touretzky, D.S. & Redish, A.D. Theory of rodent navigation based on interacting representations of space. Hippocampus 6, 247–270 (1996). 18. Eichenbaum, H., Dudchenko, P., Wood, E., Shapiro, M. & Tanila, H. The hippocampus, memory and place cells: is it spatial memory or a memory space? Neuron 23, 209–226 (1999).
19. Sharp, P.E. 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 (1997). 20. Taube, J.S. The head direction signal: origins and sensory-motor integration. Annu. Rev. Neurosci. 30, 181–207 (2007). 21. Taube, J.S., Muller, R.U. & Ranck, J.B. Jr. Head-direction cells recorded from the postsubiculum in freely moving rats. I. Description and quantitative analysis. J. Neurosci. 10, 420–435 (1990). 22. Ranck, J.B. Head direction cells in the deep cell layer of dorsal presubiculum in freely moving rats. in Electrical Activity of the Archicortex (eds G. Buzsaki & C.H. Vanderwolf) 217–220 (Akademiai Kiado, Budapest, 1985). 23. Sargolini, F. et al. Conjunctive representation of position, direction and velocity in entorhinal cortex. Science 312, 758–762 (2006). 24. Fyhn, M., Molden, S., Witter, M.P., Moser, E.I. & Moser, M.B. Spatial representation in the entorhinal cortex. Science 305, 1258–1264 (2004). 25. Hafting, T., Fyhn, M., Molden, S., Moser, M.B. & Moser, E.I. Microstructure of a spatial map in the entorhinal cortex. Nature 436, 801–806 (2005). 26. Taube, J.S., Muller, R.U. & Ranck, J.B. Head-direction cells recorded from the postsubiculum in freely moving rats. 2. Effects of environmental manipulations. J. Neurosci. 10, 436–447 (1990). 27. Johnson, A., Seeland, K. & Redish, A.D. Reconstruction of the postsubiculum head direction signal from neural ensembles. Hippocampus 15, 86–96 (2005). 28. Hargreaves, E.L., Yoganarasimha, D. & Knierim, J.J. Cohesiveness of spatial and directional representations recorded from neural ensembles in the anterior thalamus, parasubiculum, medial entorhinal cortex and hippocampus. Hippocampus 17, 826–841 (2007). 29. Yoganarasimha, D., Yu, X. & Knierim, J.J. Head direction cell representations maintain internal coherence during conflicting proximal and distal cue rotations: comparison with hippocampal place cells. J. Neurosci. 26, 622–631 (2006). 30. Fyhn, M., Hafting, T., Treves, A., Moser, M.B. & Moser, E.I. Hippocampal remapping and grid realignment in entorhinal cortex. Nature 446, 190–194 (2007). 31. Moser, E.I. & Moser, M.B. A metric for space. Hippocampus 18, 1142–1156 (2008). 32. Redish, A.D., McNaughton, B.L. & Barnes, C.A. Place cell firing shows an inertialike process. Neurocomputing 32, 235–241 (2000). 33. McNaughton, B.L., Battaglia, F.P., Jensen, O., Moser, E.I. & Moser, M.B. Path integration and the neural basis of the ‘cognitive map’. Nat. Rev. Neurosci. 7, 663–678 (2006). 34. McNaughton, B.L., Barnes, C.A. & O’Keefe, J. The contributions of position, direction, and velocity to single unit activity in the hippocampus of freely moving rats. Exp. Brain Res. 52, 41–49 (1983). 35. Hafting, T., Fyhn, M., Bonnevie, T., Moser, M.B. & Moser, E.I. Hippocampus-independent phase precession in entorhinal grid cells. Nature 453, 1248–1252 (2008). 36. Hasselmo, M.E. Grid cell mechanisms and function: contributions of entorhinal persistent spiking and phase resetting. Hippocampus 18, 1213–1229 (2008). 37. O’Keefe, J. & Burgess, N. Geometric determinants of the place fields of hippocampal neurons. Nature 381, 425–428 (1996). 38. Barry, C., Hayman, R., Burgess, N. & Jeffery, K.J. Experience-dependent rescaling of entorhinal grids. Nat. Neurosci. 10, 682–684 (2007). 39. Cheng, K. A purely geometric module in the rat’s spatial representation. Cognition 23, 149–178 (1986). 40. McGregor, A., Hayward, A.J., Pearce, J.M. & Good, M.A. Hippocampal lesions disrupt navigation based on the shape of the environment. Behav. Neurosci. 118, 1011–1021 (2004). 41. Jones, P.M., Pearce, J.M., Davies, V.J., Good, M.A. & McGregor, A. Impaired processing of local geometric features during navigation in a water maze following hippocampal lesions in rats. Behav. Neurosci. 121, 1258–1271 (2007). 42. Pearce, J.M., Good, M.A., Jones, P.M. & McGregor, A. Transfer of spatial behavior between different environments: implications for theories of spatial learning and for the role of the hippocampus in spatial learning. J. Exp. Psychol. Anim. Behav. Process. 30, 135–147 (2004). 43. Gothard, K.M., Skaggs, W.E. & McNaughton, B.L. Dynamics of mismatch correction in the hippocampal ensemble code for space: interaction between path integration and environmental cues. J. Neurosci. 16, 8027–8040 (1996). 44. Redish, A.D., Rosenzweig, E.S., Bohanick, J.D., McNaughton, B.L. & Barnes, C.A. Dynamics of hippocampal ensemble activity realignment: time versus space. J. Neurosci. 20, 9298–9309 (2000). 45. Biegler, R. Possible uses of path integration in animal navigation. Anim. Learn. Behav. 28, 257–277 (2000). 46. Samsonovich, A. & McNaughton, B.L. Path integration and cognitive mapping in a continuous attractor neural network model. J. Neurosci. 17, 5900–5920 (1997). 47. McNaughton, B.L. et al. Deciphering the hippocampal polyglot: the hippocampus as a path integration system. J. Exp. Biol. 199, 173–185 (1996). 48. Whishaw, I.Q. Place learning in hippocampal rats and the path integration hypothesis. Neurosci. Biobehav. Rev. 22, 209–220 (1998). 49. Solstad, T., Boccara, C., Kropff, E., Moser, M.B. & Moser, E.I. Representation of geometric borders in the entorhinal cortex. Science 322, 1865–1868 (2008). 50. Savelli, F., Yoganarasimha, D. & Knierim, J.J. Influence of boundary removal on the spatial representations of the medial entorhinal cortex. Hippocampus 18, 1270–1282 (2008).
1332
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
Methods Methods and any associated references are available in the online version of the paper at http://www.nature.com/natureneuroscience/. Note: Supplementary information is available on the Nature Neuroscience website.
© 2009 Nature America, Inc. All rights reserved.
Acknowledgments We thank A.M. Amundgård, I. Hammer, K. Haugen, K. Jenssen, R. Skjerpeng and H. Waade for technical assistance and T. Bonnevie and G. Pfühl for help with animal training. We thank A.D. Redish and members of the Kavli Institute for Systems Neuroscience and the Centre for the Biology of Memory for useful discussions. This work was supported by the Kavli Foundation and a Centre of Excellence grant from the Norwegian Research Council. AUTHOR CONTRIBUTIONS D.D., M.-B.M. and E.I.M. designed the study, M.F., J.R.W. and D.D. performed surgeries, D.D., J.R.W. and A.T. performed the experiments, M.F. and T.H. helped with training, D.D. analyzed the data, and D.D. and E.I.M. wrote the paper. All authors participated in planning and discussion. Published online at http://www.nature.com/natureneuroscience/. Reprints and permissions information is available online at http://www.nature.com/ reprintsandpermissions/.
ONLINE METHODS
© 2009 Nature America, Inc. All rights reserved.
Subjects and electrode implantation. Neuronal activity was recorded from MEC in 16 male Long-Evans rats (3–5 months old, 350–450 g at implantation and testing). Tetrodes were inserted above the dorsocaudal part of MEC, 4.5 mm lateral to the midline and 0.2–0.3 mm anterior to the transverse sinus, at an angle of ~10° in the anterior direction in the sagittal plane24,25. Most rats received additional tetrodes, above the dorsocaudal MEC of the other hemisphere (n = 1 rat), above CA1 (anterior-posterior, –4.0; medial-lateral, 3.1; n = 3), above CA3 (anterior-posterior, −3.2; medial-lateral, 3.1; n = 2) or above posterior parietal cortex (anterior-posterior, −4.0 and medial-lateral, 2.5; n = 9). In two additional rats, activity was only recorded from CA3. These experiments were approved by the National Animal Research Authority of Norway. Data collection. General data collection procedures are described in refs. 24 and 35. Rats collected crumbs of chocolate cereal thrown randomly into a black 1.5- × 1.5- × 0.5-m open-field arena surrounded by a black curtain. A white cuecard (95 × 45 cm) was placed on the curtain 110 cm above the floor. The floor was a removable linoleum mat. Training in the hairpin maze began when the rat regularly covered the entire open field on a 20-min trial. The linoleum floor was taken out and nine Perspex walls were inserted into parallel grooves carved into the underlying floor. The distance between each pair of adjacent walls was 14 cm. Walls were transparent (three rats) or opaque (13 rats). The walls were 135 × 50 × 1 cm (all rats in the transparent condition, three rats in the opaque condition) or 135 cm × 30 cm × 1 cm (ten rats). The rats were trained to run from east to west and from west to east on alternating trials without interruption. Food crumbs were initially administered by the experimenter at the south end of each arm (15 rats) or at both the south and north ends (one rat). Later, the food protocol was changed in 13 rats such that food was administered only at the end of the final arms (number 1 and 10), with one crumb sometimes being randomly administered on the way to keep the rat motivated. During testing, the rats ran for 20 min in the open field, followed by two 20-min runs in the hairpin maze, followed by another 20-min run in the open field. Between runs, the rats rested in a flower pot next to the maze or in their home cage. On shortcut trials, the 135-cm long wall between two of the central arms was replaced with an 85-cm wall touching the same outer wall as the replaced wall and a 29-cm long barrier was inserted perpendicular to the running direction 100 cm from the touch point. The junction between the truncated arms, in front of the barrier, was similar to junctions between the longer arms. Truncated arms were either numbers 4 and 5 (Fig. 5b) or numbers 5 and 6 from the west (Fig. 5a). The rats were trained and tested as in the regular hairpin task. In the ‘virtual hairpin’ task, the rats were tested in the same open field as in the pretraining stage, but instead of scattering food randomly across the arena, we delivered chocolate morsels, one at a time, at the north and south walls, in an alternating north-south sequence and at successive positions from west to east or vice versa. Food was delivered by two individuals, one at either side of the maze. The distance between adjacent reward locations was similar to the width of the alleys in the hairpin maze. This training regime resulted in a running pattern of ten north-south laps that was similar to the paths in the hairpin maze (Fig. 7 and Supplementary Fig. 6). Data analysis. Spike sorting was performed offline using graphical clustercutting software24. Position estimates were based on tracking of one LED on the headstage24. The tracked positions were smoothed offline with a 15-point mean filter. Tracking errors were removed offline by an interactive MATLAB script that assumed that the path did not cross walls. Analysis was performed separately for eastbound and westbound trajectories. A trajectory was defined as being a westbound path if the rat traversed all the way from arm 9 (in the east) to arm 2 (in the west) without returning back to arm 9. An eastbound path was one in which the rat traversed all the way from arm 2 to arm 9 without returning. The outer arms (1 and 10) were not analyzed because these arms contained long segments in which the rat was not running. Rate maps and spatial autocorrelation analyses. The position data were sorted into 1- × 1-cm bins and the firing rate was determined for each bin in the open field and in the hairpin maze24. A spatial autocorrelogram was calculated for the smoothed rate map of each cell in the open field and the hairpin task23. Grid spacing was determined as described previously23. A similar approach was used
doi:10.1038/nn.2396
for calculating linear autocorrelations in individual arms; in those analyses, the x parameter was held constant and only the spatial lag in the y direction was changed. The end arms (arms 1 and 10) were not used for this calculation. Local maxima in the autocorrelation map were identified by taking the spatial derivatives. If all points between two local maxima were high (>20% of the local maximum), then the lower local maximum was deleted. The degree of spatial periodicity in the autocorrelation map for each cell was determined in two ways. First, a gridness score was calculated23. This score is sensitive to the rotational symmetry of the grid. In addition, we estimated grid structure by a second mea sure in which the autocorrelation map is correlated with a perfect theoretical grid template. In those analyses, we first determined the spatial frequency (spacing) from the autocorrelation map for each cell. Grid spacing was taken as the median of the six nearest distances from local maxima to the center. The grid score was then determined by rotating the autocorrelation plot so that the closestto-the-center local maximum would lie on the x axis. Finally, the rotated plot was correlated with a perfect grid constructed from three cosines, in a ring with minimum radius equal to half the distance between two peaks and a maximum radius equal to 1.5-fold larger than the distance between two peaks. The perfect grid was calculated as 3 2p p i 2p p i G (x , y ) = ∑ cos x sin + y cos 3 d d 3 i =1 +
(1)
where d is the distance between two grid points. To estimate the periodicity of firing within individual arms of the hairpin maze, we further analyzed the linear spatial autocorrelations for each cell. First the global maximum point of the autocorrelogram was chosen (except for the edges) and all of the pixels around it that were within 50% of that maximum were checked for maxima. Out of the remaining pixels, a new global maximum was determined (with the condition that there will be a 50% drop from it on both sides) and all of the pixels around it that were within 50% of that maximum were checked, repeating this process recursively until no more maxima were found. Directional tuning The rat’s head direction was calculated for each tracker sample from the projection of the relative position of the two LEDs onto the horizontal plane, corrected for the possible angle between the placement of the two LEDs and the rat’s true heading. The directional turning function for each cell was calculated as described in ref. 49, except that head direction cells were defined as the cells with h > 0.25, that is, those cells in which the arc containing half of the distribution was smaller than 135°. The criterion was deliberately looser than that used in ref. 49 to ensure that no head direction cells were included in the turning point analyses. Firing rates in the hairpin maze. The firing rate of the cell was defined as a function of the arm number and the y position in each arm. For display, some smoothing was used. The pixel size that we used was 15 cm (width of arm) × 1 cm (in the y direction along arm), and a Gaussian kernel was used with a smoothing factor of h = 5 cm (smoothing was only applied along the arm). For correlation analyses, to not create dependencies between adjacent pixels, we did not use smoothing. In these cases, we used a larger pixel size of 15 cm (width of arm) × 10 cm (along the arm). Population vector construction and correlation procedures. The rate istributions were used to define population vectors, one for each arm of the maze d (Fig. 2b), and the Pearson product-moment correlation was calculated for each pair of arms, which resulted in a correlation matrix (for example, Fig. 2c). Frequency distributions for correlations (as in Fig. 3e) were made by determining the mean correlation between arms that were separated by one arm (same direction condition), except for arms 1 and 10, which were not used, or between adjacent arms (opposite direction condition). The frequency distributions were compared with a Monte-Carlo shuffled distribution obtained by circularly shifting the rate vector for each arm randomly by ±150 cm in the north-south direction and with a distribution obtained by reflecting the rate maps of alternating arms around the x axis at the middle of the arms. Correlations for the hippocampal data were calculated only for arms whose mean rates were larger than 0.25 Hz.
nature NEUROSCIENCE
Turning point analysis. The linear position l(x,y) of the rat along the path was calculated for westbound and eastbound paths separately. Inside the arms, the linear position was added to the linear position at the previous turning region. At the regions where two arms met, the turning area was defined as the 15-cm-long and 30-cm-wide region between the end of an inner flap, the two adjacent walls and the outer wall. Inside this region, the linear distance was defined as the projection of the path on a hemi-circle or radius of 7.5 cm whose center was positioned at the end of the inner flap. This linear distance was added to the linear position of the path at the point where the rat entered the turning region. Turning points were determined separately for eastbound and westbound trajectories. For every turn of a path segment in the turning region, a point of maximum curvature was defined as | x ′y ′′ − y ′x ′′ | max (k) = max 3 (x ′ 2 + y ′ 2 )2
(2)
Histology. Histology procedures were similar to those described in ref. 49.
© 2009 Nature America, Inc. All rights reserved.
where x and y are the positions of the path segment inside the turning region. The linear position of maximum curvature l(xmax,ymax) was also determined.
The linear position of the turning point in a specific turning region was defined as L = median(l(xmax,ymax)), for all trajectories passing through the turning region. For each path segment, x and y of the turning point were then defined such that l(x,y) = L. Once the turning points were defined, we calculated the mean linear rate of each cell triggered on the turning points, in 6-cm bins, separately for eastbound and westbound paths, with position = 0 at the turning point. A normalized rate population vector was constructed by aligning all of the linear rates for all nondirectional grid cells such that at each distance from the turning point there was a vector of n rates, where n was the number of cells. Grid cells that were modulated by head direction, as described above, were discarded. The rates of the included cells were normalized by the median. Cells with mean rate below 1 Hz were discarded. On the basis of previously published procedures5,32,43, we built a correlation matrix by correlating each rate vector at a specific position with the rate vector in a different position. The correlations were compared with those obtained in a shuffling Monte-Carlo procedure in which the trigger point (and thus the entire rate map) for each cell was shifted randomly by ± 60 cm. This procedure was repeated 1,000 times.
nature NEUROSCIENCE
doi:10.1038/nn.2396
a r t ic l e s
Transformation of nonfunctional spinal circuits into functional states after the loss of brain input
© 2009 Nature America, Inc. All rights reserved.
