Novartis Foundation Symposium 227
MECHANISMS AND BIOLOGICAL SIGNIFICANCE OF PULSATILE HORMONE SECRETION
2000
JOHN WILEY & SONS, LTD Chichester · New York · Weinheim · Brisbane · Singapore · Toronto
MECHANISMS AND BIOLOGICAL SIGNIFICANCE OF PULSATILE HORMONE SECRETION
The Novartis Foundation is an international scienti¢c and educational charity (UK Registered Charity No. 313574). Known until September 1997 as the Ciba Foundation, it was established in 1947 by the CIBA company of Basle, which merged with Sandoz in 1996, to form Novartis. The Foundation operates independently in London under English trust law. It was formally opened on 22 June 1949. The Foundation promotes the study and general knowledge of science and in particular encourages international co-operation in scienti¢c research. To this end, it organizes internationally acclaimed meetings (typically eight symposia and allied open meetings and 15^20 discussion meetings) and publishes eight books per year featuring the presented papers and discussions from the symposia. The Foundation’s headquarters at 41 Portland Place, London W1N 4BN, provide library facilities, o¡ers accommodation and meeting facilities to visiting scientists and their societies. Information on all Foundation activities can be found at http://www.novartisfound.org.uk
Novartis Foundation Symposium 227
MECHANISMS AND BIOLOGICAL SIGNIFICANCE OF PULSATILE HORMONE SECRETION
2000
JOHN WILEY & SONS, LTD Chichester · New York · Weinheim · Brisbane · Singapore · Toronto
Copyright & Novartis Foundation 2000 Published in 2000 byJohn Wiley & Sons Ltd, Ba⁄ns Lane, Chichester, West Sussex PO19 1UD, England National 01243 779777 International (+44) 1243 779777 e-mail (for orders and customer service enquiries):
[email protected] Visit our Home Page on http://www.wiley.co.uk or http://www.wiley.com All Rights Reserved. No part of this book may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording, scanning or otherwise, except under the terms of the Copyright, Designs and Patents Act 1988 or under the terms of a licence issued by the Copyright Licensing Agency, 90 Tottenham Court Road, London,W1P 9HE, UK, without the permission in writing of the publisher. Other Wiley Editorial O⁄ces John Wiley & Sons, Inc., 605 Third Avenue, NewYork, NY 10158-0012, USA WILEY-VCH Verlag GmbH, Pappelallee 3, D-69469 Weinheim, Germany Jacaranda Wiley Ltd, 33 Park Road, Milton, Queensland 4064, Australia John Wiley & Sons (Asia) Pte Ltd, 2 Clementi Loop #02-01, Jin Xing Distripark, Singapore 129809 John Wiley & Sons (Canada) Ltd, 22 Worcester Road, Rexdale, Ontario M9W 1L1, Canada Novartis Foundation Symposium 227 ix+269 pages, 70 ¢gures, 11 tables Library of Congress Cataloging-in-Publication Data Mechanisms and biological signi¢cance of pulsatile hormone release / editors, DerekJ. Chadwick and Jamie A. Goode. p. cm. ^ (Novartis Foundation symposium ; 227) Symposium on Mechanisms and Biological Signi¢cance of Pulsatile Hormone Secretion, held at the Novartis Foundation, London, 2^4 March 1999. Includes bibliographical references and index. ISBN 0-471-99918-0 (hc: alk. paper) 1. Hormones ^Secretion^Congresses. 2. Biological rhythms ^Molecular aspects ^Congresses. I. Chadwick, Derek. II. Goode, Jamie. III. Symposium on Mechanisms and Biological Signi¢cance of Pulsatile Hormone Secretion (1999 : London, England) IV. Series. QP571.M4252000 99-057033 612.4’05^dc21 British Library Cataloguing in Publication Data
A catalogue record for this book is available from the British Library ISBN 0 471 99918 0 Typeset in 1012 on 1212 pt Garamond by DobbieTypesetting Limited, Tavistock, Devon. Printed and bound in Great Britain by Biddles Ltd, Guildford and King’s Lynn. This book is printed on acid-free paper responsibly manufactured from sustainable forestry, in which at least two trees are planted for each one used for paper production.
Contents Symposium on Mechanisms and biological signi¢cance of pulsatile hormone secretion, held atthe Novartis Foundation, London, 2^4 March1999 Editors: Derek J. Chadwick (Organizer) and Jamie A. Goode This symposium is based on a proposal made by Franz Schaeferand Johannes D.Veldhuis Johannes D.Veldhuis
Introduction 1
Nicholas S. Foulkes, Nicolas Cermakian, David Whitmore and Paolo Sassone-Corsi Rhythmic transcription: the molecular basis of oscillatory melatonin synthesis 5 Discussion 15 Albert Goldbeter, Genevie've Dupont and Jose¤ Halloy of pulsatility 19 Discussion 36
The frequency encoding
Pierre De Meyts and Ronald M. Shymko Timing-dependent modulation of insulin mitogenic versus metabolic signalling 46 Discussion 57 David J.Waxman Growth hormone pulse-activated STAT5 signalling: a unique regulatory mechanism governing sexual dimorphism of liver gene expression 61 Discussion 75 Steven M. Pincus Discussion 96
Orderliness of hormone release
82
G. Brabant and K. Prank Prediction and signi¢cance of the temporal pattern of hormone secretion in disease states 105 Discussion 114 Francis Le¤vi Therapeutic implications of circadian rhythms in cancer patients 119 Discussion 136 v
vi
CONTENTS
Georges Copinschi, Karine Spiegel, Rachel Leproult and EveVan Cauter Pathophysiology of human circadian rhythms 143 Discussion 157 Johannes D.Veldhuis Nature of altered pulsatile hormone release and neuroendocrine network signalling in human ageing: clinical studies of the somatotropic, gonadotropic, corticotropic and insulin axes 163 Discussion 185 Peter Butler Pulsatile insulin secretion 190 Discussion 199 Iain C. A. F. Robinson secretagogues 206 Discussion 220
Control of growth hormone (GH) release by GH
Franz Schaefer Pulsatile parathyroid hormone secretion in health and disease Discussion 239
225
S. L. Lightman, R. J.Windle, M. D. Julian, M. S. Harbuz, N. Shanks, S. A.Wood, Y. M. Kershaw and C. D. Ingram Signi¢cance of pulsatility in the HPA axis 244 Discussion 257 Johannes D.Veldhuis
Closing remarks 261
Index of contributors
263
Subject index
265
Participants Georg Brabant Abt. Klinische Endokrinologie, Medizinische Hochschule Hannover, Carl-Neubergstr. 1, D-30623 Hannover, Germany David Brown Laboratory of Computational Neuroscience,The Babraham Institute, Cambridge CB2 4AT, UK Peter Butler* Department of Medicine, University of Edinburgh,Western General Hospital, Edinburgh EH4 2XU, UK Iain J. Clarke Prince Henry’s Institute of Medical Research, PO Box 5152, Clayton,VIC 3168, Australia Georges Copinschi Centre for the Study of Biological Rhythms and Laboratory of Experimental Medicine, Universite¤ Libre de Bruxelles, 808 Route de Lennik, B-1070 Brussels, Belgium Pierre De Meyts Department of Receptor Biology, Hagedorn Research Institute, Niels Steensens Vej 6, DK-2820 Gentofte, Denmark Sta¡an Ede¤n Department of Physiology, University of G˛teborg, Medicinaregatan 11, 413 90 Go«teborg, Sweden Albert Goldbeter Unite¤ de ChronobiologieThe¤ orique, Faculte¤ des Sciences, Universite¤ Libre de Bruxelles, Campus Plaine, CP 231, B-1050 Brussels, Belgium Allan E. Herbison Laboratory of Neuroendocrinology,The Babraham Institute, Cambridge CB2 4AT, UK Lise Lund Kjems (Bursar) Department of Medicine,The University of Edinburgh,Western General Hospital, Edinburgh EH4 2XU, UK *Present address: Division of Endocrinology and Diabetes, Department of Medicine, University of Southern California, 1333 San Pablo Street, BMT-B11, Los Angeles, CA 900899113, USA vii
viii
PARTICIPANTS
Gareth Leng Department of Physiology, University Medical School,Teviot Place, Edinburgh EH8 9AG, UK Francis Le¤vi Laboratoire Rythmes Biologiques et Chronothe¤ rapeutique, Institut du Cancer et d’Immunoge¤ ne¤ tique. Ho“pital Paul Brousse, 94800 Villejuif, France Julio Licinio Clinical Neuroendocrinology Branch, National Institutes of Mental Health, Bldg 10, Room 2D46, 10 Center Drive, Bethesda, MD 20892-1284, USA Sta¡ord L. Lightman Dorothy Crowfoot Hodgkin Laboratories, University of Bristol, Divison of Medicine, Bristol Royal In¢rmary, Bristol BS2 8HW, UK John C. Marshall Department of Internal Medicine, University of Virginia Medical Center, (Box 612), Charlottesville,VA 22908, USA David R. Matthews Chairman, Oxford Centre for Diabetes Endocrinology and Metabolism,The Radcli¡e In¢rmary,Woodstock Road, Oxford OX2 6HE, UK Agneta Mode Department of Medical Nutrition, Karolinska Institute, Novum, Huddinge S-14186, Sweden Steven M. Pincus
990 Moose Hill Road, Guilford, CT 06437, USA
Iain C. A. F. Robinson Division of Neurophysiology, National Institute of Medical Research,The Ridgeway, Mill Hill, London NW7 1AA, UK Paolo Sassone-Corsi Institut de Ge¤ ne¤ tique et de Biologie Mole¤ culaire et Cellulaire, CNRS-INSERM-ULP, 1 rue Laurent Fries, 67404 IllkirchStrasbourg, France Franz Schaefer Division of Pediatric Nephrology, University Children’s Hospital, Im Neuenheimer Feld 150, D-69120 Heidelberg, Germany Johannes D.Veldhuis (Chair) Division of Endocrinology and Metabolism, Department of Internal Medicine General Clinical Research Center, Center for Biomathematical Technolgy, University of Virginia, Charlottesville,VA 22908, USA
PARTICIPANTS
David J.Waxman Division of Cell and Molecular Biology, Department of Biology, Boston University, Boston, MA 02215, USA Frederick C.W.Wu Department of Endocrinology, Manchester Royal In¢rmary, Oxford Road, Manchester M12 9WL, UK
ix
Introduction Johannes D. Veldhuis Division of Endocrinology and Metabolism, Department of Internal Medicine General Clinical Research Center, Center for Biomathematical Technology, University of Virginia, Charlottesville, VA 22908, USA
One of my favourite de¢nitions from the Oxford English Dictionary is that of wisdom. At this Symposium our assignment is to address the ‘wisdom’ of the neuroendocrine axis that is, to assess ‘aptness in the choice of ends and the pursuit of means’ of these pulsatile signalling systems. I take this challenge to include discussion of the mechanisms by which neuroendocrine signals are generated, communicated, transduced and fed back. Because of the broad range of time-scales inherent in di¡erent biological signalling systems, the challenge is particularly complex. By way of an overview, a prototypical signalling axis functions as a well integrated homeostatic unit. Exactly how this is accomplished in any given signalling system is not yet well understood. Indeed, whereas one’s research focus is often directed toward understanding a gene, an enzyme, a structural feature of a cell or a collection of cells, the physiological merit of homeostatic signalling is the dynamic ability of the axis to make minute and frequent adjustments, as necessary, to maintain the life of the organism as a whole. Given the hierarchical controls within an axis, it will be a formidable task in the immediate future to assemble the details of any given signalling system into a coherent, well articulated and comprehensive ensemble. Because most available data consist of successive measurements of output from only a single signalling element in the system over time, I believe that success in this lofty aim will require the broad and concerted interdisciplinary contributions of biomedical, computational and mathematical colleagues. One proximate goal of this colloquium is to examine critically the more plausible, and experimentally tractable, hypotheses that predict the biological relevance of pulsatile signalling mechanisms. Most of our current notions about signal regulation are not complete, and undoubtedly some are incorrect. Discussions in this colloquium should challenge ill-supported views and propose others, when appropriate. Many feedback systems show dominant frequency and/or amplitude control. Indeed, signal variation is a hallmark of all axes. However, the exact signi¢cance of the individual and multiple mechanisms of signal variation is not always clear. 1
2
INTRODUCTION
The present Novartis Foundation symposium examines this issue in depth by bringing together a wide range of signalling expertise. Specialized neuroglandular interfaces, such as the hypothalamic^pituitary junction, constitute a unique paradigm of signal transfer and reconstitution. For example, episodic neuronal impulses from the brain must be conveyed to and transduced by the pituitary gland, which is a sparingly excitable target tissue. Discussants explore the mechanisms of this physiological ‘interface problem’ in relation to several distinct neuroendocrine axes. For example, one ostensible consequence of the CNS^hypothalamic or brain^pituitary nexus is integration of short-term (ultradian) pulse mechanisms with long-term (e.g. circadian) rhythms. Whereas such integration can be de¢ned by analytical means (e.g. circadian variations in neurohormone pulse interval or mass), precisely how the organism accomplishes this neuro^glandular connectivity between divergent (ultradian and circadian) time-scales is poorly understood. A recent construct in neuropeptide signalling is the concept of burst-like neurohormone secretion. This notion posits that neuroendocrine cells store hormones in granules, which can be discharged abruptly when signalled by a relevant incoming stimulus. In some sense, this model is very economical, since a hormone with a reasonably long half-life in blood can be released only intermittently and still generate more or less unchanging target-cell exposure. Stimulus-dependent burst-like release also allows for a rapid increment in secretion, if cellular stores of the hormone are adequate. However, in the burst construct, more long-lived e¡ector molecules create an impediment to rapid signal turn-o¡. In this regard, the stress-adaptive corticotropic axis achieves rapidity both of turn-on and turn-o¡, given the rather brief half-life of adrenocorticotropic hormone (12^16 min in the human). The more prolonged half-life of cortisol (60^90 min) allows consistency of glucocorticoid exposure to this life-sustaining hormone. Much also needs to be learned about the role of signalling at the level of the receptor. In some cases, intermittent neurohormone signals allow receptor recycling, and drive hormone, cell and time-dependent second messengers. This important domain of work is addressed by several discussants. The mathematical biologists must join in deciphering how interlinked, timedelayed and non-linear stochastic feedback/feedforward systems are intrinsically stable. Of course, we know intuitively that, in health, physiological signalling systems behave well for 6^8 decades of life. However, subtle disruptions become evident in the course of ageing, and marked or complete loss of stability can emerge in disease states. Timing within these whole-organismic feedback control systems is often supervised by neural (CNS) pulse-generating mechanisms. The driving principles that organize such neuronal pulsatility are by no means established, and may di¡er among cells; e.g. the gonadotropin-releasing hormone (GnRH)
INTRODUCTION
3
neuron, pancreatic b cell, or pinealocyte. An intuitive view is that joint intracellular and intercellular mechanisms supervise the ensemble output, but precise mechanisms are incompletely understood. Indeed, the complexity of intracellular signal control alone is substantial at the second-messenger level. Multiple intracellular signal arborizations extend complexity to tertiary, quaternary and more distal ‘cascading’ e¡ects within the cell. Thus, beginning to bridge the current gulf between transcriptional gene regulation and bifurcating second, third, fourth etc. messenger pathways that act singly, jointly, and multiply is a critical long-term challenge in the signalling ¢eld. An exciting proposition discussed further in this symposium is that the ¢nely structured time pattern of neurohormone output can be used as a barometer of the integrative behaviour of a feedback unit. This is illustrated by the ability of time series that are equivalent in their mean to show very di¡erent behaviour microscopically; e.g. progressive corruption of a recurrent sine or cosine wave can produce new series with the same mean but with graded distinctions in ¢ne structure. Quantifying this subordinate structure (e.g. via approximate entropy, ApEn) is one promising avenue for extracting insights into the behaviour of an integrated network without analysing output of all regulatory nodes simultaneously, a formidable experimental prospect. Some practical medical applications are evident in this new strategy; e.g. all endocrine neoplasms studied so far show reduced orderliness of the ¢ne structure of their secretory patterns, whether assessed by the strategy of ApEn or the independent technique of predictive networks. Analogously, loss of subpattern reproducibility of neurohormone release is one of the earliest quanti¢able ageing-associated alterations in neuroendocrine control, developing before any measurable changes in mean blood hormone concentrations. These observations raise such questions as whether so-called signalling (or system) noise increases with rising age; what subtle defects in feedback and/or feedforward control occur in ageing; and/or whether small homeostatic shifts in dose^response linkages emerge with ageing. This symposium uses interdisciplinary interactions to approach some of the foregoing fundamental mechanistic and integrative implications of signalling. The physiological impact of the pulsatile mode of signal delivery is most evident for the GnRH^luteinizing hormone interface and the growth hormone^target tissue interface (especially in the rodent), and likely to be signi¢cant for insulin, glucagon and parathyroid hormone actions as well. Much remains to be learned concerning the pathophysiology of signal disruption; the nature of the basic pulse-generating mechanisms; the varying implications to the organisms of di¡erent hierarchies of rhythms (e.g. circadian, ultradian, seasonal); the complexity of controls among multiple interactive messenger systems within an individual cell; and the molecular mechanisms of signal-speci¢c gene promoter regulation. These issues constitute the ¢rst stage in
4
INTRODUCTION
ultimately unravelling signalling behaviour at the level of molecules and atoms, which subserve the regulation of genes, enzymes, cells, organs, systems and a thriving organism. The Novartis Foundation is to be congratulated for its prescience and wisdom in launching a multidisciplinary strategy as the means to the apt pursuit of these ends.
Rhythmic transcription: the molecular basis of oscillatory melatonin synthesis Nicholas S. Foulkes, Nicolas Cermakian, David Whitmore and Paolo Sassone-Corsi* Institut de Ge¤ ne¤ tique et de Biologie Mole¤ culaire et Cellulaire, CNRS-INSERM-ULP, 1 rue Laurent Fries, 67404 Illkirch-Strasbourg, France
Abstract. Pulsatile hormone synthesis and secretion are characteristic features of various oscillatory biological systems. Circadian rhythms are critical in the regulation of most physiological functions, and much interest has been centred on the understanding of the molecular mechanisms governing them. Adaptation to a changing environment is an essential feature of physiological regulation. The day^night rhythm is translated into hormonal oscillations governing the metabolism of all living organisms. In mammals the pineal gland is responsible for the circadian synthesis of the hormone melatonin in response to signals originating from the endogenous clock located in the hypothalamic suprachiasmatic nucleus (SCN). The molecular mechanisms involved in rhythmic synthesis of melatonin involve the CREM gene, which encodes transcription factors responsive to activation of the cAMP signalling pathway. The CREM product, ICER, is rhythmically expressed and participates in a transcriptional autoregulatory loop which also controls the amplitude of oscillations of serotonin N-acetyl transferase, the rate-limiting enzyme of melatonin synthesis. Thus, a transcription factor modulates the oscillatory levels of a hormone. 2000 Mechanisms and biological signi¢cance of pulsatile hormone secretion. Wiley, Chichester (Novartis Foundation Symposium 227) p 5^18
Day^night and seasonal changes in the environment dominate the lives of plants and animals, thus many facets of physiology are adapted to anticipate these changes. In vertebrates, the endocrine system plays a key role in synchronizing physiology with the environment. Circadian and seasonal rhythmicity characterize the action of many hormones which ultimately direct long-term changes in gene expression (Felig et al 1987, Krieger 1979). Thus, the properties of transcription factors and the signalling pathways which regulate them, constitute an essential link in the relay of temporal information. Fundamental to temporal adaptations made by animals is the presence of an internal circadian clock
*Corresponding author 5
6
FOULKES ET AL
(Ascho¡ 1981). Daily input of light and other stimuli continually reset this clock and synchronize it with the environment. Clock output pathways subsequently modulate various aspects of physiology. One key hormonal output from the clock is the night-time production of melatonin. In vertebrates this is synthesized by the retina and pineal gland. The pineal gland The pineal gland in vertebrates is a small structure located in the centre of the skull between the two cerebral hemispheres. In birds, reptiles, amphibia and ¢sh it is directly light-sensitive indeed one of its popular names is ‘the third eye’ (Collin 1971, Dodt 1973, Oksche 1984). In addition, it possesses an independent circadian clock (Takahashi et al 1980, Menaker & Wisner 1983). In contrast, in mammals, pinealocytes are neither light-sensitive nor possess a clock. The clock is instead located in the hypothalamic suprachiasmatic nucleus (SCN). Light stimuli are conveyed to the SCN indirectly via the retinohypothalamic pathway. The pineal gland is a principal site for melatonin production. Amongst the primary roles of melatonin in mammals are: (a) the regulation of seasonal changes in reproductive activity in response to changes in daylength; (b) entrainment of the SCN clock to ensure synchronicity with the environment; and (c) in the retina, where melatonin is also synthesized, this hormone has been implicated in the regulation of photopigment disc shedding, phagocytosis and the inhibition of retinal dopamine release (Cahill 1996, Tosini & Menaker 1996). Melatonin mediates its biological e¡ects by binding to high a⁄nity receptors belonging to the seven-transmembrane G protein-coupled receptor superfamily. The synthesis of melatonin begins with the N-acetylation of serotonin followed by addition of a methyl group at the 5-hydroxy position via the enzyme, hydroxyindole-Omethyltransferase (HIOMT) (Axelrod & Weissbach 1960). The melatonin synthetic pathway is shown in Fig. 1. In the pineal gland there is a strong circadian rhythm of melatonin synthesis. Whereas serotonin levels are much lower at night than day, melatonin concentrations display a reversed rhythm with highest concentration at night associated with elevated circulating levels of melatonin. The link between these two reciprocal rhythms is the rate-limiting enzyme for melatonin synthesis, serotonin N-acetyltransferase (AANAT). This displays a diurnal rhythm of activity with levels at the night-time peak up to 100 times higher than the daytime (Klein & Weller 1970). The AANAT and melatonin rhythms derive from activation at night of the pineal’s sympathetic innervation in mammals. Norepinephrine binds to b-adrenoceptors and thus stimulates adenylate cyclase activity. The resulting increase in cAMP levels have been shown to stimulate
RHYTHMIC TRANSCRIPTION
7
FIG. 1. Melatonin synthesis pathway. Serotonin is acetylated by serotonin N-acetyltransferase (AANAT) to produce N-acetylserotonin. N-acetylserotonin is in turn methylated by the enzyme hydroxyindole-O-methyltransferase (HIOMT).
AANAT transcription, translation and also maintain the enzyme in an active form (Takahashi 1994). a1-adrenergic receptors also participate in AANAT stimulation (Klein et al 1983), apparently by activating the phosphoinositide (PI) cycle and protein kinase C (PKC) which potentiates b-receptor-induced cAMP production. The AANAT enzyme was ¢rst cloned from sheep by screening a cDNA expression library using an enzymatic assay for acetylation of arylalkylamine substrates (Coon et al 1995). The rat cDNA was subsequently isolated using a PCR-based subtractive hybridization technique involving day and night pineal gland RNA (Borjigin et al 1995). The night-enriched AANAT cDNA displayed a diurnal rhythm of expression in the pineal gland, identical to that of AANAT activity. Interestingly, AANAT expression in the rat contrasts with the ovine pineal where levels of the AANAT transcript change only slightly between day and night while AANAT activity oscillates strongly (Coon et al 1995, Borjigin et al 1995, Roseboom et al 1996). The rapid rise and fall in rat AANAT mRNA levels indicate that transcriptional regulation is a primary determinant of AANAT function in this species.
8
FOULKES ET AL
Linking the clock to melatonin synthesis: cAMP-responsive transcription factors Night-time release of norepinephrine in the mammalian pineal activates b-adrenoreceptors and thereby stimulates adenylate cyclase activity. The associated intracellular rise in cAMP is a key step in the subsequent up-regulation of AANAT and melatonin synthesis (Klein 1985, Sugden et al 1985, Vanecek et al 1985). In the avian pineal gland these changes in cAMP occur under endogenous circadian clock control. Increases in intracellular cAMP levels lead to activation of cAMPdependent protein kinase A (PKA) and the transport of active catalytic subunits to the nucleus (Krebs & Beavo 1979). Nuclear phosphorylation targets include a group of transcription factors which modulate the expression of cAMP responsive genes (Sassone-Corsi 1995). These factors constitute a family of both activators and repressors which bind as homo- and heterodimers to cAMP-responsive elements (CREs). They belong to the basic leucine zipper (bZip) class of transcription factors and their function is tightly regulated by phosphorylation (Sassone-Corsi 1995). Constitutively expressed factors such as CREB (CRE-binding protein) are phosphorylated by PKA and thereby converted into transcriptional activators. Their transcriptional activation domain contains a phosphorylation box (P-box) with consensus phosphorylation sites for several protein kinases, including PKA (Sassone-Corsi 1995). This is £anked by glutamine-rich regions, termed Q1 and Q2, which are believed to make contacts with the basal transcriptional machinery. The CRE-modulator (CREM) gene is closely related to CREB and in common with other cAMP responsive factors generates a family of alternatively spliced isoforms (Foulkes et al 1991, Laoide et al 1993). A unique feature of CREM, however, is the presence of two alternative DNA binding domains, interchanged by the alternate use of splicing acceptor sites (Foulkes et al 1991). CREM isoforms function as both activators and repressors of cAMP-directed transcription and have a characteristic cell- and tissue-speci¢c pattern of expression with high levels of CREM activators, notably in the testis (Foulkes et al 1992). In addition, the use of an alternative cAMP-inducible promoter (P2) at the 3’ end of the CREM gene generates the factor ICER (Inducible cAMP Early Repressor) (Stehle et al 1993, Molina et al 1993). This small factor contains only the DNA binding domain consisting of the leucine zipper and basic region and functions as a dominant repressor of cAMP-induced transcription (Fig. 2). It acts by binding to CREs either as a homodimer or as heterodimeric complexes with other CRE activators. Since it lacks the activation domains, ICER repression function is primarily regulated by its intracellular concentration (Stehle et al 1993, Molina et al 1993). Stimuli that increase CREB phosphorylation have thus been associated with increases of ICER levels. ICER subsequently repress the same genes that are activated by phospho-CREB. Furthermore, ICER participates in a negative
RHYTHMIC TRANSCRIPTION
9
FIG. 2. Two promoters within the CREM gene. The CREM gene encodes various isoforms using alternative splicing, alternative translation initiations and alternative promoters (SassoneCorsi 1995). P1 is a constitutive, housekeeping promoter which directs the expression of the activator form CREMt, highly abundant in testis. S indicates the serine residue at position 117, the phosphoacceptor site for PKA and other kinases. The 3’ UTR contains various AUUUA sequences which confer instability to the transcript (small boxes) and three alternative polyadenylation sites. The P2 promoter, located in an intron upstream from the DNA-binding domain, directs the synthesis of the ICER transcript. The P2 promoter is cAMP-inducible by virtue of four CREs. The ICER protein acts as a powerful repressor of cAMP-induced transcription (Sassone-Corsi 1995).
autoregulatory loop (Molina et al 1993) since ICER protein binds to the CRE elements in its own promoter and represses its own transcription.
ICER and rhythmic melatonin synthesis Like AANAT, ICER mRNA expression displays diurnal rhythmicity in the pineal gland (Stehle et al 1993). The peak of ICER mRNA occurs during the second part of the night, just preceding the decline of melatonin synthesis. Interestingly, this pattern is developmentally regulated, being absent at birth and maturing only between the ¢rst and second week of postnatal development (Stehle et al 1995). This coincides with the maturation of a functional sympathetic innervation linking the SCN and pineal as well as maturation of cAMP inducibility of gene expression within the pineal and the appearance of elevated night-time melatonin synthesis (Stehle et al 1995). Together, these observations suggested that ICER might function as a down-regulator of melatonin production by repressing cAMP-induced AANAT transcription at the end of the night (Stehle et al 1995). A direct evaluation of the relationship between ICER and AANAT has been made
10
FOULKES ET AL
possible by the generation of mice which carry a null mutation at the CREM locus (Nantel et al 1996). Using mice to assess AANAT regulation posed potential problems since the majority of inbred strains used for transgenic and homologous recombination experiments have genetic defects in melatonin synthesis (Goto et al 1989). By biochemical and genetic analyses, defects in AANAT or AANAT regulators and HIOMT have been implicated in this de¢ciency (Ebihara et al 1986). Thus the ¢rst step was to test whether AANAT mRNA is expressed in the 129/sv strain used for the CREM knockout studies. By using a RNAse protection assay to analyse mouse pineal RNA, it was demonstrated that this mouse strain does indeed show a nighttime induction in AANAT expression, the timing of which is identical to that in the rat. The same AANAT expression pattern was encountered in C3H/He mice an outbred mouse strain which does produce melatonin (Foulkes et al 1996a). This indicates that the genetic defect in melatonin biosynthesis cannot be accounted for at the level of AANAT transcription. The two mice strains also display equivalent pro¢les of elevated ICER night-time expression, again with the timing being the same as that in the rat (Foulkes et al 1996a). Expression of the transcription factor Fra-2 (Fos-related antigen) was also tested in the mouse pineal. Fra-2 mRNA and protein have been documented to vary diurnally in the rat pineal gland with an elevation in the early part of the night which appears to be directed by adrenergic signals (Baler & Klein 1995). Furthermore, Fra-2 has been implicated as a negative regulator of AANAT expression (Baler & Klein 1995). The kinetics of Fra-2 expression in the mouse are the same as those in the rat. Thus, the patterns of ICER, AANAT and Fra-2 expression indicate that rat and mouse pineal glands can be considered equivalent in terms of adrenergically regulated gene expression (Foulkes et al 1996a). With the exception of time points during the day where mutant and wild-type control animals display an equivalent low basal level of expression, the mutant animals have signi¢cantly higher levels of night-time AANAT mRNA than their wild-type counterparts (Foulkes et al 1996a). Speci¢cally, in the CREM null mutants a rise in AANAT transcript is detected earlier at the beginning of the night, achieves at a higher peak of expression and then persists longer than in wild-type siblings. Thus the consequence of removal of ICER protein seems to be the relief of a general dampening e¡ect upon night-time AANAT expression. In contrast, the timing and magnitude of Fra-2 expression is equivalent in wildtype and mutant animals. Normal Fra-2 expression in the mutant animals demonstrates that the deregulation of AANAT expression does not extend to all adrenergically regulated genes. Furthermore the Fra-2 result indicates that clockderived adrenergic signals are not grossly altered in the knockout animals. To analyse the molecular mechanisms whereby ICER down-regulates AANAT expression, we cloned and sequenced the AANAT promoter. A CRE
RHYTHMIC TRANSCRIPTION
11
(TGACGCCA), di¡erent from the consensus (TGACGTCA; Sassone-Corsi 1995) by only one mismatch, was identi¢ed at position 108. A 378 bp promoter fragment including this region is su⁄cient to direct cAMP-inducible transcription of a reporter gene and also the down-regulation of cAMP-activated transcription by coexpressed ICER. ICER protein generated in bacteria binds to the AANAT CRE. Moreover a high mobility complex binds to this CRE in nuclear extracts prepared from mouse and rat pineal glands which is absent in extracts prepared from the CREM knockout mice. Thus, ICER protein binds to the AANAT promoter at the CRE in vivo. The AANAT transcript is up-regulated at all stages of its night-time induction in CREM knockouts relative to wild-type controls. This indicates that ICER dampens AANAT transcription throughout the night and not as originally predicted, at the end of the night when melatonin synthesis falls. Consistent with this function, the ICER protein persists throughout the day^night cycle in contrast to the strong diurnal variations in its mRNA (Foulkes et al 1996b). These ¢ndings support the following scenario for ICER function in the rat pineal gland (Fig. 3). Adrenergic stimulation at the onset of the night induces
FIG. 3. The role of the CREM feedback loop in transducing a rhythmic clock-directed signal into rhythmic hormone synthesis. Schematic representation of the regulatory pathway responsible for generating rhythmic melatonin synthesis. Night-time adrenergic signals originating from the clock, activate PKA and thus phosphorylate CREB. During the day, dephosphorylation is achieved by phosphatase action. Thus clock-directed signals determine the equilibrium position. Phosphorylated CREB activates the P2 promoter of the CREM gene and thus induces the expression of ICER. ICER down regulates its own expression constituting the CREM feedback loop. The balance between the proportion of phosphorylated CREB (positive e¡ect) and ICER protein levels (negative e¡ect) determines the transcriptional activity of the AANAT promoter. Thus the promoter cycles between activated and repressed states as a function of time. In this way, AANAT mRNA oscillates between high night-time and low basal daytime levels and determines the characteristic day^night oscillation of AANAT activity. This ensures rhythmic melatonin synthesis.
12
FOULKES ET AL
CREB phosphorylation while the termination of adrenergic stimulation towards morning is associated with CREB dephosphorylation (Foulkes et al 1996a). Abundant evidence indicates that CREB phosphorylation involves PKA (Sassone-Corsi 1995) while a phosphatase that dephosphorylates CREB in the pineal has yet to be identi¢ed. Phosphorylated CREB binds to the CREM P2 promoter and thereby activates night-time transcription of ICER. Dephosphorylation of CREB and the instability of the ICER transcript causes ICER mRNA levels to fall to low basal levels by the beginning of the day. In contrast, the ICER protein is more stable and persists at elevated levels throughout the day and night. Via binding to the CRE in the AANAT promoter, ICER modulates the rate and magnitude of melatonin induction in response to adrenergic signals by exerting a dampening e¡ect (Fig. 2). Thus the negative regulatory role of ICER operates throughout the 24 h cycle and not exclusively during the down-regulation of melatonin synthesis that occurs at the end of the night. Normal Fra-2 expression in the CREM mutant mice indicates that negative regulation by ICER in the pineal gland does not extend equally to all adrenergically regulated genes. Di¡erential binding a⁄nities of activators and repressors to the respective CREs may explain this observation. A central conclusion of our studies concerns the oscillations of hormonal synthesis. Indeed, regulation of AANAT by the CREM feedback loop may constitute a paradigm for how transcriptional autoregulatory loops control oscillatory responses (Fig. 3) (Sassone-Corsi 1994). Summary and perspectives The cloning of the AANAT gene and the demonstration that it is regulated by ICER has provided important insights into the molecular mechanisms underlying rhythmic expression. A number of questions concerning pineal gland physiology might be resolved based upon these ¢ndings. For example, in some species such as the sheep, AANAT transcripts oscillate only weakly compared with the rat. Also in the mouse, overall levels of AANAT transcript are substantially lower than in rat. This implies an inherent variability in the mode of AANAT regulation between di¡erent mammalian species. These di¡erences might re£ect di¡erent relative contributions of ICER and other transcriptional regulators to AANAT transcriptional regulation. Although the transcriptional mechanisms within mammalian pinealocytes documented here are potentially of much wider importance, the function of the mammalian pineal gland is regulated only indirectly by the SCN clock. It will thus be of great interest to assess the relative contribution of ICER and other transcriptional regulators to rhythmic melatonin production in lower vertebrates where the cAMP £uxes which drive melatonin synthesis are generated by an
RHYTHMIC TRANSCRIPTION
13
endogenous pinealocyte clock. The diurnal oscillation of AANAT mRNA in chick pinealocyte cultures combined with the pattern of cAMP inducibility of AANAT expression has lead to the speculation that cAMP regulation of AANAT activity is primarily post-transcriptional in the chick pineal (Bernard et al 1997). Furthermore, it has been proposed that transcriptional regulation in the chick pinealocyte is clock driven via a mechanism independent of cAMP (Bernard et al 1997). It will be of great interest to determine whether CREB and ICER might still play a role in this chick pinealocyte clock output pathway.
Acknowledgements D. W. was supported by an EEC fellowship and N. C. by a HFSP postdoctoral fellowship. Our studies are funded by CNRS, INSERM, CHUR, Fondation pour la Recherche Me¤ dicale and Association pour Recherche sur le Cancer (P. S.-C.).
References Ascho¡ J 1981 Biological rhythms. In: Ascho¡ J (ed) Handbook of behavioral neurobiology, vol 4: Biological rhythms. Plenum Press, New York, p 3^10 Axelrod J, Weissbach H 1960 Enzymatic O-methylation of N-acetylserotonin to melatonin. Science 131:1312 Baler R, Klein DC 1995 Circadian expression of transcription factor Fra-2 in the rat pineal gland. J Biol Chem 270:27319^27325 Bernard M, Klein DC, Zatz M 1997 Chick pineal clock regulates serotonin N-acetyltransferase mRNA rhythm in culture. Proc Natl Acad Sci USA 94:304^309 Borjigin J, Wang MM, Snyder SH 1995 Diurnal variation in mRNA encoding serotonin Nacetyltransferase in pineal gland. Nature 378:783^785 Cahill GM 1996 Circadian regulation of melatonin production in cultured zebra¢sh pineal and retina. Brain Res 708:177^181 Collin JP 1971 Di¡erentiation and regression of the cells of the sensory line in the epiphysis cerebri. In: The pineal gland. Churchill Livingstone, Edinburgh (Ciba Found Symp 125) p 79^125 Coon SL, Roseboom PH, Baler R et al 1995 Pineal serotonin N-acetyltransferase: expression cloning and molecular analysis. Science 270:1681^1683 Dodt E 1973 In: Handbook of sensory physiology. Elsevier, New York, p 113^140 Ebihara S, Marks T, Hudson DJ, Menaker M 1986 Genetic control of melatonin synthesis in the pineal gland of the mouse. Science 231:491^493 Felig P, Baxter JD, Broadus AE, Frohman LA 1987 Endocrinology and metabolism. McGrawHill, New York Foulkes NS, Borrelli E, Sassone-Corsi P 1991 CREM gene: use of alternative DNA-binding domains generates multiple antagonists of cAMP-induced transcription. Cell 64:739^749 Foulkes NS, Mellstr˛m B, Benusiglio E, Sassone-Corsi P 1992 Developmental switch of CREM function during spermatogenesis: from antagonist to transcriptional activator. Nature 355:80^84 Foulkes NS, Borjigin J, Snyder SH, Sassone-Corsi P 1996a Transcriptional control of circadian hormone synthesis by the CREM feedback loop. Proc Natl Acad Sci USA 93:14140^14145
14
FOULKES ET AL
Foulkes NS, Duval G, Sassone-Corsi P 1996b Adaptive inducibility of CREM as transcriptional memory of circadian rhythms. Nature 381:83^85 Goto M, Oshima I, Tomita T, Ebihara S 1989 Melatonin content of the pineal gland in di¡erent mouse strains. J Pineal Res 7:195^204 Klein DC 1985 Photoneural regulation of the mammalian pineal gland. In: Photoperiodism, melatonin and the pineal. Pitman, London (Ciba Found Symp 117), p 38^56 Klein DC, Weller JL 1970 Indole metabolism in the pineal gland: a circadian rhythm in Nacetyltransferase. Science 169:1093^1095 Klein DC, Sugden D, Weller JL 1983 Postsynaptic a-adrenergic receptors potentiate the badrenergic stimulation of pineal serotonin N-acetyltrasferase. Proc Natl Acad Sci USA 80:599^603 Krebs EG, Beavo JA 1979 Phosphorylation^dephosphorylation of enzymes. Annu Rev Biochem 48:923^959 Krieger DT 1979 Endocrine rhythms. Raven Press, New York Laoide BM, Foulkes NS, Schlotter F, Sassone-Corsi P 1993 The functional versatility of CREM is determined by its modular structure. EMBO J 12:1179^1191 Menaker M, Wisner S 1983 Temperature-compensated circadian clock in the pineal of Anolis. Proc Natl Acad Sci USA 80:6119^6121 Molina CA, Foulkes NS, Lalli E, Sassone-Corsi P 1993 Inducibility and negative autoregulation of CREM: an alternative promoter directs the expression of ICER, an early response repressor. Cell 75:875^886 Nantel F, Monaco L, Foulkes NS et al 1996 Spermiogenesis de¢ciency and germ-cell apoptosis in CREM-mutant mice. Nature 380:159^162 Oksche A 1984 Evolution of the pineal complex: correlation of structure and function. Ophthalmic Res 16:88^95 Roseboom PH, Coon SL, Baler R, McCune SK, Weller JL, Klein DC 1996 Melatonin synthesis: analysis of the more than 150-fold nocturnal increase in serotonin N-acetyltransferase messenger ribonucleic acid in the rat pineal gland. Endocrinology 137:3033^3044 Sassone-Corsi P 1994 Rhythmic transcription and autoregulatory loops: winding up the biological clock. Cell 78:361^364 Sassone-Corsi P 1995 Transcription factors responsive to cAMP. Annu Rev Cell Dev Biol 11:355^377 Stehle JH, Foulkes NS, Molina CA, Simonneaux V, Pe¤ vet P, Sassone-Corsi P 1993 Adrenergic signals direct rhythmic expression of transcriptional repressor CREM in the pineal gland. Nature 365:314^320 Stehle JH, Foulkes NS, Pe¤vet P, Sassone-Corsi P 1995 Developmental maturation of pineal gland function: synchronized CREM inducibility and adrenergic stimulation. Mol Endocrinol 9:706^716 Sugden D, Vanecek J, Klein DC, Thomas TP, Anderson WB 1985 Activation of protein kinase C potentiates isoprenaline-induced cyclic AMP accumulation in rat pinealocytes. Nature 314:359^361 Takahashi JS 1994 Circadian rhythms. ICER is nicer at night (sir!). Curr Biol 4:165^168 Takahashi JS, Hamm H, Menaker M 1980 Circadian rhythms of melatonin release from individual superfused chicken pineal glands in vitro. Proc Natl Acad Sci USA 77:2319^2322 Tosini G, Menaker M 1996 Circadian rhythms in cultured mammalian retina. Science 272: 419^421 Vanecek J, Sugden D, Weller JL, Klein DC 1985 Atypical synergistic alpha 1- and betaadrenergic regulation of adenosine 3’,5’-monophosphate and guanosine 3’,5’-monophosphate in rat pinealocytes. Endocrinology 116:2167^2173
RHYTHMIC TRANSCRIPTION
15
DISCUSSION Veldhuis: To me, one of the most spectacular results you have published is the discovery of Clock gene expression in tissues such as the spleen and kidney (Whitmore et al 1998). What are these genes doing there? Sassone-Corsi: I agree that these are impressive results. We have recently cloned the partners for Clock, two transcription factors which work with Clock to regulate transcription. These partners oscillate di¡erentially in various tissues. Even more impressive is the fact that the same gene oscillates with di¡erent phases in di¡erent tissues. For example, it will peak at the middle of the night in the liver and the middle of the day in the heart. This tells us that there are intrinsic clocks in each tissue. These receive and process signals coming from the central clock in an independent fashion. As to their function, they are likely to regulate endogenous functions in each tissue. Veldhuis: In vivo are they oscillating at these di¡erent phases, or are their phases re-set centrally? Sassone-Corsi: The situation is a little more complicated than that. Classically, there is phase re-set in vivo in various systems. In vitro experiments show that there is oscillation in the retina, the pineal and in other tissues, and possible phase resetting. Clarke: In the light^dark phases, the Clock gene seems to be most highly expressed in the latter part of the light phase, before the onset of dark. What is the dark phase doing in terms of regulation of expression? Sassone-Corsi: This is an important point. In the case of the CREM gene, expression regulation precedes the signal. The gene is almost ‘expecting’ night to come, so it is already preparing for a new wave of expression. We have done experiments in which we changed the photoperiod in the rat and looked at CREM expression, and found that the gene needs some time to adjust. In reality, the gene is able to change and respond di¡erently to various photoperiods. Plasticity at the transcription level allows the gene to respond as an early response gene in a short photoperiod situation and almost as a constitutive gene in a long photoperiod situation. Clarke: But in continuous dark you have two rhythms? Sassone-Corsi: In continuous darkness after entrainment the animal will ‘remember’ the previous cycle. Matthews: How much phase drift occurs ex vivo with di¡erent tissues? Sassone-Corsi: That’s a tough question. You can only work in vitro for ¢ve or six days. Matthews: What happens to the phase relationship over this time? Sassone-Corsi: There is very little phase shift, although it is technically di⁄cult to study this.
16
DISCUSSION
Waxman: Last year a paper was published in Cell showing that if you give a high serum pulse to cells in culture, in the absence of any central signals, the cells are stimulated to go through multiple circadian rounds of transcription that last for two or three days (Balsalobre et al 1998). How do you envisage this might occur? Sassone-Corsi: This is work by Ueli Schibler in Geneva. We envisage this might occur in a similar way, by entrainment. The response to pulses of serum might mimic in cultured cells the response to pulses of light or adrenergic signalling in the animal. Waxman: If I remember correctly, in that study a single pulse of serum could stimulate the cells to go through a circadian rhythm, in the absence of light or central signalling. Sassone-Corsi: This is true: the serum is the signal at that point. Waxman: It is a single signal, which would suggest that there is an intrinsic timekeeper in the cells. Sassone-Corsi: Absolutely. My opinion is that every cell has an endogenous timekeeper. Is it a coincidence that the cell cycle is about 24 h? Although cell cycle and circadian rhythms are certainly distinct processes, it is evident that each cell has the endogenous property of ‘ticking’. In my mind, over the course of evolution single cells became responsive to day^night cycles and so developed a system intrinsic to the cell which allows it to conserve oscillatory function, even in complete darkness. De Meyts: If you say that every cell has in the course of evolution adapted to day^ night rhythms, are you suggesting that all cells still have a mechanism for photoreception? Sassone-Corsi: I believe so, although during evolution most cells have lost it. Brabant: Did you ever look at other means of entrainment of these genes? Sassone-Corsi: You can re-entrain clock function by other means, such as thermoregulation. One of the phenotypes of the CREM-knockout mice is a delayed liver regeneration due to defects in cell cycle. CREM has a function in regulating melatonin oscillation in the pineal, and in another it helps regulate cell cycle in hepatocytes. Marshall: You have shown that the stimulation and dephosphorylation is completed in about 2 h, and then protein synthesis continues for almost 24 h. This de¢nes the response time of the cAMP response system. Then we translate that to whether cells see day or night. Perhaps the timing is a function of that type of system and has a cyclicity of somewhere in the range of 2^24 h. Have you had the opportunity to look in some of those pineal cells at other second messenger systems? Are there very di¡erent cyclical time courses ongoing? Sassone-Corsi: We put pinealocytes in culture and observed oscillation. Pinealocytes respond very strongly to adrenergic signals. Experiments involving serum factors clearly induce di¡erent pathways activated by growth factors. In my lab we have shown that one subunit of p90rsk, Rsk2, is able to phosphorylate
RHYTHMIC TRANSCRIPTION
17
CREB at the same PKA site, thereby eliciting similar responses not in response to PKA and cAMP, but to the Ras/Raf/MAP kinase signalling pathway. Indeed, we now know that transcription factors are converging targets of various pathways at the same time. There are at least ¢ve di¡erent pathways leading to phosphorylation of CREB or CREM, including the stress pathway. When you deprive cells of serum, you induce the stress pathway and CREB phosphorylation occurs at the PKA phosphoacceptor serine with kinetics similar to the early response. Matthews: Have you got any views about the speed of transcription? What is the fastest and the slowest that you can run a transcription pulsatility clock? Sassone-Corsi: In the case of MAP kinase pathways, phosphorylation occurs within one minute. Activation and transcription cannot go faster than 10 or 15 min. Matthews: By the time you have got your inhibitor running back, what’s the periodicity of the cycle? Sassone-Corsi: We have done experiments in the rat by changing the photoperiod. You can have the same gene, with the same promoter, working in a short photoperiod as an early response gene, and in the long photoperiod with constitutive transcription. The animal adapts to di¡erent light conditions within only a week. So there is a direct adaptation to the photoperiod at a transcriptional level. Matthews: But what are the limits of that? What is the fastest you can do it? Sassone-Corsi: The fastest inducibility for an early response gene is 15 min. Matthews: But what I’m asking is, what is the fastest you can run the complete cycle? Sassone-Corsi: In experiments in cultured cells it is about 5 h. Goldbeter: I have a question regarding melatonin synthesis in the pineal. You emphasized the role of the ICER regulatory loop. It is known that feedback loops play a role in the regulation of circadian rhythmicity in Drosophila and Neurospora, for example. But it isn’t clear that the ICER loop plays a role in your system. You previously described the case where circadian rhythmicity was driven by the SCN in the pineal. Have you looked at the possible role of the ICER loop in avian pineal cell cultures, which were shown by Takahashi et al (1989) to behave as an autonomous pacemaker? Sassone-Corsi: That is a good question. We have done work in collaboration with Joe Takahashi on chick pineal cells, and have shown various things. One is that the NAT promoter in the chick has the same site for ICER binding. ICER oscillates in chick pineal in a similar way. In comparison with mammalian cells, chick cells have a better oscillation of CREB phosphorylation, which in mammalian cells is not great. It would appear that in the chick pinealocytes the situation is even more enhanced.
18
DISCUSSION
Leng: You have been talking about the molecular basis of pulsatility as elements that are intrinsic to a single cell. I guess that in the pineal gland, cell^cell interactions are also important in synchronizing the rhythms. For most endocrine systems, homotypic cell^cell interactions are important for synchronizing rhythms. Are they important in the pineal, and if so how do they a¡ect the dynamic processes of pulse generation? Sassone-Corsi: All rhythms that appear in the pineal are clearly synchronized: all pinealocytes work at the same time. In vivo, the oscillation of gene expression or melatonin synthesis are synchronized. Each single cell has the property of oscillating but if you take two independent cells they will probably become asynchronous after a while. Leng: This demonstrates that cell^cell interactions are important, but how do they a¡ect the dynamics of this molecular mechanism? Sassone-Corsi: That’s a tough one. There must be ways for di¡erent cells in di¡erent tissues to respond di¡erently from a central clock signal, because we have found that Clock and the Clock partners oscillate in di¡erent ways in di¡erent issues. The signal coming from the central clock must therefore be processed separately by each tissue, of it must arrive at the various tissues in a di¡erent way. Leng: What happens if you mix cultures which have established di¡erent rhythms? Sassone-Corsi: I don’t think anyone has done that. References Balsalobre A, Damiola F, Schibler U 1998 A serum shock induces circadian gene expression in mammalian tissue culture cells. Cell 93:929^937 Takahashi JS, Murakami N, Nikaido SS, Pratt BL, Robertson LM 1989 The avian pineal, a vertebrate model system of the circadian oscillator: cellular regulation of circadian rhythms by light, second messengers and macromolecular synthesis. Recent Progr Horm Res 45: 279^352 Whitmore D, Foulkes NS, Strhle U, Sassone-Corsi P 1998 Zebra¢sh Clock rhythmic expression reveals independent peripheral circadian oscillators. Nat Neurosci 1:701^707
The frequency encoding of pulsatility Albert Goldbeter, Genevie' ve Dupont and Jose¤ Halloy Unite¤ de Chronobiologie The¤ orique, Faculte¤ des Sciences, Universite¤ Libre de Bruxelles, Campus Plaine, C.P. 231, B-1050 Brussels, Belgium
Abstract. Examples of pulsatile signalling abound in intercellular communication, suggesting that this phenomenon represents a major function of biological rhythms. Pulsatile signals can be encoded in terms of their frequency and prove more e⁄cient than monotonous ones whenever constant stimulation induces desensitization of target cells. We address the main examples of frequency encoding of pulsatility, besides those of neuronal nature. Considered in turn are cAMP oscillations in the slime mould Dictyostelium discoideum, the pulsatile secretion of hormones such as gonadotropinreleasing hormone or growth hormone, intracellular Ca2+ oscillations, and circadian rhythms. Models based on receptor desensitization show the possibility of optimizing cellular responses to cAMP signals in Dictyostelium or to pulsatile hormonal stimulation. The models indicate how the optimal duration of the pulsatile signal and the optimal interval between successive pulses vary as a function of the rates or receptor desensitization and resensitization and of the maximum ligand level during stimulation. The frequency encoding of intracellular Ca2+ oscillations appears to rely on another molecular mechanism. Models based on protein phosphorylation by a Ca2+-calmodulin activated kinase show that the mean level of phosphorylated protein increases with the frequency of calcium spikeswhich itself rises with the degree of stimulationand that distinct levels of di¡erent phosphorylated proteins can be reached for a Ca2+ signal of given frequency. 2000 Mechanisms and biological signi¢cance of pulsatile hormone secretion. Wiley, Chichester (Novartis Foundation Symposium 227) p 19^45
A major role of biological rhythms is to provide support mechanisms for pulsatile signalling in intercellular communication. Periodic, pulsatile signals prove more e⁄cient in cases where constant signals induce desensitization of target cells. At the same time pulsatile signals prove superior to monotonous ones in that they can be encoded in terms of their frequency. Frequency encoding of pulsatile signals has been observed for many types of intercellular communication, as exempli¢ed by the temporal coding of trains of action potentials in the brain (Buszaki et al 1994). The frequency encoding phenomenon, however, is not restricted to neural signalling (Goldbeter 1996). Thus, it is also observed for a large variety of hormones in cell to cell communication and, intracellularly, for pulsatile Ca2+ signals which occur in 19
20
GOLDBETER ET AL
many cell types in response to external stimulation by a hormone or neurotransmitter. Communication by pulsatile signals also takes place between cells of the slime mould Dictyostelium discoideum which aggregate after starvation in response to pulses of cAMP emitted by cells which behave as aggregation centres. The purpose of this chapter is to give an overview of the frequency encoding of pulsatile signals in intercellular communication. We will focus on signals other than those of a neuronal nature. Thus, we shall consider in turn the frequency encoding of cAMP pulses in Dictyostelium cells, which provide a primitive example of pulsatile intercellular communication, before turning to the frequency encoding of pulsatile hormonal signals and of intracellular Ca2+ spikes. Finally, we shall brie£y discuss the frequency encoding of circadian rhythms, stressing their role in adaptation to a periodic environment as well as their clinical implications for chronotherapeutics. The study of theoretical models for pulsatile signalling shows that such a mode of intercellular communication allows for optimization in terms of frequency encoding of the pulsatile signals. A variety of molecular mechanisms may underlie the frequency encoding process, while di¡erent physiological responses may correspond to di¡erent optimal frequencies. Pulsatile signals of cAMP in Dictyostelium discoideum In many respects, the mechanism of intercellular communication in cellular slime moulds resembles hormone signalling in higher, multicellular organisms. While cAMP is often used as a second messenger synthesized intracellularly in response to hormonal stimuli, in the slime mould D. discoideum, it behaves at the same time as ¢rst and second messenger (Konijn 1972). Indeed, these amoebae synthesize cAMP in response to a cAMP stimulus (Roos et al 1975); the latter extracellular signal controls both the aggregation and the di¡erentiation of the amoebae after starvation (Gerisch 1987, Devreotes 1989). D. discoideum cells grow and divide in the unicellular stage as long as food is present. Starvation triggers a developmental program that leads these amoebae from aggregation on a solid support such as agar to the formation of a fruiting body consisting of a stalk surmounted by a mass of spores. The unicellular phase of the life cycle thereafter resumes with spore germination. The aggregation process is periodic: cells aggregate around centres which secrete cAMP pulses with a periodicity of about 10 min (Gerisch & Wick 1975) and relay these signals (Roos et al 1975) toward the periphery of the aggregation ¢eld. Concentric or spiral waves of amoeboid movement can be observed, which superimpose on waves of cAMP (Gerisch 1987, Devreotes 1989). The latter molecule serves as chemotactic
FREQUENCY ENCODING OF PULSATILITY
21
factor governing the aggregation of as many as 105 amoebae around a centre (Konijn 1972). The observation that pulses of cAMP promote cell di¡erentiation when delivered with a periodicity of 5 min, in contrast with constant stimuli (Darmon et al 1975, Gerisch et al 1975) or with signals delivered with the higher frequency of one pulse every 2 min (Wurster 1982) suggests the existence of an optimal frequency of periodic stimulation. The e⁄ciency of periodic signalling is associated with the process of desensitization of target cells under constant stimulation. Here, desensitization appears to involve the reversible phosphorylation of the receptor, following binding of extracellular cAMP (Devreotes & Sherring 1985). The origin of cAMP oscillations and the possible frequency encoding of cAMP pulses can be investigated by means of a theoretical model for cAMP signalling based on receptor desensitization (Martiel & Goldbeter 1987). The model, schematized in Fig. 1, takes into account the existence of the cAMP receptor in an active (R) or desensitized (D) state, only the former being able, upon binding of extracellular cAMP, to activate adenylate cyclase (C) which transforms ATP into cAMP. Cyclic AMP is transported into the extracellular medium where it binds to its receptor and thereby enhances its own production. Such a positive feedback loop forms the core of the instability mechanism that is responsible for the occurrence of sustained autonomous oscillations of cAMP. The variables considered in the model are the fraction of active cAMP receptor (rT ) and the normalized concentrations of intracellular (b) and extracellular (g) cAMP. When integrating numerically the kinetic equations which govern the
FIG. 1. Model for cAMP signalling in Dictyostelium discoideum. The model (Martiel & Goldbeter 1987, Goldbeter 1996) is based on the binding of extracellular cAMP to the plasma membrane receptor which exists in the active (R) and desensitized (D) states. Binding of cAMP to the active receptor state triggers a G protein-mediated activation of adenylate cyclase (C) which produces intracellular cAMP from ATP. Arrows refer to the hydrolysis of intracellular and extracellular cAMP by phosphodiesterase, and to the transport of intracellular cAMP into the extracellular medium. The latter process creates a positive feedback loop that forms the core of the mechanism of pulsatile synthesis of cAMP.
22
GOLDBETER ET AL
time evolution of the signalling system, sustained oscillations in the concentrations of intra- and extracellular cAMP are obtained (Martiel & Goldbeter 1987, Goldbeter 1996); as observed in the experiments (Klein et al 1985), these oscillations are accompanied by a periodic variation of the receptor between its active and desensitized states (Fig. 2A). Because of cAMP-induced desensitization, the fraction of active receptor drops as soon as the level of extracellular cAMP rises; the resulting decrease in cAMP synthesis and drop in cAMP level allow for receptor resensitization which eventually brings about a new burst in cAMP synthesis. The model can also be used to assess the in£uence of the frequency of cAMP pulses on the responsiveness of target cells. To this end, and to confer to their
FIG. 2. Oscillations and waves of cAMP generated by the model for cAMP signalling in Dictyostelium discoideum schematized in Fig. 1. (A) Oscillations of intracellular cAMP (of normalized concentration b) are accompanied by a periodic variation of the total fraction rT of active cAMP receptor (Martiel & Goldbeter 1987). (B) When di¡usion of extracellular cAMP is incorporated into the model, waves of cAMP can be obtained which are either concentric (left panel) or take the form of large spirals (middle panel) or numerous small spirals (right panels) depending on the degree D of cell desynchronization along the developmental path (Lauzeral et al 1997, Halloy et al 1998).
FREQUENCY ENCODING OF PULSATILITY
23
analysis a more general character, Li & Goldbeter (1989) determined the response of the signalling system in (non-physiological) conditions where the positive feedback process is suppressed in the scheme of Fig. 1. The capability of maximizing the number of signi¢cant responses, in the form of synthesis of large-amplitude pulses of cAMP, was determined in these conditions as a function of the duration t1 of a square-wave signal of extracellular cAMP and of the interval t0 between two such pulses. The optimal pattern (t0 ,t1 ) of duration and amplitude of the square-wave pulsatile signal was obtained for three sets of values of the rate constants k2 and k1 characterizing, respectively, the kinetics of desensitization and resensitization of the receptor. When inserting into the model the values taken from the literature for the kinetic constants governing the reversible phosphorylation of the cAMP receptor in D. discoideum, one obtains (see Table 1) an optimal pattern of periodic stimulation corresponding to the values t1 ¼ 2:7 min and t0 ¼ 4:75 min, for the pulse duration and the pulse interval (Li & Goldbeter 1989). These values are close to those observed for the periodic signal of cAMP generated by cells which behave as aggregation centres after starvation (Gerisch 1987, Devreotes 1989). The analysis indicates that the optimal pattern of pulsatile stimulation markedly depends on the desensitization and resensitization rate constants k2 and k1 : as indicated in Table 1, the optimal duration of the pulsatile stimulus and the optimal interval between pulses decrease as desensitization and resensitization become more rapid, respectively. TABLE 1 Optimal pattern of pulsatile stimulation predicted by the model for cAMP signalling in D. discoideum, as a function of the rates of receptor desensitization and resensitization Desensitization rate constant, k2 (min1 )
Resensitization rate constant, k1 (min1 ) 0.072
0.1332
0.666
3.33
t1 ¼ 13:50
3.32 20.3 2.70 4.75 2.28 1.24
0.78 19.2 0.66 4.06 0.53 0.93
t0 ¼ 23:65
0.36
t1 ¼ 11:3 t0 ¼ 6:18
1.8
t1 ¼ 9:50 t0 ¼ 1:60
From Li & Goldbeter (1990). The values (in min) of the optimal duration t1 and interval t0 were determined for the experimentally determined values of 0.36 and 0.666 min1 for the desensitization (k2) and resensitization (k1 ) ¢rst-order rate constants, and for values of these constants multiplied or divided by a factor of 5.
24
GOLDBETER ET AL
The above results on oscillations and frequency encoding have been obtained in a model based on receptor desensitization. Recent observations suggest that intracellular regulatory processes also play a role in controlling the response of the cells to cAMP signals (S˛derbom & Loomis 1998). We are currently extending the model for cAMP signalling to incorporate the control of adenylate cyclase by a cascade of phosphorylations triggered by the binding of extracellular cAMP to its receptor. In regard to frequency encoding of pulsatile signals, another interesting aspect of cAMP signalling in Dictyostelium bears on the desynchronization of cells in the course of development after starvation. A recent theoretical study (Lauzeral et al 1997, Halloy et al 1998) of wavelike aggregation in response to cAMP pulses released by aggregation centres indicates that the concentric or spiral nature of the waves can be directly related to the degree of desynchronization (D) of the cells along a developmental path corresponding to an increase in adenylate cyclase and phosphodiesterase during the hours that follow the onset of starvation. Such continuous increases in enzyme activities are responsible for the discontinuous developmental transitions no relay^relay^oscillations^relay observed during that phase of the life cycle. Cell desynchronization results from the fact that cells are caught in di¡erent phases of the cell cycle when they stop growing at the onset of starvation. When cells are only slightly desynchronized, concentric waves form. When desynchronization, measured by parameter D, progressively increases, the waves ¢rst acquire the form of large spirals before transforming into a large number of much smaller spirals (Fig. 2B). The model therefore suggests the existence of an optimal degree of cell desynchronization for the formation of large-amplitude spiral waves in the course of D. discoideum aggregation. Pulsatile hormone signalling As discussed and illustrated in several chapters of this volume, most hormones are secreted in a pulsatile rather than continuous manner. High-frequency hormonal pulses are often superimposed on a circadian hormone pro¢le (Van Cauter & Ascho¡ 1989). The prototype of pulsatile hormone secretion is that of the gonadotropin-releasing hormone (GnRH) which is secreted with a frequency of about 1 pulse per hour in the rhesus monkey and humans (Knobil 1980, Lincoln et al 1985). Other prominent examples include the growth hormone (GH), secreted with a frequency of about 1 pulse every 3 h (Tannenbaum & Martin 1976), and insulin which is released in a periodic manner with a frequency of 1 pulse every 13 min in a variety of organisms including humans (Goodner et al 1977). The case of GH presents the further interest that the temporal pulsatile pattern di¡ers in male and female rats. Such di¡erences appear to underlie the distinct growth
FREQUENCY ENCODING OF PULSATILITY
25
patterns and liver enzyme expression pro¢les observed in males and females (Waxman et al 1991, Waxman 2000, this volume). The physiological e⁄ciency of hormones is generally related to the frequency of their pulsatile secretion. Thus, the classical work of Knobil and co-workers (Pohl et al 1983) has shown that the release of luteinizing hormone (LH) and folliclestimulating hormone (FSH) by the pituitary can be induced by GnRH pulses delivered at the physiological, circhoral frequency but not when this frequency changes to two or three pulses per hour, or one pulse every two hours. As emphasized by Knobil (1981), the temporal pattern of the hormone can be as important as its concentration. These ¢ndings have led to clinical applications in the treatment of infertility. Women su¡ering from abnormal GnRH secretion are routinely implanted with a pump delivering one pulse of this hormone per hour; such a treatment restores LH and FSH patterns required for ovulation (Leyendecker et al 1980). Similar frequency-dependent physiological e¡ects have been demonstrated for growth hormone (Clark et al 1985, Hindmarsh et al 1990). The increased e⁄ciency of pulsatile insulin stimulation has also been documented (Matthews et al 1983). The molecular basis of frequency encoding of hormone pulses has been investigated in a general model based on receptor desensitization (Li & Goldbeter 1989, 1992). The response of such a receptor subjected to a squarewave, pulsatile variation of the hormonal ligand (Fig. 3) has been determined as a function of the duration (t1 ) of a pulse and the interval (t0 ) between successive pulses. When de¢ning target cell responsiveness as the capability to generate as many large-amplitude responses in a given amount of time under such pulsatile stimulation, the model shows that there exists an optimal pattern (t1 ,t0 ) that maximizes cell responsiveness (Fig. 4). Much as for the model discussed in the previous section for cAMP signalling in Dictyostelium cells, this optimal pattern not only depends on the kinetics of desensitization and resensitization, but also on the amplitude of the ligand pulse, as shown in Fig. 4 in which the three panels relate to a maximum ligand level during the pulse equal, from top to bottom, to 10 Kd, Kd, and 0.1Kd, where Kd denotes the dissociation constant of the ligand for the receptor in its active state. When the condition that the optimal pattern should correspond to a 5 min pulse spaced from the next pulse by about 1 h (as observed for GnRH) is inserted into the model, it predicts that desensitization of pituitary gonadotrophs should occur in minutes while resensitization should occur in about 20 min (Li & Goldbeter 1989). These predictions hold with the experimental values recently determined for these processes (Heinze et al 1998). Beyond the case of GnRH, the model of Fig. 3 and the results of Fig. 4 are of general signi¢cance for the frequency encoding of pulsatile hormonal signals. In particular, the analysis should apply to the cases of GH and insulin. The GH receptor has been shown to undergo desensitization.
26
GOLDBETER ET AL
FIG. 3. General model for pulsatile stimulation of a receptor undergoing reversible desensitization. The square wave variation of ligand L is characterized by a duration t1 of each pulse and an interval t0 between successive pulses; R and D denote the active and desensitized receptor states, respectively (Li & Goldbeter 1989, 1992).
Resensitization of this receptor takes about 3 h, which ¢ts with the interval that separates successive GH pulses (Bick et al 1989). Likewise, the period of insulin oscillations holds with the time, of the order of 13 min, required by the insulin receptor on adipocytes to return to the plasma membrane after hormone-induced internalization (Goodner et al 1988). Intracellular Ca2+ pulses Oscillations of cytosolic Ca2+ represent one of the most widespread examples of periodic behaviour at the cellular level (Berridge 1990, Meyer & Stryer 1991, Goldbeter 1996). These oscillations occur either spontaneously, as in cardiac cells, or in response to some hormonal signal like vasopressin in hepatocytes or histamine in endothelial cells. The mechanism of Ca2+ oscillations is generally based on the self-amplifying process of Ca2+-induced Ca2+ release that characterizes the transport of Ca2+ between the cytosol and intracellular stores such as the endoplasmic or sarcoplasmic reticulum.
FREQUENCY ENCODING OF PULSATILITY
27
FIG. 4. Cell responsiveness as a function of pulse duration (t1 ) and pulse interval (t0 ), at three di¡erent values of the maximum ligand level during stimulation (Li & Goldbeter 1992). The curves are generated for the model schematized in Fig. 3 for a maximum ligand level equal to 10 Kd (top), Kd (middle panel), and 0.1 Kd (bottom panel), where Kd represents the dissociation constant of the complex formed by the ligand with the receptor in its active state.
28
GOLDBETER ET AL
Depending on the cell type, the period of the oscillations ranges from seconds to minutes. Signal-induced Ca2+ spiking appears to be involved in a variety of physiological processes, including secretion, gene expression, the control of egg development after fertilization and the operation of metabolic pathways such as the phosphorylation^dephosphorylation cascade that governs the switch between glycogen synthesis and degradation in the liver. The most obvious physiological advantage of such repetitive Ca2+ increases is to allow a signi¢cant activation of Ca2+-mediated cellular responses, while minimizing the extent of any Ca2+dependent cell injury (Berridge et al 1998). As in the case of pulsatile hormone signalling, the Ca2+-mediated responses are also governed by the frequency of Ca2+ spikes. In most cell types indeed, an increase in the level of stimulation is associated with both a higher frequency of Ca2+ spiking and a larger cellular response (Fig. 5). For example, an increase in the frequency of local Ca2+ increases can induce vasodilatation of arteries (Porter et al 1998). Another well-known example of frequency coding of Ca2+ oscillations is that of gene expression. It has been shown indeed that in T lymphocytes, the various transcription factors are di¡erentially stimulated, depending on the frequency of Ca2+ oscillations. Also, the e⁄ciency of gene expression is increased
FIG. 5. Schematic representation of the relationship between the level of external stimulation, the frequency of cytosolic Ca2+ oscillations and the physiological response of the cell. Both the frequency of Ca2+ oscillations and the extent of the response increase with the level of stimulation. This phenomenon is often referred to as ‘frequency encoding of Ca2+ oscillations’. Redrawn after Berridge (1990).
FREQUENCY ENCODING OF PULSATILITY
29
by Ca2+ oscillations as compared to a constant elevation of the Ca2+ concentration at the same average value (Dolmetsch et al 1998, Li et al 1998). Frequency encoding of Ca2+ spikes can probably rely on a variety of molecular mechanisms. One of these involves the Ca2+-calmodulin activated protein kinase (CaM kinase), which acts as a widespread mediator between the Ca2+ spikes and the Ca2+-dependent physiological response (Meyer & Stryer 1991). Recent experiments have shown that CaM kinase II activity is sensitive to the temporal pattern of high frequency Ca2+ spikes (De Koninck & Schulman 1998). Such a capability of decoding the frequency of Ca2+ oscillations can be ascribed to the complex mode of regulation of CaM kinase II activity by Ca2+, in the form of autophosphorylation and CaM trapping (Hanson et al 1994). Second, more generally, a couple of converter enzymes consisting of a Ca2+activated protein kinase and an associated phosphatase can provide a mechanism for the frequency encoding of Ca2+ pulses (see scheme in Fig. 6). The encoding of Ca2+ oscillations then occurs in terms of the mean level of a phosphorylated target protein (Goldbeter et al 1990, Dupont & Goldbeter 1992). Given the same Ca2+activated kinase, the mean level of phosphorylated target protein will be governed by the rate of dephosphorylation by the phosphatase. Shown in Fig. 7 are three phosphorylation curves corresponding to the same kinase coupled to three di¡erent phosphatases characterized by di¡erent maximum rates. The results indicate that in such conditions the same frequency of Ca2+ oscillations can be transduced into three di¡erent levels of phosphorylated target proteins, which in turn induce various levels of activation of Ca2+-mediated responses involving the same kinases. This kind of e¡ect can also allow the pattern of Ca2+ oscillations (rates of rise and decline, half-width) to modulate the nature of the cellular response. Moreover, as shown previously (Dupont & Goldbeter 1992), the level of each of the three phosphorylated target proteins increases with the frequency of Ca2+ oscillations in the way schematized in Fig. 5. Thus, Ca2+ oscillations coupled to phosphorylation^dephosphorylation allow at the same time for frequency encoding as well as speci¢city in response.
FIG. 6. Protein phosphorylation driven by Ca2+ oscillations. Schematized is a protein W phosphorylated by a kinase into the form W*; the latter form is dephosphorylated by a phosphatase. If the kinase is activated by Ca2+, oscillations in cytosolic Ca2+ will be accompanied by a periodic increase in the level of phosphorylated substrate, which could mediate the cellular response to stimulation.
30
GOLDBETER ET AL
FIG. 7. Phosphorylation curves of three substrates phosphorylated by the same kinase but dephosphorylated by distinct phosphatases, during the course of Ca2+ oscillations. If the phosphatase is less active (a), the level of phosphorylated target protein will be larger. The same frequency of Ca2+ oscillations can thus be transduced into di¡erent levels of phosphorylated proteins. Results have been obtained by numerical integration of the equations presented in Dupont & Goldbeter (1992), with the same parameter values as in Fig. 8 of that paper, except for K1 ¼ K2 ¼ 1, and for VP ¼ 5 mM s1 , 1.5 mM s1 and 0.1 mM s1 for curves (c), (b) and (a), respectively.
