Kandel and Hawkins

Home , Ivan Pavlov

© 1992 SCIENTIFIC AMERICAN, INC The Biological Basis of Learnin andIndividuality Recent discoveries suggestg that learning engages a simple set of rules that modify the strength of connections between neurons in the brain. These changes play an important role in making each individual unique

by Eric R. Kandel and Robert D. Hawkins

ver the past several decades, there surgeon at the Montreal Neurological Institute. has been a gradual merger of two In the 1940s Penfield began to use electri originally separate fieldsof science: cal stimulation to map motor, sensory and neurobiology,O the science of the brain, and language functions in the cortex of patients cognitive psychology, the science of the mind. undergoing neurosurgery for the relief of Recently the pace of unification has quick epilepsy. Because the brain itself does not ened, with the result that a new intellectual have pain receptors, brain surgery can be car framework has emerged for examiningpercep ried out under local anesthesia in fully con tion, language, memory and conscious aware scious patients, who can describe what they ness. This new framework is based on the abil experience in response to electric stimuli ap ity to study the biological substrates of these plied to different cortical areas. Penfield ex mental functions. A particularly faSCinating plored the cortical surface in more than 1,000 example can be seen in the study of learning. Elementary as patients. Occasionally he found that electrical stimulation pects of the neuronal mechanisms important for several dif produced an experiential response, or flashback, in which ferent types of learning can now be studied on the cellular the patients described a coherent recollection of an earlier and even on the molecular level. The analysis of learning may . experience. These memorylike responses were invariably therefore provide the first insights into the molecular mecha elicited from the temporal lobes. nisms underlying a mental process and so begin to build a Additional evidence for the role of the temporal lobe in bridge between cognitive psychology and molecular biology. memory came in the 1950s from the study of a few patients Learning is the process by which we acquire new knowl who underwent bilateral removal of the hippocampus and edge, and memory is the process by which we retain that neighboring regions in the temporal lobe as treatment for knowledge over time. Most of what we know about the world epilepsy. In the first and best-studied case, Brenda Milner of and its civilizations we have learned. Thus, learning and mem the Montreal Neurological Institute described a 27-year-old ory are central to our sense of individuality. Indeed, learning assembly-line worker, H.M., who had sufferedfrom untreat goes beyond the individual to the transmission of culture able and debilitating temporal lobe seizures for more than from generation to generation. Learning is a major vehicle for 10 years. The surgeon William B. Scoville removed the medi behavioral adaptation and a powerful force for social prog al portion of the temporal lobes on both sides of H.M.'s ress. Conversely, loss of memory leads to loss of contact with brain. The seizure disturbance was much improved. But im one's immediate self, with one's life historyand with other hu mediately after the operation, H.M. experienced a devastat man beings. ing memory deficit: he had lost the capacity to form new Until the middle of the 20th century, most students of be long-term memories. havior did not believe that memory was a distinct mental Despite his difficulty with the formation of new memories, function independent of movement, perception, attention H.M. still retained his previously acquired long-term memory and language. Long after those functions had been localized store. He remembered his name, retained a perfectly good use to differentregions of the brain, researchers still doubt of language and kept his normal vocabulary; his IQ remained ed that memory could ever be assigned to a specific region. in the range of bright·normal. He remembered well the The firstperson to do so was Wilder G. Penfield, a neuro-

ERIC R. KANDEL and ROBERT D. HAWKINS have collaborated on studies of the neurobiology of learning. Kandel is University MIRROR DRAWINGEXPERIMENT in patients with temporal Professor at the College of Physicians and Surgeons of Columbia lobe lesions gave the first hint, in 1960, that there are two University and senior investigator at the Howard Hughes Medical distinct types of learning systems. One form, which is spared Institute. He received anA.B. from Harvard College, anM.D . from by the lesions, involves tasks that have an automatic quality the New York University School of Medicine and psychiatric train· such as the skilled movements illustrated in this experiment. ing at Harvard Medical School. Hawkins received a B.A. from The subject, who can see his hand only in the mirror, tries to Stanford University and a Ph.D. in experimental psychology from trace the shape of a star. The second type of learning depends the University of California, San Diego. He is associate professor on conscious awareness and cognitive processes and is abol in the Center for Neurobiology and Behavior at Columbia. ished by the lesions.