Grégoire Courtine1,2, Yury Gerasimenko3,4, Rubia van den Brand1,2, Aileen Yew5, Pavel Musienko1,2,4, Hui Zhong3, Bingbing Song6, Yan Ao6, Ronaldo M Ichiyama3, Igor Lavrov3, Roland R Roy3,6, Michael V Sofroniew5,6 & V Reggie Edgerton3,5,6 After complete spinal cord transections that removed all supraspinal inputs in adult rats, combinations of serotonergic agonists and epidural electrical stimulation were able to acutely transform spinal networks from nonfunctional to highly functional and adaptive states as early as 1 week after injury. Using kinematics, physiological and anatomical analyses, we found that these interventions could recruit specific populations of spinal circuits, refine their control via sensory input and functionally remodel these locomotor pathways when combined with training. The emergence of these new functional states enabled full weightbearing treadmill locomotion in paralyzed rats that was almost indistinguishable from voluntary stepping. We propose that, in the absence of supraspinal input, spinal locomotion can emerge from a combination of central pattern-generating capability and the ability of these spinal circuits to use sensory afferent input to control stepping. These findings provide a strategy by which individuals with spinal cord injuries could regain substantial levels of motor control. Severe spinal cord injuries that remove all supraspinal input to lumbosacral spinal circuits lead to permanent paralysis of the legs in adult rodents1–3 and humans. Nevertheless, networks of neurons in the lumbosacral spinal cord retain an intrinsic capability to oscillate and generate coordinated rhythmic motor outputs. Circuits under lying such rhythmic and oscillatory outputs are commonly referred to as central pattern generators (CPGs) and are found in all invertebrate and vertebrate animals4,5. Although the anatomical architecture of locomotor CPGs remains poorly understood, especially in mammals5, the functional phenomenon, central pattern generation, has been documented extensively. Indirect evidence suggests that CPGs are present in human spinal cord6,7. These observations offer the possibil ity of directly accessing and activating spinal cord CPGs to facilitate locomotor recovery after a severe spinal cord injury (SCI). Several experimental strategies have been tested to activate loco motor circuits in mammals after a complete spinal cord transection, including pharmacological treatments8–10, epidural2,11,12 or intraspi nal13,14 electrical stimulation, and motor training1,2,8,15,16. Serotonin or agonists of 5-HT2A and 5-HT1A/7 receptors can activate the qui escent locomotor circuitry in neonatal rodent fictive locomotion preparations17,18 and can facilitate treadmill stepping with limited weight bearing in adult rats9 and mice10 with SCI. Epidural electrical stimulation (EES) applied dorsally at the lumbar (L2) 2,11 or sacral (S1)12,19 spinal segments induces rhythmic hindlimb movements2,12. Locomotor training, notably in conjunction with pharmacological8,9 or electrical stimulation2 interventions, can promote use-dependent plastic changes in sensorimotor circuits below the injury 16,20,21 that lead to specific improvements of stepping patterns. These
i nterventions, however, have shown limited potential for promoting weight-bearing capacities and there have been few attempts to cor relate the specific functional states induced pharmacologically9,10, electrically11–13,19,22 or by locomotor training2,20 with distinct char acteristics of stepping motor patterns. When studied in sufficient sta tistical detail, analyses of kinematics and electromyographic (EMG) features revealed that such induced spinal locomotion differed from voluntary stepping in many important aspects2,9,13. In addition, it remains unknown whether lumbosacral neuronal networks in the absence of brain input could sustain full weight-bearing locomotion that resembles nondisabled stepping. Considering the diffusely dis tributed5 and heterogeneous4 character of the spinal locomotor sys tem, it is likely that multiple complementary approaches, both acute and chronic, would be required to attain the full possible expression of effective stepping in the absence of supraspinal input. We tested the hypothesis that combinations of specific pharmaco logical and electrical stimulation interventions, together with locomo tor training, may interact synergistically to activate and functionally remodel spinal locomotor circuits, possibly enabling a coordinated and context-dependent function of the paralyzed hindlimbs of adult rats after a complete spinal cord transection. We tested combinations of 5-HT2A and 5-HT1A/7 serotonin agonists and EES at two different positions distal to the lesion, each of which individually exerted some facilitating effects on hindlimb function. Using detailed kinematics, EMG and anatomical analyses, we found that such combinatorial interventions induced unique functional states that correlated with distinct patterns of locomotion in paralyzed rats. We demonstrate for the first time, to the best of our knowledge, the ability of rats
1Neurology Department, University of Zurich, Zurich, Switzerland. 2Rehabilitation Institute and Technology Center Zurich, Zurich, Switzerland. 3Department of Physiological Science, University of California Los Angeles, Los Angeles, California, USA. 4Motor Physiology Laboratory, Pavlov Institute of Physiology, St. Petersburg, Russia. 5Department of Neurobiology, 6Brain Research Institute, University of California Los Angeles, Los Angeles, California, USA. Correspondence should be addressed to G.C. (
[email protected]).
Received 3 March; accepted 20 August; published online 20 September 2009; doi:10.1038/nn.2401
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1333
a r t ic l e s
1′
1′
n Full combination
1′
0%
5
k
–10 10 5 10 PC 0 5 2: –5 –5 0 21% 15 –10 –10 % 1: PC iance) (var
Increasing quality of stepping 8
0 –5
BWS: 20%
* 0
* –8
*
*
l
EES at L2 versus S1 6
*
m
**
0
–6
BWS: 30%
BWS: 30%
BWS: 15%
Specificity of quipazine
**
2 0
–5
*
BWS: 50% Kinematics
BWS: 20%
BWS: 5%
PC 3 scores
a rdin ti oo
BWS: 0%
Gait timing
quipazine 8-OHDPAT
Endpoint trajectory
10′
quipazine 8-OHDPAT
4 cm Stance
PC 2 scores
4 mV
PC 3: 11%
OR
g EES at S1 h EES at L2 i
ing
C
© 2009 Nature America, Inc. All rights reserved.
10′
f
Sw
ag Dr
50% BWS: 1 s on 100%
j
1′
e
EES at L2 + S1 Quipazine 8-OHDPAT 8-OHDPAT
L2 + S1 quipazine
Quipazine EES S1 + L2 8-OHDPAT EES S1 + L2 Quipazine + 8-OHDPAT EES S1 Quipazine + 8-OHDPAT EES L2 Full combination
1 5 Variability 10 15 20 Value 25 30 35 40 Variability 45 50 55 60 65 Excursions 70 75 80 Similar 85 to pre-injury 90 95 Variability100 Timing 105 Limb osc 110 Amplitude 115 120 Timing 125 130 Stability 135 BWS Value
0.8
0.4
0
Factor loading
4 cm
s–1 Sol TA
2 cm
Ankle MTP
1 week
d EES at
EMG
No intervention EES at L2 + S1
Voluntary Swing Hip Knee
1 week after a complete interruption of supraspinal input
c
PC 1 scores
a
b
Pre-injury
–0.4
–0.8 PC 1 PC 2PC 3
Figure 1 Accessing spinal locomotor circuits 1 week after the interruption of all supraspinal input. (a–i) EMG and kinematic characteristics underlying locomotion recorded pre-injury (a) and 7–8 d post-injury (b) without any intervention, as well as under various combinations of serotonergic agonists and/or EES (c–i). The full combination (i) included quipazine, 8-OHDPAT and EES at L2 and S1. Recordings were performed sequentially in the same rat. Horizontal arrows indicate the chronology of the different recordings. A representative stick diagram decomposition of hindlimb motion during swing is shown for each condition with successive color-coded trajectories of limb endpoint. Vectors represent the direction and intensity of the limb endpoint velocity at swing onset. A sequence of raw EMG activity from tibialis anterior (TA) and soleus (Sol) muscles is shown below. Grey and red bars indicate the duration of stance and drag phases, respectively. The BWS of the represented rat under each condition is shown. Finally, a polar plot representation documents the coordination between the left and right hindlimbs (thin arrow, single gait cycle; thick red arrow, average of all gait cycles; 50%, out of phase). (j) Three-dimensional statistical representation of locomotor patterns. Each small colored label represents the gait pattern from an individual rat under a given combination of interventions. The area defined by individual points under a given condition is traced to emphasize the differences between gait patterns under specific combinations. This analysis revealed that each combination of interventions resulted in distinct, but reproducible, patterns of locomotion. (k–m) Bar graphs of average scores on principal components 1–3. (n) Color-coded representation of factor loadings that identify the variables that contributed most to the differences observed between the experimental conditions. For example, principal component 2 captured the differences between stepping with EES at L2 versus S1. Variables associated with changes in joint angles toward flexion and limb endpoint trajectory (left and right step heights, 18–19) clustered on principal component 2, indicating that EES at L2 enhanced flexion, whereas EES at S1 enhanced extension. All of the computed kinematic and EMG variables (n = 135) are reported in Supplementary Table 1. Error bars represent s.e.m. * P < 0.05, different from all other conditions. ** P < 0.05, significantly different conditions.
Accessing spinal locomotor circuits We first examined the ability of combined pharmacological and electrical stimulations to transform spinal circuits from a dormant to a functional state. These experiments were conducted sequen tially (Fig. 1a–i) on the same rats to identify the specificity of and synergy between each intervention on the modulation of step ping patterns (Supplementary Fig. 2). To facilitate stepping with
electrical stimulation, we applied EES over the dorsal surface of L2 and S1 (Supplementary Fig. 1). We previously found that stimula tion at either site can facilitate treadmill locomotion in paralyzed rats with SCI after 5–7 weeks of recovery post-injury11,19. Applied 7–8 d post-injury, EES (40–50 Hz, 0.2 ms, 1–4 V) at L2 and/or S1 modestly increased EMG bursting patterns or tonic activity in the hindlimb muscles (P < 0.05; Fig. 2), but failed to generate any step-like movements (Fig. 1c). To engage locomotor networks phar macologically, we administrated quipazine, a predominantly 5-HT2A receptor agonist, and 8-OHDPAT, a 5-HT1A and 5-HT7 receptor agonist. Both serotonin agonists can facilitate treadmill locomo tion in spinal rodents after some weeks of recovery from SCI9,10, but when systemically administered at 7–8 d post-injury, quipazine (0.3 mg per kg of body weight, intraperitoneal) and/or 8-OHDPAT (0.1–0.3 mg per kg, subcutaneous) induced only brief periods of erratic hindlimb movements in response to treadmill motion (Fig. 1f), with limited weight bearing (Fig. 2c) and small EMG bursts in both extensor and flexor muscles (Fig. 2g,h). Thus, pharmacologi cal or EES interventions alone, administrated 1 week after a complete interruption of supraspinal input, could not induce functional states that would enable stepping. In contrast, several combinations of serotonin agonists and EES were strongly synergistic, resulting in hindlimb locomotion whose features clearly varied with the proce dures used (Fig. 1d–i).
1334
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
with SCI to generate full weight-bearing bipedal treadmill locomotion that is almost indistinguishable from voluntary stepping recorded in the same rats prior to injury. We also found that sensory input determines the formation of adaptive motor patterns in the absence of supraspinal influences. RESULTS Experiments were conducted on adult rats that received a complete mid-thoracic (~T7) spinal cord transection, permanently removing all supraspinal input below the level of the lesion. At 7–8 d after injury, all of the rats showed flaccid paralysis of the hindlimbs; no bursts of EMG were observed in extensor or flexor muscles (Fig. 1) when a rat was positioned in a bipedal posture on a moving treadmill belt (9 cm s−1) while secured in a jacket that was attached to a robotic arm used to control and measure the amount of hindlimb body weight support (BWS; Supplementary Fig. 1)2.
a r t ic l e s a
b
Cross-correlation (r value at t = 0)
f
We extensively quantified the kinematics and EMG characteris tics of stepping movements caused by specific treatment combina tions (n = 135; Supplementary Table 1). However, because a large number of these parameters changed substantially across combina tions, we resorted to principal component analysis (PCA) to iden tify the specific contributions of individual interventions to the observed variations in hindlimb stepping. PCA transforms the raw dataset to a new coordinate system such that the explained variance is maximized on each axis. In the PCA-based statistical representa tion, the differences between gait patterns are related to the distance between the data points. The distinct spatial locations (P < 0.05; Fig. 1k–m) occupied by each combination (Fig. 1j) indicated that
h
Normalized amplitude
g
Normalized amplitude
each intervention promoted a specific pattern of hindlimb locomo tion. The distinct characteristics captured by each component can be extracted from the analysis of factor loadings, that is, correlations between each variable and each component. To visualize factor load ings, we developed a color-coded representation that clearly high lighted the variables that cluster on each component (Fig. 1n). For example, principal component 3 (Fig. 1j) differentiated the effects of quipazine from the effects of the other interventions (P < 0.05; Fig. 1j). Variables associated with joint extension (variables 56, 57 and 76) and EMG activity of extensor muscles (soleus, P < 0.01; Fig. 2g) clustered on principal component 3 (Fig. 1n), indicating that quipazine primarily facilitated extension components. 8-OHDPAT
Figure 3 Site-specific effects of EES during Standing Foot/step height standing and stepping. (a) Stick diagram Stepping Standing Stepping L2 S1 decomposition of hindlimb movements and time Optimal L2 S1 ** 5.0 ** * ** course of changes in hindlimb joint angles when 5 ms 50 ms * * delivering EES at L2 (left) or S1 (right) during 2.5 standing (20% of weight bearing). Each diagram –1 2 cm s is separated by 5 ms (L2) or 50 ms (S1). The dark 0 No L2 S1 L2 S1 L2 S1 gray shaded areas indicate the period during which EES Opt +0.4V +0.8V 100 100 Hip EES-induced changes in hindlimb posture were Angular changes 4 cm (deg) 80 80 ** ** 100 100 observed. Light gray shaded areas represent periods 80 70 Ext ** Knee * * Ext * 60 60 during which EES-induced posture was maintained. 150 150 Ankle * 60 0 50 50 (b) Effects of increasing EES by 0.4 and 0.8 V 150 150 Fle MTP at L2 versus S1 on hindlimb movements during 100 100 Fle 50 –80 STIM * S1 locomotion enabled by the full combination of L2 S1 L2 S1 No L2 200 ms Opt +0.4V +0.8V EES interventions. Data are presented as in Figure 1. Arrows indicate increased flexion with EES at L2 versus increased extension with EES at S1. (c) Bar graph of average values of maximum foot height during standing and step height during locomotion. Opt, optimal EES intensity to encourage stepping. (d) Bar graphs of average values of angular changes during standing and stepping. For standing, values were obtained by measuring, for each joint angle, the difference between positions at EES onset and at the time of maximum EES-induced change in hindlimb posture at each joint and then averaging these values across joints. Ext, extension. Fle, flexion. For stepping, the maximum position of the hip joint angle in flexion during swing was computed. Error bars represent s.e.m. * P < 0.05, different from all the other conditions. ** P < 0.05, significantly different conditions.
c
Max hip flexion (deg)
Whole limb (deg)
d
+0.8V
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
(cm)
b
+0.4V
a
2 cm
© 2009 Nature America, Inc. All rights reserved.
Normalized values
e
d
Correlation (r value )
Maximum of body weight (%)
c
Height (cm)
Swing
Drag
Stance
Duration (s)
Figure 2 Kinematics and EMG features of Pre-injury Inter-limb coordination Gait timing locomotor patterns. (a–h) Bar graphs of average Voluntary 0.2 3 ** values (n = 7 or 8 rats per group) for locomotor 0 2 7–8 d post-injury parameters computed under the different –0.2 No intervention experimental conditions. In each graph, the 1 ** –0.4 1 EES S1 + L2 green horizontal bar represents the pre-lesion 2 Quipazine + 8-OHDPAT –0.6 0 ** baseline recorded in the same rat 1 week pre3 Quipazine and EES S1 + L2 123456 123456 ** 4 8-OHDPAT and EES S1 + L2 injury. The duration of stance (light gray), swing 5 Quipazine + 8-OHDPAT and EES S1 Weight-bearing capacities Kinematic similitude (dark gray) and drag (red) phases is shown in a. 6 Quipazine + 8-OHDPAT and EES L2 ** ** 100 1.0 Full combination The r values at t = 0 for the cross-correlation ** ** ** ** function between oscillations of the left and ** 9 weeks post-injury ** 50 0.7 ** ** right hindlimbs computed over a gait sequence Non-trained ** of ten steps is shown in b. A r value of −0.5 Trained with EES S1 + L2 ** Trained with quipazine and indicates out of phase coupling between the 0 0.4 8-OHDPAT 123456 123456 limbs. The maximum level of weight bearing Trained with full combination Limb endpoint trajectory Variability (percentage of body weight) at which the rat ** ** 5.0 3 could perform ten successful steps is shown in c. ** ** ** Cross-correlation functions were computed ** ** ** 2 ** ** 2.5 between datasets obtained pre-injury and under ** ** 1 a given experimental condition for the hip, knee, ankle and MTP joint angles, and associated joint 0 0 123456 123456 velocity profiles (d). Maximum r values were Extensor EMG Flexor EMG extracted from each cross-correlation function ** ** ** ** 1.5 1.5 and averaged across joint angle and joint P = 0.07 angle velocity profiles. The variability of gait ** ** 1.0 1.0 ** ** parameters computed as the mean coefficient 0.5 ** 0.5 ** of variation for all the computed parameters normalized to the pre-injury baseline are shown 0 0 123456 123456 in e. Step height, defined as the maximum vertical distance between the foot (MTP marker) and the stepping surface, is shown in f. The average EMG burst amplitude for left and right soleus (g) and tibialis (h) anterior muscles normalized to pre-injury values are shown. Error bars represent s.e.m. ** P < 0.05, significantly different conditions.
1335
a r t ic l e s
Swing
Locomotor training enabled by
c
Quipazine 8-OHDPAT
EES L2 + S1
d
Full combination
f
1
Gait timing
b
Value
Endpoint trajectory
Nontrained
a
Value
5
0.8
10
Variability
15 20
4 cm
4 cm s–1
70%
100%
35
50
60
50%
ina ord tio
n
Excursions
Co Kinematics
Locomotor training enabled by Quipazine 8-OHDPAT
50%
50%
Full combination
EES L2 + S1
50%
60%
60%
100%
Similar to pre-injury
70%
60%
70%
60%
70%
60%
70%
60%
70%
90%
60%
100%
80%
**
–4 6 PC 2
0
(1
7% –6 –8 )
0 ) (24% PC 1
8
PC 1 scores
–6 0
P = 0.06 0
–6
i
**
Deterioration –6
**
**
0
–4
Limb osc
100%
j
*
90 95 –0.4
105 110
Amplitude 115
100%
Timing Stability BWS
100%
120 125 130
–0.8
135 PC 1 PC 2 PC 3
Training specificity 3
PC 3 scores
4
Improved gait
PC 2 scores
h
85
Variability 100
100%
EMG
50%
0
70 80
Timing 50%
65
75
Nontrained
45%
PC 3 (14%)
45
Factor loading
e
g
0.4
40 Variability
55 1s
© 2009 Nature America, Inc. All rights reserved.
