The past, the future and the biology

of memory storage

Eric R. Kandel and Christopher Pittenger

Howard Hughes Medical Institute, Center for Neurobiology and Behavior, Columbia University, College of Physicians and Surgeons,
1051 Riverside Drive, New York, NY 10032, USA ([email protected])

We here brie£y review a century of accomplishments in studying memory storage and delineate the two
major questions that have dominated thinking in this area: the systems question of memory, which
Downloaded from https://royalsocietypublishing.org/ on 26 December 2024

concerns where in the brain storage occurs; and the molecular question of memory, which concerns the
mechanisms whereby memories are stored and maintained. We go on to consider the themes that memory
research may be able to address in the 21st century. Finally, we re£ect on the clinical and societal import
of our increasing understanding of the mechanisms of memory, discussing possible therapeutic approaches
to diseases that manifest with disruptions of learning and possible ethical implications of the ability,
which is on the horizon, to ameliorate or even enhance human memory.
Keywords: explicit; implicit; memory

These two independent uni¢cations stand at the

1. INTRODUCTION
extremes of the biological sciences: the one at the inter-
In the past several decades of this century we have face between biology and chemistry; the other at the
witnessed a remarkable increase in the explanatory power interface between biology and psychology. This raises a
of biology, including the biology of the brain. The ability question: To what degree can these two disparate strands
of biology to address central issues of brain function will be brought together? Can molecular biology enlighten
only increase in the 21st century and is likely to have a the study of mental processes as it has enlightened the
broad impact on many aspects of our lives. As a result, study of development and oncogenesis? Can we anticipate
when the intellectual historians of The Royal Society look an even broader synthesis in the 21st century, a synthesis
back on these decades, they are likely to acknowledge ranging from molecules to mind? In this article we
that the deepest insights into the nature of mental outline the possibility of a molecular biology of cognition,
processes will not have come from the disciplines and suggest that it will occupy centre stage in the early
traditionally concerned with mind. They will have come part of the 21st century. We outline this development by
not from philosophy, from the arts, or even from using as an example the study of memory.
psychology or psychoanalysis, but from biology. This is We begin with a historical perspective because many of
because in the past two decades biology has participated the themes that dominate current research on memory ö
not simply in one but in two major uni¢cations of thought including the distinction between systems and molecular
which bear on our understanding of mind. approaches to memory ö emerge best in a historical over-
First, there has been a remarkable scienti¢c synthesis, view. We will then describe more recent molecular bio-
achieved through molecular biology, that has brought logical investigations of memory. Here we will focus in
together the disciplines of cell biology, biochemistry, particular on one component of the memory process: the
developmental biology and oncogenesis. This uni¢cation switch from short- to long-term memory. This component
derives from major advances in our understanding of the can in fact now be analysed by combining cognitive
gene, which have revealed how its structure determines psychology with modern molecular biology. Finally, we
heredity and how its regulation determines development will suggest some possible future directions in memory
and function. These remarkable insights have given us a research, focusing on the drawing together of strands of
marvellous sense of the conservation between di¡erent research that have historically been separated.
cells in any one organism and di¡erent organisms across
evolution.
2. THE PROBLEM OF MEMORY HAS A SYSTEMS
Second, there has been a parallel uni¢cation between
COMPONENT AND A MOLECULAR COMPONENT
neurobiology, the science of the brain, and cognitive
psychology, the science of the mind. This second The work of Ramon y Cajal at the beginning of the
uni¢cation is far less mature than that brought about by century (Cajal 1893) and of Donald Hebb in 1949 (Hebb
molecular biology, but it is potentially equally profound, 1949) established a useful conceptual framework for the
for it promises to provide us with a new framework for study of memory, based on the idea that memory is stored
the analysis of a variety of mental functions, such as as changes in the strength of speci¢c synaptic connec-
perception, action, language, learning and memory. tions. This framework divides the study of memory into

Phil. Trans. R. Soc. Lond. B (1999) 354, 2027^2052 2027 & 1999 The Royal Society
 2028 E. R. Kandel and C. Pittenger The past, the future and the biology of memory storage

two components: the systems problem and the molecular (a)

problem. The systems problem of memory is concerned
with where in the brain memory is stored and how neural
circuits work together to create, process and recall
memories. The molecular problem of memory is
concerned with the mechanisms whereby synapses change
and information is stored. Most early work on memory
focused on the systems problem, focusing on the question
`Where is memory stored?' We therefore begin our histor-
ical perspective with this question.