In some cases where cells are coupled through gap junctions, Ca2+ increases can even be synchronized at the intercellular level. Such propagation of intercellular Ca2+ waves has been observed in airway epithelial cells (Sanderson et al 1988), astrocytes (Charles et al 1992), pancreatic acinar cells (Yule et al 1996) or hepatocytes (Nathanson et al 1995). In such cases, the Ca2+ waves e¡ectively coordinate the physiological response (in the form of beating or secretion, for example) and can thus induce a pulsatile behaviour at the level of a whole organ. Circadian rhythms Although they do not belong to the class of high-frequency pulsatile signals discussed above, circadian rhythms ¢nd, in a larger sense, their place in this chapter devoted to frequency encoding, for at least two reasons. First, these rhythms play a signi¢cant role in the adaptation of living organisms to their periodically changing environment; varying the period of endogenous circadian rhythmicity may deeply a¡ect such adaptation. Second, in many instances it
FREQUENCY ENCODING OF PULSATILITY
31
appears that these rhythms bear on the optimization of temporal patterns of drug delivery. Recent experiments on circadian rhythms in cyanobacteria have shown that wild-type and short- or long-period mutant populations will compete with each other in mixed suspensions growing in light^dark cycles (Ouyang et al 1998). The outcome of such competition experiments depends on the period of the imposed light^dark cycle. Generally, the (wild-type or mutant) species possessing the period closest to that of the imposed light^dark cycle will eventually win over and eliminate the other. These results, which can be accounted for by a theoretical model based on the release of a growth inhibitor during the light phase (Roussel et al 2000), demonstrate the importance of the frequency of circadian rhythms in the optimal adaptation of the organism toward its periodically varying environment. A similar role of frequency has been demonstrated in several instances for the optimization of drug delivery patterns. Periodic drug delivery can be ¢nely tuned to take into account circadian rhythms within the organism which result in circadian variations of drug tolerance or e⁄cacy (Lemmer 1989). A case in point, discussed elsewhere in this volume (Le¤ vi 2000, this volume), is that of cancer chronopharmacology (Le¤ vi et al 1999). Beyond the case of circadian rhythms, another aspect of chronopharmacology pertains to the use of chronomodulated treatments by hormones such as GnRH or GH in endocrinology. Conclusions Frequency encoding represents a most important function of many key biological rhythms, particularly those that provide support for pulsatile signaling in intercellular communication. Besides neuronal oscillations, not considered in this chapter, examples of frequency encoding of pulsatile signals range from cAMP signalling in the cellular slime mould D. discoideum to the pulsatile secretion of a large number of hormones among which the most prominent are GnRH, LH and FSH, as well as GH and insulin. In the majority of these cases the frequency of pulsatile release governs the physiological e⁄ciency of the signal. We have brie£y recalled in this chapter several molecular mechanisms on which frequency encoding can be based. Thus, the ubiquitous phenomenon of receptor desensitization allows the adaptation of cellular responses to constant stimuli and at the same time provides a robust mechanism for the frequency encoding of pulsatile signals. Theoretical models based on receptor desensitization show how the latter process allows for the frequency encoding of cAMP pulses in Dictyostelium cells aggregating after starvation and, more generally, for the frequency encoding of pulsatile hormonal signals. In the slime mould system, the model (Fig. 1) indicates the existence of an optimal pattern of pulsatile stimulation that
32
GOLDBETER ET AL
maximizes the capability of target cells to synthesize cAMP in response to a cAMP pulse. This pattern corresponds to an optimal duration of the pulse and an optimal interval between successive pulses. The model indicates that these optimal values are unique for a given set of conditions but change when increasing or decreasing the rate of receptor desensitization or resensitization (Table 1). The more general two-state receptor model considered for the response to pulsatile signalling (Fig. 3) also showed the existence of an optimal pattern of pulsatile stimulation. In addition to the dependence of this pattern on the kinetics of receptor desensitization and resensitization, the model indicates how the optimal pattern changes with the level of ligand at the peak of stimulation. Thus, the top of the mound that de¢nes the optimal pattern at saturating ligand levels (Fig. 4, top panel) moves to another location in the pulse duration-interval plane when the maximum ligand level decreases down to mid-saturation (middle panel) or much below saturation (bottom panel). When applied to the case of GnRH signalling (Li & Goldbeter 1989), this model yields agreement with the receptor desensitization^resensitization kinetics associated with the optimal pattern of one 5 min pulse delivered every hour, which corresponds to the physiological GnRH signal released by the hypothalamus in rhesus monkey and humans. The desensitized receptor model indicates that the notion of an optimal frequency is by no means all-or-none. As shown in Fig. 5, a graded responsiveness is observed as a function of pulse duration and interval: 90% of the maximum responsiveness (which may already su⁄ce in physiological conditions) can be obtained in a sizeable region around the pair of values (t0 ,t1 ) corresponding to the optimal pattern. This region increases as the required percentage of maximum responsiveness progressively diminishes. Another mechanism allowing the frequency encoding of pulsatile signals is based on phosphorylation^dephosphorylation. Thus, a Ca2+-activated kinase can transform Ca2+ oscillations of increasing frequency into an increase in the mean level of a phosphorylated protein. Moreover, as shown in Fig. 7, such a kinase can transform Ca2+ oscillations of a given frequency into distinct levels of phosphorylated target proteins when coupled to phosphatases acting with di¡erent rates on their respective phosphorylated substrates. Mechanisms allowing the frequency encoding of pulsatile signals may involve other processes mediating the adaptation of target cells to constant stimuli, such as the inactivation of receptor-coupled ion channels, or reversible receptor internalizationas may be the case for insulin (Goodner et al 1988) and GH (Bick et al 1989)rather than receptor desensitization through conformational change with or without covalent modi¢cation. The comparative study of cAMP signalling in Dictyostelium cells, pulsatile hormone secretion and intracellular Ca2+ spiking shows that common concepts
FREQUENCY ENCODING OF PULSATILITY
33
emerge as to the frequency encoding of pulsatile signals (Goldbeter 1988, 1996). In mechanisms involving receptor desensitization or internalization, the reason why pulsatile signals often prove superior to non-pulsatile ones can be related to the fact that a constant stimulus can only trigger a maximal response once; thereafter the response declines owing to desensitization. A pulsatile signal allows the receptor to produce maximal responses in a repetitive manner, because the receptor can resensitize between successive pulses provided that it has su⁄cient time to do so. The existence of an optimal frequency results from the interplay between the antagonistic constraints of producing as many responses as possible in a given amount of timewhich requires decreasing the time interval between pulses and preserving the capability of producing large-amplitude responses, which requires increasing this interval (Li & Goldbeter 1989). Di¡erent types of physiological response may correspond to di¡erent optimal frequencies of a given pulsatile signal in the same or in di¡erent types of target cells. The role of the frequency of the physiological rhythm extends beyond the case of pulsatile signals. Thus, intracellular Ca2+ oscillations occur in many cell types in response to extracellular signals which need not be themselves of a pulsatile nature. As shown above, these signals can be transduced via Ca2+ spikes into frequency-dependent intracellular responses. Finally, the period of circadian rhythms also appears to confer adaptation to the periodically changing environment, as recently demonstrated for cyanobacteria, while in humans such rhythms govern the responsiveness to certain types of medication according to the temporal pattern of their administration. In Dictyostelium cells which communicate by means of cAMP pulses during the transition from the unicellular to the multicellular stage of the life cycle, an optimal frequency of these signals exists that maximizes the capability of cells to relay cAMP pulses (Li & Goldbeter 1990). Beyond temporal responses, the concept of optimal signalling patterns extends to spatiotemporal aspects of cellular behaviour. Thus, as shown in Fig. 2B, the analysis of a model for cAMP signalling indicates (Lauzeral et al 1997) that an optimal desynchronization of the cells with regard to developmental changes after starvation can maximize the formation of largeamplitude spiral waves of cAMP in the early stages of aggregation. Acknowledgments This work was supported by grant 3.4607.99 from the Fonds de la Recherche Scienti¢que Me¤ dicale (FRSM, Belgium), and by the Programme ‘Actions de Recherche Concerte¤ e’ (convention 94/99^180) launched by the Division of Science and Higher Education, French Community of Belgium. G.D. is Chercheur quali¢e¤ du FNRS.
References Berridge MJ 1990 Calcium oscillations. J Biol Chem 265:9583^9586
34
GOLDBETER ET AL
Berridge MJ, Bootman M, Lipp P 1998 Calciuma life and death signal. Nature 395:645^648 Bick T, Youdim MBH, Hochberg Z 1989 Adaptation of liver membrane somatogenic and lactogenic growth hormone (GH) binding to the spontaneous pulsation of GH secretion in the male rat. Endocrinology 125:1711^1717 Buzsaki G, Llinas R, Singer W, Berthoz A, Christen Y (eds) 1994 Temporal coding in the brain. Springer-Verlag, Berlin Charles A, Naus C, Zhu D, Kidder G, Dirksen E, Sanderson M 1992 Intercellular calcium signaling via gap junctions in glioma cells. J Cell Biol 118:195^201 Clark RG, Jansson JO, Isaksson O, Robinson ICAF 1985 Intravenous growth hormone: growth responses to patterned infusions. J Endocrinol 104:53^61 Darmon M, Brachet P, Da Silva LH 1975 Chemotactic signals induce cell di¡erentiation in Dictyostelium discoideum. Proc Natl Acad Sci USA 72:3163^3166 De Koninck P, Schulman H 1998 Sensitivity of CaM kinase II to the frequency of Ca2+ oscillations. Science 279:227^230 Devreotes PN 1989 Dictyostelium discoideum: a model system for cell^cell interactions in development. Science 245:1054^1058 Devreotes PN, Sherring JA 1985 Kinetics and concentration dependence of reversible cAMPinducedmodi¢cationofthesurfacecAMPreceptorinDictyostelium.JBiolChem260:6378^6384 Dolmetsch R, Xu K, Lewis R 1998 Calcium oscillations increase the e⁄ciency and speci¢city of gene expression. Nature 392:933^936 Dupont G, Goldbeter A 1992 Protein phosphorylation driven by intracellular calcium oscillations: a kinetic analysis. Biophys Chem 42:257^270 (erratum: 1995 Biophys Chem 54:291) Gerisch G 1987 Cyclic AMP and other signals controlling cell development and di¡erentiation in Dictyostelium. Annu Rev Biochem 56:853^879 Gerisch G, Wick U 1975 Intracellular oscillations and release of cyclic AMP from Dictyostelium cells. Biochem Biophys Res Commun 65:364^370 Gerisch G, Fromm H, Huesgen A, Wick U 1975 Control of cell contact sites by cAMP pulses in di¡erentiating Dictyostelium cells. Nature 255:547^549 Goldbeter A 1988 Periodic signaling as an optimal mode of intercellular communication. News Physiol Sci 3:103^105 Goldbeter A 1996 Biochemical oscillations and cellular rhythms: the molecular bases of periodic and chaotic behaviour. Cambridge University Press, Cambridge Goldbeter A, Dupont G, Berridge MJ 1990 Minimal model for signal-induced Ca2+ oscillations and for their frequency encoding through protein phosphorylation. Proc Natl Acad Sci USA 87:1461^1465 Goodner CJ, Walike BC, Koerker DJ et al 1977 Insulin, glucagon, and glucose exhibit synchronous, sustained oscillations in fasting monkeys. Science 195:177^179 Goodner CJ, Sweet IR, Harrison HC Jr 1988 Rapid reduction and return of surface insulin receptors after exposure to brief pulses of insulin in perifused rat hepatocytes. Diabetes 37:1316^1323 Halloy J, Lauzeral J, Goldbeter A 1998 Modeling oscillations and waves of cAMP in Dictyostelium discoideum cells. Biophys Chem 72:9^19 Hanson P, Meyer T, Stryer L, Schulman H 1994 Dual role of calmodulin in autophosphorylation of multifunctional CaM kinase may underlie decoding of calcium signals. Neuron 12:943^956 Heinze K, Keener RW, Midgley AR Jr 1998 A mathematical model of luteinizing hormone release from ovine pituitary cells in perifusion. Am J Physiol 275:E1061^E1071 Hindmarsh PC, Stanhope R, Preece MA, Brook CGD 1990 Frequency of administration of growth hormonean important factor in determining growth response to exogenous growth hormone. Horm Res 33 (suppl 4):83^89
FREQUENCY ENCODING OF PULSATILITY
35
Klein P, Theibert A, Fontana D, Devreotes PN 1985 Identi¢cation and cyclic AMP-induced modi¢cation of the cyclic AMP receptor in Dictyosteliumdiscoideum. J Biol Chem 260:1757^1764 Knobil E 1980 The neuroendocrine control of the menstrual cycle. Recent Prog Horm Res 36:53^88 Knobil E 1981 Patterns of hormone signals and hormone action. N Engl J Med 305:1582^1583 Konijn TM 1972 Cyclic AMP as ¢rst messenger. Adv Cyclic Nucleotide Res 1:17^31 Lauzeral J, Halloy J, Goldbeter A 1997 Desynchronization of cells on the developmental path triggers the formation of spiral waves of cAMP during Dictyostelium aggregation. Proc Natl Acad Sci USA 94:9153^9158 Lemmer B (ed) 1989 Chronopharmacology: cellular and biochemical interactions. Marcel Dekker, New York Le¤ vi F 2000 Therapeutic implications of circadian rhythms in cancer patients. In: Mechanisms and biological signi¢cance of pulsatile hormone secretion. Wiley, Chichester (Novartis Found Symp 227) p 119^142 Le¤ vi F, Zidani R, Brienza S et al 1999 A multicenter evaluation of intensi¢ed, ambulatory, chronomodulated chemotherapy with oxaliplatin, 5-£uorouracil, and leucovorin as initial treatment of patients with metastatic colorectal carcinoma. International Organization for Cancer Chronotherapy. Cancer 85:2532^2540 Leyendecker GL, Wildt L, Hansmann M 1980 Pregnancies following intermittent (pulsatile) administration of GnRH by means of a portable pump (‘Zyklomat’) a new approach to the treatment of infertility in hypothalamic amenorrhea. J Clin Endocr Metab 51:1214^1216 Li W, Llopis J, Whitney M, Zlokarnik G, Tsien RY 1998 Cell-permeant caged InsP3 ester shows that Ca2+ spike frequency can optimize gene expression. Nature 392:936^941 Li YX, Goldbeter A 1989 Frequency speci¢city in intercellular communication. The in£uence of patterns of periodic signaling on target cell responsiveness. Biophys J 55:125^145 Li YX, Goldbeter A 1990 Frequency encoding of pulsatile signals of cAMP based on receptor desensitization in Dictyostelium cells. J Theor Biol 146:355^367 Li YX, Goldbeter A 1992 Pulsatile signaling in intercellular communication. Periodic stimuli are more e⁄cient than random or chaotic signals in a model based on receptor desensitization. Biophys J 61:161^171 Lincoln DW, Fraser HM, Lincoln GA, Martin GB, McNeilly AS 1985 Hypothalamic pulse generators. Recent Prog Horm Res 41:369^419 Martiel JL, Goldbeter A 1987 A model based on receptor desensitization for cyclic AMP signaling in Dictyostelium cells. Biophys J 52:807^828 Matthews DR, Naylor BA, Jones RG, Ward CM, Turner RC 1983 Pulsatile insulin has greater hypoglycemic e¡ect than continuous delivery. Diabetes 32:617^621 Meyer T, Stryer L 1991 Calcium spiking. Annu Rev Biophys Biophys Chem 20:153^174 Nathanson M, Burgstahler A, Mennone A, Fallon M, Gonzalez C, Saez J 1995 Ca2+ waves are organized among hepatocytes in the intact organ. Am J Physiol 269:G167^G171 Ouyang Y, Andersson CR, Kondo T, Golden SS, Johnson CH 1998 Resonating circadian clocks enhance ¢tness in cyanobacteria. Proc Natl Acad Sci USA 95:8660^8664 Pohl CR, Richardson SW, Hutchison JS, Germak JA, Knobil E 1983 Hypophysiotropic signal frequency and the functioning of the pituitary^ovarian system in the rhesus monkey. Endocrinology 112:2076^2080 Porter VA, Bonev AD, Knot HJ et al 1998 Frequency modulation of Ca2+ sparks is involved in regulation of arterial diameter by cyclic nucleotides. Am J Physiol 274:C1346^C1355 Roos W, Nanjundiah V, Malchow D, Gerisch G 1975 Ampli¢cation of cyclic AMP signals in aggregating cells of Dictyostelium discoideum. FEBS Lett 53:139^142 Roussel M, Gonze D, Goldbeter A 2000 Modeling di¡erential ¢tness of cyanobacterial strains whose circadian oscillators have di¡erent free-running periods: comparing mutual inhibition and substrate-depletion hypotheses, in preparation
36
DISCUSSION
Sanderson M, Chow I, Dirksen E 1988 Intercellular communication between ciliated cells in culture. Am J Physiol 254:C63^C74 S˛derbom F, Loomis WF 1998 Cell^cell signaling during Dictyostelium development. Trends Microbiol 6:402^406 Tannenbaum GS, Martin JB 1976 Evidence for an endogenous ultradian rhythm governing growth hormone secretion in the rat. Endocrinology 98:562^570 Van Cauter E, Ascho¡ J 1989 Endocrine and other biological rhythms. In: DeGroot LJ (ed) Endocrinology, vol 3. WB Saunders, Philadelphia, PA, p 2658^2705 Waxman DJ 2000 Growth hormone pulse-activated STAT5 signalling: a unique regulatory mechanism governing sexual dimorphism of liver gene expression. In: Mechanisms and biological signi¢cance of pulsatile hormone secretion. Wiley, Chichester (Novartis Found Symp 227) p 61^81 Waxman DJ, Pampori NA, Ram PA, Agrawal AK, Shapiro BH 1991 Interpulse interval in circulating growth hormone patterns regulates sexually dimorphic expression of hepatic cytochrome P450. Proc Natl Acad Sci USA 88:6868^6872 Wurster B 1982 On induction of cell di¡erentiation by cyclic AMP pulses in Dictyostelium discoideum. Biophys Struct Mechan 9:137^143 Yule D, Stuenkel E, Williams J 1996 Intercellular calcium waves in rat pancreatic acini: mechanism of transmission. Am J Physiol 271:C1285^C1294
DISCUSSION Veldhuis: Your talk highlights the tremendous challenge of this ¢eld: bridging the gap from rhythms at the level of Ca2+-calmodulin kinase up to the level of seasonal changes. I would like to begin the discussion with an area I had trouble visualizing intuitively, namely how the spiral waves are generated. Is it possible to sketch a bit more intuitively how you would get from concentric to spiral waves? Goldbeter: We have generated a computer simulation of the dynamic behaviour of Dictyostelium cells after starvation. The simulation is based on the model schematized in Fig. 1 and incorporates the fact that over time cells undergo certain biochemical changes. These changes pertain to the activity of adenylate cyclase, the enzyme that makes cAMP, and phosphodiesterase, the enzyme that destroys cAMP. Both enzyme activities increase in the hours that follow starvation. In the parameter plane where adenylate cyclase is plotted versus phosphodiesterase, cells follow a developmental path that corresponds to a continuous increase in the activity of the two enzymes. This path brings the cells, successively, across a domain of no relay, then a domain of relay in which cells amplify a suprathreshold pulse of extracellular cAMP in a pulsatory manner, and later into a domain of autonomous oscillations of cAMP, before the amoebae leave the oscillatory domain and become excitable again. A key factor in the origin of cAMP waves is that the cells are not fully synchronized when following this developmental path (Lauzeral et al 1997). Desynchronization here does not pertain to any lack of synchrony in pulsatile cAMP signalling, but rather it refers to the fact that cells do not possess the same amounts of enzymes or cAMP receptor
FREQUENCY ENCODING OF PULSATILITY
37
at the onset of starvation, and thus start moving from di¡erent initial conditions on their developmental path. The origin of such heterogeneities is that when the cells starve, they are blocked at di¡erent stages of the cell cycle. The simulation shows that concentric waves ¢rst appear, due to the oscillations generated by cells behaving as aggregation centres (these cells are the ¢rst to enter the oscillatory domain in parameter space). However, because the cells are not synchronized and are thus characterized by di¡erent enzyme activities at a given time, the waves are blurred because of local variations in frequency of oscillations or in refractoriness. The defects which occur serve as nucleation sites for the development of spiral waves of cAMP. When cells are not highly desynchronized, only concentric waves are observed. As the degree of desynchronization increases, spiral waves form. The stronger the desynchronization, the smaller the spirals and the larger their number (see Fig. 2B). Once formed the spirals are extremely stable, as they remain sustained as long as cells are excitable, even in the absence of centres generating autonomously pulsatile signals of cAMP. The model suggests that if one were to synchronize the cells, one should observe concentric waves instead of spirals. Beyond the case of Dictyostelium, understanding the origin of spiral waves is important because the phenomenon plays a signi¢cant role in other physiological systems. Thus, as shown by work from the groups of Winfree and Jalife, an important example of spiral wave formation is provided by ¢brillation in the heart (Davidenko et al 1992, Winfree 1994). Pincus: There is an excellent reference on this general area of three-dimensional waves: Arthur Winfree (University of Arizona) has written a book entitled When time breaks down (Winfree 1987). Waxman: You mentioned the frequency encoding of the response. It is clear that if the frequency becomes too rapid and there’s insu⁄cient time in the receptor system for resensitization, the result is a reduced e¡ect and ultimately no e¡ect. What happens if you go in the other direction, where the frequency becomes distant, and there is a non-physiological long time between pulses? Would you expect to have a loss of responsiveness? Goldbeter: When the interval between pulses is very large, a maximum response is generated by each pulse. However, a cell may need to generate a certain number of responses in a given time. For example, if Dictyostelium cells were stimulated every three hours instead of every ¢ve minutes, obviously the interval between two pulses would be so large that each stimulus would elicit a maximum intracellular cAMP response. But in between two stimuli the level of cAMP would probably remain too low for triggering the physiological response produced by a rise in intracellular cAMP. Hence the probable existence of an optimum interval: if the interval between two pulses is too short, then desensitization ensues; if the interval is too long, the mean intracellular response over time may not be su⁄cient to
38
DISCUSSION
establish the desired level of intracellular messenger needed, for example, for gene expression linked to an increased mean level of cAMP. Waxman: That’s interesting, because in the GH system, for example, Agneta Mode demonstrated in the early 1980s that twice per day GH treatment given to a hypophysectomized rat, which is much less frequent than the physiological frequency of six or seven times per day, was su⁄cient to achieve a biological response. In later studies that we carried out, we showed that as we increased the frequency everything was ¢ne until we exceeded even only slightly the natural GH pulse frequency, in which case the response was totally lost. In that type of model, which is a more complex whole-animal model than Dictyostelium, when you go to more frequent stimulation you lose responsiveness. However, anything that’s much less frequent is, surprisingly, fully active. Goldbeter: Perhaps in that system a reduced number of pulses already su⁄ces to establish the required intracellular response. In Dictyostelium, experiments have pointed to a minimum time interval between pulses, of the order of 2 min, but less information is available as to the e¡ect of intervals longer than the physiological one, which is of the order of 5 min. Brown: In your theoretical studies, where you are relating desensitization to experimental ¢ndings, it is interesting that you used a linear function of the four di¡erent receptor types as a measure of activity. Have you considered any non-linear functions? In particular, when we were modelling GH release, we found that you could get a better predictor of release by using the rate of transfer from one state to another, rather than the actual numbers in each state. Goldbeter: We have not considered non-linear functions, nor have we considered the interesting aspect of the rate of transition between the receptor states. The reason why we focused on the linear combination of receptor states is that we made the simplest assumption that it is the amounts of complexes between the ligand and the receptor in its various states that brings about the physiological response. It would be interesting to compare the predictions of this model with your alternative approach. Clarke: You mentioned with respect to the data of Knobil that the frequency would determine the amount of hormone produced. However, we have used a similar animal model with a similar sort of paradigm, and as you change the frequency of input but maintain a constant amount of stimulus, the response is much greater at the lower frequency, and the amount of hormone that is released at the lower frequency is exactly the same as the amount that is released at the higher frequency, if you take into account the number of pulses. In actual fact, I’m not sure that we should be saying that the frequency of stimulus will determine the absolute amount of output because of this frequency^amplitude relationship.
FREQUENCY ENCODING OF PULSATILITY
39
Goldbeter: In the simulations we performed, because the amplitude of each pulse remained constant, the mean level of stimulation over a period changed when the frequency of stimulation varied. The clearance rate is another factor that might also be important for accounting for the di¡erences that you are seeing. Clarke: It’s not a function of clearance of rate. If you look speci¢cally at the GnRH/LH system, over quite a wide range of frequencies, the frequency^ amplitude relationship is correlated 0.99. It is also correlated to 0.99 with the amount of releasable hormone in the cell. I also wanted to suggest that we shouldn’t think about ‘optimal responses’ in at least some systems, because with the GnRH/LH system there are profound alterations across the oestrus cycle in the frequency and the amplitude. These alterations have physiological meaning in the sense that they’re more frequent or less frequent depending on the stage of the cycle and the steroidal background. This doesn’t mean to say that the higher amplitude is receiving optimal stimulation. Goldbeter: I agree. I was using the term ‘optimal frequency’ in relation to the experiments on cAMP signalling in Dictyostelium and on the induction of appropriate levels of LH and FSH by circhoral GnRH signals. For the latter phenomenon, the fact that the frequency is a key factor has clinical implications. Obviously, to elicit a physiologically e¡ective response, there is not a unique set of values for pulse generation and pulse interval, but a range. In the picture I showed (see Fig. 4) there was a kind of kind of hill, and you can draw a circumference for 90% of maximum response, 80%, and so on. It is likely that to a certain extent such submaximal responses will also work in physiological terms. Moreover, as you mention it, the optimal pattern of pulsatile stimulation may well change in the course of time, owing to developmental changes in the target system. Finally, as shown in Fig. 4, the characteristics of the optimal pattern depend on the amplitude of stimulation which may also vary in the course of time. Robinson: It’s a little dangerous to imagine that there is only one optimal pattern in such a complicated pro¢le. I noticed in some of your simulations with phosphorylase and Ca2+ that although you have an increased sensitivity to Ca2+, you also have a broader slope. If one of the features of the system is to sharpen the response relationship to a change in Ca2+, a steeper relationship between this and the output might actually be the optimal response for some of the targets. Also, commenting on what David Waxman was saying when he was talking about increasing the growth hormone frequency, as I shall show later, it’s not that you lose responsiveness, but that you generate another set of responses. I would like to suggest that we should not be thinking about an ‘optimal’ response, but ways of encoding more than one response for the same receptor transduction system. Goldbeter: I fully agree with your remarks. Frequency encoding de¢nitely has wider implications than the mere existence of an optimal frequency. When the
40
DISCUSSION
response depends on the frequency of the stimulus in a graded manner, di¡erent physiological e¡ects can be elicited using a range of di¡erent frequencies, for a given type of cell or in di¡erent target cells. Robinson: What about clustering? You talked about randomizing your patterns, but quite often you see clustering of di¡erent frequencies. In Dictyostelium, for instance, can you can trigger it with certain clusters of optimal frequency and then have quite long periods without any signal? In other words, is there a memory in terms of a clustering of the optimal signals? Goldbeter: There is no evidence for such clustering in Dictyostelium. What has been tried experimentally, and also studied by numerical simulations, is a variation of a few minutes above and below the physiological period, as well as random variation of the interval between successive pulses. Robinson: The reason I raised this point is because a lot of hormone bursts are actually multicomponent bursts. Leng: Your paper linked phenomena that operate on quite di¡erent time-scales. You talked about the importance of receptor desensitization mechanisms in establishing the biological e¡ectiveness of pulsatile signalling, and you also talked about Ca2+ oscillations and the capacity for encoding information there. Linking these two is di⁄cult, and it is not clear what the consequences are. I noticed that the example you showed of encoding the information by Ca2+ spikes was in a receptor system which actually doesn’t show desensitizationthe response to vasopressin. If you have a system which shows strong desensitization, I wonder whether you can encode information reliably by Ca2+ oscillations. Goldbeter: This should depend on the relative time-scales of the phenomena. With Ca2+ oscillations one is dealing with seconds or minutes. The e¡ect of the oscillations could well be encoded before receptor desensitization sets in. As mentioned by Paolo Sassone-Corsi in his paper (Foulkes et al 2000, this volume), if a transcription factor is phosphorylated, then a signal can be switched on inside the cell even in the presence of subsequent desensitization of the hormone receptor. Sassone-Corsi: I was wondering about the phosphorylation^dephosphorylation by protein kinase A (PKA) in Dictyostelium. Would you also invoke the inducibility of phosphatases? It is so quick you would expect not only phosphorylation by PKA, but also that there must be a system where phosphatases need to be induced. Goldbeter: Di¡erent sets of kinases and phosphatases are at work in Dictyostelium: PKA is active intracellularly, and in addition there is a receptor kinase closely related to the b-adrenergic receptor kinase. Less is known about associated receptor phosphatases. The kinetics of receptor phopshorylation and dephosphorylation has been studied by Devreotes and co-workers (Devreotes & Sherring 1985). Sassone-Corsi: Is that phosphatase inducible?
FREQUENCY ENCODING OF PULSATILITY
41
Goldbeter: I don’t know about phosphatase inducibility. The activity of a number of enzymes in Dictyostelium does increase in the course of development after starvation. Many of the components of the machinery for aggregation which incorporates not only adenylate cyclase, phosphodiesterase and the cAMP receptor, but probably also the receptor kinase and phosphatase, are progressively expressed during the 6 h that lead to aggregation. Kjems: You showed beautifully how changing the frequency within your pulsatile patterns will a¡ect the subsequent steps of the cascade. One point of major interest is that in some pathophysiological states it is not the frequency that is changed, but the amplitude. This is something you should focus on. Goldbeter: Any pulsatile signal is indeed characterized by its amplitude and by its frequency; both aspects are clearly important for the control of normal or pathological responses. When Knobil wrote his editorial in the NEJM (Knobil 1981), he was focusing on the frequency because this was a concept vastly underestimated at that time. The importance of amplitude should certainly be emphasized as well. As shown by the model discussed in my presentation (see Fig. 4), the e¡ects of the amplitude and of the frequency of a pulsatile stimulus are closely intertwined. Kjems: You mentioned the intracellular Ca2+ spikes. For example, you could keep the frequency constant, and change the amplitude dramatically. This must eventually have an e¡ect on the whole signalling pattern. What is it, for example, in the pathophysiological state that will lead to that reduction? Goldbeter: In some systems, such as with Ca2+ signalling, it is generally the frequency rather than the amplitude of the oscillations that changes with the agonist concentration. Veldhuis: In the GnRH system, the product of the frequency and amplitude is more or less a constant. It is proportionate to the output in any given system. In puberty we ¢nd a 30-fold increase in amplitude and a 30% increase in frequency (Wu et al 1996). This system has an explosive range of amplitude drive. It has surprised me that so many other hormones have a strong amplitude component. Eventually, a marriage of concepts will be required to involve joint amplitude and frequency control. Marshall: Along those lines, perhaps in some systems, such as the GnRH system, frequency is a means of sending the same signal and getting two di¡erent responses, in this case LH and FSH. In a more general form, an input signal to a hormonereleasing cell has two functions: one is hormone release, which we commonly look at as a product, but the other is an input signal, to either produce more hormone or replenish hormone pools, i.e. to have an e¡ect on transcription machinery. You are looking at behaviour in Dictyostelium as the output signal. Is there a transcriptional response that you can parallel to that? Do they follow parallel pathways or are they divergent?
42
DISCUSSION
Goldbeter: There are indeed di¡erent levels of response to cAMP signals in Dictyostelium. One e¡ect of the signals, which occurs on a longer time-scale, is to act on gene transcription; another, more rapid e¡ect is to elicit the response of the relay machinery which governs cell^cell signalling. We focused on the intercellular communication system, which responds over a time-scale of minutes. This response involves the binding of cAMP to its receptor and the subsequent activation of adenylate cyclase and release of cAMP. Besides participating in intercellular communication, the cAMP signal also behaves as a second messenger: it carries information through to gene expression. Thus there are two di¡erent responses triggered by the pulsatile cAMP signal, a rapid one for the communication system, and a slower one that has to do with gene expression and di¡erentiation during the hours that follow starvation. Marshall: But the latter may be important in tomorrow’s response. Goldbeter: The two types of responses are indeed linked. This is the basis for the observed changes in dynamic properties of the cAMP signalling system which, in the hours after starvation, starting from a state of no response to cAMP pulses, becomes successively capable of relay (i.e. amplifying suprathreshold cAMP pulses) and autonomous generation of periodic cAMP pulses. We considered the coupling between the two types of response when we looked at the origin of spiral waves of cAMP associated with the progression of cells along their developmental path (see above). Movement along this path is brought about by gene transcription, and corresponds to an increase in adenylate cyclase and phosphodiesterase; such an increase in enzyme activities is itself under the control of cAMP pulses. Robinson: Can I pursue the question about the pathophysiological changes you might see. One of the things that I’m curious about, is what happens in a situation where you model an increase in continuous exposure with pulses superimposed upon it? If you have a continuous exposure which invokes a desensitization, is the increment in activation that you then get with your pulses on top of that, much less? In other words, are you diminishing your ability to evoke activation by a dominant baseline desensitization? This may be relevant for just how e¡ective high amplitude secretion from tumours is, if in fact a large proportion of that is continuous and therefore not particularly e¡ective. Goldbeter: In the presence of receptor desensitization, a continuous exposure to a baseline level of stimulation will result in reducing the steady-state fraction of receptor in active state. The larger the baseline level of ligand, the smaller this fraction will be. As long as a su⁄cient amount of receptor remains in the active state, even if it is 5%, ligand pulses should be capable of eliciting responses, even though the amplitude of the latter will generally be smaller than in the absence of baseline stimulation. If baseline stimulation between pulses is so large that it leads
FREQUENCY ENCODING OF PULSATILITY
43
to full receptor desensitization on target cells, pulses of ligand superimposed on the baseline signal should remain une¡ective. Wu: I was going to expand on John Marshall’s point. The importance of frequency and amplitude has to be taken in the context of the target cell response, even within the same system, such as the GnRH/LH system. The testis is actually sensitive to a pulsatile signal, even though the LH drive is cAMP dependent. What do you think is the di¡erence between a pulsatile signal-sensitive and an insensitive system using the same second messenger? Goldbeter: If you take a receptor system which does not undergo desensitization, then the di¡erence between continuous and pulsatile signalling would be much less. Such a receptor will also generate a pulsatile response if you stimulate it in a pulsatile manner, but then there’s no striking advantage in using a pulsatile signal versus a continuous one. When there is desensitization of the target cell, a pulsatile signal can prove more e⁄cient that a continuous one, since the latter will necessarily lead to a decreasing response while the former allows for repetitive large-amplitude responses. Desensitization can of course take many aspectsit is not necessarily associated with receptor phosphorylation, but can also be based on inactivation or deactivation of an ion channel coupled to a receptor, or receptor internalization. Matthews: If you have a whole range of cells which are di¡erentially sensitive, then a pulsatile signal will move across all of the domains of sensitivity. You can envisage that that would be an optimal way of triggering many di¡erent domains of signals. As you say some will work at 5%, some will work at 95%, but you can trigger them all. And you can imagine that if you have a domain that simply runs between 50% and 100%, cells below that domain will not be recruited. So you can imagine that a pulsatile system would work extremely well within systems which might have quite a wide ranging sensitivity domain, whereas cellular systems which have a packed domain would tend to show more of an on/o¡ phenomenon. De Meyts: We should not forget when we try to extrapolate from the concept of pulsatile secretion and the way the target cell responds to it, that in many hormone systems, such as GH, insulin-like growth factor and steroids, there are binding proteins which interfere with the direct relationship. We don’t yet fully understand how these binding proteins may dampen the e¡ect of pulsatility. Lightman: We have talked a lot about the frequency of the pulses and the height of the pulses, but what we don’t seem to have talked about very much is how low the nadir levels go. Presumably, this can have a major e¡ect on responsiveness. Have you done studies investigating how low it drops? Goldbeter: We have taken that into account in some of our studies with the twostate receptor model subjected to a square wave stimulus. Changing the level of the minimum between pulses can have a tremendous e¡ect indeed. As indicated above,
44
DISCUSSION
if this minimum is large, it will induce a sizeable degree of desensitization and will thereby reduce the amplitude of the response. In most of our studies, however, we removed the stimulus altogether between pulses, or kept it well below the receptor Kd. Lightman: In most biological systems it won’t fall to that level. Goldbeter: I would think that 10-fold changes can sometimes be observed. For example, for cytosolic Ca2+ oscillations (which admittedly may di¡er from pulsatile hormone variations) there is a change roughly from 0.1^1 mM. Matthews: Nuclear receptors don’t seem to down-regulate in quite the same way as cytosolic and membrane receptors. Have you views about whether pulsatile systems work better on membrane receptors rather than nuclear receptors? Insulin receptors are membrane receptors and one can demonstrate up and downregulation of those sensitivities. Whereas with nuclear receptors, can you demonstrate that to the same extent? Sassone-Corsi: Nuclear receptors are based on a completely di¡erent system. These are transcription factors which upon binding of a ligand get activated, initiating a whole cascade of transcriptional events. They contain a DNA binding domain that will determine where they are going to bind DNA. Matthews: But have you views whether they up and down-regulate in the way that membrane receptors do? Sassone-Corsi: They are up-regulated by the ligand. Herbison: We need to remember that there’s a lot of shuttling of nuclear receptors. They are not just sitting there. Sassone-Corsi: This is indeed the case for some, such as the glucocorticoid receptor. However, the majority of nuclear receptors are thought to be in the nucleus for most of the time. De Meyts: Then it may be the kinase that phosphorylates them that shuttles, like MAP kinase. Herbison: Certainly, over the long-term, there’s very good evidence for regulation of oestrogen/progesterone receptors. Sassone-Corsi: In the case of nuclear receptors, phosphorylation has not been strongly linked to activity. Only recently have there been indications of phosphorylation modulating the function of nuclear receptors. It is the ligand that changes the conformation of the ligand-binding domain, which then allows activation by another set of transcription factors. Marshall: Following on from the issue of whether you can make one messenger do several jobs, in the example you used of the pinealocytes there are many genes which have cAMP response elements. What determines which gene responds? When you generate that cascade of transcription factors something is determining that the gene that is activated by that response element is gene X and not genes Y or Z?
FREQUENCY ENCODING OF PULSATILITY
45
Sassone-Corsi: If you are talking about cAMP response elements, the sequence TGACGTCA, a perfect palindrome, is the consensus site. However, as I showed in the case of NAT, the second T is a C (TGACGCCA). This changes the a⁄nity of binding. Just to give you an idea of the complexity of the system, there are several isoforms of the transcription factors CREB, CREM and ATF. Each one of these genes can make something like 10 proteins by di¡erential splicing and alternative usage of introns, dramatically increasing the combinatorial possibilities of interaction, as these proteins dimerize. In addition, the CRE site can also bind members of the Fos/Jun family. References Davidenko JM, Pertsov AV, Salomonsz R, Baxter W, Jalife J 1992 Stationary and drifting spiral waves of excitation in isolated cardiac muscle. Nature 355:349^351 Devreotes PN, Sherring JA 1985 Kinetics and concentration dependence of reversible cAMPinducedmodi¢cationofthesurfacecAMPreceptorinDictyostelium. JBiolChem260:6378^6384 Foulkes NS, Cermakian N, Whitmore D, Sassone-Corsi P 2000 Rhythmic transcription: the molecular basis of oscillatory melatonin synthesis. In: Mechanisms and biological signi¢cance of pulsatile hormone secretion. Wiley, Chichester (Novartis Found Symp 227) p 5^18 Knobil E 1981 Patterns of hormone signals and hormone action. New Engl J Med 305:1582^1583 Lauzeral J, Halloy J, Goldbeter A 1997 Desynchronization of cells on the developmental path triggers the formation of spiral waves of cAMP during Dictyostelium aggregation. Proc Natl Acad Sci USA 94:9153^9158 Winfree AT 1987 When time breaks down: the three-dimensional dynamics of electrochemical waves and cardiac arrhythmias. Princeton University Press, Princeton, NJ Winfree AT 1994 Electrical turbulence in three-dimensional heart muscle. Science 266:1003^1006 Wu FCW, Butler GE, Kelnar CJH, Huhtaniemi I, Veldhuis JD 1996 Patterns of pulsatile luteinizing hormone secretion from childhood to adulthood in the human male: a study using deconvolution analysis and an ultrasensitive immuno£uorometric assay. J Clin Endocrinol Metab 81:1798^1805
Timing-dependent modulation of insulin mitogenic versus metabolic signalling Pierre De Meyts and Ronald M. Shymko* Department of Receptor Biology, Hagedorn Research Institute, Niels Steensens Vej 6, DK2820 Gentofte, and *Department of Scienti¢c Computing, Novo Nordisk A/S, DK-2760 Mlv, Denmark
Abstract. This chapter will not deal sensu stricto with the mechanisms and biological signi¢cance of pulsatile hormone secretion, the general theme of this book. Rather, we will attempt to demonstrate that timing events at the receiving end of the hormonal signal, i.e. the kinetics and duration of receptor activation in target cells and subsequent downstream signalling, can play an equally important role as that of the timing aspects of secretion, in determining the qualitative and quantitative aspects of hormonal responses. We will focus on the mechanisms that determine signalling speci¢city by the receptor tyrosine kinases, especially the insulin receptor and the type I insulin-like growth factor receptors (IGF-I receptor). We will be succinct and refer the reader to our recent reviews and publications on this topic and references therein. 2000 Mechanisms and biological signi¢cance of pulsatile hormone secretion. Wiley, Chichester (Novartis Foundation Symposium 227) p 46^60
Receptor tyrosine kinase activation and downstream signalling The receptor tyrosine kinases (RTKs) transduce a variety of extracellular signals that regulate various cellular processes such as metabolism, proliferation, di¡erentiation, motility and cytoskeletal rearrangement. The signalling process is initiated upon ligand binding, which induces dimerization of the receptors. Homoand heterodimerization events are an important subset of protein^protein interactions and are frequently employed in transducing signals from the cell surface to the nucleus, not only by cell surface receptors but by many components of the signalling cascades including transcription factors (reviewed in Klemm et al 1998). One interesting consequence of ligand-induced dimerization is the occurrence a bell-shaped dose^response curve for biological e¡ects due to self-antagonism at 46
MITOGENIC VERSUS METABOLIC INSULIN SIGNALLING
47
high ligand concentration, as discussed in details elsewhere (Fuh et al 1992, Ilondo et al 1994, De Meyts et al 1995). Although the insulin and insulin-like growth factor (IGF)-I receptors are covalent dimers even in the absence of ligand, it has been proposed that one ligand molecule bridges the two a subunits, and that the mechanism of activation therefore resembles that of monomeric RTKs and cytokine receptors (De Meyts 1994, Sch¡er 1994). In the case of the RTKs, dimerization (or cross-linking by ligand in the case of the insulin and IGF receptors) results in autophosphorylation of a number of cytoplasmic tyrosines by a mechanism referred to as transphosphorylation (Schlessinger 1988, Yarden & Ullrich 1988, Ullrich & Schlessinger 1990, Schlessinger & Ullrich 1992, Heldin 1995, Heldin & stman 1996). This autophosphorylation has two consequences. First, the intrinsic activity of the tyrosine kinase is enhanced, by removing from the active site an inhibitory pseudo-substrate loop (Hubbard et al 1994). Second, the phosphorylated tyrosines become binding sites for a variety of intracellular molecules (Myers & White 1996, White 1996, Virkamki et al 1999) such as: . SH2 domain-containing proteins such as the p85 subunit of phosphatidylinositol 3’-kinase (PI3K), Ras GTPase-activating protein (RasGap), phospholipase C g 1 (PLCg1), the phosphatase SHP2 and the Src family kinases; . SH2^SH3 adaptor proteins, such as Grb2, Grb7, Shc and Nck, which link the activated receptor to other downstream signalling molecules such as SOS, a Ras guanine nucleotide exchange factor; . PTB (phosphotyrosine-binding domain)-containing docking proteins such as those of the IRS (insulin receptor substrate) family, of which four members (IRS-1 to 4) have been cloned, and Gab1 (Myers & White 1996, White 1996). In the case of insulin and IGF-I receptors, the tyrosine phosphorylated IRS represents an intermediate link between the receptor and the above-mentioned SH2 domain-containing signalling and adaptor molecules. The signalling molecules become activated by binding to the receptor (or IRS) either because they get phosphorylated by the RTK or because of their relocation at the inner face of the plasma membrane (Fambrough et al 1999). Determinants of signalling speci¢city The major downstream pathways activated by the interactions described above include the Ras pathway, the PI3K pathway and the PLCg pathway. All these pathways are ubiquitous and can be activated by a variety of ligands and receptors, not only from the RTK family but also from the growth
48
DE MEYTS & SHYMKO
hormone/cytokine receptor family and even some G protein-coupled receptors (De Meyts & Seedorf 1997). This fact, and the extensive degree of cross-talk between the di¡erent pathways, makes it di⁄cult to fully understand the nature of signalling speci¢city, that is, how a given ligand stimulates a response domain (e.g. metabolic signalling by insulin) that is di¡erent from that of another ligand (e.g. mitogenic signalling by IGF-I), especially when both receptors coexist on the same cell. In a recent provocative study using an advanced technique that allows the global expression monitoring of thousands of genes simultaneously on oligonucleotide arrays (A¡ymetrix, Inc.), Fambrough et al (1999) have shown that the same set of 66 immediate early genes is induced by both the PDGFb receptor (or more precisely a chimera of its cytoplasmic domain with the extracellular ligandbinding domain of the human M-CSF receptor) and the FGF receptor in ¢broblasts. In contrast, only a subset of these genes was stimulated by EGF (which usually activates the same pathways). Even more strikingly, a mutant PDGFb receptor with all ¢ve tyrosines essential for RasGap, PI3K, SHP2 and PLCg signalling mutated to phenylalanine, still activated exactly the same broad array of genes, but to an average level of 59% ( 30%) of that seen with the control receptor. This suggests that none of the above-mentioned signalling pathways is absolutely required, and also that they induce broadly overlapping groups of immediate early genes rather than independent modules. Such complexity makes it clear that the simplest model of signalling speci¢city as may have been suggested by early ‘second messenger’ concepts, that would feature linear and parallel £ows of information from speci¢c receptors through unique signalling pathways to speci¢c response domains, is inadequate in providing a description of the dynamics of intracellular signalling networks. A summary of some of the factors that may contribute to explaining signalling speci¢city is provided in Table 1. In the following discussion, we would like to emphasize one aspect of signalling speci¢city, the importance of which has now started to be recognized: the role of the kinetics of activation of signalling molecules in determining response speci¢city, as exempli¢ed in studies of the mitogenic and metabolic signalling by insulin analogues. Mitogenic versus metabolic signalling by insulin and insulin analogues The role of insulin as a growth factor/mitogen has been debated (for review see Moses 1991, Ish-Shalom et al 1997). The conventional view is that insulin acts primarily as a hormone that regulates metabolic pathways in an endocrine manner, while the related IGF-I and -II act primarily as growth factors/mitogens that regulate cell growth, di¡erentiation, apoptosis and motility in an endocrine as
MITOGENIC VERSUS METABOLIC INSULIN SIGNALLING
TABLE 1
49
Determinants of signalling speci¢city
Types of receptors expressed in a given cell. Level of expression of a given receptor. Stoichiometry of receptors/substrates. Types of substrates/signalling molecules expressed in a given cell. Relative a⁄nity of receptors for substrates/signalling molecules; mass action (competition for same substrates depending on relative levels of receptor expression). Selectivity of recognition domains (SH2, SH3, PH, PTB). Hybrid receptors, hetero-oligomers. Feedback loops (phosphatases, serine/threonine kinases). Localization of signalling molecules within the cell. Activation kinetics of signalling molecules (transient versus sustained, time delays). Modi¢ed from De Meyts et al (1995).
well as paracrine and autocrine manner. Insulin and the IGFs share each other’s properties at high concentration due to a weak a⁄nity for the non-cognate receptor. However, recent data have unequivocally established that in the proper cellular context insulin can act as a mitogen in the absence of IGF-I receptors, as shown in rare cell lines devoid of IGF-I receptors such as the T-lymphoma LB line (IshShalom et al 1997). Reciprocally, it was clearly demonstrated that IGF-I can exert typical insulin-like metabolic e¡ects such as stimulation of glucose transport or glycogen synthase at low ligand concentrations through the IGF-I receptor in ¢broblasts of mice with a deletion of the insulin receptor gene (IR knockout mice) (Lamothe et al 1998). This raises the question of what mechanisms regulate the speci¢city of insulin and IGF signalling, and the balance of mitogenic versus metabolic potencies, considering that insulin and the IGFs share similar signalling pathways including the MAPK and PI3K pathways, although there are quantitative di¡erences, e.g. in the way they interact with IRS-1 and IRS-2 in various cell types (E. Van Obberghen, personal communication 1999). One clue to this problem came from recent studies comparing the mitogenic and metabolic potencies of a number of recombinant human insulin analogues bearing various amino acid mutations (Drejer 1992). De Meyts et al (1993) showed using the above-mentioned LB cell line devoid of IGF-I receptors, that a number of insulin analogues displayed mitogenic potencies far in excess of their metabolic potencies and receptor binding a⁄nity. These ‘supermitogenic’ analogues all had in common a slower dissociation rate (i.e. an increased residence time) from the insulin receptor than wild-type human insulin. One such analogue, Asp B10 human insulin, had in fact been shown to induce both benign and malignant mammary tumours in rats upon chronic treatment (Drejer 1992). A quantitative
50
DE MEYTS & SHYMKO
relationship between slow dissociation rates and mitogenic potency was further established by Hansen et al (1996) who showed in CHO cells overexpressing the insulin receptor that an increased disproportion between mitogenic potency and metabolic potency was observed as the dissociation rate of the ligand slowed down (Fig. 1), resulting in an exponential increase in the mitogenic/metabolic potency ratio with decreasing o¡-rate (Fig. 2). Hansen et al (1996) also showed that this enhanced mitogenic potency correlated with more sustained activation of key signalling events such as the insulin receptor tyrosine kinase and Shc. In contrast with this dependence of mitogenic signalling potency on a prolonged residence time on the insulin receptor, Holst and De Meyts have shown that the metabolic potency of insulin analogues (e.g. in stimulating lipogenesis in rat adipocytes) is best correlated with event occurring within the ¢rst minutes of ligand binding, possibly IRS-1^3 activation; thus, analogues with slow association kinetics display reduced potencies proportional to their on rates (Holst et al 1998). Other studies have also demonstrated a determinant e¡ect of the timing of signal activation on the relative potency of ligands in activating divergent pathways. One
FIG. 1. Relationship between dissociation rate constants and biological potencies of insulin analogues. Initial dissociation rate constants (Kd), expressed as a percentage of that of insulin, were plotted against mitogenic and metabolic potencies (ED50) values as a percentage of that of insulin. Closed circles: mitogenic potency; open triangles: metabolic potency (3-O-methylglucose uptake). Values are presented as means SEM. Reproduced with permission from Hansen et al (1996) Biochemical Journal, 315, 271^279. & the Biochemical Society.
MITOGENIC VERSUS METABOLIC INSULIN SIGNALLING
51
FIG. 2. Correlation between Kd and the mitogenic/metabolic potency ratio of insulin analogues. The ratio of the relative mitogenic to metabolic biological potencies of a number of insulin analogues (means SEM) were plotted as a function of their relative dissociation rate constant (Kd) (in percentage of that of insulin). The relationship is best described by an exponential function. Reproduced with permission from Hansen et al (1996) Biochemical Journal, 315, 271^279. & the Biochemical Society.
well-known example is the contrasting e¡ects of NGF and EGF in the neuronal PC12 cell line. NGF slows growth and stimulates di¡erentiation, while EGF has the opposite e¡ect. Both ligands stimulate the MAPK pathway, but EGF induces only a transient activation while NGF induces a sustained activation that allows translocation of MAPK to the cell nucleus and the activation of immediate early genes (Traverse et al 1992, Nguyen et al 1993). The EGF receptor can mimic the NGF di¡erentiating e¡ect and induce sustained signalling if overexpressed in the PC12 cells (Traverse et al 1994). Unfortunately, few studies of signalling mechanisms have addressed the quantitative and kinetic aspects of signalling events, usually relying on single time co-immunoprecipitation of various signalling molecules after simultaneous overexpression in a given cell type. While these studies have generated useful information, some of them may demonstrate interactions that could take place, but not necessarily those that are actually taking place under physiological levels of expression.
52
DE MEYTS & SHYMKO
Logical analysis of timing-dependent receptor signalling speci¢city It is not intuitively obvious by which mechanism, e.g. in the case of insulin analogues binding to the insulin receptor, the relative ratio of mitogenic potency to metabolic potency would be shifted in favour of mitogenic signalling with increased residence time on the receptor. In other words, is it possible to demonstrate that the timing of a key signalling event (including the kinetics of the receptor binding step) may a¡ect ‘decision making’ at branching signalling pathways? Figures 3 and 4 illustrate in a schematic way that two analogues of equal receptor a⁄nity, but respectively with fast and slow kinetics, can be distinguished qualitatively both at the single receptor level (Fig. 3) and at the receptor population level (Fig. 4) (Shymko et al 1999). Figure 3 shows that although the average time a given receptor is occupied is the same for both ligands, there are fewer binding and dissociation events per unit time for ligands with slow kinetics. Figure 4 shows that although the fraction of receptors occupied at any instant is the same in both cases, for slow kinetics the same receptors are occupied for longer times, whereas for fast kinetics di¡erent receptors are occupied over time. The experiments described above indicate that the metabolic e¡ects are similar in the two cases (except for those analogues with such slow association rates that they do not reach equilibrium during the ¢rst few minutes where the signal is committed), whereas mitogenesis is enhanced for slow-dissociating insulin analogues.
FIG. 3. Comparison of ligand occupancy at a single receptor for ligands with equal equilibrium a⁄nities but fast or slow binding kinetics. There are fewer binding and dissociation events per unit time for ligands with slow kinetics, but the average time that a given receptor is occupied is the same for ligands with slow and fast kinetics. Reproduced with permission from Shymko et al (1999) Biochemical Journal, 339, 675^683. & the Biochemical Society.
MITOGENIC VERSUS METABOLIC INSULIN SIGNALLING
53
FIG. 4. Comparison of receptor-population occupancy for ligands with equal equilibrium a⁄nities but fast or slow binding kinetics. Open and ¢lled circles represent unoccupied and occupied receptors, respectively. For slow kinetics, the same receptors are occupied for longer times, whereas for fast kinetics di¡erent receptors are occupied over time. The fraction of receptors occupied at any instant is the same in both cases. Experiments indicate that metabolic e¡ects are the same in the two cases, whereas mitogenesis is enhanced for slow-dissociating analogues. Reproduced with permission from Shymko et al (1999) Biochemical Journal, 339, 675^683. & the Biochemical Society.
In order to provide a simple method for the formal quantitative description of timing-dependent signalling networks, we have used (Shymko et al 1997) a logical formalism based on the approach developed by Thomas et al (Thomas 1973, 1991, Thomas & D’Ari 1989) for genetic networks, which emphasizes the importance of asynchronous switching and feedback loops in interacting networks. Space is lacking for a comprehensive description of this approach in this brief overview and we refer the interested reader to the original publication (Shymko et al 1997). Brie£y, in the basic formulation of a simple form of logical switching theory, we considered a three-step process: ligand binding to the receptor, activation of the receptor, and transmission of a mitogenic signal if the receptor is activated long enough. The state of this system is de¢ned in terms of three Boolean variables which can take the value of 0 or 1: b = receptor occupied by ligand (1) or not (0) a = receptor activated (1) or not (0), s = signal (i.e. mitogenic signal) transmitted (1) or not (0). The system can be represented as follows:
54
DE MEYTS & SHYMKO
The application of logical transition rules to this system allows the construction of a transition diagram as shown in Fig. 5 to show all logically possible state transitions, and where circled states represent the steady states of the system. There are two possible biological end states, one with no signalling (state 000, reached either through Path 1 or Path 2) and one with signalling (state 001, reached through Paths 3 or 4). Therefore the conditions for the absence of mitogenic signalling, or its occurrence, will be expressed as conditions for traversing either Path 1 or Path 2 versus traversing Path 3 or Path 4. At each branch, the arm requiring the least time will be traversed, and the condition for choosing one arm instead of another can be expressed as an inequality between the times for traversing the respective arms of the branch. To derive these conditions we take the starting time at state 100 (i.e. ligand has just bound to an inactive receptor) as t = 0, and calculate the total time for reaching each succeeding state in terms of the characteristic transition times of the Boolean state variables. The conditions for selection of each branch arm are then expressed in terms of inequalities among these total times. A time for reaching every state in the transition diagram can be written in terms of the characteristic state variable switching times, and the inequality conditions for traversing any particular arm of a branch are thus directly derived. Since any complete pathway leading to a biological endpoint involves several successive branching decisions, the complete condition for traversal of a pathway becomes the logical AND of the conditions for selection of each branch along the path. The conditions for the
FIG. 5. Transition diagram showing all possible state transitions in the logical system. Under each state is shown the time to reach each state relative to time 0, in terms of the characteristic state variable transition times. See Shymko et al (1997) for a detailed explanation. Reproduced with permission from Shymko et al (1997) Biochemical Journal, 326, 463^469. & the Biochemical Society.
MITOGENIC VERSUS METABOLIC INSULIN SIGNALLING
55
alternative biological endpoints (mitogenic signalling or no mitogenic signalling) can be logically deduced. Thus, it can be logically demonstrated (see Shymko et al 1997) that for mitogenic signalling, the residence time of the ligand on the receptor must be greater than the maximum of two quantities, the ¢rst of which is the receptor activation time, and the second of which is the sum of activation time plus proliferation commitment time minus activation decay time. The logical system developed above shows how the timing of biological events in the signalling pathway can be taken into account in determining the speci¢city of receptor signalling. This approach, however, does not provide an explanation in molecular terms of this timing-dependent speci¢city. In subsequent theoretical considerations, Shymko et al (Shymko et al 1999) have shown that if signalling is transmitted through a single e¡ector that binds coincidentally with the ligand to the receptor and whose association and dissociation kinetics are slow relative to the hormone dissociation rate, the resulting biological e¡ect is predicted to be dependent on ligand-binding kinetics. However, known primary e¡ector molecules associating with the insulin receptor bind and interact rapidly with the receptor, and therefore a single-e¡ector model will not provide the required timing-dependence. Shymko et al (1999) showed that a model with two e¡ectors which must bind coincidentally with the ligand for signalling to occur, gives the required dependence of signalling on ligand binding kinetics, provided that at least one of the e¡ectors has slow binding kinetics relative to hormone binding. In this case, the other e¡ector can have rapid kinetics, which is consistent with the properties of the major known substrates of the insulin receptor, such as the IRS molecules. In fact, it is not necessary that the slow kinetic component be a second e¡ector; it could be a signalling event located further down the signalling chain. The entire signalling chain can thus be thought of as a coincidence detector for the association and activation of its multiple molecular components; the behaviour of such a system is sensitive to the kinetics of the slowest reaction in the chain. Conclusion and perspectives Although there has been a massive increase in the last few years in our understanding of the molecular components of the intracellular signalling pathways that mediate the e¡ects of extracellular ligands, and of their interactions, the quantitative aspects of the signalling network and the exact mechanisms that determine speci¢city still elude us. Further progress may require better methodologies that allow fast kinetic measurements of signalling interactions, measurements of translocation of signalling molecules between various cell compartments and new approaches to integrated network analysis, including for example applications of queuing theory and logical switching theory.
56
DE MEYTS & SHYMKO
Acknowledgements We are grateful to Rene¤ Thomas, Jacques Dumont and Erik Dumont for collaboration in some of the work discussed here, and to Stephane Swillens and Albert Goldbeter for helpful comments and discussions. The Hagedorn Research Institute is an independent basic research component of Novo Nordisk A/S.
References De Meyts P 1994 The structural basis of insulin and insulin-like growth factor-I (IGF-I) receptor binding and negative cooperativity, and its relevance to mitogenic versus metabolic signaling. Diabetologia 37(suppl 2):S135^S148 De Meyts P, Seedorf K 1997 The mechanism of insulin receptor binding, activation and signal transduction. In: Zahnd GR, Wollheim CB (eds) Contributions of physiology to the understanding of diabetes. Springer-Verlag, Berlin, p 89^107 De Meyts P, Christo¡ersen CT, Urs B et al 1993 Insulin potency as a mitogen is determined by the half-life of the insulin-receptor complex. Exp Clin Endocrinol 101:22^23 De Meyts P, Urs B, Christo¡ersen CT, Shymko RM 1995 Mechanism of insulin and IGF-I receptor activation and signal transduction speci¢city. Receptor dimer cross-linking, bellshaped curves, and sustained versus transient signaling. Ann N Y Acad Sci 766:388^401 Drejer K 1992 The bioactivity of insulin analogues from in vitro receptor binding to in vivo glucose uptake. Diabetes Metab Rev 8:259^285 Fambrough D, McClure K, Kazlauskas A, Lander ES 1999 Diverse signaling pathways activated by growth factor receptors induce broadly overlapping, rather than independent, sets of genes. Cell 97:727^741 Fuh G, Cunningham BC, Fukunaga R, Nagata S, Goeddel DV, Wells JA 1992 Rational design of potent antagonists to the growth hormone receptor. Science 256:1677^1680 Hansen BF, Danielsen GM, Drejer K 1996 Sustained signalling from the insulin receptor after stimulation with insulin analogues exhibiting increased mitogenic potency. Biochem J 315:271^279 Heldin CH 1995 Dimerization of cell surface receptors in signal transduction. Cell 80:213^223 Heldin CH, O«stman A 1996 Ligand-induced dimerization of growth factor receptors: variations on the theme. Cytokine Growth Factor Rev 7:3^10 Holst P, Lars L, De Meyts P 1998 Potency of insulin analogues on lipogenesis is determined early in the binding process. Diabetes 47 (suppl 1):1562 Hubbard SR, Wei L, Ellis L, Hendrickson WA 1994 Crystal structure of the tyrosine kinase domain of the human insulin receptor. Nature 372:746^754 Ilondo MM, Damholt A, Cunningham BA, Wells JA, De Meyts P, Shymko RM 1994 Receptor dimerization determines the e¡ects of growth hormone in primary rat adipocytes and cultured IM-9 lymphocytes. Endocrinology 134:2397^2403 Ish-Shalom D, Christo¡ersen CT, Vorwerk P et al 1997 Mitogenic properties of insulin and insulin analogues mediated by the insulin receptor. Diabetologia 40 (suppl 2):S25^S31 Klemm JD, Schreiber SL, Crabtree R 1998 Dimerization as a regulatory mechanism in signal transduction. Annu Rev Immunol 16:569^592 Lamothe B, Baudry A, Christo¡ersen CT et al 1998 Insulin receptor-de¢cient cells as a new tool for dissecting complex interplay in insulin and insulin-like growth factors. FEBS Lett 426:381^385 Moses AC 1991 Is insulin a growth factor? In: LeRoith D (ed) Insulin-like growth factors: molecular and cellular aspects. CRC Press, Boca Raton, FL, p 245^270 Myers MG, White MF 1996 Insulin signal transduction and the IRS proteins. Annu Rev Pharmacol Toxicol 36:615^658
MITOGENIC VERSUS METABOLIC INSULIN SIGNALLING
57
Nguyen TT, Scimeca JC, Filloux C, Peraldi P, Carpentier JL, Van Obberghen E 1993 Coregulation of the mitogen-activated protein kinase, extracellular signal-regulated kinase 1, and the 90 kDa ribosomal S6 kinase in PC12 cells. Distinct e¡ects of the neurotrophic factor, nerve growth factor, and the mitogenic factor, epidermal growth factor. J Biol Chem 268:9803^9810 Sch¡er L 1994 A model for insulin binding to the insulin receptor. Eur J Biochem 221:1127^1132 Schlessinger J 1988 Signal transduction by allosteric receptor oligomerization. Trends Biochem Sci 13:443^447 Schlessinger J, Ullrich A 1992 Growth factor signaling by receptor tyrosine kinases. Neuron 9:383^391 Shymko RM, De Meyts P, Thomas R 1997 Logical analysis of timing-dependent receptor signalling speci¢city: application to the insulin receptor metabolic and mitogenic signalling pathways. Biochem J 326:463^469 Shymko RM, Dumont E, De Meyts P, Dumont JE 1999 Timing-dependence of insulin-receptor mitogenic versus metabolic signalling: a plausible model based on coincidence of hormone and e¡ector binding. Biochem J 339:675^683 Thomas R 1973 Boolean formalization of genetic control circuits. J Theor Biol 42:563^585 Thomas R 1991 Regulatory networks seen as asynchronous automata: a logical description. J Theor Biol 153:1^23 Thomas R, D’Ari R 1989 Biological feedback. CRC Press, Boca Raton, FL Traverse S, Gomez N, Paterson H, Marshall C, Cohen P 1992 Sustained activation of the mitogen-activated protein (MAP) kinase cascade may be required for di¡erentiation of PC12 cells. Comparison of the e¡ects of nerve growth factor and epidermal growth factor. Biochem J 288:351^355 Traverse S, Seedorf K, Paterson H, Marshall CJ, Cohen P, Ullrich A 1994 EGF triggers neuronal di¡erentiation of PC12 cells that overexpress the EGF receptor. Curr Biol 4:694^701 Ullrich A, Schlessinger J 1990 Signal transduction by receptors with tyrosine kinase activity. Cell 61:203^212 Virkamki A, Ueki K, Kahn CR 1999 Protein^protein interaction in insulin signaling and the molecular mechanism of insulin resistance. J Clin Investig 103:931^943 White MF 1996 The IRS-signalling system in insulin and cytokine action. Philos Trans R Soc Lond B Biol Sci 351:181^189 YardenY,UllrichA1988Growthfactorreceptortyrosinekinases.AnnuRevBiochem57:443^478
DISCUSSION Veldhuis: I see analogies with some of the other hormone receptors. One that comes to mind is hCG, which has a very slow dissociation from the luteinizing hormone (LH) receptor compared to LH, and has a much greater downregulating e¡ect. This is sort of analogous in that the residence time on the receptor is greatly extended. Another is the gonadotropin-releasing hormone antagonist, which binds in an almost non-dissociable way. Marshall: When you use your high a⁄nity insulin analogue, are those receptors not desensitized or down-regulated? De Meyts: That’s a good question. I don’t think there is any evidence in the insulin receptor family of a receptor being desensitized by prolonged residence time, unlike the situation with the b-adrenergic receptor. What seems to be
58
DISCUSSION
important in the insulin receptor regulation is a rapid internalization of the ligand^ receptor complex in the endosome. At the acid pH of the endosome, there is fast dissociation of insulin from the receptor. This is a classic mechanism of downregulation, because once the ligand dissociates there are phosphatases which quickly inactivate the receptor, and then some of the receptor is recycled, some is degraded, and some of the insulin is also re-excreted or degraded. It has been shown that a chronic increase in insulin concentration will induce an enhancement of this ligand-induced endocytosis and then a down-regulation of the receptor concentration. This can be pretty fast in some cell types. There is not much evidence of desensitization by kinases; it is more due to a cell biological phenomenon of internalization and recycling. We do not yet have full information with our insulin analogues as to what extent their internalization, dissociation in endosomes and recycling is comparable to wild-type insulin. Copinschi: The di¡erence between the metabolic and mitogenic actions is quite consistent with the fact that the half-life of insulin is short and the half-life of IGF-I is long. I have a rather na|« ve question: what happens when you treat with the longacting insulins? De Meyts: That is a good question. However, we have to distinguish two things: when we talk of short-acting and longer-acting insulins, we are talking of its circulating life in the blood, not of what happens once it is bound to the receptor. Once bound to the receptor, the long-acting and short-acting insulins have similar a⁄nity and kinetics. So we are talking here of integrated in vivo life, as opposed to increased residence time on a single receptor. It is this staying longer on a single receptor which is more mitogenic, not staying longer in the circulation. Copinschi: But the pancreatic secretion of insulin is pulsatile, and I imagine that the fact that it is not secreted continuously has physiological signi¢cance. When insulin is released continuously by long-acting preparations, there should be some di¡erence. De Meyts: Here we are discussing what happens when it arrives at the single cell, and that’s where the discrimination regarding signalling speci¢city seems to happen. There is no evidence that making the insulin shorter or longer acting in vivo has any of those di¡erential e¡ects on signalling; it is quite a di¡erent level of integration. And when we compare insulin and IGF-I, it is probably more signi¢cant that IGF-I has as many as eight di¡erent binding proteins. There are two major ones in blood which keep the half-life much longer, and then quite a number at the cellular level modifying the actions of IGF-I. We don’t quite know what e¡ects this has on the metabolic versus mitogenic signalling potencies. The same is true for growth hormone (GH): I don’t think anyone has really presented a clear perspective as to how the GH binding protein modulates the action of GH with respect to pulsatile secretion versus cell sensitivity. Licinio: Why is the e¡ect so di¡erent if the receptors are occupied longer?
MITOGENIC VERSUS METABOLIC INSULIN SIGNALLING
59
De Meyts: We don’t really know at this point which downstream signalling step is sensitive to timing in the case of insulin’s mitogenic signalling; this will require careful quantitative studies of the kinetics of signalling pathways. Robinson: I wonder what the logical extension of your argument is if you do the same sort of thing with IGF-Ithat is, to shorten the IGF-I duration of action. Has this sort of experiment been done to make a more speci¢c metabolic analogue? De Meyts: We are very interested in pursuing this and are currently collaborating with several groups trying to make appropriate IGF-I analogues. This is di⁄cult, because the recombinant molecule has trouble folding properly. But theoretically we hope to be able to make IGF-I analogues with increased metabolic and decreased mitogenic potencies by playing with the receptor binding kinetics. Robinson: Then I would like to provoke you into speculation, because such a molecule would still have the long-acting half-life, assuming it’s still bound to all the binding proteins. So you would have the kinetics of the long-lasting form, but a short-acting IGF-I on the receptor. Will you then really be able to say that those two receptors are mediating basically the same signal? De Meyts: That is exactly what we would like to look at. I think that an IGF-I with fast kinetics and which would have a much more metabolic than mitogenic e¡ects may be useful in cases of insulin resistance, because it would bypass the insulin receptor and induce metabolism through the IGF-I receptor. This is purely conceptual, however. Veldhuis: Just to be sure we understand the problem, would you tell us several ways in which you would design antagonists. The best example I know is the GH receptor antagonist, which is fairly clever. Could you remind us how that works, and perhaps propose some novel strategies whereby we could block hormone signalling? De Meyts: In the case of the receptors which are activated by ligand-induced dimerization (essentially RTKs and GH/cytokine receptors), one can make antagonists in two ways. The ¢rst is by mutating the ligand itself so that the binding surface that binds last in the sequential binding process (site 2) becomes incapable of binding the signalling receptor, like the G120R mutation in the growth hormone molecule (Fuh et al 1992). The other is to make a small peptide or small molecule which will bind to one of the two receptor binding sites and interfere with the cross-linking by the main ligand. Matthews: Reverting to the question of binding and pulsatility in insulin, and whether that’s a¡ecting or modulating insulin resistance, we did some work a while ago where we were giving steady-state overnight insulin and pulsed insulin. It seems that you get a better glucose clearance e¡ect with pulsed insulin, which comes back to the question of whether you can actually detect these things in vivo. But, of course, you need to give intravenous insulin to detect this. I’m sure that even fast-acting insulin such as Lis-pro or the new Novo insulins are simply
60
DISCUSSION
not going to be absorbed with a subcutaneous kinetic that is ever going to demonstrate that su⁄ciently, though it would be nice if they did. Indeed, just on that point, it has been a pharmaceutical company misnomer to talk about longacting and short-acting insulins, because they’re actually long-releasing and short-releasing insulins: the circulating insulin, until recently, has had identical molecules. With regard to the other point about insulin resistance, and whether in fact if you are using the same amount of insulin overnight and you get di¡erent e¡ects this is technically a change in insulin resistance, we examined speci¢c monocyte binding in humans and demonstrated a small di¡erence between pulsatile and continuous delivery (Matthews et al 1983). But this actually wasn’t su⁄cient to explain the whole e¡ect, and so I’m sure that Pierre is right in thinking that there is something downstream from the binding domains that actually changes the insulin resistance. There is no doubt that receptor binding is a small part of the insulin resistance problem, and most of the domains are intracellular probably nuclear receptor domains. De Meyts: It is clear that if you look at pulsatile versus continuous availability of insulin it may have very strong di¡erential e¡ects on the receptor down-regulation. But we know very little at this point about down-regulation of downstream signalling elements, which could also be a¡ected independently of receptor binding. I agree that you could possibly modulate insulin resistance depending on the way insulin is phased. I don’t know if anyone has really looked at that. Matthews: We looked at it from the point of view of seeing whether membrane binding was enough to explain it, and clearly it isn’t. Robinson: Do you think the residence time e¡ects the e⁄ciency of internalization of the receptor, independent of concentration? De Meyts: It may depend on the cell type. In some cell types the kinetics of internalization is fast, so we would expect that even analogues with a relatively short residence time are internalized fast. The question is then what is the residence time of internalized insulin at the pH of the endosome, and whether this a¡ects signalling from the internalized insulin^receptor complex. We have evidence that some analogues with a prolonged residence time also have a prolonged residence time at lower pH. References Fuh G, Cunningham BC, Fukunaga R, Nagata S, Goeddel DV, Wells JA 1992 Rational design of potent agonists to the growth hormone receptor. Science 256:1677^1680 Matthews DR, Naylor BA, Jones RG, Ward GM, Turner RC 1983 Pulsatile insulin has a greater hypoglycaemic e¡ect than continuous delivery. Diabetes 32:617^621
Growth hormone pulse-activated STAT5 signalling: a unique regulatory mechanism governing sexual dimorphism of liver gene expression David J. Waxman Division of Cell and Molecular Biology, Department of Biology, Boston University, Boston, MA 02215, USA Abstract. Growth hormone (GH) exerts sexually dimorphic e¡ects on liver gene transcription that are regulated by the temporal pattern of pituitary GH release; this release is intermittent in male rats and nearly continuous in females. Comparisons of liver nuclear protein tyrosine phosphorylation in male and female rats have led to the discovery that the liver transcription factor STAT5b is tyrosine phosphorylated in male but not female rats in response to GH pulses. Intermittent plasma GH pulses trigger a rapid and repeated tyrosine phosphorylation and nuclear translocation of liver STAT5b in intact male rats, while the more continuous pattern of GH exposure down-regulates the STAT5b signalling pathway in female rat liver. The central importance of STAT5b for the physiological e¡ects of GH pulses has been veri¢ed using a mouse gene knockout model. STAT5b gene disruption leads to a major loss of multiple sexually di¡erentiated responses associated with the sexually dimorphic pattern of pituitary GH secretion. Malecharacteristic body growth rates and male-speci¢c liver gene expression are decreased to wild-type female levels in STAT5b7/7 males, while female-predominant liver gene products are increased in males to near female levels. STAT5b is thus a liver-expressed, latent cytoplasmic transcription factor that undergoes repeated tyrosine phosphorylation and nuclear translocation in response to intermittent plasma GH stimulation, and is a key intracellular mediator of the stimulatory e¡ects of GH pulses on male-speci¢c liver gene transcription. Other studies indicate, however, that STAT5a and STAT5b are both required for constitutive expression in female, but not male liver, of certain GHregulated CYP enzymes. GH activation of both STAT5 proteins, which in turn form distinct homodimeric and heterodimeric DNA-binding complexes, is thus an important determinant of the sex-dependent and gene-speci¢c e¡ects that GH has on the liver. 2000 Mechanisms and biological signi¢cance of pulsatile hormone secretion. Wiley, Chichester (Novartis Foundation Symposium 227) p 61^81
Pulsatility of plasma growth hormone (GH) secretory pattems and impact on liver gene expression The polypeptide hormone GH signals to hepatocytes and other target cells via its plasma membrane-bound receptor, GHR, a member of the cytokine/growth factor 61
62
WAXMAN
receptor superfamily (Waxman & Frank 2000). GH regulates the transcription of a large number of genes involved in somatic growth, carbohydrate and lipid metabolism, and liver function. One unique aspect of GH action is the di¡erential responsiveness of target tissues to GH’s sexually dimorphic plasma pro¢le, which can be observed in many species, including humans, but is particularly striking in rodents. In adult male rats, GH is secreted by the pituitary gland in an intermittent manner, to give plasma GH pulses that peak at 200 ng/ml each 3^3.5 h, whereas in females GH is secreted more frequently, resulting in a near continuous presence of GH in circulation at an average level of 40^50 ng/ml (Jansson et al 1985). Pulsatile GH is more e¡ective than continuous GH in promoting pubertal weight gain associated with long bone growth (Jansson et al 1982, Waxman et al 1991), however, the underlying mechanisms by which the temporal plasma pro¢le of GH regulates this important physiological response are not well understood. Liver metabolic function, in particular cytochrome P450 (CYP)-catalysed steroid and foreign chemical metabolism, also exhibits sexual dimorphism in response to the sexually dimorphic plasma GH pro¢les (Skett 1987, Waxman & Chang 1995). CYP2C11 and CYP2C12 are steroid hydroxylase P450 enzymes that are exclusively expressed in male and female rat liver, respectively, and have served as prototypic examples of sexually dimorphic, GH plasma pattern-regulated liver gene products (Waxman 1992). In vivo models have established that CYP2C11 gene transcription can be induced in hypophysectomized (GH-depleted) rats given GH in a pulsatile manner that mimics the intact male plasma GH pro¢le, whereas the same gene is suppressed and gene CYP2C12 is activated when GH is given in a continuous, female-like pattern (Legraverend et al 1992, Sundseth et al 1992). While some e¡ects of GH are mediated indirectly through intermediary factors produced in response to GH stimulation, such as insulin-like growth factor (IGF) I, attention has recently focused on intracellular signalling events mediated by JAK2 kinase, a GH receptor-associated tyrosine kinase that undergoes autophosphorylation following GH stimulation, and subsequently phosphorylates its STAT protein substrates (Darnell 1997). Seven individual STATs have been described as being activated by various cytokines, growth factors and hormones in a JAK kinase-dependent manner. Four of these STATs (STATs 1, 3, 5a and 5b) can be tyrosine phosphorylated following GH binding to its cell surface receptor, GHR. This tyrosine phosphorylation reaction, in turn, induces STAT protein homo- and heterodimerization, and it concomitantly activates the nuclear localization, DNA-binding and transcriptional activation potential of the STATs. GH-activated STAT proteins may thus serve as direct signal transducers to the nucleus that activate transcription by binding to de¢ned DNA response elements adjacent to target genes (Davey et al 1999).