50%
30
4 mV
Sol TA
4 cm BWS: 50%
25
**
1 week post-injury
Nontrained
Locomotor training enabled by
* 0
Serotonergic agonists EES Full combination
–3
Figure 4 Rehabilitation locomotor training enabled by pharmacological and EES interventions improves stepping ability. (a–d) Representative illustrations of EMG and kinematic characteristics underlying bipedal hindlimb locomotion recorded at 9 weeks post-injury under the full combination of interventions for a nontrained rat with SCI that did not receive pharmacological or EES interventions for the entire duration of the post-injury period until the day of testing (a), a rat with SCI that was trained with serotonergic agonists only (b), a rat with SCI that was trained with EES at L2 and S1 only (c), and a rat with SCI that was trained with the full combination of interventions (d). Data for this rat are also shown pre-injury in Figure 1a and at 1 week post-injury in Figure 1i. (e) Successive limb endpoint trajectories from the right hindlimb are shown for all of the other rats from each experimental group. The BWS of each rat is reported below each limb endpoint trajectory. (f) Color-coded representation of factor loadings of each variable on principal components 1–3 (as shown in Fig. 1n). Principal component 1 identified improved gait in rats tested at 1 week post-injury and then trained with the full combination compared with the other groups. The analysis of variables that clustered on principal component 1 indicated that reduced variability of gait parameters, improved gait stability, increased amplitude of EMG activity and recovery of full weight-bearing capacities were the more salient features for explaining the improved stepping performances of rats trained with the full combination. (g) Three-dimensional statistical representation of locomotor patterns. The near absence of spatial interceptions between the different groups indicates that each group of rats had unique stepping patterns. (h–j) Bar graphs of average scores on principal components 1–3, which each captured specific effects. Error bars represent s.e.m. * P < 0.05, different from all the other conditions. ** P < 0.05, significantly different conditions. Data are presented as in Figure 1.
was significantly more effective in facilitating rhythmic movements of the hindlimbs under EES at S1 plus L2 compared with quipazine (principal component 1, P < 0.05; Figs. 1k and 2b,e). However, step ping patterns were less variable (P < 0.001; Fig. 2e), showing mark edly improved interlimb coordination (P < 0.01; Fig. 2b) and higher levels of weight bearing (P < 0.01; Fig. 2c) when combining quipazine, 8-OHDPAT and EES at either site. In turn, principal component 2 differentiated gait patterns under EES at the L2 versus S1 levels (P < 0.001; Fig. 1i). Analysis of the variables that clustered on princi pal component 2 (Fig. 1m) showed that EES at L2 facilitated flexion
and resulted in an enhanced swing phase (P < 0.01; Fig. 2f), whereas EES at S1 was more biased toward extension. To further investigate the modulation of hindlimb movements by site-specific EES, we compared the changes in limb postures produced by EES at L2 compared with EES at S1. During standing (20% weight bearing), EES at L2 (Fig. 3) induced a rapid flexion (mean duration to peak flexion, 263 ± 96 ms) at all joints (P < 0.001; Fig. 3c,d) that was maintained tonically during the duration of the stimulation (3 s). In contrast, EES at S1 promoted a progressive whole-limb exten sion (mean duration to peak extension, 1,050 ± 361 ms; P < 0.001;
1336
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
a r t ic l e s a
Normalized amplitude
Trained
4 ms
200
b Nontrained
Tibialis anterior
1 week
9 weeks
+
100 0
Preinjury
1 9 week(s)
Nontrained Trained with full combination
Trained
Normalized amplitude
300
+
200 100 0
L2
No injury
Nontrained
Trained with full combination
d Count
c
e
*
350
* * ****
1 9 week(s)
* * * * ** ** **** ** *
L1 L2 L3 L4 L5 L6 S1 S2 Spinal segment Noninjured Nontrained Trained with full combination
4,000 Overall count
Preinjury
FOS expression per segment
700
0
L4
Fig. 3a,d) that persisted as long as the stimulation was continued (3 s) and was comparatively slower (P < 0.001) and less stable than the effects of EES at L2. Next, we tested the effects of increasing intensity at L2 versus S1 during locomotion enabled by quipazine, 8-OHDPAT and dual-site EES. Gradually increasing the intensity of EES at L2 progressively increased hip flexion (P < 0.05; Fig. 3d), step height (P < 0.05; Fig. 3c) and foot velocity (P < 0.05) during swing. Opposite effects (P < 0.05; Fig. 3c,d) were obtained when increasing EES at S1 (Fig. 3b). Together, these results demonstrate that upper lumbar stim ulation engages the circuits controlling flexion, whereas upper sacral stimulation primarily recruits the circuits controlling extension. When EES was simultaneously applied to L2 and S1 in the presence of quipazine and 8-OHDPAT (Fig. 1i), the variability of stepping motion was reduced compared with EES application at either site alone (P < 0.01; Fig. 2e). Weight-bearing ability increased twofold (P < 0.01; Fig. 2c) and EMG activity in extensor and flexor muscles reached levels similar to those observed during voluntary locomo tion recorded in the same rats pre-injury (Fig. 2g,h). Although some defects persisted, for example, toe dragging at swing onset (P < 0.05; Figs. 1d and 2a) and only partial weight-bearing ability (P < 0.001; Fig. 2c), the resulting locomotor patterns resembled voluntary step ping (Fig. 1a,i). These results indicate that there is a strong, synergistic potential for combined pharmacological and electrical stimulations to functionally engage spinal locomotor circuits as early as 1 week after a complete spinal cord transection, promoting weight-bearing locomotion with plantar placement of the paws in the hindlimbs of paralyzed adult rats (Supplementary Video 1).
Monosynaptic motor-evoked potentials 300 + +
Post-injury 1 week 9 weeks
2 mV
Nontrained
Pre-injury Soleus
S1
© 2009 Nature America, Inc. All rights reserved.
Figure 5 Functional remodeling of spinal circuits after rehabilitative locomotor training. (a) Representative average (n = 10) traces of monosynaptic motor-evoked potentials recorded from the soleus muscle pre-injury and at 1 and 9 weeks post-injury. Data are shown for one nontrained and one rat trained with the full combination of interventions. Dark shaded areas indicate the amplitude of pre-injury motor-evoked potentials. Bar graphs report the average amplitude of motor-evoked potentials recorded in the soleus muscle at the different time points. (b) Data are presented as in a for the tibialis anterior muscle. (c) Representative example of camera lucida drawings of FOS-positive cells in spinal segments L2, L4 and S1 of a noninjured rat, a nontrained rat with SCI and a rat with SCI trained with the full combination of interventions. (d) Average values for the total FOS-positive cell count (all laminae) per spinal segment. (e) Correlation between the total number of FOS-positive cells (all laminae from L1 to S2) and gait performance measured as individual scores along the principal component 1 axis. PCA was applied on locomotor data (n = 135) recorded from the same rats 3–5 d before the FOS experiments under the same conditions, that is, no intervention for noninjured rats and under the full combination for rats with SCI. Error bars represent s.e.m. * P < 0.05, different from trained group. ** P < 0.05, different from noninjured group.
r = –0.83
2,000 0 –12
–6 0 Gait performance
6
Learning in spinal locomotor circuits Next, we sought to optimize the use of spinal circuits for recovery of locomotor function in the absence of supraspinal input. To attain this goal, we subjected rats with SCI to the above pharma cological and/or EES interventions in combination with wellestablished rehabilitative locomotor training procedures. We hypothesized that use-dependent mechanisms would functionally remodel the pharmaco-electrically activated spinal circuits and further improve stepping ability. We exposed the rats to 20-min locomotor training sessions every other day for 8 weeks, starting 7-8 d post-injury. Stepping was enabled by the full combination of serotonergic agonists and dual-site EES. We compared the locomotor performance of these rats (Fig. 4) with
their performance measured at 1 week post-injury, before any training (Fig. 1i), with that of nontrained rats that received no pharmaco logical or electrical stimulation (Fig. 4a) and with that of rats trained with either the serotonin agonists (Fig. 4b) or dual-site EES (Fig. 4c) alone (Supplementary Fig. 2). PCA (Fig. 4f–j) identified clusters of variables on each component (Fig. 4f) that allowed us to differentiate the effects of the various condi tions on stepping abilities. Principal component 1 identified improved performance in rats that were trained with the full combination of serotonergic agonists and dual-site EES compared with nontrained rats and rats trained with either the serotonergic agonists or EES alone (P < 0.01; Fig. 4h). Principal component 2 highlighted the deteriora tion of stepping ability in nontrained rats after 8 weeks compared with the gait patterns recorded at 1 week post-injury (P < 0.001; Fig. 4i). Principal component 3 revealed the importance of combining the serotonergic agonists and EES for promoting stepping improvement with locomotor training (P < 0.01; Fig. 4j). Deterioration of stepping capacities in chronic nontrained rats (Fig. 4a) was made evident by the substantial increase in movement variability (P < 0.001; Fig. 2e), the loss of interlimb coordination (P < 0.001; Fig. 2b) and the partial coactivation of normally reciprocally activated flexor and extensor hindlimb muscles (Fig. 4a and Supplementary Fig. 3). In contrast, rats trained with the full combination of treatments (Fig. 4d) recovered the ability to sustain full weight-bearing locomotion (Fig. 2c) with kinematic profiles of hindlimb joint angles (P > 0.2; Fig. 2d) and limb endpoint trajectories (Fig. 4d) that were nearly indistinguishable from those observed during voluntary bipedal treadmill stepping in the same rats pre-injury (Fig. 1a). Although locomotor training with the full combination did not completely prevent the hindlimb muscle
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1337
a
0 cm s–1
60
Hip (deg)
5 cm s–1
15 cm s–1
25 cm s–1
0 cm s–1
4 cm
8 cm s–1
First step
5 cm
0 100
Knee
50 150
Ankle
50 200
MTP
50 MG TA
2 mv
VL St Stance
b 1.4 0.7
25 0 1s Gait timing –1 25 cm s 22 Stance 19 16 13 10 7 r = 0.97 ± 0.02 r = 0.61 ± 0.08 Swing
0 0.6
1.0 Cycle duration (s)
1.4
c
1.4
EMG timing Extensor r = 0.85 ± 0.09
0.7
r = 0.44 ± 0.12 Flexor 0 0.6 1.0 1.4 Cycle duration (s)
d Inter-limb lag (%)
Speed (cm s–1)
Burst duration (s)
Figure 6 Effects of velocity-dependent afferent input on motor patterns. (a) Representative example of hindlimb kinematics and EMG activity recorded from a continuous sequence of steps during which the speed of the treadmill belt was changed gradually (0, 5, 15, 25 and 0 cm s −1). Data are presented as in Figure 1, except that changes in hindlimb joint angles are also shown. Stick diagram decomposition of the first step is shown to demonstrate the smooth transition from standing to stepping. MG, medial gastrocnemius; St, semitendinosus; VL, vastus lateralis. (b) The durations of the swing and stance phases are plotted against the cycle duration. Colorcoded labels indicate the measured treadmill belt speed during the performance of the represented gait cycles. (c) The durations of flexor (TA) and extensor (MG) EMG bursts are plotted against the cycle duration. (d) The temporal lag between oscillations (with respect to the direction of gravity) of adjacent hindlimb segments is plotted against the cycle duration. Inter-limb lags were computed by means of cross-correlation functions and expressed as a percent of cycle duration. b–d are shown for a representative rat. Mean ± s.e.m. correlation values computed by averaging values obtained from linear regressions performed on each rat (n = 6) individually are reported in each plot. All rats were trained with the full combination of interventions for 3 weeks before the experimental testing.
Duration (s)
© 2009 Nature America, Inc. All rights reserved.
a r t ic l e s
12
Kinematic timing r = –0.85 ± 0.05 Leg-foot
0 r = 0.69 ± 0.1 Foot-toe
–12 0.6
Thigh-leg r = 0.82 ± 0.1 1.0 Cycle duration (s)
1.4
atrophy that accompanies a SCI, the amount of muscle weight loss was significantly less (mean for all muscles, 14%) in trained versus nontrained rats (P < 0.01; Supplementary Fig. 4). Deterioration of locomotor performance was reduced in rats trained with either the pharmacological or EES intervention compared with nontrained rats (P < 0.01; Fig. 4i). However, locomotor training enabled by the individual interventions failed to promote substantial improvement in stepping ability. We next investigated use-dependent plastic changes in the spinal circuits that mediate the observed functional improvements. To assess physiological changes, we chronically measured the efficacy of mono synaptic inputs to extensor and flexor motoneurons in awake standing (20% weight bearing) rats23. The rats showed a significant decrease in the amplitude of monosynaptic responses in extensor motoneuron pools 1 week post-injury (P < 0.05; Fig. 5a), whereas these values were unchanged in flexor motoneuron pools (Fig. 5b) compared to pre-lesion values. Before the onset of locomotor training, no signifi cant differences could be detected between the two groups (P > 0.4; Fig. 5a,b). After 9 weeks of the absence of weight bearing in non trained injured rats, we observed no significant changes in the ampli tude of the responses in flexor motoneuron pools (Fig. 5b), whereas there was a moderate facilitation of the extensor motoneuron pools (P < 0.05; Fig. 5a). In contrast, after 8 weeks of locomotor training, all of the rats demonstrated a substantial increase in the efficacy of monosynaptic inputs to both extensor (P < 0.01; Fig. 5a) and flexor (P < 0.01; Fig. 5b) motoneuron pools. We further assessed the func tional remodeling of lumbosacral locomotor networks anatomically by examining FOS expression patterns2 induced by 45 min of continuous
bipedal stepping under the full combination of treatments. Although FOS-positive nuclei were found mainly in laminae I–IV in the vicinity of the site of stimulation when EES was delivered with the rats in a prone, suspended non–weight-bearing position (Supplementary Fig. 5), FOS-positive nuclei were present in all of the laminae of the examined lumbar and sacral segments in response to locomotor activ ity (Fig. 5c,d). However, the number of FOS-positive cells per segment (L1 to S2) was two- to threefold higher in nontrained than in trained (P < 0.01) and noninjured (P < 0.001) rats (Fig. 5d and Supplementary Fig. 5). Moreover, the total number of FOS-positive cells was inversely correlated with the level of locomotor recovery (P < 0.001; Fig. 5e). These results confirm earlier observations2 that motor learning can occur in adult rodent lumbosacral circuits after a complete spinal cord transection and extend these findings by showing that, in the absence of any supraspinal input, use-dependent learning mechanisms can pro mote the recovery of full weight-bearing treadmill locomotion that is kinematically very similar to pre-injury voluntary stepping in the same rats (Supplementary Video 2).
1338
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
Control of spinal locomotor circuits We next investigated how successful control of spinal locomotor circuits occurs in the absence of supraspinal input in adult rats. Cats with SCI can recover hindlimb locomotion when lumbosacral circuits have access to sensory information8,15. This is usually attributed to the sensory systems providing feedback correction 24 and reinforcement25 of motor patterns generated by the spinal circuitry4,9,26. However, the precise contributions of sensory input during spinal locomotion have not been defined. To investigate the
a r t ic l e s a
b
Force (N)
Force (N)
MG (mV)
MG (mv)
TA
TA
VL
St
Ankle (deg)
Hip
2 cm
6
cm
s
–1
Figure 7 Effects of load-dependent afferent 40% 0% 0% 20% 60% 100% input on motor patterns. (a) Representative (no belt motion) (air) example of hindlimb kinematics and EMG Swing 10 activity induced by the full combination while the rat transited from suspended in the air 0 0.4 (0% of body weight) to contact with the Stance 4 cm 120 immobile treadmill belt (40% of body weight 0 100 support). (b) Representative example of limb 0.4 1 endpoint trajectories, mean EMG activity and 0 mean vertical reaction forces during stepping 0.4 –1 with 0, 20, 60 or 100% weight bearing. 0.5 0 (c) The mean vertical reaction force measured 0.8 during stance is plotted against the amplitude –0.5 0 2 8 of the medial gastrocnemius EMG burst for a representative rat that demonstrated full 0 0 weight-bearing capacities after 3 weeks of training with the full combination of 0 100 200 1s Cycle duration (%) interventions. Color-coded labels represent the amount of body weight supported by the St VL TA MG Vertical force 100 % 2 6 14 2 3 0.5 ** 80 hindlimbs of the rat. A strong relationship ** ** 60 ** 40 was observed in all of the rats (n = 6). The 20 2 ** ** ** 0 mean ± s.e.m. correlation value computed ** ** 1 1 3 7 ** ** ** ** by averaging the values obtained from linear 1 ** ** correlation computed on each rat is reported. r = 0.82 ± 0.06 (d) Bar graphs of average amplitudes (n = 6) 0 0 0 0 0 0 0 5 of EMG bursts in selected hindlimb muscles. Mean vertical force Values are normalized to values measured during stance (N) Stepping during standing (40% of body weight support). The 100% weight-bearing condition is not represented because only two rats could step without any support after 3 weeks of training. (e) Bar graphs of average values (n = 6) of vertical reaction forces measured during stance. Error bars represent s.e.m. ** P < 0.05, significantly different conditions.
a
b
Forward
g
ing
D
ra
–6
Sw
Stance
+
5 cm
90 70 100 0 200 0 300
Ankle MTP
100 300 100
Hip abd /add
1.0
TA
0.3
VL
0.3
St
1.0
100 200 Cycle duration (%)
mV
0
MG
1s 1
TA EMG
1.0 0.5 0
Sideway
+
FW OW BW IW
4 cm
Foot motion during swing (cm)
c +
0
0
0.5 1.0 MG EMG
Mean (N)
Stand 40 0 20 40 60 80
For this purpose, an additional group of rats (n = 6) was trained to step for 3 weeks before testing (Supplementary Fig. 2). We first compared the effects of dynamic versus static patterns of sensory input. In the absence of treadmill motion, but under
Swing
6
Hip (deg) Knee
Backward
e
Normalized EMG amplitude
Extensor (MG) EMG amplitude (mV)
d
influence of sensory input on the initiation and control of standing and stepping in the absence of supraspinal input, we imposed sub stantial changes in the patterns of afferent information while the rats’ spinal circuits were engaged by the full combination of interventions.