3. THE SYSTEMS COMPONENT OF MEMORY

(a) Early attempts to localize mental function in the
brain
The attempt to understand where memory is stored
Downloaded from https://royalsocietypublishing.org/ on 26 December 2024

goes back at least 200 years and is part of a larger

question: can any mental process be localized to a speci¢c
region of the brain? The modern debate about the
localization of brain function began in the early part of
the 19th century with Franz Joseph Gall, a German physi-
cian and neuroanatomist who taught for a long while at
the University of Vienna. Gall started the school of
`organology' (later `phrenology'). He made two remark-
able and enduring contributions to our understanding of (b)
brain (Gall & Spurzheim 1810). First, he appreciated that
all mental processes are biological and arise from the
brain. He therefore opposed the Cartesian mind^brain
dualism of his day and argued for a materialist view of
the mind (a view that later led to his expulsion from the
University of Vienna). Second, he posited that mental
functions are localized to di¡erent regions within the
brain. Gall argued that the brain (and speci¢cally the
cerebral cortex) is not homogeneous but is subdivided
into functionally distinct regions, each of which serves as
the organ for one or another mental function. Based on
contemporary academic psychology, Gall ascribed each of
35 mental faculties to a speci¢c cortical region. Thus,
Gall argued that even the most abstract and complex of
human behaviours, such as cautiousness, secretiveness,
hope, sublimity and parental love, are mediated by
di¡erent, individual cortical regions. Gall therefore was
the ¢rst strong proponent of the localization of function
within the cortex, and he thereby initiated a debate that
persisted for the next century (¢gure 1) (Cooter 1984; Figure 1. (a) Franz Joseph Gall theorized that cortical
Young 1970). regions have specialized function. (b) Gall divided the cortex
While Gall's theory of localization was prescient and is into 35 regions, each dedicated to a speci¢c psychological
echoed in the functionally identi¢ed areas that tile much characteristic. Taken from Kandel et al. (1995).
of the cortex in today's neuroanatomy texts, his theory
was £awed in two ways. First, as we now appreciate, the well endowed with speci¢c faculties, Gall sought to
35 abstract properties for which Gall was seeking to iden- identify the centres for those faculties. This led to `phre-
tify cortical organs, such as parental love and conscien- nology,' with its preoccupation with the shape of the skull
tiousness, are too complex to be represented by individual as indicating the structure of personality and character.
discrete areas. Second, his method for identifying the In the late 1820s, Gall's ideas were subjected to experi-
function of speci¢c areas proved simplistic. Gall mental analysis by the French experimental neurologist
distrusted lesion experiments and therefore ignored clin- Pierre Flourens, who readily showed that the bumps on a
ical ¢ndings. He rather was guided by the theory that the person's skull bear little relation to the shape of the brain
size of a given area of the brain is related to usage of that underneath (¢gure 2). Flourens removed cortical areas
area by the mental faculty which it represents. Exercise of corresponding to certain of the functional areas de¢ned
a given mental faculty causes the corresponding critical by Gall from experimental animals and failed to ¢nd any
brain region to grow. This growth in a brain region of the de¢cits that Gall's theories predicted. In fact,
would in turn cause the overlying skull to protrude. By Flourens was unable to identify any speci¢c de¢cits in
examining the bumps and ridges on the skulls of people behaviour that were associated with speci¢c lesions of the

Phil. Trans. R. Soc. Lond. B (1999)

 The past, the future and the biology of memory storage E. R. Kandel and C. Pittenger 2029
Downloaded from https://royalsocietypublishing.org/ on 26 December 2024

Figure 2. Pierre Flourens tested Gall's theories and concluded

that the cortex is equipotential.