GROWTH HORMONE PULSE REGULATION OF LIVER GENE EXPRESSION
63
FIG. 1. Activation of liver STAT5b by pulsatile GH. STAT5b activation is demonstrated by the presence of tyrosine phosphorylated STAT5b (‘p93’) in liver nuclei of adult male (M) but not female (F) rats, as shown by Western blotting using anti-phosphotyrosine (anti-pY) antibody (panel A). Panel B presents Western blots probed with anti-STAT5b (upper panel) or antiphosphotyrosine antibody. A pulse of GH given to hypophysectomized (Hx) male rats is shown to stimulate rapid conversion of STAT5b from its cytosolic, non-tyrosine phosphorylated form (lanes 4, 5) to a lower mobility tyrosine phosphorylated form (lanes 6, 7) which translocates into and accumulates in the nucleus (lane 3 vs. lanes 1, 2). Prolactin (PRL) and lipopolysaccharide (LPS) have no such e¡ect on liver STAT5b (lanes 8, 9). Data shown are based on Waxman et al (1995).
As summarized in this review article, studies on the e¡ects of GH on liver STAT5b, carried out in this laboratory in vivo using an adult rat model, have led to the discovery that STAT5b is uniquely responsive to the temporal pattern of plasma GH stimulation and in a manner that may enable it to serve as a direct transcriptional activator of male-speci¢c, GH pulse-activated liver-expressed genes (Waxman et al 1995). Intermittent plasma GH pulses, such as those which occur naturally in adult male rats, trigger a rapid, and repeated, tyrosine phosphorylation and nuclear translocation of liver STAT5b, while continuous plasma GH exposure, similar to the pattern that is found in adult female rats, leads to desensitization of this tyrosine phosphorylation pathway and consequently, a low steady-state level of the active, nuclear STAT5b transcription factor (Waxman et al 1995). This pattern of response suggests that STAT5b is a key intracellular mediator that can transduce the male GH pulse signal from the plasma membrane-bound GHR to male-expressed target genes within the nucleus. GH can also activate two other STATs in liver tissue, STAT1 and STAT3 (Gronowski & Rotwein 1994, Ram et al 1996), but in contrast to STAT5b, the activation of these other STATs is largely independent of GH’s temporal plasma pro¢le (Ram et al 1996). Other studies presented below
64
WAXMAN
include the ¢nding that targeted disruption of STAT5b leads to a selective loss of the male-speci¢c pattern of liver gene expression, as well as the loss of GH pulseinduced male whole body growth rates (Udy et al 1997), providing strong support for the hypothesis that STAT5b is a direct mediator of a broad range of physiological responses to male plasma GH pulses. Identi¢cation of liver STAT5b as a GH pulse-activated mediator of intracellular GH signalling We initially compared the patterns of liver nuclear protein tyrosine phosphorylation in male and female rats to ascertain whether any sex-dependent di¡erences could be observed. We thus discovered that a M r 93 000 nuclear protein, designated p93, is tyrosine phosphorylated to a high level in male but not female rat liver nuclei (Fig. 1A). A single physiological GH replacement pulse given to hypophysectomized rats rapidly increased p93 tyrosine phosphorylation, whereas prolactin was without e¡ect (Fig. 1B) (Waxman et al 1995). To investigate why phosphotyrosine-p93 can be detected in male but not female liver nuclei, we compared the e¡ects of intermittent vs. continuous GH treatment on p93 tyrosine phosphorylation. Intermittent pulses of GH, spaced 4 h apart, triggered repeated p93 phosphorylation in rat liver in vivo. By contrast, continuous GH exposure over an 8^24 h period desensitizes hepatocytes and leads to a dramatic decline in the steady-state level of phosphotyrosine-p93. We next investigated whether p93 corresponds to a GH-activated STAT protein, since various cytokines and growth factors were known to activate these tyrosine phosphorylatable latent cytoplasmic transcription factors. p93 was found to be identical to STAT5, whose tyrosine phosphorylation is stimulated by prolactin in mammary cells (Gouilleux et al 1994). Intermittent GH pulsation, but not continuous GH treatment, e⁄ciently translocated liver STAT5/p93 protein from the cytosol to the nucleus (Fig. 1B) and also activated its DNAbinding activity (Waxman et al 1995), which can be assayed using a STAT5binding DNA sequence from the rat b-casein promoter. Moreover, we observed a striking variability in liver nuclear STAT5 levels in individual male rats, with an excellent correlation between the presence of phosphotyrosine-STAT5 protein in the nucleus and the presence of a GH pulse in plasma at the time of death. Liver STAT5 thus undergoes repeated cycles of tyrosine phosphorylation and nuclear translocation in response to naturally occurring male plasma GH pulses, leading us to propose that STAT5b is an important intracellular mediator of the stimulatory e¡ects of plasma GH pulses on male-speci¢c liver CYP gene expression (Fig. 2). Subsequent to these initial studies, two closely related STAT5 genes, STAT5a and STAT5b (490% identical in their coding sequences) were
GROWTH HORMONE PULSE REGULATION OF LIVER GENE EXPRESSION
65
FIG. 2. Plasma GH pulses activate STAT5b by a mechanism that involves dimerization of GHR at the cell surface, recruitment of JAK2 tyrosine kinase, followed by tyrosine phosphorylation of JAK2, GHR and then STAT5b. These initial steps are followed by STAT5b dimerization, nuclear translocation and stimulation of the transcription of GH pulseinduced male-speci¢c target genes, such as those belonging to the cytochrome P450 superfamily.
66
WAXMAN
identi¢ed in several species (Liu et al 1995, Mui et al 1995). Using STAT5 formspeci¢c antibody probes and RT-PCR primers, it is now apparent that STAT5b is the major STAT5 form that is expressed and responds to GH pulses in both rat and mouse liver (relative mRNA abundance in liver: 85^90% STAT5b, 10^15% STAT5a) (Park et al 1999). STAT5a is, however, also activated in rat liver by plasma GH stimulation to a much higher level in males compared to females in a manner similar to STAT5b (Choi & Waxman 1999).
STAT5b activation pathway in CWSV-1 cells, a liver-derived cell culture model While the rat liver model greatly facilitated the discovery and characterization, in an in vivo, physiological context, of the GH pulse-activated STAT5b pathway, mechanistic questions proved to be more di⁄cult to address using the animal model. These questions include further details about the factors that underlie the GH pattern-dependence of STAT5b activation, the deactivation of STAT5b following termination of a GH pulse, and the desensitization of the GH receptor/JAK2 kinase/STAT5b pathway in liver cells exposed to GH continuously. We therefore utilized CWSV-1 cells, an SV40-immortalized rat hepatocyte-derived cell line that is GH responsive (Kempe et al 1995), to carry out more detailed investigations of STAT5b and its role in GH pulse-induced intracellular signalling. Using the CWSV-1 model, we found that GH pulses activate STAT5b, the major STAT5 form present in these liver cells, by tyrosine phosphorylation followed by a secondary serine or threonine phosphorylation reaction. Repeat cycles of GH pulsation led to repeat cycles of STAT5b activation then deactivation, similar to rat liver in vivo (Gebert et al 1997). Full responsiveness to succeeding GH pulses required a minimum GH o¡-time of about 2.5 h, but did not require new protein synthesis, indicating that the key components (GHR, JAK2, STAT5b) are probably recycled, rather than degraded after conclusion of a GH pulse. Continuous GH exposure led to down-regulation of activated STAT5b, consistent with the desensitization of this GH pulse-activated pathway that is seen in female rat liver. We also learned that the rapid deactivation of STAT5b following termination of a GH pulse (complete within 30 min) involves phosphotyrosine dephosphorylation as a key ¢rst step and can be blocked by pervanadate, a tyrosine phosphatase inhibitor (Gebert et al 1997). CWSV-1 cells thus provide a very useful liver cell model for investigation of the STAT5b activation pathway and its regulation by the temporal patterns of GH stimulation.
GROWTH HORMONE PULSE REGULATION OF LIVER GENE EXPRESSION
67
Termination of GH pulse-induced STAT5b signalling The studies described above establish that GH-activated STAT5b can be repeatedly activated by tyrosine phosphorylation catalysed by JAK2 kinase, deactivated by dephosphorylation via the action of a tyrosine phosphatase, and then reactivated by tyrosine rephosphorylation in response to a series of physiologic GH pulses. In order for a GH pulse to induce a full cycle of STAT5b activation and deactivation, several upstream signalling events need to occur (Fig. 2) (Waxman & Frank 2000). First, GH must bind to and dimerize its receptor (Cunningham et al 1991), then JAK2 is recruited (Argetsinger et al 1993), phosphorylating itself and GHR on multiple tyrosine residues (Hansen et al 1996). STAT5b is subsequently recruited to the GHR^JAK2 signalling complex (Yi et al 1996) and then is activated by a JAK2-dependent phosphorylation of tyrosine 699. GH-activated STAT5b also undergoes an H7-sensitive secondary phosphorylation on serine (Yamashita et al 1999, Gebert et al 1997, Ram et al 1996) that may modulate its transcriptional activity. GH-activated STAT5b dimerizes via SH2 domain interactions and then is transported to the nucleus (Ram & Waxman 1997), where it binds directly to DNA response elements upstream of GH target genes (Bergad et al 1995, Ganguly et al 1997, Subramanian et al 1998). The repeated activation in vivo of STAT5b by physiological GH pulses (Waxman et al 1995) requires that this JAK^STAT signalling pathway not only respond rapidly to the stimulatory e¡ect of a GH pulse, but also that it deactivate between GH pulses. As noted above, new protein synthesis is not required for the repeated activation of STAT5b by GH pulses, provided there is a GH-free interval of at least 2.5 h (Gebert et al 1997), corresponding to the physiological spacing of GH pulses in adult male rats (Tannenbaum & Martin 1976, Agrawal et al 1995, Jansson et al 1985). Since STAT5b reactivation is independent of protein synthesis under conditions where cellular STAT5b is quantitatively converted to its tyrosine phosphorylated, nuclear form, it must, at a minimum, be accompanied by dephosphorylation of activated STAT5b molecules and their recycling from the nucleus back to the cytosol (Fig. 2).
Interaction of GH-activated STATs with tyrosine phosphatase SHP-1 and nuclear JAK2 The precise mechanism whereby GH pulse-induced STAT5b signalling is terminated is of biological importance, given the apparent need for GH target cells to respond repeatedly to intermittent plasma GH pulses in order to achieve a male pattern of long bone growth and liver gene expression. Since STAT5b phosphotyrosine dephosphorylation is a critical step, both for deactivation of the
68
WAXMAN
nuclear STAT and for its subsequent return to the cytosol (Fig. 2), we investigated the tyrosine phosphatase that catalyses this dephosphorylation step based on the hypothesis that it contains an SH2 domain (src homology 2 domain¼phosphotyrosine-binding protein module), which would facilitate its direct binding to GH-activated STAT5b and/or other phosphotyrosinesignalling molecules (Fig. 3). These studies demonstrated that GH can activate by three- to fourfold the tyrosine phosphatase SHP-1, which is one of two known SH2 domaincontaining tyrosine phosphatases (Ram & Waxman 1997). We further discovered that a single pulse of GH stimulates a rapid translocation of SHP-1 from the cytosol to the nucleus and induces the speci¢c binding of SHP-1 to tyrosinephosphorylated nuclear STAT5b. Together, these ¢ndings suggest that the GHactivated tyrosine phosphatase SHP-1 may dephosphorylate nuclear STAT5b following the termination of a male GH pulse (Fig. 3). In other studies, we observed that JAK2 kinase is present both in the cytosol and in the nucleus, and showed that GH-activated STAT3, but not STAT5b, becomes speci¢cally associated with nuclear JAK2 (Ram & Waxman 1997). The function of nuclear JAK2 is uncertain, but we speculate that it could be related to some of the e¡ects that GH has on nuclear signalling molecules, perhaps including the activation of nuclear STAT3 via a direct JAK2 kinase interaction. This hypothetical STAT activation mechanism contrasts with the classic pathway of cytosolic STAT activation via a plasma membrane-associated receptor^JAK kinase complex. Moreover, it could further explain the substantially higher GH dose requirement for STAT3 (and STAT1) activation compared to STAT5b activation that is seen in liver in vivo (Ram et al 1996), given the comparatively low e⁄ciency with which GH is internalized to the nucleus of hormone-treated cells.
FIG. 3. Proposed mechanism for activation of tyrosine phosphatase (PTPase) SHP-1 upon binding of tyrosine phosphorylated STAT5b to one of its SH2 domains. The activated phosphatase is then hypothesized to dephosphorylate STAT5b, as shown.
GROWTH HORMONE PULSE REGULATION OF LIVER GENE EXPRESSION
69
STAT5b activation/deactivation cycle Further investigation of the cellular events involved in the cycle of STAT5b activation/deactivation in response to a GH pulse (Fig. 2) demonstrated that a brief exposure to GH and the associated activation of JAK2 tyrosine kinase are both necessary and su⁄cient to initiate all of the downstream steps associated with STAT5b tyrosine phosphorylation and the subsequent deactivation of both the GHR^JAK2 signalling complex and STAT5b (Gebert et al 1999a). GHR^ JAK2 signalling to STAT5b at the conclusion of a GH pulse could be sustained by the protein synthesis inhibitor cycloheximide or by the proteasome inhibitor MG132, indicating that termination of this JAK2-catalysed activation loop requires synthesis of a labile or GH-inducible protein factor and is facilitated by the proteasome pathway. In other experiments, the tyrosine phosphatase inhibitor pervanadate also sustained the GH pulse-induced STAT5b signal, but in that case the protective e¡ect was primarily at the STAT5b dephosphorylation step, rather than at the GHR^JAK2 signalling complex deactivation step. Finally, the serine kinase inhibitor H7 blocked down-regulation of GHR^JAK2 signalling to STAT5b in a manner that enabled liver cells to respond to a subsequent GH pulse even without the *3 h interpulse interval normally required for full recovery of GH pulse responsiveness (Gebert et al 1999a). Termination of GH pulse-induced STAT5b signalling thus involves multiple biochemical events: down-regulation of GHR^JAK2 signalling via a cycloheximide- and H7-sensitive step, proteasome-dependent internalization/ degradation of the receptor^kinase signalling complex, and dephosphorylation leading to deactivation of the complex’s STAT5b substrate via the action of a tyrosine phosphatase. Further study is required to elucidate the cellular and biochemical events associated with this down-regulation of STAT5b signalling late in a GH pulse, which helps reset the GHR^JAK^STAT signalling cascade so that the hepatocyte can respond to the next GH pulse.
Mechanism of suppression of liver JAK2^STAT5b pathway by continuous GH The e¡ects of continuous GH treatment on STAT5b activation have been investigated in CWSV-1 cells. These experiments were carried out to model the GH-dependent down-regulation of STAT5b activity that is seen in female rat liver, which experiences a near-continuous stimulation by plasma GH. Continuous GH treatment of CWSV-1 cells decreased by 80^90% the level of phosphotyrosine-STAT5b compared to the peak level activated by a pulse of GH (Gebert et al 1997, 1999b). Moreover, in contrast to the robust STAT5b activation loop which is established by a GH pulse, continuous GH-supported
70
WAXMAN
GHR^JAK2 signalling to STAT5b that is short-lived, depends on the continuous presence of GH for generation of new GHR^JAK2 receptor-kinase complexes, and requires ongoing protein synthesis. Increased activation of STAT5b occurred in cells treated with the proteasome inhibitor MG132, indicating that at least one component of the GHR^JAK2^STAT5b signalling pathway is labile in continuous GH-treated cells (Gebert et al 1999b). Further experiments demonstrated that the tyrosine phosphatase inhibitor pervanadate can fully reverse the down-regulation of STAT5b signalling by continuous GH, indicating that elevated tyrosine phosphatase activity makes an important contribution to this down-regulation response. Studies using the tyrosine kinase inhibitor genistein demonstrated that this phosphatase activity acts both at the level of STAT5b, and at the level of the GHR^JAK2 complex which generates newly activated STAT5b molecules. Moreover, the requirement for continuous GH stimulation and ongoing protein synthesis to maintain the low ‘female level’ of activated STAT5b are both eliminated in pervanadate-treated cells (Gebert et al 1999b), leading us to hypothesize that tyrosine dephosphorylation is an obligatory ¢rst step that ‘gates’ the internalization/degradation pathway for the GHR^JAK2 complex. These ¢ndings suggest a model where continuous GH exposure down-regulates GHR^JAK2 signalling to STAT5b by a mechanism that involves enhanced dephosphorylation of both STAT5b and GHR^JAK2, with the latter step leading to increased internalization/degradation of the receptor^kinase complex. STAT5b and STAT5a gene knockout mouse models A STAT5b gene knockout mouse model has been developed and characterized with the goal of testing the hypothesis that the GH pulse-responsive STAT5b is an important mediator of the e¡ects of GH pulses on male-speci¢c liver gene expression in vivo (Udy et al 1997). These studies led to the exciting discovery that the loss of STAT5b expression leads to a global loss of the GH-regulated, malespeci¢c pattern of liver gene expression. Male-predominant liver gene products are decreased to wild-type female levels in STAT5b7/7 males (Fig. 4, anti-CYP2D9, top panel), while female-predominant liver gene products are increased in STAT5b7/^ males to a level close to that of wild-type female liver (anti-CYP3A, top panel). These responses are in agreement with those predicted by the hypothesis that STAT5b is the major, if not the sole, STAT protein that mediates the sexually dimorphic e¡ects of GH on the liver. Moreover, the ¢nding that female-predominant liver gene products are increased to near female levels in STAT5b7/7 male mice, as also occurs in GH-de¢cient Little mice, lends strong support to Negishi’s earlier proposal that, in the mouse model, femalepredominant liver gene expression can be regulated by the negative e¡ects of
GROWTH HORMONE PULSE REGULATION OF LIVER GENE EXPRESSION
71
FIG. 4. Di¡erential e¡ect of STAT5b knockout compared to STAT5a knockout on GHregulated mouse liver P450 gene products. Shown are Western blots probed with the indicated anti-P450 antibodies. Loss of STAT5b (top) but not STAT5a (bottom) increases the expression in male mouse liver of the female-speci¢c CYP3A, band b. Similarly, STAT5b but not STAT5a is required to maintain the male-speci¢c expression of CYP2D9, band b (lanes 4^6 vs. 1^3, top; lanes 4^7 vs. 1^3, bottom). Data shown are based on Udy et al (1997) and Park et al (1999).
72
WAXMAN
male plasma GH pulses, rather than as a positive response to the female GH pattern per se (Noshiro & Negishi 1986). This situation may contrast to that of the female rat, where continuous GH can apparently exert an additional positive stimulatory e¡ect that may be mediated by a novel, non-STAT-containing GH-regulated nuclear factor termed GHNF (Waxman et al 1996). A further important ¢nding from the STAT5b knockout studies is that the body growth rate pro¢le that is characteristic of males is abolished in STAT5b7/7 males: in the absence of STAT5b, male mice grow during the pubertal and post-pubertal periods at a rate that is considerably lower than STAT5b+/+ males and is quite similar to that of wild-type females and STAT5b7/7 females (Udy et al 1997). STAT5b thus appears to be an obligatory mediator of two biologically important, sexually di¡erentiated responses associated with the sexually dimorphic pattern of pituitary GH secretion: liver gene expression and male body growth rate patterns. Interestingly, although STAT5a is very closely related to STAT5b (490% amino acid sequence identity) and is also expressed in liver, the phenotype of the STAT5b7/7 mice discussed above implies that STAT5a alone is insu⁄cient to maintain normal sexually dimorphic GH responses. Other studies support this conclusion and demonstrate that liver STAT5a is fully dispensable for the malepattern of liver gene expression, as can be seen by analysing liver CYP expression patterns in a STAT5a7/7 mice (Fig. 4, lower panel) (Park et al 1999). Since STAT5a is non-essential for GH pulse-regulated male gene expression, STAT5b homodimers, but not STAT5a^STAT5b heterodimers, probably mediate the sexually dimorphic e¡ects of male GH pulses on liver gene expression. In female mice, however, disruption of either STAT5a or STAT5b results in a striking decrease in several constitutively expressed, female liver P450-catalysed testosterone hydroxylase activities. STAT5a or STAT5b gene disruption also leads to the loss of a female-speci¢c, GH-regulated hepatic CYP2B enzyme (Park et al 1999). These studies indicate that STAT5b, as well as STAT5a, are both required for constitutive expression in female, but not male mouse liver of certain GH-regulated steroid hydroxylase P450 enzymes. The simplest interpretation of these ¢ndings is that STAT5a and STAT5b heterodimerize to regulate expression of certain female-speci¢c, GH-regulated liver gene products. Further study will be required to elucidate at the gene-regulatory level the di¡erential e¡ects that GH-activated STAT5a and STAT5b have on CYP and other sexually dimorphic liver enzymes.
Acknowledgement Work carried out in the author’s laboratory was supported in part by National Institutes of Health grant DK33765.
GROWTH HORMONE PULSE REGULATION OF LIVER GENE EXPRESSION
73
References Agrawal AK, Pampori NA, Shapiro BH 1995 Neonatal phenobarbital-induced defects in ageand sex-speci¢c growth hormone pro¢les regulating monooxygenases. Am J Physiol 268:E439^E445 Argetsinger LS, Campbell GS, Yang X et al 1993 Identi¢cation of JAK2 as a growth hormone receptor-associated tyrosine kinase. Cell 74:237^244 Bergad PL, Shih HM, Towle HC, Schwarzenberg SJ, Berry SA 1995 Growth hormone induction of hepatic serine protease inhibitor 2.1 transcription is mediated by a Stat5-related factor binding synergistically to two gamma-activated sites. J Biol Chem 270:24903^24910 Choi HK, Waxman DJ 1999 Continuous GH, but not prolactin, maintains low-level activation of STAT5a and STAT5b in female rat liver. Endocrinology 140:5126^5135 Cunningham BC, Ultsch M, de Vos AM, Mulkerrin MG, Clauser KR, Wells JA 1991 Dimerization of the extracellular domain of the human growth hormone receptor by a single hormone molecule. Science 254:821^825 Darnell JE Jr 1997 STATs and gene regulation. Science 277:1630^1635 Davey HW, Wilkins RJ, Waxman DJ 1999 STAT5 signalling in sexually dimorphic gene expression and growth patterns. Am J Hum Genet 65:959^965 Ganguly TC, O’Brien ML, Karpen SJ, Hyde JF, Suchy FJ, Vore M 1997 Regulation of the rat liver sodium-dependent bile acid cotransporter gene by prolactin. Mediation of transcriptional activation by Stat5. J Clin Invest 99:2906^2914 Gebert CA, Park SH, Waxman DJ 1997 Regulation of signal transducer and activator of transcription (STAT) 5b activation by the temporal pattern of growth hormone stimulation. Mol Endocrinol 11:400^414 Gebert CA, Park SH, Waxman DJ 1999a Termination of growth hormone pulse-induced STAT5b signaling. Mol Endocrinol 13:38^56 Gebert CA, Park SH, Waxman DJ 1999b Down-regulation of liver JAK2-STAT5b signaling by the female pattern of continuous GH stimulation. Mol Endocrinol 13:213^227 Gouilleux F, Wakao H, Mundt M, Groner B 1994 Prolactin induces phosphorylation of Tyr694 of Stat5 (MGF), a prerequisite for DNA binding and induction of transcription. EMBO J 13:4361^4369 Gronowski AM, Rotwein P 1994 Rapid changes in nuclear protein tyrosine phosphorylation after growth hormone treatment in vivo. Identi¢cation of phosphorylated mitogen-activated protein kinase and STAT91. J Biol Chem 269:7874^7878 Hansen LH, Wang X, Kopchick JJ et al 1996 Identi¢cation of tyrosine residues in the intracellular domain of the growth hormone receptor required for transcriptional signaling and Stat5 activation. J Biol Chem 271:12669^12673 Jansson JO, Albertsson-Wikland K, Ede¤n S, Thorngren KG, Isaksson O 1982 E¡ect of frequency of growth hormone administration on longitudinal bone growth and body weight in hypophysectomized rats. Acta Physiol Scand 114:261^265 Jansson JO, Ede¤ n S, Isaksson O 1985 Sexual dimorphism in the control of growth hormone secretion. Endocrine Rev 6:128^150 Kempe KC, Isom HC, Greene FE 1995 Responsiveness of an SV40-immortalized hepatocyte cell line to growth hormone. Biochem Pharmacol 49:1091^1098 Legraverend C, Mode A, Westin S et al 1992 Transcriptional regulation of rat P-450 2C gene subfamily members by the sexually dimorphic pattern of growth hormone secretion. Mol Endocrinol 6:259^266 Liu X, Robinson GW, Gouilleux F, Groner B, Hennighausen L 1995 Cloning and expression of Stat5 and an additional homologue (Stat5b) involved in prolactin signal transduction in mouse mammary tissue. Proc Natl Acad Sci USA 92:8831^8835
74
WAXMAN
Mui ALF, Wakao H, O’Farrell AM, Harada N, Miyajima A 1995 Interleukin-3, granulocytemacrophage colony stimulating factor and interleukin-5 transduce signals through two STAT5 homologs. EMBO J 14:1166^1175 Noshiro M, Negishi M 1986 Pretranslational regulation of sex-dependent testosterone hydroxylases by growth hormone in mouse liver. J Biol Chem 261:15923^15927 Park SH, Liu X, Hennighausen L, Davey HW, Waxman DJ 1999 Distinctive roles of STAT5a and STAT5b in sexual dimorphism of hepatic P450 gene expression. Impact of Stat5a gene disruption. J Biol Chem 274:7421^7430 Ram PA, Waxman DJ 1997 Interaction of growth hormone-activated STATs with SH2containing phosphotyrosine phosphatase SHP-1 and nuclear JAK2 tyrosine kinase. J Biol Chem 272:17694^17702 Ram PA, Park SH, Choi HK, Waxman DJ 1996 Growth hormone activation of Stat 1, Stat 3, and Stat 5 in rat liver. Di¡erential kinetics of hormone desensitization and growth hormone stimulation of both tyrosine phosphorylation and serine/threonine phosphorylation. J Biol Chem 271:5929^5940 Skett P 1987 Hormonal regulation and sex di¡erences of xenobiotic metabolism. Prog Drug Metab 10:85^139 Subramanian A, Wang J, Gil G 1998 STAT 5 and NF-Y are involved in expression and growth hormone-mediated sexually dimorphic regulation of cytochrome P450 3A10/lithocholic acid 6beta-hydroxylase. Nucleic Acids Res 26:2173^2178 Sundseth SS, Alberta JA, Waxman DJ 1992 Sex-speci¢c, growth hormone-regulated transcription of the cytochrome P450 2C11 and 2C12 genes. J Biol Chem 267:3907^3914 Tannenbaum GS, Martin JB 1976 Evidence for an endogenous ultradian rhythm governing growth hormone secretion in the rat. Endocrinology 98:562^570 Udy GB, Towers RP, Snell RG et al 1997 Requirement of STAT5b for sexual dimorphism of body growth rates and liver gene expression. Proc Natl Acad Sci USA 94:7239^7244 Waxman DJ 1992 Regulation of liver-speci¢c steroid metabolizing cytochromes P450: cholesterol 7alpha-hydroxylase, bile acid 6beta-hydroxylase, and growth hormoneresponsive steroid hormone hydroxylases. J Ster Biochem Molec Biol 43:1055^1072 Waxman DJ, Chang TKH 1995 Hormonal regulation of liver cytochrome P450 enzymes. In: Ortiz de Montellano PR (ed) Cytochrome P450: structure, mechanism, and biochemistry, 2nd edn. Plenum Press, New York, p 391^417 Waxman DJ, Frank SJ 2000 Growth hormone action: signaling via a JAK/STAT-coupled receptor. In: Conn PM, Means A (eds) Molecular regulation. Humana Press, Totowa, NJ, in press Waxman DJ, Pampori NA, Ram PA, Agrawal AK, Shapiro BH 1991 Interpulse interval in circulating growth hormone patterns regulates sexually dimorphic expression of hepatic cytochrome P450. Proc Natl Acad Sci USA 88:6868^6872 Waxman DJ, Ram PA, Park SH, Choi HK 1995 Intermittent plasma growth hormone triggers tyrosine phosphorylation and nuclear translocation of a liver-expressed, Stat 5-related DNA binding protein. Proposed role as an intracellular regulator of male-speci¢c liver gene transcription. J Biol Chem 270:13262^13270 Waxman DJ, Zhao S, Choi HK 1996 Interaction of novel sex-dependent, growth hormoneregulated liver nuclear factor with CYP2C12 promoter. J Biol Chem 271:29978^29987 Yamashita H, Xu J, Erwin RA, Farrar WL, Kirken RA, Rui H 1998 Di¡erential control of the phosphorylation state of proline-juxtaposed serine residues Ser725 of Stat5a and Ser730 of Stat5b in prolactin-sensitive cells. J Biol Chem 273:30218^30224 Yi W, Kim SO, Jiang J et al 1996 Growth hormone receptor cytoplasmic domain di¡erentially promotes tyrosine phosphorylation of signal transducers and activators of transcription 5b and 3 by activated JAK2 kinase. Mol Endocrinol 10:1425^1443
GROWTH HORMONE PULSE REGULATION OF LIVER GENE EXPRESSION
75
DISCUSSION Veldhuis: I’m not fully clear what the continuous GH signal is doing. I understood you to say that STAT5b has a rapid turnover in the presence of continuous GH. What does that re£ect? Activation of a phosphatase? Waxman: We have recently demonstrated enhanced phosphatase activity at the level of the GHR^JAK2 kinase complex in liver cells treated with GH continuously (Gebert et al 1999). Treatment of these cells with the phosphotyrosine phosphatase inhibitor pervanadate increases the activated STAT5b signal from the 10% level found in continuous GH-treated cells back to a 100% signal within a few minutes, even if GH was no longer present in the system. Our interpretation of this result is that a more e⁄cient dephosphorylation of the GHR^JAK2 kinase complex becomes established within two hours of continuous GH stimulation, and that pervanadate treatment enables the receptor^kinase complex to accumulate in its active (phosphorylated) form. Veldhuis: Are you taking this back to a residence time issue? When GH is bound to the receptor for two hours or more, do you then get this phosphatase activation, or is there something else happening here? Waxman: On the basis of the simpli¢ed GH signalling pathway slide that I presented, many things could happen. It is quite possible that with chronic STAT5b activation, a signalling pathway is engaged that leads to stimulation of phosphatase activity. Another possibility is that a SOCS or CIS protein is induced by GH-activated STAT5b, and this SOCS or CIS protein inhibits further signalling by the GHR^JAK2 complex (Ram & Waxman 1999). SOCS/CIS mRNAs are induced by GH with di¡erent kinetics and exhibit markedly di¡erent apparent stabilities: some SOCS/CIS mRNAs have an apparent turnover of less than 30 min and others are maintained in the cell for many hours following the initial cytokine or GH pulse. One model that we should consider is that continuous GH stimulation, but not GH pulses, may activate expression of a SOCS/CIS gene that turns o¡ signalling by the GHR^JAK2 kinase complex. Brabant: Could this mean that it is the nadir which matters more than the pulse? Waxman: Is the nadir important? Clearly the GH o¡-time is critical. We learned that from the studies carried out in hypophysectomized rats, where the frequency of GH pulse stimulation was shown to be critical for inducing a male pattern of liver P450 gene expression (Waxman et al 1991). In that model, six GH pulses per day gave a full biological response with respect to expression of CYP 2C11, a maleexpressed rat liver P450 gene, but seven GH pulses per day were ine¡ective. On the basis of these studies we concluded that a critical GH ‘o¡-time’ was required for appropriate signalling to stimulate male liver gene expression. This point is further emphasized by a simple comparison of the plasma GH pro¢les of male versus female mice. In mice, in both sexes there is a nadir, when plasma GH is
76
DISCUSSION
essentially undetectable, but the nadir is of shorter duration in the female. This di¡erence alone is apparently su⁄cient to impart the observed dramatic sex di¡erences in GH-regulated mouse liver gene expression. We have proposed that when the ‘GH o¡-time’ is too short, there’s insu⁄cient time for the system to re-set itself and to respond appropriately to a subsequent plasma GH pulse. Matthews: Are these ¢ndings consistent with gigantism in humans, where GH pulses are about the same amplitude in gigantism individuals as they are in normals, but the nadir in these people never comes at all (Matthews et al 1991)? In other words, these individuals have continuous exposure to GH, similar to female rats. You would expect this plasma GH pattern to down-regulate GH signalling, and in fact it doesn’t it works in the opposite direction. Veldhuis: It is not quite the same, but we ¢nd the same theme in acromegaly; there, the single best correlate of high IGF-I concentrations, and presumably of the adverse tissue events that are associated with acromegaly, is the mean interpeak serum nadir GH concentration (Hartman et al 1990, 1994). Matthews: It works in the opposite direction. Veldhuis: Yes, it works in the opposite direction in the human. Sassone-Corsi: In terms of activating STAT5, is it clear whether the time between the pulses is crucial? Waxman: Yes, it is. In the in vivo hypophysectomized rat model that I described a few moments ago, six GH pulses per day, or fewer than six pulses, worked equally well as physiological GH replacement, whereas more frequent pulsation, i.e. seven GH pulses a day, was without e¡ect. Sassone-Corsi: I was wondering whether ¢ve pulses in two hours is the same thing as ¢ve pulses in 10 hours. Is the time between pulses critical? Waxman: We haven’t modelled every possible parameter, but from our studies of plasma GH pulse width, height and frequency, it’s clear that pulse frequency is the most important determinant for signalling a male pro¢le of liver gene expression (Waxman et al 1991). We’ve now modelled this in more detail in cell culture, where we looked directly at the GH pattern dependence of STAT5b activation. In that model, we ¢nd that unless we wait about two-and-a-half hours, which more-or-less corresponds to the GH o¡-time in the male, we don’t regain the full GH pulse responsiveness that characterizes the initial pulse (Gebert et al 1997). Again, the take home message seems to be that the GH o¡-time is most important. Sassone-Corsi: In the STAT5b knockout mice, as in many other gene knockout studies, the mice are chronically missing the STAT gene product in utero and after birth. You are therefore looking at the consequences that may have accumulated throughout the development of the animal, and so you may not be able to compare that situation to one in which the function of this gene is suddenly missing, as it would be in a conditional knockout.
GROWTH HORMONE PULSE REGULATION OF LIVER GENE EXPRESSION
77
Waxman: There are certainly many caveats that apply to the interpretation of our STAT5b knockout studies, as is the case for all targeted gene disruptions. Of note, however, the slower pubertal body growth phenotype that we reported for male STAT5b-disrupted mice doesn’t emerge until puberty. In addition, the male GH pulse-regulated genes that we have been studying are not activated until puberty. Moreover, until that point in time the animals seem to be healthy and essentially normal. Our supposition, therefore, is that the pubertal phenotypes that emerge are not re£ective of some developmental defect that occurred neonatally or in utero, but rather are the result of defects that emerge at the point where the plasma GH pulsatility begins, which is at puberty. Sassone-Corsi: You have nicely shown that STAT5a levels don’t seem to change in the STAT5b knockouts. Does the activation and nuclear translocation of STAT5a still occur in the STAT5b knockout mice, i.e. even in the absence of STAT5b? Waxman: Yes. We have demonstrated this biochemically. Sassone-Corsi: Do they dimerize in the nucleus? Waxman: STAT5a and STAT5b can both homodimerize and heterodimerize upon tyrosine phosphorylation, but the dimerization and activation of STAT5a is not dependent on the presence of STAT5b. Sassone-Corsi: Are the JAK kinases also working ¢ne in the knockouts? Waxman: There’s no reason to think that they wouldn’t. We haven’t looked at this directly, but the fact that GH activates STAT5a in the same STAT5b knockout mouse strongly suggests that JAK2 kinase and its regulation are unimpaired. Copinschi: What would be the e¡ect on GH of manipulations of the circulating concentrations of sex steroids, for instance? Waxman: Many early studies investigated the roles of androgens and oestrogens in establishing adult plasma GH patterns. The general conclusion that has emerged is that neonatal androgen exposure is a critical factor in establishing the hypothalamic regulation of pituitary secretory patterns that become manifest later in life, at puberty. In addition, there appears to be a reversible e¡ect androgens and oestrogens in the post-pubertal period in maintaining the full level of pituitary GH secretion. However, we haven’t tried to manipulate the levels of sex steroids in the STAT5b-de¢cient animals to see whether they have any rescue e¡ects. My prediction, however, would be that androgens or oestrogens would be ine¡ective in this respect. I would say this because early studies clearly demonstrate that for liver responses to gonadal hormones to occur, androgens and oestrogens must act indirectly, via the pituitary^GH axis. If male-speci¢c liver gene expression is indeed mediated by STAT5b, then you would predict that androgens, for example, would not be able to circumvent the absence of this STAT transcription factor.
78
DISCUSSION
Copinschi: I remember older studies in which male humans who were given oestrogens for a couple of weeks responded in the same way as females in certain stimulation tests. Robinson: It is easy to run away with the idea that female rats don’t grow. There is quite a di¡erence between the growth rates of an intact female and a hypophysectomized female rat. There is a signi¢cant amount of GH stimulation of growth by whatever pattern; whether it is continuous or frequent pulsatility. This is also true for humans. There’s no real problem with acromegaly, or with transgenic animals that have continuous GH exposure and show gigantism: continuous GH is clearly a biologically e¡ective signal. The interest is that it may not be inducing some of the sexually dimorphic signals; this is the point I think your studies bring out. How do these knockout animals respond to continuous GH? Are they in fact GH insensitive? You alluded to the possibility that because of the feedback problem, they may actually be producing a female continuous pattern anyway. Have you done the other control, which is to do a continuous GH replacement and look at GH sensitivity, what might happen to IGF-I and whether you get some peculiar e¡ects on liver enzymes you might not expect from the simple model? Waxman: Those are excellent ideas that we will most certainly want to pursue. Robinson: I would like to press you on the relevance of a continuous signal for growth. The IGF-I levels are up. They may be related to continuous GH, but you wouldn’t expect high IGF-I with GH insensitivity. Continuous GH is a very good signal for a lot of other hepatic targets. Waxman: First, let me correct one point. I mentioned that IGF-I levels are actually decreased in the STAT5b-de¢cient mice (Udy et al 1997). As you mention, the STAT5b knockout animals do grow. The female animals grow essentially normally, while the males grow normally until puberty, but thereafter grow at the slower, female rate. By contrast, the hypophysectomized animals don’t grow at all, but if you give them continuous GH they grow at their normal (female) rates. We are not implying that growth per se is exclusively dependent on the presence of a functional STAT5b protein, but rather that STAT5b is required for one of the two main components of body growth, namely that component which responds to GH pulses at puberty and gives rise to the male pubertal growth spurt. Thus, there is a GH response that is independent of STAT5b, and there is the additional enhanced long bone growth that is stimulated by pulses of GH (seen primarily in the males), which is STAT5b dependent. Wu: You said that the male/female GH secretion pattern is not di¡erent until puberty. Has the experiment been done in which you neonatally castrate the animal and give them sex steroids to convert them?
GROWTH HORMONE PULSE REGULATION OF LIVER GENE EXPRESSION
79
Waxman: Yes, we carried out that experiment many years ago. We found that if you castrate rats at birth, and then give them a replacement dose of androgen neonatally, on days 1 and 3 of life, and then no further sex hormone treatment, then the neonatal androgen replacement is su⁄cient to restore at least 50% of the normal expression of male-speci¢c P450 enzymes at puberty (Waxman et al 1985). This e¡ect of neonatal androgen exposure can be observed both in castrated males and in ovarectomized females. Wu: Are you looking at the liver response in these studies? Waxman: The e¡ects of neonatal androgen on liver gene expression are actually mediated by the e¡ects of this androgenic exposure on the postnatal development of the hypothalamus, which in turn leads to a masculinization of pituitary GH secretion. The general conclusion from these types of experiments is that androgenic imprinting, which in the rat takes place in the ¢rst week of life, is su⁄cient to establish the GH secretory pattern that later emerges at puberty. This response has been termed neonatal androgenic imprinting. Ede¤ n: You can turn a male rat into a female by oestrogens, as measured by GH pattern. Some part of this is the imprinting part, but you can actually do it in a week. Waxman: If you treat with gonadal steroids during adulthood, then the e¡ects are generally reversible once the steroids are removed, whereas when the animals are exposed to androgens neonatally then the e¡ect is essentially irreversible. Clarke: You’re sexually di¡erentiating the brain, surely. In a rodent you only need to give one neonatal injection of androgen and you have got a male brain forever. Lightman: Since all these changes take place at puberty, when you’re talking about male- or female-speci¢c liver enzymes, are they really male or female speci¢c, or are they secondary to the changes over a prolonged period in the pulsatile nature of GH secretion? If you have a female on which you are superimposing a male pattern of GH secretion, will the enzymes in the liver be male or female? Waxman: Clearly the liver enzyme pattern will be male. If you give a female rat the male pattern of plasma GH, either by manipulation of the sex steroids, e.g. by androgenic imprinting or by hypophysectomy followed by an imposition of the male plasma GH pattern, then the rat will exhibit a clear male pattern of liver gene expression. Herbison: You began by showing the GH actions on liver, where there are some nice feedback e¡ects. There is a line of thought that GH feedback e¡ects are themselves involved in the sexually dimorphic patterns of the hypothalamus and pituitary secretions. How speci¢c is this STAT5b phenomena to liver cells? Is there STAT5b signalling in other tissues?
80
DISCUSSION
Waxman: It is likely that there is. STAT5b is expressed in many tissues. In the STAT5b knockout mice, the STAT5b gene is globally disrupted and the expression of STAT5b protein is lost in all cells. It would be interesting for someone to look at the e¡ect of the absence of STAT5b on feedback mechanisms induced by GH occurring in the hypothalamus, because in some of those feedback mechanisms STAT5b itself might be involved. Veldhuis: I would also love to see Iain Robinson test GH feedback in the STAT5b knockout. Robinson: Actually, it’s hard to show pattern sensitivity of feedback. But, where people have done pulses of GH to entrain the animals, it is possible to phase-lock the generator if you give them GH at the appropriate frequency. On the other hand, if you give continuous exposure to high levels of GH, the net result is that the hypothalamic generation of the pulsatility is shut down. To this extent, there does seem to be a central pattern dependence of that phenomenon. Whether or not it is mediated by STAT5b, we shall have to see. De Meyts: It seems to me that in order to understand signalling responses to the pulsatile GH secretion pattern, it is going to be essential to understand the nature and kinetics of negative feedback loops induced in the system. The way we have been looking at signalling networks so far has been very much as a sort of forwardly-active branching type of pathway, especially kinases, operating in some cases against a background of constitutively active phosphatases. Therefore we wouldn’t expect to introduce any kind of responses to pulsatility. Now, as we are beginning to realize that some of the negative feedback loops are in fact themselves signal generated, it becomes obvious that there must be some optimal pattern of signal generation. Veldhuis: David Brown, do you want to comment on any of this matter of timedelayed feedback, which seems to be critical to any oscillatory system? Brown: It is well known in many systems that delayed feedback will result in oscillations. There a many possible di¡erences, depending on the exact way in which feedback occurs. In some neuronal systems, for example the FitzHugh^ Nagumo model, which is that is thought of as an archetypal example of excitability or of an intrinsic oscillator, this does behave rather di¡erently in some important ways from the classical Hodgkin^Huxley model, which it is meant to be a simpli¢cation of.
References Gebert CA, Park SH, Waxman DJ 1997 Regulation of signal transducer and activator of transcription (STAT) 5b activation by the temporal pattern of growth hormone stimulation. Mol Endocrinol 11:400^414
GROWTH HORMONE PULSE REGULATION OF LIVER GENE EXPRESSION
81
Gebert CA, Park SH, Waxman DJ 1999 Down-regulation of liver JAK2-STAT5b signaling by the female pattern of continuous GH stimulation. Mol Endocrinol 13: 213^227 Hartman ML, Veldhuis JD, Vance ML, Faira ACS, Furnaletto RW, Thorner MO 1990 Somatotropin pulse frequency and basal growth hormone concentrations are increased in acromegaly. J Clin Endocrinol Metab 70:1375^1384 Hartman ML, Pincus SM, Johnson NL et al 1994 Enhanced basal and disorderly growth hormone secretion distinguish acromegalic from normal pulsatile growth hormone release. J Clin Investig 94:1277^1288 Matthews DR, Hindmarsh PC, Pringle PJ, Brook CGD 1991 A distribution method for analysing the baseline of pulsatile endocrine signals as exempli¢ed by 24-hour growth hormone pro¢les. Clin Endocrinol 35:245^252 Ram PA, Waxman DJ 1999 SOCS/CIS protein inhibition of growth hormone-stimulated STAT5 signaling by multiple mechanisms. J Biol Chem, in press Udy GB, Towers RP, Snell RG et al 1997 Requirement of STAT5b for sexual dimorphism of body growth rates and liver gene expression. Proc Natl Acad Sci USA 94:7239^7244 Waxman DJ, Dannan GA, Guengerich FP 1985 Regulation of rat hepatic cytochrome P-450: age-dependent expression, hormonal imprinting, and xenobiotic inducibility of sex-speci¢c isoenzymes. Biochemistry 24:4409^4417 Waxman DJ, Pampori NA, Ram PA, Agrawal AK, Shapiro BH 1991 Interpulse interval in circulating growth hormone patterns regulates sexually dimorphic expression of hepatic cytochrome P450. Proc Natl Acad Sci USA 88:6868^6872
Orderliness of hormone release Steven M. Pincus 990 Moose Hill Road, Guilford, CT 06437, USA
Abstract. ApEn, approximate entropy, is a recently formulated family of parameters and statistics quantifying regularity (orderliness) in serial data, with developments both within theoretical mathematics, as well as numerous applications to multiple biological contexts. ApEn appears to have broad application to hormone pulsatility analysis within endocrinology, bringing a new perspective to the assessment of secretory patterns. ApEn is complementary to pulse detection algorithms widely employed to evaluate hormone secretion time-series it is scale-invariant and model-independent, evaluates both dominant and subordinant patterns in data, discriminates series for which clear pulse recognition is di⁄cult, and often provides a direct barometer of feedback between subsystems. ApEn is applicable to systems with at least 50 data points and to broad classes of models: it can be applied to discriminate both general classes of correlated stochastic processes, as well as noisy deterministic systems. Moreover, ApEn is complementary to spectral and autocorrelation analyses, providing e¡ective discriminatory capability in instances in which the aforementioned measures exhibit minimal distinctions. We present some basic background on the above, and illustrate various facets of ApEn utility via several representative endocrinological studies. 2000 Mechanisms and biological signi¢cance of pulsatile hormone secretion. Wiley, Chichester (Novartis Foundation Symposium 227) p 82^104
Series of sequential data arise throughout biology in multifaceted contexts, such as hormonal secretory dynamics, heart rate rhythms, EEGs and DNA sequences. Enhanced capabilities to quantify di¡erences among such series would be quite valuable, since in their respective contexts, these series re£ect essential biological information. Although researchers typically quantify mean levels, and often the extent of variability, it is recognized that in many instances, the persistence of certain patterns, or shifts in an ‘apparent ensemble amount of randomness’, provide the fundamental insight of subject status. Despite this recognition, formulas and algorithms to quantify an ‘extent of randomness’ have not been developed or utilized in the above contexts, primarily since even within mathematics itself, such a quanti¢cation technology was lacking until very recently. Thus except for the settings in which egregious (changes in) serial features presented themselves, which specialists are trained to visually detect, 82
ORDERLINESS OF HORMONE RELEASE
83
subtler changes in patterns would largely remain undetected, unquanti¢ed, and/or not acted upon. Recently, a new mathematical approach and formula, Approximate Entropy (ApEn), has been introduced as a quanti¢cation of regularity of data, motivated by both the above application needs (Pincus 1991), and by fundamental questions within mathematics (Pincus & Singer 1996, Pincus & Kalman 1997). This approach calibrates an ensemble extent of sequential interrelationships, quantifying the continuum that ranges from totally ordered to completely random. The central focus of this paper is to discuss ApEn, and subsequently cross-ApEn (Pincus & Singer 1996, Pincus et al 1996a), a measure of twovariable asynchrony that is thematically similar to ApEn. Before presenting a detailed discussion of regularity, we consider a set of timeseries to illustrate what we are trying to measure. In Fig 1, the data in (a^f) represent time-series of growth hormone (GH) levels from rats in six distinct physiological settings, each taken at 10 min samples during a 10 h lights o¡ (‘dark’) state (Gevers et al 1998). The endpoints (a) and (f) depict, respectively, intact male and intact female serum dynamics; (b) and (c) depict two types of neutered male rats; while (d) and (e) depict two classes of neutered female rats. It appears that the time-series are becoming increasingly irregular as we proceed from (a) to (f), although speci¢c feature di¡erences among the sets are not easily pinpointed. We ask (i) how do we quantify the apparent di¡erences in regularity?; (ii) do the regularity values signi¢cantly distinguish the data sets?; (iii) how do inherent limitations posed by moderate length time-series, with noise and measurement inaccuracy present as in Fig. 1, a¡ect statistical analyses?; (iv) is there some general mechanistic hypothesis, applicable to diverse contexts, that might explain such regularity di¡erences? As stated above, we also discuss cross-ApEn (Pincus & Singer 1996, Pincus et al 1996a), a quanti¢cation of asynchrony or conditional irregularity between two signals. Cross-ApEn is thematically and algorithmically quite similar to ApEn, yet with a critical di¡erence in focus: it is applied to two time-series, rather than a single series, and thus a¡ords a distinct tool from which changes in the extent of synchrony in interconnected systems or networks can be directly determined. This quanti¢cation strategy is thus especially germane to many biological feedback and/ or control systems and models for which cross-correlation and cross-spectral methods fail to fully highlight markedly changing features of the data sets under consideration. Quanti¢cation of regularity To quantify irregularity, we utilize approximate entropy, ApEn, a modelindependent statistic de¢ned in Pincus (1991), with further mathematical properties and representative biological applications given in Kaplan et al (1991),
84
PINCUS
ORDERLINESS OF HORMONE RELEASE
85
Pincus et al (1991, 1993), Pincus & Viscarello (1992), Ryan et al (1994), Pincus & Singer (1996), Pincus & Kalman (1997). ApEn is complementary to pulse detection algorithms widely employed to evaluate hormone secretion time-series (Urban et al 1988). ApEn evaluates both dominant and subordinant patterns in data; notably, it will detect changes in underlying episodic behaviour not re£ected in peak occurrences or amplitudes (Pincus & Keefe 1992). Additionally, ApEn provides a direct barometer of feedback system change in many coupled systems (Pincus 1994, Pincus & Keefe 1992). Among numerous endocrine applications to hormone secretion time-series data based on as few as N ¼60 points, ApEn has shown vivid distinctions (P 5 10710; nearly 100% sensitivity and speci¢city in each study) between normal and tumour-bearing subjects for GH (Hartman et al 1994), adrenocorticotropic hormone (ACTH) and cortisol (Van den Berg et al 1997), and aldosterone (Siragy et al 1995), with the tumorals markedly more irregular; a pronounced and consistent gender di¡erence in GH irregularity in both human and rat (Pincus et al 1996b); highly signi¢cant di¡erences between follicle-stimulating hormone (FSH) and luteinizing hormone (LH) in both sheep (Pincus et al 1998) and in human women and men (Pincus et al 1997); and a positive correlation between advancing age and each of greater irregularity of (i) GH (Veldhuis et al 1995) and of (ii) LH and testosterone (Pincus et al 1996a). ApEn assigns a non-negative number to a time-series, with larger values corresponding to greater apparent process randomness or serial irregularity, and smaller values corresponding to more instances of recognizable features or patterns in the data. Two input parameters, a run length m and a tolerance window r, must be speci¢ed to compute ApEn. Brie£y, ApEn measures the logarithmic likelihood that runs of patterns that are close (within r) for m contiguous observations remain close (within the same tolerance width r) on next incremental comparisons; the precise mathematical de¢nition is given in Pincus (1991). The opposing extremes are perfectly regular sequences, (e.g. sinusoidal behaviour, very low ApEn), and independent sequential processes (very large ApEn). It is imperative to consider ApEn(m,r) as a family of parameters; comparisons are intended with ¢xed m and r. For the studies described herein, we calculated ApEn values for all data sets with m ¼1 or m ¼2 and r a ¢xed percentage of the standard deviation (SD) of the individual subject time-series, typically r ¼20% SD. Normalizing r to each time-series SD gives ApEn a translation- and scale-invariance to absolute serum
FIG. 1. (Opposite) Representative serum growth hormone (GH) concentration pro¢les, in ng/ ml, measured at 10 min intervals for 10 h in the dark. From, in ascending order of ApEn values (hence increasing irregularity or disorderliness): (a) intact male; (b) triptorelin-treated male; (c) gonadectomized male; (d) ovariectomized female; (e) triptorelin-treated female; and (f) intact female rats.
86
PINCUS
concentration levels (Pincus et al 1993), in that it remains unchanged under uniform process magni¢cation, reduction, or constant shift higher or lower. Additionally, ApEn is nearly una¡ected by noise of magnitude below r, the de facto ¢lter level; and it is robust or insensitive to artifacts or outliers. Multiple previous studies that included both theoretical analysis (Pincus 1991, Pincus & Huang 1992, Pincus & Goldberger 1994) and clinical applications (in Pincus et al 1991, 1993, Pincus & Viscarello 1992, Hartman et al 1994, Siragy et al 1995, Veldhuis et al 1995, Van den Berg et al 1997) have demonstrated that the choice of m ¼1 and r ¼20% SD as input parameters (above) produce good statistical reproducibility for ApEn for time-series of lengths N 5 50. To develop a more intuitive, physiological understanding of the ApEn de¢nition, a multistep description of its typical algorithmic implementation, with ¢gures, is developed in Pincus & Goldberger (1994). Further technical discussion of mathematical and statistical properties of ApEn, including robustness to noise and artifacts, mesh interplay, relative consistency of (m,r) pair choices, asymptotic normality under general assumptions, statistical bias, and error estimation for general processes can be found elsewhere (Pincus & Huang 1992, Pincus & Goldberger 1994). Representative endocrinological applications ApEn has recently been applied to numerous settings both within and outside endocrinology. We next discuss brie£y the gender di¡erence ¢ndings in GH, to further develop intuition for ApEn in an application context. Sample application: gender di¡erences in GH serum dynamics In two distinct human subject studies (employing, respectively, immunoradiometric assays and immuno£uorimetric assays), females exhibited signi¢cantly greater irregularity than their male counterparts (P 5 0.001 in each setting), with almost complete gender segmentation via ApEn in each context (Pincus et al 1996b). ApEn likewise sharply discriminates male and female GH pro¢les in the adult intact rat (Gevers et al 1998, Pincus et al 1996b) (P 5 1076, with nearly 100% sensitivity and speci¢city; Fig. 2A). Remarkably, in rats that had been castrated prior to puberty, the ApEn of GH pro¢les in later adulthood is able to separate genetically male and female animals (Gevers et al 1998). Among intact animals and rats treated prepubertally either with a long-acting gonadotropin-releasing hormone (GnRH) agonist or surgical castration, the following rank order of ApEn of GH release emerged, listed from maximally irregular to maximally regular: intact female, GnRH-agonist-treated female, ovariectomized female, orchidectomized male, GnRH-agonist-treated male and intact male (Gevers et al
ORDERLINESS OF HORMONE RELEASE
87
FIG. 2. Scatterplots of mean serum GH concentrations (ng/ml) vs. ApEn(1, 20% SD) values in (A) individual intact male and female rats; and (B) in surgically gonadectomized (gnx) or pharmocologically neutered (GnRH agonist triptorelin treatment) male and female rats.
1998), illustrated in Fig. 1. ApEn was highly signi¢cantly di¡erent between the pooled groups of neutered females and neutered males (P 5 1074, con¢rmed visually in Fig. 2B). More broadly, this application to the rat studies indicates the clinical utility of ApEn. ApEn both agrees with intuition, con¢rming di¡erences that are visually
88
PINCUS
‘obviously distinct’, as in the comparisons in Fig. 1 (a) and (f), intact males vs. females; and importantly, can uncover and establish graded, often subtle distinctions, as in comparisons of Fig. 1(b^e), the neutered subject time-series. Furthermore, these analyses accommodated both a point-length restriction of N ¼60 samples (10 h dark period, 10 min sampling protocol) and a typically noisy environment (due to assay inaccuracies and related factors), representative of the types of constraints that are usually present in clinical and laboratory settings. A mechanistic hypothesis for altered regularity It seems important to determine a unifying theme suggesting greater signal regularity in a diverse range of complicated neuroendocrine systems. We hardly expect a single mathematical model, or even a single family of models, to govern a wide range of systems; furthermore, we would expect that in vivo, each physiological signal would usually represent the output of a complex, multinodal network with both stochastic and deterministic components. Our mechanistic hypothesis is that in a variety of systems, greater regularity (lower ApEn) corresponds to greater component and subsystem autonomy. This hypothesis has been mathematically established via analysis of several very di¡erent, representational (stochastic and deterministic) mathematical model forms, conferring a robustness to model form of the hypothesis (Pincus & Keefe 1992, Pincus 1994). Restated, ApEn typically increases with greater system coupling and feedback, and greater external in£uences, thus providing an explicit barometer of autonomy in many coupled, complicated systems. Many endocrine hormone ¢ndings, including those indicated above, suggest that hormone secretion pathology usually corresponds to greater signal irregularity. Accordingly, a possible mechanistic understanding of such pathology, given this hypothesis, is that healthy, normal endocrine systems function best as relatively closed, autonomous systems (marked by regularity and low ApEn values), and that accelerated feedback and too many external in£uences (marked by irregularity and high ApEn values) corrupt proper endocrine system function. Cross-ApEn Cross-ApEn is a measure of asynchrony between two time-series (Pincus & Singer 1996, Pincus et al 1996a). As for ApEn, it is a two parameter family of statistics, with m and r taking the same meaning as in the ApEn setting, herein ¢xed for application to the paired time-series {u(i)},{v(i)}. Cross-ApEn measures, within tolerance r, the (conditional) regularity or frequency of v patterns similar to a
ORDERLINESS OF HORMONE RELEASE
89
FIG. 3. Plasma concentrations of ACTH (dotted line) and cortisol (continuous line) in a female patient with Cushing’s disease (upper panel) and a control subject (lower panel), each sampled at 10 min intervals for 24 h.
given u pattern of window length m. It is typically applied to standardized u and v time-series, with input parameter speci¢cations m ¼1, r ¼20% SD. Greater asynchrony indicates fewer instances of (sub)pattern matches, quanti¢ed by larger cross-ApEn values. Figure 3, taken from a recent study of paired ACTH^ cortisol dynamics in Cushing’s disease (Roelfsema et al 1998), illustrates the crossApEn quanti¢cation, with greater ACTH^cortisol secretory asynchrony in the diseased subject, compared to the control. Cross-ApEn is generally applied to compare sequences from two distinct yet intertwined variables in a network. Thus we can directly assess network, and not just nodal, evolution, under di¡erent settings e.g. to evaluate uncoupling, and/ or changes in feedback and control. Hence, cross-ApEn facilitates analyses of
90
PINCUS
output from myriad complicated networks, avoiding the requirement to model the underlying system. This is especially important, since accurate modelling of (biological) networks is often nearly impossible even full description of all system nodes and pathways is typically unknown in most biological systems, to say nothing of subsequent good mathematical approximations of the resultant inter-network dynamics. The key point, similarly for ApEn, is that full model speci¢cation is not required to realize an e¡ective discrimination strategy. Furthermore, of course, there is a paucity of general multivariate time-series statistical tools. The precise de¢nition, given in Pincus & Singer (1996), De¢nition 5, is thematically similar to that for ApEn. As a representative example of application of cross-ApEn to biological data, we now consider the following study. LH^testosterone study, males In recent years, there has been considerable study of LH and testosterone (T) serum concentration time-series in younger and older males, both to better understand the physiology of reproductive capacity and in assessing e.g. a loss of libido, or decreased reproductive performance. Furthermore, there is considerable interest in determining whether a hypothesized male climacteric (so-called andropause) at least partially analogous to menopause in the female exists, and if so, in what precise sense. While considerable insight has already been gained, there remain non-trivial controversies concerning several classes of ¢ndings, including primary determinations of whether overall mean levels of LH and T decrease with increasing age. A study was performed recently to determine possible secretory irregularity shifts with ageing within the LH^T axis (Pincus et al 1996a). Serum concentrations were derived for LH and T in 14 young (21^34 years) and 11 older (62^74 years) healthy men. For each subject, blood samples were obtained at frequent (2.5 min) intervals during a sleep period, with an average sampling duration of 7 h. Although mean (and SD) of LH and T concentrations were indistinguishable in the two age groups, for each of LH and T, older males have consistently and highly signi¢cantly more irregular serum reproductive-hormone concentrations than younger males: for LH, aged subjects had greater ApEn values (1.5250.221) than younger individuals (1.2070.252), P 5 0.003, while for testosterone, aged subjects had greater ApEn values (1.6220.120) than younger counterparts (1.3840.228), P 5 0.004. Probably a yet mechanistically more important ¢nding in this study was seen via cross-ApEn analysis. Cross-ApEn was applied to the paired LH^T time-series; statistically, even more vividly than for the irregularity (ApEn) analyses, older subjects exhibited greater cross-ApEn values (1.9610.121) compared to
ORDERLINESS OF HORMONE RELEASE
91
FIG. 4. Individual subject cross-ApEn values versus cross-correlation (Pearson r), applied to the joint LH^testosterone time-series in healthy young versus older men.
younger subjects (1.5740.249) (P 5 1074, with nearly 100% sensitivity and speci¢city), indicating greater LH^T asynchrony in the older group (Fig. 4). Moreover and notably, no signi¢cant LH^T linear correlation (Pearson r) di¡erences were found between the younger and older cohorts, P 4 0.62 (Fig. 4). Several possibilities for the source of the erosion of LH^T synchrony are discussed (Pincus et al 1996a), although a determination of this source awaits future study. Mechanistically, the results implicate (LH^T) network uncoupling as marking male reproductive ageing, which we now have several quanti¢able means to assess. As another example of cross-ApEn utility, in a study of 20 Cushing’s disease patients vs. 29 controls (Roelfsema et al 1998), cross-ApEn of ACTH^cortisol was greater in patients (1.6860.051) than in controls (1.0770.039) (P 5 10715, with nearly 100% sensitivity and speci¢city), suggesting compromise of hormonal pathways and feedback control in diseased subjects, atop that previously seen for more localized nodal secretory dynamics of each hormone individually (Van den Berg et al 1997). Figure 3 displays representative serum pro¢les from this study. Additionally, healthy men and women showed progressive erosion of bihormonal ACTH^cortisol synchrony with increased ageing via cross-ApEn, similar to the LH^T erosion of synchrony in men noted above, suggesting that increased cross-ApEn (greater asynchrony) of paired secretory dynamics is a ubiquitous phenomenon with advancing age.
92
PINCUS
Complementarity of ApEn and cross-ApEn to correlation and spectral analyses Mathematically, the need for ApEn, and particularly for cross-ApEn, is clari¢ed by considering alternative parameters that might address similar concepts. In comparing two distinct signals or variables, primary parameters that one might employ include the cross-correlation function (e.g. Pearson r), and the crossspectrum (Chat¢eld 1989), with single variable counterparts, the autocorrelation function and the power spectrum. Evaluation of these parameters often is insightful, but with relatively small data sets, statistical estimation issues are nontrivial and, moreover, interpretation of the sample cross-correlation function is highly problematic, unless one utilizes a model-based pre¢ltering procedure (Chat¢eld 1989). Most importantly, the autocorrelation function and power spectrum, and their bivariate counterparts, are most illuminating in linear systems, e.g. ARIMA (autoregressive integrated moving average) models, for which a rich theoretical development exists (Box & Jenkins 1976). For many other classes of processes, these parameters often are relatively ine¡ective at highlighting certain model characteristics, even apart from statistical considerations. To illustrate this point, consider the following simple model, denoted as a ‘variable lag’ process: this consists of a series of quiescent periods, of variable length duration, interspersed with identical positive pulses of a ¢xed amplitude and frequency. Formally, recursively de¢ne an integer time-valued process denoted VarLag whose ith epoch consists of (a quiescent period of) values¼0 at times ti1 +1, ti1 + 2, . . ., ti1 + lagi, immediately followed by the successive values sin(p/6), sin(2p/ 6), sin(3p/6), sin(4p/6), sin(5p/6), sin(6p/6)¼0 at the next 6 time-units, where lagi is a random variable uniformly distributed on (randomly chosen between) the integers between 0 and 60, and ti1 denotes the last time-value of the (i1)st sine-pulse. Figure 5A displays representative output from this process, with Fig. 5B a closer view of this output near time t¼400. The power spectrum and autocorrelation function calculations shown in Figs 5C and 5E were calculated from a realization of length N ¼100 000 points. Processes consisting of alternating quiescent and active periods would seem reasonable for biologists to consider, as they appear to model a wide variety of phenomena. However, within mathematics, such processes with a variable quiescent period are not commonly studied. To the biologist, output from the above model would be considered smoothly pulsatile, especially with the identical pulses; the variable lag process would be most readily distinguished from its constant lag counterpart (for which lagi ¼30 time units for all i) via a decidedly positive SD for the interpulse duration time-series, in the variable lag setting, as opposed to SD¼0 (constant interpulse duration) in the constant lag setting. The critical point here, however, is that for VarLag, the power spectrum
ORDERLINESS OF HORMONE RELEASE
93
FIG. 5. (A) Representative time-series for a ‘variable lag’ sine-wave process denoted VarLag (see text for de¢nition); (B) close-up view of (A), near time t ¼ 400; (C) power spectrum for VarLag; (D) power spectrum for a constant (¢xed) lag analogue of VarLag; (E) autocorrelogram corresponding to (C); and (F) autocorrelogram corresponding to (D). Figures (C)^(F) are all derived from time-series of length N ¼ 100 000 points.
94
PINCUS
and autocorrelation function somewhat confound, as seen in Figs 5C and 5E. Based on these ¢gures alone, the pulsatile nature of the time-series realizations is hardly evident, and for all k 5 6, the autocorrelation coe⁄cient rk at lag k is insigni¢cantly di¡erent from 0. In contrast, the power spectrum and autocorrelation function con¢rm the periodicity of the constant lag analogue, shown in Figs 5D and 5F, as expected. Signi¢cantly, the issues here are in the parameters, rather than statistical inadequacies based on an insu⁄ciently long output, or on artefacts (outliers), since Figs 5C^F were derived from calculations based on 100 000 points from a purely theoretical model. Similar limitations of the spectra and autocorrelation function are inherent to wide classes of processes, e.g. non-linear (deterministic and stochastic) di¡erential equations, in Poisson clumping models, and in output variables in typical (adaptive) control theory models and queuing network models. Notably, for many two-dimensional analogues of variable lag processes, and indeed for many two-dimensional systems in which no small set of dominant frequencies encapsulates most of the total power, the cross-spectrum and the crosscorrelation function often will similarly fail to highlight episodicities in the underlying model and data, and thus fail to highlight concomitant changes to such episodic components. In contrast to the autocorrelation function and spectral di¡erences between the above variable lag and constant lag processes, the respective ApEn(1, 20% SD) values for the two processes are in close agreement: mean ApEn¼0.195 for the variable lag process, while ApEn¼0.199 for the constant lag setting. This agreement in ApEn values manifests the primary requirement of matching (sub)patterns within data, while relaxing the requirement of a dominant set of frequencies at which these subpatterns occur. The two-variable analogue of ApEn, given by cross-ApEn, similarly enables one to assess synchrony in many classes of models. It thus should not be surprising that in many studies, e.g. the LH^T study (Pincus et al 1996a), cross-correlation (Pearson r) does not show signi¢cant group di¡erences, whereas cross-ApEn does (as in Fig. 4). Summary and conclusion The principal focus of this paper has been the description of both ApEn, a quanti¢cation of serial irregularity, and of cross-ApEn, a thematically similar measure of two-variable asynchrony. Several properties of ApEn facilitate its utility for biological time-series analysis: (i) ApEn is nearly una¡ected by noise of magnitude below a de facto speci¢ed ¢lter level; (ii) ApEn is robust to outliers; (iii) ApEn can be applied to time series of 50 or more points, with good reproducibility; (iv) ApEn is ¢nite for stochastic, noisy deterministic and composite (mixed) processes, these last of which are likely models for
ORDERLINESS OF HORMONE RELEASE
95
complicated biological systems; (v) increasing ApEn corresponds to intuitively increasing process complexity in the settings of (iv); and (vi) changes in ApEn have been shown mathematically to correspond to mechanistic inferences concerning subsystem autonomy, feedback and coupling, in diverse model settings. The potential uses of ApEn to provide new insights in endocrinological settings are thus myriad, from a complementary perspective to that given by classical statistical methods. Lastly, cross-ApEn, which assesses paired signal asynchrony, can be employed to directly quantify network uncoupling, and/or changes in feedback and control for a diverse variety of complicated systems.