St EMG
© 2009 Nature America, Inc. All rights reserved.
c
0 VL EMG
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
Figure 8 Effects of direction-dependent afferent input on motor patterns. (a–c) Representative example of mean (+ s.d.) hindlimb kinematics and raw EMG activity during continuous locomotion in the forward (a), backward (b) and sideways (c) directions. The same limb from the same rat is shown for the three directions, which corresponds to the leading (front) limb during sideward locomotion. Bar graphs show the average (n = 6 rats) linear distance traveled by the foot during swing with respect to the pelvis orientation for the different directions of stepping. Backward (BW) and forward (FW) motions correspond to displacements in the sagittal plane (defined by the pelvis orientation), whereas outward (OW) and inward (IW) motions correspond to displacements in the medio-lateral direction. Probability density distributions of normalized EMG amplitudes between the semitendinosus and medial gastrocnemius muscles, and the tibialis anterior and vastus lateralis muscles are shown at the bottom. L-shape patterns indicate reciprocal activation between the pair of muscles, whereas line-shape patterns indicate coactivation. Abd, abduction (increasing value); Add, adduction. Data are presented as in Figure 1, except that stick diagrams are represented in three dimensions, with the main plane oriented with the direction of the treadmill belt motion.
1339
© 2009 Nature America, Inc. All rights reserved.
a r t ic l e s weight-bearing conditions (37% ± 3% of weight bearing), pharmacological and EES interventions facilitated the tonic recruit ment of extensor muscles, whereas the flexor muscles were quiescent or weakly active, behaviorally apparent as standing (Fig. 6a). When treadmill belt motion (5 cm s−1) was initiated, all hindlimb joints underwent changes toward extension (limb moving backward), creat ing dynamic proprioceptive input that immediately transformed the motor patterns from a tonic to a rhythmic state (Fig. 6a). To further test the influence of velocity-dependent afferent input on motor pat tern formation, we incrementally increased the treadmill belt speed from 5 cm s−1 (slow walking) to 25 cm s−1 (running). Increasing tread mill speed induced a velocity-dependent lengthening of the stride (P < 0.01; Fig. 6a), an increase in stepping frequency (P < 0.001; Fig. 6a,b), a progressive decrease in the duration of the stance phase (P < 0.001; Fig. 6b) and of extensor EMG bursts (P < 0.001; Fig. 6c), and a progressive adjustment in the relative timing between hind limb segment oscillations (P < 0.05; Fig. 6d). In contrast, the dura tion of the swing phase (Fig. 6b) and of flexor bursts (Fig. 6c) were unchanged across speeds. Similar adjustments of the kinematics and EMG parameters were observed in noninjured rats (Supplementary Fig. 6). All of the rats with SCI (n = 6) accommodated displacement of the limbs and recruitment of motor pools to changing treadmill belt speeds in a single step and were able to locomote for extended periods of time on the treadmill even at the fastest speeds (25 cm s−1; Supplementary Fig. 7). When the treadmill belt was stopped at either slow or fast speeds, rhythmic hindlimb movements arrested instantly (Fig. 6a). Ongoing bursts in extensor muscles persisted as sustained tonic activity, whereas flexor muscles became quiescent, similar to what would occur in noninjured rats. Next, we assessed the influence of weight-bearing input on the for mation of motor patterns in rats with SCI (Fig. 7). When hindlimbs were suspended above the treadmill belt, that is, in the absence of load-related input, pharmacological and EES interventions induced step-like movements with alternate recruitment of extensor and flexor motor pools in all of the rats (n = 6; Fig. 7a). In the absence of belt motion, contact of the hindlimbs with the treadmill immediately arrested the rhythmic movements (Fig. 7a), stopped the recruitment of flexor muscles (P < 0.001; Fig. 7d), and induced tonic levels of EMG activity in extensor muscles (P < 0.001; Fig. 7d) that resulted in verti cal reaction forces (P < 0.001; Fig. 7e), enabling a sustained stand ing posture. Stepping movements continued if the treadmill belt was moving at the time of hindlimb contact, even with minimal weight bearing (Fig. 7b). Increasing the level of weight bearing resulted in a progressive increase in the amplitude of the extensor bursts (P < 0.001; Fig. 7b–d), an increase in the vertical reaction forces during stance (P < 0.001; Fig. 7b,c,e) and adjustments in hindlimb endpoint trajectories (longer stride, lower height) (P < 0.01; Fig. 7b). Finally, we assessed the effects of direction-dependent afferent input on locomotor pattern formation (Fig. 8). When reversing the direction of the treadmill belt from forward to backward, all of the rats (n = 6) showed alternate stepping movements with backward-oriented motion of the foot during swing (P < 0.001; Fig. 8b) and significant adaptations (timing and amplitude, P < 0.001) in muscle activity (Fig. 8b). Notably, a substantial coactivation of antagonist muscles was observed systematically during stance (Fig. 8b), as also occurred, although less frequently, in noninjured rats (Supplementary Fig. 6). In contrast, reciprocal recruitment of extensor and flexor motor pools was the only pattern observed during forward locomotion (Fig. 8a). Likewise, rotating the rat perpendicular to the treadmill belt direction did not arrest stepping movements. Instead, all of the rats (n = 6) showed sideways displacements of the feet (P < 0.001;
Fig. 8c) with a reversed pattern of hip abduction and adduction (P < 0.001; Fig. 8c) compared with forward locomotion (Fig. 8a). Although variable among rats, a major reorganization in the coor dination patterns of proximal and distal flexor and extensor motor pools was observed during sideways locomotion. Well-defined coac tivation patterns (Fig. 8c) between motor pools that were activated reciprocally during forward locomotion (Fig. 8a) also occurred dur ing sideways stepping. These results reveal a previously unrecognized ability of velocity-, load- and direction-dependent sensory inputs to regulate both the control of standing and the control of adaptive and flexible stepping movements with an exquisite degree of refinement and without any influences from supraspinal centers in adult rats (Supplementary Video 3).
1340
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
DISCUSSION Accessing the circuits and receptors in lumbosacral spinal cord Although several studies have documented the potential of pharma cological9,10 and/or electrical stimulation11–14,19,27, as well as motor training2,8,16, to facilitate stepping in rats with a complete spinal cord transection, limited weight-bearing capacities have been achieved, and when investigated in detail, kinematics and EMG patterns dif fered substantially from those underlying voluntary stepping. A major result of our study is that a combination of pharmacological and electrical stimulations was able to acutely transform rodent lumbo sacral circuits from nonfunctional to highly functional and adaptive states as early as 1 week post-injury. Furthermore, we found that, in conjunction with rehabilitative motor training, the same combined stimulations promoted the recovery of full weight-bearing hindlimb locomotion on a treadmill that was nearly indistinguishable from voluntary stepping, although robotic assistance was necessary for the rats to maintain balance. Together with previous evidence in cats24, our results indicate that the mammalian lumbosacral spinal cord con tains circuitry that is sufficient to generate close-to-normal hindlimb locomotor patterns in the absence of any supraspinal input. The mechanisms underlying locomotion induced by intraspinal stimulation13,14 or EES11,12,19,27 in rats and cats remain undetermined. Neurophysiological recordings28 and computer simulations29, how ever, suggest that the electrical stimulation engages spinal circuits primarily by recruiting afferent fibers. In fact, dorsal root stimulation can induce stepping similar to intraspinal stimulation in spinal cord– injured cats14. In humans, leg muscle vibration, which recruits largediameter afferent fibers, can evoke locomotor-like movements of the legs6. In principle, EES applied at discrete lumbosacral locations could recruit both long-range ascending and descending afferent branches in the segments below the spinal cord transection and activate pro fuse intraspinal ramifications that contact subsets of circuits localized around the stimulation site30. Accordingly, although electrical stimu lation of virtually any lumbosacral segment can facilitate step-like movements on a treadmill11, we found clear site-specific effects of EES on hindlimb locomotion. Consistent with the rostrocaudal ana tomical gradient of flexor and extensor motor pools, lumbar stimu lation facilitated flexion, whereas stimulation applied at the sacral level primarily facilitated extension. Distinct locations of serotonin receptors associated with stepping have also been reported in neo natal rats. A previous study18 found that in vitro brainstem-induced fictive locomotion can be blocked by antagonizing 5-HT2A or 5-HT7 receptors and that 5-HT7 receptors are prominent in the upper lum bar segments, whereas 5-HT2A receptors are more localized to the lower lumbar and sacral segments. Further studies will be required to identify the location and types of neurons, interneurons and receptors that are engaged by serotonergic agonists and electrical stimulation
© 2009 Nature America, Inc. All rights reserved.
a r t ic l e s during treadmill locomotion. Nevertheless, our statistical analyses indicate that each pharmacological or electrical intervention modu lates distinct functions, suggesting the fine-tuning of selective cir cuits. This specificity provides the means for the synergy between serotonergic agonist and electrical stimulation interventions, such that only combinations of both were able to functionally engage spi nal locomotor networks 1 week after injury and induce substantial functional improvements when combined with rehabilitative loco motor training. These findings support the viewpoint17,31,32 that the spinal motor infrastructure is composed of a widely distributed and heterogeneous, but highly integrated and synergistic, system of neural circuits and receptors that can generate a range of task-specific movements when recruited in different combinations. Evidence for such principles has been extracted statistically in humans33,34 and documented directly in anesthetized frogs31, rats32 and cats35. Our findings are consistent with and extend these organizational principles to the production of adaptive locomotor movements without any brain input in vivo. Consistent with this, the development of high-density electrode arrays that enable detailed, distributed and simultaneous access to the dif fusely located components of the lumbosacral circuits may offer the potential to promote a higher level of motor control than is currently possible in paralyzed subjects. Functional remodeling of the spinal locomotor circuitry A substantial reorganization of propriospinal circuits and spared descending fibers that leads to functional recovery occurs spontane ously after an incomplete SCI 3 and can be enhanced with locomo tor training36. However, there has been limited or no evidence for a similar plasticity of lumbosacral networks when the lesion inter rupts all supraspinal inputs in adult rats1,2 and humans37, contrary to functional improvements that have been repeatedly documented in cats with SCI8,15,16. We found that the combination of locomotor training with pharmacological and electrical stimulation markedly improved the functional capabilities of the sensorimotor circuits to sustain locomotion without supraspinal input. Although locomotor training with either pharmacological or electrical stimulation alone could reduce the deterioration of stepping ability observed in chronic untrained rats with SCI, only the combination of all three interven tion types could produce a substantial improvement. The decrease in the number of FOS-positive neurons, together with elevated syn aptic efficacy of monosynaptic inputs to both flexor and extensor motoneuron pools, suggests that locomotor training combined with serotonergic and electrical stimulations may reinforce selected spinal circuits and may suppress other, nonspecific circuitry. Use-depend ent selection and strengthening of neural circuits has been shown neurochemically16 and electrophysiologically20,21 after motor training in adult cats with SCI and rats with a neonatal SCI, as well as in the brain in conjunction with the practice of skilled movements38. Our results suggest that similar mechanisms underlie motor learning in the adult rat spinal cord bereft of brain input and that the extent of this functional remodeling is highly correlated with the degree of improvement of locomotor ability. We also found that the chronic absence of activity leads to a deterioration in stepping ability, as has been reported in cats15. These results corroborate clinical insights from individuals with severe SCI39 and emphasize the need to coun teract the rapid deterioration of function that occurs in the absence of weight bearing and motor activity after a SCI. Combined, these results indicate that the recovery of stepping ability after a spinal cord transection does not result simply from the activation of an ontogenetically defined hardwired and unmodified nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
circuitry that persists and recovers post-injury. Instead, specific combinations of locomotor training and pharmacological and elec trical stimulation induce de novo use-dependent functional states that enable spinal circuits to learn the motor tasks that are trained and practiced. Sensory control of stepping after the loss of brain input The recovery of hindlimb locomotion in animals with SCI is usu ally attributed to the neuronal networks responsible for central pat tern generation, that is, oscillatory output without brain or sensory input4,9,26. However, we found that near normal adaptive stepping and standing without any brain input emerged in vivo as a result of the capability of spinal neuronal networks to recognize and use taskspecific sensory input. We found that sensory information instantly transformed motor patterns in response to changing task, load, speed and direction conditions with a degree of flexibility and precision that has not been previously observed in animals with SCI24. Although the role of sensory information in feedback modulation of locomotor activity has been well recognized in cats with SCI15,16,24,25 as well as in humans40,41, our results suggest that the sensory informa tion is also instructive in a functional, primary feedforward manner. Indeed, we previously showed that pharmacological and electrical stimulations fail to elicit rhythmic recruitment of deafferented hind limb muscles in rats with SCI in vivo42 and a partial deprivation of cutaneous input strongly reduces the stepping ability of cats with SCI, but not in intact cats43. Robust changes in the functional properties of spinal locomotor circuits have been described in the lamprey when introducing sensory inputs44. However, these interactions between feedforward and feedback mechanisms have received little attention. Our results indicate that the ability of afferent information to mark edly reorganize functional connections among spinal sensorimotor pathways, both acutely and chronically, is an important property of the spinal motor infrastructure. Clinical perspectives Our findings have important implications for the understanding of motor pattern formation in the absence of supraspinal input in vivo and provide a conceptual framework for the design of strategies to ameliorate motor function in humans after SCI or in the context of other neuromotor disorders such as Parkinson’s disease45. Preliminary clinical studies have reported that EES can engage lumbosacral circuits of spinal cord–injured individuals7 and facilitate the recovery of functional walking46. Given the evidence that afferents can serve as a source of control for stepping and standing and can drive usedependent plasticity of locomotor networks in spinal cord–injured individuals37, combinatorial strategies may also promote recovery of motor function after severe SCI in humans. Such interventions could include neuroprosthetic spinal cord electrode arrays, custom ized pharmacological treatments and robotically-assisted locomotor training procedures. With the development of efficacious neural repair therapies, these neurorehabilitative strategies may become important for counteracting chronic deterioration of function39, optimizing the use of intrinsic lumbosacral circuits in regaining function37 and ensuring functional interactions between spinal mechanisms and regenerative fibers47,48. Methods Methods and any associated references are available in the online version of the paper at http://www.nature.com/natureneuroscience/. Note: Supplementary information is available on the Nature Neuroscience website.
1341
a r t ic l e s Acknowledgments We would like to acknowledge the excellent technical help provided by S. Zdunowski, L. Friedli, J. Heutschi and O. Märzendorfer for data collection and analysis, as well as M. Herrera for his expert assistance and guidance involving the care and handling of the animals. This work was supported by the Craig H. Nielsen Foundation (#20062668), the International Paraplegic Foundation (P106), the National Center of Competence in Research ‘Neural Plasticity and Repair’ of the Swiss National Science Foundation, the Christopher and Dana Reeve Foundation (VEC-2007), the US National Institute of Neurological Disorders and Stroke (NS16333), the US Civilian Research and Development Foundation (RUB1-287207), the Roman Reed Spinal Cord Injury Research Fund of California and the Russian Foundation for Basic Research (07-04-00526 and 08-04-00688).
1. Kubasak, M.D. et al. OEG implantation and step training enhance hindlimb-stepping ability in adult spinal transected rats. Brain 131, 264–276 (2008). 2. Ichiyama, R.M. et al. Step training reinforces specific spinal locomotor circuitry in adult spinal rats. J. Neurosci. 28, 7370–7375 (2008). 3. Courtine, G. et al. Recovery of supraspinal control of stepping via indirect propriospinal relay connections after spinal cord injury. Nat. Med. 14, 69–74 (2008). 4. Grillner, S. Biological pattern generation: the cellular and computational logic of networks in motion. Neuron 52, 751–766 (2006). 5. Kiehn, O. Locomotor circuits in the mammalian spinal cord. Annu. Rev. Neurosci. 29, 279–306 (2006). 6. Gurfinkel, V.S., Levik, Y.S., Kazennikov, O.V. & Selionov, V.A. Locomotor-like movements evoked by leg muscle vibration in humans. Eur. J. Neurosci. 10, 1608–1612 (1998). 7. Dimitrijevic, M.R., Gerasimenko, Y. & Pinter, M. Evidence for a spinal central pattern generator in humans. Ann. NY Acad. Sci. 860, 360–376 (1998). 8. Chau, C., Barbeau, H. & Rossignol, S. Early locomotor training with clonidine in spinal cats. J. Neurophysiol. 79, 392–409 (1998). 9. Antri, M., Mouffle, C., Orsal, D. & Barthe, J.Y. 5–HT1A receptors are involved in short- and long-term processes responsible for 5-HT-induced locomotor function recovery in chronic spinal rat. Eur. J. Neurosci. 18, 1963–1972 (2003). 10. Landry, E.S. et al. Contribution of spinal 5–HT1A and 5–HT7 receptors to locomotorlike movement induced by 8-OHDPAT in spinal cord–transected mice. Eur. J. Neurosci. 24, 535–546 (2006). 11. Ichiyama, R.M., Gerasimenko, Y.P., Zhong, H., Roy, R.R. & Edgerton, V.R. Hindlimb stepping movements in complete spinal rats induced by epidural spinal cord stimulation. Neurosci. Lett. 383, 339–344 (2005). 12. Gerasimenko, Y.P. et al. Epidural spinal cord stimulation plus quipazine administration enable stepping in complete spinal adult rats. J. Neurophysiol. 98, 2525–2536 (2007). 13. Guevremont, L. et al. Locomotor-related networks in the lumbosacral enlargement of the adult spinal cat: activation through intraspinal microstimulation. IEEE Trans. Neural Syst. Rehabil. Eng. 14, 266–272 (2006). 14. Barthélemy, D., Leblond, H. & Rossignol, S. Characteristics and mechanisms of locomotion induced by intraspinal microstimulation and dorsal root stimulation in spinal cats. J. Neurophysiol. 97, 1986–2000 (2007). 15. De Leon, R.D., Hodgson, J.A., Roy, R.R. & Edgerton, V.R. Retention of hindlimb stepping ability in adult spinal cats after the cessation of step training. J. Neurophysiol. 81, 85–94 (1999). 16. Tillakaratne, N.J. et al. Use-dependent modulation of inhibitory capacity in the feline lumbar spinal cord. J. Neurosci. 22, 3130–3143 (2002). 17. Hochman, S., Garraway, S., Machacek, D. & Shay, B. 5-HT receptors and the neuromodulatory control of spinal cord function. in Motor Neurobiology of the Spinal Cord (ed. Cope, T.C.) 47–87 (CRC Press, Boca Raton, Florida, 2001). 18. Liu, J. & Jordan, L.M. Stimulation of the parapyramidal region of the neonatal rat brain stem produces locomotor-like activity involving spinal 5-HT7 and 5-HT2A receptors. J. Neurophysiol. 94, 1392–1404 (2005). 19. Lavrov, I. et al. Epidural stimulation induced modulation of spinal locomotor networks in adult spinal rats. J. Neurosci. 28, 6022–6029 (2008).