neocortex. He thus concluded that the nervous system in Figure 3. Pierre-Paul Broca concluded from patients with
localized brain lesions that language is localized within the
general, and the cerebral hemispheres in particular, are
cortex.
equipotential: any part of the cerebral cortex participates
in and is able to perform all the functions of the whole hemispheres, although apparently symmetrical, have
(Flourens 1893). Injury to any part would, according to slightly di¡erent functions.
this theory, not a¡ect any one capacity more than others. A decade later, in 1876, Carl Wernicke, a German
The debate between the advocates of functional locali- neurologist (¢gure 4), described a second type of aphasia
zation and the advocates of an equipotential view of (Wernicke's aphasia) that is in a way the opposite of
cortex coloured thinking about the brain in the ¢rst half Broca's: an impairment not of the production of speech
of the 19th century. This debate was largely resolved in but of comprehension. Wernicke found that this syndrome
the second half of the century by two great neurologists, was caused by a lesion in the posterior superior portion of
Paul Broca and Carl Wernicke. In studies of patients with the temporal lobe of the left hemisphere, a lesion distinct
speci¢c language de¢cits (aphasias), Broca and Wernicke from that described by Broca, in an area we now call
identi¢ed cortical areas whose lesion produced remark- Wernicke's area.
ably speci¢c disorders of language, thereby demonstrated Taking his ¢ndings together with those of Broca,
convincingly the localization of at least some higher Wernicke put forward a theory of how the cortex is orga-
functions within the cerebral cortex. nized for language that, although simpler than our
current understanding, is still central to how we view the
(b) Broca, Wernicke and the localization of language brain. Wernicke proposed that any complex behaviour
In 1861, Pierre-Paul Broca, a French neurologist much requires the activity not of one but of a number of
in£uenced by Gall (¢gure 3), described the ¢rst of nine di¡erent brain areas, and that these areas are intercon-
patients who su¡ered from a language impairment (now nected in various ways. Mental activity is not unitary or
known as Broca aphasia) in which they could understand seamless as might intuitively appear to be the case, but
language but could not speak £uently. These defects were can be broken down into multiple components, and each
speci¢c to the expression of language rather than to component can be assigned to a more or less speci¢c
motor control of the vocal tract, as the patients could brain region, much as the organologists had insisted.
hum or whistle a tune but could not write £uently. Post- However, these di¡erent specialized areas do not function
mortem examination of the patients' brains revealed in by themselves but as part of large, interconnected
each case a lesion in the posterior region of the frontal networks. By application of this model Wernicke
lobe (a region of cortex now called Broca's area). Thus, predicted, correctly, the possibility of a third kind of
for the ¢rst time, Broca was able to assign a well-de¢ned aphasia, conduction aphasia, resulting from lesion not of
higher function to a speci¢c region of cortex (Broca Broca's area or Wernicke's area but of the ¢bres (the
1865). Since all of these lesions were in the left arcuate fasciculus) passing between the two. Thus, while
hemisphere, Broca was also able to establish that the two speci¢c functions are localized as Gall had insisted, the

Phil. Trans. R. Soc. Lond. B (1999)

 2030 E. R. Kandel and C. Pittenger The past, the future and the biology of memory storage
Downloaded from https://royalsocietypublishing.org/ on 26 December 2024

Figure 5. Karl Lashley investigated the e¡ects of cortical

lesions on memory in rats and determined that no portion of
cortex is specialized for memory.