References Box GEP, Jenkins GM 1976 Time series analysis, forecasting and control. Holden-Day, San Francisco Chat¢eld C 1989 The analysis of time series: an introduction, 4th edn. Chapman & Hall, London Gevers E, Pincus SM, Robinson ICAF, Veldhuis JD 1998 Di¡erential orderliness of the GH release process in castrate male and female rats. Am J Physiol 274:R437^R444 Hartman ML, Pincus SM, Johnson ML et al 1994 Enhanced basal and disorderly growth hormone secretion distinguish acromegalic from normal pulsatile growth hormone release. J Clin Invest 94:1277^1288 Kaplan DT, Furman MI, Pincus SM, Ryan SM, Lipsitz LA, Goldberger AL 1991 Aging and the complexity of cardiovascular dynamics. Biophys J 59:945^949 Pincus SM 1991 Approximate entropy as a measure of system complexity. Proc Natl Acad Sci USA 88:2297^2301 Pincus SM 1994 Greater signal regularity may indicate increased system isolation. Math Biosci 122:161^181 Pincus SM, Goldberger AL 1994 Physiological time-series analysis: what does regularity quantify? Am J Physiol 266:H1643^1656 Pincus SM, Huang WM 1992 Approximate entropy: statistical properties and applications. Commun Statist Theory Meth 21:3061^3077 Pincus S, Kalman RE 1997 Not all (possibly) ‘random’ sequences are created equal. Proc Natl Acad Sci USA 94:3513^3518 Pincus SM, Keefe DL 1992 Quanti¢cation of hormone pulsatility via an approximate entropy algorithm. Am J Physiol 262:E741^E754 Pincus S, Singer BH 1996 Randomness and degrees of irregularity. Proc Natl Acad Sci USA 93:2083^2088 Pincus SM, Viscarello RR 1992 Approximate entropy: a regularity measure for fetal heart rate analysis. Obstet Gynecol 79:249^255 Pincus SM, Gladstone IM, Ehrenkranz RA 1991 A regularity statistic for medical data analysis. J Clin Monit 7:335^345 Pincus SM, Cummins TR, Haddad GG 1993 Heart rate control in normal and aborted-SIDS infants. Am J Physiol 264:R638^R646 Pincus SM, Mulligan T, Iranmanesh A, Gheorghiu S, Godschalk M, Veldhuis JD 1996a Older males secrete luteinizing hormone and testosterone more irregularly, and jointly more asynchronously, than younger males. Proc Natl Acad Sci USA 93:14100^14105 Pincus SM, Gevers E, Robinson ICAF et al 1996b Females secrete growth hormone with more process irregularity than males in both humans and rats. Am J Physiol 270:E107^E115
96
DISCUSSION
Pincus SM, Veldhuis JD, Mulligan T, Iranmanesh A, Evans WS 1997 E¡ects of age on the irregularity of LH and FSH serum concentrations in women and men. Am J Physiol 273:E989^E995 Pincus SM, Padmanabhan V, Lemon W, Randolph J, Midgley AR 1998 Follicle-stimulating hormone is secreted more irregularly than luteinizing hormone in both humans and sheep. J Clin Invest 101:1318^1324 Roelfsema F, Pincus SM, Veldhuis JD 1998 Patients with Cushing’s disease secrete adrenocorticotropin and cortisol jointly more asynchronously than healthy subjects. J Clin Endocrinol Metab 83:688^692 Ryan SM, Goldberger AL, Pincus SM, Mietus J, Lipsitz LA 1994 Gender- and age-related di¡erences in heart rate dynamics: are women more complex than men? J Am Coll Cardiol 24:1700^1707 Siragy HM, Vieweg WVR, Pincus SM, Veldhuis JD 1995 Increased disorderliness and ampli¢ed basal and pulsatile aldosterone secretion in patients with primary aldosteronism. J Clin Endocrinol Metab 80:28^33 Urban RJ, Evans WS, Rogol AD, Kaiser DL, Johnson ML, Veldhuis JD 1988 Contemporary aspects of discrete peak-detection algorithms. I. The paradigm of the luteinizing hormone pulse signal in men. Endocr Rev 9:3^37 Van den Berg G, Pincus SM, Veldhuis JD, Fro«lich M, Roelfsema F 1997 Greater disorderliness of ACTH and cortisol release accompanies pituitary-dependent Cushing’s disease. Eur J Endocrinol 136:394^400 Veldhuis JD, Liem AY, South S et al 1995 Di¡erential impact of age, sex steroid hormones, and obesity on basal versus pulsatile growth hormone secretion in men as assessed in an ultrasensitive chemiluminescence assay. J Clin Endocrinol Metab 80:3209^3222
DISCUSSION Veldhuis: ApEn may re£ect the behaviour of a variety of inputs, even though you measure a single node (Veldhuis & Pincus 1998). In many of our systems we can’t measure multiple nodes concurrently. Steve Pincus, would you tell us the basis for that kind of assertion? Could you perhaps name a couple of situations in which we know this to be true? Pincus: There is both theory and practice. If something is going to work statistically, you want some mathematical back-up. Many of us have written papers where we want to show mathematical understanding. We have a mathematical model that has been posed, and we do parameter ¢t or parameter estimation, and we draw a mechanistic inference at the end. The point is, there are any number of model forms, from stochastic di¡erential equations, to ordinary and partial di¡erential equations, non-linear equations and di¡usion models. You don’t know what the actual biological model is, but you do expect it to be an ugly composite. Whatever statistics you do, you’d better be consistent across a range of model forms, so that you can draw mechanistically replicable inferences. In other words, if I had a situation where for di¡erential equations my conclusion was that the system was becoming uncoupled, and for di¡usion
ORDERLINESS OF HORMONE RELEASE
97
processes and cyclic queues, the conclusion was that it was becoming more increasingly coupled, I’d be worried. In theory, what I did was to take several di¡erent mathematical model forms: a composite mixed stochastic^deterministic model, a logistics model, and what is called an ARMA model (autoregressive moving average), a classical statistics model. For each of these three very di¡erent model forms, I translated the question, ‘What does it mean to say that a system is uncoupling?’, into precise mathematical formulation. I let the parameter change according to it becoming more uncoupled, ran ApEn on the output signals, and asked whether ApEn con¢rmed what the parameter was telling us. Namely, did ApEn go up as the system became more uncoupled. For each of those model forms, ApEn was consistent with the model. Therefore, I think we can be reasonably con¢dent that across large varieties of diverse models, uncoupling means increasing ApEn. In a lot of these models I’ve only quanti¢ed one node of complicated networks. All the models on the table had hidden variables that weren’t being explicitly analysed, yet we’re still getting at the uncoupling. In practice, in instances where we have had one signal going up in terms of ApEn, the other signals in closed networks do also. For the pure math inclined, there is one exception: we can write all this down in terms of languages of di¡erential equations, estimations, derivatives and Jacobians. It is only when the Jacobian becomes singularwhen the matrix that de¢nes the system has a determinant of zerothen you lose that property. This is highly non-generic: if you perturb the system a tiny bit, you regain that. What this singularity means is that in some sense you have uncoupled or unhooked the system up to ¢rst order. So in theory and practice, ApEn quanti¢es what it should. Matthews: This is an important advance in terms of understanding some of the stochastic processes going on. The reality is that people have used extremely na|« ve approaches to pulsatile phenomena. I really like this approach. Fourier transforms can be used under some circumstances, but you are right that they are less clearly useful when pulsatility is irregular. However, I’m not sure that I agree with you about cross-correlation. There are various identi¢able functions of cross-correlation which are quite good, even in the irregular cycles, and which can be demonstrated in subgroups. One of the advantages of cross-correlation is that there’s an identi¢ability to things like lag time, which I’m not sure ApEn provides. Pincus: That is right. Cross-correlation can certainly give speci¢c information that is entirely distinct from ApEn. I will never say that ApEn is better than another sound statistic. The example I give on heart-rate, when I give talks to a di¡erent audience, is that if I have a mean heart rate of 15 beats per minute, I want to be rushed to the ER. Is the mean a better or a worse statistic than ApEn? No, it is di¡erent. It has di¡erent applications. What we want is a diverse set of tools. In many instances where it looks like there is some sort of coincidence (and I’ll use that precise term instead of correlation) between variables, the correlogram has
98
DISCUSSION
not sorted that out. Most people use canned programs. Beware: canned FFT programs are there because they are very quick. There are many insidious problems in spectral estimation that are not known to anyone other than pure math specialists. There is a colleague of mine, David Thomson, who is a leader in this ¢eld. He has been able to show that on the basis of just 100^200 points of data, instances where these packages are 10 orders of magnitude wrong in their spectral estimate. The canned FFT packages are very dangerous when you’re using them on just a couple of hundred points. If you are seeing something, it is probably visible in another way. That being said, it is often physiologically useful, because you historically have the interpretation action at a speci¢c band calibrating to something of physiological or pathophysiological signi¢cance. This is particularly true in the EEG. So spectral analysis isn’t to be discounted, but rather for showing signi¢cance between groups, it’s often not quantitatively as sharp as you would want it to be. Brown: I ¢nd it di⁄cult to think of using just one measure of irregularity. I can understand why you restrict yourselves m ¼2 or m ¼1 because you have a short series of data, but if you have a very long series, you could actually use other values of m and r. Pincus: It is a family of statistics. We’re looking at pairs and triples, and aggregates of functions thereof. If you have say 100 000 points of data, maybe a good robust EEG for a period of time, you can ask about quadruples and quintuples, and ask which one is the best. The best one is a function of which is the best clinical correlate. What is more subtle and interesting mathematically is that the probability marginal distributions are not redundant with one another. You can have instances where either by looking at coarser or ¢ner r, or larger or smaller m, you £ip £op back and forth, as a function of m and r. Fortunately, mechanistically, it turns out that for most coupled systems, as you uncouple the system, ApEn tends to go up almost irrespective of m and r. At this coarse level of physiologic or mechanistic change, it is stable, but purely probabilistically you want the range. I have seen contexts in which triples do better than pairs. I’ve also seen contexts (and this is true in one instance in insulin) where, because the di¡erence seems to be more at a macroscopic than a microscopic level, you do better at a coarse r rather than a ¢ne r. To make an analogy from the art world, it would be akin to viewing a Seurat painting from up close versus far away. You see di¡erent things from close up as opposed to at a distance. Brown: In one of your earlier papers you mentioned that although the method is model free, it’s not necessarily measurement-strategy free. There are possible problems with dependence on assay error and time interval between measurements. Have you sorted these problems out? Pincus: That is actually three questions in one. How sensitive is ApEn to noise? If the r in ApEn is well above the noise threshold, ApEn changes very minimally, that
ORDERLINESS OF HORMONE RELEASE
99
is, it is very robust. It is easy to show that if the level of noise in the system is comparable to r or bigger, you can draw invalid inferences. The other issue though, which is a slightly di¡erent variation on this theme, is that sometimes you might want to see things six or seven points in a row, or over a longer timeframe, and you don’t have enough points. What you can do is to cleverly apply ApEn by looking at the ¢rst three points, taking an average of those, the second three points and average those, and formally derive time-series. There have been instances in seismology, for example, where that’s a clearer discriminator. The point is, once the technology is there to say what is irregularity, than you can do clever stunts with that. Robinson: In your cross-ApEn, would you get a lot more information if you drove one of the inputs? For example, if you put a pacemaker driving in and then made some assessment about the relative correlations between the two variables? Pincus: You would indeed. You can do cross-ApEn on A on B, or B on A. Which will be more incisive, again, is context dependent. In particular, I think it is mechanism driven. If you think A is driving B, in general do cross-ApEn of B on A, and that often shows clearer di¡erences say between normal and pathophysiologic than the opposite. But that’s in general mechanistically driven. So it certainly matters, and they’re not symmetric. Le¤ vi: ApEn has been mostly applied to hormonal time-series. Have you used this method for motor activity or temperature time-series? Pincus: My partner is a professor of Obstetrics and Gynaecology, and she specializes in the menopause. She deals clinically and theoretically with issues of hot £ushes. The reason I mention this is that you may be able to measure temperature, and to do correlates of that with the hormonal variables, to say how much the external temperature quanti¢es hot £ushes. The other more theoretical issue is, this has actually been done in pregnant women, looking at the temperature time-series. In e¡ect, there is almost nothing on the table to measure stress scienti¢cally. Certainly the catecholamines are one, temperature and hot £ushes are a very natural sort of correlate, and those are actually very relevant hormone signals. None the less, in theory and practice little has been done and it is of interest. Sassone-Corsi: There is something that bothers me here. Is there an arbitrary choice that you make in deciding the tolerance window? Pincus: That is an important point. In the ApEn de¢nition, we set a window length with m and we set the tolerance with r. What would happen if I chose a di¡erent m or a di¡erent r? Am I just doing everything in creation until I get a good answer? No, but ¢rst I want to explain why I choose the m and the r, and then what the robustness properties are. The bottom line is that the robustness properties are extremely good. My thesis 30 years ago had to do with something
100
DISCUSSION
called the Lyapunov spectrum, which has been implicated in true dynamical systems with chaos work. I ¢rst saw some of the applications in the name of chaos applied to heart rate time-series. The issue of very regular heart rate patterns is very important within fetal monitoring; there, sinusoidal heart rhythms (SHRs) being ominous is well known. But there were subtle changes sometimes to perfect SHRs; lo and behold I got answers that did not match the visual intuition. The question became, what is going on? To cut a long story short, it became clear that there was ‘blind’ technology transfer of an algorithm to compute the Kolmogorov^Sinai (K^S) entropy. People were in e¡ect looking at blocks of data 8^10 points in a row where you have no replicability. So the m and the r are driven here so that roughly speaking you’re breaking the state space into a few enough cells so you get reproducibility. m is almost always going to be one or two; r doesn’t want to be too small, because you’re saying this pattern has to exactly match that pattern, and if you have no instances of basal patterns, and if you have no instances of ‘matched’ patterns, you’ll get no robustness on your estimate of conditional probability, because ApEn is basically aggregating conditional probability. Sassone-Corsi: So it is basically a compromise. Pincus: That is not quite true. It is a compromise in the sense that statistics is a compromise of probability. If you give me 12 points of data, I can ¢t a quadratic polynomial. If you give me a million points of data I can ¢x a sixth order polynomial. If you tell me to ¢t a sixth or seventh order polynomial to my 10 points, I can do that as well, but I have no reproducibility in the higher terms. It’s a compromise in that statistics forces the low order: statistics is ‘probability in the land of limited resources’. However, ApEn is ¢nite for lots of processes for which the K^S entropy is in¢nite, thus allowing me to discriminate huge classes of processes that are indistinguishable from an ergodic theory perspective, so it allows you to grade. The other question, which is dead on the mark, relates to the choice of m and r. Suppose instead of m ¼1, I took m ¼2, and suppose instead of r ¼20% I chose r ¼ 10%. Would ApEn change non-trivially? Absolutely. But, if I’m looking at controls versus tumour growth and I do ApEn with one set of set of m and r, and I get an almost entire separation with P ¼0.001, if I use a di¡erent m and r I’ll get basically the same separation and P value. Roughly speaking in every (uncoupling) theoretical study and empirical study that I have done this on, there is absolute relative consistency to the choice of m and r, up to the limits of statistical replicability. Wu: You put a lot of emphasis on decreasing ApEn when the signals are coupled. In relation to what Iain Robinson mentioned earlier, is that always the case, depending on whether it’s a feedforward or feedback phenomenon? Pincus: No, positive feedback and negative feedback work in the opposite direction. Purely theoretically, the way to think of it is as derivative becoming
ORDERLINESS OF HORMONE RELEASE
101
positive or negative. Positive feedback loops accelerate and will increase ApEn, whereas negative feedback loops will slow down and decrease ApEn. Wu: You showed some pictures of FSH pulses. My experience in trying to interpret FSH pulses would suggest the opposite: if you have feedback, it tends to fragment these pulses, and seems to increase irregularity. Pincus: That’s exactly what my data showed. In all instances FSH was more irregular than others. Wu: If you take away the feedback then FSH tends to become more regular. Pincus: That’s a question of linguistic interpretation. I can say with quanti¢able certainty that in all instances FSH is more irregular than LH. As to the means by which a putative extra factor intersects the axis, I’m the least biological person here, so I’m not going to call you on whether it is feedforward or feedback, but I will say that the FSH always has the more subordinate wiggles and is more irregular. What requires interpretation is exactly the physiology of what inhibin will be doing. But the quantitative results are irrefutable. Veldhuis: Let me just make sure I understand Fred Wu’s question, because I might have analogous data with testosterone withdrawal. Fred, what I see with testosterone withdrawal is that FSH goes up and the ApEn is higherFSH release is more irregular (Veldhuis et al 1997). Are you saying that with inhibin withdrawal, in a ¢xed testosterone milieu, you get more irregular FSH release also? Wu: What I am saying is that without inhibin, you tend to see FSH pulses more clearly. Pincus: Are you saying that you lose the subordinate wiggles? Wu: Yes. Pincus: That would be believable. Wu: This is visual intuition. Pincus: The numbers tend to correspond with the intuition. Veldhuis: But then we have a real divergence in feedback mechanisms. It suggests that these systems are very messy, and that you’re not always dealing with singular feedbackyou may be dealing with multivalent or multi-site feedback, which then would give you any number of ways to alter ¢nal output (Veldhuis 1999). Pincus: There are many tra⁄c communication theory paradigms or concepts which, even if they haven’t been brought to bear in biology, are well known in the communication ¢eld. Any of these complications can enrich or complicate network interpretation. In particular, one such phenomenon could be adaptive control. If you’re in adaptive control to some optimal loss function, you could well imagine that this would steer or control the system in a positive way, but the loss function would be rather funky. In that case, it could be quite complicated. Licinio: With cross-ApEn, why do you stop at just two hormones?
102
DISCUSSION
Pincus: There are several reasons. The formal de¢nition of cross-ApEn was born out of pure math. In particular, the question of, what does it mean to say two ¢nite sequences are conditionally irregular? Cross-ApEn was developed primarily to answer that question. Thus the trivial answer to your question is that the pure math bore on two variables. The empirical answer is as follows. The analogy I make is that my car has stopped in mid-drive. I want to identify what went wrong. So I take it to the garage and ask them to start doing diagnostics. They are not going to write down a bunch of precise mathematical equations: they will start testing nodes, pathways or conduits. The nodes are the one-variable statistics, and the conduits are the bivariate statistics. The point is, once that I have established that it is more than one node that went down, if it is something in the category of a network, the question you are asking is is it simply a pathway, or is it something more at central control? This comes up in an interesting paper that Johannes Veldhuis and I are publishing on LH and nocturnal penile tumescence (NPT). We have looked at sex steroids (LH, testosterone and FSH) and an endpoint measure of NPT. Penile tumescence is somewhat rhythmic, and the question is, in ageing is there some sort of a disruption? It turns out that there is. But the question is, is it a mean thing? No. Is it an ApEn thing? It is to some extent, but it is much more of a coupled ApEn thing, namely an asynchronous thinga cross-ApEn thing. We are looking at FSH, LH, prolactin and NPT. What happened is that all the cross-ApEns went up, but very few of the ApEns went up. Think about this in terms of pathways. If one pathway changed a lot, you would implicate the pathway. If all the pathways changed, you would implicate the central controller. The way I like to use cross-ApEn, therefore, is to say that if I want to get at a single pathway, typically one cross-ApEn is increased. If all or many of them are increased, I think it is more central control. Marshall: I would like to ask about applying this to physiology. It seems to me as you remove inhibitory feedback systems, then the ApEn score will go up. Pincus: This is true in endocrinology, but in certain other systems, such as heart rate, pathology causes more regular signals. Marshall: We are not able to make a lot of progress in learning some of the circuitry of regulation of the human nervous system and its hormone output. Take a situation where you have an open loop feedback, such as in an adrenalectomized personthose who have been replaced with steroids. What range in change of ApEn scores would you expect, and could you potentially characterize e.g. the corticotropin-releasing factor/ACTH system as opposed to the testosterone/LH feedback system? Can you come up with some parameters that allow us to characterize those di¡erent hypothalamic regulatory systems in humans? Pincus: Let me answer this a couple of di¡erent ways. First, from the purely mathematical standpoint, since the ApEn standard deviation is typically 0.05 or
ORDERLINESS OF HORMONE RELEASE
103
smaller, anything above 0.1 of an ApEn change would be signi¢cantly di¡erent. But let me draw a picture which comes to mind. Take for instance GH administration in the elderly.I’mgoingtodrawanetworkmodeltogiveasenseofthekindsofthingsyoucan do with point processes. I am going to drawa highway network model with four nodes and ¢ve pathways, ¢ve lanes on the road, then later suppose thatthere are four, and later still, that there are only three. Suppose ageing represents this network becoming threadbare or knocked out. As you start losing roads on your highway, if your tra⁄c stays at a given level, you will get longer delays and lose performance. Physiologically, in the GH instance, if you have gone down from ¢ve pathways to one, all you might be able to do is restore one or two of the decayed pathways. If on the other hand there is simply a blockage here, such as a tumour that is excisable, you might be able to get full restoration. So precise quantitative guidelines are going to be a function of whether the network is fully restorable, or is there just degradation due to general ageing that is basically irreparable. Marshall: Intuitively, we don’t expect things to get better as we age. But what I am trying to get at is, is there a commonality of the variability when you use this tool (ApEn) to characterize feedback systems in humans and rats as you add or subtract inhibitory regulatory feedback by steroids? Pincus: Certainly in all the ageing models we’ve looked at there’s been universal disruption and ApEn has gone up. Marshall: Just take an ordinary healthy person who doesn’t have an adrenal gland. If you look at what happens when they have high ACTHs, I assume that the ApEn scores would be high. If you put back cortisol in a physiological manner (you can do the same for any of the feedback systems), do you now come down to a similar common ApEn score? Veldhuis: Steve doesn’t know that we have already used his statistic to do that. We’ve actually done it for three di¡erent neuroendocrine axes: primary hypothyroidism (so we have withdrawn the negative feedback signal of T4), ketoconazole to remove testosterone and fasting to delete insulin-like growth factor I, and the ApEn jumps 4^8 standard deviations for thyrotropin, LH and GH, respectively, in these contexts (Veldhuis et al 1996). Thus the system gets fairly frantic under those conditions, and jitters. In contrast, after metyrapone, ApEn for ACTH falls (Veldhuis et al 1998). I thought your other question was whether it is possible to tell cross-species di¡erences in network levels of integration by looking at ApEn under identical assays or under suitable r values that adjust for the di¡erent assays, to ¢lter the noise. Pincus: In the rat and human we see sexual dimorphism. Interestingly, in both the sheepandhumanchangesinLHandFSHgoawaywithageing.Therewillbeinferences that you can draw across species, but these will be the subject to the usual foibles of model di¡erences that you would expect in any other measurement assessment.
104
DISCUSSION
Goldbeter: Can you use ApEn methods to characterize chaos, or the transition to chaos? This is a question that may be relevant to the underlying mechanism. What do you think of attempts to characterize hormone time-series in terms of chaotic dynamics? Pincus: If I have a truly iterated dynamical system, which is what a truly chaotic model is (I have done several papers on dynamical systems, so it’s not that I’m friend or foe, what I am friend or foe of is doing the right application in the right context), if I have a parameter which increases from say periodic to chaotic, ApEn will go monotonically up, increasing in the same way that the entropy and the Lyapunov spectrum will. So it’s absolutely consistent with this. That being said, the modelling of especially biological data by strictly deterministic models leaves me cold. I can approximate any truly dynamical system model by a Markov chain, which is probabilistic, so from a theoretical answer I can bust it entirely. More sensibly (this is really where I really side with Pierre De Meyts), I can write down any number of reasonably simple models. I really do believe very strongly the issues of tra⁄c and queuing theory, and decision processes want to bear on biology, so I don’t want to dwell on chaos per se. Almost all invariant steady-state measures are chaotic in the sense that they are aperiodic and bounded. It’s to discriminate among that huge class of muddied aperiodic and bounded measures which is of interest to us, to get at a subtle change in this degree of irregularity from that degree of irregularity: all of those would be chaotic, but how you tell those apart is the issue here. Goldbeter: There is an interest in distinguishing between true chaotic behaviour from noise. Pincus: It’s easy to distinguish true chaotic behaviour from what is called IID, or totally random behaviour: that’s easy. The point is, just saying the data are nonrandom is of limited use. Getting at subtle changes is very important. References Veldhuis JD 1999 Recent insights into neuroendocrine mechanisms of ageing of the human male hypothalamo-pituitary-gonadal axis. J Androl 20:1^17 Veldhuis JD, Pincus SM 1998 Orderliness of hormone release patterns: a complementary measure to conventional pulsatile and circadian analyses. Eur J Endocrinol 138:358^362 Veldhuis JD, Iranmanesh A, Pincus SM 1996 Reduction of intrinsic negative feedback regulation of neuroendocrine axes increases the approximate entropy (serial irregularity) of the pituitary hormone release process for TSH, GH, and LH. Soc Neurocience Meeting, Washington DC, November 16^21, Abstr Veldhuis JD, Iranmanesh A, Urban RJ 1997 Primary gonadal failure in men selectively ampli¢es the mass of follicle stimulating hormone (FSH) secreted per burst and increases the disorderliness of FSH release: reversibility with testosterone replacement. Internat J Androl 20:297^305 Veldhuis JD, Iranmanesh A, Naftolowitz D, Carroll BJ 1998 Mechanisms of ACTH neurosecretory reactivity to abrupt withdrawal of glucocorticoid negative feedback in healthy men: pulsatile, nyctohemeral, and entropic responses. 80th Annual Meeting Endocrine Society, New Orleans, Louisiana, June 24^27, Abstr
Prediction and signi¢cance of the temporal pattern of hormone secretion in disease states G. Brabant and K. Prank Abt. Klinische Endokrinologie, Medizinische Hochschule Hannover, Carl-Neubergstr. 1, D-30623 Hannover, Germany
Abstract. Comparison of the temporal pattern of hormone secretion in health and disease reveals distinct di¡erences in many systems. Analysis of these visually apparent di¡erences conventionally rests on computer-assisted programs based on either model assumptions, or estimations of hormonal decay rates or threshold values, all of which may not accurately re£ect physiological and/or pathophysiological states. Only recently have new methods evolved which are independent of preexisting knowledge of the sytem under study. Apart from the widely used approximate entropy statistic (ApEn), a measure for the regularity of a time-series, arti¢cial neural networks are able to capture temporal structures in endocrine rhythms without any previous assumptions. In particular, non-linear dynamical systems may be delineated and separated from random behaviour. This is achieved by mapping complex input data to a given complex output by propagating data from the input layer to the output layer through a larger number of interconnections, so-called hidden layers. The networks are capable to extract relevant features from training samples and store this information in the distributed structure of interconnections. Using this approach on growth hormone (GH) rhythms of healthy controls, fasted healthy subjects, untreated acromegalic patients and acromegalics under octreotide suppressive therapy we were recently able to demonstrate the power of this approach to di¡erentiate the temporal pattern of GH secretion following normalization of the data for absolute amplitudes. In a second approach we were able to signi¢cantly reduce the number of data points required to characterize the temporal structure of these rhythms. This latter quality of the networks may help to transfer analysis of changes in the temporal pattern of hormone secretion on a more routine basis. 2000 Mechanisms and biological signi¢cance of pulsatile hormone secretion. Wiley, Chichester (Novartis Foundation Symposium 227) p 105^118
Visual inspection of hormonal time-series frequently reveals distinct di¡erences between health and disease. In order to objectively quantify the speci¢c characteristics of the temporal pattern of hormone secretion, computer-assisted programs have been developed which allow one to de¢ne the amplitude and frequency of hormonal pulses (Merriam & Wachter 1982, Veldhuis & Johnson 105
106
BRABANT & PRANK
1986, Prank & Brabant 1994). They are based on model assumptions on the shape, duration, and rates of rise and fall of a hormonal pulse in relation to assay noise, the estimation of hormonal decay rates or simply on threshold levels de¢ned to di¡erentiate noise and true secretory activity in sequential time points of a hormonal rhythm. However, all these criteria may only incompletely delineate physiology and thus may cause problems especially in the comparison of normal and pathophysiological states where these assumptions may selectively be violated. In an attempt to circumvent such problems, we recently evaluated the capability of new analytical tools for the prediction of endocrine rhythms under normal conditions and in a diseased state. Arti¢cial neural networks are e¡ective information processing tools widely used in other ¢elds to compare di¡erent situations without using preformed assumptions or models (Haykin 1994, Bishop 1995). Their analytical power rests on the mapping of complex input data to given complex output data (Fig. 1). By propagating data from the input layer to the output layer through a larger number of interconnections so-called hidden layers the networks are capable of extracting relevant features from training samples and storing this information in the distributed structure of interconnections. Arti¢cial neural networks are able to generalize from the examples trained onto unknown examples. They are adaptive to changes in di¡erent settings and allow the acummulation of information from a large number of examples. Thus, increasing the number of learning samples might not simply improve statistics but the quality of analysis as well. As a major advantage for a tool designed to analyse a biological system, neural networks are fault-tolerant. Due to the distributed nature of information storage in the network architecture, imprecise information results in less damage to the mapping process and may be successfully compensated by the remaining complex information structure. The power of arti¢cial neural networks is based on their non-linearity. As regulation through temporal-encoded endocrine feedback loops appears to be non-linear, they may have distinct advantages over other analytical approaches. In a previous report we tested the 24 h growth hormone (GH) secretory pattern in healthy subjects studied twice under basal conditions and following 3 days of fasting by sampling blood every 10 min over 24 h. We compared these rhythms to the 24 h GH patterns of untreated acromegalic patients before and after 4 days of a continuous infusion of 300 mg octreotide (Riedel et al 1992, 1995). Conventional pulse detection with the programs Cluster, Pulsar and Desade revealed distinct di¡erences in the amplitude of a GH pulse between controls and acromegalic patients under octreotide infusion, but failed to de¢ne changes in GH pulse frequency in the various conditions (Tables 1, 2). To overcome this apparent discrepancy with the visually distinct patterns of GH secretion we applied approximate entropy statistics (ApEn), a method used to characterize the regularity of a given time-series (Pincus 1991). We choose a ¢lter value of 20% of
PATTERNS OF HORMONE SECRETION IN DISEASE
107
FIG. 1. Schematic ¢gure of a feedforward neural network.
the respective standard deviation ( SD) of the individual GH rhythm to decorrelate the ApEn value from the SD of the time-series under study. The size of the window used was set to 1. Under these conditions the regularity of the GH rhythm of healthy subjects, both under basal as under fasting conditions, was signi¢cantly higher than that of acromegalic patients (Table 3). Thus, ApEn appears to be capable of distinguishing the di¡erences in the temporal pattern of GH rhythms. To further investigate the hypothesis of an unstructured patterning of GH release in acromegaly potentially due to an autonomous GH secretion from the tumour instead of a hypothalamic controlled regulation, we analysed GH rhythms in normal subjects and in acromegalic patients with the help of two
108
TABLE 1
BRABANT & PRANK
Classical pulse detection analysis: mean number of GH pulses in 24 h SD Heuristic methods
Model-based approach
Group 1 vs. group 2
CLUSTER
PULSAR
DESADE
Acromegaly, basal vs. Control, basal
13.0 5.7 10.8 4.2 NS 13.0 5.7 13.0 1.9 NS 14.8 3.5 10.8 4.2 NS 14.8 3.5 13.0 1.9 NS
14.5 8.3 9.0 2.8 NS 14.5 8.3 14.8 3.7 NS 5.5 2.9 9.0 2.8 NS 5.5 2.9 14.8 3.7 P ¼ 0.002
11.0 1.7 9.6 1.5 NS 11.0 1.7 10.8 4.6 NS 8.5 2.9 9.6 1.5 NS 8.5 2.9 10.8 4.6 NS
Acromegaly, basal vs. Control, fasting Acromegaly, SMS vs. Control, basal Acromegaly, SMS vs. Control, fasting
TABLE 2 mU/l SD
Classical pulse detection analysis: mean GH pulse amplitude in
Heuristic methods
Model-based approach
Group 1 vs. group 2
CLUSTER
PULSAR
DESADE
Acromegaly, basal vs. Control, basal
18.7 21.5 2.3 1.5 NS 18.7 21.5 5.4 1.5 NS 2.1 1.6 2.3 1.5 NS 2.1 1.6 5.4 1.5 P ¼ 0.006
15.8 13.4 2.5 1.3 NS 15.8 13.4 5.1 1.0 NS 2.8 2.1 2.5 1.3 NS 2.8 2.1 5.1 1.0 P ¼ 0.046
26.0 19.5 4.0 1.9 P ¼ 0.04 26.0 19.5 9.6 3.3 NS 4.2 3.7 4.0 1.9 NS 4.2 3.7 9.6 3.3 P ¼ 0.029
Acromegaly, basal vs. Control, fasting Acromegaly, SMS vs. Control, basal Acromegaly, SMS vs. Control, fasting
PATTERNS OF HORMONE SECRETION IN DISEASE
TABLE 3 N ¼ 144)
109
Approximate entropy (ApEn) statistics (m ¼ 1, r ¼ 0.20 *SD,
Group 1 vs. group 2
ApEn
Acromegaly, basal vs. Control, basal
1.36 0.29 0.35 0.11 P ¼ 0.001 1.36 0.29 0.85 0.23 P ¼ 0.01 1.38 0.35 0.35 0.11 P ¼ 0.001 1.38 0.35 0.85 0.23 P ¼ 0.01
Acromegaly, basal vs. Control, fasting Acromegaly, SMS vs. Control, basal Acromegaly, SMS vs. Control, fasting
methods evaluated on non-hormonal time-series, arti¢cial neural networks and ‘algorithmic complexity’. Algorithmic complexity (Rapp et al 1994) is a method comparable to ApEn which permits testing for regularity in a given time-series. In principle, the method works by compressing a data set through identi¢cation of common motifs. Time-series of higher regularity are more e¡ectively compressed than more irregular rhythms. In the example of the 24 h GH rhythms the di¡erence between acromegalics and controls were signi¢cant. The principal advantage of this approach is its ability to use smaller sets of data successfully. Mapping by arti¢cial neural networks may not only be used to delineate complex situations, but also to capture regularities in the temporal pattern of complex timeseries, particularly of non-linear dynamical systems, and to separate deterministic from random behaviour (Sugihara & May 1990, Tsonis & Elsner 1992, Weigend & Gershenfeld 1994). The technique allows the user to predict the future dynamics of a time-series from a number of past values. Di¡erences in the temporal dynamics of a system are then re£ected in an altered predictability. Such predictive approaches have been e¡ectively used in several biological systems (Sugihara 1994, Blinowska & Malinowski 1991, Lefebvre et al 1993, Chang et al 1994, Schi¡ et al 1994) and may have advantages over ApEn statistics as they can be applied for short time-series containing a very limited number of data points (Sugihara & May 1990, Tsonis & Elsner 1992). A single feedforward network was used and trained for each subgroup on pooled data by using a leave-one-out strategy. Prior to training, each individual GH rhythm was normalized between 0
110
BRABANT & PRANK
FIG. 2. Self-organized quanti¢cation of pulsatility (SOPUL). Normalized original GH timeseries and selection pattern of local experts used for prediction. (a) Healthy control, basal state; (b) acromegalic patient, basal state.
and 1 to avoid any bias by di¡erences in the absolute GH concentrations and to focus simply on the temporal pattern of the time-series. Including nine sequential input values of a GH time-series the network predicts the following time point. This predicted GH time point is then included in an iterative approach consisting this time of eight measured and one predicted time point to estimate the next step in TABLE 4
Frequency distribution of the two most probable local experts 1 and 5
Group 1 vs. group 2
Expert 1 (%)
Expert 5 (%)
Acromegaly, basal vs. Control, basal
12.7 6.7 2.5 1.6 P ¼ 0.009 12.7 6.7 4.4 1.2 P ¼ 0.02 19.7 9.6 2.5 1.6 P ¼ 0.003 19.7 9.6 4.4 1.2 P ¼ 0.007
82.2 9.7 96.0 2.1 P ¼ 0.01 82.2 9.7 94.4 1.8 P ¼ 0.02 73.0 12.6 96.0 2.1 P ¼ 0.003 73.0 12.6 94.4 1.8 P ¼ 0.005
Acromegaly, basal vs. Control, fasting Acromegaly, SMS vs. Control, basal Acromegaly, SMS vs. Control, fasting
PATTERNS OF HORMONE SECRETION IN DISEASE
111
the GH rhythm. This procedure, which may be used for any given number of time steps into the future, can be evaluated for its precision by estimating the prediction error during the multiple step-ahead prediction process. To avoid over¢tting of the networks to noise during the learning phase, we used regularization functions and cross-valdidation. In a second step we further optimized the neural network analysis by using a modular neural network system composed of several di¡erent ‘expert’ networks plus a gating network that decides which of the experts should be used for each input. Experts are specialized networks capable of capturing speci¢c features of the endocrine rhythm. During the training procedure these networks compete to generate the desired output for each input pattern. The analytical importance of an expert generating a lower error than the weighted average of the errors of all experts is increased, whereas networks with a higher error have a decreased analytical weight for certain features of the endocrine rhythm. The experts are local since their impact changes dynamically and they are not linked to the weights of other experts. To adjust the mixing proportions of the experts a gating network is trained to select the best performing expert for a given case (Fig. 2). Simulations on a complex vowel classi¢cation task have shown that adaptive mixtures of local experts are able to e¡ectively decompose a problem to yield higher classi¢cation performance than a single network (Nowlan 1991). Using this approach, we found that arti¢cial neural networks perform a quanti¢cation of hormonal pulsatility in a self-organized manner and that they are able to predict the future time-course of a GH rhythm. This self-organized quanti¢cation of hormone pulsatilty (SOPUL) is a method for analysing endocrine rhythms without any prior knowledge of the form of a pulse or the model of secretion. SOPUL analysis signi¢cantly separated the GH secretory dynamics in acromegaly under basal conditions and under octreotide treatment from that in normal controls in basal and fasting state (Table 4; Prank et al 1996). These results extended and con¢rmed data on the secretory dynamic of parathyroid hormone from patients with osteoporosis and from controls (Prank et al 1995). A major problem in the analysis of temporal rhythms in endocrinology is the high number of data points necessary to adequately describe not only the mean hormone secretion but also the temporal dynamics. As long as no sensors are available to monitor hormone concentrations ‘online’ in the circulation, discrete blood samples have to be drawn with high frequency, usually over 24 h. Analytical methods are based either on the statistical distribution of discrete hormone pulses or as demonstrated by ApEn statistics by the regularity of a time-series which ought to contain at least 100^200 data points. In an attempt to overcome this drawback for a wider and routine diagnosis of an altered temporal dynamic in disease states, we investigated the potentials of arti¢cial neural networks to reduce the number of data points required for the characterization of the
112
BRABANT & PRANK
FIG. 3. Representative 24 h serum GH concentration time-series. Healthy control, sampling interval 5 min (a) and arti¢cially subsampled at 20 min (c). Acromegalic patient, sampling interval 5 min (b) and arti¢cially subsampled at 20 min (d). The data are normalized to a mean of 0.3 and a standard deviation of 0.15.
FIG. 4. Discrimination of GH secretory dynamics in healthy controls and acromegalic patients. Area under the ROC curve over time. (a) Sampling interval 5 min, input window 2 h, prediction window 2 h; (b) sampling interval 20 min, input window 3h 20 min, prediction window 4 h.
PATTERNS OF HORMONE SECRETION IN DISEASE
113
temporal dynamic. In collaboration with the Charlottesville group, 24 h GH rhythms of patients with acromegaly and of healthy controls were systematically tested as for their predictability when the number of data points is sequentually reduced (Fig. 3). Figure 4 demonstrates the capability of the networks to signi¢cantly distinguish GH rhythms of acromegalic patients and controls by reducing data points to 20% of the original time-series. As outlined above, di¡erent absolute GH concentrations between groups were normalized to the same mean and variance. Dependent upon con¢rmation in other data sets and over a larger number of examples, these data suggest that arti¢cal neural networks may help to extract the characteristics of an endocrine time-series from a few data points and may thus represent a model to study diseases of the temporal coding of endocrine signals in future on a less arti¢cial academic basis. Acknowledgement Supported by DFG Br 915/4-4.
References Bishop CM 1995 Neural networks for pattern recognition. Oxford University Press, Oxford Blinowska KJ, Malinowksi M 1991 Non-linear and linear forecasting of the EEG time series. Biol Cybern 66:159^165 Chang T, Schi¡ SJ, Sauer T, Gossard JP, Burke RE 1994 Stochastic versus deterministic variability in simple neuronal circuits: I. Monosynaptic spinal cord re£exes. Biophys J 67:671^683 Haykin S 1994 Neural networks. Macmillan, New York Lefebvre JH, Goodings DA, Kamath MV, Fallen EL 1993 Predictability of normal heart rhythms and deterministic chaos. Chaos 3:267^276 Merriam GR, Wachter KW 1982 Algorithms for the study of episodic hormone secretion. Am J Physiol 243:E310^E318 Nowlan SJ 1991 Soft competitive adaptation: neural network learning algorithms based on ¢tting statistical mixtures. PhD thesis. Carnegie Mellon University, Pittsburgh, PA, USA Pincus SM 1991 Approximate entropy as a measure of system complexity. Proc Natl Acad Sci USA 88:2297^2301 Prank K, Brabant G 1994 Estimating thyrotropin secretory activity by a deconvolution procedure. Methods Neurosci 20:377^389 Prank K, Nowlan SJ, Harms HM et al 1995 Time series prediction of plasma hormone concentration. Evidence for di¡erences in predictability of parathyroid hormone secretion between osteoporotic patients and normal controls. J Clin Invest 95:2910^2919 Prank K, Kloppstech M, Nowlan SJ, Sejnowski TJ, Brabant G 1996 Self-organized segmentation of time series: separating growth hormone secretion in acromegaly from normal controls. Biophys J 70:2540^2547 Rapp PE, Zimmerman ID, Vining EP, Cohen N, Albano AM, Jime¤ nez-Monta•o MA 1994 The algorithmic complexity of neural spike trains increases during focal seizures. J Neurosci 14:4731^4739 Riedel M, Gˇnther T, von zur Mˇhlen A, Brabant G 1992 The pulsatile GH secretion in acromegaly: hypothalamic or pituitary origin? Clin Endocrinol 37:233^239
114
DISCUSSION
Riedel M, Hoeft B, Blum WF, von zur Mˇhlen A, Brabant G 1995 Pulsatile growth hormone secretion in normal-weight and obese men: di¡erential metabolic regulation during energy restriction. Metabolism 44:605^610 Schi¡ SJ, Jerger K, Chang T, Sauer T, Aitken PG 1994 Stochastic versus deterministic variability in simple neuronal circuits: II. Hippocampal slice. Biophys J 67:684^691 Sugihara G 1994 Nonlinear forecasting for the classi¢cation of natural time series. Phil Trans R Soc Lond A 348:477^495 Sugihara G, May RM 1990 Nonlinear forecasting as a way of distinguishing chaos from measurement error in time series. Nature 344:734^741 Tsonis AA, Elsner JB 1992 Nonlinear prediction as a way of distinguishing chaos from random fractal sequences. Nature 358:217^220 Veldhuis JD, Johnson ML 1986 Cluster analysis: a simple, versatile, and robust algorithm for endocrine pulse detection. Am J Physiol 250:E486^E493 Weigend AS, Gershenfeld NA (eds) 1994 Time series prediction: forecasting the future and understanding the past. Addison-Wesley, Reading, MA (SFI Studies in the sciences of complexity, Proceedings vol XV)
DISCUSSION Veldhuis: One of the challenges facing us is understanding how a whole collection of signals can £uently produce a sensible output for the system. We have now had experience of applying ApEn to some other tumours, such as Cushing’s and prolactinomas. What is your theory on what these endocrine tumours are doing that produces this random-looking output? I noticed from your algorithmic approach that when you shu¥ed the tumour data, you got minimal further increases: you were nearly maximally random. Brabant: That’s what we proposed. You mentioned Cushing’s. There is a big debate in the literature, stimulated by E. van Cauter (van Cauter & Refeto¡ 1985) that there might be a hypothalamic form of Cushing’s disease with a more structured ‘hyperpulsatile’ adrenocorticotropic hormone (ACTH) release and a pituitary form with autonomous tumour cells releasing ACTH in a more random fashion. It may well be that these are the examples to tackle with such methods to see whether we can distinguish them. We know from clinical experience that the outcome concerning recurrence in Cushing’s disease falls in two distinct groups. If the tumour is removed in one situation the patient is cured for the rest of their life; in the other situation frequent recurrences are to be expected without clear-cut di¡erences in the tumour size or surgical approach. It has been suggested that the more structured ‘hypothalamic’ form is prone to recurrence whereas the pituitary form with autonomous release from a clonal tumour is not. It might well be possible with the approach outlined to predict these two forms and alter the clinical therapeutic approach based on such data.
PATTERNS OF HORMONE SECRETION IN DISEASE
115
Pincus: There are a many diseases that have classically been put into one pool because of a certain overt set of clinical symptoms. However, detailed physiologic inspection has suggested these are probably several di¡erent diseases, but we don’t have the means of separating them. You can well imagine that the therapies would be di¡erent. In addition to the tumours, which George brought out, there is the example of polycystic ovary syndrome (PCO). It turns out that the classical case is obese hirsute ladies who don’t ovulate, but virtually all clinicians will say that while that may be 70% of all PCOs, it is hardly the whole cohort. There are other large subcategories that have been lumped under the umbrella of PCO. There are some thin PCOs with earlier onset. You can imagine not so much that you could simply detect from appropriate statistical or mathematical technologies that they are di¡erent, but in fact that the therapies would be di¡erent and you can keep score of restoring pulsatility to the norm. The means of restoration the therapeutic interventions would actually be di¡erent. So it’s not just an idle statistical exercise, and it’s not just limited to tumours. With tools to delineate the secretion pattern more sharply, you can actually delineate di¡erent diseases that are currently lumped into a single pot, into subgroups where you want di¡erent treatment modalities. Veldhuis: George, do you have any clinical recurrence data? One of the puzzling things that we’ve noticed in our collaboration with Ferdinand Roelfsema is that in the ¢rst three months or so post-operatively, after removing a GH-secreting pituitary tumour (presumptively nearly in toto, because the patient later appears to be cured), the ApEn remains high, and then only gradually recovers (Roelfsema et al 1998, Van de Berghe et al 1998). This raises a number of issues about prediction and whether you might err trying to make measures of ApEn too early while there’s still recovery from post-operative changes. Do you yet have any hard clinical data on recurrence prediction? That would be fairly exciting. I recently saw a large Italian study on Cushing’s disease, where there is a 15% ¢ve year recurrence rate. It would be nice to know who the patients are who will recur. It might be that early radiation or re-operation is sensible to prevent a recurrence with a subset at high risk. Brabant: I agree, but we don’t have any data yet. This is really a two-step thing. First, you need an analytical approach which gives you a chance to tackle such problems, and these approaches are now emerging. Second, from the clinical point of view, you need to know whether the patient is at risk or not. Brown: I thought it was a very interesting case study, and you do seem to achieve very good predictability of some hormone levels. But one thing which concerned me was the number of parameters in your neural network models. There were about 70 parameters in one neural network corresponding to the synaptic weights. Then you had a second neural network, and the gating network: there might be well over 100 parameters in all. This seems to be a lot of parameters to
116
DISCUSSION
estimate. Have you tried using smaller nets? Have you also done cross-validation, checking this out on independent data? Brabant: We always check with independent data. With these network approaches you need many training sets, and then you need independent data where you try to see how it behaves. I agree with you that there is a trade-o¡ between the number of parameters you use. For optimal adaptation of the network to the training examples you would like to include more parameters, and for best distinction thereafter you would like to cut down the numbers. Because this is a very important issue for the practical application of the networks, we are currently testing this trade-o¡ formally to optimize the number of parameters used. Clarke: I would like to shift the discussion a little. I have a lot of chaotic data, and it’s nice to have some way of telling whether they are more chaotic or less chaotic, which is what I understand techniques such as ApEn do. But what most of us still want to know about our data sets is how many pulses there are, and what frequency and amplitude of these pulses is. These sorts of analyses give us an index, but I would like to know what other sort of analysis we should use to provide us with amplitude and frequency data. Pincus: From my intuition, in those instances where you can £ag pulses fairly clearly, the classical pulse methods such as spectral analysis are more intuitive. Furthermore, they’re estimable with typically fewer points and don’t need such sharp statistics. It’s often in the normative physiological setting where you do get the clear pulses and you can identify them, but it’s in the pathophysiologic instances in which the data are nearly maximally irregular, that it’s very hard to identify a pulse. In these cases the algorithms are arbitrary, and the consequence of this is that you get poor replicability. In those instances where you can clearly delineate pulses, I don’t think there’s a particular statistical discriminatory advantage of either ApEn or the prediction methods. They’re basically consistent and complementary and give about the same power. It is in the pathophysiology that these methods are better. Clarke: But you’re still not telling me what a pulse is. I want to know what is the best method for identifying a pulse. The pulsar analysis is one that’s subjective in the ¢rst instance, because you choose the settings depending on what you think is a pulse! Brabant: That was the reason for my introduction. You may get nowhere by counting pulses, because you have a pre-formed knowledge of the pulses which may not truly represent physiology. I think we should leave a formal counting of pulses or a de¢nition of a pulse by height, form and/or duration because, as we discussed after David Waxman’s paper, we have to know the sensitivity of the responding cell or tissue which might be altered by many cofactors. Furthermore, a long interval of no stimulation by a given stimulus may
PATTERNS OF HORMONE SECRETION IN DISEASE
117
completely alter the sensitivity as compared to a rapid series of completely identical pulses. So I would like to question the importance of a general pulse de¢nition. What we urgently need is a more global understanding of signalling in interconnected dynamic systems. I do believe that we have to move away from descriptive to more predictive way of approaching these questions where a single pulse in its detailed characteristics is embedded in a complex signalling network. Clarke: If I send my paper o¡ without any information about frequency or amplitude of pulses, the referees will laugh. Are you are telling us that we should get right away from the idea of pulses? Pincus: No, we are saying that in the instances where you can clearly ¢nd a pulse, go with that if you’re comfortable with it. However, the problem here is that there are lot of very messy processes that are known to mathematicians who do stochastic processes, where you have very broad banded spectra yet which are not white. If you looked in the time domain at those, you would not clearly ¢nd pulses at all. Clarke: In the case where you can £ag a pulse, I want to know what we should be using to £ag the pulse, that’s all. Pincus: In the instances where pulses are fairly clearly delineated most of the algorithms give about the same signi¢cance between cohorts. It is in those muddy instances where you’re fairly irregular that you get any meaningful di¡erences among the algorithms. It is in all of those where you don’t get the power that you do here. Veldhuis: This reaches a deep issue, which I feel is an issue of scale. If you sample GH every 30 s you ¢nd micro-bursts that are summating into an overall big volley or peak (Holl et al 1991). Now at a certain scale, those bursts represent say 1000^ 10 000 cells more-or-less ¢ring together. The macro-bursts, however, represent a greater coherence of cell discharge of hormone, representing yet a larger population synchrony. In a sense there is almost a fractal element to the pulse analysis. If you go back even further, you come to single cell Ca2+ oscillations. Thus, these are events that are non-uniform over time, and are distributed over space. The scale is a tremendous challenge to us at this meeting. One of the things that I enjoyed yesterday is the scale of the quick feedback at the intracellular signalling level, on up to the macroscopic scale.
References Holl RW, Hartman ML, Veldhuis JD, Taylor WM, Thorner MO 1991 Thirty-second sampling of plasma growth hormone in man: correlation with sleep stages. J Clin Endocrinol Metab 72:854^861 Roelfsema F, Van den Berghe G, van Dulken H, Veldhuis JD, Pincus SM 1998 Pituitary apoplexy in acromegaly. A long-term follow-up study on two patients. J Endocrinol Invest 21:298^303
118
DISCUSSION
Van Cauter E, Refeto¡ S 1985 Evidence for two subtypes of Cushing’s disease based on the analysis of episodic cortisol secretion. N Engl J Med 312:1343^1349 Van den Berghe G, Pincus SM, Frolich M, Veldhuis JD, Roelfsema F 1998 Reduced disorderliness of growth hormone release in biochemically inactive acromegaly after pituitary surgery. Eur J Endocrinol 138:164^169
Therapeutic implications of circadian rhythms in cancer patients Francis Le¤ vi Laboratoire Rythmes Biologiques et Chronothe¤ rapeutique, Institut du Cancer et d’Immunoge¤ ne¤ tique, Ho“ pital Paul Brousse, 94800 Villejuif, France Abstract. Drug absorption, transport, metabolism and/or elimination usually show 24 h changes in both laboratory rodents and human beings. These variations in target cell exposure to drugs, as well as the rhythms which modulate cellular detoxi¢cation functions, account for the chronopharmacology of most medications, including anticancer agents, and have warranted the exploration of the relevance of the chronotherapy principle. Most of the cellular detoxication rhythms appear to be coupled to the rest^activity cycle, both in nocturnally active rodents and in diurnally active healthy subjects as well as in cancer patients. For instance, a 24 h rhythm was found in the activity of dehydropyrimidine dehydrogenase (DPD), both in rodent liver and in human circulating mononuclear cells, with a maximum located in the early rest span in either species. This cellular enzyme catabolizes 5-£uorouracil (5-FU), hence protects normal cells against damage produced by this widely used antimetabolite drug. Although DPD amplitude was nearly threefold in rodent liver and 40% in human lymphocytes, the adaptation of 5-FU administration to this rhythm largely improved tolerability both in rodents and in patients. The results thus support the coupling of the DPD rhythm and other chronopharmacology mechanisms to the average rest^activity cycle across species. The clinical relevance of such group chronotherapy has been further validated in Phase I, II and III clinical trials involving nearly 1500 patients. Multicentre randomized clinical trials have demonstrated that chronotherapy was both better tolerated and more e¡ective than constant rate infusion in patients with metastatic colorectal cancer. Nevertheless 24 h rhythms in plasma cortisol or rest^activity could be altered in nearly 30% of cancer patients. Results from a prospective study performed in 200 patients with metastatic colorectal cancer indicated that poor circadian coordination constitutes an independent prognostic factor of both treatment tolerability and e⁄cacy. Novel chronotherapeutic approaches targeted at circadian system coordination should then be devised in these patients. 2000 Mechanisms and biological signi¢cance of pulsatile hormone secretion. Wiley, Chichester (Novartis Foundation Symposium 227) p 119^142
Circadian rhythms Biological functions of living beings are organized along a 24 h time-scale. These circadian rhythms are endogenous, since they persist in constant environmental conditions. 119
120
LE¤VI
The regular alternation of light and darkness over 24 h is a potent synchronizer of the circadian system. It calibrates the endogenous period to precisely 24 h through the e¡ects of light and melatonin, a hormone mostly secreted by the pineal gland during darkness (Touitou & Haus 1992, Klein et al 1991). Under such synchronization, mammals with normal circadian function display circadian rhythms in cellular metabolism and proliferation with predictable amplitude and times of peak and trough (Fig. 1). These rhythms in£uence anticancer drug pharmacology and ultimately tolerability and/or antitumour e⁄cacy of cancer treatments (Le¤ vi 1997). Conversely, a lack of synchronization, a de¢ciency in its perception, or an alteration of circadian clock function make rhythm peaks and troughs unpredictable, and may require speci¢c measures for chronotherapy to improve therapeutic index (Depre¤ s-Brummer et al 1998). The rest^activity cycle is one of the most obvious rhythm. Its endogenicity was demonstrated as it persisted in constant environmental conditions in £ies, rodents and humans. This rhythm is controlled by a number of genes, including the per gene in Drosophila and by the clock gene in mouse. Direct pharmacological actions targeted at the suprachiasmatic nucleus (SCN) level in rodents can modify the timing of biochemical or molecular rhythms in these nuclei. These changes translate into a phase shift of the rest^activity cycle of these animals. These and other experimental facts (see above) clearly demonstrate the dependency of this latter rhythm upon SCN function. The easy recording of this rest^activity cycle has further supported its use as a reference rhythm for circadian timing of medications and more recently for assessing circadian system function.
Implications of rhythms for cancer therapy: group chronotherapy Experimental and clinical prerequisites Chronopharmacology of anticancer drugs in laboratory rodents. The circadian time of administration of over 30 anticancer agents in£uences the extent of toxicity and anticancer activity in mice or rats, kept under controlled circadian synchronization (Le¤ vi 1997). The latter usually consists of an alternation of 12 h of light and 12 h of darkness (Fig. 2). For all these drugs, the dosing time-related di¡erence in toxicity of the same dose usually ranges from two- to eightfold.These rhythms in drug tolerability result from circadian changes in drug pharmacokinetics and/or susceptibility rhythms of target tissues. The cellular rhythms in enzymatic activities and those involved into the cell cycle regulation appear as the main determinants of the chronopharmacology of anticancer drugs (Zhang et al 1993, Li et al 1997, Ohdo et al 1997,Tampellini et al 1998) (Table 1). Quite strikingly, the administration of a drug at the circadian time when it is best tolerated has usually achieved the best antitumour activity. This was found for
CIRCADIAN RHYTHMS IN CANCER THERAPY
121
FIG. 1. Schematic view of the circadian system. The suprachiasmatic nucleus (SCN) is a biological clock located at the £oor of the hypothalamus. It is able to maintain an approximate 24 h cycle in its electrical activity in vitro. Its period (cycle duration) is calibrated by the alternation of light (directly) and darkness (through melatonin secretion by the pineal gland). The SCN controls or coordinates the circadian rhythms in the body. The main circadian rhythm is the rest^activity cycle. Cellular metabolism and proliferation also display rhythms in normal tissues, which may be keyed to the rest^activity cycle. Chronopharmacologic intervention can consist of the adaptation of drug delivery to circadian rhythms or of the modulation of circadian mechanisms, or both. The former approach is the only one that has been used so far for the chronomodulation of cancer chemotherapy. In this case, chemotherapy delivery was increased when normal tissue proliferation was at its low point along the 24 h time-course. RHT, retinohypothalamic tract.
122
LE¤VI
FIG. 2. Circadian rhythms in anticancer drug tolerability in laboratory mice or rats. The least toxic dosing time is indicated for each cytostatic or immunologic agent as a function of the light^ dark schedule, the main synchronizer of the rest^activity circadian cycle.
antimetabolites, such as 5-£uorouracil (5-FU) or arabinofuranosylcytosine, for intercalating agents such as doxorubicin, for alkylating drugs such as melphalan or cisplatin and for antimitotic drugs such as vinorelbine or docetaxel (Le¤ vi 1997, Tampellini et al 1998, Filipski et al 1999). This improvement in e⁄cacy has usually been achieved, because drug doses could be safely and selectively increased by 30^ 50% at the circadian time of best tolerability. In summary, the experimental model tells us that the circadian rhythm in drug tolerability can be used for two purposes. An improvement in quality of life can result from the dosing time-related reduction of chemotherapy toxicity, while dose and e⁄cacy remain similar to standard schedules of delivery. An improvement in survival can result from the administration of a higher maximum tolerated dose at the least toxic circadian time as compared to other dosing times. From mice to cancer patients: coupling of cellular rhythms to the rest^activity cycle. Cells engaged in DNA synthesis usually display an increased susceptibility to antimetabolites or intercalating agents. The proportions of bone marrow, gut,
CIRCADIAN RHYTHMS IN CANCER THERAPY
123
TABLE 1 Cellular determinants of circadian rhythms in tolerability for cancer chemotherapy Biological function
Drug
Reduced glutathione Non-protein sulphydril groups
Cisplatin, oxaliplatin
Enzymatic activities Dehydropyrimidine dehydrogenase Deoxythymidine kinase
5-£uorouracil, £oxuridine
Dehydrofolate reductase Topoisomerase I O6-alkylguanine methyltransferase
Methotrexate Irinotecan Cystemustine
Cellular proliferation DNA synthesis (S-phase)
Bcl-2 expression
5-£uorouracil, theprubicin, irinotecan, docetaxel Docetaxel
skin and oral mucosa cells engaged into this S-phase of the cell division cycle vary by 50% or more along the 24 h time-scale in healthy human subjects. For all these tissues, lower mean values occur between midnight and 04:00, and higher mean values between 08:00 and 20:00 (Smaaland et al 1991, Buchi et al 1991, Bjarnason et al 1999, reviewed in Le¤ vi 1997). Dehydropyrimidine dehydrogenase (DPD) is the rate-limiting enzyme of 5-FU catabolism. Its activity in circulating mononuclear cells increases by nearly 50% between 10:00 and midnight, both in healthy subjects and in cancer patients (Harris et al 1990, reviewed in Le¤ vi 1997). These mechanisms of anticancer drug chronopharmacology display a similar phase relationship with the rest^activity cycle in mice and in humans, despite the fact that the former are active at night and the latter during daytime (Fig. 3). For instance, DPD activity peaks during early light in mice or rats and at early night in humans. Similarly, the proportion of S-phase bone marrow cells peaks in the second half of darkness in mice and near 16:00 in humans. In addition, constant rate infusion of 5-FU results in a circadian rhythm in plasma level both in mice and in cancer patients. Peak concentration in 5-FU occurs in the early rest span in both species, if the drug is infused continuously over 1 week or less. Such apparent coupling between the circadian rest^activity cycle and several chronopharmacology mechanisms across species has been the basis for the current chronotherapy schedules which have been given to cancer patients.
124
LE¤VI
FIG. 3. Coupling between chronopharmacology mechanisms and the rest^activity cycle in nocturnally active rodents (left) and diurnally active humans (right). The arrow indicates the time of maximum in corresponding function. [5-FU], plasma concentration of 5-£uorouracil during constant rate infusion for several days; DPD, dehydropyrimidine dehydrogenase activity in the liver (mice or rats) and in circulating mononuclear cells (humans); AGT, O6alkylguanine methyltransferase activity in liver (mice) and in circulating mononuclear cells (humans); %S phase, proportion of cells in S-phase.
Multichannel pumps that are time-programmable have allowed us to test the clinical relevance of the chronotherapy principle in fully ambulatory patients, and now to administer this treatment modality routinely to these patients. In doing this, we assumed that patients exhibit similar circadian rhythms. This assumption was supported by the fact that groups of patients with metastatic or advanced breast, ovarian or colorectal cancer displayed signi¢cant circadian rhythms in blood cell counts and in plasma or serum concentrations of cortisol, liver enzymes and creatinine (Fig. 4) (Mormont et al 1998a, Benavides et al 1991). The magnitude of these average circadian changes and their relationship to the average rest^activity cycle were quite similar in cancer patients and in normal subjects. Group chronotherapy Two clinical trials compared the toxicity of two dosing times of anthracyclines and cisplatin in 30 patients with advanced ovarian cancer. Both studies have demonstrated that doxorubicin or theprubicin were better tolerated near 06:00 and cisplatin between 16:00 and 20:00 than 12h apart (Hrushesky 1985, Le¤ vi et al
CIRCADIAN RHYTHMS IN CANCER THERAPY
125
FIG. 4. Total serum cortisol (mean SEM) as a function of circadian sampling time in a group of 19 healthy subjects, 18 patients with metastatic colorectal cancer and 20 patients with advanced ovarian cancer. Rhythms were statistically validated with both analysis of variance and cosinor (after Mormont et al 1998a).
1990). Other limited-scale non-randomized trials were performed in patients with lung, breast or kidney cancer, with apparent improvement in therapeutic index. Nevertheless, it became clear that the clinical relevance of the chronotherapy principle had to be tested in a large patient population using the common clinical pharmacology methodology. Metastatic colorectal cancer is the second highest cause of cancer deaths in both genders and its prognosis is poor with conventional treatment methods. Time-programmable multichannel pumps allowed the sinusoidal delivery of two or three anticancer drugs in the patient’s home or during their usual activities. The protocols involved the
LE¤VI
126
TABLE 2 cancer Phase of study
Clinical trials of chronotherapy in 1275 patients with metastatic colorectal
Drug(s)
Schedule
No. of patients
Reference
I
5-FU 5-FU-LV l-OHP 5-FU-LV-l-OHP
5 d every 3 wks 5 d every 3 wks 5 d every 3 wks 4 d every 2 wks
35 34 25 114
Le¤ vi et al (1995) Garu¢ et al (1997) Caussanel et al (1990) F. Le¤vi, unpublished results
II ma m m m
5-FU-LV id. id. l-OHP 5-FU-LV-l-OHP id. id.
5 d every 3 wks 4 d every 2 wks 14 d every 4 wks 5 d every 3 wks 5 d every 3 wks 4 d every 2 wks 4 d every 2 wks
43 100 67 30 93 54 50
id. id.
4 d every 2 wks 4 d every 2 wks
62 90
Chollet et al (1994) Cure¤ et al (1998) Bjarnason et al (1998) Le¤ vi et al (1993) Le¤ vi et al (1992) Brienza et al (1993) Bertheault-Cvitkovic et al (1996) F. Le¤vi, unpublished results Le¤ vi et al (1999)
5-FU-LV-l-OHP £at vs. chrono id. Chrono 5-FU-LV l-OHP
5 d every 3 wks
92
Le¤ vi et al (1994)
5 d every 3 wks
186 200
m m III m m m
am,
Le¤ vi et al (1997) Giacchetti et al (1997, 1999a)
multicentric.
chronomodulated infusion of 5-FU and leucovorin (LV), the reference combination chemotherapy for this disease, eventually associated with oxaliplatin (l-OHP), a recent active drug. Maximum delivery rate of 5-FU and LV was scheduled at 04:00 for 5-FU and LV and at 16:00 for l-OHP, based upon an extrapolation from experimental data. Courses lasted 4 or 5 days and were repeated every 2 or 3 weeks (Fig. 4). The tolerability, maximum dose intensities and antitumour activity of these chronotherapy schedules were evaluated in Phase I, II and III clinical trials, involving over 1000 patients with metastatic colorectal cancer (Caussanel et al 1990, Le¤vi et al 1992, 1993, 1994, 1995, 1997, 1999, Garu¢ et al 1997, BertheaultCvitkovic et al 1996) (Table 2). We devised eight chronomodulated schedules with di¡erent peak times of chemotherapy delivery rate and allocated 114 patients with metastatic colorectal cancer to receive one of them. All the patients had failed on
CIRCADIAN RHYTHMS IN CANCER THERAPY
127
conventional chemotherapy. Severe toxicity (WHO grade 3^4) occurred in 21% of the patients receiving peak 5-FU at night (4:003 h) as compared to 46% of those treated with peak 5-FU infusion in the afternoon (13:003 h). Quite strikingly, the antitumour activity (objective response rate, from extramural review of computed tomography scans) was 41% in the best tolerated schedules, and 26% in the most toxic ones (F. Le¤ vi, unpublished results). Two randomized multicentre studies were performed in a total of 278 patients with previously untreated metastatic colorectal cancer. All the patients received the three-drug regimen either as a constant rate infusion or as a chronomodulated administration. Chronotherapy reduced the incidence of severe mucositis ¢vefold, halved that of functional impairment from peripheral sensory neuropathy and reduced threefold the incidence of grade 4 toxicity requiring hospitalization, as compared to the £at infusion regimen. This improvement in tolerability was accompanied with a signi¢cant increase in objective response rate from 29% to 51% (Table 3) (Le¤ vi et al 1994, 1997). Further dose escalation of chronotherapy was possible and achieved 66% of objective tumour responses in a multicentre setting (Bertheault-Cvitkovic et al 1996, Le¤ vi et al 1999). This antitumour e⁄cacy was three- to fourfold greater than that achieved with conventional regimens involving 5-FU and LV. The results clearly emphasize that group chronotherapy improves the therapeutic index of the chemotherapy of colorectal cancer. The high activity of chronotherapy further allowed the surgical resection of liver metastases in previously inoperable patients. This novel therapeutic strategy did result in longterm survival in this population with liver-only metastatic disease: median survival was not reached after 5 years, while usually no survivors remain after conventional treatment (Bismuth et al 1996, Giacchetti et al 1999b). Nevertheless, one can wonder whether the magnitude of improvement brought about by such ¢xed chronotherapy schedule varies according to individual circadian system function. Thus altered circadian rhythms could be observed in tumour-bearing animals and in cancer patients. TABLE 3 Main results from randomized multicentre trial comparing £at vs. chronomodulated chemotherapy in 186 patients with metastatic colorectal cancer E¡ect
Flat
Chrono
P
Hospitalization for toxicity Severe mucositis Functional impairment (peripheral sensory neuropathy) Tumour response450%
31 76 31
10 14 16
0.001 0.0001 0.01
29
51
0.003
Figures represent percentage of patients (After Le¤ vi et al 1997).
128
LE¤VI
Circadian system alterations during cancer processes Experimental data Rhythms with periods of about 24 h or shorter (ultradian rhythms, with a 12 h or an 8 h period for instance) have been documented in more than a dozen murine tumour models. These studies have indicated that the circadian periodicity in cellular proliferation indices or metabolic activity is usually retained in slowgrowing or well-di¡erentiated tumours, yet with reduced amplitude and sometimes a shift in phase. Conversely, the circadian organization tends to be lost and possibly replaced with an ultradian periodicity in rapidly growing or advanced-stage tumours. The presence of a cancer also altered the normal rhythms in plasma corticosterone, body temperature or bone marrow DNA synthesis in some rodent tumour models, but not in others (reviewed in Mormont & Le¤ vi 1997). Rhythms in human tumours As early as 1953, daily variations in the mitotic index of human mammary carcinoma and squamous or basal cell carcinomas were described, with interindividual variations (Voutilainen 1953, Tahti 1956). Reanalysis of these data validated a circadian rhythm in a group of six women with breast cancer, with a maximum near 15:00 h. Ultradian rhythms were found in a group of 31 patients with squamous or basal cell carcinoma (Garcia-Sainz & Halberg 1966). Progressive dampening of skin mitotic activity was also suggested in patients with actinic keratoses or skin cancer (Zagula-Mally et al 1979). More recently, cell cycle-related parameters of tumour cells and normal mesothelial cells were studied around the clock within the peritoneal lavage £uid from 30 patients with ovarian cancer (Klevecz et al 1987). A circadian maximum in DNA synthesis of both diploid and aneuploid tumour cells was found between 12:00 and 16:00 h. This time was almost 12 h out of phase with the peak of DNA synthesis in mesothelial cells. Ultradian rhythms, with 8 h and 12 h periods, were also documented in the aneuploid tumour cell population (Klevecz & Braly 1991). 24 h changes were described for DNA synthesis in malignant lymph nodes from 24 patients with non-Hodgkin’s lymphomas. The maximum occurred near midnight, while the peak of S phase was usually found between 12:00 h and 16:00 h in the bone marrow of healthy subjects (Smaaland et al 1993). Interestingly, when patients were classi¢ed according to tumour stage, a circadian rhythm in DNA synthesis was validated in the group of patients with an early-stage tumour, but not in the group of patients with stage IV lymphoma. This observation suggested a link between cancer stage and circadian rhythm alteration.
CIRCADIAN RHYTHMS IN CANCER THERAPY
TABLE 4
129
Rhythms in human tumours
Cancer type
Variable
Circadian rhythm
Peak di¡erent from normal tissue
Breast
Surface temperature P32 uptake Mitotic index
Yes Yes Yes
Yes Yes NS
Cervix
Surface temperature
Yes
Yes
Ovary
Cell cycle distribution Yes Yes (variable according to ploidy) Cell cycle distribution Yes Yes (variable according to stage)
Non-Hodgkin lymphoma
NS, not speci¢ed. (After Mormont & Le¤ vi 1997.)
Noteworthy observations on breast skin temperature indicated that circadian rhythms persisted on the surface of breast cancers that were slow growing and well di¡erentiated. Rhythms were however dampened, with maxima occurring 6 h earlier than in the non-cancerous breast. Conversely, fast growing, poorly di¡erentiated tumours displayed a shortening in the period of tumour temperature rhythm (Gautherie & Gros 1974). Another study on 14 patients also showed a phase advance of 6 h in the rhythm of skin temperature of the cancerous breast, as compared to the contralateral one (Mans¢eld et al 1973). Table 4 summarizes the main results published on human tumour circadian and ultradian rhythms.
Individual rhythms in cancer patients Although the mean proportion of bone marrow cells in S phase was signi¢cantly greater at noon as compared to midnight in 11 patients with advanced cancer, no time-related di¡erence was apparent in four of these patients, whose cortisol rhythm was suppressed or inverted (Smaaland et al 1992). Studies on cortisol and other blood circadian rhythm were performed in 51 patients with advanced or metastatic ovarian, breast or colorectal cancer, with a minimum of 10 blood samples collected over 36^48 h in order to also estimate individual rhythmicity (Mormont & Le¤vi 1997, Mormont et al 1998a, Touitou et al 1995, 1996). Once more a circadian rhythm was statistically validated for each group of patients (Fig. 4). Nevertheless, the 24 h rhythms in plasma cortisol and other variables
130
LE¤VI
FIG. 5. Chronomodulated chemotherapy with 5-£uorouracil (5-FU), leucovorin (LV) and oxaliplatin (l-OHP) as it is automatically delivered by programmable-in-time pumps to patients with colorectal cancer.
were prominent in some patients and apparently suppressed in others, despite the lack of any glucocorticoid medication (Fig. 6). These rhythm alterations were mostly found in patients with poor performance status (graded 2 to 4, according to the WHO scale) and/or large tumour burden. A large study was undertaken in 200 patients with metastatic colorectal cancer in order to estimate the incidence of circadian system alterations, as assessed from rest^activity and cortisol rhythms, in a population of patients eligible for clinical trials, i.e. with a performance status 4 2. This circadian system assessment was as little invasive as possible, and did not require hospitalization. Motor activity was continuously monitored for 3 days, using a wrist-worn actigraphy bracelet, and a blood sample was obtained at 08:00 and at 16:00 on two consecutive days in each patient. The rest^activity cycle can be appropriately measured with a 3 day recording. The strength of the circadian component was assessed with an autocorrelation coe⁄cient at 24 h (r24). The relative di¡erence between cortisol levels at 08:00 and at 16:00 had been shown to be a good estimator of the circadian amplitude of this rhythm, with 40% being a low normal limit (Mormont et al 1998a). Thirty percent of the 200 patients had an abnormal
CIRCADIAN RHYTHMS IN CANCER THERAPY
131
FIG. 6. Individual 24 h variation in plasma cortisol and total proteins and in circulating leukocytes in breast cancer patients with good performance status (NC) or poor performance status (JL). (After Mormont & Le¤vi 1997.)