20. Petruska, J.C. et al. Changes in motoneuron properties and synaptic inputs related to step training after spinal cord transection in rats. J. Neurosci. 27, 4460–4471 (2007). 21. Côté, M.P. & Gossard, J.P. Step training–dependent plasticity in spinal cutaneous pathways. J. Neurosci. 24, 11317–11327 (2004). 22. Barthélemy, D., Leblond, H., Provencher, J. & Rossignol, S. Nonlocomotor and locomotor hindlimb responses evoked by electrical microstimulation of the lumbar cord in spinalized cats. J. Neurophysiol. 96, 3273–3292 (2006). 23. Lavrov, I. et al. Plasticity of spinal cord reflexes after a complete transection in adult rats: relationship to stepping ability. J. Neurophysiol. 96, 1699–1710 (2006). 24. Rossignol, S., Dubuc, R. & Gossard, J.P. Dynamic sensorimotor interactions in locomotion. Physiol. Rev. 86, 89–154 (2006). 25. Pearson, K.G. Generating the walking gait: role of sensory feedback. Prog. Brain Res. 143, 123–129 (2004). 26. Barrière, G., Leblond, H., Provencher, J. & Rossignol, S. Prominent role of the spinal central pattern generator in the recovery of locomotion after partial spinal cord injuries. J. Neurosci. 28, 3976–3987 (2008). 27. Iwahara, T., Atsuta, Y., Garcia-Rill, E. & Skinner, R.D. Spinal cord stimulationinduced locomotion in the adult cat. Brain Res. Bull. 28, 99–105 (1992). 28. Gaunt, R.A., Prochazka, A., Mushahwar, V.K., Guevremont, L. & Ellaway, P.H. Intraspinal microstimulation excites multisegmental sensory afferents at lower stimulus levels than local alpha-motoneuron responses. J. Neurophysiol. 96, 2995–3005 (2006). 29. Rattay, F., Minassian, K. & Dimitrijevic, M.R. Epidural electrical stimulation of posterior structures of the human lumbosacral cord. 2. Quantitative analysis by computer modeling. Spinal Cord 38, 473–489 (2000). 30. Rudomin, P. Central control of information transmission through the intraspinal arborizations of sensory fibers examined 100 years after Ramon y Cajal. Prog. Brain Res. 136, 409–421 (2002). 31. Giszter, S.F., Mussa-Ivaldi, F.A. & Bizzi, E. Convergent force fields organized in the frog’s spinal cord. J. Neurosci. 13, 467–491 (1993). 32. Tresch, M.C. & Bizzi, E. Responses to spinal microstimulation in the chronically spinalized rat and their relationship to spinal systems activated by low-threshold cutaneous stimulation. Exp. Brain Res. 129, 401–416 (1999). 33. Courtine, G. & Schieppati, M. Tuning of a basic coordination pattern constructs straight-ahead and curved walking in humans. J. Neurophysiol. 91, 1524–1535 (2004). 34. Ivanenko, Y.P., Cappellini, G., Dominici, N., Poppele, R.E. & Lacquaniti, F. Modular control of limb movements during human locomotion. J. Neurosci. 27, 11149–11161 (2007). 35. Lemay, M.A. & Grill, W.M. Modularity of motor output evoked by intraspinal microstimulation in cats. J. Neurophysiol. 91, 502–514 (2004). 36. Engesser-Cesar, C. et al. Wheel running following spinal cord injury improves locomotor recovery and stimulates serotonergic fiber growth. Eur. J. Neurosci. 25, 1931–1939 (2007). 37. Harkema, S.J. Plasticity of interneuronal networks of the functionally isolated human spinal cord. Brain Res. Rev. 57, 255–264 (2008). 38. Rioult-Pedotti, M.S., Friedman, D. & Donoghue, J.P. Learning-induced LTP in neocortex. Science 290, 533–536 (2000). 39. Dietz, V. & Muller, R. Degradation of neuronal function following a spinal cord injury: mechanisms and countermeasures. Brain 127, 2221–2231 (2004). 40. Harkema, S.J. et al. Human lumbosacral spinal cord interprets loading during stepping. J. Neurophysiol. 77, 797–811 (1997). 41. Beres-Jones, J.A. & Harkema, S.J. The human spinal cord interprets velocitydependent afferent input during stepping. Brain 127, 2232–2246 (2004). 42. Lavrov, I. et al. Facilitation of stepping with epidural stimulation in spinal rats: role of sensory input. J. Neurosci. 28, 7774–7780 (2008). 43. Bouyer, L.J. & Rossignol, S. Contribution of cutaneous inputs from the hindpaw to the control of locomotion. II. Spinal cats. J. Neurophysiol. 90, 3640–3653 (2003). 44. Grillner, S. & Wallen, P. Cellular bases of a vertebrate locomotor system-steering, intersegmental and segmental coordination and sensory control. Brain Res. Brain Res. Rev. 40, 92–106 (2002). 45. Fuentes, R., Petersson, P., Siesser, W.B., Caron, M.G. & Nicolelis, M.A. Spinal cord stimulation restores locomotion in animal models of Parkinson’s disease. Science 323, 1578–1582 (2009). 46. Carhart, M.R., He, J., Herman, R., D’Luzansky, S. & Willis, W.T. Epidural spinal-cord stimulation facilitates recovery of functional walking following incomplete spinalcord injury. IEEE Trans. Neural Syst. Rehabil. Eng. 12, 32–42 (2004). 47. Courtine, G. et al. Can experiments in nonhuman primates expedite the translation of treatments for spinal cord injury in humans? Nat. Med. 13, 561–566 (2007). 48. Maier, I.C. et al. Differential effects of anti-Nogo-A antibody treatment and treadmill training in rats with incomplete spinal cord injury. Brain 132, 1426–1440 (2009).
1342
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
AUTHOR CONTRIBUTIONS G.C., Y.P.G., I.L., M.V.S. and V.R.E. designed the study. G.C., R.R.R., H.Z. and P.M. performed the surgeries. G.C., Y.P.G., R.v.d.B., P.M. and A.Y. carried out the experiments. G.C., A.Y., B.S., Y.A., R.I. and M.V.S. conducted the anatomical assessments. G.C. analyzed the data. G.C., R.R.R., M.V.S. and V.R.E. wrote the manuscript. G.C. supervised the study.
© 2009 Nature America, Inc. All rights reserved.
Published online at http://www.nature.com/natureneuroscience/. Reprints and permissions information is available online at http://www.nature.com/ reprintsandpermissions/.
ONLINE METHODS
© 2009 Nature America, Inc. All rights reserved.
Animals and animal care. All procedures followed the US National Institutes of Health’s Guide for the Care and Use of Laboratory Animals and were approved by the Animal Use Committee at the University of California, Los Angeles and the Veterinary Office of the Canton of Zurich. These experiments were conducted on adult female Sprague-Dawley rats (~300 g), housed individually on a 12-h light/dark cycle with access to food and water ad libitum. Supplementary Figure 2 describes the time line of the surgical and experimental procedures, as well as the different groups of rats that were included in the present study. Surgical procedures and post-surgical care. All procedures have been described in detail previously2,12,19. Briefly, a combination of ketamine (100 mg per kg) and xylazine (10 mg per kg) was used to induce anesthesia, which subsequently was maintained with isoflurane (1–2.5%) via facemask. Under aseptic conditions, a partial laminectomy was performed over spinal segments L2 and S1. Tefloncoated stainless steel wires (AS632, Cooner Wire) were passed under the spinous processes and above the dura matter of the remaining vertebrae between the partial laminectomy sites. After removing a small portion (~1 mm notch) of the Teflon coating to expose the stainless steel wire on the surface facing the spinal cord, we secured the electrodes at the midline of the spinal cord at spinal level L2 and S1 by suturing the wire to the dura mater above and below the electrode (Supplementary Fig. 1). A common ground wire (~1 cm of the Teflon removed at the distal end) was inserted subcutaneously in the mid-back region. Bipolar intramuscular EMG electrodes using the same wire type as above were inserted bilaterally in the mid-belly of the soleus and deep (close to the bone) mid-belly of the tibialis anterior muscles, or unilaterally into the deep mid-belly of the vastus lateralis, distal deep compartment of the semitendinosus and medial deep region of the medial gastrocnemius. All electrode wires were connected to a percutaneous amphenol connector cemented to the skull of the rat. The proper location of the epidural and EMG electrodes was verified post-mortem. After 2– 3 weeks of recovery from surgery, baseline recordings were obtained (see below). The rats then underwent a second surgical procedure to receive the SCI. A partial laminectomy was made at a mid-thoracic level (~T7) and the spinal cord was completely transected. Gel foam was inserted into the gap created by the transec tion as a coagulant and to separate the cut ends of the spinal cord. The complete ness of spinal cord transections was verified by two surgeons by lifting the cut ends of the cord during the surgery as well as histologically post-mortem. Pharmacological and EES interventions. To encourage locomotion via EES, we delivered rectangular pulses (0.2 ms duration) at 40–50 Hz19 using a dual-output Grass S88 stimulator (Grass Instruments) through two constant-voltage isola tion units (Grass SIU5, Grass Instruments) connected to the L2 and S1 epidural electrodes. The stimulation intensity (1–4 V) was adjusted to obtain optimal facilitation of stepping visually. Pharmacological modulation was induced by administrating quipazine (0.2–0.3 mg per kg) intraperitoneally and 8-OHDPAT (0.05–0.3 mg per kg) subcutaneously9. Serotonin agonists were injected 10 and 15 min before locomotor training and behavioral testing, respectively. Locomotor training. An upper-body harness was used to position the rats over a treadmill belt and to partially support their body weight during bipedal loco motion (Supplementary Fig. 1). An automated, servo-controlled body-weight support system measured and provided the optimal amount of body-weight support that each rat needed to step. Rats were trained every other day, 20 min per session for 8 weeks, starting 8 d after spinal cord transection. Locomotor training was enabled by EES at S1 and L2, by quipazine and 8-OHDAPT, and by the combination of both. Treadmill belt speed was set at 9 cm s−1. Nontrained rats were housed similarly to the trained rats, but did not receive serotonin agonists or EES and were not placed in the treadmill apparatus, except for the final testing session when all rats were tested under the same conditions. Spinal cord reflexes. Monosynaptic motor potentials were evoked during bipedal standing with constant weight bearing (20% of body weight) by deliv ering rectangular pulses (0.5-ms duration) through the S1 electrode at 0.2 Hz49. Before the injury, we identified the stimulus intensity that elicited the largest monosynaptic responses in the absence of direct muscle responses (direct stimu lation of the motor nerve, equivalent to M waves), typically 1.5–2-fold greater than motor threshold49, and used this intensity to test reflexes at 1 and 9 weeks
doi:10.1038/nn.2401
post-injury. The efficacy of monosynaptic inputs to flexor and extensor motone uron pools was measured as the peak-to-peak amplitude of averaged (n = 10) motor-evoked potentials recorded from the tibialis anterior and soleus muscles bilaterally (10 kHz). Kinematics and EMG analyses of locomotion. Three-dimensional video recordings (100 Hz) were made using four cameras (Basler Vision Technologies) oriented at 45° and 135° bilaterally with respect to the direction of locomo tion or the motion capture system VICON by means of 8 infrared television cameras (200 Hz). Reflective markers were attached bilaterally at the iliac crest, greater trochanter, lateral condyle, lateral malleolus, MTP and the tip of the toe (Supplementary Fig. 1). SIMI motion capture software (SIMI Reality Motion Systems) and Nexus (Vicon) were used to obtain three-dimensional coordinates of the markers. The body was modeled as an interconnected chain of rigid seg ments and joint angles were generated accordingly. EMG signals (2 kHz) were amplified, filtered (10–1,000-Hz bandpass), stored and analyzed off-line to compute the amplitude, duration and timing of indi vidual bursts. To evaluate coactivation between muscles, we generated probability density distributions of normalized EMG amplitudes of specific pairs of muscles during continuous treadmill stepping sequences, as described previously3. In some recordings (Fig. 4), vertical reaction forces were monitored using a biomechanical force plate (2 kHz, HE6X6, AMTI) located below the treadmill belt (Supplementary Fig. 1). During the long bouts of recording under different load and speed conditions, a rigid circular (diameter, 0.5 mm) stick was inserted between the rat’s limbs to prevent them from crossing. Ten successive step cycles were extracted for both the left and right hindlimbs from a continuous sequence of treadmill stepping for each rat under each condition. A 10-s interval was used when no stepping movements were observed. A total of 135 parameters quantifying gait timing, joint kinematics, limb endpoint trajectory and EMG activity were computed for each gait cycle according to methods described previously (Supplementary Table 1)2,3,50. Gait parameters. Gait cycles were defined as the time interval between two suc cessive paw contacts of one limb. Successive paw contacts were visually defined by the investigators with an accuracy of ±1 video frame (100 Hz). The onsets of the swing phases were set at the zero crossings of the rate of change of the eleva tion angle of the limb axis (virtual segment connecting the crest to the MTP), that is, at the onset of forward oscillation. Cycle duration and stance and swing durations were determined from the kinematic recordings. Footfall patterns were used to compute the coupling between the hindlimbs. In particular, the time at which the contralateral limb contacted the treadmill belt was expressed as a percentage of the duration of the ipsilateral gait cycle and represented in polar coordinates (Fig. 1). On these polar representations, 50% indicates that the limbs move out of phase, whereas 0% typically corresponds to jumps after which both feet contact the treadmill belt simultaneously or to the absence of movements. Interlimb coordination was statistically computed as the r value at t = 0 of the cross-correlation function between the oscillation of the left and right limb axis. An r value of −0.5 indicates that both hindlimbs oscillate perfectly out of phase. The variability of all parameters was measured as the coefficient of variation computed separately for left and right hindlimbs. Limb endpoint trajectory. Limb endpoint trajectories were analyzed on the basis of the xyz motion of the MTP marker. The computed spatial parameters included stride length, three-dimensional linear path, step height, and distance at swing and stance onset between the limb endpoint and the hip in the direction of walk ing50. The extent of paw dragging during swing was computed as the time during which the paw was in contact with the treadmill belt after swing onset. Intensity and direction of foot velocity at swing onset were computed as the module and pitch angle of the MTP velocity vector in the sagittal plane (direction of walk ing), respectively. Limb endpoint acceleration at swing onset and maximum limb endpoint velocity during swing were measured. We used principal component analysis to quantify spatial consistency of hindlimb endpoint trajectory. Spatial coordinates of hindlimb endpoint trajectory were extracted from the selected sequence of stepping and separated into their x, y and z components. For each gait cycle, each step component was resampled to 100 data points, thus removing the temporal information. The data were then arranged into three m × n matrices, with each column containing the data for a single step and each row containing
nature NEUROSCIENCE
the interpolated position values at each time step. Matlab (Mathworks) scripts were written to extract principal components from each dataset. The consistency of limb endpoint trajectory during swing was measured as the amount of variance explained by the first principal component.
© 2009 Nature America, Inc. All rights reserved.