SCIENTIFIC AMERICAN September 1992 79 © 1992 SCIENTIFIC AMERICAN, INC events that preceded the surgery, such psychological distinction-a division in task better after five days of practice as the job he had held, and he remem the way all of us acquire knowledge. than on the first day, he may respond, bered vividly the events of his child Although it is still not clear how many "What are you talking about? I've never hood. Moreover, H.M. still had a com distinct memory systems there are, re done this task before." pletely intact short-term memory. What searchers agree that lesions of the tem Whereas explicit memory requires H.M. lacked, and lacked profoundly, poral lobes severely impair forms of structures in the temporal lobe of ver was the ability to translate what he learning and memory that require a con . tebrates, impliCit memory is thought to learned from short-term to long-term scious record. In accordance with the be expressed through activation of the memory. For example, he could con suggestion of Neal j. Cohen of the Uni particular sensory and motor systems verse normally with the hospital staff, versity of Illinois and Larry R. Squire of engaged by the learning task; it is ac but he did not remember them even the University of California at San Diego quired and retained by the plasticity in though he saw them every day. and of Daniel L. Schacter of the Univer herent in these neuronal systems. As a The memory deficit following bilater sity of Toronto, these types of learning result, impliCit memory can be studied al temporal lobe lesions was original are commonly called declarative or ex in various reflex systems in either ver ly thought to apply equally to all forms plicit. Those forms of learning that do tebrates or invertebrates. Indeed, even of new learning. But Milner soon dis not utilize conscious participation re simple invertebrate animals show ex covered that this is not the case. Even main surprisingly intact in patients with cellent reflexive learning. though patients with such lesions have temporal lobe lesions; they are referred profound deficits, they can accomplish to as nondeclarative or implicit. he existence of two distinct forms certain types of learning tasks as well Explicit learning is fast and may take of learning has caused the reduc as normal subjects can and retain the place after only one training trial. It of tionists among neurobiologists to memory of these tasks for long periods. ten involves association of simultane askT whether there is a representation on Milner first demonstrated this residual ous stimuli and permits storage of in the cellular level for each of these two memory capability in H.M. with the dis formation about a single event that types of learning process. Both the neu covery that he could learn new motor happens in a particular time and place; ral systems that mediate explicit memo skills normally [see illustration on page it therefore affords a sense of familiar ry and those that mediate impliCit mem 181. She, and subsequently Elizabeth K. ity about previous events. In contrast, ory can store information about the as Warrington of the National Hospital for implicit learning is slow and accumu sociation of stimuli. But does the same Nervous Diseases in London and Law lates through repetition over many tri set of cellular learning rules guide the rence Weiskrantz of the University of als. It often involves association of se two memory systems as they store as Oxford, found that patients such as quential stimuli and permits storage of sociations, or do separate sets of rules H.M. can also acquire and retain memo information about predictive relations govern each system? ry for elementary kinds of learningthat between events. Implicit learning is ex An assumption underlying early stud involve changing the strength of reflex pressed primarily by improved perfor ies of the neural basis of memory sys responses, such as habituation, sensiti mance on certain tasks without the sub tems was that the storage of associa zation and classical conditioning. ject being able to describe just what tive memory, both implicit and explicit, It immediately became apparent to has been learned, and it involves mem required a fairly complex neural circuit. students of behavior that the difference ory systems that do not draw on the One of the first to challenge this view between types of learning that emerged contents of the general knowledge of was the Canadian psychologist Don from studies of patients with temporal the individual. When a subject such as ald O. Hebb, a teacher of Milner. Hebb lobe lesions represented a fundamental H.M. is asked why he performs a given boldly suggested that associative learn ing could be produced by a simple cel lular mechanism. He proposed that as sociations could be formed by coinci PRE-POST COINCIDENCE PRE-MODULATORY COINCIDENCE dent neural activity: "When an axon of (HEBB SYNAPSE) (ACTIVITY-DEPENDENT FACILITATION) cell A ...excite[sl cell B and repeated ly or persistently takes part in firing PRESYNAPTIC ____ MODULATORY it, some growth process or metabolic NEURON -NEURON change takes place in one or both cells such that A's efficacy, as one of the cells firing B, is increased." According to Hebb's learning rule, coincident ac PRESYNAPTIC PRESYNAPTIC tivity in the presynaptic and postsyn NEURON N aptic neurons is critical for strengthen ing the connection between them (a so called pre-post associative mechanism) [see illustration at left1. ======: Ladislav Tauc and one of us (Kandel) d proposed a second associative learning rule in 1963 while working at the Insti TWO CEllULAR MECHANISMS are hypothesized for associative changes in synap tute Marey in Paris on the nervous sys tic strength during learning. The pre-post o coincidence mechanism, proposed by tem of the marine snail Aplysia. They Donald O. Hebb in 1949, posits that coincident activity in the presynaptic and post found that the synaptic connection be synaptic neurons is critical for strengthening the connections between them. The tween two neurons could be strength pre-modulatory coincidence mechanism proposed in1963, based on studies in Aplysia, holds that the connection can be strengthened without activity of the ened without activity of the postsynap postsynaptic cell when a thirdneuron, the modulatory neuron, is active at the tic cell when a third neuron acts on the same time as the presynaptic neuron. Stripes denote neurons in which coincident presynaptic neuron. The third neuron, activity must occur to produce the associative change. called a modulatory neuron, enhances

he marine snail Aplysia (top left) is used in studies the enlargement at the right. Serotonin released from the of the biological basis of learning because its sim modulatory neuron by the unconditioned stimulus acti ple nervous system consists of only 20,000 rela vates adenylyl cyclase in the sensory neuron. When the tively large neurons. The diagram (bottom left) traces one sensory neuron is active, levels of calcium are elevated ofT the pathways involved in classical conditioning of the within the cell. The calcium binds to calmodulin, which in gill-withdrawal reflex in Aplysia. An increase in the release turn binds to adenylyl cyclase, enhancing its ability to syn of neurotransmitter due to activity-dependent facilitation thesize cyclic AMP. The cyclic AMP activates protein kinase, is a mechanism that contributes to conditioning. The mo which leads to the release of a substantially greater amount lecular steps in activity-dependent facilitation are shown in of transmitter than would occur normally.