132
LE¤VI
A
B
FIG. 7. Example of individual actigraphy records of two patients with metastatic colorectal cancer (after Mormont 1998). The dark vertical bars represent amount of activity per unit time. Activity records for two consecutive days in a patient with circadian rhythm (A) and in a patient with altered rhythmicity and lack of any reproducible pattern from one day to the next (B). Both the autocorrelation coe⁄cient (r24) and the dichotomy index (l50) constitute statistical estimates of circadian rhythmicity.
CIRCADIAN RHYTHMS IN CANCER THERAPY
133
cortisol rhythm using this criterion (Mormont 1998). The rest^activity pattern ranged from marked to completely disrupted 24 h rhythmicity (cf. examples in Fig. 7). Approximately 30% of the patients displayed a profoundly disturbed cycle, with r24 5 0.30 (Mormont 1998). Nevertheless, only a weak correlation was found between cortisol rhythm and rest^activity cycle alterations. This suggests that both rhythms are controlled by di¡erent circadian oscillators and/or circadian clock pathways.
Clinical relevance of individual circadian system function The status of the circadian system was ¢rst tested as an estimate of a cancer patient’s prognosis in two pilot studies, one involving 20 patients with advanced ovarian cancer and the other 13 patients with metastatic breast cancer. Signi¢cant correlations were found between individual circadian amplitude of plasma cortisol or circulating leukocyte count and well-known prognostic factors of response and survival (Benavides 1991). The above-mentioned study prospectively investigated the relevance of circadian system function for quality of life and survival in 200 patients with metastatic colorectal cancer. Results have indicated that the circadian distribution of activity was well correlated to several quality of life parameters from the EORTC QLQ-30 questionnaire and constituted a joint prognostic factor of survival, independently from the well-known prognostic factors in this disease (Mormont et al 1997, 1998b). The results suggest that circadian system function may play an important role for cancer patient outcome, an issue which deserves further investigation over the next years. Thus, speci¢c treatments of circadian dysfunctions may help improve the status and outcome of cancer patients, and contribute to enhance the therapeutic e⁄cacy of chronomodulated chemotherapy. The clinical relevance of chronotherapy for the outcome of cancer patients is currently being investigated along these lines within the EORTC for colon, pancreas and breast carcinomas.
References Benavides M 1991 Cancer avance¤ de l’ovaire: approche chronobiologique comme nouvelle strate¤ gie du traitement et de la surveillance clinique et biologique. PhD thesis, Universite¤ Paris XI, Paris, France Bertheault-Cvitkovic F, Jami A, Ithzaki M et al 1996 Biweekly intensi¢ed ambulatory chronomodulated chemotherapy with oxaliplatin, 5-£uorouracil and leucovorin in patients with metastatic colorectal cancer. J Clin Oncol 14:2950^2958 Bismuth H, Adam R, Le¤ vi F et al 1996 Resection of initially unresectable liver metastases from colorectal cancer following systemic chemotherapy. Ann Surg 224:509^522
134
LE¤VI
Bjarnason GA, Marsh R, Chu NM, Kerr I 1998 Phase II study of 5-£uorouracil (FU) by a 14 day chronomodulated infusion in patients with metastatic colorectal cancer. Proc Am Soc Clin Oncol 17:275a Bjarnason GA, Jordan RCK, Sothern RB 1999 Circadian variation in the expression of cell-cycle proteins in human oral epithelium. Am J Pathol 154:613^622 Brienza S, Le¤ vi F, Valori VM et al 1993 Intensi¢ed (every 2 weeks) chronotherapy with 5£uorouracil (5-FU), folinic acid (FA) and oxaliplatin (L-OHP) in previously treated patients with metastatic colorectal cancer. 29th Annual Meeting Am Soc Clin Oncol, Orlando, FL, USA, May 16^18. Proc Am Soc Clin Oncol 12:197 (abstr) Buchi KN, Moore JG, Hrushesky WJM, Sothern RB, Rubin NH 1991 Circadian rhythm of cellular proliferation in the human rectal mucosa. Gastroenterology 101:410^415 Caussanel JP, Le¤vi F, Brienza S et al 1990 Phase I trial of 5-day continuous venous infusion of oxaliplatin at circadian rhythm-modulated rate compared with constant rate. J Natl Cancer Inst 82:1046^1050 Chollet Ph, Cure¤ H, Garui C et al 1994 Phase II trial with chronomodulated 5-£uorouracil (5-FU) and folinic acid (FA) in metastatic colorectal cancer. Proceedings 6th International conference in Chronopharmacology and Chronotherapy, Amelia Island, FL, USA, July 5^9 1994, abstract VIIIb-4 Cure¤ H, Adenis A, Tubiana-Mathieu N et al 1998 Phase II trial of chronomodulated (cm) high dose 5-£uorouracil (5-FU) and l-folinic acid (l-FA) in patients with metastatic colorectal cancer (MCC). Proceedings 34th Annual Meeting Am Soc Clin Oncol, Los Angeles, CA, USA, May 16^19 1998, abstract 1048 Depre¤ s-Brummer P, Metzger G, Le¤ vi F 1998 Pharmacologic restoration of suppressed temperature rhythms in rats by melatonin, melatonin receptor agonist S20242 or 8-hydroxy2-(di-n-propylamino)tetralin (8-OH-DPAT). Eur J Pharmacol 347:57^66 Filipski E, Amat S, Lemaigre G, Vincenti M, Breillout F, Le¤ vi F 1999 Relationship between circadian rhythm of vinorelbine toxicity and e⁄cacy in p388-bearing mice. J Pharm Exp Ther 289:231^235 Garcia-Sainz M, Halberg F 1966 Mitotic rhythms in human cancer, reevaluated by electronic computer programs. Evidence for chronopathology. J Natl Cancer Inst 37:279^292 Garu¢ C, Le¤ vi F, Aschelter AM et al 1997 A phase I trial of 5-day chronomodulated infusion of 5£uorouracil and 1-folinic acid in patients with metastatic colorectal cancer. Eur J Cancer 33:1566^1571 Gautherie M, Gros C 1974 Circadian rhythm alteration of skin temperature in breast cancer. Chronobiologia 4:1^17 Giacchetti S, Zidani R, Perpoint B et al 1997 Phase III trial of 5-£uorouracil (5-FU), folinic acid (FA), with or without oxaliplatin (OXA) in previously untreated patients (pts) with metastatic colorectal cancer (MCC). Proceedings 33rd Annual Meeting Am Soc Clin Oncol May 17^20 Denver, CO, USA, abstract 805 Giacchetti S, Perpoint B, Zidani R et al for the International Organization of Cancer 1999a Phase III multicenter randomized trial of oxaliplatin addition to chronomodulated 5-£uorouracil^ leucovorin as ¢rst line treatment of metastatic colorectal cancer. J Clin Oncol, in press Giacchetti S, Itshaki M, Gruia G et al 1999b Long term survival of patients with unresectable colorectal liver metastases following infusional chemotherapy with 5-£uorouracil, folinic acid, oxaliplatin and surgery. Ann Oncol 10:1^7 Harris B, Song R, Soong S, Diasio RB 1990 Relationship between dihydropyrimidine dehydrogenase activity and plasma 5-£uorouracil levels with evidence for circadian variation of plasma drug levels in cancer patients receiving 5-£uorouracil by protracted continuous infusion. Cancer Res 50:197^201 Hrushesky W 1985 Circadian timing of cancer chemotherapy. Science 228:73^75
CIRCADIAN RHYTHMS IN CANCER THERAPY
135
Klein DC, Moore RY, Reppert SM 1991 Suprachiasmatic nucleus: the mind’s clock. Oxford University Press, Oxford Klevecz R, Braly P 1991 Circadian and ultradian cytokinetic rhythms of spontaneous human cancer. Ann NY Acad Sci 618:257^276 Klevecz R, Shymko R, Blumenfeld D, Braly P 1987 Circadian gating of S phase in human ovarian cancer. Cancer Res 47:6267^6271 Le¤ vi F 1997 Chronopharmacology of anticancer agents. In: Redfern PH, Lemmer B (eds) Physiology and pharmacology of biological rhythms. Springer-Verlag, Berlin, p 299^331 Le¤ vi F, Benavides M, Chevelle C et al 1990 Chemotherapy of advanced ovarian cancer with 4’-Otetrahydropyranyl doxorubicin (THP) and cisplatin: a phase II trial with an evaluation of circadian timing and dose intensity. J Clin Oncol 8:705^714 Le¤ vi F, Misset JL, Brienza S et al 1992 A chronopharmacologic phase II clinical trial with 5£uorouracil, folinic acid and oxaliplatin using an ambulatory multichannel programmable pump. High antitumor e¡ectiveness against metastatic colorectal cancer. Cancer 69:893^900 Le¤ vi F, Perpoint B, Garu¢ C et al 1993 Oxaliplatin activity against metastatic colorectal cancer. A phase II study of 5-day continuous venous infusion at circadian rhythm modulated rate. Eur J Cancer 29:1280^1284 Le¤ vi F, Zidani R, Vannetzel JM et al 1994 Chronomodulated versus ¢xed infusion rate delivery of ambulatory chemotherapy with oxaliplatin, 5-£uorouracil and folinic acid in patients with colorectal cancer metastases. A randomized multiinstitutional trial. J Natl Cancer Inst 86:1608^1617 Le¤ vi F, Soussan A, Adam R et al 1995 A Phase I^II trial of ¢ve-day continuous intravenous infusion of 5-£uorouracil delivered at circadian rhythm modulated rate in patients with metastatic colorectal cancer. J Infus Chemother 5:153^158 Le¤ vi F, Zidani R, Misset JL 1997 Randomized multicentre trial of chronotherapy with oxaliplatin, £uorouracil, and folinic acid in metastatic colorectal cancer. International Organization for Cancer Chronotherapy. Lancet 350:681^686 Le¤ vi F, Zidani R, Brienza S et al 1999 A multicentre evaluation of intensi¢ed ambulatory chronomodulated chemotherapy with oxaliplatin, £uorouracil and leucovorin as initial treatment of patients with metastatic colorectal cancer. International Organization for Cancer Chronotherapy. Cancer 85:2532^2540 Li XM, Metzger G, Filipski E et al 1997 Pharmacologic modulation of reduced glutathione circadian rhythms by buthionine sulfoximine: relationship with cisplatin toxicity in mice. Toxicol Appl Pharmacol 143:281^290 Mans¢eld CM, Carabasi RA, Wells W, Borman K 1973 Circadian rhythm in the skin temperature of normal and cancerous breasts. Int J Chronobiol 1:235^243 Mormont C 1998 Se¤ lection et validation d’un test pre¤ dictif de la survie des patients atteints de cancer colorectal me¤ tastase¤ , fonde¤ sur l’alte¤ ration du syste' me circadien. PhD thesis, Universite¤ Paris XI, Paris, France Mormont MC, Le¤ vi F 1997 Circadian-system alterations during cancer processes: a review. Int J Cancer 70:241^247 Mormont MC, Bleuzen P, Lellouch J et al 1997 Prognostic value of circadian rhythm assessment for survival of patients with metastatic colorectal cancer. Proceedings of the 33rd Annual Meeting of the American Society of Clinical Oncology, Denver, CO, USA, May 17^20 1997, abstract 956 Mormont MC, Hecquet B, Bogdan A, Benavides M, Touitou Y, Le¤vi F 1998a Non-invasive estimation of the circadian rhythm in serum cortisol in patients with ovarian or colorectal cancer. Int J Cancer 78:421^424 Mormont MC, Giacchetti S, Waterhouse J et al 1998b Activity circadian rhythm as an independent prognostic factor of survival in patients (pts) with metastatic colorectal cancer
136
DISCUSSION
(MCC). Proceedings of the 34th Annual Meeting of the American Society of Clinical Oncology, Los Angeles, CA, USA, May 16^19 1998, abstract 1032 Ohdo S, Makinosumi T, Ishizaki T et al 1997 Cell cycle-dependent chronotoxicity of irinotecan hydrochloride in mice. J Pharmacol Exp Ther 283:1383^1388 Smaaland R, Laerum OD, Lote K, Sletvold O, Sothern RB, Bjerknes R 1991 DNA synthesis in human bone marrow is circadian stage dependent. Blood 77:2603^2611 Smaaland R, Abrahamsen JF, Svardal AM, Lote K, Ueland PM 1992 DNA cell cycle distribution and glutathione (GSH) content according to circadian stage in bone marrow of cancer patients. Br J Cancer 66:39^45 Smaaland R, Lote K, Sothern RB, Laerum OD 1993 DNA synthesis and ploidy in nonHodgkin’s lymphomas demonstrate intrapatient variation depending on circadian stage of cell sampling. Cancer Res 53:3129^3138 Tahti E 1956 Studies of the e¡ect of X-irradiation on 24 hour variations in the mitotic activity in human malignant tumors. Acta Path Microbiol Scand 177:1^61 Tampellini M, Filipski E, Liu X-H et al 1998 Docetaxel chronopharmacology in mice. Cancer Res 58:3896^3904 Touitou Y, Haus E 1992 Biological rhythms in clinical and laboratory medicine. SpringerVerlag, Berlin Touitou Y, Le¤ vi F, Bogdan A, Benavides M, Bailleul F, Misset JL 1995 Rhythm alteration in patients with metastatic breast cancer and prognostic factors. J Cancer Res Clin Oncol 121:181^188 Touitou Y, Bogdan A, Le¤ vi F, Benavides M, Auze¤ by A 1996 Disruption of the circadian patterns of serum cortisol in breast and ovarian cancer patients: relationships with tumor marker antigens. Br J Cancer 74:1248^1252 Voutilainen A 1953 Uber di 24-stunden-rhythmik der mitozfrequenz in malignen tumoren. Acta Path Microbiol Scand (suppl) 99:1^104 Zagula-Mally ZW, Cardoso SS, Williams D et al 1979 Time point di¡erences in skin mitotic activity of actinic keratoses and skin cancers. In: Reinberg A, Halberg F (eds) Chronopharmacology. Pergamon Press, New York, p 399^402 Zhang R, Lu Z, Liu T et al 1993 Relationship between circadian-dependent toxicity of 5£uorodeoxyuridine and circadian rhythms of pyrimidine enzymes: possible relevance to £uoropyrimidine chemotherapy. Cancer Res 53:2816^2822
DISCUSSION Veldhuis: Clearly this ¢eld is moving along substantially and it is a delight to see your randomized prospective trials. For the most part, you are adjusting the timing of the dose to enhance tolerability. Do you have a strategy to enhance tumour killing which might involve taking out and assessing the rhythmicity of the cancer cells? At this point you are mainly trying to preserve the normal cells from toxicity, but could you ¢nd a vulnerable interval in the cancer cell biology and strike that? Le¤ vi: The problem with this approach is that cancer cell biology is very complex, and cancer cell chronobiology is largely unknown. Using very crude measurements such as the proportion of cells undergoing DNA synthesis, we can tell whether a tumour is well di¡erentiated and has retained circadian rhythmicity or whether it is undi¡erentiated and has lost rhythmicity. However, we must also remember that
CIRCADIAN RHYTHMS IN CANCER THERAPY
137
the tumour cells carry mutations of genes, most of which are involved in the regulation of the cell cycle. With regard to anticancer drug susceptibility, the relevance of a given cell cycle stage may thus be very di¡erent for a mutated tumour as opposed to a normal tissue. Normally there are checkpoints in the cell cycle when the cell can stop to repair DNA lesions or to die. Tumour cells have often lost this ability. We have studied vinorelbine, an antimitotic agent which stimulates p53 expression in a di¡erential manner according to its dosing time. The drug is less toxic in the activity span. If we now treat the mice with leukaemia cells with mutated p53, the tumour is unable to undergo p53dependent repair processes. This may explain why we have a good coincidence between the circadian rhythm in tolerability and that in antitumour activity (Filipski et al 1999). My feeling is that more knowledge of the tumour biology, even without chronobiology, will be helpful for directing the chronotherapy strategy you mention. Sassone-Corsi: You have to take into account the proliferation of these cancer cells. Depending on the stage of their tumorogenicity, they may proliferate in di¡erent ways. The link between the circadian clock and cell cycle is clearly interesting, but in a tumorous cell you may have no link with the circadian rhythm. Le¤ vi: The tumour cell is one thing, but the microenvironment of the tumour cells is at least as important as the tumour cell behaviour. We have tried to go with the most simple strategy so far, but we need to investigate further. Sassone-Corsi: Did you take into account that rodents are nocturnal and humans aren’t? Are you just reversing what you learn from mice and rats? Le¤ vi: There is a 12 h di¡erence in the circadian phase of several mechanisms involved in anticancer drug chronopharmacology between mice or rats and humans, if we consider rest onset as a phase reference. A simple formula which was used to extrapolate the ‘best’ time rodent data to humans is ‘rat + 12 h¼human’! Waxman: Obviously it’s useful to reduce the toxicity of these drugs in a cancer patient. But if you have the same chronobiology in the tumour as you do in the host tissues that you are trying to protect, then reducing toxicity for the patient may correspondingly decrease toxicity to the tumour. Are there rodent models in which you can address this? Le¤ vi: Most of the studies that have been performed in animal models have found a very close coincidence between the time of best tolerability and that of best antitumour e⁄cacy. Perhaps this is because these animal tumours have gene mutations which are responsible for an alteration of tumour circadian rhythms, as compared to the normal tissue they derive from. Waxman: In Fig. 2 you showed the clock and many drugs. I noted that for one drug that my laboratory has been studying, ifosfamide, you indicated that the ideal time of drug administration is close to midnight. In the case of ifosfamide, there are
138
DISCUSSION
two alternative pathways of metabolism: one pathway, 4-hydroxylation, activates the drug, whereas a second, alternative pathway is an N-dechloroethylation reaction that detoxi¢es it. These two pathways are catalysed by di¡erent liver P450 enzymes. Do your data indicate that ifosfamide should be administered at midnight due to a more favourable expression of the alternative pathway’s P450 enzymes? Le¤ vi: This study was reported in 1981 by the team of M. Smolensky in Houston (Snyder et al 1981). No mechanistic analysis has been performed since then, to the best of my knowledge. Nevertheless, there are plenty of possible mechanisms, other than P450-related ones. Rhythms in cellular glutathione content which protect cells against cytotoxic damage (Li et al 1997) and rhythms in O6alkylguanine transferase activity which repairs alkyl lesions (Martineau-Pivoteau et al 1996), likely contribute to the ifosfamide tolerability rhythm. Matthews: I was slightly worried about your statements relating activity, disease and outcome. I wonder whether all you’re measuring is what the poet Yeats wrote: ‘When I am old and grey and full of sleep’. Those slumped in a chair, in the process of dying, wouldn’t be expected to be doing much activity. What you have described could simply be a very crude measure of what you would see from the end of the bed. Le¤ vi: You refer to the fact that the prognosis of cancer patients with a poor rest^activity circadian cycle is worse than for those with a marked rhythm. May I emphasize that these days many cancer patients can go to work and are active like most of us. Nevertheless, our ¢rst interpretation of these data was that we were indeed measuring the consequence of the tumour burden upon the patient. This is why we carefully examined the results of the multivariate analysis, which comprised not only the ‘classical’ prognostic factors of metastatic colorectal cancer, but also these novel rest^activity cycle parameters. Performance status of a patient is graded by the medical oncologist according to a scale ranging from 0 to 4, which was developed by the WHO. When a patient has a performance status of zero, it means that he or she can work. We found that the rest^activity cycle displayed the same predictive value in this latter category of patients as in the whole population of 200 patients. Thus, we feel that rest^ activity cycle recording provides additional information, as compared with the performance status of the patient, and possibly points towards a cancer-associated circadian clock disturbance. Licinio: Using this chronobiological approach, have you been able to increase the combination of drugs given at any one time? Usually in chemotherapy the number of drugs given is limited by toxicity, and this may open the way to new modes of chemotherapy. Le¤ vi: This hasn’t been studied in humans, although in the animal model, combinations of up to ¢ve drugs could be safely given, but only at speci¢c
CIRCADIAN RHYTHMS IN CANCER THERAPY
139
circadian dosing times. Here, we didn’t follow the approach of increasing the number of drugs, but instead increased the doses of the two or three drugs we gave. As I showed, this could be done only at speci¢c dosing times. Schaefer: In the last part of your paper you showed us patients with loss of motor activity rhythms. These are probably the patients with dampened circadian hormonal rhythms. You showed also the link between the loss of cortisol rhythm and DPD activity. Will those patients who have a loss of the motor activity rhythms be the ones least likely to bene¢t from chronotherapy? Le¤ vi: It’s too early to say. I would like to point out that those patients who have apparently lost or have a damped rest^activity cycle are not necessarily the same as those who have altered cortisol rhythms. We also have investigated the melatonin rhythm in these patients. It turns out that there is no common pattern of circadian system alteration in these patients, up to this point. This is why we are very interested in the possibility of having a method for assessing the coupling of several rhythms. Robinson: Listening to your ability to control the entry rate much more precisely, I wondered whether you might comment on whether the agents that you’re using might have been optimized partly for duration of action. Perhaps less e⁄cient versions might be more bene¢cial in the situation where you control that narrow temporal window of action. If what you want to do is to have a fast on and a fast o¡, you might consider that some of the agents that you are using have too broad a lifetime (or dwell time) at their target for your chronobiological way of delivering the agent. Perhaps short-acting, maybe less potent, agents in larger quantities might be more e¡ective when you’re trying to window-down their time of exposure. Le¤ vi: 5-FU has a short half-life of just 10^20 min. As a result, its plasma concentration follows quite closely the pattern of drug delivery, in particular if the latter is chronomodulated over 24 h. The same is not necessarily true for drugs with longer half-lives. However, what determines cellular injury, and ultimately toxicity, is not the plasma level of the drug, but its pharmacodynamics in the target cells. In animal models there were sometimes apparent discrepancies between drug chronopharmacokinetics and pharmacodynamics. For instance, the dosing time which produced a shorter half-life of free drug was associated with a lower toxicity of carboplatin but a higher toxicity of oxaliplatin, as compared to other dosing times (Boughattas et al 1994). Lightman: In looking at the rest^activity cycles, one of the things that occurred to me is that a proportion of your patients may well be very depressed. There is an increasing literature showing that patients who are depressed following myocardial infarction are much more likely to die than the ones who aren’t. This may also be true in oncology. Is there any correlation between the patients’ degree of depression and how they respond to your treatment?
140
DISCUSSION
Le¤ vi: The correlation between depression and response to treatment is not very clear. There was a good correlation between the parameters for rest^activity and depression state as assessed with the Hospital Anxiety^Depression scale. Nevertheless, it is di⁄cult to answer your question because those patients did not complain of depression, but rather this was documented in some patients using a systematic questionnaire. Goldbeter: Have you also addressed the possibility of using non-circadian chronotherapy? For example, the cell cycling time of some cancer cells may be less than 24 h. There have been some reports by groups from Israel and Russia, who use shorter chronotherapy for tumour control in animal models. Le¤ vi: We haven’t done this, but it could be one of the ways to further improve chronotherapy e⁄cacy. Indeed, if we are able to show that some tumours have a shorter circadian cycle, it could be a valid method to deliver drugs along both the circadian time-scale and a shorter period time-scale. Goldbeter: The idea that they proposed and tested theoretically some 15 years ago (Dibrov et al 1985, Agur 1986, Agur et al 1988) was rather simple. Suppose the cell cycle duration of cancer cells is 11 h, and in normal cells it is 17 h. They propose to give an antimitotic drug every 17 h. If the cells were synchronized this would kill the same fraction of normal cells, while progressively eliminating cancer cells. I wonder whether this approach has ever been tried clinically. Le¤ vi: Not to my knowledge. Nevertheless, this approach assumes the synchronization of cell cycle events in all cancer cells and in all the normal cells. I understand that this model does not involve any circadian control of cell cycle events, which we know is the case, at least. Goldbeter: I also have a question related to survival. You said that in patients, although the situation improved in many respects, there was no gain in survival: could you comment on that? Le¤ vi: There are two reasons which likely explain the lack of di¡erence in survival in the trial comparing constant rate versus chronomodulated delivery of chemotherapy. Patients with metastatic colorectal cancer who are left without treatment can spontaneously live from 3^24 months according to the initial tumour size, the number of sites involved and so on. So there is large variability in patient survival, which relates to the tumour and patient biology. Conversely, chemotherapy-induced tumour shrinkage is only determined by treatment, since no spontaneous shrinkage of tumour has been observed in metastatic colorectal cancer. As a result, the number of patients needed to demonstrate a di¡erence in survival is much larger than the number required to demonstrate a di¡erence in tumour response rate. In the multicentre trial you refer to, the main criterion was tumour response, the hypothesis being that it would be 20% greater in the chronotherapy arm, and indeed it was. The apparent lack of correlation with survival also stemmed from a unidirectional cross-over to chronotherapy in
CIRCADIAN RHYTHMS IN CANCER THERAPY
141
those patients in whom £at infusion had failed. We are currently assessing the survival issue further in a larger-scale multicentre study with a planned accrual of 554 patients. Ede¤ n: Coming back to the question about the chronobiology of the cancer cells versus the normal cells. Looking at your outcomes, side-e¡ects and tumour shrinkage, could you discriminate the consequences of the therapy on side-e¡ects versus tumour shrinkage, and in that way see if the aim to kill the cells could be discriminated from the aim to avoid side-e¡ects? This may enable you to further re¢ne your treatment regimes. I’m really asking whether you can discriminate the endpoints in terms of your treatments. Is it possible that within your data, you actually have results showing that you have a more speci¢c tumour killing e¡ect by pulsing the treatments, for example, that is di¡erent from the e¡ects on the normal cells? Le¤ vi: This possibility certainly exists, and it may account for the fact that the median survival achieved in the chronotherapy trials have always been longer than that reported with conventional administration schedules. Matthews: I think it might move in the opposite direction. If you say your tumour cells have chaotic chronobiology, and that the normal cells have a reasonable circadian rhythm, perhaps you could synchronize your normal cells even better by infusing arginine at some point in the day, for example. Then you give cisplatin, for example, at a time when the normal cells are protected. Le¤ vi: This is an alternative explanation, which can hardly be distinguished from the former put by Dr Ede¤n.
References Agur Z 1986 The e¡ect of drug schedule on responsiveness to chemotherapy. Ann NY Acad Sci 504:274^277 Agur Z, Arnon R, Schechter B 1988 Reduction of cytotoxicity to normal tissue by new regimens of cell-cycle phase-speci¢c drugs. Math Biosci 92:1^15 Boughattas NA, Hecquet B, Fournier C et al 1994 Comparative pharmacokinetics of oxaliplatin (L-OHP) and carboplatin (CBDCA) in mice with reference to circadian dosing time. Biopharm Drug Dispos 15:761^773 Dibrov BF, Zhabotinsky AM, Yu A, Neyfakh A, Orlova MP, Churikova LI 1985 Mathematical model of cancer chemotherapy. Periodic schedules of phase-speci¢c cytotoxic agent administration increasing the selectivity of therapy. Math Biosci 73:1^31 Filipski E, Amat S, Lemaigre G, Vincenti M, Breillout F, Le¤ vi FA 1999 Relationship between circadian rhythm of vinorelbine toxicity and e⁄cacy in P388-bearing mice. J Pharmacol Exp Ther 289:231^235 Li XM, Metzger G, Filipski E et al 1997 Pharmacologic modulation of reduced glutathione circadian rhythms with buthionine sulfoximine: relationship with cisplatin toxicity in mice. Toxicol Appl Pharmacol 143:281^290
142
DISCUSSION
Martineau-Pivoteau N, Cussac-Buchdahl C, Chollet P et al 1996 Circadian variation in O6-methylguanine-DNA methyltransferase activity in mouse liver. Anticancer Drugs 7: 703^709 Snyder NK, Smolensky MH, Hsi BP 1981 Circadian variation in the susceptibility of male Balb/ C mice to ifosfamide. Chronobiologia 8:33^44
Pathophysiology of human circadian rhythms Georges Copinschi*, Karine Spiegel*{, Rachel Leproult*{ and Eve Van Cauter*{ *Centre for the Study of Biological Rhythms and Laboratory of Experimental Medicine, Universite¤ Libre de Bruxelles, 808 Route de Lennik, B-1070 Brussels, Belgium, and {Department of Medicine, MC 1027, University of Chicago, 5841 South Maryland Avenue, Chicago, IL 60637, USA
Abstract. The 24 h pro¢les of hormonal secretions represent a good model for the study of the human circadian system. Diurnal hormonal variations generally re£ect the modulation of ultradian or pulsatile release at 1^2 h intervals by signals occurring at nearly 24 h periods and result from the interaction of an internal timekeeping system or circadian clock with the sleep^wake homeostasis and various environmental factors, including the light^dark cycle, periodic changes in activity levels and the meal schedule. This temporal organization is altered in many pathophysiological conditions, including ageing, sleep loss, night or shift work, jet lag, a¡ective disorders and endocrine diseases. Both photic and non-photic stimuli may a¡ect the regulation of the circadian pacemaker and, therefore, the diurnal pattern of hormonal secretions. Appropriately timed stimuli may induce either a phase-advance or a phase-delay of the circadian clock, according to the timing of administration. Phase-shifting e¡ects have been shown in humans for light and for dark pulses, physical exercise, melatonin and melatonin agonists, and benzodiazepine hypnotics. These results open new perspectives for the treatment of a variety of disorders involving dysregulation of the circadian rhythmicity. 2000 Mechanisms and biological signi¢cance of pulsatile hormone secretion. Wiley, Chichester (Novartis Foundation Symposium 227) p 143^162
The circadian clock A prominent feature of all living organisms is their high degree of temporal organization. Indeed, far from obeying the Claude Bernard’s concept of homeostasis, nearly every aspect of the internal environment of the organism undergoes pronounced £uctuations over the course of the 24 h day, in synchrony with the 24 h periodicities in the physical environment. However, these daily or diurnal rhythms are not simply a response to the environmental periodicities imposed by celestial mechanics, but instead are generated by an internal timekeeping system, often referred to as the circadian clock system (for review, see 143
144
COPINSCHI ET AL
Van Cauter & Turek 1995). The term circadian is derived from the Latin circa diem (meaning ‘approximately one day’) and was coined because laboratory studies revealed that in the absence of any environmental time cues, the period of the endogenous rhythm generated by the circadian pacemaker is rarely exactly 24 h. In the present chapter, the term ‘circadian rhythm’ will refer to the endogenous 24 h periodicities generated by the circadian clock while the term ‘diurnal rhythm’ will refer to the 24 h periodicities which re£ect the superposition of various internal and external stimuli on the circadian signal. In mammals, two small bilaterally paired nuclei located in the anterior hypothalamus immediately above the optic chiasm, and therefore called the suprachiasmatic nuclei (SCN), function as a master circadian pacemaker which is primarily responsible for the generation and the entrainment of all circadian rhythms of the body (for review, see Turek et al 1995). Recent studies have begun to elucidate the molecular and genetic mechanisms underlying the circadian rhythmicity. The ¢rst mammalian gene, called CLOCK, was identi¢ed and sequenced two years ago (King et al 1997). Since then, three more mammalian genes have been identi¢ed based on sequence homologies with known clock genes in the fruit £y Drosophila. This endogenous circadian clock needs to be synchronized and entrained by environmental signal(s) to ensure an optimal adaptation of behaviour and physiological functions to environmental conditions. Such synchronization is essential for the survival of the species. Otherwise, even a clock with a period only a few minutes longer or shorter than 24 h would soon be totally desynchronized from the environment. In addition, the circadian clock also organizes the internal milieu so that internal changes occur in coordination with one another. Such internal temporal organization is crucial to the health and wellbeing of the organism. In most species, including humans, the light^dark and rest^activity cycles constitute the major synchronizing environmental agents. Light^dark information reaches the SCN from the retina via the retinohypothalamic tract. Photic information is then transmitted from the SCN to the pineal gland to regulate melatonin rhythmicity, which in turn exerts synchronizing e¡ects on the circadian clock as an indirect photic Zeitgeber (for review, see Van Cauter & Turek 1999). Circadian rhythmicity plays an important role in the timing of sleep onset and o¡set, the distribution of rapid eye movement (REM) sleep, and sleep spindle activity. In contrast, non-REM sleep, in particular slow-wave (SW) sleep, is primarily controlled by a recovery process dependent on the duration of prior wakefulness, often referred to as the homeostatic component. The level of this component rises during waking and decays exponentially during sleep (for review, see Van Cauter & Turek 1999).
PATHOPHYSIOLOGY OF CIRCADIAN RHYTHMS
145
Diurnal hormonal pro¢les as a model for the study of circadian rhythmicity Many diurnal hormonal patterns are dependent, to some degree, on the circadian clock. The relative contributions of circadian rhythmicity vs. homeostatic control in the temporal organization of hormonal release di¡er from one endocrine axis to another. For most pituitary hormones, the 24 h pro¢les re£ect the superposition of circadian signals on an ultradian, or pulsatile, release and result from the interaction of the circadian clock with sleep^wake homeostasis. Several rhythmic and nonrhythmic factors, such as periodic food intake, postural changes and levels of physical activity may also exert modulatory e¡ects on diurnal hormonal secretory patterns. It had been suggested in early studies that diurnal pro¢les of certain hormones, such as growth hormone (GH) and prolactin, were entirely regulated by the sleep^ wake cycle, without any e¡ect of circadian rhythmicity, while other hormones, such as cortisol, were entirely dependent on the circadian clock, without any in£uence of sleep^wake homeostasis. However, current evidence indicates that both sleep and circadian inputs may be evidenced in the diurnal pro¢les of most hormones, but that their relative contribution is di¡erent for each hormone (for review, see Van Cauter 1995). To di¡erentiate the e¡ects of sleep^wake homeostasis from the e¡ects of circadian rhythmicity, experimental strategies involving large abrupt shifts of sleep^wake and light^dark cycles were developed. Indeed, the circadian pacemaker needs several days to adjust to such shifts. Thus, these studies allow for the e¡ects of circadian rhythmicity to be observed in the absence of sleep and for the e¡ects of sleep to be observed at an abnormal circadian time. Statistical methods have been developed to better analyse circadian changes independently of more rapid £uctuations (Cleveland 1979, Van Cauter 1979). These methods are used to build best-¢t curves which allow us to quantify the period, the amplitude and the phase of the hormonal circadian variation. This is illustrated for cortisol and melatonin in Fig. 1. Phase reference points commonly used include the timing of the ¢tted maximum (or acrophase) and the timing of the ¢tted minimum (or nadir). The timing of the onset and the timing of the o¡set of the circadian rise are also often estimated, either from the actual circulating concentrations, or from the best-¢t curves (Fig. 1). The characteristics of diurnal variations of melatonin, cortisol, thyrotropin (TSH), GH and prolactin are brie£y reviewed in the next sections of this chapter.
Melatonin Not surprisingly, the 24 h pro¢le of plasma melatonin is a robust marker of the human circadian pacemaker (Rosenthal 1991). Daytime levels are stable and low.
146
COPINSCHI ET AL
FIG. 1. (Upper panel) Diurnal pro¢le of plasma cortisol obtained at 20 min intervals in a normal subject and quanti¢ed by a best-¢t curve shown in dashed line. The acrophase, nadir and amplitude of the circadian variation were de¢ned using the best-¢t curve rather than the raw data. The quiescent period of cortisol was de¢ned as starting when plasma concentrations lower than 50 ng/ml were observed for at least one hour and ending when concentrations higher than 50 ng/ml were observed for at least one hour. The onset of the circadian rise was de¢ned as the start of the ¢rst signi¢cant pulse after the nocturnal nadir. (Lower panel) Diurnal pro¢le of plasma melatonin obtained at 20 min intervals in a normal subject and quanti¢ed by a best-¢t curve shown in dashed line. The acrophase of the circadian variation was de¢ned using the best-¢t curve rather than the raw data. The onset of the melatonin circadian rise was de¢ned as the timing of the ¢rst plasma level exceeding 10 pg/ml not followed by a return to lower concentrations before the acrophase. The melatonin o¡set was de¢ned as the timing of the last value occurring after the acrophase that exceeded 10 pg/ml.
PATHOPHYSIOLOGY OF CIRCADIAN RHYTHMS
147
The circadian rise starts in the evening, between 21:00 and 23:00, and the maximum normally occurs around the middle of the sleep period. Thereafter, melatonin levels return to low daytime values between 08:00 and 09:00 (Fig. 1). Melatonin rhythmicity is largely dependent on the circadian clock and appears to be una¡ected by sleep and non-photic stimuli (for review, see Van Cauter & Turek 1995). In contrast, and independently of its e¡ects on the clock, exposure to light of su⁄cient intensity (4200^500 lux) exerts immediate direct inhibitory e¡ects on the pineal gland, resulting in a dose-dependent suppression of nocturnal melatonin secretion (Lewy et al 1980).
Cortisol Diurnal plasma cortisol pro¢les re£ect the circadian pattern of adrenocorticotropic activity (which results, in turn, from periodic changes in level of stimulation by corticotropin-releasing hormone). The 24 h rhythm of plasma cortisol represents an excellent model to estimate the temporal organization of the corticotropic axis because of its reproducibility and large amplitude. Diurnal cortisol pro¢les show an early morning maximum around 07:00^08:00, declining levels during daytime, a prolonged period of minimal levels (sometimes referred to as the quiescent period) lasting approximately 4 h and centred around midnight, followed by an abrupt elevation referred to as the circadian rise during the last part of the night, towards the morning acrophase (Fig. 1) (for review, see Van Cauter & Turek 1995). This pattern is primarily controlled by circadian rhythmicity and is produced by modulation of the height of successive secretory pulses (Veldhuis et al 1990a). However, modulatory e¡ects are exerted by the sleep^wake homeostasis. Indeed, sleep onset is reliably associated with decreasing cortisol secretion (Weitzman et al 1983, Van Cauter et al 1991). Conversely, during the second part of the night, awakenings and particularly the ¢nal morning awakening are consistently followed by pulses of cortisol secretion (Van Cauter et al 1991). Nevertheless, adaptation of cortisol diurnal rhythmicity to abrupt shifts of the sleep^wake cycle takes several days and therefore the 24 h plasma cortisol pro¢le may be considered to be a robust marker of circadian timing (for review, see Van Cauter & Turek 1995). Alterations of the 24 h pro¢les of plasma cortisol constitute a hallmark of Cushing’s syndrome (for review, see Van Cauter & Turek 1995). As illustrated in Fig. 2, the diurnal cortisol variation is markedly dampened, or even abolished. While an important overlap exists over the major part of the 24 h span between individual cortisol levels recorded in normal subjects and in patients with Cushing’s syndrome, a good discrimination is attained in a 4 h interval centred around midnight, i.e. during the normal quiescent period. On the other hand,
148
COPINSCHI ET AL
FIG. 2. Mean (SEM) 24 h pro¢les of plasma cortisol for 60 normal subjects and 56 patients with Cushing’s syndrome. Note the important overlap between both groups except in a 4 h interval centred around midnight (from Van Cauter & Turek 1995, with permission).
when diurnal variations can still be detected, the phase of the cortisol circadian variation appears to be essentially una¡ected in this condition. Depressive illness may also be associated with abnormalities of the 24 h cortisol pro¢le. Interestingly, these alterations are similar, albeit more severe, than those observed in normal ageing. In patients with major endogenous depression in the acute phase of the illness, the quiescent period is markedly shortened and a phaseadvance of the cortisol circadian rhythmicity is frequently observed. These
PATHOPHYSIOLOGY OF CIRCADIAN RHYTHMS
149
alterations are normalized after successful antidepressant treatment, suggesting that they are state rather than trait dependent (Linkowski et al 1987). Thyrotropin (TSH) The 24 h pattern of plasma TSH levels appears to be generated by amplitude as well as frequency modulation of secretory pulses (Veldhuis et al 1990b). Daytime levels are low and relatively stable. The nocturnal rise starts in the early evening and culminates in a maximum around the beginning of the sleep period. Indeed, this pre-sleep elevation is interrupted by the onset of sleep and a progressive decline is observed throughout the sleep period towards low daytime values (Brabant et al 1990). Because the onset of this nocturnal rise occurs well before the time of sleep onset, it is considered to re£ect a circadian e¡ect. However, sleep clearly exerts an inhibitory in£uence on TSH secretion (Parker et al 1976). The decline in TSH levels normally observed following sleep onset does not occur in case of sleep deprivation (Parker et al 1976). Interestingly, when sleep occurs during daytime hours, TSH levels are not suppressed signi¢cantly below normal daytime levels (Hirschfeld et al 1996). Thus, the inhibitory action of sleep on TSH secretion appears to be essentially operative only when the circadian elevation has occurred. The sleeprelated TSH inhibition appears to be associated to SW stages (Goichot et al 1992). While the onset of the TSH nocturnal rise may be considered to be a robust marker of the circadian clock, the diurnal TSH pro¢le appears to be a good illustration of the interaction between sleep and circadian rhythmicity. Growth hormone (GH) It has been recognized for more than 30 years that GH secretion is markedly stimulated during sleep (Takahashi et al 1968). In normal adults, the 24 h pro¢le of plasma GH consists of stable low levels abruptly interrupted by secretory bursts (Takahashi et al 1968). In normal young men, the largest and most reproducible burst occurs shortly after sleep onset, while in normally cycling young women, daytime GH pulses are more frequent and the sleep-associated pulse, while also present, does not generally account for the majority of the 24 h secretion. The close association between sleep-onset and GH secretory bursts persists in subjects submitted to a variety of manipulations of the sleep^wake cycle, including phase advances and delays as well as interruptions followed by re-initiations of sleep, so that a shift of the sleep^wake cycle is immediately followed by a shift of the GH rhythm (Fig. 3). In night workers, the major GH secretory pulse still occurs at sleep onset, i.e. during the daytime. In subjects living in free-running conditions, i.e. temporal isolation without any environmental time cues, the sleep-onset associated GH release is largely maintained (for review, see Van Cauter et al
150
COPINSCHI ET AL
FIG. 3. Individual pro¢le of plasma GH obtained at 20 min intervals for 52 h in a normal young man. This 52 h period included 8 h of nocturnal sleep, 28 h of sleep deprivation and 8 h of daytime sleep. Note the immediate adaptation of GH secretion to the 12 h delay shift of the sleep^wake cycle, and the persistence of a modest GH pulse in the early part of the night during nocturnal sleep deprivation (data from Van Cauter et al 1998a).
1998a). Extensive evidence indicates that the sleep-associated stimulation of GH secretion results from the existence of a robust relationship between SW activity and GH release (Holl et al 1991,Van Cauter et al 1992, 1998a). While sleep is clearly a major determinant of the 24 h pro¢le of GH secretion, there is also evidence for the existence of a modulation of the somatotropic axis by circadian rhythmicity. During nocturnal sleep, the sleep-onset GH pulse is caused by a surge of hypothalamic GH-releasing hormone which coincides with a circadian period of relative somatostatin disinhibition. This circadian rhythmicity of somatostatinergic tone is likely to explain that modest increases in GH pulsatility persist during waking in the late evening and the early part of the night following abrupt delays of the sleep period (Ja¡e et al 1995, Van Cauter et al 1998a). Prolactin Under normal conditions, the 24 h pro¢le of prolactin levels follows a bimodal pattern, with minimal concentrations around noon, a modest increase in the afternoon and a major nocturnal elevation starting shortly after sleep onset to reach a maximum around mid-sleep (Van Cauter et al 1981). Diurnal prolactin variations are essentially regulated by the sleep^wake cycle (Sassin et al 1972). Sleep onset, irrespective of the time of the day, has a stimulatory e¡ect on
PATHOPHYSIOLOGY OF CIRCADIAN RHYTHMS
151
prolactin release (Van Cauter & Refeto¡ 1985). As for GH, a shift of the sleep^ wake cycle is immediately followed by a shift of the prolactin rhythm. However, the amplitude of the prolactin rise associated with daytime sleep is somewhat dampened and modest increases in prolactin levels persist during waking around the time of usual sleep onset, particularly in women, suggesting that prolactin secretion is also modulated by circadian rhythmicity (De¤ sir et al 1982, Waldstreicher et al 1996). Maximal stimulation is observed only when sleep and circadian e¡ect are superimposed. There is evidence for a temporal relationship between the sleep-associated prolactin secretion and SW activity (Spiegel et al 1995). Ageing of the circadian clock system Circadian rhythmicity persists in healthy elderly subjects, but ageing is associated with a signi¢cant 1.0^1.5 h phase advance and with a marked dampening of the amplitude of circadian signal (for review, see Van Cauter et al 1998b). This has been shown in particular for several hormonal pro¢les primarily controlled by the circadian pacemaker, such as melatonin, cortisol and TSH (van Coevorden et al 1991). These alterations may result, at least partially, from age-related changes in the circadian system. Morphological and neurochemical alterations consistent with both an earlier phase and a reduced amplitude of the circadian signal have been evidenced in the SCN of older animals, in some but not all studies. However, in healthy older humans, the endogenous period of the circadian pacemaker appears to be similar to that of young adults and, therefore, the advanced phase and lower amplitude of overt 24 h rhythms may re£ect age-related changes in entrainment mechanisms rather than intrinsic clock function. Thus, changes in life habits, e.g. absence of constraining professional and social schedules and diminished exposure to the synchronizing e¡ects of the light^dark and rest^activity cycles, may alter the circadian function in the elderly (for review, see Van Cauter et al 1998b). E¡ects of sleep loss on diurnal hormonal pro¢les Chronic sleep loss constitutes an increasingly common condition in industrialized societies, that is primarily due to socioeconomic and cultural pressures and a¡ects at least one-third of adults living in industrial societies. While sleep loss was generally thought to a¡ect mood and cognitive performances rather than physiological functions, it has recently been shown that partial sleep loss for several days is associated with signi¢cant and potentially deleterious alterations of endocrine function, including marked disturbances of the temporal organization of hormonal release in the hypothalamo^pituitary axes (Leproult et al 1997, Spiegel et al 1999). The normal nocturnal TSH elevation is markedly dampened and free thyroxin levels are elevated. The cortisol pro¢les display a shorter quiescent
152
COPINSCHI ET AL
period, due to a delay in its onset, with increased cortisol secretion in the afternoon and early evening, and a more abrupt early morning elevation (Fig. 4). This later disturbance suggests that sleep loss a¡ects the rate of recovery, i.e. the resiliency, of the hypothalamo^pituitary^adrenal axis to endogenous stimulation by circadian signals. Decreased resiliency of the hypothalamo^pituitary^adrenal axis is thought to accelerate the development of central and peripheral disturbances associated with glucocorticoid excess, such as memory de¢cits due to impaired hippocampal function (McEwen & Sapolsky 1995) and insulin resistance (Dallman et al 1993). Conditions of misalignment between the circadian pacemaker and the rest^activity cycle Misalignment between endogenous circadian rhythmicity and the rest^activity cycle occurs in particular after transmeridian £ights (resulting in the jet lag syndrome) and in shift work conditions. Both are associated with a variety of symptoms, including fatigue, sleep disturbances, subjective discomfort,
FIG. 4. Individual 24 h pro¢les of plasma TSH (top panels) and plasma cortisol (lower panels) in a representative subject. The black bars represent the sleep periods. In the sleep debt condition, the normal nocturnal elevation of plasma TSH levels was markedly dampened, and the overall 24 h mean TSH level was reduced (0.72 vs. 1.75 mU/ml). The quiescent period of cortisol secretion (the period during which plasma cortisol levels were below 50 ng/ml; indicated by a horizontal line) was shorter (685 vs. 730 min) largely due to a 40 min delay in its onset (at 17:45 vs. 17:05 h), and levels of cortisol in the afternoon and early evening were higher in the sleep debt condition than in the sleep credit condition (shaded area; 54 vs. 35 ng/ml).
PATHOPHYSIOLOGY OF CIRCADIAN RHYTHMS
153
gastrointestinal disorders, and reduced mental and psychomotor performance (for review, see Van Reeth 1998). The malaise partially results from the desynchrony between endogenous and environmental rhythms, but is also partially due to internal desynchrony because di¡erent physiological systems adapt to abrupt shifts of the sleep^wake and light^dark cycles at di¡erent rates. Occasional jet lag is not believed to have long term adverse e¡ects on health, although complete adaptation requires several days. After transmeridian £ights, all environmental cues are generally equally displaced in the same direction and cooperate to rapidly shift the circadian system. However, the re-entrainment rate di¡ers among variables. Adaptation is almost immediate for rhythms markedly modulated by sleep^wake homeostasis (e.g. GH and prolactin), but takes several days for rhythms which are strongly dependent on the circadian system (e.g. melatonin, cortisol) (for review, see Van Cauter & Turek 1995). Adaptation appears to occur faster after a delay (i.e. westward) shift than after an advance (i.e. eastward) shift, possibly because the endogenous human circadian period is slightly longer than 24 h. On the contrary, shift work (which is generally maintained over long periods) is a major health hazard, involving increased risks of cardiovascular and gastrointestinal disorders, reduced immune function, infertility and drug abuse. Shift workers sleep generally less per scheduled sleep period than daytime workers and are therefore in a condition of chronic sleep deprivation (for review, see Van Reeth 1998). Shift work usually creates conditions in which some Zeitgebers (e.g. the arti¢cial light^dark cycle, the rest^activity cycle) are shifted while others (e.g. the natural light^dark cycle, familial and social routines) are not. This situation of con£icting Zeitgebers almost never allows a complete shift of the circadian system. Thus, workers on permanent or rotating night shifts do not completely adapt to these schedules, leading to a chronic state of internal desynchrony (Roden et al 1993). Phase-shifting e¡ects of photic and non-photic stimuli Light may be used to reset human circadian rhythms under a variety of experimental conditions (Czeisler et al 1986, Van Cauter et al 1994). Appropriately timed light pulses induce either a phase-advance or a phase-delay of the circadian clock, according to the timing of administration, and may be used to facilitate adaptation to shift work or to jet lag. A schematic representation of the phase response curve of the mammalian circadian clock to light exposure is illustrated in Fig. 5. Brie£y, in humans, maximal e¡ects are observed around the circadian phase, i.e. at the timing of the nadir of the 24 h pro¢le of body temperature (around 05:00 h in young healthy subjects). Light exposures during the ¢rst part of the night and in the early morning result in a
154
COPINSCHI ET AL
180
Phase shifts (min)
120
60 n=15
n=14
n=15
0
60
120
180 0
6
12 18 Circadian time
24
FIG. 5. Schematic representation of the phase response curve of the circadian clock to light exposure in hamsters. Note that light is e¡ective in resetting the clock only during a relatively short time interval over the 24 h span, with a very rapid inversion of e¡ects (adapted from F. W. Turek, personal communication).
phase delay and in a phase advance, respectively. It has recently been shown that human circadian phase is also sensitive to dark exposure. Afternoon exposure of young normal subjects to a sleep^dark period results in an advance of melatonin onset (Van Cauter et al 1998c). Not surprisingly, phase-shifting e¡ects of melatonin (Lewy & Sack 1997) and melatonin agonists (Kruchi et al 1997) on human circadian rhythms parallel those of dark exposure. When melatonin is given to young adults in the afternoon or in the evening, a phase advance is observed. On the contrary, a phase delay is obtained when melatonin is administered in the early morning. We have recently demonstrated that a melatonin agonist is also able to phase-shift the circadian clock in older subjects (G. Copinschi, R. Leproult, A. Van Onderbergen, M. L’Hermite-Bale¤ riaux & E. Van Cauter, unpublished results). Other pharmacological and behavioural stimuli may also exert phase-shifting e¡ects on the human circadian pacemaker. This has been shown for triazolam, a short-acting benzodiazepine (O. Buxton, G. Copinschi, A. Van Onderbergen,
PATHOPHYSIOLOGY OF CIRCADIAN RHYTHMS
155
T. G. Karrison & E. Van Cauter, unpublished data), and for physical exercise (Van Reeth et al 1994). These data indicate that several agents could possibly be used to reset human circadian rhythms. This may open new perspectives for the treatment of a variety of disorders involving dysregulation of the circadian rhythmicity. Acknowledgements This work was supported in part by grants from the Belgian Fonds de la Recherche Scienti¢que Me¤ dicale, the Universite¤ Libre de Bruxelles and the U.S. National Institutes of Health (NIDDK DK-41814 and NIA AG-11412).
References Brabant G, Prank K, Ranft U et al 1990 Physiological regulation of circadian and pulsatile thyrotropin secretion in normal man and woman. J Clin Endocrinol Metab 70:403^409 Cleveland WS 1979 Robust locally weighted regression and smoothing scatterplots. J Am Stat Assoc 74:829^836 Czeisler CA, Allan JS, Strogatz SH et al 1986 Bright light resets the human circadian pacemaker independent of the timing of the sleep^wake cycle. Science 233:667^671 Dallman MF, Strack AL, Akana SF et al 1993 Feast and famine: critical role of glucocorticoids with insulin in daily energy £ow. Front Neuroendocrinol 14:303^347 De¤ sir D, Van Cauter E, L’Hermite M et al 1982 E¡ects of ‘jet lag’ on hormonal patterns. III. Demonstration of an intrinsic circadian rhythmicity in plasma prolactin. J Clin Endocrinol Metab 55:849^857 Goichot B, Brandenberger G, Sainio J, Wittersheim G, Follenius M 1992 Nocturnal plasma thyrotropin variations are related to slow-wave sleep. J Sleep Res 1:186^190 Hirschfeld U, Moreno-Reyes R, Akseki E et al 1996 Progressive elevation of plasma thyrotropin during adaptation to simulated jet lag: e¡ects of treatment with bright light or zolpidem. J Clin Endocrinol Metab 81:3270^3277 Holl RW, Hartmann ML, Veldhuis JD, Taylor WM, Thorner MO 1991 Thirty-second sampling of plasma growth hormone in man: correlation with sleep stages. J Clin Endocrinol Metab 72:854^861 Ja¡e CA, Turgeon DK, Friberg RD, Watkins PB, Barkan AL 1995 Nocturnal augmentation of growth hormone (GH) secretion is preserved during repetitive bolus administration of GHreleasing hormone: potential involvement of endogenous somatostatin a clinical research center study. J Clin Endocrinol Metab 80:3321^3326 King DP, Zhao Y, Sangoram AM et al 1997 Positional cloning of the mouse circadian Clock gene. Cell 89:641^653 Kra« uchi K, Cajochen D, M˛ri D, Graw P, Wirz-Justice A 1997 Early evening melatonin and S20098 advance circadian phase and nocturnal regulation of core body temperature. Am J Physiol 272:R1178^R1188 Leproult R, Copinschi G, Buxton O, Van Cauter E 1997 Sleep loss results in an elevation of cortisol levels the next evening. Sleep 20:865^870 Lewy A, Sack R 1997 Exogenous melatonin’s phase-shifting e¡ects on the endogenous melatonin pro¢le in sighted humans: a brief review and critique of the literature. J Biol Rhythms 12:588^594 Lewy AJ, Wehr TA, Goodwin M, Newsome DA, Markey SP 1980 Light suppresses melatonin secretion in humans. Science 210:1267^1269
156
COPINSCHI ET AL
Linkowski P, Mendlewicz J, Kerkhofs M et al 1987 24-hour pro¢les of adrenocorticotropin, cortisol, and growth hormone in major depressive illness: e¡ect of antidepressant treatment. J Clin Endocrinol Metab 65:141^152 McEwen BS, Sapolsky RM 1995 Stress and cognitive function. Curr Opin Neurobiol 5:205^216 Parker DC, Pekary AE, Hershman JM 1976 E¡ect of normal and reversed sleep^wake cycles upon nyctohemeral rhythmicity of thyrotropin: evidence suggestive of an inhibitory in£uence in sleep. J Clin Endocrinol Metab 43:318^329 Roden M, Koller M, Pirich K, Vierhapper H, Waldhauser F 1993 The circadian melatonin and cortisol secretion pattern in permanent night shift workers. Am J Physiol 265:R261^R267 Rosenthal NE 1991 Plasma melatonin as a measure of the human clock. J Clin Endocrinol Metab 73:225^226 Sassin J, Frantz A, Weitzman E, Kapen S 1972 Human prolactin: 24-hour pattern with increased release during sleep. Science 177:1205^1207 Spiegel K, Luthringer R, Follenius M et al 1995 Temporal relationship between prolactin secretion and slow-wave electroencephalographic activity during sleep. Sleep 18:543^548 Spiegel K, Leproult R, Van Cauter E 1999 Impact of sleep debt on metabolic and endocrine function. Lancet 354:1435^1439 Takahashi Y, Kipnis DM, Daughaday WH 1968 Growth hormone secretion during sleep. J Clin Invest 47:2079^2090 Turek FW, Pinto LH, Vitaterna MH, Penev PD, Zee PC, Takahashi JS 1995 Pharmacological and genetic approaches for the study of circadian rhythms in mammals. Front Neuroendocrinol 16:191^223 Van Cauter E 1979 Method for characterization of 24-h temporal variation of blood components. Am J Physiol 237:E255^E264 Van Cauter E 1995 Hormones and sleep. In: Kales A (ed) The pharmacology of sleep. SpringerVerlag, Berlin, p 279^306 Van Cauter E, Refeto¡ S 1985 Multifactorial control of the 24-hour secretory pro¢les of pituitary hormones. J Endocrinol Invest 8:381^391 Van Cauter E, Turek FW 1995 Endocrine and other biological rhythms. In: DeGroot LJ (ed) Endocrinology, 3rd edn. WB Saunders, Philadelphia, PA, p 2487^2548 Van Cauter E, Turek F 1999 Roles of sleep^wake and dark^light cycles in the control of endocrine, metabolic, cardiovascular and cognitive function. In: McEwen BS (ed) Coping with environment. Humana Press, Totowa, NJ (Handbook of Physiology series), in press Van Cauter E, L’Hermite M, Copinschi G, Refeto¡ S, De¤ sir D, Robyn C 1981 Quantitative analysis of spontaneous variations of plasma prolactin in normal man. Am J Physiol 241:E355^E363 Van Cauter E, Blackman JD, Roland D, Spire JP, Refeto¡ S, Polonsky KS 1991 Modulation of glucose regulation and insulin secretion by circadian rhythmicity and sleep. J Clin Invest 88:934^942 Van Cauter E, Kerkhofs M, Caufriez A, Van Onderbergen A, Thorner MO, Copinschi G 1992 A quantitative estimation of growth hormone secretion in normal man: reproducibility and relation to sleep and time of day. J Clin Endocrinol Metab 74:1441^1450 Van Cauter E, Sturis J, Byrne MM et al 1994 Demonstration of rapid light-induced advances and delays of the human circadian clock using hormonal phase markers. Am J Physiol 266: E953^E963 Van Cauter E, Plat L, Copinschi G 1998a Interrelations between sleep and the somatotropic axis. Sleep 21:553^556 Van Cauter E, Plat L, Leproult R, Copinschi G 1998b Alterations of circadian rhythmicity and sleep in aging: endocrine consequences. Horm Res 49:147^152 Van Cauter E, Moreno-Reyes R, Akseki E et al 1998c Rapid phase advance of the 24-h melatonin pro¢le in response to afternoon dark exposure. Am J Physiol 275:E48^E54
PATHOPHYSIOLOGY OF CIRCADIAN RHYTHMS
157
van Coevorden A, Mockel J, Laurent E et al 1991 Neuroendocrine rhythms and sleep in aging men. Am J Physiol 260:E651^E661 Van Reeth O 1998 Sleep and circadian disturbances in shift work: strategies for their management. Horm Res 49:158^162 Van Reeth O, Sturis J, Byrne MM et al 1994 Nocturnal exercise phase delays circadian rhythms of melatonin and thyrotropin in normal men. Am J Physiol 266:E964^E974 Veldhuis JD, Iranmanesh A, Johnson ML, Lizarralde G 1990a Amplitude, but not frequency, modulation of adrenocorticotropin secretory bursts gives rise to the nyctohemeral rhythm of the corticotropic axis in man. J Clin Endocrinol Metab 71:452^463 Veldhuis JD, Iranmanesh A, Johnson ML, Lizarralde G 1990b Twenty-four-hour rhythms in plasma concentrations of adenohypophyseal hormones are generated by distinct amplitude and/or frequency modulation of underlying pituitary secretory bursts. J Clin Endocrinol Metab 71:1616^1623 Waldstreicher J, Du¡y JF, Brown EN, Rogacz S, Allan JS, Czeisler CA 1996 Gender di¡erences in the temporal organization of prolactin (PRL) secretion: evidence fort a sleep-independent circadian rhythm of circulating PRL levels a clinical research center study. J Clin Endocrinol Metab 81:1483^1487 Weitzman ED, Zimmerman JC, Czeisler CA, Ronda JM 1983 Cortisol secretion is inhibited during sleep in normal man. J Clin Endocrinol Metab 56:352^358
DISCUSSION Waxman: I have a two-part question. First, do you know why the endogenous rhythm is not exactly 24 h? Related to this, have studies been done to investigate what the hormonal consequences might be of seasonal changes in day length? Are hormonal patterns that may have been set by light more appropriate to accommodate to the natural daylight cycle? Copinschi: In answer to the ¢rst question, the length of the endogenous circadian period is genetically determined. My guess is that natural selection of circadian clocks occurred, and only individuals with an endogenous circadian clock which was around 24 h were able to cope with the environment and survive. Waxman: But not precisely 24 h. Copinschi: There is no reason why it should be precisely 24 h. We are the result of billions of years of natural selection. Matthews: But it does have to be longer. If you have got light^dark cycle stimuli which are positive triggers, then that only works against a background of an intrinsically slower system. Clarke: But the natural rhythm is exactly 24 h if you take the environmental cues into account. It is longer if you put the individual in a constant environment. Goldbeter: But in other organisms it can be shorter than 24 h: in Neurospora it is close to 21 h. Copinschi: With regard to the di¡erences between the di¡erent seasons, few studies have been performed on this. The only well documented hormonal
158
DISCUSSION
consequence is the seasonal variation in duration of the nocturnal melatonin secretion. Sassone-Corsi: The classical example is what happens to blind people. There are extremely few people who are completely blind: most have enough sensation of light to entrain their clocks. But people who are totally blind are arrhythmic. Licinio: You mentioned that occasional jet lag has no adverse health e¡ects: what about constant jet lag in people such as pilots and £ight attendants? Le¤ vi: It has been reported repeatedly that £ight attendants have an increased risk of breast cancer (Pukkala et al 1995). Could this relate to circadian disruption? Copinschi: I don’t know, but it has been shown in shift workers independently of jet lag that they are at increased risk of a number of diseases (Van Reeth 1998). Veldhuis: In terms of shift work, there are data from the steel industry indicating increased accident rates in workers having revolving shift times versus those with more or less consistent schedules (G. Block, M. Straume & J. D. Veldhuis, unpublished results). Kjems: If you look at shift workers and see changes in the cortisol pattern, how does that correlate to catecholamine patterns? Do you see a change in the pattern of the stress hormones: are they in synchrony, or are they out of phase? What about the regularity? Shift work is not just about changing your 24 h rhythm, there is also the impact of stress. Copinschi: Trying to study shift workers as a group in a natural environment is very di⁄cult, because they are not a homogeneous group. This has been done by several groups. Weibel et al (1997) have studied shift workers in their natural environment, and found all kinds of pro¢les, probably corresponding to the fact that each worker adapted more or less to the shift of their activities: it depended on whether they were working in bright light or sleeping in good conditions and so forth. Marshall: I want to comment on the age-related changes. Humans seem to be less dependent on rhythmicity than other species where reproduction, for example, is tightly controlled by rhythms. One of the interesting aspects is whether some of the changes, for example the cortisol changes in older people, could have some deleterious e¡ects. Steroids are a potent inhibitor of collagen synthesis, for example. Can you reverse that for example, by changing the lifestyles of older people? As we age, we do much less exercise and we don’t sleep as well: there are some studies that show that if you exercise a lot more during the day you sleep much better. Does this reverse the age-related changes? Can you, for example, reduce the insulin resistance that we propose may be associated with some of those insulin changes? Copinschi: In older people we know that the amplitude of the rest^activity cycle is less important. Also many older people don’t go out from the house, so the intensity amplitude of the light^dark cycle is lower. In addition there are what I
PATHOPHYSIOLOGY OF CIRCADIAN RHYTHMS
159
call the ‘environmental’ cues that are probably part of the entrainment which older people lose when they don’t go to work anymore. The maintenance of adequate entrainment may slow down the ageing of the circadian clock. Marshall: The link with the changes which we see as being deleterious has, at least to my mind, not quite been made yet. Is there evidence that if you take an older pattern and reverse that pattern that you could change insulin resistance or cortisol, for example? Copinschi: This hasn’t been done. Veldhuis: I was going to comment further on the issue John Marshall has raised, because I’ve struggled with it also: do these quantitatively de¢nable variations in 24 h rhythms have implications for health and disease? In my own attempt to review this, the matter of intellectual vigilance appears the one best documented. NASA has some of the best data here, because these astronauts are going through light^dark cycles every 90 min. Not only this, but they are up at 3 a.m. in the morning, heavily stressed, since they’re going to be ¢red o¡ into space at about 12 000 mph. Their vigilance goes to pot after about 30 h. The question is, how do you emulate this on earth? How much is due to not to the rhythm disruption but to the cortisol, the adrenaline, the sleep deprivation and having nutrients delivered as semi-liquids in a bag, and so forth? Many factors make this di⁄cult to sort out. Yet the ground studies suggests that at least the vigilance element is disrupted by a collection of this disarray of rhythms. Robinson: Could you comment on the role of timing of nutrition in this? Has anyone done any studies on Muslim groups who fast during daylight hours during Ramadan. They maintain their activity cycles during the day, but shift for a period their nutritional input. How many of these hormonal cycles are related to metabolic rhythmicity? Copinschi: It’s a rather complicated matter. There is a relationship with food schedule and also, for some hormones, with the composition of the food. Normally, these cues are linked together. It has not really been investigated in the human. Robinson: I would be curious to know what happens if you have a sustained period of non-daytime eating, but you still maintain normal physical activity entrained to the normal light^dark cycle. This is a way of shifting those cues and asking which of them is driving metabolic rhythms. Copinschi: It will depend on the hormones you look at. Licinio: Still on the subject of nutrition, Jenkins et al (1989) took the total amount of food a person consumed during the day and divided it into 14 meals taken at 1 hour intervals during the day. The levels of cortisol were altered. Not only the content, but also the timing of the food is important. A more indirect way that nutrition can regulate endocrine rhythms is through leptin secreted by adipose tissue. This is more of a chronic integrator of nutritional status, and has a
160
DISCUSSION
profound e¡ect on nutritional rhythm. We’ve studied the only adults who have been identi¢ed so far who have a leptin gene mutation: they have very low, £at leptin levels through the day, and their other endocrine rhythms are also disrupted. Veldhuis: This masking e¡ect of food has been sort of unmasked by Charles Cszeizler’s constant tube feedings at Massachusetts General Hospital. He puts the patient in constant dim illumination with a 24 h slow dribble of nutrients by tube into the duodenum. Such studies help unveil rhythms that are otherwise partially masked. The food-driven release of cortisol is one such example. Masking by these external/social activities can obscure the circadian variation and make the data messy. Sometimes, these intrusions don’t actually eliminate the 24 h trends. Robinson: I could imagine that a continuous nutritional supply would do that sort of thing, but my question concerned the e¡ects of variations in the episodic nutritional challenge which your endocrine system has also got to cope with, as well as a circadian rhythmicity. Pincus: I’m intrigued/concerned about the endpoint measure. In many of the ageing studies you’re dampening the degree of variation. But it would be nice to see something about the di¡erence between ageing and senescence. Take for example, a cohort of healthy 80 year-olds who are vibrant, like Erte¤ who died parasailing at age 98, as opposed to someone who has been extremely sedentary since age 42. Have studies been done comparing one aged cohort to another which is clearly more senesced? Also, have studies been done, for example, in Uzbekistan where there is a large group who live to 100-plus, to see the degree to which their circadian rhythms have retained a degree of large diurnal amplitude? Copinschi: Not really. There are some reports of individual cases. For instance, the Chicago group has studied a subject aged 75 who had a GH pattern that was quite similar to that of young people and he was in a very good physical shape (E. Van Cauter, unpublished results). Le¤ vi: I was interested in the phase advance of the cortisol rhythms in the depressive state. We also saw this in some cancer patients, and it has been reported in patients with HIV infection (Villette et al 1990). Do you think it is speci¢c to depression, or do you think it is the reaction of the circadian system to an ongoing disease? Copinschi: I suspect that it is a reaction to the disease, because if it was speci¢c to depression it would not be ‘cured’ after successful treatment. Veldhuis: It is interesting that this late afternoon rise in cortisol might be the least speci¢c. We have seen this in alcohol withdrawal, fasting and also in ageing (Iranmanesh et al 1989, Veldhuis et al 1998, Bergendahl et al 1996). But the phase advance in a couple of those conditions does not occur. The phase advance might be a bit more speci¢c than this late afternoon elevation. Why is this late afternoon nadir period so vulnerable to any of a collection of stressors?
PATHOPHYSIOLOGY OF CIRCADIAN RHYTHMS
161
Copinschi: Probably because the hypothalamic^adrenal axis has diminished capacity to adapt to a stress. Veldhuis: Do you think there’s a feedback change at that time? In the rat, for equivalent adrenocorticotropic hormone (ACTH) levels in the late day you get larger incremental adrenal cortisol secretion, despite similar corticosteroid binding globulin levels, suggesting some splanchnic variation in adrenal responsiveness. Is this a feature also in the human, or is the bioactivity of the ACTH released late in the day di¡erent, or is the feedback sensitivity altered? Copinschi: It is probably the feedback sensitivity, but I’m not aware of any data in humans. Kjems: I have a comment regarding insulin resistance. A Danish study has shown that physical training increases insulin action in skeletal muscle in patients with type 2 diabetes by changes in metabolism (Dela et al 1995, Dela 1996). The insulin sensitivity is improved and normal values may be obtained. I am not an expert in insulin resistance, but I think the defects that lead to insulin resistance are far more complicated than just the processes that take place in skeletal muscle. Lightman: Recently, Born et al (1999) published an interesting paper: they were looking at ACTH rhythms and designed a study to see whether the early morning rise anticipates expected time of awakening. Before going to sleep, the subjects in this study were informed of speci¢c times at which they were expected to wake up. When they were expecting to wake up earlier their ACTH rise started earlier; when they were expecting to wake up later, the ACTH rise actually occurred later. This suggests another regulatory mechanism which could actually could entrain the ACTH cortisol rhythm. No one seems to have addressed the issue of what it is that causes the system to know that the ACTH should be going up at that particular time. Pincus: I have an oblique comment to that, dealing with the heart rate axis, rather than the endocrine axis. There are some old studies of REM sleep, in which the same anticipatory question was posed. They found that there was much less deep sleep the hour or two before wake up. This could be a possible explanation. Veldhuis: All the medics who have been on call are familiar with this: you expect that bloody phone to ring, and it usually does, unfortunately!
