SCIENCE'S COMPASS
28. F. Desdouits, J. C. Siciliano, P. Greengard , J. A. Girault, 40. A. A. Fienberg eta/., Science 281, 838 (1998). have worked in our laboratory. I would particularly Proc. Nat/. Acad. Sci. U.S.A. 92, 2682 {1995). 41. P. B. Allen, C. C. Ouimet, P. Greengard, Proc. Nat/. like to mention A. C. Naim, who has been a dose 29. A. Nishi, G. L Snyder, P. Greengard, }. Neurosci. 17, Acad. Sci. U.S.A. 94, 9956 (1997). colleague and friend for more than 20 years. Th1s 8147 (1997). 42. Z. Van et al.. Nature Neurosd. 2, 13 {1999). work has also benef1ted enormously from collabora 30. G. L Snyder, A. A. Fienberg. R. L Huganir, P. Green 43. J.-A. Girault, H. C. Hemmings Jr., K. R. Williams, A. C. tions w1th eKcellent scientists at several other uni gard,j. Neuroscl. 18, 10297 (1998) Nairn, P. Greengard, }. Bioi. Chem. 264, 21748 versities. Our work on regulation of 10n pumps was 31 . P. Svenningsson eta/., Neurosd. 84. 223 {1998). (1989). carried out in collaboration with A. Aperia at the 32. A. Nishi, G. L Snyder, A. C. Naim, P. Greengard, 44. F. Desdouits, D. Cohen, A. C. Nalm, P. Greengard, J.·A. Karolinska Institute. We continue to collaborate with }. Neurochem. 72, 2015 {1999). Girault, }. Bioi. Chem. 270, 8772 (1995). R. L Huganir, who was at The Rockefeller University, 33, P. Svenningsson et al .. j. Neurochem. 75, 248 {2000). 45. J, A. Bibb eta/., Nature 402(6762), 669 (1999). and Is now at The johns Hopkins University School of 34. G. l. Snyder et at .. j. Neurosci. 20, 4480 (2000). 46. E. F. da Cruz e Silva et a/.,}. Neurosci. 1 5(5), 3375 Medicine and with E. J. Nestler, who was at the Yale 35. A. A. Flenberg. P. Greengard, Brain Res. Rev. 31, 313 (1995). University School of MediCine and is now at the (2000). 47. P. Greengard, J, jen, A. C. Naim. C. F. Stevens, Science University of Texas Southwestern Medical Center. 36. P. Svenningsson et a/., Proc. Nat/. Acad. Sci. U.S.A. 253, 1135 (1991). Much of our electrophysiological work has been done 97, 1856 (2000). 48. l.-Y Wang. M. W. Salter, J. F. MacDonald, Science in collaboration with D. J. Surmeier at Northwestern 37. A. Nishi eta/., Proc. Nat/. Acad. Sci. U.S.A. 97, 12840 253. 1132 (1991). University. The work of our research group has been (2000). 49. l. Hsieh-Wilson et al .. unpublished data. very generously supported for over 30 years by the 38. J. A. Bibb eta/., Nature 410, 376 {2001). SO. C. Rosenmund eta/., Nature 368, 853 (1994). National Institutes of Health, including the National 39. H. C. Hemmings Jr .. P. Greengard, H. Y. L Tung. P. 51. The work summarized here reflects outstanding con· Institute of Mental Health, the National Institute on Cohen, Nature 310, 503 (1984). tributions from many highly gifted associates who Drug Abuse, and the National Institute on Aging.
REVIEW: NEUROSCIENCE The Molecular Biology of Memory Storage: A Dialogue Between Genes and Synapses
Eric R. Kandel* form. This was the approach that AJden Spen One of the most remarkable aspects of an animal's behavior is the ability to modify cer and l took when we joined forces at Nil I in that behavior by learning. an ability that reaches its highest form in human beings. For 1958 to study the cellular properties of the me, learning and memory have proven to be endlessly fascinating mental processes hippocampus, the part of the mammalian brain because they address one of the fundamental features of human activity: our ability thought to be most directly involved in aspects to acquire new ideas from experience and to retain these ideas over time in memory. of complex memory (/). We initially asked. Moreover, unlike other mental processes such as thought, language, and conscious rather na'ivcly: Are the electrophysiological ness, learning seemed from the outset to be readily accessible to cellular and properties of the pyramidal cells of the hip molecular analysis. I, therefore, have been curious to know: What changes in the brain pocampus. which were thought to be the key when we learn? And, once something is learned, how is that information retained in hippocampal cells involved in memory storage, the brain? I have tried to address these questions through a reductionist approach that fundamentally different from other neurons in would allow me to investigate elementary forms of learning and memory at a cellular molecular level-as specific molecular activities within identified nerve cells. the brain? With study, it became clear to us that all nerve cells, including the pyramidal cells of the hippocampus. have similar signaling prop erties. Therefore. the inoinsic signaling proper first became interested in the study of mem the National Institutes of Health (NlH) in Be ties of neurons would themselves not give us ory in 1950 as a result of my readings in thesda from 1957 to 1960, I focused on learning key insights into memory storage (2). The Ip sychoanalysis while still an undergraduate more about the biology ofthe brain and became unique functions of the hippocampus had to at Harvard College. Later, dunng medical train interested in knowing how learning produces arise not so much from the intrinsic properties ing. I began to find the psychoanalytic approach changes in the neural networks of the brain. of pyramidal neurons but from the pauem of limiting because it tended to treat the brain, the My purpose in translating questions about functional interconnections of these cells. and organ that genertltes behavior, as a black box. In the psychology of learning into the empirical how those interconnections are affected by the mid- 1950s, while sti ll in medical school. I language of biology was not to replace the logic learning. To tackle that problem we needed to began to appreciate that during my lifetime the of psychology or psychoanalysis with the logic know how sensory information about a learning black box of the brain would be opened and that of cellular molecular biology, but to try to join task reaches the hippocampus and how infor the problems of memory storage. once the ex these two disciplines and to contribute to a new mation processed by the hippocampus influenc clusive domain of psychologists and psychoan synthesis that would combine the mentalistic es behavioral output. This was a fom1idable alysts, could be investigated with the methods psychology of memory storage with the biology challenge, since the hippocampus bas a large of modem biology. As a result. my interest in of neuronal signaling. I hoped further that the number of neurons and an immense nurnber of memory shifted from a psychoanalytic to a biological analysis of memory might carry with interconnections. lt seemed unlikely that we biological approach. As a postdoctoral fe llow at it an extra bonus, that the study of memory would be able to work out in any reasonable storage might reveal new aspects of neuronal period of time how the neural networks. in signaling. Indeed, this has proven true. which the hippocampus was embedded. partic Howard Hughes Medical Institute, Center for Neuro ipate in behavior and how those networks are biology and Behavior, College of Physicians and Sur· A Radical Reductionist Strategy to geons of Columbia University, New York State Psy affected by learning. chiatric Institute, 1051 Riverside Drive, New York, NY learning and Memory To bring the power of modem biology to 10032, USA. E-mail: [email protected] At first thought, someone interested in learning bear on the sn1dy of learning, it seemed nec *This essay is adapted from the author's address to and memory might be tempted to tackle the essary to take a very different approach- a the Nobel Foundation. December 2000. problem in its most complex and interesting radically reductionist approach. We needed