PRESYNAPTIC ENHANCED SENSORY TRANSMITIER NEURON RELEASE PHOSPHOLIPASE A2

MODULATORY NEURON " SEROTONIN J�RECEPTOR \ DIACYLGLYCEROL ---

ADENYL YL - ""'"""r..,....,� CYCLIC AMP CYCLASE PROTEIN CALMODULIN KINASE

CALCIUM ...... - SENSORY CHANNEL PROTEIN NEURON KINASE C

SIPHON POTASSIUM CONVER AREA \ CHANNEL� GENCE SENSORY CALCIUM NEURON � I CHANNEL TAIL AREA \ NEURO j TRANSMITIER VESICLE

UNCONDITION D RELEASE STIMULUS >===u==,,"" � OTOR SITE PATHWAY M NEURON " MODULATORY CONTROL NEURON � PATHWAY �------�

MANTLE / POSTSYNAPTIC SENSORY SHELF AREA ; MOTOR NEURON NEURON

transmitter release from the terminals ciative mechanism in Aplysia, where it neural networks. Rather the ability to of the presynaptic neuron. They sug contributes to classical conditiOning, an detect associations may Simply reflect gested that this mechanism could take implicit form of learning. Then, in 1986, the intrinsic capability of certain cellu on associative properties if the electri Holger ]. A. Wigstrtim and Bengt E. W. lar interactions. Moreover, these find cal impulses known as action potentials Gustafsson, working at the University of ings raised an intriguing question: Are in the presynaptic cell were coincident Gtiteborg, found that the pre-post asso these apparently different mechanisms with action potentials in the modulato ciative mechanism occurs in the hippo in any way related? Before consider ry neuron (a pre-modulatory associative campus, where it is utilized in types of ing their possible interrelation, we shall mechanism). synaptic change that are important for first describe the two learning mecha Subsequently, we and our colleagues spatial learning, an explicit form of nisms, beginning with the pre-modula Thomas J. Carew and Thomas W. Ab learning. tory mechanism contributing to classi rams of Columbia University and Edgar The finding of two distinct cellular cal conditioning in Aplysia. T. Walters and John H. Byrne of the learning rules, each with associative Classical conditioning was firstde University of Texas Health Science Cen properties, suggested that the associa scribed at the turn of the century by ter found experimental confirmation. tive mechanisms for implicit and ex the Russian physiologist Ivan Pavlov, We observed the pre-modulatory asso- plicit learning need not require complex who immediately appreciated that con-

SCIENTIFIC AMERICANSept ember 1992 81 © 1992 SCIENTIFIC AMERICAN, INC pathway, the mantle shelf, can then be used as a control pathway. The control FORN IX pathway is stimulated the same num ber of times, but the stimulus is not paired (associated) with the tail shock.

TEMPORAL LOBE After five pairing trials, the response to stimulation of the siphon (the paired pathway) is greater than that of the mantle (the unpaired pathway). If the procedure is reversed and the mantle shelf is paired rather than the siphon, the response to the mantle shelf will be greater than that to the Siphon. This differential conditioning is remarkably similar in several respects to that seen

HIPPOCAMPUS in vertebrates. / To discover how this conditioning / SCHAFFER COLLATERAL PATHWAY works, we focused on one component: \ the connections between the sensory HIPPOCAMPUS \ neurons and their target cells, the in \ \ terneurons and motor neurons. Stimu lating the sensory neurons from either the Siphon or the mantle shelf gener PERFORANT FIBER PATHWAY ates excitatory synaptic potentials in the interneurons and motor cells. These synaptic potentials cause the motor cells to discharge, leading to a brisk re