References Bergendahl M, Vance ML, Iranmanesh A, Thorner MO, Veldhuis JD 1996 Fasting as a metabolic stress paradigm selectively ampli¢es cortisol secretory burst mass and delays the time of maximal nyctohemeral cortisol concentrations in healthy men. J Clin Endocrinol Metab 81:692^699 Born J, Hansen K, Marshall L, M˛lle M, Fehm H 1999 Timing the end of nocturnal sleep. Nature 397:29^30
162
DISCUSSION
Dela F 1996 On the in£uence of physical training on glucose homeostasis. Acta Physiol Scand 635:1^41 Dela F, Larsen JJ, Mikines KJ, Ploug T, Petersen LN, Galbo H 1995 Insulin-stimulate muscle glucose clearance in patients with NIDDM. E¡ects of one-legged physical training. Diabetes 44:1010^1020 Iranmanesh A, Veldhuis JD, Johnson ML, Lizarralde G 1989 Twenty-four hour pulsatile and circadian patterns of cortisol secretion in alcoholic men. J Androl 10:54^63 Jenkins DJ, Wolever TM, Vuksan V et al 1989 Nibbling versus gorging: metabolic advantages of increased meal frequency. N Engl J Med 321:929^934 Pukkala E, Auvinen A, Wahlberg G 1995 Incidence of cancer among Finnish airline cabin attendants, 1967^92. BMJ 311:649^652 Van Reeth O 1998 Sleep and circadian disturbances in shift work: strategies for their management. Horm Res 49:158^162 Veldhuis JD, Yoshida K, Iranmanesh A 1998 The e¡ect of mental and metabolic stress on the female reproductive system and female reproductive hormones. In: Hubbard J, Workman EA (eds) Handbook of stress medicine: an organ system approach. CRC Press, Boca Raton, FL, p 115^140 Villette JM, Bourin P, Doinel C et al 1990 Circadian variations in plasma levels of hypophyseal, adrenocortical and testicular hormones in men infected with human immunode¢ciency virus. J Clin Endocrinol Metab 70:572^577 Weibel L, Spiegel K, Gronifer C, Follenius M, Brandenberger G 1997 Twenty-four hour melatonin and core body temperature rhythms: their adaptation in night workers. Am J Physiol 272:R948^954
Nature of altered pulsatile hormone release and neuroendocrine network signalling in human ageing: clinical studies of the somatotropic, gonadotropic, corticotropic and insulin axes Johannes D. Veldhuis Division of Endocrinology and Metabolism, Department of Internal Medicine General Clinical Research Center, Center for Biomathematical Technology, University of Virginia, Charlottesville, VA 22908, USA Abstract. Recent clinical investigations have implemented an array of new analytical tools to evaluate the neuroregulation of endocrine axes. These studies demonstrate multifold disruption within the growth hormone (GH), luteinizing hormone (LH)^testosterone, adrenocorticotropin (ACTH)^cortisol and the insulin axes in healthy ageing men and women. Novel research strategies in ageing include such developments as the indirect in vivo assessment of neuroendocrine network integration, via the approximate entropy (ApEn) statistic to monitor the unihormonal orderliness and bihormonal synchronicity of hormone release, and thus infer stability of network-integrative processes. For example, ApEn calculations show that the individual orderliness of GH, insulin or LH release falls progressively in older men and women, and the conditional synchrony between LH and testosterone (or LH and follicle-stimulating hormone/prolactin) release, and LH secretion and the neurogenically organized signal, nocturnal penile tumescence (NPT), all decline markedly in older men. Evaluation of the ACTH^ cortisol axis points additionally to disrupted bihormonal synchrony within this stressresponsive system in healthy ageing. A complementary investigative tool, viz. a stochastic di¡erential equation random-e¡ects feedback construct of the interactive male gonadotropin-releasing hormone^LH^testosterone axis, predicts that only certain extant postulates of ageing in the male reproductive axis will give rise to the observed erosion of LH^testosterone synchrony. Collectively, available clinical data suggest a general model of early neuroendocrine ageing in the human, in both the male and female, wherein ageing is marked by variable disruption in the time-delayed feedback and feedforward interconnections among neuroendocrine glands, which constitute an integrated axis and which control the joint synchrony of hormone release. 2000 Mechanisms and biological signi¢cance of pulsatile hormone secretion. Wiley, Chichester (Novartis Foundation Symposium 227) p 163^189 163
164
VELDHUIS
Ageing modi¢es the function and structure of the reproductive, musculoskeletal, cardiovascular, central nervous, hepato^renal, pulmonary and other integrated organ systems in part via pathophysiological changes in endocrine glands. Overall homeostasis is still maintained by adaptations within neuroendocrine axes, a notion developed further below as outlined earlier (Evans et al 1992, Giustina & Veldhuis 1998, Urban et al 1988, Veldhuis 1999a). Here, we will present, review and appraise a new formulation of neuroendocrine ageing, which posits that multisite disruption within dynamic neurohormone axes constitutes one of the ¢rst evident and signi¢cant regulatory defects in ageing. Such postulated defects include speci¢cally the coupling processes that interlink the primary control loci within feedback systems. This thesis will be illustrated for the male pulsatile gonadotropin-releasing hormone (GnRH)^luteinizing hormone (LH)^sex steroid (testosterone) axis (Pincus et al 1996, 1997), the male and female adrenocorticotropin (ACTH)^cortisol axis (Roelfsema et al 1998), the growth hormone (GH)^insulin-like growth hormone type I (IGF-I) axis (Friend et al 1996, Iranmanesh et al 1998, Veldhuis et al 1995), and episodic insulin release in healthy ageing humans, (Meneilly et al 1997, 1999). Network notion of pulsatile neuroendocrine axes Multiple control loci or regulatory nodes Endocrine glands signal their remote target tissues via the episodic or pulsatile release of chemical e¡ectors (Veldhuis et al 1987). Hormone-secreting cells within any given gland do not function in homeostatic isolation. Indeed, at the intraglandular level, important paracrine, autocrine and juxtacrine mechanisms serve to integrate cellular secretory activity. In addition, dominant extraglandular inputs of neuroendocrine and systemic origin impinge upon any particular gland. Such blood-borne in£uences typically encompass both feedforward (positive or trophic signals directing the gland) and feedback (negative or inhibitory signals, often of end-product origin). Consider, for example, the male GnRH^LH^ testosterone feedforward and feedback axis. The anatomically critical control loci (or nodes) in this network comprise hypothalamic neural regulatory centres governing the pulsatile secretion of GnRH, pituitary gonadotrope cells secreting LH (and follicle-stimulating hormone, FSH), and testicular Leydig cells that synthesize and release testosterone (Fig. 1). Evaluating any one primary locus within the GnRH^LH^testosterone axis in isolation would fail to account for critical internodal connectivity (Keenan & Veldhuis 1998). Connectivity allows dynamic coupling, which speci¢cally mediates homeostasis of the axis as a whole (Veldhuis 1999a). Recent clinical experiments in ageing point to disruption of such linkages, as well as alterations in the regulatory nodes themselves (Pincus et al 1996).
PULSATILE HORMONE SECRETION IN AGEING
165
FIG. 1. Schematic representation of the (human) male neuroendocrine GnRH^LH^ testosterone, feedback/feedforward, time-lagged control network. The various ‘H’ terms denote dose-responsive interface functions, as de¢ned in (Keenan & Veldhuis 1998). Both feedback (negative) and feedforward (positive) linkages connect the principal macroscopic control nodes; viz. hypothalamic GnRH neurons, pituitary gonadotrope cells, and testicular Leydig cells. (Adapted with permission from Keenan & Veldhuis 1998.)
Dose-responsive linkages within a pulsatile neurohormone axis Largely on the basis of broken-system studies of isolatedendocrine glands or dispersed cells, interglandular functional connectivity is known to be orchestrated via speci¢c dose^response functions operating within an axis. For example, experiments using synthetic GnRH decapeptide in vivo or in vitro to stimulate whole pituitary, hemipituitary or enzymatically harvested (variably puri¢ed) gonadotrope cells (evaluated in populations or as single cells), have de¢ned a monotonic, sigmoidal and ascending (agonistic) feedforward dose-responsive action of GnRH on LH bsubunit gene expression, biosynthesis of LH glycoprotein and subsequent LH secretion (Evans et al 1992, Urban et al 1988) (Fig. 2). GnRH drives LH secretion by the anterior pituitary gland into the inferior petrosal sinus, from whence LH enters the systemic circulation and ultimately exerts an agonistic dose-dependent e¡ect on gonadal testosterone biosynthesis (Evans et al 1995, Veldhuis 1999a). Secreted testosterone associates with low- (albumin) and high- (sex-hormone binding globulin) a⁄nity transport proteins in the blood (Veldhuis et al 1994a). After a time delay, circulating testosterone acts on hypothalamic and pituitary sites by way of feedback and on multiple peripheral target tissues (e.g. on muscle, kidney) by way of
166
VELDHUIS
FIG. 2. Illustrative schema of feedback and feedforward dose-responsive coupling functions operating within the male GnRH^LH^testosterone axis. Adapted with permission from (Veldhuis 1999b).
feedforward actions. The feedback impact of androgen on CNS and hypophyseal loci is exerted by corresponding dose-dependent inhibitory-response functions (Fig. 2). These pivotal macroscopic linkages complete a so-called ‘closed-loop’ GnRH^LH^ testosterone feedforward/feedback interactive network or control system, consisting
PULSATILE HORMONE SECRETION IN AGEING
167
FIG. 3. Nominal bases for multiple time-scales within a dynamic neuroendocrine axis, such as that for GnRH^LH^testosterone.
of three primary regulatory nodes and four interlinking (dose^response) interface functions. Other neuroendocrine axes function analogously; e.g. the ACTHreleasing hormone (CRH)/arginine vasopressin (AVP)^ACTH^cortisol axis, and the GH-releasing hormone (GHRH)/somatostatin^GH^IGF-I network (Giustina & Veldhuis 1998). We postulate (below) that ageing alters nodal (control-site) function and/or relevant dose^response linkages among key regulatory glands. Within-axis time delays Whereas neurohormone secretion can be enacted rapidly at the neuronal level, endocrine glands proper respond with ¢nite time delays to relevant (negative) feedback and (positive) feedforward inputs (Fig. 3). Circulating feedback and feedforward signals are subject to strong dissipative force within the systemic blood pool, such as molecular di¡usion, linear advection, geometric dilution in the vascular tree, and tissue-speci¢c irreversible elimination (Keenan & Veldhuis 1998). Even after arrival at the target organ, hormones are temporally and topographically dispersed in their access to speci¢c cellular receptors. Time delays arise from not only the foregoing circulatory delivery and within-gland non-uniformities of signal access, but also from multiple serial
168
VELDHUIS
and parallel-arrayed biochemical e¡ector pathways triggered within any given responsive target cell; e.g. GnRH and LH activate respective cell-surface receptors, promote receptor microaggregation, enlist heterotrimeric guaninenucleotide binding protein interactions with the receptor, alter ion-channel £ux, recruit second and third messenger signalling elements, and eventually alter the expression of pertinent genes and proteins. Although the exact time constants for these various distributive processes are not well known, interglandular communication is clearly not instantaneous. In addition, any particular feedback or feedforward endocrine signal secreted into the circulation exhibits a ¢nite and hormone-speci¢c lifetime, which can typically be approximated by a (pluri-) exponential decay function and/or corresponding half-lives (Keenan & Veldhuis 1998, Shah et al 1999, Veldhuis et al 1986, Veldhuis & Johnson 1994). In the circumstance of concurrent basal (or time-invariant) hormone release, a minimum stimulus/inhibitor signal strength is maintained nearly constantly at the target cells. For some axes, such as that of GHRH/somatostatin^GH^IGF-I, unvarying albeit minimal (agonist) signalling induces speci¢c and signi¢cant gene expression in selected target tissues; e.g. low and constant levels of GH evoke hepatic low-density lipoprotein (LDL) and GH receptor expression more e⁄ciently than pulsatile GH stimuli (Giustina & Veldhuis 1998) (see Table 1). Accordingly, the time dimension of interglandular^target tissue communication is complex, and may represent a pertinent locus of alteration in ageing. The largely unexplored interplay between sequential and parallel time delays within a neuroendocrine network would further allow for putative age-related variability in time latencies in feedback and/or feedforward interactions. With the exception of a tendency for some glycoprotein hormones to exhibit a prolonged elimination half-life in older subjects (e.g. LH in postmenopausal women) (Keenan et al 1998), little is known about age-speci¢c alterations in signalling time constants in ageing humans. However, in principle, either heightened variability of, and/or altered latencies of feedback or feedforward signalling within, a neuroendocrine axis could contribute to the evidently more disorderly patterns of release of LH, testosterone, ACTH, cortisol, GH and insulin, as documented by approximate entropy (ApEn) analyses, in older men and women (above). Feedback linkages In neuroendocrine networks, both feedforward and feedback control is a hallmark of e¡ective homeostatic regulation. Typically, end-product inhibition serves to limit otherwise unrestrained hormone secretion. For example, in stress-adaptive states, the initial stress elicits marked (3^10-fold) overproduction of hypothalamic CRH/AVP, pituitary ACTH and ACTH-driven adrenal cortisol
PULSATILE HORMONE SECRETION IN AGEING
169
TABLE 1 exposure
Distinctive target tissue responses to pulsatile versus continuous GH
Pulsatile GH (male)
Continuous GH ( female)
Linear growth Body weight Hepatic EGF receptors Skeletal muscle IGF-I Hepatic P450 2C11 steroid hydroxylase STAT5b tyrosine phosphorylation Hepatic aldehyde oxidase Carbonic anhydrase III Etc.
Hepatic GH and LDL receptors Hepatic sulfatase CBG Glutathione-S-transferase (several isotypes suppressed) Hepatic 5-a-reductase Hepatic P450 2C12 steroid hydroxylase 6-lithocothic acid Etc.
(Adapted with permission from Giustina & Veldhuis 1998.)
secretion. The subsequent rise in cortisol concentrations secondarily imposes (negative) feedback both on hypothalamic centres that direct CRH/AVP secretion and on pituitary corticotrophs that synthesize and secrete ACTH. This key feedback activity is illustrated for the CRH/AVP^ACTH^cortisol axis in Fig. 4. The general precept of feedback regulation applies more broadly to all neuroendocrine axes. Recent clinical studies reveal either augmented or reduced negative feedback activity in ageing; e.g. augmentation of testosterone negative feedback on LH secretion in older men (Deslypere et al 1987, Winters et al 1984, Winters & Atkinson 1997), and reduced cortisol rapid feedback on ACTH release (Gudmundsson & Carnes 1997). Based on current neuroendocrine network constructs, (e.g. Keenan & Veldhuis 1998), disruption of normal feedback timedelays and/or signalling strength would result in disorderly nodal (i.e. single gland) hormone output, as observed in older men for LH and testosterone secretion (Pincus et al 1996) and in older men and women for ACTH and cortisol release (Roelfsema et al 1998). On the other hand, IGF-I’s feedback on GH appears not to be altered in the ageing human (Chapman et al 1997). Feedforward neuroendocrine linkages Whereas ACTH’s feedforward drive of adrenal cortisol secretion is not known to be impaired in the ageing human (Copinschi & Van Cauter 1994), earlier clinical studies have indicated that hCG/LH’s stimulation of Leydig cell testosterone
170
VELDHUIS
FIG. 4. Multinodal, time-lagged, joint feedback and feedforward linkages within a simpli¢ed CRH/AVP^ACTH^cortisol dynamic axis. Negative feedback by cortisol may impinge on either CRH/AVP and/or ACTH biosynthesis and secretion.
production is de¢cient in older men (Deslypere & Vermeulen 1984, Chen et al 1981, Winters & Troen 1982). Because LH’s feedforward drive of testosterone secretion is a critical interlinking function within the male GnRH^LH^androgen axis, any attenuation this agonistic linkage would be predicted to disrupt the bihormonal synchrony (or conditional orderliness) of LH^testosterone co-release in older men (Keenan & Veldhuis 1998). Indeed, this prediction is veri¢ed by two independent analytical techniques to appraise coordinate LH^testosterone release in older men (Keenan & Veldhuis 1997, Pincus et al 1996). Importantly, whereas ACTH’s drive of cortisol secretion is not known to be impaired in older men or women, the dual synchrony expected between ACTH and cortisol secretion deteriorates quanti¢ably in older individuals (Roelfsema et al 1998). According to the notion that the CRH/AVP^ACTH^cortisol axis operates homeostatically as a dynamic, coupled, time-delayed joint feedback and feedforward network, the erosion of joint synchrony of ACTH/cortisol secretion with ageing implies signi¢cant loss of within-network regulatory controls, and/or attenuation of (one or more) nodal function(s). Among such possible mechanisms, at least reduced
PULSATILE HORMONE SECRETION IN AGEING
171
FIG. 5. Schema of principal sources of postulated stochastic variability of measured output within a neurohormone axis.
rate-sensitive (rapid) cortisol feedback (inhibition) of ACTH secretion has been suggested in ageing volunteers (Gudmundsson & Carnes 1997). However, the behaviour of each control node (CRH, AVP, ACTH and cortisol) and the reactivities of various feedback and feedforward interfaces will need to be examined systematically in older humans, in order to explore adequately the full matrix of possible disturbances in ageing. Moreover, dynamic feedback and feedforward studies will be required to better understand age-related changes in other neuroendocrine axes. Stochastic elements within a neuroendocrine axis The foregoing discussion simpli¢es the network behaviour of neuroendocrine axes in several respects; viz. by way of the scale of view (largely macroscopic, rather than molecular), and by an assumption of restricted variability (Keenan & Veldhuis 1998). However, several plausible sources of uncertainty can impact observed hormone output. Unrecognized sources of variability would give rise to observations that are not captured expressly by any idealized model. Fig. 5 highlights three domains of uncertainty that are capable of yielding unexplained
172
VELDHUIS
PULSATILE HORMONE SECRETION IN AGEING
173
(stochastic) variability. The term ‘domain’ is used here, since each category of uncertainty is multifaceted. For example, ¢rst, technical experimental imprecision arises collectively from sample collection, processing and assay technology. Secondly, glandular secretory variability occurs due to cellular heterogeneity, the anatomic dispersion of cells, microvascular non-uniformities, and turbulent admixture of secreted hormone within the circulation. And, thirdly, additional between-subject and within-subject biological variability likely exists about the expected mean feedback, feedforward and time-delay functions (Keenan & Veldhuis 1998). An important unexplored postulate in ageing is that disorderly hormone secretion patterns, evident both unihormonally and bi- (or multi-) hormonally, re£ect greater secretory and/or feedback/feedforward stochastic variability. To address this speculation will require new techniques to quantify the relative magnitudes of putative stochastic biological variations inherent in the regulatory linkages within neuroendocrine axes in young and older individuals. Paradigms of disrupted orderliness of neurohormone release in the ageing human LH^testosterone coupling Figure 6 highlights the quanti¢ably greater irregularity or disorderliness of serial LH release considered univariately, and LH^testosterone synchrony considered bivariately, in healthy older men compared to young individuals. The data illustrate a vivid age contrast, independently of detrending technique (if any), wherein release of LH singly and secretion of LH^testosterone jointly show quanti¢ably greater process randomness of measures sampled every 10 min over 24 h in older volunteers (D. Keenan & J. D. Veldhuis, unpublished results). These observations corroborate earlier results of overnight 2.5 min blood sampling for LH and testosterone in an independent cohort of healthy men (Pincus et al 1996). FIG. 6. (Opposite) Heightened irregularity or disorderliness of LH, but not testosterone (T), release univariately (panels A and B), and LH^testosterone jointly (panel C). Monohormonal irregularity and bihormonal asynchrony were quantitated by the approximate entropy (ApEn) statistic and cross-ApEn, respectively. ApEn provides a model-free and scale-invariant measure of process randomness (univariate) or bihormonal uncoupling (bivariate). Thus, higher ApEn or cross-ApEn values denote greater disorderliness or system uncoupling (Pincus et al 1996). Calculations were performed from 10 min sampled LH and testosterone measures made over 24 h in young and older men. Data were either detrended (by the heat equation, or cosine regression) or not transformed prior to ApEn analyses. LH^testosterone uncoupling in older men was also inferred independently earlier by 2.5 min blood sampling overnight (Pincus et al 1996) , and by cross-correlation analysis (Keenan & Veldhuis 1997). (D. Keenan, T. M. Mulligan, A. Iranmanesh & J. D. Veldhuis, unpublished results).
174
VELDHUIS
FIG. 7. Algebraic di¡erences between ApEn (see legend of Fig. 6) of LH and ApEn of FSH in older (postmenopausal) versus young women (panel A), and as a linear function of age in healthy men (panel B). The interrupted lines denote a null hypothesis of equal ApEn values for LH and FSH. (Data adapted with permission from Pincus et al 1997.)
LH^FSH co-release In the young human as well as the sheep, pulsatile LH release is more orderly than that of FSH (Pincus et al 1997, 1998). However, as shown in Fig. 7, this di¡erence in LH^FSH orderliness of release is abolished in (postmenopausal) women (Panel A), and is progressively reduced in ageing men (Panel B). LH and FSH are co-secreted
PULSATILE HORMONE SECRETION IN AGEING
175
by gonadotrope cells, and both hormones can be colocalized within the same pituitary cell. GnRH can stimulate the secretion of both LH and FSH, but other endocrine, paracrine and autocrine regulators (e.g. inhibin, activin, follistatin, sex steroids) stimulate LH and FSH release di¡erentially (Veldhuis 1999a). The latter distinctions would allow for age-related asynchrony of LH and FSH release via: (i) age-related endocrine modulation of GnRH-stimulated LH versus FSH release; and/or (ii) age-related di¡erential paracrine regulation of LH or FSH release. In contrast, the disorderliness of FSH secretion considered univariately in men and women is high in young adults, and remains elevated in older individuals. Bihormonal synchrony analysis using the cross-ApEn statistic (as de¢ned above, for LH and testosterone) also corroborates such an age-dependent loss of LH^FSH coupling in older men (Fig. 8A). Analogous cross-entropy data have not yet been presented in older women to our knowledge. LH^prolactin bihormonal synchrony of release Since either the LH^testosterone or the LH^FSH joint asynchrony reported in older men (Veldhuis et al 1999) could originate from defective gonadotropic cell regulation in ageing, we have also examined the bivariate orderliness of LH and prolactin release. LH and prolactin are synthesized in distinct (but potentially interacting) cell types within the anterior pituitary gland. Figure 8B depicts the marked attenuation of LH^prolactin joint pattern synchrony evident in healthy older men (Veldhuis et al 1999). The dissociation in expected LH^prolactin coupling suggests an age-related deterioration in gonadotrope^lactotroph interactions and/or in their coordinate control by hypothalamic and/or other (intrapituitary or systemic) factors. In favour of an explanation of disrupted hypothalamic^CNS control pathways is the joint asynchrony also observed between LH secretion and the neurogenically organized signal, nocturnal penile tumescence (NPT), in older men (Fig. 8C). A schematized postulate of the hypothalamic regulatory alterations in ageing that would account for these ¢ndings collectively is given in Fig. 9. In particular, a provisional concept would include an age-related loss of CNS^hypothalamic coordination among sleep^ wake^activity cycles (and hence NPT), the hypothalamic GnRH pulse generator and a putative (neural) prolactin pulse generator (Veldhuis et al 1999). ACTH^cortisol axis In analogy with joint asynchrony in the older male’s LH^testosterone axis (above), the bihormonal coupling of ACTH and cortisol release also deteriorates with ageing in both men and women (Fig. 10). The disorderliness of ACTH (slightly) and cortisol (not at all) secretion each considered univariately declines far less than
176
VELDHUIS
PULSATILE HORMONE SECRETION IN AGEING
177
FIG. 9. Hypothetical relationships among pivotal CNS regulatory centres that control: (a) GnRH pulsatility; (b) a putative neural prolactin (PRL) pulse generator; and (c) sleep^ wake activity cycles and the neurogenically organized NPT (nocturnal penile tumescence) re£ex. (Adapted with permission from Veldhuis et al 1999.)
FIG. 8. (Opposite) Cross-ApEn, a measure of joint (bihormonal) asynchrony (see Fig. 7), for LH and FSH (panel A), LH and prolactin (panel B), and LH and NPT (nocturnal penile tumescence, panel C) in 11 young and 8 older men, each of whom underwent blood sampling and NPT-monitoring at 2.5 min intervals overnight. P values are for unpaired Student’s ttesting. (Adapted with permission from Veldhuis et al 1999.)
178
VELDHUIS
PULSATILE HORMONE SECRETION IN AGEING
179
that of bivariate ACTH^cortisol synchrony (Roelfsema et al 1998). The last observation would point to primary network disruption more than an attenuation of single nodal function. GH^IGF-I axis Akin to the quantitatively consistent attrition in the regularity of monohormonal release patterns of LH or ACTH in ageing (above), 24 h serum GH concentration pro¢les in older men manifest marked loss of pattern orderliness (Iranmanesh et al 1998, Veldhuis et al 1995). The rise in ApEn for GH is age-correlated in men over a range of 18^72 years and is evident in the basal state as well as after treatment with a somatostatin inhibitor or pulsatile GHRH (Fig. 11). The mechanisms underlying this age-associated disruption of GH axis homeostasis are likely multiple, as reviewed elsewhere recently (Giustina & Veldhuis 1998). Foremost considerations include somatostatin excess, GHRH de¢ciency, putative GHreleasing peptide de¢ciency and reduced plasma IGF-I concentrations. The lastmentioned change would produce relative IGF-I feedback withdrawal, and thereby alter within-axis regulation. However, numerous other modulators govern GH secretion, and which of these is/are altered in ageing is not known (Giustina & Veldhuis 1998). Insulin The secretion of insulin is characterized by high-frequency pulsatility in both the fasting (post-absorptive) and fed states (Prksen et al 1994, 1995a,b, Shapiro et al 1988, Veldhuis et al 1994b). Discrete and punctuated episodes of insulin release with a 4^12 min periodicity account for at least 70% of total insulin secretion in the dog and human. In healthy (non-diabetic) individual, such ultradian insulin oscillations are relatively uniform and ad seriatim insulin concentrations are highly ordered (Schmitz et al 1997). This temporal organization is diminished in euglycaemic relatives of patients with adult-onset diabetes mellitus (Schmitz et al 1997), and in healthy older men and women with no personal or family history of diabetes mellitus (Meneilly et al 1997, 1999) (Fig. 12). Since intracellular Ca2+ oscillations in single pancreatic islets and individual b cells correlate with cyclic
FIG. 10. (Opposite) Progressive age-related rise in joint asynchrony of 24 h ACTH and cortisol release in healthy men and women, as monitored quantitatively by cross-ApEn (approximate entropy) values (see legend of Fig. 6). Higher cross-ApEn de¢nes greater uncoupling or asynchrony of the two hormones. Data were analysed by linear regression. Corresponding plots are given for univariate ApEn of ACTH and cortisol each considered singly. Adapted with permission from Roelfsema et al (1998).
180
VELDHUIS
PULSATILE HORMONE SECRETION IN AGEING
181
insulin release (e.g. as inferred by transmembrane capacitance changes), and pulsatile insulin secretion persists in isolated islets of Langerhans, disruption of orderly insulin release patterns in prediabetic patients and in healthy older (nondiabetic) volunteers would suggest altered intercellular or inter-islet secretory coordination. Deterioration of orderly insulin release may presage ine¡ectual tissue actions of insulin, as discussed by Butler elsewhere in this symposium (Butler 2000, this volume). General propositions regarding the impact of ageing on pulsatile neuroendocrine systems Measurements of hormone secretion over time reveal consistent disruption of orderly release patterns for GH, insulin, ACTH and LH considered individually, as well as LH and testosterone (as well as LH and FSH, and LH and prolactin) and ACTH and cortisol bivariately in older men and women. A general notion of network-wide alterations in ageing is thus postulated here. According to a general statement of this concept, normal neuroendocrine axes are time-lagged, functionally interlinked and multinodal ensembles, which are homeostatically maintained by coupled feedback and feedforward interactions among key control sites. Disorderly mono- and/or joint hormonal secretion would then represent one of the earliest quanti¢able alterations in ageing, prior to any decrement (or rise) in mean serum hormone concentrations in the case of (univariately) LH, ACTH, GH, insulin, and (bivariately) LH^testosterone, LH^FSH, LH^prolactin and ACTH^ cortisol control. On the basis of a network concept of normal young adult neuroendocrine within-axis regulation, measurable attenuation of expected coordinate or synchronous hormone release in ageing thus likely signi¢es: (a) disruption in the function of one or more discrete control nodes within the network; (b) impairment in feedforward and/or feedback linkages among control nodes; and/or (c) heightened stochastic biological variability within the time-lagged interface/ coupling functions. Extensive further clinical and laboratory studies will be required to con¢rm, re¢ne and extend these inferences beyond the foregoing neuroendocrine axes. Thus, further investigations are needed to elucidate the putative multisite nature of disturbances in integrative network function,
FIG. 11. (Opposite) Age-associated rise in the disorderliness of minute-to-minute GH release in healthy men. Serum GH concentrations were assayed by ultrasensitive chemiluminescence assay in blood sampled every 10 min for 24 h. Disorderliness increases with age whether or not the GH axis is exogenously driven by pyridostigmine (to inhibit somatostatin release) or pulsatile GHRH infusions (Friend et al 1997, Iranmanesh et al 1998). Analysis is by linear regression. (Adapted with permission from Veldhuis et al 1995.)
FIG. 12. Disruption of the minute-by-minute orderliness of fasting insulin release in healthy older individuals compared to young volunteers. Insulin was assayed in blood samples withdrawn every minute in young and older subjects. Illustrative plasma insulin concentration versus time pro¢les are shown for two young and two older subjects. Quantitation by the approximate entropy (ApEn) statistic established a strong age contrast, with insulin secretion in older adults exhibiting elevated ApEn values, thus denoting loss of subpattern reproducibility for ad seriatim insulin measures. (Adapted with permission from Meneilly et al 1997.)
PULSATILE HORMONE SECRETION IN AGEING
183
whether adaptive or pathologic, which unfold in the ageing human and experimental animal. Failure of an end-organ gland alone will no longer fully explicate the complex pathophysiology documented in ageing neuroendocrine axes. Acknowledgements We thank Patsy Craig for her skillful preparation of the manuscript; Paula P. Azimi for the deconvolution analysis, data management, and graphics; Brenda Grisso and Ginger Bauler for performance of the immunoassays; and Sandra Jackson and the expert nursing sta¡ at the University of Virginia General Clinical Research Center for conduct of the research protocols. This work was supported in part by NIH Grant MO1 RR00847 (to the General Clinical Research Center of the University of Virginia Health Sciences Center), the National Science Foundation Center for Biological Timing (Grant DIR89-20162), the NIH U-54 Specialized Cooperative Centers Program in Reproductive Research (HD-28934), Veterans A¡airs Merit Review Research Funds, and NIH NIA RO1 AG14799.
References Butler P 2000 Pulsatile insulin secretion. In: Mechanisms and biological signi¢cance of pulsatile hormone secretion. Wiley, Chichester (Novartis Found Symp 227) p 190^205 Chapman IM, Hartman ML, Pezzoli SS et al 1997 E¡ect of aging on the sensitivity of growth hormone secretion to insulin-like growth factor-I negative feedback. J Clin Endocrinol Metab 82:2996^3004 Chen GC, Murono E, Osterman J, Cole BT, Nankin H 1981 The aging Leydig cell: II. Two di¡erent populations of Leydig cells and the possible site of defective steroidogenesis. Steroids 37:63^72 Copinschi G, Van Cauter E 1994 Pituitary hormone secretion in aging: role of circadian rhythmicity and sleep. Eur J Endocrinol 131:441^442 Deslypere JP, Vermeulen A 1984 Leydig cell function in normal men: e¡ect of age, lifestyle, residence, diet, and activity. J Clin Endocrinol Metab 59:955^962 Deslypere JP, Kaufman JM, Vermeulen T, Vogelaers D, Vandalem JL, Vermeulen A 1987 In£uence of age on pulsatile luteinizing hormone release and responsiveness of the gonadotrophs to sex hormone feedback in men. J Clin Endo Metab 64:68^73 Evans, WS, Sollenberger MJ, Booth RA Jr et al 1992 Contemporary aspects of discrete peak detection algorithms. II. The paradigm of the luteinizing hormone pulse signal in women. Endocr Rev 13:81^104 Evans WS, Booth RA, Ho KKY et al 1995 Growth hormone economy in normally cycling women. In: Adashi EY, Thorner MO (eds) The somatotropic axis and the reproductive process in health and disease. Springer-Verlag, New York, p 107^123 Friend K, Iranmanesh A, Veldhuis JD 1996 The orderliness of the growth hormone (GH) release process and the mean mass of GH secreted per burst are highly conserved in individual men on successive days. J Clin Endocrinol Metab 81:3746^3753 Friend K, Iranmanesh A, Login IS, Veldhuis JD 1997 Pyridostigmine treatment selectively ampli¢es the mass of GH secreted per burst without altering the GH burst frequency, halflife, basal GH secretion or the orderliness of the GH release process. Eur J Endocrinol 137:377^386 Giustina A, Veldhuis JD 1998 Pathophysiology of the neuroregulation of GH secretion in experimental animals and the human. Endocr Rev 19:717^797
184
VELDHUIS
Gudmundsson A, Carnes M 1997 Pulsatile adrenocorticotrpic hormone: an overview. Biol Psychiatry 41:342^365 Iranmanesh A, South S, Liem AY et al 1998 Unequal impact of age, percentage body fat, and serum testosterone concentrations on the somatotropic, IGF-I, and IGF-binding protein responses to a three-day intravenous growth-hormone-releasing-hormone pulsatile infusion. Eur J Endocrinol 139:59^71 Keenan D, Veldhuis JD 1997 Stochastic model of admixed basal and pulsatile hormone secretion as modulated by a deterministic oscillator. Am J Physiol 273:R1182^R1192 Keenan DM, Veldhuis JD 1998 A biomathematical model of time-delayed feedback in the human male hypothalamic-pituitary-Leydig cell axis. Am J Physiol 275:E157^E176 Keenan DM, Veldhuis JD, Yang R 1998 Joint recovery of pulsatile and basal hormone secretion by stochastic nonlinear random-e¡ects analysis. Am J Physiol 275:R1939^R1949 Meneilly GS, Ryan AS, Veldhuis JD, Elahi D 1997 Increased disorderliness of basal insulin release, attenuated insulin secretory burst mass, and reduced ultradian rhythmicity of insulin secretion in older individuals. J Clin Endocrinol Metab 82:4088^4093 Meneilly GS, Veldhuis JD, Elahi D 1999 Disruption of the pulsatile and entropic modes of insulin release during an unvarying glucose stimulus in elderly individuals. J Clin Endocrinol Metab 84:1938^1943 Pincus SM, Mulligan T, Iranmanesh A, Gheorghiu S, Godschalk M, Veldhuis JD 1996 Older males secrete luteinizing hormone and testosterone more irregularly, and jointly more asynchronously, than younger males. Proc Natl Acad Sci USA 93:14100^14105 Pincus SM, Veldhuis JD, Mulligan T, Iranmanesh A, Evans WS 1997 E¡ects of age on the irregularity of LH and FSH serum concentrations in women and men. Am J Physiol 273:E989^E995 Pincus SM, Padmanabhan V, Lemon W, Randolph J, Midgley AR Jr 1998 Follicle-stimulating hormone is secreted more irregularly than luteinizing hormone in both humans and sheep. J Clin Invest 101:1318^1324 Prksen N, Munn S, Ferguson D, O’Brien T, Veldhuis JD, Butler P 1994 Coordinate pulsatile insulin secretion by chronic intraportally transplanted islets in the isolated perfused rat liver. J Clin Invest 94:219^227 Prksen N, Munn S, Steers J, Vore S, Veldhuis JD, Butler P 1995a Pulsatile insulin secretion accounts for 70% of total insulin secretion during fasting. Am J Physiol 269:E478^E488 Prksen N, Munn S, Steers J, Veldhuis JD, Butler PC 1995b Impact of sampling technique on appraisal of pulsatile insulin secretion by deconvolution and Cluster analysis. Am J Physiol 269:E1106^E1114 Roelfsema F, Pincus SM, Veldhuis JD 1998 Patients with Cushing’s disease secrete adrenocorticotropin and cortisol jointly more asynchronously than healthy subjects. J Clin Endocrinol Metab 83:688^692 Schmitz O, Prksen N, Nyholm B et al 1997 Disorderly and nonstationary insulin secretion in glucose-tolerant relatives of patients with NIDDM. Am J Physiol 35:E218^E226 Shah N, Aloi J, Evans WS, Veldhuis JD 1999 Time-mode of growth hormone (GH) entry into the bloodstream and steady-state plasma GH concentrations rather than sex, estradiol, or menstrual-cycle stage primarily determine the GH elimination rate in healthy young women and men. J Clin Endocrinol Metab 84:2862^2869 Shapiro ET, Tillil H, Polonsky KS, Fang VS, Rubenstein AH, Van Cauter E 1988 Oscillations in insulin secretion during constant glucose infusion in normal man: relationship to changes in plasma glucose. J Clin Endocrinol Metab 67:307^314 Urban RJ, Evans WS, Rogol AD, Kaiser DL, Johnson ML, Veldhuis JD 1988 Contemporary aspects of discrete peak-detection algorithms. I. The paradigm of the luteinizing hormone pulse signal in men. Endocr Rev 9:3^37
PULSATILE HORMONE SECRETION IN AGEING
185
Veldhuis JD 1999a Male hypothalamic^pituitary^gonadal axis. In: Yen SSC, Ja¡e RB, Barbieri RL (eds) Reproductive endocrinology. WB Saunders, Philadelphia, PA, p 622^631 Veldhuis JD 1999b Acromegaly on CD-ROM. Ipsen Publishing, Milan, Italy Veldhuis JD, Johnson ML 1994 Testing pulse detection algorithms with simulations of episodically pulsatile substrate, metabolite, or hormone release. Methods Enzymol 240:377^415 Veldhuis JD, Fraioli F, Rogol AD, Dufau ML 1986 Metabolic clearance of biologically active luteinizing hormone in man. J Clin Invest 77:1122^1128 Veldhuis JD, Carlson ML, Johnson ML 1987 The pituitary gland secretes in bursts: appraising the nature of glandular secretory impulses by simultaneous multiple-parameter deconvolution of plasma hormone concentrations. Proc Natl Acad Sci USA 84:7686^7690 Veldhuis JD, Faunt LM, Johnson ML 1994a Analysis of nonequilibrium dynamics of bound, free, and total plasma ligand concentrations over time following nonlinear secretory inputs: evaluation of the kinetics of two or more hormones pulsed into compartments containing multiple variable-a⁄nity binding proteins. Methods Enzymol 240:349^377 Veldhuis JD, Iranmanesh A, Wilkowski MJ, Samojlik E 1994b Neuroendocrine alterations in the somatotropic and lactotropic axes in uremic men. Eur J Endocrinol 131:489^498 Veldhuis JD, Liem AY, South S et al 1995 Di¡erential impact of age, sex steroid hormones, and obesity on basal versus pulsatile growth hormone secretion in men as assessed in an ultrasensitive chemiluminescence assay. J Clin Endocrinol Metab 80:3209^3222 Veldhuis JD, Iranmanesh A, Mulligan T, Pincus SM 1999 Disruption of the young-adult syncrhony between luteinizing hormone release and oscillations in follicle-stimulating hormone, prolactin, and nocturnal penile tumescence (NPT) in healthy older men. J Clin Endocrinol Metab 84:3498^3505 Winters SJ, Atkinson L 1997 Serum LH concentrations in hypogonadal men during transdermal testosterone replacement through scrotal skin: further evidence that aging enhances testosterone negative feedback. The testoderm study group. Clin Endocrinol (Oxf) 47:317^322 Winters SJ, Troen P 1982 Episodic luteinizing hormone (LH) secretion and the response of LH and follicle-stimulating hormone to LH-releasing hormone in aged men: evidence for coexistent primary testicular insu⁄ciency and an impairment in gonadotropin secretion. J Clin Endocrinol Metab 55:560^565 Winters SJ, Sherins RJ, Troen P 1984 The gonadotropin-suppressive activity of androgen is increased in elderly men. Metabolism 33:1052^1059
DISCUSSION Clarke: With respect to the networking issue, my understanding is that Phyllis Wise is telling us that the suprachiasmatic nucleus (SCN) is one of the key determinants in the ageing issue (Wise et al 1997). By disruption of that nucleus and the loss of diurnal rhythm, you can see a lot of the things that you have demonstrated here. Have you looked into this? Veldhuis: That is a fairly convoluted question: I think you have been brewing that for a while! This is a strong suggestion in the rat, where the circadian coupling to the prooestrus surge is critical. In the human it’s far less critical: we just see a modest blunting of 24 h rhythmicity of LH and testosterone with ageing (Veldhuis et al 1992, 1994a). Most of you are probably familiar with a paper from Bill
186
DISCUSSION
Bremner showing blunting of testosterone rhythms in older men with one-hourly sampling (Veldhuis 1997). With 10 minute sampling in 20 men, we ¢nd that some older men have a 30% testosterone rhythm which is right in the middle of the young adult normal range (Veldhuis 1999a). Thus, this rhythm is not always disrupted. If you put these older men on a GnRH pump, you can partially recover that rhythm (Mulligan et al 1996). However, you do not recover normal ApEn under a GnRH drive, suggesting that there’s a conduit defect somewhere that is not overcome by re-imposing that one segment, GnRH^LH. None of this ApEn technology has yet been applied in the ageing rat. In the male rat it is even a little more confusing in that there’s evidence for bi-level failure. Various investigators have taken out Leydig cells from ageing rats and found steroidogenic defects at several levels (Veldhuis 1999b). On the other hand, in the Brown Norway male rat, you get a failure of LH rise, just as in the human (Veldhuis 1999a). Thus one concern I have is that the technical procedures have not yet been applied adequately to the rat; a second concern is that the ageing male rat is not so well studied as the female in some respects. Thirdly, I’m not sure how representative the male rat is of the healthy older human male. In addition, the fourth issue, is that I don’t know about the coupling that normally pertains between the circadian clock and the ultradian output in the human. The closest I have to any data on this personally is the short-tau mutant hamster from Mike Menaker (Loudon et al 1994). In the codominant with a 22 h clock, this animal has accelerated LH pulsatility. I left that comment until last because this most closely supports your notion that perhaps a disruption in the circadian clock can a¡ect ultradian rhythms. This is the only such example I know. One of the approaches we have tried in order to address this query is to de-trend the data and look only at the ¢ne structure. The contrasts are there. So I personally don’t think it’s the SCN, although in an ageing female rat SCN alterations may be fairly important. Matthews: When you were looking at feedforward and feedback, you demonstrated some really nice cross-correlation data, but I wasn’t clear about the way that you were using that. The problem is that, with cross-correlation data you demonstrate negatives and positives, but with any repeating pattern you’re going to get positives and negatives regardless of whether there is negative feedback as well as feedforward, because you get the next pulses coming through. A positive cross-correlation will usually have a negative domain as well, and it is impossible to deduce causation from these domains, although the association and time lag data may be very clear. Veldhuis: What we did to try to handle that is implement an alternative strategy that is a little more physiological. I should have pointed that out on the slide. The title of the slide should have read ‘LH concentration cross-correlated with
PULSATILE HORMONE SECRETION IN AGEING
187
waveform-independent deconvolution of estimated testosterone (T) secretion rates on a sample-by-sample basis’. We have decorrelated the T concentration by calculating its secretion. There are little spikes, but there’s essentially no autocorrelation in secretory data. This is one approach. But we haven’t removed the correlation in the LH. We did this a few years ago, using ARMA (autoregressive moving average) models, where we ¢nd that the ¢rst-order term explains most of the LH drive on T. We set up the standard matrix of LH on LH autocorrelation, T on T autocorrelation, and LH on T and T on LH crosscorrelations. Solving that matrix, we found that most of the co-variance in the data is explained by LH^T drive within 10 min. This is the correct way, I think, to approach the problem. We were satis¢ed statistically that LH drove T. Under those conditions we could show that T inhibited LH in ¢ve of six men, in a small series of individuals sampled every 10 min for 36 h (Veldhuis et al 1987). But I agree; I think you want to be careful about spurious cross-correlation when there’s strong autocorrelation in one or both series. Pincus: Cross-ApEn is asymmetric, which isn’t a bad or good thing it tells you di¡erent things. You can do cross-ApEn of T on LH, or conversely. Basically, this tells you which is driving the other more and, in particular, in perturbation, which drive is more signi¢cantly disrupted. Matthews: That will normally also work with cross-correlation. The data from that would be asymmetric as well. Pincus: In Cushing’s, when you look at the fact that there’s diurnal variation both in the controls and the Cushing’s in ApEn during the day, the question is, mechanistically, why does this happen? There is a reasonably simple putative explanation from tra⁄c theory. If that variation corresponds to a peak activity time of day as opposed to a low activity time of day, from tra⁄c theory, what you would experience if you were in heavy tra⁄c versus light tra⁄c is much more heterogeneity, as opposed to relatively modest tra⁄c conditions of normal £ow where there’s much less variation. This (the heavy tra⁄c period) is an extra heterogeneity or, if you will, an extra tra⁄c-induced form of noise. If the di¡erence in time of day is correlated with actually the peak versus lower output that could explain physiologically what happens. Robinson: Going back to your experiments in which you were driving the GH system with GHRH or trying to manipulate somatostatin, you still saw in increase in the disorder of the ¢ne structure of those peaks. How much of that might be due to a heterogeneity in the kinetics of the aggregate of the pituitary response? In other words, if in ageing the pituitary is unable to secrete in such a monotypic, modelled structure, how much of that would explain the change in ApEn? Veldhuis: I have intuitively thought that this was the explanation in part for acromegaly. There is a very irregular structure of GH release over time (Hartman et al 1994, Van de Berghe et al 1998). There are micro-bursts from functionally
188
DISCUSSION
coupled clumps of cells that are dribbling out irregular, erratic and poorly consolidated amounts of hormone at variable amounts and intervals. There may be a bit of that micro-autonomy even with ageing (Veldhuis et al 1994b, 1995). I would appreciate some thought on how to address that. We ¢nd this also for several other hormones, including outside the pituitary, like the pancreas, where you ¢nd the same irregularity, since those cells could be breaking down with loss of intrapancreatic networking too (Meneilly et al 1997, 1999). Marshall: On that same point, when you gave GnRH to those older individuals, your ApEn score did not change. I’ve been thinking of this as a neat way to look at some disorganization of a neural circuit that otherwise I can’t get my hands on. I would have thought that the ApEn score should be as low as it possibly could be if you are giving GnRH pulses. Does this mean to say that your other analyses are a function of responsiveness to GnRH? Veldhuis: Actually, ApEn rose on GnRH. What I believe this means is that for a sieving size of one point, m ¼1, the ApEn is picking up super¢ne structure. Even a moderately genuine (infused) GnRH signal thus cannot actually coherently organize the gonadotroph population in older men (Mulligan et al 1996). We cannot match the accomplishments of endogenous GnRH pulses delivered centrally (Veldhuis 1996). We have now shown this to be true also for GHRH’s drive of GH: all the ApEns jumped up in 19 men on GHRH (Iranmanesh et al 1998). When we inject a bolus of GHRH we are not mimicking the coalesced response of the somatotroph population, judged by this kind of peripheral blood sampling. Pincus: Are you saying that the physical mechanism of GnRH is more complicated than a simple superposition, no matter how cleverly you do the superposition? Veldhuis: I think that’s what I’m saying, but I can’t reject Iain Robinson’s hypothesis, that all cells in the pituitary are partially shut down in ageing.
References Hartman ML, Pincus SM, Johnson NL et al 1994 Enhanced basal and disorderly growth hormone secretion distinguish acromegalic from normal pulsatile growth hormone release. J Clin Investig 94:1277^1288 Iranmanesh A, South S, Liem AY et al 1998 Unequal impact of age, percentage body fat, and serum testosterone concentrations on the somatotrophic, IGF-I and IGF-binding protein responses to a three-day intravenous growth-hormone-releasing-hormone (GHRH) pulsatile infusion. Eur J Endocrinol 139:59^71 Loudon ASI, Wayne NL, Kreig R, Iranmanesh A, Veldhuis JD, Menaker M 1994 Ultradian endocrine rhythms are altered by a circadian mutation in the Syrian hamster. Endocrinology 135:712^718
PULSATILE HORMONE SECRETION IN AGEING
189
Meneilly GS, Ryan AS, Veldhuis JD, Elahi D 1997 Increased disorderliness of basal insulin release, attenuated insulin secretory burst mass, and reduced ultradian rythmicity of insulin secretion in older individuals. J Clin Endocrinol Metab 82:4088^4093 Meneilly GS, Veldhuis JD, Elahi D 1999 Disruption of the pulsatile and entropic modes of insulin release during an unvarying glucose stimulus in elderly individuals. J Clin Endocrinol Metab 84:1938^1943 Mulligan T, Godschalk M, Iranmanesh A, Veldhuis JD 1996 Pulsatile GnRH (iv) unmasks intact gonadotrope but impaired Leydig cell responsiveness in healthy aged men. American Geriatrics Society Annual Meeting, Chicago, IL, May 4, abstr Van den Berghe G, Pincus SM, Frolich M, Veldhuis JD, Roelfsema F 1998 Reduced disorderliness of growth hormone release in biochemically inactive acromegaly after pituitary surgery. Eur J Endocrinol 138:164^169 Veldhuis JD 1996 Male hypothalamo-pituitary-gonadal axis. In: Lipshultz LI, Howards SS (eds) Infertility in the male. Mosby-Year Book, Philadelphia, PA, p 23^58 Veldhuis JD 1997 Novel modalities for appraising individual and coordinate pulsatile hormone secretion: the paradigm of luteinizing hormone and testosterone release in the ageing male. Mol Psychiatry 2:70^80 Veldhuis JD 1999a Recent insights into neuroendocrine mechanisms of aging of the human male hypothalamo-pituitary-gonadal axis. J Androl 20:1^17 Veldhuis JD 1999b Male hypothalamo-pituitary-gonadal axis. In: Yen SSC, Ja¡e RB, Barbieri RL (eds) Reproductive endocrinology. WB Saunders, Philadelphia, PA, p 622^631 Veldhuis JD, King JC, Urban RJ et al 1987 Operating characteristics of the male hypothalamopituitary-gonadal axis: pulsatile release of testosterone and follicle-stimulating hormone and their temporal coupling with luteinizing hormone. J Clin Endocrinol Metab 65:929^941 Veldhuis JD, Urban RJ, Lizarralde G, Johnson ML, Iranmanesh A 1992 Attenuation of luteinizing hormone secretory burst amplitude is a proximate basis for the hypoandrogenism of healthy ageing in men. J Clin Endocrinol Metab 75:52^58 Veldhuis JD, Urban RJ, Dufau ML 1994a Di¡erential responses of biologically active LH secretion in older versus younger men to interruption of androgen negative feedback. J Clin Endocrinol Metab 79:1763^1770 Veldhuis JD, Iranmanesh D, Lizarralde G, Urban RJ 1994b Combined de¢cits in the somatotropic and gonadotropic axes in healthy older men: an appraisal of neuroendocrine mechanisms by deconvolution analysis. Neurobiol Aging 15:509^517 Veldhuis JD, Liem AY, South S et al 1995 Di¡erential impact of age, sex-steroid hormones, and obesity on basal versus pulsatile growth hormone secretion in men as assessed in an ultrasensitive chemiluminescence assay. J Clin Endocrinol Metab 80:3209^3222 Wise PM, Kashon ML, Krajnak KM et al 1997 Aging of the female reproductive system: a window into brain aging. Recent Prog Horm Res 52:279^303
Pulsatile insulin secretion Peter Butler* Department of Medicine, University of Edinburgh, Western General Hospital, Edinburgh EH4 2XU, UK
Abstract. Insulin is secreted in a pulsatile manner. Recently it has been shown that almost all (*70^100%) of insulin is secreted in discrete insulin secretory bursts occurring approximately every 6 min. Furthermore, it has been revealed that regulation of the rate of insulin secretion is achieved primarily through modulation of the mass of these discrete insulin bursts. Thus meal ingestion increases insulin secretion by enhancing insulin burst mass by *50% but also increases pulse frequency by *50%. Interestingly, the hepatic clearance of insulin is also apparently closely related to the pattern of insulin delivery to the liver. It has been known for many years that the pattern of insulin delivery is abnormal in patients with type 2 diabetes. Recently, with new more sensitive insulin assays (ELISA) and validated methods for pulse detection, it has been possible to examine more precisely the abnormalities of pulsatile insulin in patients with type 2 diabetes. These recent studies suggest that the principal defect of insulin secretion is a de¢cient pulse mass of insulin with no changes in pulse frequency, and that this defect can be overcome by a period of b cell rest. 2000 Mechanisms and biological signi¢cance of pulsatile hormone secretion. Wiley, Chichester (Novartis Foundation Symposium 227) p 190^205
Insulin secretion is closely regulated in order to maintain carbohydrate, fat and protein metabolism. A total de¢ciency of insulin secretion is responsible for the pathogenesis of type 1 diabetes and a relative de¢ciency of insulin secretion is present in type 2 diabetes. Approximately 30 years ago, it was observed that insulin concentrations oscillate in the peripheral circulation (Goodner et al 1977). Once it was shown that these oscillations were accompanied by oscillations of C-peptide (Koerker et al 1978), it was clear that the £uctuating insulin concentrations were due to intermittent secretion rather than clearance. As well as these high frequency insulin pulses (pulse interval *6 min), insulin is also secreted in ultradian pulses (pulse interval *90 min) (Polonsky et al 1988). The present review focuses on the
*Present address: University of Southern California School of Medicine, Division of Endocrinology & Diabetes, 2025 Zonal Avenue, BMT-B11, Los Angeles, CA 90089-9326, USA. 190
PULSATILE INSULIN SECRETION
191
high frequency pulses. Robert Turner’s group in Oxford con¢rmed that this pulsatile secretion of insulin was also present in humans (Lang et al 1979) and that this was abnormal in patients with type 2 diabetes (Lang et al 1981). Subsequently it was shown that pulsatile insulin secretion was abnormal in patients with evolving type 1 diabetes (Bingley et al 1992) and abnormal pulsatile insulin secretion was reported in animal models with partial b cell de¢ciency (Goodner et al 1989). The presence of pulsatile insulin release posed several questions: (1) (2) (3) (4)
What is the pacemaker responsible for the pulsatile release of insulin? How are the islets dispersed throughout the endocrine pancreas co-ordinated? What proportion of insulin is secreted as discrete insulin discretory bursts? To what extent is regulation of insulin secretion achieved through modulation of the pulsatile component of secretion? (5) What is the mechanism of abnormal insulin secretion in patients with diabetes mellitus? (6) Is the pulsatile delivery of insulin important in hepatic and extra-hepatic insulin sensitive tissues? Pacemaker Initially it was thought possible that oscillations of the plasma glucose concentration may be responsible for oscillations of the insulin concentration present in plasma. However, after it was shown that the isolated perfused pancreas secreted insulin in a pulsatile manner despite perfusion with glucose at a constant concentration (Stagner et al 1980), it became clear that the pancreas had an intrinsic ability to secrete insulin in a pulsatile fashion which did not require intermittent £uctuations in glucose concentration. Furthermore, the isolated perfused pancreas studies revealed that the pulsatile insulin release did not require pancreatic extrinsic innervation. Subsequently, studies with isolated periperfused islets, either in groups (Bergstrom et al 1989, Chou & Ipp 1990, Chou et al 1991) or in single islet preparations (Bergsten et al 1994), revealed a pulsatile pattern of insulin release which implied that pacemaker properties are present in each islet. Taken to another level, electrophysiological studies of individual b cells reveal spontaneous intermittent discharge consistent with pacemaker function (Cook 1983). One area of debate is whether the spontaneous intermittent depolarization of membrane potential leading to pulsatile release of insulin is driven by a primary membrane-driven event or whether the depolarization of the membrane is caused by oscillations in glycolysis (Detimary et al 1998, Tornheim 1997). Since in vivo the two processes are inextricably linked, it is likely that there is an interplay between the resting membrane potential and glycolytic oscillations mediated through
192
BUTLER
oscillations in cytoplasmic Ca2+ (Bergsten et al 1998). The latter increase in response to closure of the ATP sensitive K+ channel which leads to b cell depolarization and Ca2+ in£ux. Since restoration of membrane potential is an energy-dependent process, the latter will lead to a fall in ATP availability and tend to provide a further forward pulse of glycolysis. Coordination As net insulin secretion is pulsatile, there must be coordination of secretion between the islets scattered throughout the pancreas. The observation that there is coordinate pulsatile release present in the isolated perfused pancreas implies that extrinsic innervation is not required for islet-to-islet coordination. There is an intrinsic neural network in the pancreas connected to well-de¢ned pancreatic ganglia (Miller 1981) and it has been suggested that this serves in a similar manner to Purkinje tissue in the heart by allowing islet-to-islet communication. Recent support for this notion is the observation that suppression of neural activity with tetratotoxin in the isolated perfused pancreas results in loss of net pulsatile release, while in contrast, application of the same agent to the isolated perfused islet does not interfere with pulsatile insulin secretion (Sundsten et al 1998). The other ¢ndings which support a potential role of pancreatic ganglia in coordination of islets include the observation of spontaneous depolarizations in these ganglia at a comparable frequency to pulsatile insulin release (King et al 1989). Another experiment which supports the role of the neural network between islets to allow for coordinate pulsatile insulin secretion was reported by Prksen (Prksen et al 1994). In his study, pancreatic islets were transplanted into the liver of rats where they did not secrete insulin in a coordinate pulsatile fashion until the islets had become re-innervated. Proportion of insulin secreted in a pulsatile fashion Quanti¢cation of pulsatile insulin release is complicated by the fact that insulin is discharged into the portal circulation and undergoes extensive hepatic clearance prior to its delivery into the peripheral circulation, which is the usual sampling site. We overcame this problem by developing a canine model with direct portal vein catheterization and were able to show that virtually all insulin secretion is derived from discrete insulin discretory bursts (Fig. 1) (Prksen et al 1995). Subsequently, we were able to validate an established deconvolution method to allow for measurement of pulsatile insulin release. This revealed that humans also secrete insulin exclusively in discrete insulin secretory bursts. These data are consistent with studies of single isolated islets which also show virtually all insulin is released in discrete bursts (Bergsten et al 1994). Taken together, the
PULSATILE INSULIN SECRETION
193
FIG. 1. Portal vein insulin concentration pro¢le in a dog where oscillations in insulin concentrations are clearly separated by plateaus (A). When insulin secretion rate was calculated from this data set by use of Clustcath (B) or deconvolution (C), pulsatile insulin secretion accounted for *100% of insulin secreted (from Prksen et al 1995, with permission).
latter two observations suggest that the coordination between islets in vivo is highly e⁄cient since virtually all islets must be coordinated for net insulin secretion to be almost exclusively pulsatile. Regulation of insulin secretion As virtually all insulin secretion is derived from the pulsatile mode of secretion, it is predictable that regulation of insulin secretion must be achieved by modulation of
194
BUTLER
the pulsatile component of insulin release. It transpires that the circumstance for the pancreas is very comparable to that of the pituitary gland with most modulation of hormone secretion from each endocrine system being made by perturbation of pulse mass rather than pulse frequency (Fig. 2) (Prksen et al 1996a). For example, following meal ingestion, increased insulin secretion is achieved almost exclusively through the speci¢c mechanism of increased pulse mass with only minor changes in
Insulin (pmol/l)
2.000
1.000
0
Insulin (pmol/L)
4.000
2.000
0
0
25
50
75 Minutes
100
125
150
FIG. 2. Portal vein insulin concentration pro¢les before and after glucose ingestion in two representative dogs. There is a marked increase in pulse amplitude after glucose ingestion (30 g at arrow) (from Prksen et al 1996a, with permission).
PULSATILE INSULIN SECRETION
195
pulse frequency. Likewise, inhibition of insulin secretion with somatostatin is achieved by suppression of insulin pulse mass with no change in pulse frequency (Fig. 3) (Prksen et al 1996b). Enhancement of insulin secretion with the sulfonylurea class of compounds is achieved by ampli¢cation of insulin pulse mass with no change in pulse frequency (Prksen et al 1996c). Glucagon-like peptide 1 also increases insulin secretion through the speci¢c mechanism of increasing insulin pulse mass (Prksen et al 1998). Abnormal pulsatile insulin secretion in diabetes
Insulin secretion rates (pmol/min)
Pulsatile insulin secretion could be abnormal in patients with type 2 diabetes for a number of reasons. One possibility is that there is a failure of the pacemaker and/or A
B
600
600
Dog 3
500
500
400
400
300
300
200
200
100
100
0 200
0 Dog 4
200
150
150
100
100
50
50
0 150
Dog 3
Dog 5
0 150
125
125
100
100
75
75
50
50
25
25
0
Dog 4
Dog 5
0 0
10 20 30 40 50 60 Time (min)
0
10 20 30 40 50 60 Time (min)
FIG. 3. Insulin secretion rates in three dogs during saline (A) or low-dose somatostatin infusion (B), calculated by deconvolution analysis. Low dose somatostatin markedly suppressed the insulin pulse mass but did not a¡ect the frequency of the bursts (from Prksen et al 1996b, with permission).
196
BUTLER
coordination system within the pancreas in patients with type 2 diabetes. Against the latter hypothesis, recent studies using highly speci¢c ELISA insulin assays indicate that the insulin pulse frequency in patients with type 2 diabetes is normal but that pulse mass is de¢cient (Laedtke et al 1999). This decrease in pulse mass might theoretically be due to the insulin resistance present in patients with type 2 diabetes. However, in both non-diabetic obese humans (Polonsky et al 1988) and obese monkeys (Hansen et al 1982) (which have insulin resistance), pulsatile insulin secretion is intact. In patients with type 2 diabetes, there is a speci¢c islet pathology consisting of diminished b cell mass associated with islet amyloid (Butler 1996). It is plausible that the presence of the islet amyloid disrupts the coupling between b cells in the islet. It is also possible that the decreased b cell mass in the presence of increased insulin demand leads to diminished insulin stores in the immediately secretable insulin pool present within b cells. The concept of the immediately secretable pool is derived from the observation that insulin secretion is biphasic with immediate ¢rst-phase release over the ¢rst 10^15 min after an intravenous secretagogue followed by a later second phase of increased insulin secretion (Porte & Pupo 1969). First-phase secretion is believed to be derived from insulin vesicles that are docked and ready for secretion, whereas second-phase section is assumed to arise from stored pools of insulin or de novo insulin synthesis. Since pulsatile release of insulin would be from the docked secretory vesicles, any perturbation of the size of the docked insulin pool would presumably also be apparent in both ¢rst-phase insulin release as well as in pulse mass. Patients with type 2 diabetes have diminished ¢rst-phase insulin secretion (Seltzer et al 1967, Ward et al 1984). This defect in ¢rst-phase insulin secretion has been shown to be corrected after a period of inhibition of insulin secretion using diazoxide (Greenwood et al 1976). Interestingly, pulsatile insulin release and ¢rst-phase insulin release in patients with type 2 diabetes are restored after an overnight period of inhibition of insulin secretion using somatostatin (Laedkte et al 1999). These data support the hypothesis that the cause of diminished pulsatile insulin release in type 2 diabetes is reduced insulin in the immediately secretable insulin pool. Since a comparable defect of pulsatile insulin release has been observed in patients with evolving type 1 diabetes (also characterized by a diminished b cell mass), the mechanism of abnormal pulsatile insulin release may be comparable in both circumstances. Importance of pulsatile release on insulin action It is well recognized that the pulsatile pattern of pituitary hormone delivery is important in the action of these hormones. One of the obstacles to determine whether this is the case for the pattern of pulsatile insulin release on insulin action is the di⁄culty of delivering insulin in a pulsatile manner which reproduces
PULSATILE INSULIN SECRETION
197
pulsatile insulin release in vivo, i.e. directly into the portal vein. None the less, when insulin was delivered in small intermittent pulses into the systemic circulation, it was found to be more e⁄cacious in reducing blood glucose through the mechanism of suppression of hepatic glucose release (Matthews et al 1983, Paolisso et al 1991). The magnitude of the oscillations of insulin concentration in the portal vein when these have been directly measured, far exceed those concentrations of insulin that are conventionally considered ‘physiologic’ with respect to insulin dose^response relationship for the liver (Prksen et al 1995, 1996a, Storch et al 1993). It remains to be determined how important pulsatile portal vein insulin delivery is for normalization of hepatic glucose or hepatic fat metabolism. This may be of some therapeutic relevance when islet allografts or other islet replacement systems are being designed for transplantation in patients with type 1 diabetes.
Summary and conclusions In summary, insulin secretion is derived from coordinate discrete pulsatile insulin secretory bursts. The pacemaker that is responsible for these intermittent pulses of insulin release appears to exist within the islets, any of which may serve as a pacemaker. The network of islets communicate with each other through the intrinsic pancreatic neural network via the intra-pancreatic ganglia. This system results in coordinated secretion of insulin pulses into the portal vein approximately once every 6 min. While passing through the liver, most of the insulin within this insulin concentration wave front is cleared by the liver with a small proportion being released into the systemic circulation in an alternated pulsatile pattern. Regulation of insulin secretion is achieved through modulation of the mass of insulin discharged by the pancreas in each burst while the latter also appears to in£uence hepatic insulin uptake. Abnormal pulsatile insulin delivery in diabetes is characterized by a diminished mass of insulin release in each pulse which may contribute towards hepatic and perhaps extra-hepatic insulin resistance since pulsatile insulin delivery enhances insulin action. This decreased pulse mass in diabetes can at least temporarily be reversed by prior inhibition of insulin secretion with somatostatin. One explanation for a diminished insulin pulse mass in diabetes is a decrease in the magnitude of the immediately secretable insulin pool. The dynamics of vesicle membrane docking and undocking are now becoming accessible to novel electrophysiological research techniques (Heinemann et al 1993). E¡orts to better understand the regulation of the magnitude of the immediately secretable insulin pool in relation to the larger stored insulin pool may provide an opportunity for therapeutic interventions in patients with type 2 diabetes mellitus.
198
BUTLER
References Bergsten P, Grapengiesser E, Gylfe E, Tengholm A, Hellman B 1994 Synchronous oscillations of cytoplasmic Ca2+ and insulin release in glucose-stimulated pancreatic islets. J Biol Chem 269:8749^8753 Bergsten P, Lin J, Westerlund J 1998 Pulsatile insulin release: role of cytoplasmic Ca2+ oscillations. Diabetes Metab 24:41^45 Bergstrom RW, Gukimoto WY, Teller DC, De Hae« n C 1989 Oscillatory insulin secretion in perifused isolated rat islets. Am J Physiol 257:E479^E485 Bingley PJ, Matthews DR, Williams AJK, Bottazzo GF, Gale EAM 1992 Loss of regular oscillatory insulin secretion in islet cell antibody positive non-diabetic subjects. Diabetologia 35:32^38 Butler PC 1996 Islet amyloid and its potential role in the pathogenesis of type II diabetes mellitus. In: LeRoith D, Olefsky J, Taylor S (eds) Diabetes mellitus: a fundamental and clinical text. Lippincott-Raven, Philadelphia, PA, p 113^117 Chou HF, Ipp E 1990 Pulsatile insulin secretion in isolated rat islets. Diabetes 39:112^117 Chou HF, McGivern R, Berman N, Ipp E 1991 Oscillations of circulating plasma insulin concentrations in the rat. Life Sci 48:1463^1469 Cook DL 1983 Isolated islets of Langerhans have slow oscillations of electrical activity. Metabolism 32:681^685 Detimary P, Gilon P, Henquin JC 1998 Interplay between cytoplasmic Ca2+ and the ATP/ADP ratio: a feedback control mechanism in mouse pancreatic islets. Biochem J 333:269^274 Goodner CJ, Walike BC, Koerker DJ et al 1977 Insulin, glucagon and glucose exhibit synchronous, sustained oscillations in fasting monkeys. Science 195:177^179 Goodner CJ, Koerker DJ, Weigle DS, McCulloch DK 1989 Decreased insulin- and glucagonpulse amplitude accompanying b-cell de¢ciency induced by streptozocin in baboons. Diabetes 38:925^931 Greenwood RH, Mahler RF, Hales CN 1976 Improvement in insulin secretion in diabetes after diazoxide. Lancet 444^447 Hansen BC, Jen K-LC, Koerker DJ, Goodner CJ, Wolfe RA 1982 In£uence of nutritional state on periodicity in plasma insulin levels in monkeys. Am Physiol Soc 242:R255^R260 Heinemann C, von Rˇden L, Chow RH, Neher E 1993 A two-step model of secretion control in neuroendocrine cells. Eur J Physiol 424:105^112 King BF, Love JA, Szurszewski JH 1989 Intracellular recordings from pancreatic ganglia of the cat. J Physiol (Lond) 419:379^403 Koerker D, Goodner C, Hansen B, Brown A, Rubenstein A 1978 Synchronous, sustained oscillation of C-peptide and insulin in the plasma of fasting monkeys. Endocrinology 102:1649^1652 Laedtke T, Kjems L, Prksen N, Schmitz O, Veldhuis J, Butler PC 1999 An overnight somatostatin-imposed inhibition of b-cell secretion restores orderliness and pulsatility of insulin secretion and the proinsulin to insulin ratio in patients with type 2 diabetes. In review Lang DA, Matthews DR, Peto J, Turner RC 1979 Cyclic oscillations of basal plasma glucose and insulin concentrations in human beings. N Engl J Med 301:1023^1027 Lang DA, Matthews DR, Burnett M, Turner RC 1981 Brief, irregular oscillations of basal plasma insulin and glucose concentrations in diabetic man. Diabetes 30:435^439 Matthews DR, Naylor BA, Jones RG, Ward CM, Turner RC 1983 Pulsatile insulin has greater hypoglycaemic e¡ects than continuous hormone delivery. Diabetes 32:617^621 Miller RE 1981 Pancreatic neuroendocrinology: peripheral neural mechanisms in the regulation of the islets of Langerhans. Endocr Rev 2:471^494
PULSATILE INSULIN SECRETION
199
Paolisso G, Scheen AJ, Guigliano D et al 1991 Pulsatile insulin delivery has greater metabolic e¡ects than continuous hormone administration in man: importance of pulse frequency. J Clin Endocrin Metab 72:607^615 Polonsky KS, Given BD, Van Cauter E 1988 Twenty-four-hour pro¢les and pulsatile patterns of insulin secretion in normal and obese subjects. J Clin Invest 81:442^448 Prksen N, Munn S, Ferguson D, O’Brien T, Veldhuis J, Butler PC 1994 Coordinate pulsatile insulin secretion by chronic intraportally transplanted islets in the isolated perfused rat liver. J Clin Invest 94:219^227 Prksen N, Munn S, Steers J, Vore S, Veldhuis J, Butler PC 1995 Pulsatile insulin secretion accounts for 70% of total insulin secreted during fasting. Am J Physiol 269:E478^E488 Prksen N, Munn S, Steers J, Veldhuis J, Butler PC 1996a E¡ects of glucose ingestion versus infusion on pulsatile insulin secretion. The incretin e¡ect is achieved by ampli¢cation of insulin secretory burst mass. Diabetes 45:1317^1323 Prksen N, Munn S, Steers J, Veldhuis J, Butler PC 1996b E¡ects of somatostatin on pulsatile insulin secretion; selective inhibition of insulin burst mass. Am J Physiol 270:E1043^E1049 Prksen N, Munn S, Steers J , Schmitz O, Veldhuis J, Butler PC 1996c Mechanisms of sulfonylurea’s stimulation of insulin secretion in vivo: selective ampli¢cation of insulin secretory burst mass. Diabetes 45:1792^1797 Prksen N, Grfte T, Nyholm B et al 1998 Glucagon-like peptide 1 increases mass but not frequency or orderliness of pulsatile insulin secretion. Diabetes 47:45^49 Porte D Jr, Pupo AA 1969 Insulin responses to glucose: evidence for a two pool system in man. J Clin Invest 48:2309^2319 Seltzer HS, Allen EW, Herron AL Jr, Brennan MT 1967 Insulin secretion in response to glycemic stimulus: relation of delayed initial release to carbohydrate intolerance in mild diabetes mellitus. J Clin Invest 46:323^325 Stagner JI, Samols E, Weir GC 1980 Sustained oscillations of insulin, glucagon, and somatostatin from the isolated canine pancreas during exposure to a constant glucose concentration. J Clin Invest 65:939^942 Storch MJ, Rossie M, Kerp L 1993 Pulsatile insulin secretion into the portal vein in liver cirrhosis. Dtsch Med Wochenschr 118:134^138 Sundsten T, Orsater H, Pergsten P 1998 Inhibition of intrapancreatic ganglia causes sustained and non-oscillatory insulin release from the perfused pancreas. Diabetologia 41:294 Tornheim K 1997 Are metabolic oscillations responsible for normal oscillatory insulin secretion? Diabetes 46:1375^1380 Ward WK, Bolgiano DC, McKnight B, Halter JB, Porte D Jr 1984 Diminished b-cell secretory capacity in patients with noninsulin-dependent diabetes mellitus. J Clin Invest 74:1318^1328
DISCUSSION Robinson: Can you speculate about the relevance of the pulsatility in the portal vein, where the variation is probably much less than in the peripheral circulation, in terms of target sensitivity requiring pulsatility in the liver, and less so in peripheral tissues? Butler: There is no question that the magnitude of the excursion of the insulin concentration pro¢le in the portal vein is very large. We have shown (but not yet published) that hepatic glucose release oscillates in relation to the pulses of insulin delivered to the liver. It is therefore clear that the liver responds to those pulses on a minute-by-minute basis. However, the periphery also responds to the
200
DISCUSSION
smaller pulses. Richard Bergman’s group at the University of Southern California, Los Angeles, have shown that lipolysis oscillates at a similar frequency (personal communication). Since the dose^response curve of lipolysis (compared to hepatic glucose release) to insulin is much more sensitive, even small pulses of insulin in the periphery would cause changes in lipolysis. Use of immunochemoluminescent assays for insulin is able to reveal pulses in the periphery which, although small, are very distinct. However, they are hard to see using the older insulin assays. De Meyts: What fraction of the insulin is actually extracted by the liver after the ¢rst passage? Butler: The vast majority about 80% or more of the pulse is extracted. The insulin that has got through the liver once is more likely to get through again. It seems that there is a dual mechanism for regulating insulin delivery to the periphery. With evolving type 2 diabetes, you lose insulin pulse mass through reduced b cell mass, but a greater proportion of the insulin that is secreted will make it through to the periphery. This may be one of the things that protects these patients from developing diabetic keto-acidosis, because more of the insulin that they secrete makes it through to the periphery where it inhibits lipolysis. But in a healthy individual about 80% of a pulse is cleared. De Meyts: The point I was trying to make (I was discussing this with Lise Kjems earlier) is that when insulin arrives in the liver, 80% is extracted by the process of insulin binding to the receptor, and is rapidly endocytosed, and then dissociates from the receptor at acid pH. Part of it goes to the lysosomes to be degraded, part of it is recycled and then on top of that, part of the insulin which is bound on the surface is also degraded at the receptor level. I was amazed that the pulsatile pattern of insulin arriving at the liver would survive this entire process and that there would still be a pulsatile pattern in the periphery. Butler: But not every insulin molecule will see a liver cell. Presumably there are some that are going to make it through the hepatic sinusoid without interacting with a receptor. Marshall: I was impressed by your somatostatin experiment, where you restored very dramatically the mass of insulin secreted. In your analogy to weight loss amelioration of diabetes, is that a mechanism by which you would see a restoration? Butler: Those experiments are ongoing. If you enhance insulin sensitivity through either weight loss or through drugs, you decrease the b cell demand per minute in the surviving b cells, and therefore you may theoretically restore pulsatile secretion. In a sense, this is all old news if you consider ¢rst-phase insulin secretion as a quick and dirty way of measuring pulses. In every circumstance examined to date, ¢rst-phase insulin release ends up giving the same qualitative information as measuring pulsatile insulin release. Arguably, both are a measure of the
PULSATILE INSULIN SECRETION
201
immediately secretable insulin pool. It has been shown that ¢rst-phase insulin release is decreased in patients with type 2 diabetes, but after a weight loss or after drugs that enhance insulin sensitivity, the ¢rst-phase insulin release is restored. If we accept ¢rst-phase insulin secretion as a surrogate measure of pulsatile insulin release, the answer is yes. To restore pulsatile insulin secretion you either need to give back more b cells or decrease the demand. Coming back to the notion that the liver likes to see pulses, the advantage of backing-o¡ on the demand is that the insulin that is secreted is then secreted in a more appropriate fashion, so that the liver is now more sensitized to it. You probably therefore get more insulin action for the same number of insulin molecules. Licinio: Is glucose also pulsatile? Butler: Yes, glucose also oscillates. Licinio: Does this correlate with insulin? Butler: It correlates with it, but it’s not the driving factor. It seems to be oscillating in response to insulin, rather than the other way round. Matthews: It £uctuates about 0.2 mmol, which is a tiny oscillation. Butler: The glucose pool is very big, and the turnover of glucose through that pool is very small. This is why we have measured hepatic glucose release directly across the liver. If you look at hepatic glucose release by this method, it is quite dramatic: it is almost on^o¡. But if you inject a small amount of glucose into someone, there is very little change in the glucose concentration, simply because of the magnitude of the glucose pool. Goldbeter: It has been suggested that the pulsatile release of insulin might be driven by glycolytic oscillations which have been observed in b cells (Chou et al 1992, Tornheim 1997). Glycolytic oscillations have a period in the order of 10 min, which is also that of pulsatile insulin secretion. Via the modulation of the ATP/ ADP ratio and the subsequent variation of a K+ conductance, these oscillations could bring about a periodic in£ux of Ca2+ and thereby trigger pulsatile insulin release (Corkey et al 1988). The question would remain as to the nature of the coupling mechanism that would synchronize glycolytic oscillations in di¡erent b cells and di¡erent islets. Butler: The notion here (and this is speculative) is that the islets are electrically coupled through the intrapancreatic neural network and that the islets are interconnected to each other through the pancreatic ganglia. One islet somewhere within the pancreas discharges ¢rst, and sends a wave of depolarization via the ganglia to the other islets, which causes electrical depolarization of these. Sundsten et al (1998) have demonstrated that coordinate pulsatile insulin release requires intact pancreatic ganglia. Goldbeter: You mentioned about the docking of the vesicles as a defect. Alternatively, are there any defects that have been seen in glycolytic enzymes?