Stability. Stance width was measured for each gait cycle as the perpendicular distance (medio-lateral plane) between left and right MTP markers at the time of paw contact. The degree of pelvis stability was evaluated as the s.d. of pelvis oscillations in the sagittal and medio-lateral planes during the analyzed sequence of stepping. Hindlimb kinematics. Maximum joint angle positions in flexion and extension were computed as the minimum and maximum values of each angle over the time course of each gait cycle, respectively. The amplitude of joint angular excur sions was measured as the difference between maximum positions in flexion and extension. We used cross-correlation functions to assess the coordination among joints. Cross-correlations among all unique pairs of degrees of freedom (that is, hip, knee, ankle and MTP) were calculated and maximum r values extracted to quantify the degree of correlation among a given pair of joints1. The timing in the coupling between oscillations of adjacent segments (thigh, leg, foot and toe) was quantified for each gait cycle as the relative lag at the maximum r value of the cross-correlation function between both angles. To quantify the degree of similitude in the profile of joint angles and joint angle velocities between pre- and post-injury stepping patterns, we computed cross-correlation functions between mean waveforms. The procedure was repeated for each rat, joint angle, joint angle velocity and side. The highest positive r value determined the degree of similitude between pre- and post-injury waveforms. Variability of joint motion. To assess the variability of joint angles across con secutive gait cycles, we computed the s.d. of time-normalized joint angles every 10% of the normalized time base. Variability of each joint angle from both left and right sides was measured as the mean of s.d. PCA. Performance of gait implies the rhythmic repetition of stereotypical pat terns of leg motion. Several parameters need to be used to thoroughly character ize a given gait pattern; here, we computed 135 variables that provided detailed quantification of kinematic and EMG features underlying specific patterns of locomotion. Such high dimensionality substantially complicates the extraction of relevant parameters to account for differences between experimental condi tions. However, reduction of such multidimensional datasets can be achieved via multivariate statistical analysis such as PCA. PCA is mathematically defined as an orthogonal linear transformation that transforms the original dataset to a new coordinate system, such that the variance is maximized on each new coordinate axis. Here, data were analyzed using the correlation method, which adjusts the mean of the data to 0 and the s.d. to 1. This is a conservative method and is appropriate for variables that differ in their variance. For each rat and condition, we computed the averaged values of the 135 original variables (Supplementary Table 1) and ordered them in a matrix, with the variables along the columns and the observations (data from each rat under a given condition) along the rows. Principal components were then extracted on different sets of conditions (for example, combinations of interventions or different locomotor training condi tions). A few principal components, typically three, were sufficient to account for more than 50% of the total variance of the studied samples, demonstrating the high correlation between the variables describing gait. The degree of similarities and differences between the rats and conditions was evaluated as the difference in the factor coordinates of each observation (scores) on each principal compo nent axis. To visualize the differences between rats and conditions, we plotted coordinates from each rat under a specific condition (observation) in the new space created by the first three principal components. In this representation, the distance between the data points increases (nonlinearly) with the difference between the underlying locomotor characteristics, allowing us to readily visualize
nature NEUROSCIENCE
differences and similarities between rats and conditions. Next, we identified the variables that contributed to account for the differences between the experimental conditions. This classification is based on the analysis of factors loadings, that is, correlations between each variable and each principal component. To display the results of this analysis, we created a color-coded representation (Figs. 1n and 4f) of correlation values for all the computed variables with principal components. In this representation, variables that correlate positively or negatively with a given principal component identified clusters of parameters that account for a specific difference between conditions. Accordingly, each principal component tends to explain distinct differences between rats and experimental conditions. FOS immunohistochemistry. FOS immunoreactivity of spinal neurons was deter mined as previously described2. Briefly, after a 45-min bout of continuous hind limb bipedal stepping under pharmacological (quipazine and 8-OHDPAT) and EES (S1 and L2) interventions, the rats were returned to their cages and were anes thetized and killed by intracardial perfusion of 4% paraformaldehyde (wt/vol) in phosphate buffer about 60 min later. After perfusion, the spinal cords were grossly dissected with the vertebral column and post-fixed overnight (4 °C). Spinal cords then were dissected carefully and cryoprotected in 30% sucrose (wt/vol) in phos phate buffered saline (PBS) for at least 3 d. Segments L1 through S2 were mounted and frozen and 30-µm coronal sections were obtained using a cryostat. We proc essed one section of each spinal segment for locomotor-trained, nontrained and noninjured rats simultaneously for FOS immunoreactivity. Free-floating sections were submerged in 0.3% H2O2 (wt/vol, 10 min) and subsequently washed in PBS (three times for 10 min). Sections were incubated in the primary antibody to the c-fos protein product FOS (rabbit, polyclonal; Santa Cruz Biotechnology) at 1:1,000 for 48 h (4 °C). Sections then were washed in PBS (30 min) and incu bated in the secondary antibody (biotinylated goat antibody to rabbit; Jackson ImmunoResearch) at 1:100 for 1 h. After PBS washes (30 min), sections were incu bated for 1 h in a 1:100 Vectastain Elite ABC (avidin–biotin peroxidase complex) solution (Vector Laboratories). Sections were washed again in PBS and reacted with diaminobenzidine solution (Sigma-Aldrich) for about 2–3 min. Finally, sections were rinsed in PBS (30 min), mounted on microscope slides, air dried, counterstained with Methyl Green and coverslipped with Permount. FOS-positive nuclei were counted manually using camera lucida techniques and superimposed onto Molander’s cytoarchitectonic maps of the rat lumbosacral cord. Counts were performed blindly for each spinal segment and for each rat. Muscle weights. After perfusion, the left hindlimb was dissected blindly and the following muscles and muscle groups extracted: triceps surae (soleus, medial gastrocnemius, lateral gastrocnemius), quadriceps (rectus femoris, vastus latera lis, vastus medialis, vastus intermedius), tibialis anterior, plantaris and extensor digitorum longus. After removal of excess fat and connective tissues, the muscles or muscle groups were weighed and normalized to the body weight of the rat at the time of perfusion. Statistical analyses . All data are reported as mean values ± s.e.m. One way ANOVAs, repeated-measures ANOVAs or one way repeated-measures ANOVAs were used, depending on the conditions, to test differences between groups and/or conditions on the experimental parameters. The factors examined were the presence of EES (L2 and/or S1) and serotonin agonists (quipazine and/or 8-OHDPAT), body sides (left, right), time points (pre-injury and 1 and 9 weeks post-injury), training status (nontrained, trained), and task conditions (stand ing, stepping, load, speed, direction). Post hoc differences were assessed using the Newman-Keuls test. The software package Statistica was used for all analyses. 49. Gerasimenko, Y.P. et al. Spinal cord reflexes induced by epidural spinal cord stimulation in normal awake rats. J. Neurosci. Methods 157, 253–263 (2006). 50. Courtine, G. et al. Performance of locomotion and foot grasping following a unilateral thoracic corticospinal tract lesion in monkeys (Macaca mulatta). Brain 128, 2338–2358 (2005).
doi:10.1038/nn.2401
a r t ic l e s
Classical conditioning in the vegetative and minimally conscious state
© 2009 Nature America, Inc. All rights reserved.
Tristan A Bekinschtein1–3,6,7, Diego E Shalom1,4,5,7, Cecilia Forcato3, Maria Herrera5, Martin R Coleman2, Facundo F Manes1,3,4 & Mariano Sigman5 Pavlovian trace conditioning depends on the temporal gap between the conditioned and unconditioned stimuli. It requires, in mammals, functional medial temporal lobe structures and, in humans, explicit knowledge of the temporal contingency. It is therefore considered to be a plausible objective test to assess awareness without relying on explicit reports. We found that individuals with disorders of consciousness (DOCs), despite being unable to report awareness explicitly, were able to learn this procedure. Learning was specific and showed an anticipatory electromyographic response to the aversive conditioning stimulus, which was substantially stronger than to the control stimulus and was augmented as the aversive stimulus approached. The amount of learning correlated with the degree of cortical atrophy and was a good indicator of recovery. None of these effects were observed in control subjects under the effect of anesthesia (propofol). Our results suggest that individuals with DOCs might have partially preserved conscious processing, which cannot be mediated by explicit reports and is not detected by behavioral assessment. A pressing issue in neuroscience, with important theoretical and practical implications, is whether an objective manner or reliable test can be established to assess awareness without relying on explicit reports1,2. A specific implementation of the classical conditioning procedure, trace conditioning, is considered to be one of the best candidates for such a test1,3. Classical conditioning is a simple form of associative learning in which contingencies are established between a behaviorally important stimulus (unconditioned stimulus, UCS) and a closely paired neutral stimulus (conditioned stimulus). In the trace conditioning of the eyeblink response, the conditioned stimulus is a tone that is presented several hundred milliseconds before the UCS, which is an air puff to the cornea. The temporal demand imposed by the silent trace interval between both stimuli has been shown to engage a broad cerebral network, including the cerebellum, neocortex and hippocampus4–6, and to require awareness of the contingencies between stimuli7–9. Subjects with bilateral mediotemporal lobe lesions fail to acquire the conditioning as well as normal subjects that did not show explicit knowledge of contingencies between stimuli4,6. DOCs10 describe a heterogeneous group of individuals who have survived severe brain damage. The clinical assessment of these individuals is challenged with the task of behaviorally distinguishing those who demonstrate no evidence of awareness (vegetative state) from those who demonstrate inconsistent, but reproducible, evidence of awareness (minimally conscious state, MCS). Vegetative state patients typically emerge from coma and remain in a state in which they seem to be awake, but are unaware of themselves and/or their environment11. They have a preserved capacity for spontaneous or
stimulus-induced arousal, but no evidence of purposeful (voluntary) behavior in response to visual, auditory, tactile or noxious stimuli12,13. The MCS differs from the vegetative state by the presence of inconsis tent, but reproducible, purposeful behavior, which might include a response to command. In the upper bound of the DOC spectrum, an MCS patient who starts to communicate is reclassified as being severely disabled (SED)14. The neuroscience community has recently made great efforts to facilitate the assessment of individuals with DOCs by developing objective tools through which an individual’s level of retained awareness can be assessed without requiring an overt movement on their part. Brain integrative processing, a proposed prerequisite of awareness15,16, has been observed in MCS patients using functional imaging procedures presenting simple tones17,18 and speech19–23. In the vegetative state, brain activations are mostly restricted to primary sensory cortices17,18,24–26. However, some exceptions have been found. Three vegetative state patients activated the inferior frontal gyrus and posterior superior temporal gyrus in response to speech stimuli 27, individuals in the vegetative state have shown differential responses to syntactic violations20 and a single case report of an individual in the vegetative state showed specific brain activations in functional magnetic resonance imaging (fMRI) when asked to imagine playing tennis in her head or to imagine moving around her house28. Here, we sought to test whether individuals with DOCs have the capacity to learn trace conditioning to determine whether these individuals might exhibit partially preserved conscious processing, which cannot be mediated by explicit reports. We found that individuals
1Institute
of Cognitive Neurology, Buenos Aires, Argentina. 2Impaired Consciousness Research Group, Department of Clinical Neurosciences, University of Cambridge, Cambridge, UK. 3Raul Carrea Institute, Buenos Aires, Argentina. 4Institute of Neuroscience, Favaloro Foundation, Buenos Aires, Argentina. 5Laboratory of Integrative Neuroscience, Physics Department, University of Buenos Aires, Argentina. 6Present address: The Medical Research Council, Cognition and Brain Sciences Unit, Cambridge, UK. 7These authors contributed equally to this work. Correspondence should be addressed to T.A.B. (
[email protected]). Received 12 June; accepted 3 August; published online 20 September 2009; doi:10.1038/nn.2391
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1343
a r t ic l e s Figure 1 Stimulus design and different stages of the EMG response. The procedure consisted of a total of 140 conditioning trials (70 conditioned stimulus (CS+) and 70 unpaired tones (CS−)). Tones lasted for 250 ms. The conditioned stimulus was followed, after an interval of 500 ms, by an aversive stimulus, the UCS. We divided the EMG response from the eye into four temporal intervals: the period before tone onset (baseline, −400 to 0 ms), a short period that started with the onset of the tone and lasted for 300 ms (α response), a subsequent period (anticipatory interval) following this transient response and before the presentation of the puff (conditioned response) and the period following the onset of the air puff (unconditioned response). The EMG signal was rectified and normalized, on a trial by trial basis, to the s.d. of the baseline window for each trial. We refer to this as the nEMG, measured in s.d.
RESULTS Subjects were trained with 140 trials of a trace-conditioning eye-blink response procedure (70 tones paired with an airpuff (conditioned stimulus) and 70 unpaired tones). Only the conditioned stimulus was followed, after an interval of 500 ms, by an aversive stimulus, the UCS (Fig. 1). The main group of the study, the DOC group, included 22 individuals with DOCs who met recognized criteria defining the vegetative state (12, 13), MCS or SED (14). We carried out the same experimental procedure in two control groups: a control to assess learning in subjects with intact conscious processing, the controlled consciousness group (n = 16), and a controlled unconsciousness group (n = 12), which was under the influence of the anesthetic agent propofol during standard endoscopic procedures. The controlled
a
BL
α
EA LA
CS+ CS– 4
nEMG
3 2 1 0
–200 0 200 400 600 Time (ms)
b
c Counts
6
CC
4 2 BL
Counts
DOC
2 1 BL 2.0
Counts
1.0 0.5
1344
BL
EA LA
d
CS– 15
10
10
5
5 1
2 3 4 >5 nEMG value
0
15
15
10
10
5
5
0
EA LA
1.5
CS+ 15
0
EA LA
3
CU
© 2009 Nature America, Inc. All rights reserved.
with DOCs were able to learn trace-conditioning associations, that learning was specific to stimulus and temporal contingencies, and that learning was a good indicator of recovery.
1
2 3 4 >5 nEMG value
0
15
15
10
10
5
5
0
1
2 3 4 >5 nEMG value
0
1
2
3
4
>5
1
2
3
4
>5
1
2
3
4
>5
100 ms
250 ms 500 ms
CS+
250 ms CS–
Baseline
Alpha
Early-late anticipatory
Unconditioned response
unconsciousness group was included in this study as a negative control, that is, a baseline for comparison in a group in which we expected no trace learning. It must be emphasized, however, that anesthesia leads to a broad state change that affects distinct factors, including a general amnesia that must be considered when comparing these groups29. Learning during the anticipatory interval We first measured the event-related response (ERR), averaging the normalized electromyographic (nEMG) response across all trials, for the conditioned stimulus and unpaired tones (Fig. 2a). The ERR reflects the magnitude and the specificity of learning. An increase in the anticipatory interval as compared with baseline is indicative of learning; if this increase is present for the conditioned stimulus, but not for the unpaired tones, then learning is specific. We examined the ERR of a subject who showed specific learning in the controlled consciousness group (Fig. 2a). The conditioned stimulus response ramped during the anticipatory interval. This effect was tuned close to the onset of the air puff, reflecting the temporal specificity of the contingency. These observations in a single subject serve as a guide for the subsequent analysis. For each experimental group, we calculated the average nEMG activity during three segments of the response: baseline, early (first half) and late (second half) in the anticipatory interval. Averaged across the controlled consciousness group, the nEMG response increased during the anticipatory interval. This increase was specific to the conditioned stimulus and was more pronounced in the second half of the anticipatory interval (Fig. 2b). The DOC group showed the same pattern, although the main effect and the stimulus specificity were less pronounced. In contrast with these observations, we did not see any trace of learning in the controlled unconsciousness group. This observation summarizes our main finding: individuals with DOCs, as a group, showed learning during the anticipatory period, which was restricted to the few hundred milliseconds before the puff and thus reflected the temporal specificity of the contingency. When compared with the conscious control subjects, the effect size and the specificity were considerably reduced. Figure 2 Learning during the anticipatory interval. ERRs of nEMG activity in response to conditioned stimulus (black) and unpaired tones (gray). (a) Example of a representative normal volunteer showing significant learning (P < 0.05). The conditioned stimulus response was larger in the anticipatory interval and this effect was more pronounced in the second half of this interval, close to the onset of air puff. (b) Average nEMG activity for each group (control consciousness (CC), DOC subjects and controlled unconsciousness (CU)) during three segments of the response: baseline and the early (EA) and late (LA) anticipatory intervals. The conditioned stimulus response was larger in the late anticipatory interval for the controlled consciousness and DOC groups. The response in the subjects was weaker (note the change of scale of nEMG). The controlled unconsciousness group showed no trace of anticipatory learning. (c,d) Individual contribution to the mean nEMG for each group. The numbers of subjects showing distinct values of nEMG activity (from 1 to 5) during the baseline (white) and late anticipatory interval (filled bars) are shown.
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
a r t ic l e s BL LA Figure 3 Single-subject measures of learning Learner during the anticipatory interval. (a) SingleCS+ Nonspecific learner subject summary of nEMG responses for Marginally significant learner CS– three healthy volunteers (left) and three DOC Nonlearner subjects (right). Each panel shows the ERRs 6.0 1.7 CC1 VS1 of the nEMG during the baseline and late learner learner anticipatory interval (left). These two intervals are separated by a vertical dotted line. The average nEMG for the conditioned stimulus 0.7 0.7 (black) and unpaired tones (gray) and for the –400 –50 545 745 baseline (white face–color) and late anticipatory 2.6 2.0 MCS1 CC2 interval (filled bars) is shown in the center of learner learner each panel. Error bars indicate standard errors. Learning is indicated by an increase in the anticipatory interval activity compared with 0.7 0.7 baseline. If this increase is selective to the 4.3 4.7 conditioned stimulus, then learning is specific. CC6 VS4 nonspecific learner Scatter plots of nEMG activity during the late learner anticipatory interval (y axis) and nEMG activity during the baseline (x axis) for the conditioned stimulus (black) and unpaired tones (gray) are 0.7 0.7 shown on the right side of each panel. Each dot above the diagonal indicates a trial in which 4 4 4 activity during late anticipatory was higher than during the baseline. The figure shows two subjects with specific learning (top two panels) 0 0 0 and a subject with nonspecific learning (bottom panel). (b) Comparison between late anticipatory –4 –4 –4 and baseline (t values of paired t tests) for –4 0 4 –4 0 4 –4 0 4 conditioned stimulus (y axis) and unpaired CC DOC CU 2 4.5 1.2 tones (x axis) for all subjects in the controlled consciousness (left), DOC (center) and controlled unconsciousness (right) groups. Learners are depicted as black circles, nonspecific learners 1.0 as crossed white circles, marginally significant learners as gray circles and nonlearners as 1 1.0 white circles (see text for details on learning criterion). Dashed lines indicate the t scores corresponding to P values of 0.05. (c) Learning group averages. Bars are colored as in a for the controlled consciousness (left), DOC (center) and controlled unconsciousness (right) groups. For a summary of results for each individual subject, see Supplementary Figures 1–3.
b
c
nEMG
© 2009 Nature America, Inc. All rights reserved.
nEMG
a
Single-subject measures of learning Because individuals with DOCs form a heterogeneous population that is classified by the vegetative state and MCS criterion, but have a high rate of misdiagnosis12,14, we conducted a subject by subject analysis to determine which subjects showed significant learning. On the basis of our previous results, we focused our analysis on the late anticipatory interval (200 ms before the presentation of the puff), where the effects were more pronounced. A comparison of the first segment of the anticipatory interval yielded similar results with an overall smaller effect size for all measures. To quantify the individual contributions to the group results, we first measured nEMG activity for each subject for the conditioned stimulus and unpaired tones during baseline and the late anticipatory interval (Fig. 2c,d). In the controlled consciousness group, 6 out of 16 subjects had a nEMG response to the conditioned stimulus that was greater than three, that is, a threefold increase of activity during the late anticipatory interval, as compared with fluctuations before the onset of the trial. Only 1 out of 16 had a nEMG response to the unpaired tones that was greater than three. A similar result was found in the DOC group, where 5 out of 22 individuals had a nEMG response to the conditioned stimulus (and only one to the unpaired tones) that was greater than three. These data indicate that there was a great heterogeneity in the amount of learning (even in the control group), emphasizing the necessity of a subject by subject analysis.
For each subject, we carried out a 2 × 2 ANOVA with interval (baseline or late anticipatory) and stimulus (conditioned stimulus or unpaired tones) as main factors (see Supplementary Tables 1–3 for the results of the 50 subjects involved in this study). We considered an effect as significant in a single subject if P value was below 0.05. With this criterion, nine of the 16 subjects in the controlled consciousness group showed a significant effect of interval, eight showed a signi ficant effect of stimulus and six showed an interaction between these two factors (P < 0.001 for all cases; binomial probability of obtaining more than nine, eight or six positive results out of 16 tosses, each one of P = 0.05). Of the 22 subjects in the DOC group, 10 showed a significant effect of interval, 4 showed a significant effect of stimulus and 1 showed an interaction between these two factors (P < 0.001, P = 0.004 and P > 0.1; binomial probability of obtaining more than 10, 4 or 1 positive results out of 22 tosses, each one of P = 0.05). In the controlled unconsciousness group (n = 12), none showed a significant effect of interval, one showed a significant effect of stimulus and none of them showed an interaction between these two factors (P > 0.1 for all cases; binomial probability of obtaining 0 or 1 positive results out of 12 tosses, each one of P = 0.05). We determined a learning criterion from the ANOVA, in which we considered a subject to be a learner if he/she showed a significant effect of stimulus and interval. We considered a subject to be a nonspecific learner if he/she showed an effect of interval without reaching significance for the stimulus factor (Fig. 3).
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1345
a r t ic l e s
© 2009 Nature America, Inc. All rights reserved.