studies. In 1938, the neurosurgeon Wilder Pen¢eld

Figure 4. Carl Wernicke developed the idea of interconnected (¢gure 6), working at the Montreal Neurological Institute,
specialized areas of cortex that is still the dominant framework
developed methods for the surgical treatment of focal
today.
epilepsy, a form of epilepsy in which seizure is restricted
to a relatively small region of the cortex. To functionally
function of the brain as a whole requires distributed map the areas surrounding the epileptic centre so as to
processing somewhat reminiscent of that propounded by avoid later damage to critical areas, such as Broca's and
Flourens (Wernicke 1908). Wernicke's model of a distrib- Wernicke's, Pen¢eld electrically stimulated the cortical
uted network of specialized areas has emerged as a domi- surface. Because the brain contains no pain receptors, the
nant theme in the study of the brain. patients were unanaesthetized and could report what
they experienced during stimulation. In this way Pen¢eld
(c) Can memory storage be localized to speci¢c studied over 1000 patients and mapped out in each of
regions of the brain? them most of the exposed cortical surface. On rare occa-
The ¢nding that the language can be localized within sions during such mappings, Pen¢eld found a region of
the brain led to the hunt for other areas concerned with temporal cortex where stimulation gave rise to speci¢c
speci¢c higher functions. Areas concerned with motor experiential responses, memory-like perceptions that the
control and with each of the senses were soon identi¢ed. patients could describe. Pen¢eld concluded that portions
It was only a matter of time before e¡orts to localize of the temporal lobe were speci¢cally involved in memory
cognitive function would turn to memory (Ferrier 1890; (Pen¢eld & Perot 1963).
Jackson 1884). However, attempts in the ¢rst half of the More conclusive evidence for the involvement of
20th century to localize memory failed. The dominant temporal lobe structures in memory came in 1957, when
¢gure in this period was Karl Lashley, Professor of William Scoville, a neurosurgeon in£uenced by Pen¢eld,
Psychology at Harvard (¢gure 5). Lashley began the and Brenda Milner, a psychologist and long-term
experimental search for the locus of memory storage by collaborator of Pen¢eld's (¢gure 7), reported the now
training rats on speci¢c memory tasks, systematically famous case of H.M. (Scoville & Milner 1957). At age 9,
removing portions of cortex, and then testing the rats' H.M. sustained a head injury after being hit by a bicycle;
recall. He repeatedly failed to ¢nd any particular brain over the next 18 years he su¡ered progressive seizures
region that was special to or necessary for the storage of until he was completely incapacitated. As a last resort,
memory. On the basis of these ¢ndings, Lashley formu- H.M. underwent complete bilateral removal of the
lated the law of mass action, according to which the medial temporal lobe (where his seizures initiated). The
extent of a memory de¢cit is correlated with the size of a surgery relieved his epilepsy, but he was left with a
cortical lesion but not with the speci¢c site of that lesion profound memory de¢cit: from the time of his surgery
(Lashley 1929). This law was reminiscent of the views of until this day he has been unable to form any new
Flourens a century earlier. memories of people, facts or events.
The ¢rst clear evidence for the localization of memory Brenda Milner studied H.M. and demonstrated that
came not from experimental animals but from clinical structures in the medial temporal lobe that Scoville had

Phil. Trans. R. Soc. Lond. B (1999)

 The past, the future and the biology of memory storage E. R. Kandel and C. Pittenger 2031
Downloaded from https://royalsocietypublishing.org/ on 26 December 2024

Figure 6. Wilder Pen¢eld, a neurosurgeon working at McGill

University, used stimulation of the cortical surface of epilepsy
patients to map functional areas.
Figure 7. Brenda Milner's seminal studies of the patient H.M.
removed are specialized for memory. Her further studies conclusively identi¢ed the medial temporal lobe as a region
with H.M. not only controverted Lashley's in£uential specialized for memory.
views but also cast new light on the systems problem of
memory: there are, within the brain, multiple, function-
ally specialized memory systems. brain is the distinction between conscious and uncon-
scious memories.
(d) Memory is not a unitary faculty of mind: there Thus, even prior to 1960 fractionation of memory had
are multiple memory systems already been proposed on the basis of content, function
The idea that there may be multiple memory systems and temporal pro¢le. Nevertheless, the concept of
is old, but it did not enter mainstream psychological multiple memory systems only drew the attention of the
thinking until Milner's work in the 1960s. In the early scienti¢c community with the studies of H.M. After the
parts of the 19th century, the French philosopher Maine profound nature of his memory de¢cit was recognized,
de Biran argued that memory can be subdivided into Milner made the further discovery that despite his
di¡erent systems for ideas, feelings and habits (Copple- impairment he could learn a surprising amount of new
stone 1977). William James emphasized the idea that information. First, H.M. was found to have perfectly
memory has distinct temporal phases (James 1890). In good short-term memory: he could accurately repeat
the 20th century, Henri Bergson developed the distinc- back a telephone number and he can carry on a normal
tion between conscious memory and habit (Bergson conversation. It is only when he is distracted from the
1913). In 1949, the British philosopher Gilbert Ryle topic or task at hand that his memory de¢cit reveals itself.
proposed a similar distinction between `knowing that' Thus, the temporal lobe structure, which H.M. lacks, is
(conscious recall of knowledge for facts and events) and not required for short-term (or `working') memory. This
`knowing how' (knowledge of performance or skills ¢nding validated the early distinction between short- and
without recourse to conscious awareness) (Ryle 1949). A long-term memory.
similar distinction was made by the psychologist Jerome Second, Milner found that H.M. has reasonably good
Bruner, who termed `knowing that' a memory with long-term memory for events prior to his operation. He
record and `knowing how' a memory without record maintains his overall intelligence and has good
(Milner et al. 1998). The de¢ning characteristic of command of English. He remembers events from his
memories with record is the ability to summon up a childhood and adult life before his surgery. There is a
more or less detailed conscious recollection of facts and period of retrograde amnesia for events shortly before
events about persons, places and objects. The de¢ning the surgery, but for the most part H.M.'s symptoms
characteristic of a memory without record, by contrast, revealed that the medial temporal lobe is not the ulti-
is a change in the way an organism responds to a situa- mate storage site for previously acquired knowledge. This
tion or a stimulus, without access to the speci¢c circum- ¢nding supports the idea implied in Wernicke's model
stance under which the memory was formed. The idea of that knowledge is ultimately stored in whatever area of
distinct memory systems is, in a sense, implicit in Freud's the cortex processes the relevant sort of information (see,
psychoanalytic writing. Central to Freud's view of the for example, Zeki 1993).