1030 2 NOVEMBER 2001 VOL 294 SCIENCE www.sciencemag.org SCIENCE'S COMPASS to study not the most complex but the sim tern i!; made up of a small number of nerve ing (5-7). As we examined these three plest mstances of memory storage, and to cells: many of these are gtgantic: and (as forms of learning, we were struck by the study them in animals that were most tracta became evident to me later) many are unique resemblance each had to corresponding ble experimentally. Such a reductionist ap ly identifiable (3, 4). Whereas the mammali forms of learning in h1gher vertebrates and proach was hardly new in 20th-century biol an brain has a trillion central nerve cells. humans. As with vertebrate learning. mem ogy. One need only think of the use of Dro Aplysia has only 20,000. and the simplest ory storage for each type of learning in sophila in genetics, of bacteria and bacterio behaviors that can be modified by learning Aplysia has two phases: a transient memory phages in molecular biology. and of the squid may directly involve less than 100 central that lasts minutes and an enduring memory giant axon in the study of the conduction of nerve cells. ln addition to being few in num that lasts days. Conversion of short-term to nerve impulses. Nevertheless, when tt ean1e bers, these cells are the largest nerve cells in long-tenn memory storage requires spaced to the study of behavior, many mvestigators the animal kingdom. reaching up to 1000 tJ.m repetition-practice makes perfect. even in were reluctant to use a reductionist strategy. in diameter, large enough to be seen with the snails (Fig. I B) (6-8). In the 1950s and 1960s many biologists and naked eye. One can record from these large We focused initially on one type of learn most psychologists believed that learning was cells for many hours without any difftcuJty, ing. Sensitization is a form of learned fear m the one area of biology in which the use of and the same cell can be returned to and which a person or an experimental animal simple animal models, particularly inverte recorded from over a period of days. The learns to respond strongly to an otherwise brate ones, was least likely to succeed. They cells can easily be dissected out for biochem neutral stimulus (5, 6, 8). For example, if a argued that only higher animals exhibit inter ical studies, so that from a single cell one can person is suddenly exposed to an averstve esting forms of learning and that these forn1s obtain sufficient mRNA to make a eDNA stimulus, such as a gunshot going otT nearby, require new·onal organizations and neuronal library. Finally, these identtfied cells can that person will be sensitized by the unex mechanisms qualitatively different from readily be injected with labeled compounds, pected noise. As a result. that person will be those found in simple animals. antibodies, or genetic constructs, procedures frightened and will now startle to an other It was my belief. however, that concerns which opened up the molecular study of sig wise innocuous stimulus like a tap on the about the use of a simple experimental sys nal transduction within individual nerve cells. shoulder. Similarly, on receiving an aversive tem to study learning were misplaced. Lf el Irving Kupfermann and I soon delineat shock to a part of the body such as the tail, an ementary forms of learning are common to all ed a very simple defensive reflex: The Aplysia recognizes the stimulus as aversive animals with an evolved nervous system, withdrawal of the gill upon stimulation of and learns to enhance its defensive reflex there must be conserved features in the mech the siphon, an action that is like the quick responses to a variety of subsequent stimuli anisms of learning at the cell and molecular withdrawal of a hand from a hot object. applied to the siphon, even innocuous stimuli level that can be studied effectively even in When a weak tactile stimulus is applied to (Fig. 1A) (9). The antmal remembers the simple invertebrate animals. the siphon, both the siphon and gill are shock, and the duration of this memory is a withdrawn into the mantle cavity for pro function of the number of repetitions of the A Simple Learned Behavior in an tection under the mantle shelf (fig. 1A) (5). noxious experience (Fig. I B). A single shock Invertebrate Kupfermann, Harold Pinsker, and later gives rise to a memory lasting only minutes; After an extensive search for a suitable ex Tom Carew, Robert Hawkins, and I found this short-term memory does not require the perimental animal, l settled on the giant ma that this simple reflex could be modified by synthesis of new protein. ln contrast, four or rine snail Aplysia (Fig. lA) because it offers three different forms of learning: habitua five spaced shocks to the tail give rise to a three important advantages: Its nervous sys- tiOn, sensitization. and classical condition- memory lasting several days: this long-term
A Gill Withdrawal Reflex Sensitization 8 1000 4 trains I day for 4 days I
Mantle shelf
4 single shocks /
100 Tactile Tactile stimulus stimulus 0 0 4 7 Days after training Fig. 1. A simple learned behavior. (A) A dorsal view of Aplysia showing tion converts short-term memory into long-term memory in Aplysia. the gill, the animal's respiratory organ. A light touch to the siphon Before sensitization training, a weak touch to the siphon causes only with a fine probe causes the siphon to contract and the gill to a weak, brief siphon and gill withdrawal reflex. Following a single withdraw. Here, the mantle shelf is retracted for a better view of the noxious, sensitizing, shock to the tail, that same weak touch produces gill. Sensitization of the gill-withdrawal reflex, by applying a noxious a much larger siphon and gill reflex withdrawal response, an enhance stimulus to another part of the body, such as the tail, enhances the ment that lasts about 1 hour. More tail shocks increase the size and withdrawal reflex of both the siphon and the gill. (B) Spaced repeti- duration of the response. [Modified from (79)]