VS
2
b
5
1 BL+ LA+ BL– LA– 2
MCS/SED
t score of LA versus BL
4
nEMG
a
nEMG
Figure 4 Learning, clinical measures and prediction of recovery. (a) The learning pattern did not show a significant difference between subgroups of DOC subjects (vegetative state (VS) and MCS) (P > 0.05). For both groups, we observed significant learning that was, at the group level, less specific than for control subjects (see also Table 1). (b) Learning was a good estimate of recovery probability. Each data point indicates (on the y axis) the t value of the conditioned stimulus response (late anticipatory versus baseline), measured for each individual subject. The x axis indicates a measure of recovery. Improvement, subjects who had higher scores in the CRS after several months; no change, subjects who showed no signs of improvement; state change, subjects who changed from vegetative state to MCS or SED and subjects who changed from MCS to SED. We observed a clear relationship between learning and recovery. Learners are depicted as black circles, nonspecific learners as crossed white circles, marginally significant learners as gray circles and nonlearners as white circles.
3 2 1 0 –1
1
–2 State change Improvement
No change
We then followed the ANOVA with post hoc specific tests comparing conditioned stimulus and unpaired tones nEMG activity in the different epochs. When specifically looking at the anticipatorybaseline comparison, we found that many subjects in the controlled consciousness and DOC groups showed a significant difference for the conditioned stimulus (late anticipatory-baseline) stimulus (13 out of 16 for the controlled consciousness and 13 out of 22 for the DOC group). Subjects passing this criterion and not reaching significance in the ANOVA test were referred to as being marginally significant learners. Considerably fewer subjects (5 out of 16 for the controlled consciousness and 7 out of 22 for the DOC group) showed a substantial difference between the late anticipatory interval and the baseline for the unpaired tones (late anticipatory-baseline). For the controlled unconsciousness group, only 1 subject out of 12 showed a difference for the conditioned stimulus (anticipatory-baseline). To simultaneously visualize the amount of learning for all subjects in all groups, we plotted, for each individual subject, the t score of the difference between the anticipatory interval and the baseline for the conditioned stimulus versus the unpaired tones (Fig. 3b). As in our previous analysis, these data reflect the same pattern. Learning was significant and highly specific in the controlled consciousness group and learning was significant, although considerably less prominent, in the DOC group. Although learning was specific, the specificity of learning was considerably less than for the conscious control group. Virtually no learning was seen in the controlled unconsciousness group. This group effect was seen synthetically after averaging all subjects in each condition (Fig. 3c). We then set out to quantitatively study the differences in learning across groups and, in the DOC group, assess which clinical factors relate to the amount of anticipatory learning.
learning than the vegetative state group for the conditioned stimulus (late anticipatory-baseline) comparison (P < 0.05). This difference was not significant for the unpaired tones comparison (P > 0.1). The comparison between the vegetative state and controlled unconsciousness groups showed significantly higher learning in the vegetative state for both contrasts (P < 0.025 for both). Notably, the unpaired tones (late anticipatory-baseline) contrast showed significant differences (P < 0.025) between the vegetative state and controlled unconsciousness groups, but not between control and anesthesia groups. This further supports our previous finding of lesser learning specificity in the vegetative state group and suggests that this does not result exclusively from a decreased response to the conditioned stimulus, but rather that individuals in the vegetative state fail to inhibit the anticipatory response to unpaired tones. The MCS/SED subjects also showed partial learning as a group, with a slightly different pattern than the vegetative state. None of the contrasts showed a difference between the MCS/SED and the controlled consciousness groups, although the conditioned stimulus (anticipatory-baseline) was close to significance (P = 0.07). The comparison with the controlled unconsciousness group showed an effect of both contrasts, indicating that, for the vegetative state group, there was significant, but not specific, learning (P < 0.025 for both). The comparisons between the MCS/SED and vegetative state group showed no significant differences (P > 0.1 for both). In summary, this analysis confirmed our previous qualitative observations, that the control group showed the most significant and specific learning effect. The vegetative state and MCS groups were very similar and showed a significant effect of learning over the anes thesia group in the conditioned stimulus and unpaired tones contrast, revealing a less specific form of learning.
Comparison across different groups The nEMG patterns for the vegetative state and MCS groups (Fig. 4a) showed virtually no differences and were identical to the two groups pulled together (Fig. 3c). To compare learning between the four groups of subjects (controlled consciousness, vegetative state, MCS/ SED and controlled unconsciousness), we applied a Kruskal-Wallis one-way ANOVA for each of two learning contrasts defined previously. The conditioned stimulus late anticipatory-baseline showed a significant difference between groups (χ2 = 16.78, P < 0.001), whereas the same measure for unpaired tones was not significant (χ2 = 6.58, P < 0.083). Specific comparisons between pairs of these measures revealed distinct patterns of conditioning between the controlled consciousness, DOC and controlled unconsciousness groups (Table 1). The controlled consciousness group showed marginally significantly higher
Learning and clinical measures We then explored the relationship of learning in the DOC group with four relevant clinical markers, age, cortical atrophy score (CAS),
1346
Table 1 Learning differences at the group level CC versus CU CC versus VS CC versus MCS VS versus MCS VS versus CU MCS versus CU
CS+ LA/BL
CS– LA/BL
z = −3.95*** z = −1.93* z = −1.70 ns z = −0.17 ns z = −2.26** z = −1.97*
z = −1.42 ns z = −0.81 ns z = −1.10 ns z = −0.03 ns z = −2.26** z = −2.06*
Each cell in the table shows the z value for the Mann-Whitney test. MCS group includes two SED subjects. CC, controlled consciousness group (volunteers); CU, controlled unconsciousness group (anesthetized subjects); VS, vegetative state group. ns, not significant (P > 0.05). *P < 0.05, **P < 0.025, ***P < 0.001.
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
a r t ic l e s group of patients. CRS scores and age do not significantly explain variance for conditioned stimulus (anticipatory-baseline) and none of the independent measures (TFI, CAS, CRS and age) showed sigCS+ LA/BL CS– LA/BL nificant values for unpaired tones. Age Beta = −0.037 Beta = −0.285 We performed two different analyses to evaluate the predictive power t = −0.54 t = 0.84 of learning measures to discriminate between groups, one for clinical P = 0.601 P = 0.420 CAS Beta = −0.773 Beta = −0.327 diagnosis (vegetative state or MCS/SED) and the other for cause of t = −5.28 t = −1.33 injury (traumatic brain injury (TBI) or non-TBI). A binary logistic P = 0.001 P = 0.204 regression was undertaken to assess the power of conditioned stimulus TFI Beta = −0.277 Beta = −0.178 late anticipatory-baseline to classify subjects in two groups. Conditioned t = −1.89 t = −0.67 stimulus late anticipatory-baseline contrast failed to accurately classify P = 0.081 P = 0.514 the subject population in the vegetative state or MCS/SED groups. The CRS Beta = −0.131 Beta = −0.337 t = −0.69 t = −1.34 model incorrectly classified 2 out of 11 individuals in the vegetative P = 0.505 P = 0.201 state and 4 out of 9 non–vegetative state subjects, leaving the model Beta, t and P values of the backward regressions are shown for each of the four clinical with an accuracy of 72.7% (χ2 = 3.61, P = 0.057). This is consistent with independent variables (age, CAS, TFI and CRS) scores in both conditioned stimulus our previous observation that learning in vegetative state and MCS was late anticipatory-baseline and unpaired tones late anticipatory-baseline. CS+, not significantly different. This same measure was used to differenticonditioned stimulus; CS–, unpaired tones; LA/BL, late anticipatory-baseline. ate between TBI and anoxic/hypoxic events as the cause of the DOC. time from ictus (TFI) and coma recovery scale (CRS) scores. In a The model distinguished between TBI and non-TBI subjects with 82% correlation analysis, age and CAS significantly correlated with condi- accuracy. The contribution from conditioned stimulus late anticipationed stimulus (late anticipatory-baseline; r = −0.384, P = 0.039 and tory-baseline contrast significantly increased the accuracy of the model r = −0.640, P = 0.001, respectively), but when placed in conjunction (χ2 = 4.52, P = 0.033). The mode correctly classified 11 out of 12 TBI with the other clinical variables in a regression analysis, only CAS subjects, but 3 out of 10 non-TBI subjects were incorrectly assigned contributed to explain the variance of learning (Table 2). (those who showed some degree of learning, VS2, VS6 and MCS3). Two backward regressions, one for late anticipatory-baseline in Finally, we explored whether learning correlated with probability of conditioned stimulus and the other for the control comparison late recovery. We classified our subjects in two groups in relation to their clinanticipatory-baseline unpaired tones, showed that CAS—which ical outcome, with no recovery comprising those subjects who showed explains ~40% of the learning measure conditioned stimulus (anti no change in CRS scores (after 6 months to 2 years) and recovery concipatory-baseline) variance independently of the regression model sisting of those subjects who changed from the vegetative state to MCS —and, to a lesser degree, TFI can partially predict learning in our or from the vegetative state or MCS to SED and those who increased their behavioral portfolio (CRS scores) without changing conscious state (Fig. 4b). We performed a b 20 a logistic regression to evaluate whether condi40 tioned stimulus late anticipatory-baseline could 10 differentiate between recovery and no recovery. 30 Amplitude Learning (conditioned stimulus late anticipatory0 baseline, χ2 = 5.02, P = 0.025) indicated, with an 20 accuracy of 86%, whether a subject had shown –10 Latency 10 signs of recovery or not. The change of conscious state subgroup from –20 the recovery group was composed of subjects VS3 700 800 900 1,000 1,100 1,200 700 800 900 1,000 1,100 1,200 Time (ms) Time (ms) and VS8, who changed to MCS, VS1 and VS4, who CC DOC CU moved to SED, and MCS1, MCS2 and MCS5, who c showed signs of improvement after the learning 2 2 2 evaluation (Supplementary Table 4). These last 0 0 0 three individuals could communicate (fulfilling the criteria for SED) at the time this manuscript was –2 –2 –2 accepted for publication. The other three individu0 10 20 0 10 20 0 10 20 als from the recovery group showed greater CRS Trial number Trial number Trial number scores a few months after the full assessment; VS5 d 0.2 Amplitude showed 3 to 6 in his CRS score, VS7 changed from 3 Figure 5 Nonassociative learning in patients and Latency to 7 and MCS4 changed from 11 to 13. All patients control subjects: changes in the response to the aversive in the recovery group suffered from TBI (and they stimulus. (a) Raster plot of single-trial unconditioned 0 all showed learning), the patients in the no recovresponses illustrates how the unconditioned response changed with learning. Both the latency and the ery group did not suffer TBI (except for MCS6 and amplitude of the response decreased as learning MCS7), with only 2 out of 10 showing learning. nEMG
Trial number
Slope
nAmp and nLat
© 2009 Nature America, Inc. All rights reserved.
Table 2 Cortical atrophy partially explains anticipatory activity (learning) in trace conditioning
–0.2 progressed. The vertical white line corresponds to CC DOC CU the onset of the UCS. (b) A single trial showing the Group measured amplitude and latency. (c) Normalized amplitude (nAmp) and latency (nLat) of the response decreased as training proceed during the first 20 trials for the controlled consciousness and DOC groups. (d) Individual contributions to the decrease in latency and amplitude (measured by the slope).
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
Nonassociative learning To further understand the global aspects of learning in individuals with DOCs, we carried out an analysis of the change in the eye blink response 1347
a r t ic l e s P = 0.607). However, the amplitude of the response in DOC subjects was significantly smaller than in control subjects (negative peak z = −1.886, P = 0.066; positive peak z = −3.185, P < 0.001; Fig. 6).
Voltage (µV)
6 3
Auditory processing in individuals with DOCs A critical aspect of trace conditioning is that it establishes a temporal gap between the conditioned stimulus and UCS. Because there is an overall attenuation of neural activity in individuals with DOCs17,18,20, it is conceivable that it also presents long latencies and that some processing of the first stimulus may still be going on by the time the UCS occurs. To rule out this possibility, we measured the latency of evoked potentials in an electroencephalography experiment. We tested nine individuals with DOCs in an oddball procedure, using the same stimuli as in the trace-learning experiment. This experiment served two purposes. First, we could measure the latency and the amplitude of the response. Second, we were able to determine whether the subjects reallocate attention to a different stimulus by measuring the mismatch negativity (MMN). Eight out of nine subjects (all except SED1) elicited an early electroencephalography response to the auditory stimulus, referred as a N1 component. There were significant scalp electrophysiological differences (P < 0.05) between tones, indicating a significant MMN in seven out of nine individuals with DOCs (SED1, VS12, MCS3, VS5, MCS4, MCS7 and VS4, but not VS7 and MCS6). In particular, this indicates that sensory responses in these subjects distinguished the two tones used in the trace-learning experiment. The latencies in the response components of the DOC subjects were not significantly different from the latencies of normal volunteers (negative peak z = −1.653, P = 0.113; positive peak z = −0.59,
DISCUSSION The maintenance of information in time is one of the main psychological attributes of conscious processing15,30. In fact, when there is no temporal interval between a conditioned stimulus and a UCS (delay conditioning), learning the conditioned stimulus–UCS relationship becomes automatic and reflexive and no longer requires declarative knowledge1,31. In this context, our finding that clinically defined vegetative state patients can acquire trace conditioning is surprising. As with other studies involving individuals with DOCs, this result has two interpretations. Individuals with DOCs may have partially preserved conscious processing, which cannot be exhibited overtly via intentional movement or verbal responses, or, alternatively, trace conditioning can indeed be acquired in the absence of consciousness. Some considerations suggest that the first interpretation is more likely. First, subjects in a pharmacologically controlled unconscious state were incapable of eliciting trace-conditioning learning. However, as mentioned previously, this comparison must be made cautiously because of the many factors affected by anesthesia, including amnesia31,32. Another argument suggesting preserved conscious processing in individuals with DOCs comes from single-subject analysis. First, learning was a good predictor of recovery. Second, subjects that showed significant learning also showed other indications of partially preserved conscious processing in complementary neuroimaging studies. Three of the subjects fulfilling the clinical criteria for vegetative state that acquired trace conditioning had brain activity that may indicate partial preservation of awareness. Subject VS4 (Fig. 3b) showed specific brain activations when asked to imagine playing tennis or to imagine moving around the rooms of their house that were consistent with those of volunteers performing the same task28,33. Activation of the lateral premotor cortex was seen in subjects VS3 and VS8 when asked to move the opposite hand, suggesting motor intention, even when no muscle activity could be detected (unpublished data, T.A.B., F.F.M., M. Villarreal, A.M. Owen and V. Della Maggiore). The remaining vegetative state subjects who showed significant learning had activity in the lateral temporal cortex in response to speech stimuli34. Despite the previous arguments, we cannot exclude the possibility that trace conditioning is acquired, to a lesser extent, in the absence of awareness. Lower-order organisms, who most probably have nothing akin to human consciousness, can show this type of learning35, suggesting that different learning mechanisms might be engaged when human conscious awareness is offline. In humans, other manners of acquiring trace conditioning in the absence of awareness have been demonstrated using emotionally salient stimuli as conditioned stimulus. Fear trace conditioning can be learned even when the conditioned stimulus is masked36, but is not learned when the masked conditioned stimulus does not have a strong emotional content37. The functional neuroanatomy involved in trace learning seems to be consistent with its psychological attributes; bridging the cognitive gap generated by the silence between the tone and the air puff entails a functional neural network involving the cerebellum, the hippocampus and certain prefrontal cortex areas3–5,38. Because individuals with DOCs are a very heterogeneous group, classified mainly by clinical measures and resulting from a wide range of brain lesion patterns, it is important to understand which neuroanatomical and physiological indicators may be predictive of learning. In this study, we found that the degree of brain atrophy was a good predictor of learning capabilities. In addition, we observed a clear relationship between learning
1348
VOLUME 12 | NUMBER 10 | october 2009 nature NEUROSCIENCE
0 –3 –6 100
150
200
250
300
Time (ms)
© 2009 Nature America, Inc. All rights reserved.
Figure 6 Intact latencies, but smaller amplitudes, in event-related auditory potentials in DOCs. Mean MMN negative (open symbols) and positive (filled symbols) related peaks’ latencies and intensities of the difference between the responses to frequent and infrequent tones. Latencies and amplitudes were estimated for each individual subject for healthy volunteers (triangles) and individuals with DOC (circles). The positive peak amplitude was significantly smaller (P < 0.001) in DOC subjects than in normal volunteers.
(unconditioned response) to the air puff, which constitutes a nonassociative (not related to the conditioned stimulus) form of learning. Both the latency and amplitude of the response change during the course of learning (Fig. 5a,b). To quantify this observation, we estimated, for each individual trial, the latency and amplitude of the response. Amplitude and latency decreased for the DOC and the controlled consciousness groups during the course of learning (Fig. 5c,d). This decrease was confined to the first 20 trials. When we considered the first 20 trials, the slope of the regression was significant for the control group (amplitude: t = −3.1, P < 0.001, degrees of freedom = 18; latency: t = −4.4, P < 0.001, degrees of freedom = 18) and for the DOC group (amplitude: t = −4.8, P < 0.001, degrees of freedom = 18; latency: t = −2.5, P = 0.02, degrees of freedom = 18) and was not significant for the anesthesia group (amplitude: t = 1.3, P > 0.1, degrees of freedom = 18; latency: t = −0.9, P > 0.1, degrees of freedom = 18). When we examined trials 20–40 (or beyond), none of these regressions were significant (P > 0.05). For the controlled consciousness and DOC groups, most subjects showed negative slopes, although there was a broad dispersion (Fig. 5d).
© 2009 Nature America, Inc. All rights reserved.
a r t ic l e s and cause of injury: there was a very high probability of learning in TBI patients and a very low probability in hypoxic patients. One of the most ambitious goals in DOC investigations is to establish measures that may work as indicators of recovery. We found that learning was a predictor of future recovery, with an accuracy of 86%. Moreover, individuals who showed learning and subsequently changed their state of consciousness (recovered) were reassessed behaviorally at least 2 years after the learning assessment; individuals who only showed improvement on the behavioral scale, but not to the point of a change in conscious state, were assessed less than a year after learning. These findings suggest there is room for late recovery in DOC patients39 and that trace conditioning could predict this type of recovery in post acute DOC patients. These results suggest that an adequate classification of individuals with DOCs requires anatomical, functional and behavioral measures. Trace learning may be an important indicator in this integrative diagnostic battery. Although it requires further investigation, our finding that individuals with chronic pathologies of awareness can acquire trace conditioning (and may recover) suggests that there is a window for cognitive neuro-rehabilitation. The underlying idea of clinical rehabilitation is to train the networks involved in the specific pathology. Although this has a straightforward implementation in muscular, sensory or mnemonic disorders, this becomes less clear when awareness is the process to be trained. Although extremely speculative, it is important to investigate whether training the circuits involved in awareness may help the recovery of consciousness.