Phil. Trans. R. Soc. Lond. B (1999)

 2032 E. R. Kandel and C. Pittenger The past, the future and the biology of memory storage

Third, and most surprising, H.M. is able to form intact semantic memory but no episodic memory
certain types of long-term memory. In 1962, Milner and (Tulving et al. 1988).
the psychologist Suzanne Corkin found that H.M. was Implicit memory similarly is not a single memory
able to acquire new motor skills (speci¢cally, the ability system but a large collection of systems, as is made more
to trace a complex ¢gure in a mirror) (Corkin 1965). clear by Squire's designation non-declarative memory
When asked, he would deny that he had encountered or (Squire & Kandel 1999). Fractionation of implicit
practised the task before; but his performance showed memory is made on the basis of speci¢c learning tasks,
unequivocal improvement over time. This ¢nding showed most of which can be assigned to speci¢c systems. Habi-
that learning of this skill is preserved after severe tuation, sensitization and classical conditioning involve
temporal lobe damage and in the presence of profound sensory and motor systems recruited for the speci¢c task
amnesia for facts and events, and thus it demonstrated for tested. These are the most elementary forms of learning
the ¢rst time a fractionation of memory on the basis of and can be studied e¡ectively even in invertebrates.
content rather than just duration. Milner and Corkin Motor learning involves the cerebellum (Thompson et al.
thus validated the distinction between conscious memory 1998). Emotionally based memory involves the amygdala
and habit propounded by Bergson 50 years earlier and by (LeDoux 1996). Operant conditioning, the association of
Ryle in 1949. a stimulus or action with a reward, involves the ventral
Downloaded from https://royalsocietypublishing.org/ on 26 December 2024