www.sciencernag.org SCIENCE VOL 294 2 NOVEMBER 2001 1031 SCIENCE'S COMPASS memory does requ1re the synthesis of new the cell biological study of learning: How can animal underwent sensitization. cla:.stcal condi protem. Further trammg, four brief trains a leammg occur m a neural circutt that is so tJorung, or hab1tuauon. Our ~tudics provtdcd day for four days, g1ves nse to an even more precisely wired? clear evidence for the 1dea proposed by Ramon endurmg memory lastmg weeks. wlucb also In 1894. Sant1ago Ramon y Cajal pro y Cajal. that learnmg results from changes m the requ1res nC\\ protem synthests. Thus, JUSt as posed a theory of memory storage according strength of the synapttc connect1onJ> between in complex learnmg m mammal ( 10, I 1 ). the to which memory is stored in the growth of precisely mterconnccted cells (I 2. 20). Thus. long-term memory tor sensitization differs ne\\ connections (19). Th1s presc1ent idea while the organ1sm's developmental program from the short-term memory in requiring the was neglected in good part for half a century assures that the connections !>~!tween cells are synthesis of new protems. Th1s was our ftrSt as students of learning fought over newer mvanant. 11 docs not spcc1fy their prec1sc clear evidence for the conservation of bio competmg ideas. First, Karl Lashley, Wolf strength. Rather, expencnce alter.. the strength chemical mechamsms between Aplysia and gang Kohler, and a number of Gestalt psy and effccttvcncss of these preex1sttng chem1cal vertebrates. chologists proposed that learning leads to connections. ccn in the perspective of these Kupfermann, Castellucci, Carew, Hawktns, changes in electric fields or chemical gradi three forms of learning, synaptic plasticity John Byrne, and I worked out sign.iticant com ents, which they postulated surround neuro emerged as a fundamental mechanism for mfor ponents of the neural circuit g1ll-withdrawal re nal populattons and arc produced by the ag mation storage by the nervous system, a mech flex (Fig. 2). The c1reuit is located in the abdom gregate activity of cells rccntited by the anism that is butlt into the very molecular archi inal ganglion and has 24 central mechanorecep learning process. Second, Alexander Forbes tecture of chem1cal synapses (2 I). tor sensory neurons that innervate the siphon and Lorente de No proposed that memory is ~km and make direct monos}'naptic connections stored dynamically by a self-reexcitmg chain Molecular Biology of Short- and w1th 6 gtll motor cells (f1g. 2C) (12 14). The of neurons. Donald Hebb later championed Long-Term Memory Storage sensory neurons also made mdirect connections th1s 1dca as a mechantsm for short-term mem What are the molecular mechamsms whereby wtth the motor cells through small groups of ory. Fmally, Jlolger Hyden proposed that short-tenn memory 1s estabhshcd, and how is it excttatory and mb1b1tory tntemeurons (15, 16). leammg led to changes m the base composi converted to long-term memory'' lmttally, we In addttton to bcmg 1dent1fiablc. indi\1dual cells tion of D A or RNA Even though there was focused on short-tenn sensmzanon. In collabo also have :.urpnsmgly large effects on beha\10r much discussion about the merits of each of ration with James H. Sch\\ artz. \\C found that (Fig. 28)(4, 14, /7). As we examined the neural these 1deas. there was no direct evidence to the synaphc changes, like short-term behaVIor. c1rcutt of this reflex, we were struck by its support any of them (2). were expressed even when protem synthesis invanance. In every an1mal we exammed. each Kupfermann. Castellucci, Carew, Hawkins, was inhibited. This findmg firs~ suggested to us cell connected only to ccrtam target cells and not and I addressed these altemaltve 1deas directly that short-tenn synapttc plasticity might be me to others (Fig. 2C). This also was true in the by confronting the question of how learning can diated by a second messenger system such as neural c1rcwtry of other behaviors in Aplysia occur m a circuit with fixed neuronal elements. cyclic AMP (22). 1-ollowtng up on th1s 1dea, includmg inking. control of the circulation. and To address this question, we examined the neu Schwartz, Howard Cedar, and I found m 1972 locomotion (4, 18). llus ra1sed a key question in ral circuit of the gill-withdrawal reflex while the that stimulation of the modulatory pathways
A c Fig. 2. The neural circuit of the Aplysia gill-with drawal reflex. (A) In this dorsal view of the ab dominal ganglion, the six identified motor L7 Motor cells to the gill are ~uron brown and the seven sensory neurons are blue. A sensory neuron that synapses on gill motor neuron L7 is stimulated electrically Modulatory lnterneurons with an Intracellular electrode and a micro ..._._,__ electrode in the motor lnterneurons neuron records the syn aptic potential pro duced by the action po B tential in the sensory neuron (see middle Gill trace in (B)). The sensory neuron carries the input from the siphon skin; the motor neuron makes direct connections onto the gill (B) Individual cells make significant contributions to Motor Motor 1'\.. Motor Ill uJ the reflex. Stimulating a single motor neuron (traces on the left) produces a detectable Neuron Neuron Neuron Jl.! change in the gill and stimulating a single sensory neuron produces a large synaptic 1'--- potential in the motor neuron (traces in the middle). Repeated stimulation of a single Sensory Sensory l sensory neuron increases the frequency of firing in the motor neuron, leading to a visible Neuron • L--- Neuron __ reflex contraction of the gill (traces on the right). A single tactile stimulus to the skin normally activates 6 to 8 of the 24 sensory neurons, causing each to fire 1 to 2 action potentials. The repetitive firing of 10 action potentials in a single sensory neuron, designed to simulate the firing of the total population (trace on the right) simulates the reflex behavior reasonably well. {C) Diagram of the circuit of the gill-withdrawal reflex. The siphon is innervated by 24 sensory neurons that connect directly with the six motor neurons. The sensory neurons also connect to populations of excitatory and inhibitory interneurons that in turn connect with the motor neurons. Stimulating the tail activates three classes of modulatory interneurons (serotonergic neurons, neurons that release the small cardioactive peptide, and the L29 cells) that act on the terminals of the sensory neurons as well as on those of the excitatory interneurons. The serotonergic modulatory action is the most important; blocking the action of these cells blocks the effects of sensitizing stimuli. (From (25)]