1. Koch, C. The Quest for Consciousness: A Neurobiological Approach (Roberts and Company, Englewood, Colorado, 2004). 2. Laureys, S. The neural correlate of (un)awareness: lessons from the vegetative state. Trends Cogn. Sci. 9, 556–559 (2005). 3. Clark, R.E. & Squire, L.R. Classical conditioning and brain systems: the role of awareness. Science 280, 77–81 (1998). 4. Clark, R.E., Manns, J.R. & Squire, L.R. Trace and delay eyeblink conditioning: contrasting phenomena of declarative and nondeclarative memory. Psychol. Sci. 12, 304–308 (2001).
5. Knight, D.C., Cheng, D.T., Smith, C.N., Stein, E.A. & Helmstetter, F.J. Neural substrates mediating human delay and trace fear conditioning. J. Neurosci. 24, 218–228 (2004). 6. McGlinchey–Berroth, R., Carrillo, M.C., Gabrieli, J.D., Brawn, C.M. & Disterhoft, J.F. Impaired trace eyeblink conditioning in bilateral, medial-temporal lobe amnesia. Behav. Neurosci. 111, 873–882 (1997). 7. Clark, R.E., Manns, J.R. & Squire, L.R. Classical conditioning, awareness and brain systems. Trends Cogn. Sci. 6, 524–531 (2002). 8. Dawson, M.E. & Furedy, J.J. The role of awareness in human differential autonomic classical conditioning: the necessary-gate hypothesis. Psychophysiology 13, 50–53 (1976). 9. Lovibond, P.F. & Shanks, D.R. The role of awareness in Pavlovian conditioning: empirical evidence and theoretical implications. J. Exp. Psychol. Anim. Behav. Process. 28, 3–26 (2002). 10. Jennett, B. Thirty years of the vegetative state: clinical, ethical and legal problems. Prog. Brain Res. 150, 537–543 (2005). 11. Jennett, B. & Plum, F. Persistent vegetative state after brain damage. A syndrome in search of a name. Lancet 299, 734–737 (1972). 12. Royal College of Physicians. The Vegetative State: Guidance on Diagnosis and Management (Royal College of Physicians, London, 2003). 13. Multi–Society Task Force on PVS. Medical aspects of the persistent vegetative state. N. Engl. J. Med. 330, 1572–1579 (1994). 14. Giacino, J.T. et al. The minimally conscious state: definition and diagnostic criteria. Neurology 58, 349–353 (2002). 15. Baars, B.J. A Cognitive Theory of Consciousness (Cambridge University Press, New York, 1988). 16. Dehaene, S. & Naccache, L. Towards a cognitive neuroscience of consciousness: basic evidence and a workspace framework. Cognition 79, 1–37 (2001). 17. Boly, M. et al. Auditory processing in severely brain injured patients: differences between the minimally conscious state and the persistent vegetative state. Arch. Neurol. 61, 233–238 (2004). 18. Bekinschtein, T.A. et al. Neural signature of the conscious processing of auditory regularities. Proc. Natl. Acad. Sci. USA 106, 1672–1677 (2009). 19. Bekinschtein, T. et al. Emotion processing in the minimally conscious state. J. Neurol. Neurosurg. Psychiatry 75, 788 (2004). 20. Kotchoubey, B. et al. Information processing in severe disorders of consciousness: vegetative state and minimally conscious state. Clin. Neurophysiol. 116, 2441–2453 (2005). 21. Laureys, S. et al. Cerebral processing in the minimally conscious state. Neurology 63, 916–918 (2004). 22. Schiff, N.D. et al. fMRI reveals large-scale network activation in minimally conscious patients. Neurology 64, 514–523 (2005). 23. Schnakers, C. et al. Voluntary brain processing in disorders of consciousness. Neurology 71, 1614–1620 (2008). 24. Laureys, S. et al. Auditory processing in the vegetative state. Brain 123, 1589–1601 (2000). 25. Menon, D.K. et al. Cortical processing in persistent vegetative state. Wolfson Brain Imaging Centre Team. Lancet 352, 200 (1998). 26. Boly, M. et al. Perception of pain in the minimally conscious state with PET activation: an observational study. Lancet Neurol. 7, 1013–1020 (2008). 27. Coleman, M.R. et al. Do vegetative patients retain aspects of language comprehension? Evidence from fMRI. Brain 130, 2494–2507 (2007). 28. Owen, A.M. et al. Detecting awareness in the vegetative state. Science 313, 1402 (2006). 29. Alkire, M.T., Hudetz, A.G. & Tononi, G. Consciousness and anesthesia. Science 322, 876–880 (2008). 30. Tononi, G. & Edelman, G.M. Consciousness and complexity. Science 282, 1846–1851 (1998). 31. Manns, J.R., Clark, R.E. & Squire, L.R. Standard delay eyeblink classical conditioning is independent of awareness. J. Exp. Psychol. Anim. Behav. Process. 28, 32–37 (2002). 32. Veselis, R.A., Reinsel, R.A. & Feshchenko, V.A. Drug-induced amnesia is a separate phenomenon from sedation: electrophysiologic evidence. Anesthesiology 95, 896–907 (2001). 33. Boly, M. et al. When thoughts become action: an fMRI paradigm to study volitional brain activity in noncommunicative brain-injured patients. Neuroimage 36, 979–992 (2007). 34. Bekinschtein, T.A. Cognitive Processes in the Vegetative and Minimally Conscious State. (PhD thesis, Univ. Buenos Aires, 2006). 35. Carew, T.J., Hawkins, R.D. & Kandel, E.R. Differential classical conditioning of a defensive withdrawal reflex in Aplysia californica. Science 219, 397–400 (1983). 36. Morris, J.S., Ohman, A. & Dolan, R.J. Conscious and unconscious emotional learning in the human amygdala. Nature 393, 467–470 (1998). 37. Öhman, A. & Soares, J.J. Emotional conditioning to masked stimuli: expectancies for aversive outcomes following nonrecognized fear-relevant stimuli. J. Exp. Psychol. Gen. 127, 69–82 (1998). 38. Carter, R.M., O’Doherty, J.P., Seymour, B., Koch, C. & Dolan, R.J. Contingency awareness in human aversive conditioning involves the middle frontal gyrus. Neuroimage 29, 1007–1012 (2006). 39. Voss, H.U. et al. Possible axonal regrowth in late recovery from the minimally conscious state. J. Clin. Invest. 116, 2005–2011 (2006).
nature NEUROSCIENCE VOLUME 12 | NUMBER 10 | october 2009
1349
Methods Methods and any associated references are available in the online version of the paper at http://www.nature.com/natureneuroscience/. Note: Supplementary information is available on the Nature Neuroscience website. Acknowledgments We thank the care homes and rehabilitation centers in the UK and Argentina, the Cambridge Impaired Consciousness Research Group, the staff of the Wellcome Trust Research Facility for their contribution, and all the study’s participants. We especially thank F. Klein and the Anesthesia Favaloro Team. We also thank L. Naccache and C. Koch for comments on an early version of the manuscript. This study was funded by an Antorchas Foundation grant (T.A.B.), a Marie Curie IIF grant (T.A.B.), a StartUp grant (F.F.M.), the Human Frontiers Science Program (M.S.) and a Medical Research Council Acute Brain Injury Collaborative grant (G0600986). AUTHOR CONTRIBUTIONS T.A.B. and C.F. designed the study. T.A.B. and F.F.M. conducted the behavioral and neurological assessments. T.A.B., C.F., M.R.C., M.H. and D.E.S. conducted the eyeblink conditioning task in the normal volunteers group and T.A.B., M.R.C. and C.F. conducted the task in the patient group. T.A.B., D.E.S., C.F., M.H., M.R.C., F.F.M. and M.S. analyzed and interpreted the data. T.A.B., D.E.S. and M.S. performed the statistical analysis. T.A.B., D.E.S. and M.S. drafted the manuscript. All of the authors revised the manuscript for important intellectual content. Published online at http://www.nature.com/natureneuroscience/. Reprints and permissions information is available online at http://www.nature.com/ reprintsandpermissions/.
© 2009 Nature America, Inc. All rights reserved.
ONLINE METHODS Patients. We chose 22 individuals with DOCs were chosen from a pool of 62 vegetative state or MCS patients using several inclusion criteria. The subjects were between 18 and 60 years of age, had relatively preserved auditory evoked potentials (normal or mildly delayed wave V) on at least one side, were three months from ictus or more, and had a preserved eye-blink reflex. We conducted comprehensive evaluations during 5 d of hospitalization in Buenos Aires or Cambridge. All of the DOC subjects underwent a full clinical assessment that included neurological examination, CRS40 and Wessex head injury matrix41 to establish each patient’s behavioral profile. Subjects were given structural MRI, short latency auditory and somatosensory evoked potentials assessments. Additional evaluations were conducted in some patients: fMRI speech task, fMRI movement intention task, fMRI volition task, visual and motor evoked potentials, novelty evoked related potentials and a movement intention electromyographic task. We assessed the degree of cortical and subcortical atrophy using a visual rating scale42 inspired by a previously developed scale43. Briefly, we first defined, using T1 three-dimensional anatomical images, atrophy levels from 0 to 4 (0 = no atrophy, 1 = very low, 2 = mild, 3 = severe and 4 = highly severe atrophy) in a group of 12 patients with neurodegenerative disorders and applied the scale to the DOC subjects (Supplementary Table 4). The subjects’ T1-weighted images were assessed by two experienced raters (T.A.B. and F.F.M.) using a blind procedure. Signed assent from subjects’ next of kin was acquired before investigation. The Cambridge Local Research Ethics Committee, the Raul Carrea Ethics Committee and the Fundacion Favaloro Ethics Committee approved this study. Healthy volunteers (controlled consciousness group). Normal volunteers were free of physical and psychiatric illnesses, including head trauma. We carried out the trace-conditioning experiment on 16 subjects. None of them had a history of alcohol dependence or substance abuse and refrained from smoking and having caffeine intake for at least 4 h before the test. All control subjects included in the analysis matched DOC subjects in age and gender. Control subjects had normal hearing for their age, were naive to eye-blink conditioning procedures and were instructed to watch a silent video (Classic Chaplin movie) during the presentation of the stimuli. They were also told that they would receive a questionnaire about their experience during the task. Every participant gave written informed consent. No payment was received for taking part in the study. Anesthetized subjects (controlled unconsciousness group). We carried out the test on 12 patients under the effect of intravenous propofol during standard endoscopic procedures (video-gastroscopy and colonoscopy), no muscle relaxants were given during the procedure and all of the subjects showed preserved eye-blink reflexes. The depth of anesthesia was controlled using clinical markers (heart rate, respiratory rate, muscular activity and bispectral index) by a specialized anesthesiologist. The learning procedure was performed during programmed endoscopic procedures of at least 30-min duration and it started no less than 3 min after anesthetic induction or when bispectral index levels reached 60 or less44. Every participant gave written informed consent. No payment was received for taking part in the study. Experimental procedure. The stimulation apparatus was designed on the basis of the classic human eye-blink conditioner and consisted of an electronic device that controlled the delivery of the air puff and the auditory stimuli and sent a synchronization signal to the EMG recorder. Eye blinks were recorded with surface electrodes placed around the orbicularis oculi muscle of the right eye. One electrode was placed 1 cm lateral to the outer canthus and a second was placed 1 cm below the right eye. The ground electrode was aligned at the center of the subject’s forehead. Electromyographic activity was acquired with a Keypoint machine (Medtronic) at 2,000 Hz in continuous recording mode. The conditioning parameters were similar to those used in a previous study3. The conditioned stimulus was a 75-dB, 250-ms, 1- or 2-kHz tone with a 5-ms rise-fall time; delivered binaurally (when appropriate) through earphones. It was followed by a silent period of 500 ms after which a 100-ms, 3-psi corneal air puff UCS was delivered to the right eye. The unpaired tones was either a white (static) noise or a 1- or 2-kHz tone that was not paired to an air puff. All DOC subjects’ reflex responses to the air puff were tested at least 2 d before the conditioning experiment. The procedure consisted of a total of 140 conditioning trials (70 conditioned stimulus and 70 unpaired tones), with an intertrial interval ranging from 10–15 s. The order
nature NEUROSCIENCE
of conditioning trials was pseudorandom with the constraint that neither trial type occurred more than twice consecutively. No unpaired pseudo-conditioning trials were used before paired conditioning trials. The data was stored and later preprocessed with MATLAB (Mathworks). The continuous EMG data was segmented in epochs starting 1,000 ms before and ending 2,000 ms after the conditioned stimulus onset. Data was low-pass filtered with a cutoff frequency of 50 Hz. We implemented an automatic artifactdetection algorithm. For each subject, we estimated the mean values and the s.d. of the baseline interval. Trials for which these values differed in more than two s.d. from the distribution were discarded. For further analysis, the EMG signal was normalized on a trial by trial basis, subtracting the mean and dividing by the s.d. of a window of 290 ms, starting 928 ms before the beginning of the trial. We then rectified the resulting signal. We refer to this as the nEMG, which has units of s.d. We then parsed this nEMG response in four critical temporal intervals (Fig. 1): the period before tone onset (baseline, −400 to 0 ms), a short period of 300 ms following the presentation of the tone (α response), a subsequent period (anticipatory interval) following this transient response and before the presentation of the puff (conditioned response), and the period following the onset of the air puff (unconditioned response). The anticipatory interval was divided into two intervals of the same duration: early (first half of the anticipatory interval) and late (second half of the anticipatory interval). We then averaged the nEMG activity to get a single value for each trial and interval. These values were used in all data analysis of anticipatory response with the exception of Figures 2a and 3a, in which nEMG activity is averaged across trials. The averaged nEMG activity for each trial and interval was submitted to a by subject ANOVA with 2 × 2 factorial design, in which the main factors were interval (baseline or late anticipatory), and stimulus (conditioned stimulus or unpaired tones). The significance level was set to P = 0.05. This value was used even when we presented multiple comparisons. Multiple comparison corrections would decrease the number of false positives at the expense of an increase in the number of false negatives; it was critical for our analysis to avoid false negatives. The significance of our results at the group level was very high; the probability of having as many as 9 out of 16 (as in control group) or 10 out of 22 positive results (as in DOC group) just by chance is P < 10−6. As post hoc tests, we performed one-tailed paired t tests to evaluate the significance level of the difference between baseline and late anticipatory interval for both stimulus types (conditioned stimulus and unpaired tones) and for each subject (Table 1). Individual t values of these tests were plotted in Figure 3b. To compare the anticipatory interval between both stimulus types and for baseline interval between conditioned stimulus and unpaired tones, we performed unpaired t tests. To quantify the change in the unconditioned response, as a measure of nonassociative learning, we estimated the latency and amplitude of each individual trial. To estimate it, we searched the amplitude and latency of the maximum and minimum peaks in the normalized, but not rectified, EMG data. The amplitude was calculated as the difference between both peaks’ amplitudes and the latency was taken from the later of both. Because both parameters can vary broadly between subjects, we normalized them by subtracting the mean and dividing by s.d. for each subject separately. Finally, we plotted the first 20 trials of each subject (the average for the three groups is shown in Fig. 5e) and performed a linear regression to trial number. Statistical analysis at the group level. We used SPSS v16.0 for nonparametric statistics at the group level and for regressions. Because of the small sample size and nonGaussian distribution, we used Kruskal-Wallis one-way ANOVA for main group comparison in the core measure of learning (conditioned stimulus late anticipatory period/baseline) and its control condition (unpaired tones late anticipatory period/baseline). For each measure, the differences between groups were investigated using a nonparametric Mann-Whitney U with exact (that is, nonasymptotic) significance assessment method. This method has been suggested to increase the reliability of results in small sample sets by dispensing with the assumptions required for the asymptotic method45. Two backwards regressions (on conditioned stimulus late anticipatory period/baseline and unpaired tones late anticipatory period/baseline) were performed to evaluate the explanatory power of the clinical measures (age, CRS, CAS and TFI) on learning for the DOC group. Two binary logistic regression were employed to evaluate the predictive power of learning measures (conditioned stimulus Late anticipatory period/ baseline and unpaired tones Late anticipatory period/baseline) to discriminate between groups, one for vegetative state or nonvegetative state (clinical diagnosis)
doi:10.1038/nn.2391
and the other for TBI or nonTBI (cause of injury). Another binary logistic regression was undertaken to assess the power of conditioned stimulus Late anticipatory period/baseline to classify individuals in two groups. Inclusion of only one factor in the binary logistic regression allows us to maintain an approximate 1 to 20 ratio between regressors and observations46.
© 2009 Nature America, Inc. All rights reserved.
40. Giacino, J.T., Kalmar, K. & Whyte, J. The JFK coma recovery scale–revised: measurement characteristics and diagnostic utility. Arch. Phys. Med. Rehabil. 85, 2020–2029 (2004). 41. Shiel, A. et al. The Wessex head injury matrix (WHIM) main scale: a preliminary report on a scale to assess and monitor patient recovery after severe head injury. Clin. Rehabil. 14, 408–416 (2000).
42. Bekinschtein, T.A. et al. Can electromyography objectively detect voluntary movement in disorders of consciousness? J. Neurol. Neurosurg. Psychiatry 79, 826–828 (2008). 43. Galton, C.J. et al. Temporal lobe rating scale, application to Alzheimer’s disease and frontotemporal dementia. J. Neurol. Neurosurg. Psychiatry 70, 165–173 (2001). 44. Soriano, S.G., McCann, M.E. & Laussen, P.C. Neuroanesthesia. Innovative techniques and monitoring. Anesthesiol. Clin. North America 20, 137–151 (2002). 45. Siegel, S. & Castellan, N.J. Nonparametric Statistics for the Behavioral Sciences (MacGraw-Hill, New York, 1998). 46. Harrell, F.E. Jr. et al. Regression modeling strategies for improved prognostic prediction. Stat. Med. 3, 143–152 (1984).
doi:10.1038/nn.2391
nature NEUROSCIENCE