DENTATE REGION flex withdrawal of the gill. The uncon ditioned reinforcing stimulus to the tail activates many cell groups, some of HIPPOCAMPUS stores long-term memory for weeks and gradually transfers it to which also cause movement of the gill. specific regions of the cerebral cortex. The diagram illustrates an example of this Among them are at least three groups process involving a visual image. Neural input travels to the visual cortex and then to the hippocampus, where it is stored for several weeks before it is transferred back of modulatory neurons, in one of which to the cortex for long-term memory. The hippocampus (enlargement) has three ma the chemical serotonin is the transmit jor synaptic pathways, each capable of long-term potentiation (LTP), which is thought ter. (Neurotransmitters such as seroto to play a role in the storage process. LTP has different properties in the CAl and CA3 nin that carry messages between cells regions of the hippocampus. are called first messengers; other chemi cals known as second messengers relay information within the cell.) These modulatory neurons act on the ditioning represents the simplest exam mal is therefore thought to learn predic sensory neurons from both the siphon ple of learning to associate two events. tive relations between the two stimuli. and the mantle shelf, where they pro In classical conditioning, an ineffective Because Aplysiahas a nervous system duce presynaptic facilitation, that is, stimulus called the conditioned stimu containing only about 20,000 central they enhance tran�mitter release from lus (or more correctly, the to-be-con nerve cells, aspects of classical condi the terminals of the sensory neurons. ditioned stimulus) is repeatedly paired tioning can be examined at the cellular Presynaptic facilitation contributes to a with a highly effective stimulus called level. Aplysia has a number of simple nonassociative form of learning called the unconditioned stimulus. The condi reflexes, of which the gill-withdrawal sensitization, in which an animal learns tioned stimulus initially produces only a reflex has been particularly well stud to enhance a variety of defensive re small response or no response at all; the ied. The animal normally withdraws flex responses after receiving a nox unconditioned stimulus elicits a pow the gill, its respiratory organ, when a ious stimulus [see "Small Systems of erful response without requiring prior stimulus is applied to another part of Neurons," by Eric R. Kandel; SCIENTIFIC conditioning. its body such as the mantle shelf or the AMERICAN, September 1979J. This type As a result of conditioning (or learn fleshy extension called the siphon. Both of learning is referred to as nonasso ing), the conditioned stimulus becomes the mantle shelf and the siphon are in ciative because it does not depend on capable of producing either a larger re nervated by their own populations of pairing between stimuli. sponse or a completely new response. sensory neurons. Each of these popula The findingthat modulatory neurons For example, the sound of a bell (the tions makes direct contact with motor act on both sets of sensory neurons conditioned stimulus) becomes effective neurons for the gill as well as with vari those from the siphon as well as those in eliciting a behavioral response such ous classes of excitatory and inhibitory from the mantle-posed an interesting as lifting a leg only after that sound has interneurons that synapse on the mo question: How is the specific associative been paired with a shock to the leg (the tor neurons. We and our colleagues strengthening of classical conditioning unconditioned stimulus) that invariably Carew and Walters found that even this achieved? Timing turned out to be an produces a leg-lifting response. For con sinlple reflex can be conditioned. important element here. For classical ditioning to occur, the conditioned stim A weak tactile stimulus to one path conditioning to occur, the conditioned ulus generally must be correlated with way, for example, the siphon, can be stimulus generally must precede the the unconditioned stimulus and precede paired with an unconditioned stimulus unconditioned stimulus by a critical it by a certain critical period. The ani- (a strong shock) to the tail. The other and often narrow interval. For condi-

82 SCIENTIFIC AMERICAN September 1992 © 1992 SCIENTIFIC AMERICAN, INC tioning gill withdrawal by tail shock, and decreasing the activity of others. urn enters the terminal, more transmit the interval is approximately 0.5 sec The activation of the protein kinase ter is released. Second, as a result of ond. If the separation is lengthened, in sensory neurons has several impor protein kinase activity, serotonin acts shortened or reversed, conditioning is tant short-term consequences. The pro to mobilize transmitter vesicles from a drastically reduced or does not occur. tein kinase phosphorylates potassium storage pool to the release sites at the In the gill-withdrawal reflex, the spec channel proteins. Phosphorylation of membrane; this facilitates the release ificity in timingresults in part from a these channels (or of proteins that act of transmitter independent of an in convergence of the conditioned and un on these channels) reduces a compo crease in calcium influx. In this action, conditioned stimuli within individu nent of the potassium current that nor cyclic AMP acts in parallel with another al sensory neurons. The unconditioned mally repolarizes the action potential. second messenger, protein kinase C, stimulus is represented in the sensory Reduction of potassium current pro which is also activated by serotonin. neurons by the action of the modulato longs the action potential and thereby Why should the firing of action po ry neurons, in particular the cells in allows calcium channels to be activated tentials in the sensory neurons just be which serotonin is the transmitter. The for longer periods, permitting more cal fore the unconditioned stimulus en conditioned stimulus is represented by cium to enter the presynaptic terminal. hance the action of serotonin?Action activity within the sensory neurons Calcium has several actions within the potentials produce a number of chang themselves. We found that the modula cell, one of which is the release of trans es in the sensory neurons. They allow tory neurons activated by the uncon mitter vesicles from the terminal. When, sodium and calcium to move in and po ditioned stimulus to the tail produce as a result of an increase in the dura tassium to move out, and they change greater presynaptic facilitation of the tion of the action potentials, more calci- the membrane potential. Abrams and sensory neurons if the sensory neurons had just fired action potentials in re sponse to the conditioned stimulus. Ac ENHANCED tion potentials in the sensory neurons PRESYNAPTIC TRANSMITTER that occur just after the tail shock have NEURON RELEASE no effect. This novel property of presynaptic fa CALCIUM