1032 2 NOVEMBER 2001 VOL 294 SCIENCE www.sciencemag.org S CIE N CE'S COM P A SS recruited during hetcrosynaptJe facilitation led short-tem1 memory- the regulation of trans blocked the actions ofserotonin. These fmdings to an increase in cAMP in the abdominal gan mitter release (Fig. 3). provided direct evidence for the role of PKA in glion (23). Cedar and Schwartz found that the How does cAMP enhance transmitter re short-term presynaptic facilitation (31. 32). neurotransmitter candidates serotonin and dopa lease? Serotonin, or injected cAMP, leads to In an elegant series of experiments, Steven mine could simulate this action of electrical mcreased excitabthty and a broadenmg of the Siegclbaum. Joseph Camarda and Michael 4 stimulation and increase levels of cAMP (24). action potential by reducing specific K + cur Schuster identified a novel K charmel. the Later, Hawkins. Castcllucct. David Glanzman, rents. allowing greater Ca2 influx into the S-type K+ channel, and showed that it too and J delineated the modulatory system activat presynaptic terminal with each action potential could be modulated by cAMP (33) and that ed by a sensitizing stimulus to the tail (16. 25, (29). The greater ca~ influx could connibute PKA could act on the S-type K 1 chrumel di 26), and confirmed that it contains serotonergic to the enhanced transmitter release. Following rectly (34). Later, Byrne showed that serotonm intemeurons. the lead of Paul Greengard, who had proposed also modulates a delayed-rectifier K .. (32). The We next found that serotonin acts on spe that cAMP produces it~ action in the bram S-type channel mediated the increase in excit cific receptors in the presynaptic terminals of through the cAMP-dependent protein kinase ability with a minor contribution to broadening, the sensory neuron to enhance transmitter (PKA), Marc Klein and J suggested that cAMP whereas the delayed-rectifier K charmcl con release. In 1976, Marcello Brunelli, Castel may cause phosphorylation of this K+ channel tributed little to excitability but had a major role lucct, and I injected cAMP directly into the by activating PKA (29). ln collaborative exper in spike broadening. Finally, Hochner. Klein, presynaptic cells and found that it too pro Iments with Paul Greengard in 1980, Castel and J- ru1d independently Jack Byrne and Ius duced presynaptic facilitation (27. 18). This lucci, Schwartz. and I found that the active colleagu.:s-showed that, m addition to spike provided the most compelling evidence then catalytic subunit of PKA by itself produced broadening, serotonin also enhanced release by available that cAMP is involved in control broadening of the action potential and enhance an as-yet-unspecified action on the release ma hog synaptic strength and gave us our first ment of glutamate release (30). Conversely. the chinery. Tht!S, serotonin leads to an increase in instght into one molecular mechanism of specific peptide inhibitor of PKA (PKJ) presynaptic cAMP, which activates PKA and leads to synaptic strengthening through en hanced transmitter release produced by a com Sensory Neuron bination of mechanisms (Fig. 3) (32). CREB-1 mediated transcription. By sub stituting puffs of serotonin, the trru1smitter Long Term released by tail shocks. for the tail shocks themselves, Samuel Schacher. Pier Gtorgio Montarolo. Philip Goelet. and l modeled sen sitization in a culture dish consisting of a single sensory cell making synaptic connec tions with a single motor cell (35). We were able to induce both short- and long-term fa cilitation and found, as with the intact animal, Short Term that the long-term process differed from the short-term process in requiring the synthesis Tail SHT of new proteins. We used this cell culture to ask: What genes ~ -·t----<1111::3 are activated to convert the short-term to the long-term process, and what genes are essential ~ Channels .,.--- for the maintenance of the long-term process? We found that five spaced puffs of serotonin (simulating five spaced shocks to the tail) acti vate PKA, which in tum recruits the mitogen activated protein kinase (MAPK). Both translo cate to the nucleus. where they activate a tran Motor Neuron sctiptional cascade beginning with the tran scription factor CREB-1, the f;AMP response Fig. 3. Effects of short- and long-term sensitization on the monosynaptic component of the c;:lement Q.inding protein-!, so called because it gill-withdrawal reflex of Aplysia. In short-term sensitization {lasting minutes to hours) a single tail shock causes a transient release of serotonin that leads to covalent modification of preexisting binds to a cAMP response element (CRE) m the proteins. The serotonin acts on a transmembrane serotonin receptor to activate the enzyme promoters of target genes (Fig. 3) . The first clue adenylyl cyclase (AC), which converts ATP to the second messenger cyclic AMP. In turn, cAMP to the importance of CREB in long-term mem recruits the cAMP-dependent protein kinase A {PKA) by binding to the regulatory subunits ory was provided in 1990 by Prarnod Dash and (spindles), causing them to dissociate from and free the catalytic subunits (ovals). These subunits Binyamin Bochner (36). They injected, into the can then phosphorylate substrates {channels and exocytosis machinery) in the presynaptic termi nals, leading to enhanced transmitter availability and release. In long-term sensitization, repeated nucleus of a sensory neuron in culture, oligo stimulation causes the level of cAMP to rise and persist for several minutes. The catalytic subunits nucleotides carrying the CRE DNA element, can then translocate to the nucleus, and recruit the mitogen-activated protein kinase (MAPK). In thereby titrating out CREB. This treatment the nucleus, PKA and MAPK phosphorylate and activate the cAMP response element-binding {CREB) blocked long-term but not short-term facilita protein and remove the repressive action of CREB-2, an inhibitor of CREB-1. CREB-1 in turn tion (Fig. 3 ). Later. Dusan Bartsch cloned Aply activates several immediate-response genes, including a ubiquitin hydrolase necessary for regu sia CREB-la (ApCREB-la) and showed that lated proteolysis of the regulatory subunit of PKA. Cleavage of the {inhibitory) regulatory subunit results in persistent activity of PKA, leading to persistent phosphorylation of the substrate proteins injection of the phosphorylated form of this of PKA. A second immediate-response gene activated by CREB-1 is C/EBP, which acts both as a transcription factor by itself could initiate the homodimer and as a heterodimer with activating factor (AF) to activate downstream genes long-term memory process. Downstream from [including elongation factor 1o {EF1a)] that lead to the growth of new synaptic connections. ApCREB (37), Cristina Alberini and Bartsch