202
DISCUSSION
Butler: Yes. There is a rare genetic condition in which glycolysis is inhibited. It has been shown that pulsatile insulin release is abnormal in these patients. Waxman: Following up on the notion that the docking of the vesicles at the plasma membrane may be impaired, isn’t it correct that many of the SNAPs, SNAREs and related factors that are used generically in a large number of cell types that have secretory events? Butler: I must have given the wrong impression: I don’t think docking is impaired. Instead, I think that docking is rate-limiting. If I take out half of my b cells I won’t secrete insulin terribly well. Half the b cells are gone, yet there is still stored insulin. If we extract the pancreas from people who have died and who have type 2 diabetes, we ¢nd stored insulin. In neuroendocrine cells generally the vast majority of stored hormone or transmitter is in vesicles that are not docked and at any one moment are not available for secretion. The idea is to try to take advantage of this, by ¢nding a way of enhancing docking. In diabetes there is a decreased number of b cells, and it appears that proinsulin biosynthesis is not rate-limiting. The proteins involved in the docking process are a therapeutic target: if we can manipulate that system, perhaps we can enhance docking, making the surviving b cells ‘super-dockers’. Clarke: I understood with those SNAP/SNARE complexes that once a vesicle had docked and somehow aggregated with the SNAP or SNARE, that it then was on the membrane and did not leave the membrane. Butler: That’s incorrect. There is a tremendous dynamism between docking and undocking. Under quiescent circumstances, you have about an 85% chance of a vesicle, having docked, of undocking and going back into the remote pool. Robinson: Just to be provocative, is that the right way round to look at docking? You could also make the argument that you have got a dribble of insulin secretion all the time, there will be a tremendous load on the cell, and whenever there is an acceleration through that pathway: the turnover is increased, but the e¡ective mass per burst remains very low. What your somatostatin is doing is conserving your stores away from a release site, so that when you want to trigger the release you then get a big burst. It has been shown in the growth hormone (GH) system, in a very analogous way, that somatostatin pre-treatment with GH-releasing hormone (GHRH) doesn’t cause release, but it restrains the GH stores in position so that when you then give the next GHRH pulse without somatostatin, you get a much greater pulse burst. Butler: We use somatostatin as a tool, not as a therapy for patients with type 2 diabetes. Robinson: But what I’m wondering is whether what you really need is to stop dribble release.
PULSATILE INSULIN SECRETION
203
Butler: We have those experiments ongoing. The trouble with type 2 diabetes is that if I decrease insulin release in a patient, they’re going to get more hyperglycaemic. We can use tools such as diazoxide. Greenwood et al (1976) showed more than 20 years ago that treating patients with type 2 diabetes with diazoxide for 10 days restored ¢rst-phase insulin release. But as to the notion of inhibiting insulin secretion as a therapy for diabetes, I couldn’t agree with you more that this may be useful this is my prejudice of what we need to do. It is just looked upon as being weird. Goldbeter: You said that the glucose release from the liver is pulsatile because of insulin secretion itself. Is it possible that this is the synchronizing factor of b cells? Butler: No, because the isolated pancreas, perfused ex vivo with Krebs bu¡er and 10 mM of glucose, shows coordinate pulsatile insulin release. Also, the single islet shows a similar pulsatile release. One could argue that in vivo glucose and insulin oscillations could be feeding back. However, the glucose oscillations are not absolutely unnecessary to provoke pulsatile insulin release. The single islet actually ends up providing almost identical data to the whole human, if you scale it up. Veldhuis: Are you saying that you don’t need any sychronization except within an islet? Butler: Islets provide their own pacemaker but they still need coordination. Licinio: If you fast type 2 diabetic patients for prolonged periods and then give them their ¢rst meal, is the insulin release better? Butler: These experiments are ongoing. That’s an excellent idea. Licinio: Is there more or less insulin secretion in the morning as opposed to at night? Butler: Glucose tolerance worsens over the course of the day. Veldhuis: I think those experiments in Chicago showing worsening glucose tolerance at night were done by Eve Van Cauter with constant enteral carbohydrate loading, pointing to the pituitary rhythms producing this sort of phenomena of impaired glucose tolerance. Ede¤ n: I was fascinated by the pacemaker role of the islets. Have you any evidence for speci¢c islets being more responsible for pacemaker function? If these are destroyed it may explain the di¡erent sensitivity. Butler: I don’t think we have seen islets that don’t perfuse in a pulsatile fashion. They all do it. I had the simple notion that it was going to be like the Purkinje tissue in the sino atrium of the heart, so I did rather 18th century experiments where we re-sected slices of dog pancreas, thinking that we would suddenly see a change in frequency when we removed the ‘sinoatrial node’ of the pancreas. As we ¢lleted the pancreas from one end to the other so that the dog ended up with just a tiny stump of pancreas, we did observed a progressive decrease in pulse mass: the frequency was completely unchanged.
204
DISCUSSION
Matthews: We published on the data going in the opposite direction, in people who have had operations to take out pancreatic carcinomas (Matthews et al 1983). They still have insulin oscillations, although they are irregular. Whichever way you re-sect the pancreas, you don’t get to the ‘sinoatrial node’. Butler: We did try one other thing. I don’t know if you are familiar with the electrophysiological stocking that cardiologists wrap round the heart. We wrapped that around the pancreas and all we could say was that we saw multiple foci from it. Goldbeter: Is the electrical coordination within an islet or between islets? Butler: Both. Each islet presumably achieves coordination between b cells. Between islets, coordination is presumed to rely on the pancreatic intrinsic neural network. It is primitive neural tissue and it doesn’t die away after central denervation, which is just like the heart: the transplanted heart still beats, and the transplanted pancreas still shows pulsatile insulin release. The intrinsic neural network does not require CNS input. Sassone-Corsi: Is there a network between islets? Butler: Again, this is going back to old data from brave histologists who do these nerve tracings. There is an extensive neural network within the pancreas. It appears to go from islets to these ganglia and then into other islets. The ganglia appear to act as relay stations. Sassone-Corsi: Is there any subordination network within islets? Butler: This is not known. The electrophysiology of the pancreas is a nightmare. The only experiments that I’m aware of that address this are the experiments of nature that David Matthews mentioned in patients who have had their pancreas re-sected, and our unpublished observations with the electrophysiological stockings. Licinio: Is it connected to the CNS or is it autonomous? Butler: It’s all autonomous, but like every other autonomous system, it has cultural connections. It is analogous to peristalsis in the gut. It is quite selfsu⁄cient. Matthews: After truncal vagotomy, you do get some alterations in the pulse intervals, but basically they still continue. Lightman: I’m not quite clear whether or not your suggestion is that in type 2 diabetes there is a primary problem in the secretion of these secretory pulses, rather than all the other problems in terms of insulin resistance. Butler: Patients with type 2 diabetes are no more insulin resistant than comparably obese people with type 1 diabetes. The role of insulin resistance in the pathogenesis of type 2 diabetes has been hugely exaggerated, and in the ¢eld there has been a retreat from this. Undoubtedly insulin action is very important in how the pancreas behaves, but there is a consensus in the ¢eld that insulin resistance alone is not the cause of diabetes.
PULSATILE INSULIN SECRETION
205
References Chou HF, Berman N, Ipp E 1992 Oscillations of lactate released from islets of Langerhans: evidence for oscillatory glycolysis in B-cells. Am J Physiol 262:E800^E805 Corkey BA, Tornheim K, Deeney JT et al 1988 Linked oscillations of free Ca2+ and the ATP/ ADP ratio in permeabilized RINm5F insulinoma cells supplemented with a glycolyzing cellfree muscle extract. J Biol Chem 263:4254^4258 Greenwood RH, Mahler RF, Hale CN 1976 Improvement in insulin secretion in diabetes after diazoxide. Lancet I:444^447 Matthews DR, Lang DA, Burnett M, Turner RC 1983 Control of pulsatile insulin secretion in man. Diabetologia 24:231^237 Sundsten T, Ortsater H, Bergsten P 1998 Inhibition of intrapancreatic ganglia causes sustained and non-oscillatory insulin release from the perfused pancreas. Diabetologia 41:294 Tornheim K 1997 Are metabolic oscillations responsible for normal oscillatory insulin secretion? Diabetes 46:1375^1380
Control of growth hormone (GH) release by GH secretagogues Iain C. A. F. Robinson Division of Neurophysiology, National Institute for Medical Research, The Ridgeway, Mill Hill, London NW7 1AA, UK
Abstract. Despite its long-term role in postnatal growth and metabolism, pituitary growth hormone (GH) is secreted in a short-term highly episodic pulsatile pattern in all species in which it has been examined. This pattern is tightly controlled by the interplay of GH-releasing hormone (GHRH) and somatostatin (SRIF), the primary hypothalamic factors that determine GH secretion from the somatotroph and which also regulate GH synthesis and secretory reserve. The discovery of a endogenous receptor for synthetic GH secretagogues (GHS)s that di¡er from GHRH implies the existence of at least one other endogenous GHS system, though the physiological role of the hypothetical GHS ligand remains unclear. The GH secretory pattern is sexually dimorphic with sex di¡erences at many levels in the hypothalamo^pituitary somatotroph axis. Studies in transgenic animals have shown that GH output is also highly sensitive to feedback control by GH itself, as well as by insulin-like growth factor I. Peripheral responses to GH are markedly dependent on the pattern of GH exposure, which is further modi¢ed after secretion by interaction with GH binding proteins and with GH receptors, both also regulated by the pattern of GH exposure. Although the hypothalamic GH pulse generator is of central importance in the control of GH output, its origin, location and mechanisms remain to be elucidated. 2000 Mechanisms and biological signi¢cance of pulsatile hormone secretion. Wiley, Chichester (Novartis Foundation Symposium 227) p 206^224
It is surely a curious paradox that growth hormone (GH), the 22 kDa pituitary protein hormone whose main biological role is the long-term control of growth and metabolism, should be secreted in a such a rapidly changing dynamic pattern in all species in which it has been investigated. This paradox can be resolved by recognizing that the GH signal is a complex waveform, and that frequency components of the GH plasma pro¢le are as important as amplitude components (i.e. hormone concentrations above threshold) in determining the range and character of biological responses to GH. As is evident from this symposium, the endocrine signals encoded in most hormone secretory patterns enable an endocrine source to send a variety of signals to di¡erent peripheral tissues that can respond di¡erentially to these hormones. 206
CONTROL OF GH RELEASE
207
In the case of GH, this temporal regulation extends both up and down the endocrine cascade. On the one hand, interactions in timing between the hypothalamic factors determine the minute-to-minute pituitary GH secretory response. On the other hand, the pattern of GH in the circulation also a¡ects the expression of the GH receptor (GHR) and its related circulating binding protein (GHBP), thereby a¡ecting the ability of peripheral tissues to respond to GH. In this article I will review some of our studies of the hypothalamic control of GH secretory pattern, its autofeedback regulation and its pattern-dependent e¡ects, to illustrate the importance of di¡erent components of the GH secretory signal for its biological response. For more recent detailed discussion of the mechanisms involved see Giustina & Veldhuis (1998) and Robinson & Hindmarsh (1999). What factors regulate the GH secretory pattern? Both the degree of GH pulsatility and the pattern dependence of GH response vary between species, but are particularly prominent in the rat, in which most of our studies have been performed. Essentially, in male rats, GH shows regular secretory episodes every 3^3.5 h with very low GH concentrations between the bursts (Tannenbaum & Martin 1976). Female rats, in contrast, show a more continuous irregular secretory pro¢le (Ede¤ n 1979, Clark et al 1987), though they also show very high frequency higher-amplitude pulses, particularly at night (Fig. 1). The development of a method for automatic microsampling of blood from conscious rats (Clark et al 1986) made it possible to characterize in detail both spontaneous and secretagogue-induced GH secretory patterns in the rat. Despite the obvious di¡erences, a number of features of GH pulsatility documented in the rat are similar in the human, including gender di¡erences, diminution with age, and responses to GH secretagogues and inhibitors. In general, those principles that have emerged from experimental studies of GH secretion usually hold true in humans, though the individual details vary somewhat. The standard model of GH control is an episodic release of GH-releasing hormone (GHRH) in combination with a varying secretory tone of somatostatin (SRIF) which inhibits GH release. In the simplest model (Tannenbaum & Ling 1984) GH release occurs in response to pulses of GHRH occurring without SRIF. No GH release occurs without a GHRH pulse, but it is SRIF which is ‘gating’ the pituitary GH response to GHRH. Since in male rats, the GH responsiveness to exogenous GHRH varies with the endogenous secretory rhythm (Clark & Robinson 1985), a simple cyclic pattern of SRIF secretion would generate periodicity. Female rats show an unvarying GHRH responsiveness, implying a more constant SRIF tone, and imposition of a
208
ROBINSON
FIG. 1. Automatic blood microsampling in a conscious female rat shows irregular GH pulsatility which varies throughout the oestrus cycle. Assaying prolactin (PRL) in some samples identi¢es the afternoon of proestrus. From Clark et al (1987).
‘male-type’ sinusoidal SRIF infusion is su⁄cient to generate episodic GH release and stimulate growth in females (Clark & Robinson 1988). Whilst this basic model remains a good working description, it is clearly oversimpli¢ed. The e¡ects of SRIF are more complex; for example, acute withdrawal of SRIF exposure is an excellent way to generate a GHRH-dependent GH pulse (Clark et al 1988a). Since there are SRIF receptors on arcuate GHRH cells (Epelbaum et al 1989), a separate population (Willoughby et al 1989) of arcuate SRIF cells, distinct from those SRIF cells in the periventricular nucleus which secrete SRIF into portal blood, may constitute a local SRIF pulse generator that may synchronize GHRH discharge into portal blood. We have recently used a novel pure SRIF antagonist to explore this (Baumbach et al 1998). Whilst this antagonist did increase GH release acutely in non-pulsing anaesthetized rats, preliminary experiments have shown that infusion of the same antagonist to conscious pulsing male rats profoundly interrupts their episodic GH secretion (W. R. Baumbach, K. M. Fairhall, D. F. Carmignac & I. C. A. F. Robinson, unpublished data), suggesting that a functioning SRIF system is
CONTROL OF GH RELEASE
209
indeed necessary for spontaneous GH pulsatility. Since blockade of GHRH also interrupts pulsatility (Wehrenberg et al 1982), both peptides would appear to be necessary to generate a regular GH pulse rhythm. What the role of SRIF is in the more continuous irregular GH output in female rats is less clear. Although GHRH is a GH secretagogue (GHS), it has a second, probably more important role as a long-term trophic factor for the GH axis, stimulating GH cell proliferation and GH synthesis via the adenyl cyclase pathway, and this escapes SRIF blockade (Barinaga et al 1985). Inactivating mutations of the GHRH receptors lead to a hypoplastic pituitary (Lin et al 1993) and reduced pituitary GH reserves, whilst excess GHRH exposure leads to GH cell hyperplasia and excess GH secretion (Thorner et al 1982, Hammer et al 1985). Many of the GHRH pulses measured in conscious sheep hypophysial portal blood are not associated with GH release (Thomas et al 1991) though they continue to have an important stimulating e¡ect on GH stores, which would increase the amplitude of the next GH pulse that escapes SRIF blockade. Since concomitant SRIF and GHRH exposure also increases the amount of GH available for eventual release, an interaction of GHRH and SRIF directly regulates GH pulse amplitude as well as frequency. GH pulsatility with GHRH and SRIF has recently been modelled (e.g. Wagner et al 1998) but there is always a danger in such an approach that other important inputs may be excluded. For example, the existence of a synthetic GHS system (exempli¢ed by the hexapeptide GHRP-6; Bowers et al 1984) unrelated to GHRH, presents a pharmacological curiosity which interacts in an interesting way with GHRH, synergizing on GH release, and amplifying spontaneous GH secretion (Smith et al 1997). The relationship with GHRH is decidedly ambiguous, since the GHS and GHRH ligands act via independent receptors and signal transduction systems, yet GHSs are heavily dependent on an intact GHRH system and/or GHRH receptor function (Clark et al 1989, Smith et al 1997). The identi¢cation of arcuate neurons as targets for GHS (Dickson et al 1993), was followed by the cloning of an endogenous GHS receptor (Howard et al 1996). Furthermore, the expression of GHS receptors is much more prominent in the hypothalamus than pituitary and the hypothalamic expression is negatively regulated by GH in parallel with GHRH, but opposite to neuropeptide Y (NPY) (Bennett et al 1997). Circumstantial evidence suggests that an endogenous GHS system could well play a role in pulsatile GH control. Furthermore the responses to GHSs are also pattern dependent, track with infrequent regular injections (Fairhall et al 1995), but desensitize with prolonged GHS exposure (McDowell et al 1985). GHSs also activate prolactin and adrenocorticotropic hormone (ACTH) release, albeit to a lesser extent than GH. If GHS activity is present in portal blood
210
ROBINSON
and the endogenous ligand(s) identi¢ed there, this will strengthen the notion of a third GH regulator. The development of potent GHS antagonists would also facilitate studies of the physiological role of the endogenous receptor. However it remains possible that GH release is only one aspect of the true physiological role of the GHS system. NPY and leptin are also important inputs to the GH axis, and these could well be involved in a neuronal pathway regulating GH secretion in relation to nutrition and metabolic demands. Amino acids, glucose and fatty acids all regulate GH secretion directly or indirectly but the way these are channelled to the GHRH/SRIF/GHS pathways remains unclear. Another factor that must be taken into consideration is the powerful role of GH feedback on its own release. This occurs quite rapidly, and also shows some temporal relationship to the secretory cycle, such that trains of GH pulses will also entrain the GH pulse generator in conscious male rats (Carlsson & Jansson 1990), whilst continuous exposure to GH will powerfully suppress GH secretion, both by stimulation of SRIF and blockade of GHRH synthesis. GH feedback will also block SRIF-induced GH rebounds (Clark et al 1988b) and GH receptors are localized on periventricular SRIF cells as well as arcuate NPY and GHRH cells (Burton et al 1992, 1995). GH feedback can be modelled in transgenic animals with hGH expression targeted to the CNS, and we have studied this recently in transgenic rats (Flavell et al 1996). Minimal CNS hGH expression is su⁄cient to cause powerful GHRH suppression and secondary pituitary GH de¢ciency. Such animals continue to show pulsatile GH release though of reduced amplitude, but respond appropriately to GHRH, SRIF and GHS administration (Wells et al 1997). Chronic peripheral GH treatment corrects their dwar¢sm but has no further suppressive e¡ect on their hypothalamic GHRH expression (Pellegrini et al 1997) (Fig. 2). The foregoing has concentrated on the hypothalamic neuroendocrine mechanisms, but until we know more about the electrophysiological behaviour of these cells, the origin and location of the GH pulse generator remains a mystery. Given the requirement to achieve a major pulse of GHRH secretion into portal blood, the term ‘pulse generator’ may actually be a misnomer, and it may be that what is required in male rats is a temporal synchronization of individual cells to generate a population pulse of GHRH secretion, and that this may not require any absolute increase in individual cell ¢ring rates. A continuous output as in female rats could re£ect a summation of the average ¢ring rate in a desynchronized population of cells. In this model, somatostatin, GHS pulses, or exogenous GH pulses all serve to synchronize the cell population, rather than to stimulate increased ¢ring per se. However, this still begs the question of what input determines the underlying spontaneous pulse frequency.
CONTROL OF GH RELEASE
211
FIG. 2. GH feedback regulation of GHRH cells. Transgenic growth-retarded (Tgr) rats are dwarfed due to hypothalamic expression of an hGH transgene targeted to GHRH neurons (Flavell et al 1996) and suppressing GHRH, whilst dwarf (Dw) rats have a primary pituitary GH de¢cit and increased GHRH expression. Acromegaly may be induced in both dwarf models by continuous exposure to rat GH from subcutaneous implants of GH-producing cells (GC cells). Such chronic GH exposure feeds back to inhibit GHRH in normal (WF/^) and dwarf (Dw) rats but has no e¡ect in Tgr rats (WF/Tgr) in whom the number of detectable GHRH cells is already markedly suppressed by hypothalamic hGH from the transgene (Pellegrini et al 1997).
Physiological signi¢cance of the GH secretory pattern The most obvious question is whether such complex episodic secretory patterns determine the e¡ectiveness of GH on its target tissues. The growth rates in both GH-de¢cient rats and children are signi¢cantly improved by increasing the frequency of administration of GH for a given GH dose. This has lead to the conclusion that GH pulsatility is ‘the optimal’ pattern for all GH e¡ects, which is a misleading oversimpli¢cation. Continuous GH exposure is clearly e¡ective to stimulate some growth in GH-de¢cient animals, and female rats grow rapidly with their more continuous GH secretory pattern, albeit less rapidly than males. Furthermore, clinical GH replacement therapy with daily subcutaneous injections does not reproduce the physiological pattern of GH but generates marked improvements in ¢nal height in GH-de¢cient children. It still remains to be established what the optimal and appropriate GH replacement regime is in GHde¢cient adults. A major problem in adults is the greater di⁄culty in establishing a
212
ROBINSON
relevant and easily measured metabolic endpoint in the absence of a growth response. The extreme case of continuous GH exposure in acromegalic subjects teaches us that continuous GH is biologically e¡ective, and that the GH response does not down-regulate to such continuous GH exposure. However, it is also true that other GH responses may either be indi¡erent to the pattern of exposure or even be more e¡ectively elicited by a continuous exposure to GH. This may be the case for the lactogenic response in cattle in which depot formulations of slow release GH are e¡ective and convenient. On the other hand, observations in transgenic animals continuously exposed to GH certainly suggest that pathology readily follows, even when the plasma concentrations of continuous GH are modest. Similarly, the GH hypersecretion such as that seen in acromegalic subjects shows dysregulation in both amplitude and frequency components of the GH signal, and whilst most treatments reduce the amount of GH secreted, relatively low GH concentrations present continuously may still prove harmful if maintained over long periods. despite their convenience, it will be important to show that long-term use of depot preparations in humans does not generate inappropriate continuous GH exposure, with potentially deleterious e¡ects in the longer term. In physiological terms, both continuous and intermittent GH secretory patterns are utilized to target the secreted GH to di¡erent responses. In many of our studies, we have taken advantage of the spontaneous GH-de¢cient dwarf rat (Charlton et al 1988) to assess the e¡ects of di¡erent patterns of pulsatile or continuous GH exposure given separately or in combination via chronic indwelling venous catheters. By developing methods (Clark et al 1985) for delivering computercontrolled i.v. infusions to conscious hypophysectomized or dwarf rats we were able to study the parameters of the GH pattern that were important for di¡erent GH responses. Both continuous and pulsatile GH patterns show dose^response relationships that di¡er in threshold and slope for di¡erent endpoints, and in the same animal, it is possible to obtain opposite responses to the same dose of GH by varying the duration of exposure (Clark et al 1996). Studies using separate injection or infusion patterns con¢rmed that skeletal growth is dependent on pulse frequency as well as amplitude for GH (Jansson et al 1982, Clark et al 1985). More recently it has been possible to study a combination of both continuous and pulsatile contributions using mixed infusion patterns in GH-de¢cient rats, and varying the relative contribution of pulse height vs. trough within a given dose range of GH (Gevers et al 1996; Fig. 3). Whilst the pulsatile component is clearly more e¡ective for skeletal growth in the rat, this is not true for all GH responses in this species. This is well illustrated in the same rodent model in which the lipolytic e¡ects of GH could also be studied. Female dwarf rats fed a high fat diet rapidly become obese and sensitive to the fat-mobilizing e¡ects of GH (Clark et al 1996). GH injections produced an
CONTROL OF GH RELEASE
213
FIG. 3. Intravenous infusions of GH were given to dwarf (dw/dw) rats using mixed infusion patterns that varied the proportion of the dose delivered in a continuous (c) or pulsatile (p) mode (upper panel). For any given dose, the growth response was mostly responsive to the pulsatile component of the infusion. From Gevers et al (1996).
increased weight gain, as the anabolic and skeletal growth responses outweigh the lipolytic response. However, these obese rats lost weight when given the same GH dose by continuous infusion. This was because continuous GH was much more e¡ective than daily injections in reducing fat stores (Fig. 4). Thus the same dose of GH produced opposite e¡ects on weight gain depending on its pattern of administration. In order to e¡ect any of these responses, the GH signal must be transduced by its receptor, and part of the pattern sensitivity may be mediated at the GHR itself, by altered transcription, translation, cellular distribution and/or internalization. The
214
ROBINSON
FIG. 4. GH is also lipolytic, and this may be conveniently studied in vivo in female GH-de¢cient rats which become obese when fed a high fat diet. In these animals, di¡erential responses are observed to intermittent vs. continuous GH treatment, the latter treatment mode being much more e¡ective in reducing ovarian or retroperitoneal fat. From Clark et al (1996).
CONTROL OF GH RELEASE
215
same experiments (Gevers et al 1996) that focused on the di¡erential e¡ect of pulsatile and continuous components of patterned GH administration also showed that, unlike the growth response, GHR binding and plasma GHBP levels were increased by continuous, but not intermittent GH exposure (Fig. 5). It is not known what consequence these changes have for the alterations in signal transduction that have been shown to occur via this receptor, or for plasma distribution and half-life via alterations in plasma GHBP levels (and thus the proportion of a GH pulse vs. trough levels bound to GHBP). However, the fact remains that the hepatic GHR/GHBP expression is strongly regulated by the pattern of ligand exposure in the rat. One may speculate that long-term changes in GHR and GHBP in response to continuous, but not intermittent, GH exposure may underlie the apparent adaptation in GH responsiveness that is seen in response to prolonged GH treatment. It may also be an important factor in physiological adaptation to altered secretory patterns in humans as we move from childhood and puberty though adulthood to old age. One can imagine that the cellular responses to GH probably di¡er when comparing the requirements of growth and development in young individuals with those for metabolism and repair in the same individuals much later in life. If so, the appropriate pattern of GH therapy in the elderly, will probably not be that most e¡ective in stimulating growth in the young! Other examples of di¡erential e¡ects of GH patterns can be seen by measuring other biochemical targets, including numerous GH-sensitive hepatic proteins in the rodent. A fuller discussion of these is beyond the present scope of this article and the mechanisms mediating some of these e¡ects are discussed elsewhere in this volume (see this volume: Waxman 2000, De Meyts & Shymko 2000). It is worth pointing out however, that most attention is focused on the GH pulse as the ‘important’ signalling component of an episodic pattern. In my view the interpulse ‘trough’ may be equally important and has been relatively neglected, though it is easy to show that it, too, provides a signal. One of the de¢ning aspects of GH secretion in female rats is the higher irregular secretion, presenting a continuous signal to the tissues (Ede¤ n 1979, Clark et al 1987). This is compounded by higher circulating levels of plasma GHBP in female rats. We showed some years ago that it was possible to impose periods of very low (male-type) troughs in female rats by intermittent suppression with SRIF infusions (Clark et al 1988a,b). In normal rats, this is accompanied by ‘male-type’ rebound GH peaks, but when performed in dwarf female rats, the rebounds are very small. Simply inducing regular troughs in this fashion was su⁄cient to ‘masculinize’ levels of one particular hepatic GH-responsive enzyme (carbonic anhydrase III) (Je¡ery et al 1990), whereas, in the same animals, generating regular male peaks by giving three-hourly injections of GHRH (but without
216
ROBINSON
CONTROL OF GH RELEASE
217
suppressing GH secretion in the troughs) did not a¡ect this enzyme. Provided that so-called ‘trough’ values of GH are suprathreshold for GH receptor signalling, it would seem likely that they are biologically relevant for some GH responses. If so, the role of endogenous SRIF in ensuring regular periods of GH suppression is not simply to ensure e⁄cient pituitary pulsatile release: it is also sending an intermittent ‘absence of GH’ signal to some tissues. This may be even more relevant if the receptor signalling also includes a dynamic ‘on/o¡’ component responsive to rate of change of concentration, down as well as up. Control of pituitary secretion is not the only way to regulate basal exposure of the tissues to GH. The presence of GHBP may also regulate access of GH to its receptors, and very high levels of GHBP can block GH e¡ects either by complexing with GH, or possibly by forming inactive heterodimers with GHRs. GHBP injected together with GH forms a complex with a prolonged circulatory half-life, and injections of GHBP complexed with GH elicited a hepatic CYP2C response which was altered towards a more continuous type of GH exposure (Wells et al 1994). Thus even after the secretory pattern is established, the abundance of GHBP can also modulate the e¡ective pattern of exposure to GH at the peripheral tissues. Finally, it is appropriate to ask if there are physiological conditions in which continuous GH exposure is the most appropriate pattern? One example is human pregnancy. The human placenta provides a continuous source of a GH variant which is fully active, reaches acromegalic concentrations in the third trimester, and powerfully represses pituitary GH secretion, essentially converting the GH secretion from pulsatile to high continuous exposure in the mother (Eriksson et al 1989). Given the large changes in metabolic demand and the necessity of partitioning and utilizing nutrients di¡erently in pregnancy, the altered maternal metabolic requirements may be more appropriately elicited by a continuous GH supply. Following delivery of the placenta, the source of placental GH is abruptly withdrawn, and the mother may even be transiently GH de¢cient until the pituitary GH secretion recovers from the powerful feedback suppression. As yet, the physiological importance of this intriguing alteration in GH pattern remains largely unexplored.
FIG. 5. (Opposite) Both GHRs and plasma GH binding protein (GHBP) are sensitive to the pattern of GH autoregulation. Using intravenous programmed infusions in rats to vary the proportion of the GH dose delivered as a pulse (p) or continuously (c), both GHR binding and plasma GHBP (measured by RIA) were induced dose dependently by continuous, but not pulsatile GH exposure. This is particularly obvious from the experiment depicted in the middle sections of both panels in which all animals received the same total GH dose (200 mg/day) but in a mixed infusion pattern, in which the continuous component was increased at the expense of the pulsatile component.
218
ROBINSON
Acknowledgements I thank my colleagues, whose work has put substance to many of the ideas expressed here.
References Barinaga M, Bilezikjian LM, Vale WW, Rosenfeld MG, Evans RM 1985 Independent e¡ects of growth hormone releasing factor on growth hormone release and gene transcription. Nature 314:279^281 Baumbach WR, Carrick TA, Pausch MH et al 1998 A linear hexapeptide somatostatin antagonist blocks somatostatin activity in vitro and in£uences growth hormone release in rats. Mol Pharmacol 54:864^873 Bennett PA, Thomas GB, Howard AD et al 1997 Hypothalamic growth hormone secretagoguereceptor (GHS-R) expression is regulated by growth hormone in the rat. Endocrinology 138:4552^4557 Bowers CY, Momany F, Reynolds GA, Hong A 1984 On the in vitro and in vivo activity of a new synthetic hexapeptide that acts on the pituitary to speci¢cally release growth hormone. Endocrinology 114:1537^1545 Burton KA, Kabigting EB, Clifton DK, Steiner RA 1992 Growth hormone receptor messenger ribonucleic acid distribution in the adult male rat brain and its colocalization in hypothalamic somatostatin neurons. Endocrinology 131:958^963 Burton KA, Kabigting EB, Steiner RA, Clifton DK 1995 Identi¢cation of target cells for growth hormone’s action in the arcuate nucleus. Am J Physiol 232:E716^E722 Carlsson L, Jansson JO 1990 Endogenous growth hormone (GH) secretion in male rats is synchronized to pulsatile GH infusions given at 3-hour intervals. Endocrinology 126:6^10 Charlton HM, Clark RG, Robinson ICAF et al 1988 Growth hormone-de¢cient dwar¢sm in the rat: a new mutation. J Endocrinol 119:51^58 Clark RG, Robinson ICAF 1985 Growth hormone responses to multiple injections of a fragment of human growth hormone-releasing factor in conscious male and female rats. J Endocrinol 106:281^289 Clark RG, Robinson ICAF 1988 Paradoxical growth promoting e¡ects induced by patterned infusions of somatostatin in female rats. Endocrinology 122:2675^2682 Clark RG, Jansson JO, Isaksson O, Robinson ICAF 1985 Intravenous growth hormone: growth responses to patterned infusions. J Endocrinol 104:53^61 Clark RG, Chambers G, Lewin J, Robinson ICAF 1986 Automated repetitive microsampling of blood: growth hormone secretion in conscious male rats. J Endocrinol 111:27^35 Clark RG, Carlsson LMS, Robinson ICAF 1987 Growth hormone secretory pro¢les in conscious female rats. J Endocrinol 114:399^407 Clark RG, Carlsson LMS, Ra¡erty B, Robinson ICAF 1988a The rebound release of growth hormone (GH) following somatostatin infusion in rats involves hypothalamic GH-releasing factor release. J Endocrinol 119:397^404 Clark RG, Carlsson LMS, Robinson ICAF 1988b Growth hormone (GH) secretion in the conscious rat: negative feedback of GH on its own release. J Endocrinol 119:201^209 Clark RG, Carlsson LMS, Trojnar J, Robinson ICAF 1989 The e¡ects of a growth hormonereleasing peptide and growth hormone-releasing factor in conscious and anaesthetized rats. J Neuroendocrinol 1:249^255 Clark RG, Mortensen DL, Carlsson LM, Carlsson B, Carmignac D, Robinson ICAF 1996 The obese growth hormone (GH)-de¢cient dwarf rat: body fat responses to patterned delivery of GH and insulin-like growth factor-I. Endocrinology 137:1904^1912
CONTROL OF GH RELEASE
219
De Meyts P, Shymko RM 2000 Timing-dependent modulation of insulin mitogenic versus metabolic signalling. In: Mechanisms and biological signi¢cance of pulsatile hormone secretion. Wiley, Chichester (Novartis Found Symp 227) p 46^60 Dickson SL, Leng G, Robinson ICAF 1993 Systemic administration of growth hormonereleasing peptide activates hypothalamic arcuate neurons. Neuroscience 53:303^306 Ede¤ n S 1979 Age- and sex-related di¡erences in episodic growth hormone secretion in the rat. Endocrinology 105:555^560 Epelbaum J, Moyse E, Tannenbaum GS, Kordon C, Beaudet A 1989 Combined autoradiographic and immunohistochemical evidence for an association of somatostatin binding sites with growth hormone-releasing factor-containing nerve cell bodies in the rat arcuate nucleus. J Neuroendocrinol 1:109^115 Eriksson L, Frankenne F, Ede¤ n S, Hennen G, Von Schoultz B 1989 Growth hormone 24 h serum pro¢les during pregnancy lack of pulsatility for the secretion of the placental variant. Br J Obstet Gynaecol 96:949^953 Fairhall KM, Mynett A, Smith RG, Robinson ICAF 1995 Consistent GH responses to repeated injection of GH-releasing hexapeptide (GHRP-6) and the non-peptide GH secretagogue, L692,585. J Endocrinol 145:417^426 Flavell DM, Wells T, Wells SE, Carmignac DF, Thomas GB, Robinson ICAF 1996 Dominant dwar¢sm in transgenic rats by targeting human growth hormone (GH) expression to hypothalamic GH-releasing factor neurons. EMBO J 15:3871^3879 Gevers EF, Wit JM, Robinson ICAF 1996 Growth, growth hormone (GH)-binding protein, and GH receptors are di¡erentially regulated by peak and trough components of the GH secretory pattern in the rat. Endocrinology 137:1013^1018 Giustina A, Veldhuis JD 1998 Pathophysiology of the neuroregulation of GH secretion in experimental animals and the human. Endocr Rev 19:717^797 Hammer RE, Brinster RL, Rosenfeld MG, Evans RM, Mayo KE 1985 Expression of human growth hormone-releasing factor in transgenic mice results in increased somatic growth. Nature 315:413^416 Howard AD, Feighner SD, Cully DF et al 1996 A receptor in pituitary and hypothalamus that functions in growth hormone release. Science 273:974^977 Jansson JO, Albertsson-Wikland K, Ede¤ n S, Thorngren KG, Isaksson OGP 1982 E¡ect of frequency of growth hormone administration on longitudinal bone growth and body weight in hypophysectomised rats. Acta Physiol Scand 114:261^265 Je¡ery S, Carter ND, Clark RG, Robinson ICAF 1990 The episodic secretory pattern of growth hormone regulates liver carbonic anhydrase III. Studies in normal and mutant growth hormone-de¢cient dwarf rats. Biochem J 266:69^74 Lin S-C, Lin C, Gukovsky I, Lusis A, Sawchenko P, Rosenfeld M 1993 Molecular basis of the little mouse phenotype and implications for cell-type speci¢c growth. Nature 364:208^213 McDowell RS, Elias KA, Stanley MS et al 1995 Growth hormone secretagogues: characterization, e⁄cacy, and minimal bioactive conformation. Proc Natl Acad Sci USA 92:11165^11169 Pellegrini E, Carmignac DF, Bluet-Pajot MT et al 1997 Intrahypothalamic growth hormone feedback: from dwar¢sm to acromegaly in the rat. Endocrinology 138:4543^4551 Robinson ICAF, Hindmarsh PC 1999 The importance of the GH secretory pattern for statural growth. In: Kostyo JL (ed) American handbook of physiology, vol 5: Hormonal control of growth. Oxford University Press, New York, in press Smith RG, Van der Ploeg LH, Howard AD et al 1997 Peptidomimetic regulation of growth hormone secretion. Endocr Rev 18:621^645 Tannenbaum GS, Ling N 1984 The interrelationship of growth hormone (GH)-releasing factor and somatostatin in the generation of the ultradian rhythm of GH secretion. Endocrinology 115:1952^1957
220
DISCUSSION
Tannenbaum GS, Martin JB 1976 Evidence for an endogenous ultradian rhythm governing growth hormone secretion in the rat. Endocrinology 98:562^570 Thomas GB, Cummins JT, Francis H, Sudbury AW, McCloud PI, Clarke IJ 1991 E¡ect of restricted feeding on the relationship between hypophysial portal concentrations of growth hormone (GH)^releasing factor and somatostatin, and jugular concentrations of GH in ovariectomized ewes. Endocrinology 128:1151^1158 Thorner MO, Perryman RL, Cronin MJ et al 1982 Somatotroph hyperplasia. Successful treatment of acromegaly by removal of a pancreatic islet tumour secreting an growth hormone-releasing factor. J Clin Investig 70:965^977 Wagner C, Caplan SR, Tannenbaum GS 1998 Genesis of the ultradian rhythm of GH secretion: a new model unifying experimental observations in rats. Am J Physiol 275:E1046^E1054 Waxman DJ 2000 Growth hormone pulse-activated STAT5 signalling: a unique regulatory mechanism governing sexual dimorphism of liver gene expression. In: Mechanisms and biological signi¢cance of pulsatile hormone secretion. Wiley, Chichester (Novartis Found Symp 227) p 61^81 Wehrenberg WB, Ling N, B˛hlen P, Esch F, Brazeau P, Guillemin R 1982 Physiological roles of somatocrinin and somatostatin in the regulation of growth hormone secretion. Biochem Biophys Res Comm 109:562^567 Wells T, Mode A, Floby E, Robinson ICAF 1994 The sensitivity of hepatic CYP2C gene expression to baseline growth hormone (GH) bioactivity in dwarf rats: e¡ects of GHbinding protein in vivo. Endocrinology 134:2135^2141 Wells T, Flavell DM, Wells SE, Carmignac DF, Robinson ICAF 1997 E¡ects of growth hormone secretagogues in the transgenic growth-retarded (Tgr) rat. Endocrinology 138:580^587 Willoughby J, Brogan M, Kapoor R 1989 Hypothalamic interconnections of somatostatin and growth hormone releasing factor on growth hormone secretion. Neuroendocrinology 50:584^591
DISCUSSION Veldhuis: I was especially interested in your notion of hormone trapping in the interstitial space. I guess with a tissue dialysis method you could measure the complex in lymph or interstitial £uid. There are probably some reasonable predictions that you could then test. One of the curious things in the human is that, with the exception of only two or three circumstances, high blood GH production rates are associated with relatively low plasma binding proteins (BPs) and vice versa. It would be very interesting to know what the interstitial £uid levels of either free or bound BP are in some of those circumstances. Robinson: There are cases of very high BP production where you do get a dominant-negative e¡ect. But if you look biochemically at the responses, you can get both inhibition and prolongation of e¡ect with a BP. The key question is what is the relevant physiological concentration in the vicinity of the responding tissue? One of my problems with the lymph is that it is too far away from the site of action. You may need optical measures to ¢gure out how much of that growth hormone is actually complexed and how much is available. We still don’t really know the
CONTROL OF GH RELEASE
221
availability to the receptors. From some of the work that we’ve done, even though GH is complexed, it is still able to signal, and it tends to signal a more continuous signal through those sexually dimorphic enzymes than a pulsatile signal. It is a numbers game though: too much BP is bad for you. Matthews: Are your data about BP generalizable? Robinson: It is species-dependent. In humans, if you give six months of continuous GH infusion, you actually increase the BP levels, whereas if you give injections of the same dose of GH you don’t. But the degree is not so marked as in the rat. Also there’s a problem with the way the BP is generated, which is di¡erent in humans and in rodents. Matthews: Does the same thing apply for insulin-like growth factor (IGF)-I? Robinson: The same arguments are have been proposed in the IGF ¢eld for years, that we have both inhibitory and stimulatory BP. Whether they’re delivering or blocking the IGF action is still a matter of debate. Veldhuis: In the IGF BP ¢eld, an interesting point I ran across recently is that BP4 is inhibitory in all systems tested, whereas some of the other BPs are either stimulatory or inhibitory. The story is therefore extremely messy. Sassone-Corsi: If you induce with GH-releasing proteins (RPs) and elicit di¡erent pulsatility, do you still see the same kind of growth? Robinson: You get the same pattern as you would with any other factor that is stimulating the GH output. The growth is dependent on what you’ve done to the GH pattern. By giving the GHRPs in di¡erent patterns you can dial up whatever GH exposure you would like. The physiological action of the GHRPs is primarily in the hypothalamus to drive through the pulse generator, rather than controlling the pituitary output. Sassone-Corsi: In the dwarf mice, how would the GHRPs work? Robinson: This is an interesting issue, because it is di⁄cult to ¢nd models which will respond selectively. If you look at the little mice, which have a GHRH receptor mutation, or in people with a GHRH receptor mutation, they should respond to the GHRPs (because these act on a di¡erent receptor from that for GHRH), but they don’t. You need an intact GHRH signalling system here, it seems. Why this is the case, when GHRPs and GHRH use di¡erent G protein-coupled receptors and second messenger systems, is an unanswered question. Licnio: Is there any physiological situation in which you presume that the endogenous ligand for the GHRP might be acting? Robinson: The circumstances in which the ligands have been e¡ective is in ageing and obesity. In those two situations GHRPs are relatively more e¡ective than GHRH. You might then imagine that under those circumstances the GHRP system is more active, but you could also argue that the GHRP receptor system hasn’t down-regulated as much as others.
222
DISCUSSION
Clarke: We reported last year a series of studies where we used a very elaborate reporter system that Dennis Leong has developed, in which he has transfected cells with the receptor, taken my samples of portal blood and put them into this assay system. In a cross-time series we see patterns of activity that activate the GHRP, and the concordance of that activity with growth hormone is around 85%. This is fairly strong evidence that there is an endogenous ligand. We don’t know what the ligand is. Goldbeter: What is the e¡ect of GH receptor desensitization? Robinson: On peripheral cells you can demonstrate a GHR desensitization. Acutely, you can show quite a signi¢cant down-regulation of GHRH receptor, but chronically that clearly doesn’t occur. You can see a biochemical change, but I’m not sure that this functionally means that you shut o¡ your signalling. You certainly can’t silence the system with continuous GHRH. If you give continuous GHRH you amplify endogenous pulsatility. Physiologically it’s very di⁄cult to obtain a correlate of the biochemical desensitization you can demonstrate under certain circumstances. The GHRH receptor tends not to be a very strongly desensitizing receptor. Marshall: One of the points you made is that GHRH is a trophic factor, presumably meaning that it increases GH synthesis. Robinson: Yes, and cell number. Marshall: What does GHRP do? Robinson: It doesn’t do that. Marshall: Put it this way: TRH would have been PRF (prolactin-releasing factor) if we had had a prolactin assay at the time TRH was discovered. The nomenclature is a function of the technology available at the time of study. It could be that GHRP is the GHRF (GH-releasing factor) and GHRH is, as you are suggesting, a trophic factor which is governing synthesis. Robinson: I agree. We don’t really understand what the physiological role of the GHRPs are. The problem is that the GHRPs synergize with GRF. To the extent that they do that, they ought to synergize with the trophic e¡ects of GHRH, but they don’t very well. Long-term treatment with GHRPs doesn’t give you any larger pituitaries or greater GH synthesis; on the contrary the GH responses wane quite dramatically. To the extent that they’re supposed to be synergizing with GHRH, something doesn’t quite add up. What is clear is that there is no long-term trophic e¡ect. Herbison: I agree. We should move away from this idea that just because something is inhibitory it is deleterious. In terms of the coordination of GH or somatostatin populations, there are reciprocal connections between these two populations. Thus you might only have to synchronize one, to lead to synchronization of the other population. Thinking of it in this context, it is not at all surprising that when you use a somatostatin antagonist
CONTROL OF GH RELEASE
223
the whole system shuts down. Indeed, you can pharmacologically modulate somatostatin neuron activity, and this also provides this nice stopping and reentrainment of GH pulses. Robinson: I agree. I’m not even sure that it’s the periventricular somatostatin population, because as you know there’s an arcuate somatostatin cell population. It may well be that there’s an intrinsic somatostatin interneuronal system in the arcuate nucleuswhichisactuallydoingthat.Wecan’tdistinguishinanyofthesetreatmentsas to the source of the somatostatin that’s impinging on these cells. We know we can electrically stimulate from periventricular to arcuate. The question of a role for somatostatin within the arcuate nucleus as a sort of circuit generator really hasn’t been addressed: it could easily be mediating these things. I agree with you; there are somatostatin receptors on GHRH neurons, and there’s clearly a close interconnection between the two peptidergic systems. I don’t think it’s a problem; I think that’s how the system works. Both inhibitory and stimulatory systems have to know what the other is doing in order to generate a regular series of responses. My interpretation of those antagonist experiments is that when you interrupt that communication the system just doesn’t know what to do and shuts down. Copinschi: There is evidence for GHRP receptors in some peripheral tissues, including cardiovascular tissues. Could you speculate about this? Robinson: I’m glad I have the opportunity to say something about this. Although GHRP binding sites have been demonstrated in peripheral tissues, GHRP receptors to my de¢nition have not been demonstrated. And in those tissues where one studies the mRNA for the identi¢ed GHRP receptors this is not present. My view of what’s in the literature is that that GHRP binding sites of an unidenti¢ed nature and of di¡erent a⁄nity states have been identi¢ed in di¡erent tissues. Before we just assume there’s a whole family of GHRP receptors which are bona ¢de receptors for that ligand, we should check out that the GHRP at higher concentrations isn’t hitting binding sites which actually are for some other ligand. Clarke: But it is also possible that there is more than one GHRP receptor. In fact, if you do parallel studies of the e¡ects of GHRP in di¡erent species, in some you can block the e¡ect of GHRP with the GRF antagonist, whereas in others you can’t. To my mind this suggests that there are other as yet unidenti¢ed GHRP receptors. Robinson: The GHRP analogues, all of which activate the cloned GHRP receptor, are not similar on these peripheral sites. I hesitate to call them GHRP receptors until we know exactly what the molecular basis of that binding is. Veldhuis: I wonder if I might challenge the mathematicians to help us by critiquing Iain Robinson’s attractive theory of random versus synchronous ¢ring: even though there are phasic trains within the argenine vasopressin ¢ring system, the trains are out of phase. Would you like to enlarge on this presentation? Robinson: I don’t want to go too far on the vasopressin system, but the oxytocin system is interesting, because two di¡erent stimuli produce di¡erent ¢ring patterns
224
DISCUSSION
and di¡erent outputs. The oxytocin milk ejection burst which gives a very narrow large pulse of oxytocin, shows this very synchronous neuronal activation, and if you record from more than one neuron you can show that many are ¢ring at the same time. However, if you osmotically activate the oxytocin system you get a continuous output, and you also get an increase in neuronal ¢ring rate but no synchronization of the ¢ring bursts. So the same neurons (you don’t have to go from oxytocin to vasopressin) will switch their hormone output patterns, and the only way they are doing this is by timing the synchrony of that activation. If one is trying to model GRF, you don’t have to activate the GRF neurons, you just have to synchronize them to get your pulses. An asynchronous activation will give you a continuous output. Veldhuis: Presumptively likewise for gonadotropin-releasing hormone (GnRH), a small amount of secretion per neuron will provide low basal output, and with synchonization an abrupt increase. Brown: Weknow a lot more about oxytocin and vasopressin thanwe do about either the GRF or GnRH system. One thing about the oxytocin system is that stimulation, either from the uterus or from the suckling input, is essential: this seems to be what convertsthe system to being capable of being pulsatile rather than just dribbling along in a continuous fashion. Also, when the system is pulsatile, it seems that for most of the ¢ve minutes or so between the pulses, the oxytocin cells are ticking along totally independently of each other. Then, all of a sudden, they all ¢re together for two seconds. If you compromise one nucleus, then cells in other nuclei are compromised, which suggests it’s a strong network phenomenon. From simultaneous electrophysiological recordings of oxytocin cells we have much more information about the likely mechanism than for GRF or GnRH cells. Robinson: With regard to what Peter Butler was saying yesterday, I was wondering whether his islets were all disconnected, and I wondered about the parathyroids also. Where you have disparate sources which you need to generate your pulses, there’s got to be some interconnection for synchrony. It is relatively easy to see that in a neuronal system where you have interconnections. Pincus: I think interconnection has to be 2D and not 1D: you have to look at the spatial arrangement. The point about synchrony is that you can say I’ve got this bigger percentage of things that are ¢ring at about the same time. To re¢ne the notion of interconnection, you want to think of synchrony being spawned by nearest neighbour movement: it is not just the percentage of things that ¢re at the same time that matters, but also an ascending of ¢rings that triggers nearby ¢rings, descending along a chain of spatially contiguous locations. You can get good synchronicity by a clump of ¢rings over here and a clump over there, but this is very di¡erent than this chain which evolves along contiguous point activation. This is what I think is going on here. I think it can be evaluated both empirically and theoretically.
Pulsatile parathyroid hormone secretion in health and disease Franz Schaefer Division of Pediatric Nephrology, University Children’s Hospital, Im Neuenheimer Feld 150, D-69120 Heidelberg, Germany
Abstract. In humans plasma parathyroid hormone (PTH) £uctuates episodically at a frequency of 6^7 bursts per hour. Approximately 30% of circulating PTH is attributable to pulsatile secretion and 70% to tonic secretion. PTH release is tightly controlled by Ca2+. Acute hypocalcaemia elicits a biphasic wave of PTH release, with an initial selective ampli¢cation and acceleration of the pulsatile component followed by proportionate stimulation of pulsatile and tonic secretion. Acute hypercalcaemia submaximally suppresses the frequency and mass of PTH bursts as well as tonic PTH release. Patients with primary hyperparathyroidism exhibit proportionate increases in pulsatile and tonic secretion, with no change in pulse frequency. In secondary hyperparathyroidism due to renal insu⁄ciency, tonic secretion and pulsatile burst mass are also proportionately ampli¢ed, and burst frequency is increased. Moreover, the hypocalcaemia-induced increase in burst frequency and mass as well as their suppression during hypercalcaemia is diminished, suggesting partial uncoupling of hyperplastic parathyroids from physiological regulatory mechanisms. While the secretory pattern of PTH and its dysregulation in disease states is now well de¢ned, the functional signi¢cance of pulsatile PTH signalling for target tissues is still largely unexplored. Preliminary work indicates that intermittent, in contrast to continuous, PTH administration stimulates bone formation. Cell culture studies suggest PTH receptor down-regulation with tonic exposure. 2000 Mechanisms and biological signi¢cance of pulsatile hormone secretion. Wiley, Chichester (Novartis Foundation Symposium 227) p 225^243
Numerous endocrine systems use pulsatile hormone signalling for transmitting complex information into cellular compartments. In peptide hormone-dependent tissues, information processing via cell membrane receptors usually involves receptor^hormone complex internalization and resynthesis of new receptor protein. The permanent oscillation between occupied and free receptor states induced by pulsatile hormone secretion is believed to be a particularly e⁄cient and economical mode of signal transduction. While the pathophysiological importance of alterations of the physiological mode of hormone secretion has 225
226
SCHAEFER
been demonstrated in several diseases, the functional signi¢cance of pulsatility in many other endocrine systems is still unknown (Matthews 1991). The parathyroid gland quadruplet was among the last endocrine organs for which a pulsatile secretion mode was demonstrated both in animals and in humans (Fox et al 1981, Harms et al 1989, Kitamura et al 1990). It is now clear that the plasma concentrations of parathyroid hormone (PTH) £uctuate episodically. The pulses are of small amplitude and have rapid decay rates, explaining why they escaped some early studies using infrequent sampling protocols and PTH assays of limited sensitivity and speci¢city (Herfarth et al 1992). It is likely that a dual mode of PTH secretion (e.g. combined tonic and pulsatile) underlies the concentration pattern typically observed in animals and humans. Technical aspects of PTH pulse analysis Disappearance half-life The detection of episodic secretory events in a hormone concentration vs. time pro¢le critically depends on the knowledge of the plasma disappearance half-life of the hormone. For PTH, widely discrepant half-life estimates, ranging from 2^ 35 min, have been obtained in the past (Goltzman et al 1984, Brasier et al 1988). This wide range may be explained by factors such as disregard of residual PTH secretion from remaining parathyroid tissue in parathyroidectomy studies, insu⁄cient sampling frequencies and assay speci¢city problems. In an isotope study in dogs an intact PTH half-life of 1.2^2.8 min was found (Fox et al 1983). Recently, we determined PTH half-life in humans during suppression of endogenous PTH release by Ca2+ infusion. Analysing the initial exponential decay and allowing for admixed non-suppressible basal secretion, we observed a very similar plasma disappearance half-life of 2.4 min (range 2.0^2.9 min) (Schmitt et al 1996). Major clearance sites of PTH are the kidney and the liver; in patients with reduced renal function we observed an inverse relationship between plasma PTH half-life and glomerular ¢ltration rate (Schmitt et al 1998a). In patients on maintenance dialysis with less than 10% residual renal function, mean PTH halflife was 6.6 min, as compared to 2.6 min in control subjects with normal kidney function. Considering the very short half-life of plasma PTH, it is conceivable why early studies using insu⁄cient sampling frequencies were not able to detect PTH pulsatility. Theoretical considerations imply that the false-negative detection error reaches a minimum only when three samples per half-life are obtained (Veldhuis et al 1986). Accordingly, at least one sample per minute should be obtained in studies of instantaneous PTH release.
PULSATILE PARATHYROID HORMONE SECRETION
227
Deconvolution analysis In the past, several mathematical algorithms have been developed to objectively assess £uctuations of plasma hormone concentrations. The multiparameter deconvolution model is perhaps the most sophisticated approach developed for this purpose (Veldhuis et al 1987). Based on the concept that the measurable plasma concentration vs. time pro¢le of a hormone is the net result of a number of secretory events and continuously acting elimination processes, this algorithm is a quantitative estimation of various speci¢c hormone secretion parameters using all hormone concentrations and their intra-assay variances simultaneously. The pulsatile hormone secretion rate is calculated as the product of the number of secretory events and the mass of hormone secreted per burst. The tonic hormone secretion rate re£ects a baseline amount of circulating hormone that cannot be attributed to the pulsatile secretory component. The application of the multiparameter deconvolution technique has enabled us and others to analyse both spontaneous and Ca2+-modulated PTH secretion in healthy and diseased subjects, accounting for disease-speci¢c di¡erences in PTH half-life (Samuels et al 1993, Schmitt et al 1996, 1998a,b).
Dual mode of spontaneous PTH secretion In humans PTH is apparently secreted by two separate mechanisms (Samuels et al 1993, Schmitt et al 1996). Under normocalcaemic conditions about 30% of total PTH secretion is released in an episodic or pulsatile fashion with a mean frequency of 6^7 pulses per hour and a burst half-duration of approximately 2.5 min. The average maximal secretion rate during spontaneous bursts is approximately 25% of the prevailing mean plasma level. The remaining 70% of total PTH secretion is attributable to tonic (time-invariant) hormone release (see also Figs 1, 2 and 3). At present, the physiological mechanisms underlying tonic and pulsatile PTH secretion are still largely speculative. PTH is stored in cytoplasmic vesicles before release (Pesce et al 1989). It is tempting to postulate that dual mechanisms control exocytosis of these granules, e.g. one might represent a relatively unvariable constitutive or tonic form of vesicular secretion, and the other pathway a highly regulated mode of secretion of readily releasable granules primarily involved in the short-term control of PTH release within a feedback-controlled homeostatic system. In some endocrine systems exocytosis via the constitutive pathway is a continuous process at a constant rate which is limited by the biosynthesis rate of the protein. In contrast, regulated secretory processes involve granular storage of proteins at high concentrations for extended periods before release is induced by a secretagogue (Halban &
228
SCHAEFER
Irminger 1994). In some cell types the two mechanisms seem to occur simultaneously (Farnworth 1995). The primary external factor a¡ecting instantaneous PTH secretion appears to be plasma ionized Ca2+ (see below), raising the possibility of a non-linear feedback system. However, short-term £uctuations of ionized Ca2+ in temporal association with plasma PTH oscillations have not been demonstrated conclusively so far (Fox et al 1981) (Fig. 1).
FIG. 1. Concentration vs. time pro¢le of PTH plasma PTH and Ca2+ in thyroid venous e¥uent blood collected in 1 min fractions in an anaesthesized dog. PTH pulses occur at a mean interval of 8.4 min. Rise in plasma PTH levels towards end of study may have resulted from falling Ca2+ concentrations due to blood loss. (From Fox et al 1981, with permission).
PULSATILE PARATHYROID HORMONE SECRETION
229
Further, as yet hypothetical signals might include neuronal a¡erences and autocrine or paracrine feedback regulation via PTH or PTH fragments, endothelin or chromogranin A and chromogranin A-related peptides (Fasciotto et al 1990, Fuji et al 1991). Alternatively, it is possible that the pulsatile component of PTH secretion re£ects cyclic changes in neuronal input to the parathyroid glands (De Boer et al 1986). The parathyroids are densely innervated by neuronal a¡erences, and adrenergic receptors are expressed by parathyroid cells (Amenta et al 1980, Stern & Cardinali 1994). Furthermore, oscillatory changes in perfusion of the parathyroid glands due to autonomous rhythms of the arteriolar tone may be involved in the generation of plasma PTH pulses. Finally, the apparent dual mode of PTH secretion may be due to exclusively pulsatile release from multiple independent pacemakers, e.g. within each individual parathyroid gland, beating without temporal synchrony. In such a setting the discernible plasma concentration pulses would correspond to several individual secretory bursts coinciding by chance, and the apparent tonic secretory component would be the mathematical equivalent of variably phaseshifted bursts. Sensitivity of instantaneous PTH release to modulation of ionized Ca2+ The most important physiological regulator of instantaneous PTH secretion is ionized Ca2+. A speci¢c cell membrane receptor for Ca2+ has been identi¢ed on parathyroid cells (Brown et al 1993). Activation of this receptor by acute hypercalcaemia instantaneously suppresses PTH secretion, whereas hypocalcaemia results in an immediate increase of PTH release. We have used the sodium citrate/calcium gluconate clamp infusion technique to acutely modulate blood ionized Ca2+ (Schwarz et al 1993). With this we were able to study the e¡ects of dynamic changes in the ionized Ca2+ milieu on minute-to-minute PTH release (Schmitt et al 1996, 1998a,b). PTH secretion during hypocalcaemia Clamping of the ionized Ca2+ concentration at a lower level ( 0.2 mmol/l) elicits a biphasic response of PTH secretion (Fig. 2). A brisk increase in PTH concentration occurs during the ¢rst 30 min while ionized Ca2+ is decreasing to the new steadystate level. When steady-state hypocalcaemia is reached, plasma PTH concentrations gradually decrease again but remain elevated three times above baseline. The corresponding secretory changes underlying this plasma concentration pattern are shown in the lower panel of Fig. 2. In healthy subjects the induction of hypocalcaemia elicits an immediate selective increase in the pulsatile secretion component by more than 1100%, brought about by a
230
SCHAEFER
FIG. 2. Plasma PTH vs. time pro¢le measured at 1 min intervals (mean SD; upper panel) and corresponding PTH secretion pro¢le calculated by deconvolution analysis (lower panel) in a healthy young man during 75 min of normocalcaemia, subsequent 30 min of decreasing ionized calcium by 0.2 mmol/l and 75 min of steady-state hypocalcaemia (citrate clamp technique). Maximal stimulation of PTH secretion, mediated by selective increase in pulse amplitude and frequency, is observed during initial adaptation period. A magni¢cation of the baseline PTH pro¢le is inserted in the upper panel. (From Schmitt et al 1996, with permission.)
combined increase in burst frequency (from six to 12 bursts per hour) and the hormone mass secreted per burst (by ¢vefold) (Fig. 4). The tonic secretion rate remains unchanged. The increase in pulsatile PTH release during the hypocalcaemia induction period was found to be positively correlated with the rate of decrease in ionized Ca2+.
PULSATILE PARATHYROID HORMONE SECRETION
231
During the subsequent period of steady-state hypocalcaemia, the tonic PTH secretion rate increases by 150% compared to normocalcaemic conditions, burst frequency returns to the baseline level and PTH burst mass decreases to a level still twofold higher than under baseline conditions. Moreover, using the approximate entropy statistic (Pincus 1991) we observed that PTH bursts are secreted at a higher degree of orderliness during hypocalcaemia as compared to normocalcaemic conditions.
FIG. 3. Plasma PTH concentration vs. time pro¢le (mean SD, upper panel) and PTH secretion pro¢le estimated by deconvolution analysis (lower panel) during hypercalcaemic clamp study in a healthy young adult. Incomplete, proportionate suppression of pulsatile and tonic PTH secretion. Small PTH pulses reappear after 30 min of hypercalcaemia. (From Schmitt et al 1996, with permission.)
232
SCHAEFER
FIG. 4. Comparison of relative changes in PTH secretion characteristics during hypocalcaemic clamp study in 13 healthy control subjects (*, uninterrupted line) and 16 patients with secondary hyperparathyroidism (*, dotted line). Asterisks denote signi¢cant di¡erences between groups (P50.05). (From Schmitt et al 1998b, with permission.)
PTH secretion during hypercalcaemia An increase of serum ionized Ca2+ by 0.2 mmol/l abolishes pulsatile PTH secretion during the initial adaptation period, resulting in a marked decrease of plasma PTH
PULSATILE PARATHYROID HORMONE SECRETION
233
concentrations to a non-suppressible baseline concentration of 0.6 pmol/l (Fig. 3). During the subsequent 75 min of steady-state hypercalcaemia, secretory bursts reoccur less frequently and with a lower mass per burst ( 30% and 80% respectively compared to the baseline) (Fig. 5). The regularity of
FIG. 5. Comparison of relative changes in PTH secretion characteristics during hypercalcaemic clamp study in 13 healthy control subjects (*, uninterrupted line) and 16 patients with secondary hyperparathyroidism (*, dotted line). Asterisks denote signi¢cant di¡erences between groups (P50.05). (From Schmitt et al 1998b, with permission.)
234
SCHAEFER
pulsatile PTH secretion is not altered during steady-state hypercalcaemia. The pulsatile and tonic components of PTH secretion are proportionately reduced by approximately 80%. The pattern of response to hypo- and hypercalcaemic stimuli suggests a highly sensitive trigger mechanism that is more responsive to acute changes in ionized Ca2+ concentration than to a tonic resetting of absolute Ca2+ levels. The Ca2+sensing membrane protein may be involved in this immediate response. While the precise cellular mechanisms governing the modulation of PTH secretion remain to be elucidated, the transient increase in burst frequency during acute hypocalcaemia and the enhanced process regularity of pulsatile PTH release during steady-state hypocalcaemia mandate some degree of intercellular and perhaps even interglandular synchronization. Possible mechanisms explaining the more gradual rise in tonic PTH release during steady-state hypocalcaemia are changes in the intracellular degradation of PTH, regulating the availability of newly synthesized hormone (Habener et al 1975) or increased recruitment of parathyroid cells in the active phase by synchronization (Ritchie et al 1992, Roth & Raisz 1966).
E¡ect of vitamin D treatment on pulsatile PTH release Besides ionized calcium, active vitamin D (calcitriol, 1a,25-dihydroxy-vitamin D3) is an important regulator of PTH secretion. Its de¢ciency is the major cause, and its substitution the mainstay of prevention and treatment, of secondary hyperparathyroidism in patients with chronic renal failure. Calcitriol reduces PTH availability by suppressing PTH transcription (Silver et al 1986) and by inhibiting parathyroid cell proliferation. These e¡ects are mediated both directly via a speci¢c vitamin D receptor and indirectly by stimulation of intestinal Ca2+ resorption. We were able to demonstrate that exogenous calcitriol also a¡ects spontaneous PTH secretion as well as the sensitivity of the parathyroid to Ca2+ (Schmitt et al 1998b). After 5 d of oral calcitriol treatment, healthy young adults showed a 30% decrease of plasma PTH concentrations by 35%, mediated by a decrease in the frequency of PTH bursts and of tonic secretion rate. During hypocalcaemic stimulation, the pulsatile secretory component was selectively reduced, mainly due to a reduction in PTH burst mass ( 30%). In contrast, acute intravenous injections of calcitriol did not a¡ect PTH secretion. These ¢ndings are consistent with a genomic e¡ect of calcitriol, and make a direct interference with the regulation of pulsatile PTH release unlikely. Calcitriol treatment, by virtue of its anti-transcriptional action, probably reduces the amount of granule-stored PTH available for intermittent exocytosis.
PULSATILE PARATHYROID HORMONE SECRETION
235
PTH secretion in bone disease Osteoporosis Alterations in the dynamics of the PTH signal have been demonstrated in early postmenopausal women. The abnormalities seem to concern the rhythmicity of PTH release: time-series analysis disclosed a less ‘predictable’, irregular pattern of PTH secretion in osteoporotic patients in comparison to healthy subjects (Prank et al 1995). It was suggested that a loss of PTH rhythmicity may be causally involved in the alteration of bone remodelling that is characteristic of postmenopausal osteoporosis. Hormonereplacementtherapyinpostmenopausalwomen seemstodecreasePTH burst mass and tonic secretion, but not burst frequency (Harms et al 1994a). The authors did not report on any e¡ect of sex steroid replacement on the rhythmicity of PTH release. A more recent investigation did not observe any in£uence of either osteoporosis or hormone replacement therapy on the orderliness of PTH secretion as assessed by the approximate entropy method (Samuels et al 1997). Primary hyperparathyroidism Interestingly, patients with primary hyperparathyroidism retain a dual (i.e. tonic and pulsatile) pattern of PTH secretion. Harms et al (1994b) observed a proportionate increase in tonic and pulsatile secretion in nine patients with primary hyperparathyroidism. The latter was due to a selective increase in the amplitude of PTH pulses, whereas burst frequency was unchanged. Crosscorrelation analysis of PTH and Ca2+ concentration pro¢les indicated an impaired feedback regulation in primary hyperparathyroidism. Secondary hyperparathyroidism In patients with secondary hyperparathyroidism due to chronic renal failure, we observed a consistently higher baseline PTH burst frequency and a decreased responsiveness to increases as well as decreases in ionized Ca2+ compared with euparathyreote, non-uraemic controls (Schmitt et al 1998b). During acute hypocalcaemia, both the acceleration and the relative ampli¢cation of PTH bursts was 50% lower in the patient group than in the controls (Fig. 4). The hypercalcaemic clamp study showed a less marked suppression of PTH secretion in the patients, in whom PTH burst frequency was una¡ected by hypercalcaemia (Fig. 5). These ¢ndings provide evidence that not only cell mass-dependent (burst mass), but also regulation-dependent elements of PTH release (burst frequency, modulation by Ca2+) are altered in uraemia, suggesting partial uncoupling of the hyperplastic parathyroid glands from the physiological regulatory mechanisms of pulsatile PTH release.