IBOSYL __ cilitation is called activity dependence. ADP-R �->,""'->..,--___ _ TRANSFERASE Activity-dependent facilitation requires CYCLIC � GUANYL YL --" ___-... the same timing on the cellular level as � ,,,-<- GMP CYCLASE -.,- does conditioningon the behavioral lev • el and may account for such condition ing. These results suggest that a cellu lar mechanism of classical conditiOning of the withdrawal reflex is an elabora CALCIUM· tion of presynaptic facilitation, a mech MAGNESIUM RETROGRADE · • anism used for sensitization of the re MESSENGER c1 :• flex. These experiments provided an ini GLUTAMATE --- ..,, t/ (NITRIC OXIDE) tial suggestion that there might be a cellular alphabet for learning whereby the mechanisms of more complex types of learning may be elaborations or com . . binations of the mechanisms of simpler .. types of learning. �ALCIUM • · The next piece in the puzzle of how r · · POSTSYNAPTIC classical conditiOning occurs was to dis NEURON CALMODULIN cover why the firing of action poten tials in the sensory neurons just before l the unconditioned tail stimulus would CA IUM/ 1 TYROSINE enhance presynaptic facilitation. We had CALMODULIN KINASE KINASE� PROTEIN / previously found that when serotonin KIN EC is released by the modulatory neurons � in response to tail shock, it initiates � t a series of biochemical changes in the sensory neurons [see illustration on page NITRIC OXIDE SYNTHASE 81]. Serotonin binds to a receptor that activates an enzyme called adenylyl cy clase. This enzyme in turn converts ATP, one of the molecules that pro NITRIC OXIDE vides the energy needed to power the various activities of the cell, into cyclic IN LONG-TERM POTENTIATION the postsynaptic membrane is depolarized by the AMP. CycliC AMP then acts as a second actions of the non-NMDA receptor channels. The depolarization relieves the magne messenger (serotonin is the first mes sium blockade of the NMDA channel, allowing calcium to flow through the channel. senger) inside the cell to activate an The calcium triggers calcium-dependent kinasesthat lead to the induction of LTP. other enzyme, a protein kinase. Kinas The postsynaptic cell is thought to release a retrograde messenger capable of pene es are proteins that phosphorylate (add trating the meIl)brane of the presynaptic cell. This messenger, which may be nitric a phosphate group to) other proteins, oxide, is believed to act in the presynaptic terminal to enhance transmitter (gluta thereby increasing the activity of some mate) release, perhaps by activating guanylyl cyclase or ADP-ribosyl transferase.

SCIENTIFIC AMERICAN September 1992 83 © 1992 SCIENTIFIC AMERICAN, INC Kandel found that the critical function gle-gene mutants have been discovered because, unlike classical conditioning, of the action potential for activity de that are deficient in learning. One such expliCit learning is often most success pendence was the movement of calcium mutant, called rutabaga, has been stud ful when the two events that are associ into the sensory neurons. Once in the ied by William G. Quinn of the Massa ated occur simultaneously. For example, cell, calcium binds to a protein called chusetts Institute of Technology and we recognize the face of an acquaintance calmodulin, which amplifies the acti Margaret Uvingstone of Harvard Univer most easily when we see that acquain vation of the enzyme adenylyl cyclase sity and by Yadin Dudai of the Weiz tance in a speCific context. The stimuli by serotonin. When calcium/calmodu mann Institute in Israel. The gene encod of the face and of the setting actsimulta lin binds to the adenylyl cyclase, the en ing the defective protein in this mutant neously to help us recognize the person. zyme generates more cyclic AMP . This has now been shown to be a calcium/ As we have seen, explicit learning capacity makes adenylyl cyclase an im calmodulin-dependent adenylyl cyclase. in humans requires the temporal lobe. portant convergence site for the condi As a result of the mutation in ruta Yet it was unclear at first how extensive tioned and the uncoriditioned stimuli. baga, the cyclase has lost its ability to the bilateral lesion in the temporal lobe Thus, the conditioned and the uncon be stimulated by calcium/calmodulin. had to be to interfere with memory stor ditioned stimuli are represented within Moreover, Ronald L. Davis and his col age. Subsequent studies in humans and the cell by the convergence of two dif leagues at Cold Spring Harbor Labora in experimental animals by Mortimer ferent signals (calcium and serotonin) tory have found that this form of the Mishkin of the National Institutes of on the same enzyme. The O.S-second adenylyl cyclase is enriched in the mush Health and by Squire, David G. Amaral interval between the two stimuli essen room bodies, a part of the fly brain criti and Stuart Zola-Morgan of the Univer tial for learning in the gill-withdrawal cal for several types of associative learn sity of California at San Diego help to reflex may correspond to the time dur ing. Thus, both cell biological studies in answer the question. They suggest that ing which calcium is elevated in the pre Aplysia and genetic studies in Drosophi one structure within the temporal lobe synaptic terminal and binds to calmod la point to the Significance of the cyclic particularly critical for memory storage ulin so as to prime the adenylyl cyclase AMP second-messenger system in cer is the hippocampus. And yet lesions of to produce more cyclic AMP in response tain elementary types of implicit learn the hippocampus interfere only with the to serotonin. ing and memory storage. storage of new memories: patients like Activity-dependent amplification of H.M. still have a reasonably good mem the cyclic AMP pathway is not unique hat about expliCit forms of ory of earlier events. The hippocampus to the gill- or tail-withdrawal reflexes of learning? Do these more com appears to be only a temporary depos Aplysia. Genetic studies in the fruit fly plex types of associative learn itory for long-term memory. The hip Drosophila have implicated a similar Wing also have cellular representations for pocampus processes the newly learned molecular mechanism for conditioning. associativity? If so, they must differ from information for a period of weeks to Drosophila can be conditioned, and sin- the mechanisms for impliCit learning, months and then transfers the infor mation to relevant areas of the cerebral cortex for more permanent storage [see "Brain and Language," by Antonio R. Da ACTIVITY-DEPENDENT FACILITATION IN APLYSIA masio and Hanna Damasio, page 88]. As discussed by Patricia S. Goldman-Rakic, MODULATORY the memory stored at these different NEURON cortical sites is then expressed through the working memory of the prefrontal SENSORY MOTOR cortex [see "Working Memory and the NEURON NEURON Mind," page 110]. In 1973 Timothy Bliss and Terje L0mo, working in Per Andersen's labora tory in Oslo, Norway, first demonstrated that neurons in the hippocampus have remarkable plastic capabilities of the d:=== kind that would be required for learn LONG-TERM POTENTIATION ====:(] ing. They found that a brief high-fre IN THE HIPPOCAMPUS�::=z= quency train of action potentials in one 8======(] of the neural pathways within the hip pocampus produces an increase in syn aptic strength in that pathway. The in crease can be shown to last for hours in an anesthetized animal and for days :=:=====: and even weeks in an alert, freely mov � � ing animal. Bliss and L0mo called this strength ening long-term potentiation (LTP). Lat (] er studies showed that LTP has differ PRESYNAPTIC NEURON 8 POSTSYNAPTIC NEURON ent properties in different types of syn apses within the hippocampus. We will focus here on an associative type of po ASSOCIATIVE PROCESSES believed to contribute to learning in Aplysia and in the hippocampus of mammals may share similar mechanisms. Both may involve a tentiation that has two interrelated char modulatory substance that produces activity-dependent enhancement of transmit acteristics. First, the associativity is of ter release from the presynaptic neuron. Stripes denote neurons in which coinci the Hebbian pre-post form: for facilita dent activity must occur to produce the associative change. tion to occur, the contributing presyn-