www.sciencemag.org SCIENCE VOL 294 2 NOVEMBER 2001 1033 SCIENCE'S COMPASS
found two additional positive transcription reg lient features arc learned. Memory suppressors changes. Thus, despite recruitment of nuclear ulators, the CAA T box enhancer binding pro may allow for the modulation of memory stor processes, long-tem1 changes in synaptic func tein (ApCIEBP) and activation factor (Ap/AF) age by emotional stimuli, as occurs in "flash tion and structure are confined only to those (38, 39). CREB-1 activates this set of immedi bulb memories," memories of emotionally sy11apses stimulated by serotonin. ate response genes, which in turn act on down charged events that are recalled in detail, as if a How does this come about? Marttn, An stream genes. to give rise to the growth of new complete picture had been instantly and pow drea Casadio, Bailey, and I found that five synaptic connections (Fig. 3) (36, 40- 46). As erfully etched in the brain. puffs of serotonin send a signal to the nucleus first shown by Craig Bailey and Mary Chen, to activate CREB-1, which then appears to long-term memory endures by virtue of the Synapse-Specificity of long-Term send proteins to all terminals; however, only growth of new synaptic connections, a structur Facilitation those terminals that have been marked by al change that parallels the duration of the be The finding of a transcriptional cascade ex serotonin can use these proteins productively havioral memory (45- 48). As the memory plained why long-tem1 memory requires new for synaptic growth. Indeed, one puff of se fades, the connections retract over time. A typ protein synthesis immediately after training. rotonin to the previously unstimulated syn ical sensory neuron m the intact Ap(vsia bas but it posed a new cell-biological problem. A apse is sufficient to mark that synapse so that about 1200 synaptic varicosities. Following single neuron makes hundreds of contacts on it can capture a reduced fom1 of the long-tem1 long-term sensitization, the number more than many different target cells. Short-term synap facilitation induced at the other site by five doubles to about 2600; with time the number tic changes are synapse-specific. Since long puffs of serotonin (Fig. 4B). returns to about 1500. lasting synaptic changes require transcription These results gave us a new and surprising lnhibit01y constraints. ln 1995 Bartsch and thus the nucleus, is long-term memory insight into short-term facilitation. The stimulus found that positive regulators are only half the storage a cell-wide process, or are there cell that produces the short-term process has two story-there are also inhibitory constraints on biological mechanisms that maintain the syn functions (Fig. 4C). When acting alone, it pro memory (49) . Long-term synaptic facilitation apse specificity of long-term facilitation? vides a selective, synapse-specific enhancement requires not only activation of memory-enhanc To examine these questions. Kelsey Martin ofsynapt ic strength, which contributes to short er genes, but also inactivation of memory-sup cultured one Ap(vsia sensory cell with a bifur tenn memory, lasting minutes. When acting in pressor genes (Fig. 3). One of these, the tran cating axon with two motor neurons, forn'ling conjunction with the activation of CREB initi scription factor ApCREB-2, can repress two widely separated synapses (Fig. 4A). ln this ated by a long-term process in either that syn ApCREB-1 a mediated transcription; relieving culture syste114 a single puff of serotonin ap apse or in any other synapse on the same neu this repression lowers the threshold for the plied to one synapse produces transient factli ron, the stimulus locally marks those synapses long-term process. tation at that synapse only, as expected (50, 51). at which it occurs. The marked synapse can Thus, during long-term memory storage, a Five puffs of serotonin applied to one bnmch then utilize the proteins activated by CREB for tightly controlled cascade of gene activation is produces long-lasting facilitation (72 hours), synaptic growth to produce a persistent change switched on, with memory-suppressor genes also restricted to the stimulated synapse (Fig. in synaptic strength. Thus, the logic for the providing a threshold or checkpoint for memo 4B). This long-lasting synapse-specific facilita long-tetm process involves a long-range inte ry storage, presumably to ensure that only sa- tion requires CREB and also leads to structural gration that is different from the short-term
B Two Different Functions of A Initiation c c'~"'' the Short-Term Process i 1. Short-Term Memory Storage 9 • 5 X 5HT iI> Fig. 4. A single sensory neuron connects to Motor () many target cells. The requirement of a tran neurons'!( Q Q scriptional mechanism for long-term memory raises the question: What is the unit of long Q) 2. Marking lor the Capture of the Long-Term '0 12 term information storage? Is it a single syn £1 Process and the Growth of New Synapses apse, as with short-term facilitation, or the ~ Synaptic Capture: Cell B entire neuron? Is there a mechanism for re <( 150 stricting synaptic facilitation to some synaptic 0..en connections? (A) This photomicrograph shows tt 100 a culture system developed to examine the action of two independent branches of a single 50 in Aplysia sensory neuron (the small neuron in :r.t: I + I I the middle) on two different motor neurons 0 ...... (large neurons). Serotonin can be selectively 01 4 12 24 48 72 applied to one and not the other of the two Time(h) branches. The flow of the serotonin can be mon- Q Q Q itored with the dye, fast green. [From (50)) (B) Long-term facilitation is synapse-specific and can be captured at another of serotonin. (From (50)) (C) Two effects of short-term facilitation: short branch by the stimulus that initiates the short-term process. Five puffs of term memory storage when acting by itself and marking of the specific serotonin applied at the initiation site (cell A) produce a synapse-specific synapse to which it is applied for subsequent capture of the proteins facilitation shown in (B). This synapse-specific facilitation is not evident at necessary for long-term facilitation and growth when applied in conjunction the synapse of cell B unless that synapse is itself primed with a single puff with five pulses to another set of terminals. 