236
SCHAEFER
The recent ¢nding of a diminished expression of the Ca2+ membrane receptor in patients with uraemic hyperparathyroidism (Kifor et al 1996) o¡ers a possible mechanism to explain the diminished parathyroid responsiveness observed both during hypocalcaemic stimulation and during hypercalcaemic suppression. Of course, an alternative or additional post-receptor defect cannot be excluded at the current state of knowledge. Whereas defective Ca2+ sensing may readily explain the observed alterations in the modulation of PTH burst mass in uraemia, the pathophysiology of the abnormal variability of PTH burst frequency during hypo- and hypercalcaemia is less evident. The patients exhibited an elevated burst frequency at baseline, a relatively lower increase during hypocalcaemia and lack of suppressibility during hypercalcaemia. Although the physiological basis for the pulsatile organization of PTH release is unknown, the presence of neuronal ¢bres in the parathyroids and of b-adrenergic receptors on parathyroid cell surfaces (Amenta et al 1980, Stern & Cardinali 1994) is compatible with synchronization via the sympathetic nervous system. Thus, a central nervous system ‘pacemaker’ may exist in analogy to other neuroendocrine hormone ensembles. Recently, evidence has been presented that the Ca2+-sensing protein is expressed not only in the parathyroid and kidney but also in certain brain areas (Rogers et al 1997). Ca2+-sensing receptors in the brain could provide a biological basis for a central modulation of PTH burst frequency in response to acute alterations of the Ca2+ milieu. In that case, the disorders of spontaneous and Ca2+-modulated PTH burst frequency observed in the patient group would be consistent with the notion of a de¢cient expression of the Ca2+ receptor not only in the parathyroid but also in the brain in uraemia. This possibility is speculative, but merits further research. Biological signi¢cance of pulsatile PTH signalling A potential pathophysiological relevance of the pulsatile PTH secretion mode is suggested by in vitro experiments in cells derived from PTH target tissues, i.e. bone and renal tubules. The PTH receptor is coupled to both adenylate cyclase and phospholipase C, thus stimulating in parallel cAMP and IP3 production and protein kinase C (PKC) activation. Exposure of renal and bone cell cultures to high, constant levels of PTH results in a rapid decrease in PTH binding and PTH-stimulated cAMP production, suggesting down-regulation of the PTH receptor (Fukayama et al 1992). Both the PKC and the protein kinase A (PKA) pathway appear to contribute to PTH receptor down-regulation. The time courses of IP3, PKC and cAMP stimulation di¡er almost by an order of magnitude (Carvalho et al 1994). It is tempting to speculate that (1) the pattern of plasma PTH £uctuations (pulse frequency, pulsatile/tonic ratio) might control
PULSATILE PARATHYROID HORMONE SECRETION
237
PTH action by a¡ecting the receptor status, and (2) the di¡erent intracellular second messenger systems might be activated with di¡erent e⁄cacy dependent on the characteristics of the pulsatile signal. This would be a mechanism to preserve PTH signal speci¢city in the complex regulation of bone turnover. In animals, continuous infusion of PTH decreases bone mass, whereas intermittent (once daily) administration results in a net increase in trabecular bone volume (Tam et al 1982, Dobnig & Turner 1997). Several studies in humans are in accordance with these results, demonstrating a positive e¡ect of ‘pulsatile’ PTH administration on bone trabecular mass (e.g. Hesch et al 1989). The optimal exposure time to achieve an anabolic e¡ect of PTH is 1 h (Dobnig & Turner 1997). At present it is unclear whether the anabolic e¡ect of intermittently administered exogenous PTH has any analogy in the endogenous small-amplitude PTH pulses discussed above. While the markedly di¡erent time course (6 pulses/h vs. 1 pulse/d) and the much higher concentrations achieved with pharmacological dosing suggest that the two principles are not comparable, the preservation of PTH receptor expression may be a biological advantage common to both mechanisms. In summary, while the state of knowledge regarding the phenomenon of pulsatile PTH secretion and its modulation by ionized Ca2+ has improved markedly within the last few years, little is known about the mechanisms synchronizing parathyroidal PTH release. Moreover, the biological functions of the dual mode of PTH secretion still largely remain to be elucidated. References Amenta F, Cavalotti C, de Rossi M, de Santis A 1980 Beta adrenergic receptors in the parathyroid glands. Naunyn Schmiedebergs Arch Pharmacol 313:195^198 Brasier AR, Wang CA, Nussbaum SR 1988 Recovery of parathyroid hormone secretion after parathyroid adenectomy. J Clin Endocrinol Metab 66:495^500 Brown EM, Gamba G, Riccardi D et al 1993 Cloning and characterization of an extracellular Ca2+-sensing receptor from bovine parathyroid. Nature 366:575^580 Carvalho RS, Scott JE, Suga DM, Yen EH 1994 Stimulation of signal transduction pathways in osteoblasts by mechanical strain potentiated by parathyroid hormone. J Bone Miner Res 9:999^1011 De Boer RW, Karemaker JM, Strackee J 1986 On the spectral analysis of blood pressure variability. Am J Physiol 251:H685^H687 Dobnig H, Turner RT 1997 The e¡ects of programmed administration of human parathyroid hormone fragment (1-34) on bone histomorphometry and serum chemistry in rats. Endocrinology 138:4607^4612 Farnworth PG 1995 Gonadotrophin secretion revisited. How many ways can FSH leave a gonadotroph? J Endocrinol 145:387^395 Fasciotto BH, Gorr SU, Bourdeau AM, Cohn DV 1990 Autocrine regulation of parathyroid secretion: inhibition of secretion by chromogranin A (secretory protein-I) and potentiation of secretion by chromogranin-A and pancreastatin antibodies. Endocrinology 127: 1329^1335
238
SCHAEFER
Fox J, O¡ord KP, Heath H III 1981 Episodic secretion of parathyroid hormone in the dog. Am J Physiol 241:E171^E177 Fox J, Scott M, Nissenson RA, Heath H III 1983 E¡ect of plasma calcium concentration on the metabolic clearance rate of parathyroid hormone in the dog. J Lab Clin Med 102:70^77 Fuji YF, Moreira JE, Orlando C et al 1991 Endothelin as an autocrine factor in the regulation of parathyroid cells. Proc Natl Acad Sci USA 88:4235^4239 Fukayama S, Tashjian AH Jr, Bringhurst FR 1992 Mechanisms of desensitization to parathyroid hormone in human osteoblast-like SaOS-2 cells. Endocrinology 131:1757^1769 Goltzman D, Gomolin H, DeLean A, Wexler M, Meakins JL 1984 Discordant disappearance of bioactive and immunoreactive parathyroid hormone after parathyroidectomy. J Clin Endocrinol Metab 58:70^75 Habener JF, Kemper B, Potts JT Jr 1975 Calcium dependent intracellular degradation of parathyroid hormone: a possible mechanism for the regulation of hormone stores. Endocrinology 97:431^441 Halban PA, Irminger JC 1994 Sorting and processing of secretory proteins. Biochem J 299:1^18 Harms HM, Kaptaina U, Kˇlpmann WR, Brabant G, Hesch RD 1989 Pulse amplitude and frequency modulation of parathyroid hormone in plasma. J Clin Endocrinol Metab 69:843^851 Harms HM, Neubauer O, Kayser C et al 1994a Pulse amplitude and frequency modulation of parathyroid hormone in early postmenopausal women before and on hormone replacement therapy. J Clin Endocrinol Metab 78:48^52 Harms HM, Schlinke E, Neubauer O et al 1994b Pulse amplitude and frequency modulation of parathyroid hormone in primary hyperparathyroidism. J Clin Endocrinol Metab 78: 53^57 Herfarth K, Schmidt-Gayk H, Graf S, Maier A 1992 Circadian rhythm and pulsatility of parathyroid hormone secretion in man. Clin Endocrinol 37:511^519 Hesch RD, Busch U, Prokop M, Delling G, Rittinghaus EF 1989 Increase of vertebral density by combination therapy with pulsatile 1-38 hPTH and sequential addition of calcitonin nasal spray in osteoporotic patients. Calcif Tissue Int 44:176^180 Kifor O, Moore FD Jr, Wang P et al 1996 Reduced immunostaining for the extracellular Ca2+sensing receptor in primary and uremic secondary hyperparathyroidism. J Clin Endocrinol Metab 81:1598^1606 Kitamura N, Shigeno C, Shiomi K et al 1990 Episodic £uctuation in serum intact parathyroid hormone concentration in men. J Clin Endocrinol Metab 70:252^263 Matthews DR 1991 Physiological implications of pulsatile hormone secretion. Ann N Y Acad Sci 618:28^37 Pesce C, Tobia F, Carli F, Antoniotti GV 1989 The sites of hormone storage in normal and diseased parathyroid glands: a silver impregnation and immunohistochemical study. Histopathology 15:157^166 Pincus SM 1991 Approximate entropy as a measure of system complexity. Proc Natl Acad Sci USA 88:2297^2301 Prank K, Nowlan SJ, Harms HM et al 1995 Time series prediction of plasma hormone concentration. Evidence for di¡erences in predictability of parathyroid hormone secretion between osteoporotic patients and normal controls. J Clin Invest 95:2910^2919 Ritchie CK, Cohn DV, Maercklein PB, Fitzpatrick LA 1992 Individual parathyroid cells exhibit cyclic secretion of parathyroid hormone and chromogranin-A (as measured by a novel sequential hemolytic plaque assay). Endocrinology 131:2638^2642 Rogers KV, Dunn CK, Hebert SC, Brown EM 1997 Localization of calcium receptor mRNA in the adult rat central nervous system by in situ hybridization. Brain Res 744:47^56
PULSATILE PARATHYROID HORMONE SECRETION
239
Roth SI, Raisz LG 1966 The course and reversibility of the calcium e¡ect on the ultrastructure of the rat parathyroid gland in organ culture. Lab Invest 15:1187^1211 Samuels MH, Veldhuis JD, Cawley C et al 1993 Pulsatile secretion of parathyroid hormone in normal young subjects: assessment by deconvolution analysis. J Clin Endocrinol Metab 77:399^403 Samuels MH, Veldhuis JD, Kramer P, Urban RJ, Bauer R, Mundy GR 1997 Episodic secretion of parathyroid hormone in postmenopausal women: assessment by deconvolution analysis and approximate entropy. J Bone Miner Res 12:616^623 Schmitt CP, Schaefer F, Bruch A et al 1996 Control of pulsatile and tonic parathyroid hormone secretion by ionized calcium. J Clin Endocrinol Metab 81:4236^4243 Schmitt CP, Schaefer F, Huber D et al 1998a 1,25(OH)2-vitamin D3 reduces spontaneous and hypocalcaemia-stimulated pulsatile component of parathyroid hormone secretion. J Am Soc Nephrol 9:54^62 Schmitt CP, Huber D, Mehls O et al 1998b Altered instantaneous and calcium-modulated oscillatory PTH secretion patterns in patients with secondary hyperparathyroidism. J Am Soc Nephrol 9:1832^1844 Schwarz P, Srensen HA, McNair P, Transbl I 1993 Cica-clamp technique: a method for quantifying parathyroid hormone secretion: a sequential citrate and calcium clamp technique. Eur J Clin Invest 23:546^553 Silver J, Naveh-Many T, Mayer H, Schmelzer HJ, Popovtzer MM 1986 Regulation by vitamin D metabolites of parathyroid hormone gene transcription in vivo in the rat. J Clin Invest 78:1296^1301 Stern JE, Cardinali DP 1994 In£uence of the autonomic nervous system on calcium homeostasis in the rat. Biol Signals 3:15^25 Tam CS, Heersche JN, Murray TM, Parsons JA 1982 Parathyroid hormone stimulates the bone apposition rate independently of its resorptive action: di¡erential e¡ects of intermittent and continuous administration. Endocrinology 110:506^512 Veldhuis JD, Evans WS, Johnson ML, Wills MR, Rogol AD 1986 Physiological properties of the luteinizing hormone pulse signal: impact of intensive and extended venous sampling paradigms on its characterization in healthy men and women. J Clin Endocrinol Metab 62:881^891 Veldhuis JD, Carlson ML, Johnson ML 1987 The pituitary gland secretes in bursts: appraising the nature of glandular secretory impulses by simultaneous multipleparameter deconvolution of plasma hormone concentrations. Proc Natl Acad Sci USA 84:7686^7690
DISCUSSION Veldhuis: I’d like to ask the cell biologists and signalling experts to give us their opinion on whether these 6 and 8 min signals could actually be transduced successfully and repetitively. Sassone-Corsi: It sounds too short to me. As a repetitive pulse this sounds very di⁄cult. Veldhuis: We have the same issue but with a totally di¡erent signalling mode for insulin action, of course.
240
DISCUSSION
Butler: Certainly for hepatic glucose release, if you directly measure it across the liver, it responds, but of course inhibition of glycogenolysis is a pretty rapid process, so perhaps it depends on the signal. Goldbeter: In Dictyostelium, as I mentioned in my paper (Goldbeter et al 2000, this volume), pulsatile signals of cAMP delivered with a periodicity of 5^10 min can be successfully transduced in regard to synthesis and release of a new cAMP pulse by stimulated cells. The refractory period for this process is of the order of minutes and corresponds to the recovery of the cAMP receptor to its active state, probably through dephosphorylation. Sassone-Corsi: We have studied the refractory period after the ¢rst pulse of induction of gene expression. During this period things such as PKA are inactive or even some transcription factors like CREB cannot be dephosphorylated. As a ¢rst pulse this could be done within 10^15 min, but then the shortest refractory period we see is 5 h. It is di⁄cult to envisage pulses of gene expression shorter than that. Schaefer: Could PKC-mediated phosphorylation processes respond di¡erentially to pulses? Sassone-Corsi: Strangely enough, to date no transcription factors directly under the control of PKC have been found. If you look at phosphorylation of various transcription factors by the MAP kinase pathway, this also occurs rapidly, but then you need an event of dephosphorylation and reactivation of the whole transcription machinery, which takes time. Brabant: What is the time-scale of the transcription factor control you are looking at and how closely can you dissolve the temporal pattern? Even on the protein level using tools like green £uorescent protein to monitor timedependent changes there is still not a high temporal resolution for intracellular time-dependent processes. So the question is on what time-scale are we currently able to continuously monitor intracellular signalling, transcription processes and do these techniques already allow detailed answers? Sassone-Corsi: You are talking about the protein product, whereas I am talking about pure transcription, which involves looking at nascent RNA. Butler: If you can’t do it every minute, you can’t answer the question. Sassone-Corsi: There are incorrect ideas about how transcripts are made. If you are looking at making one complete transcript, that itself takes time. There are genes expressed in an oscillatory fashion in Drosophila which are 100 kb. To make one of these transcripts takes about an hour and a half. This is a problem. Transcription itself is a cyclic event. But making waves of transcription, which is really accumulating one transcript at one time and then shut it o¡, that takes time because you need one RNA to be made, terminated and then a new wave of transcript to be made. Nascent RNA can be detected within minutes, because we can do run-on experiments.
PULSATILE PARATHYROID HORMONE SECRETION
241
Veldhuis: Am I right in saying that you have an underlying assumption here that all the cells are responding uniformly and identically all the time? We could postulate a situation where 2% of the cells respond to any given 8 min pulse. We have implicitly accepted the idea that there is an instantaneous homogeneity of machinery responsiveness, which may not be true. Marshall: In addition, what determines how many run-o¡s you make per signal input? In other words, when a signal initiates transcription, presumably there is something that determines the number of cycles the polymerase can make. Is that just the availability of co-factors, or is it a ¢xed number of cycles per stimulatory signal? Sassone-Corsi: You are talking about a very complex situation, which involves the presence of co-factors, polymerase availability and chromatin opening. Marshall: The issue is, is it similar for all transcriptional mechanisms? Sassone-Corsi: Absolutely not. One commonly made assumption is to think that all genes are transcribed the same way, but this is not the case. Licinio: Could the pulse response be a¡ecting not transcription directly (because there is not time for that) but something before transcription in the cytoplasm? Then, depending on how that’s regulated, transcription will be a¡ected. Sassone-Corsi: Exactly. When you are talking about phosphorylation of transcription factors, this is what happens. NF-kB is a great example. The factor is held in the cytoplasm by a repressor, I-kB. When phorbol esters or stress signals hit the cell, I-kB stays in the cytoplasm, and allows the activator to migrate into the nucleus. The signalling event therefore happens in the cytoplasm and then the factor goes into the nucleus. Robinson: I wonder whether we’ve not gone a bit too distal in this cascade. What’s known about receptor turnover, or even inactivation kinetics for these receptors? First you have to demonstrate that a pulsatile exposure in an in vitro system at that frequency gives you a di¡erence, then you have to explain it. Instead of going down to the transcriptional level, there may be a simpler explanation, such as the availability of activatable receptors at the end of 8 min versus the beginning of 8 min. Do we know about that? Schaefer: The problem with all those cell studies is that they give a very large pulse of hormone to the cells, and then you have a maximum activation of cAMP, for instance, in the osteoblasts. This results in inactivation of the PTH receptor and this down-regulation lasts for 6^8 h at least. But this is not a physiological situation. I have di⁄culty imagining that a cell hit by a hormone in vivo will be shut-o¡ to further input for 6^8 h. Sassone-Corsi: I think that Iain Robinson’s point is well taken. We are talking about two di¡erent ballparks. One is long-term desensitization, or long-term e¡ects: these de¢nitely involve transcription. Then there are short-term e¡ects such as receptor desensitization, at the pharmacological level. We have done
242
DISCUSSION
experiments on both thyrotropin and the follicle-stimulating hormone (FSH) receptor (Lalli & Sassone-Corsi 1995, Monaco et al 1995). When you purify primary Sertoli cells from testes and put them in a plate, there are two e¡ects of FSH action. One is a short-term desensitization, but the other is a long-term e¡ect, which is transcriptional, that induces a decrease of the gene’s expression within about an hour and a half. We have shown that in the promoter of the gene encoding these receptors, there is a cAMP-responsive element which once blocked by the repressor ICER (a CREM gene product) causes that transcription in these cells to be completely non-responsive to FSH during that time. After about 5 or 6 h, the receptor expression goes back to normal and the cells again are responsive. We have determined that in the refractory period of receptor expression, the transcription repressor ICER is blocking that expression. This is a long-term e¡ect. Butler: With b cells in the liver, the receptors take about 4 min once they have bound to their ligand on the hepatocyte, they’re internalized, and then the receptor gets back to the hepatocyte surface within 4 min. The timing there is rather good, because if you have pulses coming every 6 min, it means that the receptors are available each time a pulse arrives. Coming back to the data Franz Schaefer presented, I have a slight concern about the pulse frequency modulation. You show a nice correlation between pulse mass versus pulse frequency. Is it possible that this mass/frequency relationship is an artefact caused by lower detection of small pulses? Schaefer: We have not been able to do that yet, because PTH was not available until recently. However, we have checked for this by doing triplet determinations of PTH in the suppressed phase. We obtained a very reasonable assay coe⁄cient of variation around 5^6%, even with these low serum PTH levels, so we are rather con¢dent about the pulse frequency. Butler: There is analogy here with insulin: insulin assayed by a conventional assay 20 times did not resolve the detection of small pulses in the periphery (O’Meara et al 1993). Schaefer: In some of the normals we had this problem, but not in the uraemic patients, because they don’t go down as much. If they are suppressed, they are still even higher than the normals. In these patients, frequency still increased during the hypocalcaemic stimulation. Butler: As an aside here to the endocrinologists who still think that type 2 diabetes is caused by insulin resistance, I like using the renal failure analogy, because the parathyroid gland doesn’t fail as a consequence of chronic work, nor does the anterior pituitary in Nelson’s syndrome fail. In both circumstances the endocrine glands undergo hypertrophy but do not fail. The same is true for most people with obesity: only a subset have islet failure and develop diabetes.
PULSATILE PARATHYROID HORMONE SECRETION
243
References Goldbeter A, Dupont G, Halloy J 2000 The frequency encoding of pulsatility. In: Mechanisms and signi¢cance of pulsatile hormone secretion. Wiley, Chichester (Novartis Found Symp 227) p 19^45 Lalli E, Sassone-Corsi P 1995 Thyroid-stimulating hormone (TSH)-directed induction of the CREM gene in the thyroid gland participates in the long-term desensitization of the TSH receptor. Proc Natl Acad Sci USA 92:9633^9637 Monaco L, Foulkes NS, Sassone-Corsi P 1995 Pituitary follicle-stimulating hormone (FSH) induces CREM gene expression in Sertoli cells: involvement in long-term desensitization of the FSH receptor. Proc Natl Acad Sci USA 92:10673^10677 O’Meara NM, Sturis J, Blackman JD et al 1993 Analytical problems in detecting rapid insulin pulses in normal humans. Am J Physiol 264:E231^E238
Signi¢cance of pulsatility in the HPA axis S. L. Lightman*, R. J. Windle*{, M. D. Julian*{, M. S. Harbuz*, N. Shanks*, S. A. Wood*{, Y. M. Kershaw*{ and C. D. Ingram{ *Dorothy Crowfoot Hodgkin Laboratories, University of Bristol, Division of Medicine, Bristol Royal In¢rmary, Bristol BS2 8HW and {Neuroendocrine Research Group, Department of Anatomy, University of Bristol, Bristol BS8 1TD, UK Abstract. A stress-free automated blood sampling system has been employed to demonstrate pulsatile hypothalamo^pituitary^adrenal (HPA) activity in the rat. In females, pulses of corticosterone secretion occur approximately once/hour throughout the 24 h cycle, with variation in pulse amplitude underlying a diurnal rhythm. Males show smaller pulses of secretion which become widely spaced during the early light phase nadir. Ageing does not a¡ect the occurrence of pulses but the diurnal variation is lost. Analysis of the relationship between the HPA response to an acute noise stress and its coincidence with the various phases of the pulse, suggests that pulsatile activity arises from alternating periods of activation and suppression. Responses to i.v. corticotropinreleasing factor are not a¡ected by pulse phase, indicating that this relationship is not generated at the pituitary^adrenal level. This phase relationship holds for all strains of rat except the hyperresponsive Fischer-344 in which an exaggerated stress response arises from a lack of phase-dependent suppression. Patterns of pulsatile activity are also modulated by neonatal programming or chronic HPA activation arising from adjuvantinduced arthritis, with consequent impact upon the response to acute stimuli. Thus, variations in the patterns of pulsatile activity are important determinants of both basal secretion and acute responses of the HPA axis. 2000 Mechanisms and biological signi¢cance of pulsatile hormone secretion. Wiley, Chichester (Novartis Foundation Symposium 227) p 244^260
The hypothalamo^pituitary^adrenal (HPA) axis provides a major neuroendocrine interface between the CNS and many peripheral systems within the body. The neural origin of this axis arises from the parvocellular neurons of the hypothalamic paraventricular nucleus (PVN) which synthesize either corticotropin-releasing factor (CRF) and/or arginine vasopressin (AVP), and release these peptides from their axon terminals in the median eminence into the pituitary portal blood supply. In turn, these hypophysiotropic factors act synergistically to e¡ect the release of adrenocorticotropic hormone (ACTH) from the corticotrophs of the pituitary gland. The pituitary acts to amplify the neuroendocrine signal and the ACTH released acts to increase the rate of synthesis and release of adrenal glucocorticoids which function as the e¡ector 244
PULSATILITY IN THE HPA AXIS
245
hormones. Although secretory activity can be modulated at both the pituitary and adrenal levels, it is the hypothalamic neuroendocrine signal which principally encodes information concerning the pattern of HPA activity through a complex interaction of neural inputs, gene expression and feedback control. The HPA axis is considered to have two principal modes of activity, basal and stimulated, although the distinction between the two is often hard to de¢ne. For the purposes of the current work, basal activity is that manifested as a result of both intrinsic and normal environmental signals, while stimulated activity is the response to imposed stimuli, most notably those considered to constitute a stress (physical, physiological or psychological). This chapter will consider the features of basal HPA activity, particularly focusing on the pulsatile pattern of activity and the conditions which lead to variations in this pattern, and will address the signi¢cance of the interactions between these pulses of secretory activity and the acute activation in response to a stressful stimulus. Ultradian and diurnal rhythms of HPA activity All endocrine systems show some degree of £uctuation in secretory activity which can be considered pulsatile. Yet while the importance of such £uctuations in the regulation of some hormonal systems is clearly recognized (growth hormone [GH] and luteinizing hormone [LH], for example), the majority of studies of the HPA axis have failed to address this characteristic of its activity. The principal reason for this is technical: the process of obtaining the frequent samples required for such study may generate a degree of physiological perturbation or anxiety in the subject which can itself modulate HPA activity. In order to adopt a methodology as free from stress as possible, the studies described here have been undertaken with rats placed on an automated sampling system adapted from that originally designed to study the dynamics of GH release (see Robinson 2000, this volume). This system allows the animal to remain in its home cage while sampling is achieved through an indwelling catheter using computer-driven equipment (Windle et al 1997, Fig. 1). Sampling at a frequency of once per hour reveals the classical diurnal cycle of corticosterone release with levels highest immediately prior to the onset of the active phase (Fig. 2). However, increasing the sample frequency shows a much more complex ultradian variation with corticosterone release occurring as a regular series of pulses throughout the 24 h cycle (Fig. 2). Each pulse is approximately 1 h in duration and pulses follow one after another. An important feature of this pattern is that the rate of decline of hormone levels within each pulse is in close agreement with the clearance rate for corticosterone of approximately 9 min in the rat (Windle et al 1998a). This rapid fall implies that the HPA axis is completely non-secretory and may in fact be actively inhibited for these periods. Thus, it seems likely that each pulse comprises several distinct phases: a secretory
246
LIGHTMAN ET AL
FIG. 1. Computer-driven sampling system for study of pulsatile HPA activity (adapted from Clark et al 1986). Animals [1] are catheterized in the jugular vein and the tubing exteriorized on the top of the head. The catheter is permanently attached and protected by a steel spring which is anchored to the skull. Each animal is individually caged and has complete freedom of movement, except for the area under the food hopper. Normally eight animals are sampled simultaneously using one system. Sampling is controlled by a computer programme and interface [2] which allows for complete £exibility in the timing, volume and speed of sampling, and also collects environmental data (temperature, noise and illumination). During a sample cycle the computer triggers a peristaltic pump [3] to draw a sample up through a liquid swivel [4] and three-way valve [5], and into a small reservoir [6]. The computer then switches the valve and reverses the pump so that a small blood sample (10^20 ml) is directed into the tubing leading towards a fraction collector [7]. The valve then switches back to the animal and returns the remainder of the blood plus 20^30 ml of heparinized saline [8] to maintain blood volume and keep the catheter patent. The valve ¢nally switches back towards the fraction collector and moves the blood sample along the tubing. At the appropriate time the collector arm moves to the appropriate tube position and the sample collected. This sample cycle is usually repeated every 10 min. For acute activation of the HPA axis a white noise generator may be activated [9] and the behavioural response monitored by video camera mounted over the cage.
phase when hormone levels rise, a non-secretory phase when hormone levels fall and a neutral or interpulse phase when the axis is quiescent (Fig. 5a). A clear diurnal di¡erence in pulse amplitude can be seen, with pulses being smaller in the morning than in the evening. In female rats pulse frequency does not alter over the 24 h
PULSATILITY IN THE HPA AXIS
247
FIG. 2. Ultradian and diurnal patterns of HPA activity in female rats. (Upper panel ) Plasma corticosterone levels measured in a group of animals sampled at a frequency of one sample per hour shows a characteristic diurnal rhythm, with a nadir at the time of lights on and a gradual increase to peak values which predict the onset of locomotor activity during the dark phase. Animals were maintained on a 14 h light :10 h dark illumination cycle, and dark phase being indicated by the shaded bar. (Lower panel ) Data from a single animal showing that increasing the sampling frequency to every 10 min reveals a more complex ultradian pattern which comprises episodes of secretion (pulses). The broken line indicates the pattern revealed by an hourly sampling rate. These pulses occur at a frequency of one every 60^90 min across the whole light^dark cycle, and a diurnal rhythm of secretion is generated as a result of variation in the amplitude of the pulses across the day. However, even during the period of peak secretory activity corticosterone levels fall too close to the detection limit in the interpulse periods, indicating a complete absence of adrenal secretion during these periods. Note that the large variance in the mean levels (upper panel) are due to the asynchronous nature of these pulses.
period, indicating that any pulse-initiating mechanism which generated basal HPA activity has a constant periodicity throughout the day. This suggests that the HPA axis is actively driven over the whole diurnal cycle including the nadir phase. A pulsatile pattern of HPA activity appears to be common to many species. In the human a clear pulsatile pattern of release has been seen in many studies designed to assess the contribution of episodic hormone release to basal cortisol levels in health and disease (Gallagher et al 1973, Iranmanesh et al 1989, 1993, Liu et al 1987, Schu«rmeyer et al 1996, Veldhuis et al 1989). However, as cortisol has a much longer biological half-life of around 90 min in the human, it is necessary to
248
LIGHTMAN ET AL
apply a deconvolution analysis to raw hormone measurement data in order to study the underlying pattern of HPA function. Similar pulsatile patterns of glucocorticoid release have also been seen in other rodent (Loudon et al 1994), equine (Cudd et al 1995) and simian (Carnes et al 1988, Tapp et al 1984) species. The mechanisms responsible for pulsatility within the HPA axis are unknown. Although corticosterone secretion may show some degree of modulation by sympathetic innervation at the level of the adrenal gland (Jasper & Engeland 1994), it is likely that the circhoral pulses of ACTH (Carnes et al 1989, Gambacciani et al 1987, Veldhuis et al 1990) are the major drive. These are most likely generated at the level of the hypothalamus and this is supported by reports of pulsatile release of CRF both in vivo (Ixart et al 1987, 1993, Liu et al 1994) and from isolated Macaque hypothalamic explants in vitro (Mershon et al 1992). Such pulses may be generated by inhibitory ultra-short feedback within the hypothalamus. However, since CRF itself is reported to have a stimulatory e¡ect on CRFproducing neurons in the PVN (Dunn & Berridge 1990) any such feedback would require mediation through an intermediate inhibitory factor, such as GABA or substance P. Variation in the characteristics of pulsatile HPA activity A number of factors appear to alter pulsatile HPA activity within the same species. One of the most striking of these is that related to gender. Although pulsatility is present in both genders, direct comparison of male and female Lewis rats has shown that circulating levels of corticosterone are much lower in the male than the female and that this is related to decreased pulse amplitude (Fig. 3a,b). Moreover, many studies in the male have shown that pulses are fewer in number than in the female and are interspersed by long interpulse intervals (Fig. 3a). This is particularly marked during the early light phase nadir. As a result, unlike the female, pulse frequency shows a distinct diurnal variation in the male. A relationship between genetic background and pulsatile HPA activity is also observable. Although all strains studied exhibit pulsatile corticosterone secretion (e.g. Wistar, Sprague^Dawley, Piebald^Viral^Glaxo, Lewis, Fischer-344) speci¢c di¡erences can be detected. This is particularly notable between the histocompatible Lewis and Fischer strains in which the Lewis rat displays lower levels of HPA activity, leading to lower circulating corticosterone levels and markedly smaller HPA responses to a wide range of stimuli. Rats of both strains show pulsatile secretion of corticosterone, but while the Lewis rat exhibited a clear diurnal variation in pulse amplitude (Fig. 3b), the Fischer rat showed high amplitude pulses throughout the light^dark cycle leading to a higher mean circulating corticosterone concentration (Fig. 3c). One potential mechanism for this di¡erence is that Fischer rats produce and release far more CRF than their
PULSATILITY IN THE HPA AXIS
249
FIG. 3. Gender and strain di¡erences in pulsatile HPA activity. Both male (a) and female (b) Lewis rats display pulsatile secretion of corticosterone with a diurnal variation in pulse amplitude, but the absolute levels are very di¡erent. Males show very low levels and have long inter-pulse periods during the nadir. Although a female Fischer rat displays a similar frequency of pulses (c), there is a complete absence of diurnal variation and high amplitude pulses occur across the whole light^dark cycle. The shaded bar indicates the dark period.
Lewis counterparts (Aksentijevich et al 1992, Sternberg et al 1989) such that, despite any diurnal variation, the concentration of CRF in the portal blood may always be su⁄cient to evoke a maximal response from the pituitary. However, to date, no direct comparison has measured that di¡erences in portal CRF concentrations. As well as the genetically determined di¡erences which exist between strains and gender, the HPA axis exhibits marked developmental changes. In the neonatal
250
LIGHTMAN ET AL
period the axis is subject to a degree of programming that establishes both basal and stress-induced HPA activity for the rest of the animal’s life. This has been extensively studied in the rat where it has been shown that a critical period during the ¢rst two postnatal weeks is involved. It has been shown that alterations in maternal behaviour during this critical period can lead to either a hypo- or hyperresponsive HPA axis in later life (Meaney et al 1994). Until recently it was not known to what extent this programming a¡ected the pulsatile HPA activity, but studies have now shown that extent of this programming. Neonatal treatment with a single i.p. injection of Gram-negative bacterial lipopolysaccharide (endotoxin) on days 3 and 5 post-partum was used as a model of early life infection. These animals were studied at 4 months of age and were found to exhibit a pulsatile pattern of hormone release that di¡ered quite markedly from saline-treated littermates. These animals showed increased pulse amplitude throughout the day. As well as the change in pulse amplitude, the frequency of pulses of hormone release was increased in these animals. In view of the increased CRF gene expression shown by neonatally endotoxin-treated animals, the mechanisms underlying the increased pulsatile release of corticosterone may share similarities with the Fischer rat. Furthermore, the elevated circulating glucocorticoid levels may have major consequences and indeed these animals become resistant to in£ammatory autoimmune disease in later life, which is also a feature of the Fischer rats (Shanks et al 1998). Studies of ageing animals have indicated that pulsatile HPA activity is maintained throughout life. In the study of male rats although overall circulating glucocorticoid levels were una¡ected by age, there was a loss of the diurnal rhythm of pulse amplitude due to a decrease in the evening peak levels and an increase in the morning peak levels. This loss of diurnal rhythm was seen in relatively young animals (being evident by 12 months of age), suggesting that the potentially deleterious e¡ects of ageing may start much earlier in the life cycle than previously thought. However, this loss of rhythmicity may not be simply a function of age as studies of 12 month old female Sprague^Dawley rats (Cai et al 1997) and 20^24 month old female Lewis rats, have shown no similar loss of the diurnal corticosterone rhythm. Female rats do, however, show a loss of the diurnal rhythm at a very speci¢c time during their reproductive life cycle. During pregnancy and lactation circulating corticosterone concentrations are elevated and there is a loss of the diurnal variation seen in non-pregnant, non-lactating animals (Atkinson & Waddell 1995, Lightman et al 1997, Stern et al 1973, Walker et al 1992). This is due to a £attening of circulating corticosterone levels with both a decrease in the evening peak concentrations and an increase in the morning nadir concentrations. Hormone release is still seen to occur in a pulsatile manner and these changes occur due to a modi¢cation of pulse amplitude (Windle et al 1998b). The
PULSATILITY IN THE HPA AXIS
251
function of this change in circulating corticosterone levels is unclear, but may be related to the metabolic changes associated with supporting the young and the role that glucocorticoids play in the process of lactation. Interestingly, lactating animals also show a complete loss of fast negative feedback inhibition of the HPA axis in response to exogenous injection of glucocorticoids, yet this does not disrupt the ability to generate pulses. This provides further evidence that the pulses of corticosterone secretion are not simply generated as a response to fast feedback from secreted corticosterone. E¡ect of pulsatile activity on response to acute stimuli In addition to being able to resolve the individual pulses which comprise basal HPA activity, the advantage of dynamic stress-free sampling is that responses to acute stimuli can be accurately characterized. The acute stimulus that we have employed is that of white noise, which has the advantage of being applied remotely and for a very speci¢ed time period without handling the animal (Windle et al 1997). During application of a 10 min noise stress, ACTH levels rise within 3 min of the onset and reach a peak towards the end of the stimulus. The peak corticosterone response is reached at 20 min after the stimulus onset, thereafter the levels decline rapidly to reach a nadir which is close to the limit of detection approximately 40 min later (Fig. 4). This transient response occurs within a similar time domain to a single pulse cycle and therefore the dynamic interaction between these two modes of activity can be investigated. As we have discussed, basal pulsatile secretion appears to depend upon cycles of activation and possible inhibition of the HPA axis. If this is the case then it follows that an acute stimulus which coincides with di¡erent phases of the endogenous pulse cycle will evoke very di¡erent responses, i.e. those that coincide with the neutral or activated phases when hormone levels are at a low level or rising are likely to evoke a larger response than those which coincide with the inhibited phase when hormone levels are falling (Fig. 5a). To test this hypothesis, we have sampled animals to resolve their pulsatile secretory pattern and then exposed them to acute noise. Examination of the pre-stress patterns enabled separation of those animals in which the noise coincided with the rising/interpulse phase of the cycle from those in which it coincided with the falling phase of the cycle, and the acute stress responses of these groups was compared. The outcome of this analysis is very clear. When an acute stimulus coincides with the rising phase of a basal pulse a large stimulus is evoked. However, during the falling (inhibited) phase of the pulse the acute response is markedly suppressed. This analysis has been performed on a number of di¡erent groups of rats and the distinction between the response obtained under the two phases of the pulse cycle is a consistent ¢nding for all cases, including female Lewis rats (Fig. 4a), female Sprague^Dawley rats (Windle
252
LIGHTMAN ET AL
FIG. 4. Strain di¡erences in the relationship between acute response to noise stress and pulse phase. Application of 10 min white noise stress (110 dB) evokes an acute increase in HPA activity which peaks 20 min after the onset before rapidly declining back to basal. If the onset of the noise stress coincides either with the rising phase of a basal pulse or with an interpulse period, there is a large response in both Lewis (a) and Fischer (b) rats, although the response in Fischer rats is signi¢cantly longer. If the noise coincides with the falling phase of a pulse then the Lewis rats show very much attenuated response which is not signi¢cantly above basal. In contrast Fischer rats display a response which does not di¡er from that in the other phase of the pulse. Values are the mean of six to eight female rats per group, and data have been normalized such that the mean value in the 60 min prior to the onset of the noise is equal to 100% (dotted line). See Fig. 5 for theoretical basis of this variation. (Data adapted from Windle et al 1998c).
et al 1998a), male Wistar rats of various ages (4^24 months), male Lewis rats, and male Piebald^Viral^Glaxo (PVG) rats. Only one exception to this result has been found and that is the hypersecretory Fischer rat (Windle et al 1998c). Noise stress evoked an HPA response in these animals which was both of greater magnitude
PULSATILITY IN THE HPA AXIS
253
FIG. 5. Theoretical relationship between pulse phase and acute HPA activation. (a) Individual pulses of corticosterone secretion can be considered to be generated by an activated phase (open bar) when hormone levels rise, followed by an inhibited phase (closed bar) when corticosterone levels fall at a similar rate to the plasma clearance rates. There may then be a neutral interpulse phase (shaded bar) when the axis is neither activated nor inhibited. An acute stress coinciding with the activated (‘rising’) phase will evoke a large response (dotted line), while if it coincides with the inhibited (‘falling’) phase then the response will be greatly attenuated (broken line). (b) In the case of the Lewis rat the di¡erence in the amount of CRF released by an acute stimulus may vary greatly between the two phases of the pulse, and this leads to very di¡erent pituitary^adrenal responses. (c) The Fischer rat releases much greater amounts of CRF in response to an acute stimulus and, even though this may di¡er between the di¡erent phases of the pulse, the level of CRF may be su⁄cient to evoke a maximal pituitary^adrenal response, irrespective of pulse phase.
and longer duration than that found in other strains and, furthermore, occurred irrespective of the underlying phase of the endogenous pulse cycle (Fig. 4b). A likely mechanism underlying this relationship between pulse phase and the response to acute activation involves the variation in hypothalamic production of CRF. Comparison of the magnitude of the corticosterone response to injection of exogenous CRF has shown no e¡ect of pulse phase on the response in either Lewis or Fischer rats (Windle et al 1998c), thereby excluding the pituitary or adrenal as the sites of interaction between pulse phase and responsiveness to
254
LIGHTMAN ET AL
noise stress. Instead it is likely that the amount of CRF released in response to the acute noise stimulus is considerably reduced when it coincides with the falling phase of the pulse, thus leading to an attenuated ACTH signal compared to when the axis is in the active phase of the pulse (Fig. 5b). The ability of Fischer rats to respond to stimuli irrespective of the phase of a pulse is unlikely to be due simply to a genetic loss of an inhibitory component, since if this were the case pulses would not be observed. It is more likely that the increased production of CRF in these animals means that, even when the amount of CRF released is reduced by its coincidence with the inhibitory (falling) phase of the pulse, the concentration is su⁄cient to evoke a maximal pituitary response. Thus this same mechanism can explain the lack of diurnal variation in pulse amplitude and the loss of phase relationship of the acute stress response. In view of the robust relationship between pulse phase and the magnitude of the acute response to noise stress, it follows that changes to the pattern of pulsatile secretion will have an impact on the acute response. This has been demonstrated in a study of the change in HPA activity during development of adjuvant-induced arthritis in the male PVG rat. In this model the elevated circulating glucocorticoid concentrations which accompany the onset of arthritic disease are due to an increase in the number of pulses of corticosterone release, without any change in the magnitude of these pulses. This increase in pulse number and the subsequent loss of prolonged inter-pulse periods, particularly during the nadir, is responsible for the loss of diurnal rhythmicity. Despite the raised basal secretion, mean HPA responses to acute stress are diminished in experimentally arthritic animals. Analysis shows that the phase relationship of the acute stress response is identical between control and arthritic animals. However, the increased number of pulses, the subsequent loss of inter-pulse interval (during which animals can respond to stress), and the relative shortening of the active/rising phases of HPA activity, means that acute stimuli are far more likely to coincide with the inhibited phases of the pulse cycle. Thus, the decrease in the mean response of the arthritic animals was due to the increased proportion of stimuli coinciding with these inactive phases. Conclusions The use of frequent sampling demonstrates that the HPA axis displays pulsatile activity and that the characteristics of this pattern of activity are subject to considerable genetic, physiological and pathological variation. This variation has major e¡ects both on the absolute levels of plasma glucocorticoids and on the diurnal variation. Each secretory episode (pulse) appears to be generated by periods of activity and inhibition, and the characteristics of individual pulses are such that the peak corticosterone levels may be very high but then fall rapidly to
PULSATILITY IN THE HPA AXIS
255
undetectable values. Thus, tissues of the body would be continuously exposed to very widely £uctuating corticosteroid concentrations, a fact often overlooked when considering the e¡ects of, and mechanisms regulating HPA function. Glucocorticoids have a wide variety of biological actions ranging from e¡ects on intermediate metabolism, which are mostly catabolic and involve gluconeogenesis from proteins and fatty acids, and e¡ects on the immune system, especially an inhibition of cell-mediated immunity, to alterations in mood and the state of arousal. Glucocorticoid response elements are also a feature of the promoter region of many genes. It is currently unknown whether the widely £uctuating levels of corticosterone which occur as a result of pulsatile secretion leads to similar £uctuations in these functions, but the potential for fast actions mediated through membrane receptors does raise the distinct possibility that pulsatile activity may be translated in dynamic functional e¡ects. Acknowledgements The authors are grateful to the Wellcome Trust, Neuroendocrinology Charitable Trust and United Bristol Healthcare Trust for their ¢nancial support for the work described here.
References Aksentijevich S, Whit¢eld HJ Jr, Young WS III et al 1992 Arthritis-susceptible Lewis rats fail to emerge from the stress hyporesponsive period. Brain Res Dev Brain Res 65:115^118 Atkinson HC, Waddell BJ 1995 The hypothalamic^pituitary^adrenal axis in rat pregnancy and lactation: circadian variation and interrelationship of plasma adrenocorticotropin and corticosterone. Endocrinology 136:512^520 Cai A, Scarbrough K, Hinkle DA, Wise PM 1997 Fetal grafts containng suprachiasmatic nuclei restore the diurnal rhythm of CRH and POMC mRNA in aging rats. Am J Physiol 273:R1764^R1770 Carnes M, Kalin NH, Lent SJ, Barksdale CM, Brown¢eld MS 1988 Pulsatile ACTH secretion: variations with the time of day and relationship to cortisol. Peptides 9:325^331 Carnes M, Lent S, Feyzi J, Hazel D 1989 Plasma adrenocorticotropic hormone in the rat demonstrates three di¡erent rhythms within 24 h. Neuroendocrinology 50:17^25 Clark RG, Chambers G, Lewin J, Robinson ICAF 1986 Automated repetitive microsampling of blood: growth hormone secretion in conscious male rats. J Endocrinol 111:27^35 Cudd TA, LeBlanc M, Silver M et al 1995 Ontogeny and ultradian rhythms of adrenocorticotropin and cortisol in the late-gestation fetal horse. J Endocrinol 144:271^283 Dunn AJ, Berridge CW 1990 Physiological and behavioral responses to corticotropin-releasing factor administration: is CRF a mediator of anxiety or stress responses? Brain Res Brain Res Rev 15:71^100 Gallagher TF, Yoshida K, Ro¡wang HD, Fukushima DK, Weitzman ED, Hellman L 1973 ACTH and cortisol secretory patterns in man. J Clin Endocrinol Metab 36:1058^1068 Gambacciani M, Liu JH, Swartz WH, Tueros VS, Rasmussen DD, Yen SSC 1987 Intrinsic pulsatility of ACTH release from the human pituitary in vitro. Clin Endocrinol 26:557^563 Iranmanesh A, Lizarralde G, Johnson ML, Veldhuis JD 1989 Circadian, ultradian and episodic release of beta-endorphin in men and its temporal coupling with cortisol. J Clin Endocrinol Metab 68:1019^1026
256
LIGHTMAN ET AL
Iranmanesh A, Lizarralde G, Veldhuis JD 1993 Co-ordinate activation of the corticotropic axis by insulin-induced hypoglycemia: simultaneous estimates of beta-endorphin, adrenocorticotropin and cortisol secretion and disappearance in normal men. Acta Endocrinol (Copenh) 128:521^528 Ixart G, Barbanel G, Conte-Devolx B, Grino M, Oliver C, Assenmacher I 1987 Evidence for basal and stress-induced release of corticotropin releasing factor in the push^pull cannulated median eminence of conscious free-moving rats. Neurosci Lett 74:85^89 Ixart G, Siaud P, Barbanel G, Mekaouche M, Givalois L, Assenmacher I 1993 Circadian variations in the amplitude of corticotropin-releasing hormone 41 (CRH41) episodic release measured in vivo in male rats: correlations with diurnal £uctuations in hypothalamic and median eminence CRF41 contents. J Biol Rhythms 8:297^309 Jasper MS, Engeland WC 1994 Splanchnic neural activity modulates ultradian and circadian rhythms in adrenocortical secretion in awake rats. Neuroendocrinology 59:97^109 Lightman SL, Windle RJ, da Costa APC, Shanks N, Ingram CD 1997 Lactation: a physiological model of stress hyporesponsiveness of the neuroendocrine system. In: Levy A, Grauer E, BenNathan D, de Kloet ER (eds) New frontiers in stress research. Modulation of brain function. Harwood Academic Press, Amsterdam, p 59^71 Liu JH, Kazer RR, Rasmussen DD 1987 Characterization of the twenty-four hour secretion patterns of adrenocorticotropin and cortisol in normal women and patients with Cushing’s disease. J Clin Endocrinol Metab 64:1027^1035 Liu JP, Clarke IJ, Funder JW, Engler D 1994 Studies of the secretion of corticotropin-releasing factor and arginine vasopressin into the hypophyseal-portal circulation of the conscious sheep. II. The central noradrenergic and neuropeptide Y pathways cause immediate and prolonged hypothalamus^pituitary^adrenal activation. Potential involvement in the pseudo-Cushing’s syndrome of endogenous depression and anorexia nervosa. J Clin Invest 93:1439^1450 Loudon ASI, Wayne NL, Krieg R, Iranmanesh A, Veldhuis JD, Menaker M 1994 Ultradian endocrine rhythms are altered by a circadian mutation in the Syrian hamster. Endocrinology 135:712^718 Meaney MJ, Tannenbaum B, Francis D et al 1994 Early environmental programming hypothalamic^pituitary^adrenal responses to stress. Semin Neurosci 6:247^259 Mershon JL, Sehlhorst CS, Rebar RW, Liu JH 1992 Evidence of a corticotropin-releasing hormone pulse generator in the macaque hypothalamus. Endocrinology 130:2991^2996 Robinson ICAF 2000 Control of growth hormone (GH) release by GH secretagogues. In: Mechanisms and biological sign¢cance of pulsatile hormone secretion. Wiley, Chichester (Novartis Found Symp 227) p 206^224 Schu«rmeyer TH, Brademann G, von zur Mu«hlen A 1996 E¡ect of fen£uramine on episodic ACTH and cortisol secretion. Clin Endocrinol (Oxf) 45:39^45 Shanks N, Perks P, Harbuz MS, Jessop DS, Lightman SL 1998 Neonatal handling and endotoxin exposure di¡erentially alter neuroendocrine^immune interactions and adjuvantinduced arthritis. Neuroimmunomodulation 5:46 Stern JM, Goldman L, Levine S 1973 Pituitary-adrenal responsiveness during lactation in rats. Neuroendocrinology 12:179^191 Sternberg EM, Young WS III, Bernardini R et al 1989 A central nervous system defect in biosynthesis of corticotropin-releasing hormone is associated with susceptibility to streptococcal cell wall-induced arthritis in Lewis rats. Proc Natl Acad Sci USA 86:4771^4775 Tapp WN, Holaday JW, Natelson BH 1984 Ultradian glucocorticoid rhythms in monkeys and rats continue during stress. Am J Physiol 247:R866^R871 Veldhuis JD, Iranmanesh A, Lizarralde G, Johnson ML 1989 Amplitude modulation of a burstlike mode of cortisol secretion subserves the circadian glucocorticoid rhythm. Am J Physiol 257:E6^E14
PULSATILITY IN THE HPA AXIS
257
Veldhuis JD, Iranmanesh A, Johnson ML, Lizarralde G 1990 Amplitude, but not frequency, modulation of adrenocorticotropin secretory bursts gives rise to the nyctohemeral rhythm of the corticotropic axis in man. J Clin Endocrinol Metab 71:452^463 Walker CD, Lightman SL, Steele MK, Dallman MF 1992 Suckling is a persistent stimulus to the adrenocorticol system of the rat. Endocrinology 130:115^125 Windle RJ, Wood S, Shanks N et al 1997 Endocrine and behavioural responses to noise stress: comparison of virgin and lactating female rats during non-disrupted maternal activity. J Neuroendocrinol 9:407^414 Windle RJ, Wood SA, Shanks N, Lightman SL, Ingram CD 1998a Ultradian rhythm of basal corticosterone release in the female rat: dynamic interaction with the response to acute stress. Endocrinology 139:443^450 Windle RJ, Wood SA, Kershaw YM, Lightman SL, Ingram CD 1998b Alterations in steroid negative feedback of the HPA axis during lactation and weaning in the rat. J Endocrinol (suppl) 156:C49 Windle RJ, Wood SA, Lightman SL, Ingram CD 1998c The pulsatile characteristics of hypothalamo-pituitary-adrenal activity in female Lewis and Fischer 344 rats and its relationship to di¡erential stress responses. Endocrinology 139:4044^4052
DISCUSSION Robinson: When we looked at GH-releasing proteins (RPs) that are non-speci¢c for GH (they also release ACTH and cortisol), we found basically the same sort of thing as you did: depending on when in the cycle of pulsatility you gave the secretagogue. Have you tried to model that by giving one of the two secretagogues for ACTH, for CRF or AVP? Instead of giving a stress, can you tie down whether it’s a pituitary refractoriness or an adrenal refractoriness? Lightman: If we give pulses of CRH you get the same response whenever it arrives in the endogenous cycle. This suggests that what we’re looking at must be above the pituitary. Robinson: But what happens if you ¢rst give a pulse of CRH and then give a white noise? Lightman: We haven’t done that. Robinson: Can you induce refractoriness and then test it with a stress, and show that you don’t get the response subsequent to the CRH pulse? Lightman: I don’t know. One of the things that we would like to do is repeat stresses. You adapt to stress anyway, but we’d like to see whether or not there are particular times following a stress that if you repeat it, it is particularly turned o¡. Clarke: Is the idea that you have a refractoriness based exclusively on the cortisol data? Lightman: Mainly, although we have some ACTH data on this, too. Clarke: To what extent, then, could you explain the results by looking at the releasable pool of cortisol in the adrenal? Perhaps one way of looking at that is to take the adrenal gland out of these animals across the response phase and see how they then respond in vitro.
258
DISCUSSION
Lightman: One of the nice things about the noise stress we are using is that it is a relatively minor stress. If we give a much stronger stress, we can get further release and overcome the refractoriness. There is plenty of corticosterone available to be released, and we are not having a signi¢cant e¡ect on the readily releasable pool. Veldhuis: But the adrenalectomy model would be nice, wherein you go back and measure ACTH to rule out fast cortisol feedback as the mechanism. Lightman: This feedback would have to be superfast. It is something we will have to investigate, but I don’t think it is a likely explanation. Licinio: Following on from that, in our own studies we’ve actually done the opposite. We have used maximal stressors such as immobilization, which for rats is a severe stress. In this case we might have missed subtle di¡erences because of using the maximal stress. Thus future studies should use a variety of types of stress at di¡ering intensities. Could you comment on your elegant studies comparing the CRH and AVP roles in the regulation of the stress response? Lightman: In the animals with adjuvant arthritis with chronic activation of the HPA axis, the levels of CRH mRNA actually fall to very low levels while they develop high levels of vasopressin mRNA. We also have similar ¢ndings in animals that undergo repeated (daily) restraint stress for two weeks. When we give these animals a further episode of restraint, their vasopressin goes up markedly even though their CRH mRNA and heteronuclear RNA no longer responds to the homotypic stressor. In the chronically stressed animals we have had a switch-over from a CRH-predominant to an AVP-predominant regulatory mechanism on the pituitary. Whether or not the ratio of CRH to AVP is important in governing the rate of pulses is something that we need to test. It may well be that the di¡erent ratio is one of the things that is actually a¡ecting the pulse rate in these animals. Sassone-Corsi: I was wondering whether in the Fischer rats, where you can induce a very nice peak in response to CRH at any time with a stress signal, the ¢rst peak would also induce the refractory period after a stress? Lightman: We haven’t looked at that. Butler: Why rats rather than humans? What happens in humans? Lightman: Rats are much easier to look at than humans, simply because humans have so much CBG that the pulsatility of cortisol in humans is di⁄cult to evaluate and you need to do deconvolution analysis. Also, stressing humans is phenomenally di⁄cult and ethically questionable. Furthermore, people adapt to psychological tests, so intra-subject controlled stress testing in humans is incredibly di⁄cult. Licinio: I work mostly with humans, and the problem we ¢nd is that the stressors for each human are di¡erent. Exercise is one of the few uniform paradigms for human stress.
PULSATILITY IN THE HPA AXIS
259
Butler: A related question is therefore to what extent is this important in humans? Is this an important physiological phenomenon that is just di⁄cult to measure in humans, or is important in rats and not in humans? Lightman: That is unanswerable at the moment. If the hypothesis is true that one of the most important things in stress responsiveness is how you turn o¡ the stress response, then I suspect it could be extremely important for the human situation. Studies on chronic stress in humans suggest that the nadir levels of cortisol tend to be higher. If we can think of there being a system which should be turning o¡ cortisol, which isn’t happening, and if we can understand more about the physiology of that, it could have major therapeutic potential. To date, practically all the human stress work that has been carried out has been aimed at what turns on a stress response. Licinio: Your work on the early exposure to stress is very relevant to humans. There is some interesting work done by Lerer and colleagues showing that if humans undergo early parental loss, particularly before age nine, there is increased likelihood of developing major depression in adult life (Agid et al 1999). Herbison: I thought the sex di¡erence was striking. What is known about its origin? Is it a pituitary or hypothalamic phenomenon? In terms of the circadian changes, is there any evidence for an interaction between gonadal steroids and circadian input? For example, is this di¡erent at di¡erent stages of the oestrus cycle? Lightman: These are all things that we are beginning to investigate. The sex di¡erences are indeed huge. Levi: You showed us that there was an e¡ect of age on the circadian rhythm of corticosteroid secretion in rats. Is this age e¡ect strain-dependent? And could the fact that you see no circadian rhythm of corticosterone in Fischer rats mean that there is an early ageing process in this strain? Lightman: The problem with ageing research for us is that it is phenomenally expensive to get rats that are 24 months old either to have looked after them in our animal house, or to buy them in. I would like to know the answer to your question, but at present these studies are simply too expensive for us to do. Le¤ vi: Does the Fischer rat have a shorter lifespan? Lightman: No, I don’t think they do. Le¤ vi: Is there any relationship between circadian or pulsatile rhythmicity and survival? Lightman: The way to study this would be to compare the Fischer and the Lewis rats. Since they are histocompatible and have such di¡erent pulsatile secretion, they would be very good controls for each other, but unfortunately I’m not aware of anybody who has done this. There is however an important caveat when we talk about survival, since this may depend on environmental conditions. Certainly, the Fischer rats don’t get all the autoimmune diseases that Lewis rats get, but maybe
260
DISCUSSION
they are at risk from other diseases related to glucocorticoid excess. I don’t think that changes in the HPA pulse generation will simply be either good or bad for you: it’s going to be good for some things, bad for other things. Robinson: You are advancing the concept that turning o¡ the ACTH system is what is important. If you don’t think there’s something intrinsic about the tissues that are becoming refractory, because you can test those at di¡erent times and still get responses, do you think that there is a somatostatin-like system that is yet to be discovered for the HPA axis? Lightman: I think it is more likely to be neurally induced. Rather than the somatostatin type of system, I would envisage a localized GABAergic or substance P system which is turning o¡ the CRH cells in the hypothalamus. Reference Agid O, Shapira B, Zislin J et al 1999 Environment and vulnerability to major psychiatric illness: a case control study of early parental loss in major depression, bipolar disorder and schixophrenia. Mol Psychiatry 4:163^172
Closing remarks Johannes D. Veldhuis Division of Endocrinology and Metabolism, Department of Internal Medicine General Clinical Research Center, Center for Biomathematical Technology, University of Virginia, Charlottesville, VA 22908, USA
There are multiple pivotal issues in pulsatility that should entertain our interests and intellects for some time. Some of these challenges are stated as speci¢c questions at this meeting. Collectively, I should hope that an alacrity to grapple with the unknown will inspire further multidisciplinary interchange particularly as new insights into signalling unfold. Discussions here highlight the complexity and richness of genomic, biochemical and cellular control mechanisms driven by a fairly simple primary signal; i.e. a frequency or amplitude-encoded pulse. Indeed, the colloquium unveils a multitude of mysteries and queries. For example, what mechanisms guide signal convergence and divergence within a cell? How does a single stimulus evoke such divergence in messenger responses within a cell? Indeed, is it clear that the same individual cell actually generates multiple signals? When is intercellular signal diversity required? How is a cellular signal turned o¡? What is the root role of apparent redundancies in signal transmission? When is signal complementarity critical, and when is it super£uous? How is signal information retained across frequently bifurcating (or multiply ramifying) pathways, and across disparate time-scales? What are the more critical biological linkage strategies that coalesce organismic responses across molecular, cellular, glandular and system scales? How can the experimentalist probe these connectionistic issues, which are confounded by signi¢cant structural, functional, time-delayed, dose-responsive, baseline, and stochastic variations within and between organisms? What new and central therapeutic implications will emerge from this growing knowledge base; e.g. in oncology, stress adaptation, diabetes, reproduction, ageing, etc.? A tacit assumption is that because normal physiology manifests a particular phenomenology, its disruption is pathological. Further study of pulse-signalling systems may disarm this assumption, if certain time-evolving events are epiphenomena of interface structure or represent non-critical redundancies. Genetic knockout experiments should continue to aid in identifying the latter consideration, and more subtle insights into molecular structure^function relationships should enlighten the former question. 261
262
VELDHUIS
Most importantly, I speak for each scienti¢c participant and readers in acknowledging the enormous biomedical contribution of the Novartis Foundation in orchestrating and sponsoring this international research interchange.
Index of contributors Non-participating co-authors are indicated by asterisks. Entries in bold indicate papers; other entries refer to discussion contributions.
I
B
*Ingram, C. D. 244
Brabant, G. 16, 75, 105, 114, 115, 116, 240 Brown, D. 38, 80, 98, 115, 224 Butler, P. 190, 199, 200, 201, 202, 203, 204, 240, 242, 258, 259
J *Julian, M. D. 244
C K
*Cermakian, N. 5 Clarke, I. J. 15, 38, 39, 79, 116, 117, 157, 185, 202, 222, 223, 257 Copinschi, G. 58, 77, 78, 143, 157, 158, 159, 160, 161, 223
*Kershaw, Y. M. 244 Kjems, L. L. 41, 158, 161 L
D
Leng, G. 18, 40 *Leproult, R. 143 Le¤ vi, F. 99, 119, 136, 137, 138, 139, 140, 141, 158, 160, 259 Licinio, J. 58, 101, 138, 158, 159, 201, 203, 204, 221, 241, 258, 259 Lightman, S. L. 43, 44, 79, 139, 161, 204, 244, 257, 258, 259, 260
De Meyts, P. 16, 43, 44, 46, 57, 58, 59, 60, 80, 200 *Dupont, G. 19 E Ede¤ n, S. 79, 141, 203 F *Foulkes, N. S. 5
M Marshall, J. C. 16, 41, 42, 44, 57, 102, 103, 158, 159, 188, 200, 222, 241 Matthews, D .R. 15, 17, 43, 44, 59, 60, 76, 97, 138, 141, 157, 186, 187, 201, 204, 221
G Goldbeter, A. 17, 19, 36, 37, 38, 39, 40, 41, 42, 43, 44, 104, 140, 157, 201, 203, 204, 222, 240
P H
Pincus, S. M. 37, 82, 96, 97, 98, 99, 100, 101, 102, 103, 104, 115, 116, 117, 160, 161, 187, 188, 224 *Prank, K. 105
*Halloy, J. 19 *Harbuz, M. S. 244 Herbison, A. E. 44, 79, 222, 259 263
264
R Robinson, I. C. A. F. 39, 40, 42, 59, 60, 78, 79, 99, 139, 159, 160, 187, 199, 202, 206, 220, 221, 222, 223, 224, 241, 257, 260
INDEX OF CONTRIBUTORS
V *Van Cauter, E. 143 Veldhuis, J. D. 1, 15, 36, 41, 57, 59, 75, 76, 80, 96, 101, 103, 114, 115, 117, 136, 158, 159, 160, 161, 163, 185, 186, 187, 188, 203, 220, 221, 223, 224, 239, 241, 258, 261
S Sassone-Corsi, P. 5, 15, 16, 17, 18, 40, 44, 45, 76, 77, 99, 100, 137, 158, 204, 221, 239, 240, 241, 258 Schaefer, F. 139, 225, 240, 241, 242 *Shanks, N. 244 *Shymko, R. M. 46 *Spiegel, K. 143
W Waxman, D. J. 16, 37, 38, 61, 75, 76, 77, 78, 79, 80, 137, 157, 202 *Whitmore, D. 5 *Windle, R. J. 244 *Wood, S. A. 244 Wu, F. C. W. 43, 78, 79, 100, 101
Subject index blood hormone concentrations 3 bone disease, PTH secretion in 235^236 breast cancer 129, 131, 133, 158
A acromegaly 106, 107^109, 110, 112, 113 ACTH 2, 85, 89, 91, 103, 114, 161, 168^170, 179, 244, 251, 254, 257, 260 ACTH^cortisol axis 175^179 ACTH^cortisol synchrony 179 actigraphy bracelet 130 actigraphy records 132 adenylate cyclase 236 adrenal glucocorticoids 244 b-adrenergic receptor 40, 57 adrenocorticotropic hormone see ACTH ageing 160, 185 and neurohormone release 173^181 in pulsatile neuroendocrine systems 181^183 of circadian clock 151, 159 pulsatile hormone release in 163^189 research 259 algorithmic complexity 109 androgen 166 animal models 137, 138 anticancer drugs, chronopharmacology of 120^122 approximate entropy (ApEn) 3, 83^104, 106^107, 109, 111, 114^116, 174 endocrinological applications 86^88 in correlation and spectral analyses 92^94 noise sensitivity 98 arabinofuranosylcytosine 122 arginine vasopressin (AVP) 167, 244, 257 ARIMA 92, 97 arti¢cial neural networks 106, 109, 111 ATF 45 ATP/ADP ratio 201 autocorrelation function 92 autonomous systems 204
C Ca2+ 39 concentration vs. time pro¢le 228 membrane receptor 236 oscillations 26^30, 32, 33, 40 signalling 41 spikes 41 frequency encoding of 29 signal-induced 28 waves 30 Ca2+-activated kinase 32 Ca2+-calmodulin activated protein kinase (CaM kinase) 29 Ca2+-calmodulin kinase 36 cAMP 6^7, 13, 16, 20^24, 31^33, 36^39, 41^43, 241, 242 in Dictyostelium discoideum, pulsatile signals of 20^24 production 236 pulsatile signals 240 cAMP-responsive transcription factors 8^9 cancer cells 140, 141 chronobiology 136 chronopharmacology 31 cancer patients 138 circadian rhythms in 119^120 individual rhythms 129^133 catecholamines 99 cDNA 7 chromatin 241 chronobiology, cancer cells 136 chronopharmacokinetics 139 chronopharmacology 137 anticancer drugs 120^123 cancer cells 31 mechanisms and rest^activity cycle 124 circadian clock 143^144 ageing of 151, 159
B basal cell carcinoma 128 binding proteins (BPs) 220^221 biological endpoints 55 265
266
circadian pacemaker 151, 154 misalignment 152^153 circadian rhythms 30^31, 119^120 cellular determinants 123 pathophysiology 143^162 study of 145 circadian system 121 assessment 129^130 clinical relevance 133 during cancer processes 128^133 cisplatin 122, 124 Clock 15, 18 CLOCK 144 clustering 40 CNS^hypothalamic coordination 175 CNS regulatory centres 177 colorectal cancer 126^127, 132, 133, 138, 140 corticosteroid binding globulin (CBG) 161 corticosterone secretion 248 corticotropin-releasing factor see CRF cortisol 129, 139, 146^149, 151, 152, 158^160, 247, 258 CRE 11 CREB 8, 12, 13, 17, 45, 240 CREM 8, 10, 11, 15, 16, 17, 45 CRF 244, 248^250, 253^254, 257 CRF/ACTH system 102 CRH 167, 257^260 cross-approximate entropy (ApEn) 83, 88^95, 102, 177, 179, 187 in correlation and spectral analyses 92^94 cross-correlation 97 Cushing’s disease 89, 91, 114, 115, 147, 148, 187 CWSV-1 cells 66, 69 cyanobacteria 31 CYP2C11 62 CYP2C12 62 cytochrome P450 (CYP) 62, 138 D day length 157 N-dechloroethylation 138 dehydropyrimidine dehydrogenase (DPD) 123, 139 depression 139^140, 148, 160 diabetes mellitus 179, 197, 200, 203, 204, 242 insulin secretion in 195^196 diazoxide 203
SUBJECT INDEX
Dictyostelium discoideum 36^42, 240 pulsatile signals of cAMP in 20^24 di¡erence, depression 259 dissociation rate constants 50 diurnal hormonal pro¢les 145 sleep loss e¡ects on 151^152 diurnal rhythms, hypothalamo^pituitary^ adrenal (HPA) axis 245^248 DNA 8, 44, 64 DNA lesions 137 DNA synthesis 122, 128, 136 docetaxel 122 docking and undocking 202 dose-responsive linkages 165^167 doxorubicin 122, 124 Drosophila 17, 144, 240 drug delivery patterns 31 drug tolerability 31, 120, 122, 127 E EEG 98 endocrine glands 164 endogenous circadian clock 157 endogenous rhythm 157 endoplasmic reticulum 26 EORTC QLQ-30 questionnaire 133 extraglandular inputs 164 F feedback systems 1, 101, 103, 168 feedforward-feedback systems 169^171, 186 feedforward neural network 107 FFT programs 98 5-£uorouracil (5-FU) 122, 123, 126^127, 139 £ight attendants 158 follicle-stimulating hormone see FSH food composition 159 food schedule 159^160 Fourier transforms 97 Fra-2 mRNA 10 frequency encoding of Ca2+ spikes 29 of pulsatile signals 19^45 FSH 25, 39, 41, 101, 164, 242 G gap junctions 30 gene expression pulses 240 GH see growth hormone
SUBJECT INDEX
GH^IGF-I axis 179 GHBP (GH binding protein) 215, 217 GHNF (GH-regulated nuclear factor) 72 GHR 61, 62, 65, 67, 70, 213, 215 GHR^JAK2 complex 66, 69, 70, 75 GHRF (GH-releasing factor) 222 GHRH 179, 187, 202, 207^210, 215^216, 222 GHRP 221^223, 257 glandular secretory variability 173 glucocorticoids 244, 251, 255 glucose ingestion 194 glucose release 201, 240 glucose tolerance 203 glycogenolysis 240 glycolytic enzymes 201^202 GnRH 24, 25, 41, 57, 86, 164, 165, 168, 186, 188, 224 GnRH/LH system 3, 39, 43, 164^173 gonadotropin-releasing hormone see GnRH GRF 224 group chronotherapy 120^127 clinical trials 124^127 growth factors/mitogens 48 growth hormone (GH) 24, 39, 59, 78^80, 85, 103, 106, 117, 145, 149^150, 160, 181, 187 concentration time-series 112 feedback 210, 211 gender di¡erences 86^88 pulsatile release of 61^64, 107^113, 206^224 pulse-activated STAT5 signalling 61^81 pulse amplitude 108 responses 207, 212 growth hormone binding protein see GHBP growth hormone receptor see GHR growth hormone releasing hormone see GHRH growth hormone secretagogues (GHS) 206^224 H hCG 57 hidden layers 106 homeostatic component 144 hormonal time series 105 hormone dissociation rate 55 hormone receptors 57 hormone trapping 220 hospital anxiety^depression scale 140
267
hot £ushes 99 hydroxyindole-O-methyltransferase (HIOMT) 6, 10 4-hydroxylation 138 hypercalcaemia 232^234, 235 hyperparathyroidism 235^236 hypocalcaemia 229^231, 235 hypophysiotropic factors 244 hypothalamic^adrenal axis 161 hypothalamic neural regulatory centres 164 hypothalamic^pituitary junction 2 hypothalamo^pituitary^adrenal (HPA) axis 244^260 diurnal rhythms 245^248 pulsatile activity 248^254 ultradian rhythms 245^248 I I-kB 241 ICER (inducible cAMP early repressor) 8^13, 17, 242 and rhythmic melatonin synthesis 9^12 mRNA 9 ifosfamide 137^138 IGF-I 48, 58, 59, 221 IGF-I receptors 49 IGF-II 48 IID 104 insulin 242 and insulin analogues 48^51 as growth factor/mitogen 48 extracted by liver 200 short-acting and longer-acting 58 insulin concentration 191, 199 insulin-like growth factor see IGF-I insulin mitogenic versus metabolic signalling 46^60 insulin^receptor complex 60 insulin release 179^181, 182, 190^205 in diabetes mellitus 195^196 pacemaker 191^192 pulsatile 192^193, 196^197, 201 regulation 193^195, 197 insulin resistance 59^60, 158, 159, 161, 200, 204 intracellular Ca2+ pulses 26^30 intracellular signal control 3 intraglandular level 164 IRS-1^3 activation 50 islets 203
268
SUBJECT INDEX
J
N
JAK2 67^68 JAK2 kinase 62, 67 JAK2 tyrosine kinase 65 JAK2^STAT5b pathway suppression 69^70 jet lag 153, 158
Neurospora 17, 157 NF-kB 241 nocturnal penile tumescence (NPT) 102, 175 noise stress 251^252, 258 non-circadian chronotherapy 140 non-Hodgkin’s lymphomas 128 non-photic stimuli, phase-shifting e¡ects 153^155 non-REM sleep 144 nuclear receptors 44 nutrition timing 159^160
K Kolmogorov^Sinai (K^S) entropy 100 L leucovorin (LV) 125, 127 LH^FSH co-release 174^179 LH^prolactin bihormonal synchrony of release 175 LH^testosterone coupling 90^91, 173 ligand-binding kinetics 55 ligand-induced dimerization 59 ligand^receptor complex 58 light^dark cycle 144, 145, 157, 158, 159 lipolysis dose^response curve 200 liver gene expression, sexual dimorphism of 61^81 liver STAT5b 64^66 liver-derived cell culture model 66 logical switching theory 53 logical transition rules 54 long-term desensitization 241^242 low-density lipoprotein (LDL) 168 luteinizing hormone (LH) 25, 39, 41, 43, 57, 85, 102, 103, 164, 165, 168, 173, 185, 186, 245 see also LH Lyapunov spectrum 100 lymph nodes 128 M mammary carcinoma 128^129 MAP kinase pathway 17, 240 melatonin 139, 145^147, 151 melatonin synthesis 5^18 ICER and 9^12 linking the clock to 8^9 pathway 7 melphalan 122 menopause 99 mitogenic/metabolic potency ratio of insulin analogues 51 motor activity rhythms 139 mucositis 126
O osteoporosis 235 ovarian cancer 124, 128 oxaliplatin 125, 139 P p53 expression 137 P450 enzymes 138 pancreas, electrophysiology 204 pancreatic carcinomas 204 parathyroid hormone see PTH paraventricular nucleus (PVN) 244 PDGF (platelet-derived growth factor)b receptor 48 pharmacodynamics 139 phase response curve 153^154 phosphatase inducibility 40^41 phospholipase C 236 phosphorylation 44, 240, 241 phosphorylation^dephosphorylation 32, 40 phosphotyrosine phosphatase inhibitor 74 photic stimuli, phase-shifting e¡ects 153^155 pineal gland 6^7 pituitary gonadotrope cells 164 pituitary hormone delivery 196 polycystic ovary syndrome (PCO) 115 portal vein 193, 194, 199 power spectrum 92 primary hyperparathyroidism 235 prolactin 102, 145, 150^151 prolactinomas 114 protein kinase A (PKA) 8, 17, 40, 240 protein kinase C (PKC) 236 protein phosphorylation 29 PTH 225^243
SUBJECT INDEX
in bone disease 235^236 concentration vs. time pro¢le 228 deconvolution analysis 227 hypercalcaemia 232^234 hypocalcaemia 229^231 pulse analysis 226^227 release 229^234 vitamin D 234 pulse detection analysis 108 pulse duration, cell responsiveness as function of 27 pulse frequency modulation 242 pulse identi¢cation 116^117 R receptor desensitization/resensitization 23, 26, 33, 40, 42, 43 receptor internalization 33 receptor tyrosine kinases (RTKs), activation and downstream signalling 46^47 recurrence prediction 115 REM sleep 144, 161 rest^activity cycle 120, 123, 129, 130, 133, 138, 139, 144, 158 and chronopharmacology mechanisms 124 cellular rhythms coupling to 122^124 misalignment 152^153 S sarcoplasmic reticulum 26 self-organized quanti¢cation of pulsatility (SOPUL) 110, 111 serotonin N-acetyltransferase (AANAT) 6^7, 9^13 sexual dimorphism 103 of liver gene expression 61^81 shift work 153, 158 short-term desensitization 241^242 signal-induced Ca2+ spiking 28 signal variation 1^2 signalling speci¢city determinants 47^48 single cell Ca2+ oscillations 117 sinusoidal heart rhythms (SHRs) 100 sleep loss e¡ects on diurnal hormonal pro¢les 151^152 sleep onset and o¡set 144 sleep spindle activity 144 sleep^wake cycle 145, 150 sleep^wake homeostasis 153
269
slow-wave (SW) sleep 144 SNAPs 202 SNAREs 202 SOCS/CIS gene 75 somatostatin (SRIF) 195, 200, 202, 207^210, 215, 222^223, 260 spectral analysis 116 spiral waves 37 squamous cell carcinoma 128 STAT5 signalling 61^81 STAT5a 70^72, 77 STAT5b 70^72, 75^80 activation pathway in CWSV-1 cells 66 activation/deactivation cycle 69 GH pulse-induced signalling 67 stochastic processes 97, 171^173 stress hormones 158 stress responsiveness 259 substance P 260 suprachiasmatic nucleus (SCN) 6, 17, 121, 185, 186 survival di¡erences 140^141 T temperature time-series 99 testicular Leydig cells 164 testosterone 164, 173, 185^187 theprubicin 124 three-dimensional waves 37 thyrotropin (TSH) 103, 149, 151, 152, 242 time delays 167^168 transcription factors 240, 241 TRH 222 truncal vagotomy 204 tyrosine phosphatase (PTPase) SHP-1 67^68 tyrosine phosphorylation 67 U ultradian rhythms 128 hypothalamo^pituitary^adrenal (HPA) axis 245^248 V vinorelbine 122 vitamin D e¡ect on PTH release 234 vowel classi¢cation task 111 W within-axis time delays 167^168