84 SCIENTIFIC AMERICAN September 1992 © 1992 SCIENTIFIC AMERICAN, INC aptic and postsynaptic neurons need to be active simultaneously. Second, and Representation of the Surface of the Body in the Cortex as a result, the long-term potentiation shows specificity: it is restricted in its he homunculus ("little man") is action to the pathway that is stimulated. Ta traditional way of illustrating Why is simultaneous firing of the how the surface of the body is repre presynaptic and postsynaptic cells nec sented in the somatosensory cortex. essary for long-term potentiation? The Larger areas of the cortex are devot major neural pathways in the hippo ed to parts of the body that have campus use the amino acid glutamate greater sensitivity, such as the fin as their transmitter. Glutamate produc gers and lips. es LTP by binding to glutamate recep Recently the effects of sensitivi tors on its target cells. It turns out that ty training have been shown in the there are two relevant kinds of gluta owl monkey. The monkey's digits are mate receptors: the NMDA receptors represented in areas 3b and 1 of (named after the chemical N-methyl D the somatosensory cortex (a). The diagrams (b and d) outline the regions aspartate, which also binds to these re that map the surface of the digits of an adult monkey (c) before and after ceptors) and the non-NMDA receptors. training. During training the monkey rotated a disk for one hour a day, us Non-NMDA receptors dominate most ing only digits 2, 3 and occasionally 4. After three months of this activ synaptic transmission because the ion ity, the area representing the stimulated fingers in the brain had increased channel associated with the NMDA re substantially. ceptor is usually blocked by magnesium. It becomes unblocked only when the a c d postsynaptic cell is depolarized. More 3 over, optimal activation of the NMDA re ceptor channel requires that the two sig nals-glutamate binding to the receptor and depolarization of the postsynaptic cell-take place simultaneously. Thus, the NMDA receptor has associative or coincidence-detecting properties much as does the adenylyl cyclase. But its temporal characteristics, a requirement for simultaneous activation, are better BEFORE AFTER suited for expliCit rather than impliCit DIFFERENTIAL DIFFERENTIAL STIMULATION STIMULATION forms of learning. Calcium influx into the postsynaptic cell through the unblocked NMDA re ceptor channel is critical for long-term tists. Ever since the great Spanish anato dence that nitric oxide may be such a potentiation, as was first shown by Gary mist Santiago Ramon y Cajal first enun retrograde messenger: Thomas J O'Dell Lynch of the University of California at ciated the principle of dynamic polari and Ottavio Arancio in our laborato Irvine and by Roger A. Nicoll and Robert zation, every chemical synapse studied ry, Erin M. Schuman and Daniel Madison S. Zucker and colleagues at the Univer has proved to be unidirectional. Infor of Stanford University, Paul F. Chapman sity of California at San Francisco. Cal mation flows only from the presynaptic and his colleagues at the University of cium initiates LTP by activating at least to the postsynaptic cell. In long-term po Minnesota School of Medicine and Georg three different types of protein kinases. tentiation, a new principle of nerve cell Bbhme and his colleagues in France. In The induction of LTP appears to de communication seems to be emerging. hibiting the synthesis of nitric oxide in pend on postsynaptic depolarization, The calcium-activated second-messen the postsynaptic neuron or absorbing leading to the influx of calcium and the ger pathways, or perhaps calcium act nitric oxide in the extracellular space subsequent activation of second-mes ing directly, seem to cause release of a blocks the induction of LTP, whereas ap senger kinases. For the maint enance of retrograde plasticity factor from the ac plying nitric oxide enhances transmitter LTP, on the other hand, several groups tive postsynaptic cell. This retrograde release from presynaptic neurons. of researchers have found that enhance factor then diffusesto the presynaptic In the course of studying the effects ment of transmitter from the presynap terminals to activate one or more sec of applying nitric oxide in slices of hip tic terminal is involved. These workers ond messengers that enhance transmit pocampus, we and Scott A. Small and include Bliss and his colleagues, John ter release and thereby maintain L TP Min Zhuo made a surprising finding:we Bekkers and Charles Stevens of the Salk [see illustration on page 83]. discovered that nitric oxide produces Institute and Roberto Malinow and Rich Unlike the presynaptic terminals, L TP only if it is paired with activity in ard Tsien of Stanford University. which store transmitter in vesicles and the presynaptic neurons, much as is the If the induction of LTP requires a post release it at specialized release sites, the case in activity-dependent presynaptic synaptic event (calcium influx through postsynaptic terminals lack any special facilitation in Aplysia. Presynaptic activ the NMDA receptor channels) and main release machinery. It therefore seemed ity, and perhaps calcium influx, appears tenance of L TP involves a presynaptic attractive to posit that the retrograde to be critical for nitric oxide to produce event (increase in transmitter release), messenger may be a substance that potentiation. These experiments suggest then, as first proposed by Bliss, some rapidly diffuses out of the postsynaptic that long-term potentiation uses a com message must be sent from the postsyn cell across the synaptic cleft and into bination of two independent, associative, aptic to the presynaptic neurons-and the presynaptic terminal. By 1991 four synaptic learning mechanisms: a Heb that poses a problem for neuroscien- groups of researchers had obtained evi- bian NMDA receptor mechanism and a