1034 2 NOVEMBER 2001 VOL 294 SCIENCE www.sciencemag.org SCIENCE ' S COMPASS process. In the long-tenn, the function of a contributing to short-term memory for sensi Larry Squire had shown that explicit memory, synapse is not only determined by the history of tization. Third, an even more persistent syn like implictt memory, has a short-term phase usage of that synapse. It is also detennined by aptic action, lasting days, results from repeat that does not reqwre protein synthesis and a the state of the transcriptional machinery in the ed action of a modulatory transmitter such as long-term phase that docs (55). Are these two nucleus. serotonin. With repeated applications of se components of memory storage also represent How docs one puff of serotonin mark a rotonin, second-messenger kinases translo ed at the cellular level? What rules govern synapse for long-term change? For structural cate to the nucleus, where they activate a explicit memory storage? changes to persist, local protein synthesis is cascade of gene induction leading to the A decade ago, when 1 reached my 60th required (51). Oswald Steward's important growth of new synaptic connections. This of birthday. I gathered up my courage and re work in the early 1980s had shown that den course rai~es the problem of synapse speci turned to the htppocampus. Mario Capecchi drites contain ribosomes. and that specific ficity that we have considered above. Our and Oltver Smithies, by achiev111g targeted mRNAs are transported to the dendrites and experiments, in the bifurcated culture system, gene ablation in mouse embryontc stem cells, tr.mslated there (52). Our experiments revealed a novel fourth action of neurotrans provided a superb genetic system for relating showed that one function of these locally mitters, the marking of the synapse and the individual genes to synaptic plasticity, on the translated mRNAs was to stabilize the syn regulation of local protein synthesis, which one hand, and to complex explicit memory apse-specific long-term functional and struc contributes to the establishment of synapse storage on the other. Mice have a medial tural changes. specific long-term facilitation. temporal lobe system. including a hlppocam Neurotransmiller regulation of local pro pus, that resembles that of humans, and they tein synthesis. These studies thus revealed a Explicit Memory use their hippocampus much as we do to store new, fourth type, of synaptic action mediated I have so far considered only the simplest cases memory of places and obJects (Fig. 6A). by neurotransmitter signaling (Fig. 5). Three of memory storage-those involving reflex Although we still do not know much about of these four have emerged, at least in part, es-a form called implicit or procedural mem how information is transformed as it gets into from the study ofleaming and memory. First. ory. Tmplictt memory is memory for perceptual and out of the hippocampus, it ts weU estab in 1951, Katz and Fatt opened up the modem and motor skills and is expressed through per lished that the hippocampus contains a cellular study of chemical transmission wtth their dis fomlance, without conscious recall of past epi representation of cxtrapersonal space-a cogni covery of ionotropic receptors that regulate sodes. ln contrast, the memories we hold near tive map of space--and lesions of the htp ion flux through transmitter-gated ion chan and dear are called explicit (or declarative) pocampus interfere with spatial tasks (57). nels to produce fast synaptic actions, lasting memories. These memories require conscious Moreover, in 1972, Teije Lomo and Tim Bliss milliseconds (53). Second. in the 1970s, recall and are concerned with memories for discovered that the perforant path, a major path mctabotropic receptors were found to activate people, places, objects, and events. Exphcit way within the hippocan1pus, exhibits activity second-messenger pathways. such as the memory involves a specialized anatomical sys dependent plasticity. a change now called long cAMP-PKA pathway, to produce slow syn tem in the med1al temporal lobe, and a structure term potentiation (LTP) (Fig. 6B). In the CAl aptic activity lasting minutes (54). As we deep to it, the hippocampus (Fig. 6A) (21, 55. region of the hippocampus, L TP is tnduced have seen in Aplysia, this slow synaptic ac 56). How is explicit memory stored? Louts postsynaptically by activation of an NMDA tion can regulate transmitter release, thereby Flexner. Bernard Agranoff, Sam Barondes, and receptor to glutamate. In the late 1980s Richard Morris found that blocking the NMDA receptor Fig. 5. Four consequences of the A Dialog Between Genes and Synapses pharmacologically not only mterfered with L TP action of neurotransmitters. 1. but also blocked memory storage (58, 59). Transmitter activation of a li This earher work on LTP in hippocampal gand-gated ion channel leads to slices had focused on the response to one or two a rapid synaptic action lasting trains of electrical stimuli. But in Aplv.'lia we milliseconds. 2. Transmitter acti vation of a seven transmem had found that long-term memory emerges brane receptor and a second most effectively Wlth repeated stimuli (Fig. messenger kinase leads to a !B). So when Uwe Frey, Yan-You Huang, more enduring synaptic action Peter Nguyen, and f turned to the hippocampus, lasting minutes. 3. Repeated we examined whether LTP changed with re transmitter activation of a seven peated stimulation (60 62) and found that hip transmembrane receptor leads to the translocation of the kinase pocampal LTP has phases, much like facilita to the nucleus and to activation tion in Aplysia. The early phase of LTP, pro of transcription, producing a per duced by a single train of stimuli. lasts only 1 to sistent synaptic action. 4. Trans 3 hours and does not require new protein syn mitter activation of local protein thesis (62); it involves covalent modifications synthesis to stabilize the syn of preextsting proteins that lead to the strength apse-specific facilitation. Proteins for ening of preexisting connections, similar in Growt~ :._ principle to short-term facilitation in Aplysia. By contrast, repeated trains of electrical sumuli ~--- P AA 0• produce a late phase of LTP, which has prop erties quite different from early LTP and similar i to long-term facilitation in Aplysia (Fig. 6B). The late phase of LTP persists for at least a day and requires both translation and transcription. The late phase ofLTP.like long-term storage of implicit memory, requires PKA, MAPK, and CREB and appears to lead to the growth of new synaptic connections (Fig. 6C) (60-69). www.sciencemag.org SCIENCE VOL 294 2 NOVEMBER 2001 1035 SCIENC E 'S COMPASS The late phase of LTP and ex:pficit mem hippocampus-the cells one examines artifi NMDA receptor antagonist (72). When placed ory. To explore further the specific role of cially by using electrically stimulating the in a new environment, the animals with blocked PK.A and late LTP in memory storage, Ted Schaffer collateral pathway while studying NMDA receptors formed a good spatial map Abel, Mark Barad, Rusiko Bourtcbouladze, LTP- are "place cells;" they actually encode that was still stable l hour later. However. by Peter Nguyen, and J generated transgenic extrapersonal space in the animal (7/). A 24 hours, most pyramidal cells no longer re mice that express R(AB), a mutant form of given pyramidal cell will fire only when the tained the representation of the field they had the regulatory subunit of PK.A that inhibits bead of the mouse is in a certain part of an initially. This suggested that activation of enzyme activity (70). In these R(AB) trans enclosed space--the cell's place field. When NMDA receptors-perhaps a step in modifying genic mice, the reduction in hippocampal placed in a new environment, within minutes the strength of the synapse-is required for the PKA activity was paralleled by a significant an animal develops an internal representation long-term stabilization of a place cell map, a decrease in late LTP, while basal synaptic of the space (by the coordinated firing of a result consistent with the role for the late phase transm1ssion and early LTP remained un population of place cells), which is normally of LTP in the stabilization of a place cell map. changed. Most interesting, this defic1t in the stable for days. The same cell will have the We next asked whether a selective deficit late phase of LTP was paralleled by behav same firing field each time the animal is that affects only the late phase of LTP, causes a ioral deficits in hippocampus-dependent reintroduced to that environment. When now selective abnormality in the long-term stability long-term memory for extrapersonal space, placed in a second environment, a new map is of place cells. Since only the late phase of LTP whereas learning, and short-term memory, formed-again in minutes- in part from requires PKA. Alex Rotenberg, Muller. Abel, are unimpaired (Fig. 7. A and 8). Thus, in the some of the cells that made up the map oftbe Hawkins, and 1 returned to the R(AB) trans storage of explicit memory of extrapersonal first environment and in part from pyramidal genic mice with diminished PKA activity and a space in the mammalian hippocampus. PK.A cells that had been silent previously (7/). diminished form of late LTP (73). If reduced plays a critical role in the transformation of lt struck me that the fomllltion of a new map activity of PKA affected the stability of place short-term memory into long-term memory, resembled a learning process. The map devel cells. R(AB) mice should be able to form a much as it does in the storage of implicit ops with time as the animal familiarizes itself stable map of space in a novel environment. as memory in Aplysia and Drosophila. with the space, and once learned, the map of in nom1al animals, that is stable for at least I Using the R(AB) mice we could now ask: space is retained for days and weeks. To first hour. However, the cell field should be unstable Why do animals with compromised PK.A sig test whether the molecular pathways underlying when recorded 24 hours later. This is precisely naling have difficulty with space (70)? We the late phase of LTP were important for the what we fmmd (Fig. ?C). The fact that long were influenced by the classic studies of John long-term stabilization of this map, Cliff Ken term instability in the spatial map and the deficit O'Keefe and John Dostrovsl.-y, who in l 971 tros, Robert Muller, Hawkins. and I sin1ply in long-term memory paralleled the defic1t in discovered that the pyramidal cells of the blocked LTP phannacologically with an the late phase of LTP suggested that PK.A- A c CAEB-2 CAE CAE l CAEB·1 ~ ·, __ ...... ___,_ F.----;;jein 1 j phosphatase 1 PKA rI - Phosphatase I inhibitor B =_g 400 c 8300 Late LTP (4 trains) l 0 t:_ 200 Calcineurin ~ r- Early LTP (1 train) u; 100 ...... _...J.J. ------~~---- en0.. ) 0.. w o~~--~--~--~--~--~--~~ -30 0 30 60 90 120 150 180 210 Time (min) Fig. 6. Long-term potentiation (LTP) in the hippocampus. (A) Three major train of action potentials initiates early LTP by activating NMDA receptors, pathways, each of which gives rise to LTP. The per{orant pathway from the Cal influx into the postsynaptic cell, and the activation of a set of second subiculum forms excitatory connections with the granule cells of the dentate messengers. With repeated trains of action potentials (illustrated here) the gyrus. The mossy fiber pathway, formed by the axons of the granule cells of Cal -+ influx also recruits an adenylyl cyclase (AC), which activates the the dentate gyrus, connects the granule cells with the pyramidal cells in area cAMP-dependent protein kinase. The kinase is transported to the nucleus CA3 of the hippocampus. The Schaffer collateral pathway connects the where it phosphorylates CREB. CREB in tum activates targets (C/EBPB, EPA, pyramidal cells of the CA3 region with the pyramidal cells in the CA 1 region BDNF) that are thought to Lead to structural changes. Mutations in mice that of the hippocampus. (B) The early and Late phases of LTP in the Schaffer block PKA or CREB reduce or eliminate the late phase of LTP. The adenytyl collateral pathway. A single train of stimuli for one second at 100Hz elicits cyclase can also be modulated by dopamine signals and perhaps other an early LTP, and four trains at 10-minute intervals elicit the late phase of modulatory inputs. In addition, there are constraints (in red) that inhibit LTP. The early LTP lasts about 2 hours, the Late LTP more than 24 hours. (C) L-LTP and memory storage. Removal of these constraints lowers the thresh A model for the late phase of LTP in the Schaffer collateral pathway. A single old for L-LTP and enhances memory storage. 