SCIENTIFIC AMERICANSepte mber 1992 85 © 1992 SCIENTIFIC AMERICAN, INC non-Hebbian, activity-dependent, presyn fail to learn the task. These experiments distinctive modification of brain archi aptic facilitating mechanism. Accord suggest that NMDA receptor mecha tecture, along with a unique genetic ing to this hypothesis, the activation of nisms in the hippocampus, and perhaps makeup, contributes to the biological NMDA receptors in the postsynaptic LTP, are involved in spatial learning. basis for the expression of individuality. cells produces a retrograde signal (nitric This view is best demonstrated in a oxide). The signal then initiates an activ aving now considered the mech study by Merzenich, in which he en ity-dependent presynaptic mechanism, anisms throughwhich learning couraged a monkey to touch a rotating which facilitates the release of transmit can produce changes in nerve disk with only the three middle fingers ter from the presynaptic terminals. cells, we are faced with a final set of of its hand. After several thousand disk What mightbe the functional advan quHestions. What are the mechanisms rotations, the area in the cortex devoted tage of combining two associative cellu whereby the synaptic changes produced to the three middle fingers was expand lar mechanisms, the postsynaptic NMDA by expliCit and implicit learning en ed at the expense of that devoted to the receptor and the activity-dependent pre dure? How is memory maintained in other fingers[see illustration on preced synaptic facilitation, in this way? If pre the long term? ing page]. Practice, therefore, can lead to synaptic facilitation is produced by a dif Experiments in both Aplysia and changes in the cortical repres entation fusible substance, that substance could, mammals indicate that explicit and im of the most active fingers. What mecha in theory, findits way into neighbor plicit memory storage proceed in stag nisms underlie the changes? Recent evi ing pathways. In fa ct, studies by Tobi es. Storage of the initial information, a dence indicates that the cortical con as Bonhoeffer and his colleagues at the type of short-term memory, lasts min nections in the somatosensory system Max Planck Institute for Brain Research utes to hours and involves changes in are constantly being modified and up in Frankfurt indicate that LTP initiat the strength of existing synaptic connec dated on the basis of correlated activi ed in one postsynaptic cell spreads to tions (by means of second-messenger ty, using a mechanism that appears sim neighboring postsynaptic cells. Activity mediated modifications of the kind we ilar to that which generates LTP. dependence of presynaptic facilitation have discussed). The long-term chang Indeed, as we have learned from Car could be a way of ensuring that only spe es (those that persist for weeks and la ]. Shatz [see "The Developing Brain," cific presynaptic pathways-those that months) are stored at the same site, but page 60], early results from cell bio are active-are potentiated. Any inactive they require something entirely new: the logical studies of development suggest presynaptic terminals would not be af activation of genes, the expres sion of that the mechanisms of learning may fected [see illustration on page 84]. new proteins and the growth of new carry with them an additional bonus. The changes in synapses that are connections. In Aplysia, Craig H. Bailey, There is now reason to believe that the thought to contribute to these instanc Mary C. Chen and Samuel M. Schacher fine-tuning of connections during late es of impliCit and expliCit learning raise and their colleagues at Columbia Univer stages of development may require an a surprising reductionist possibility. The sity and Byrne and his colleagues at the activity-dependent associative synaptic fact that associative synaptic changes University of Texas Health Science Cen mechanism perhaps similar to LTP. If do not require complex neural networks ter have found that stimuli that pro that is also true on the molecular level suggests there may be a direct corre duce long-term memory for sensitiza if learning shares common molecular spondence between these associative tion and classical conditioninglead to mechanisms with aspects of develop forms of learning and basic cellular an increase in the number of presynap ment and growth-the study of learning properties. In the cases that we have tic terminals. Similar anatomic changes may help connect cognitive psychology reviewed, the cellular properties seem occur in the hippocampus after LTP. to the molecular biology of the organ to derive in turn from the properties of If long-term memory leads to ana ism more generally. This broad biologi speCific proteins-the adenylyl cyclase tomic changes, does that imply that cal unification would accelerate the de and the NMDA receptor-that are capa our brains are constantly changing ana mystification of mental processes and ble of responding to two independent tomically as we learn and as we for position their study squarely within the signals, such as those from the condi get? Will we experience changes in our evolutionary framework of biology. tioned stimulus and the unconditioned brain's anatomy as a result of reading stimulus. Of course, these molecular as and remembering this issue of Scien