1036 2 NOVEMBER 2001 VOL 294 SCIENCE www.sciencemag.org SCIENCE'S COMPASS med1ated gene activation and the synthesis of se\eral days in the Intact animal und was pression, new protein synthesis. and the new protein might be esscnual for the stabili accompamed by enhanced learning and fonnation of new connections. Whereas zation of the spaual map. Navcen Agnihotri, strcngthenmg of short- and long-term mem short-term memory storage for 1mphc1t and Kentros, Hawkins, and l tested th1s 1dea. and ory on several spatial and non-spatial tasks explicit memory reqwres different signal found that inhibiting protein synthe:.1s mdced requiring the hippocampus. These results. to ing. long-term storage of both 1mplicn and destabilized the place fields m the long-tenn gether w1th prev1ous findmgs by Winder and explicit memory uses as a core s1gnaling much as docs mhibitmg PKA (81). Mansuy show10g that ovcrcxprcss10n of cai pathway PKA, MAPK , and CREB-1 At In the course of this work, Kentros and Cineurin 1mpam. PKA-dependcnt components least m the mouse, additiOnal components Agmhotn found, remarkably, that. as JS the of LTP and memory (76, 77), demonstrate are hkely recrUited. In both Implicit and case with exphct memones in humans. a that endogenou~ calcmeunn can act as a neg explicit memory the switch from shon-term key feature m the stabilization of PKA and ative regulator of synaptic plastiCity. Jearn to long-term memory IS regulated by mhlb protein ·ynthesis-dependent phase of mem mg. and memory (F1g. 6C). itory constramts. ory is attent1on (82). When a mouse does Second, the study oflearning has revealed not attend to the space It walks through, the An Overall View new features of synapt1c transm1ssion and man fonns but IS unstable after 3 to 6 hours. Our studies of the storage component of new cell-biolog1cal fi.mctions of synaptic sig When the mouse IS forced to attend to the memory. the molecular mechanism whereby naling. For example, different forn1s of learn space. however, the map invanably is sta mformat1on is stored, have led to two general ing recrUit d1fferent modulatory transmitters, ble for days! conclusions. wh1ch then act m one of three ways: (i) They lnhihztorv conHrainl.l on explicit memory F1rst. our research suggests that the cel activate second-messenger klnases that are Recently we (74) and others (75) have found lular and molecular stratcg1es used in Aply transported to the nucleus where they miuate that the threshold for hippocampal synaptiC sia for stormg short- and long-term mem processes reqUired for neuronal growth and plastictty and memory storage IS determmed ory arc conserved in mammals and that the long-term memory; (i1) they mark the specific by the balance between protem phosphoryl same molecular strategies arc employed in synapses for capture of the long-term process atiOn governed by PKA and depho~phoryl both 1mphc1t and explicit memory storage and regulate local protein synthesiS for stabi ahon (74. 76) To dctermme whether the With both implicit and e~plicit memory lization. and (1i1) they mediate, in ways we endogenous Ca' -sensitiVe phosphatase cal there are stages m memory that arc encoded are JUst begmmng to understand, attentJonal cineunn acts as a constramt on th1s balance. as changes m synaptic strength and that processes required for memory formation and we mhibtted calcmeurin and exammed the correlate w1th the bcha\ wral phases of recall. effects on synaptiC plast1c1ty and memory short- and long-term memory. The short Most important. the study of long-term storage Isabelle Mansuy. Gael Malleret, term synaptic changes involve covalent memory has made us aware of 1he extensive Danny W10dcr. T1m Bhs~. and I found that a mod1ficat10n of precx1stmg proteins, Jead dialog between the synapse and the nucleus, transient reduction of calcineurin activity re mg to modification of pre-existing synaptic and the nucleus and the synap. c (Fig. S). In sulted in facilitation of LTP both 10 v1tro and connections. whereas the long-term synap the long-term process the response of a syn 10 VIVO (74). Th1s facihtahon pers1stcd for tic changes 1nvolve activation of gene ex- apse IS not determined simply by its own Fig. 7. (A) The protocol A Context Conditioning for context condition- Training mg consists of expo- sure to the context fol Testing lowed by a tone and then a shock. The ani- mals are then tested 1 hour and 24 hours aher training. (From (70)) Exposure to Context (81 ) Mutant mice that Onset of Sound (CS) Onset of Shock (US) express the R(AB) gene Same Context in the hippocampus, 1 hour and 24 hours blocking the action of PKA. have a selective defect for long-term Context Conditioning is Selectively Impaired in R(AB) Mice Place Cell Map Stability contextual memory. Is Dependent Upon PKA Mice that express 81 82 c R(AB) were conditioned A(AB) Anisomycin Ell Saline A(AB) CWT to freeze to the con- (wild-type mice) • Amsomycln 0 4 • R(AB) text. Aher becoming ~ 50 !50 familiar with the con- !!!. oiO .. text, the mice heard a g> ~ 0 3 JO sound and received a -~ ~ 02 shock through the elec .!9 trified grid in the floor ~ ~ 0 1 10 U) As a result the animals 0 learned to associate ~ 0 0 the context of the lmm Leamong Testl h lmm Leammg Tesll h Tesl 24 h 1 h 24h space with shock and to freeze when placed in the box at a future time. These mice had good protein synthesis, during training show a similar defect for long-term short-term memory at 1 hour for freezing to context, but at 24 hours memory when tested 24 hours after conditioning. [From (70)] (C) they no longer froze to context, indicating a defect in a form of Place cell stability for R(AB) and wild-type mice. R(AB) mice with a long-term explicit (declarative) memory that requires the hippocam defect in PKA and late LTP form place fields that are stable at 1 hour. pus (82) Wild-type mice exposed to anisomycin, an inhibitor of These fields are not stable at 24 hours. (From (73, 80)) www.sciencemag.org SCIENCE VOL 294 2 NOVEMBER 2001 1037 SC I E N C E'S COMPASS history of aCII\ lty (as 10 ~hort-tcnn plastici 3 W. T FrazJer, E R. Kandel I. Kupfermann, R. Waziri, 47. C H. Baoley, M Chen, Proc. Natl Acad Sd. U.S A 8S ty). but also by the history of transcriptional and R E. CoggeshalL ). Neurophysiol 30, 1288 2373 (1988). (1967). 48. __,J. Neurosc1 9. 1774 (1989) acti~.1tion in the nucleus. <4 E. R Kandel. Cellular &uis of Behavior· An Introduc 49. D. Bartsch et al. Cel/ 83 979 (1995). I started this essay by pointing out that 40 tiOn to BehaVIOral B•ology (Freeman. San Francisco, 50. K. C. Marton et ill., Cell 91. 927 {1998) year; ago. at the beginning of m) career. I 1976). 51. A. Casadoo eta/, Cell 99 221 (1999). 5 H. Ponsker I Kupfermann. V, Castellucci, E. R. Kandel. 52 0 Steward, Neuron 18 9 (1997). thought that a rcductioni~t approach based on Soence 167 1740 (1970) 53 P Fatt, B Katz.,}. Physiol (Land.) 115, 320 (1951). the \l<;C of a simple CX[lCrimemal S) stem such as 6. H M Pon 1038 2 NOVEMBER 2001 VOL 294 SCIENCE www.sciencemag.org Copyright of Science is the property of American Association for the Advancement of Science and its content may not be copied without the publisher's express written permission except for the print or download capabilities of the retrieval software used for access. This content is intended solely for the use of the individual user. Copyright of Science is the property of American Association for the Advancement of Science and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.