sociative mechanisms do not act in iso tific American? FURTHER READING lation. They are embedded in cells that This question has been addressed by AMNESIA FOLLOWING OPERATION ON THE have rich molecular machinery for elab many investigators, perhaps most dra TEMPORAL LOBES. Brenda Milner in Am orating the associative process. And the matically by Michael Merzenich of the nesia: Clinical, Psychological and Medi cells, in turn, are embedded in complex University of California at San Francis colegal Aspects. Edited by C.W.M. Whitty neural networks with considerable re co. Merzenich examined the represen and O. L. Zangwill. Butterworths, 1966. dundancy, parallelism and computation tation of the hand in the sensory area A CELLULAR MECHANISM OF CLASSICAL CONDITIONING IN APLYSIA: ACTIVITY al power, adding substantial complexity of the cerebral cortex. Until recently, DEPENDENT AMPLIFICATION OF PRESYN to these elementary mechanisms. neuroscientists believed this represen APTIC FACILITATION. R. D. Hawkins, T. The finding that LTP occurs in the tation was stable throughout life. But W. Abrams, T. J. Carew and E. R. Kandel hippocampus, a region known to be Merzenich and his colleagues have now in SCience, Vo L 21 9, pages 400-405; significant in memory storage, made demonstrated that cortical maps are January 28, 1983. researchers wonder whether LTP is in subject to constant modification based . THE CURRENT ExCITEMENT IN LONG-TERM volved in the process of storing mem on use of the sensory pathways. Since POTENTIATION. R. A. Nicoll, J. A. Kauer C. ories in this area of the brain. Evidence all of us are brought up in somewhat and R. Malenka in Neuron, Vol. 1, No. that it is has bee n provided by Rich different environments, are exposed to 2, pages 97-103; April 1988. MEMORY AND THE HIPPOCAMPUS: A SYN ard Morris and his colleagues at the Uni different combinations of stimuli and THESIS FROM FINDINGS WITH RATS, MON versity of Edinburgh Medical School by are likely to exercise our sensory and KEYS, AND HUMANS.Larry R. Squire in means of a spatial memory task. When motor skills in different ways, the ar Psychological Review, Vol. 99, No. 2, NMDA receptors in the hippocampus chitecture of each of our brains will be pages 195-231; April 1992. are blocked, the experimental animals modified in slightly different ways. This

o Ye s, I would like to receive the single topic issue, Mind andBrain. MBSA9

(please print)

Apt.

State Zip Photocopies of this order form are acceptable.

Please send __copies @ $4.95 $ ___ _ Now available Add $ 1.00 per copy $___ _ in bulkfo r for postage and handling. industry and education Total enclosed $--- Deduct 20% on orders of 50 or more copies. Postage and handling paid by SCIENTIFIC AMERICAN. on orders SAVE of ten or more copies. Please make check or money order payable to SCIENTIFIC AMERICAN. on bulk20% orders Send orders to: SCIENTIFIC AMERICAN, Dept. M&B, of 50 or more 415 Madison Avenue, New York, NY10 017, USA