The learning tasks that amnesic patients like H.M. are striatum, or nucleus accumbens, which is also the struc-
capable of mastering have several things in common. They ture most often implicated in drug addiction (Robbins &
have an automatic quality, and the formation or expression Everitt 1999). Acquisition of certain forms of habit, by
of the memories is not dependent on awareness or cogni- contrast, requires the dorsal striatum (Graybiel 1998).
tive processes such as comparison and evaluation. This Finally, the various sensory cortices are involved in the
type of memory typically builds up slowly over many phenomenon of priming (Tulving et al. 1982). Implicit
trials and is expressed primarily by improved performance memory is therefore a heterogeneous collection of
on certain tasks. The psychologist Lawrence Weiskrantz memory functions, each performed by a di¡erent brain
has noted that the spared learning skills are re£exive structure, adapted to a di¡erent role, and dissociable from
rather than re£ective ö typically the patient need only the others by lesions or clever task design.
produce a physical response to a stimulus or cue.
This distinction was soon validated on normal subjects.
4. THE MOLECULAR PROBLEM OF MEMORY
Larry Squire has framed the distinction particularly well
STORAGE
by emphasizing the ability of humans to report verbally
the contents of explicit memory but not of implicit As we have just summarized, an examination of
memory: explicit memory is thus declarative whereas memory storage at the systems level revealed that
implicit memory is non-declarative (Squire & Zola- memory is not a unitary faculty but has at least two
Morgan 1991). Daniel Schachter framed Bruner's forms: implicit (non-declarative) and explicit (declara-
distinction using the terms implicit for `knowing how' and tive). This distinction raises a question: what are the cell
explicit for `knowing that' (Schacter 1996). These are the and molecular mechanisms by which implicit and explicit
most widely used terms, and while speci¢c de¢nitions of memories are stored in the brain? Are the molecular
these various terms can di¡er in di¡erent contexts, all storage mechanisms as di¡erent as is the logic of the
these authors are describing the basis distinction that was explicit and implicit memory systems ?
revealed in the studies of H.M. Our most detailed knowledge about the mechanisms of
implicit memory storage has come from studies of simple
(e) Implicit and explicit memory systems can be forms of implicit memory storage in the invertebrates
further subdivided Aplysia and Drosophila and from parallel studies of simple
The distinction between implicit (non-declarative) and memory storage in mammals. We shall ¢rst consider the
explicit (declarative) memory, as foreseen by James, studies in invertebrates.
Bergson, Ryle and Bruner early in the century and then
revealed by studies of H.M. and other patients with (a) Implicit memory storage is best studied in simple
amnesia, is now generally accepted as being well-founded associative and non-associative forms of learning
biologically. Indeed, once the non-unitary nature of Implicit memory refers to memory for perceptual and
memory had been established, the search was on to motor skills. The simplest instances of such storage are
discover further subdivisions. In 1972, the cognitive elementary forms of non-associative and associative
psychologist Endel Tulving proposed that explicit memory memory. These were ¢rst clearly identi¢ed in studies of
can be further subdivided into episodic and semantic the family of learning processes related to classical
memory (Tulving 1972). Episodic memory is autobiogra- conditioning.
phical memory for speci¢c events; semantic memory is Learning is the acquisition of new information about
more abstract memory for facts, be they autobiographical the world, and memory is the retention of that informa-
or more general. Di¡erent humans are predisposed to tion over time. Classical conditioning is a form of
remember their own pasts in one or the other way; and learning in which an animal learns to associate two
over time autobiographical memory often shades into sensory stimuli, a neutral initiating sensory stimulus
semantic memory, as we can no longer remember the (called the to-be-conditioned stimulus or CS by beha-
experience of a speci¢c event but can unmistakably viourist psychologists), which produces little or no
remember that it did happen to us. Tulving has described response in a naive animal, and a highly e¡ective sensory
a patient, K.C., who after a motorcycle accident possesses stimulus (the unconditioned stimulus or US), which

Phil. Trans. R. Soc. Lond. B (1999)

 The past, the future and the biology of memory storage E. R. Kandel and C. Pittenger 2033

produces an unlearned re£ex response (the unconditioned (a) tail

response or UR). Upon pairing the two sensory stimuli,
the animal learns to strengthen its pre-existing response
to the neutral sensory stimulus (the CS) or to develop a
completely new response to the CS.
In the course of studying classical conditioning, Ivan facilitatory
Pavlov and others discovered that when each of these interneurons { L29 5-HT SCP

two stimuli were repeatedly presented alone, they each

siphon skin
gave rise to distinctive forms of learning and memory
storage. Repeated presentations of the neutral stimulus SN
24
(CS) by itself gives rise to a form of learning called (LE)
INTS
gill
habituation, wherein the animal learns to recognize a
stimulus as innocuous and comes to ignore it. By MN
contrast, presentation of the aversive stimulus (US) 6

alone gives rise to sensitization, a form of learning

wherein the animal enhances its defensive and escape
Downloaded from https://royalsocietypublishing.org/ on 26 December 2024

responses to otherwise innocuous stimuli (CS). Thus,

simple forms of learning take two forms: (i) non- (b)
associative learning, such as habituation and sensitiza-
tion, in which the animal learns about the properties of
a single stimulus; and (ii) associative learning, in which
the animal learns about the relationship between two
stimuli (Squire & Kandel 1999).
What are the cellular changes that result in the brain
when animals learn these simple tasks? Initial insights
into the cell biological mechanisms of each of these three
forms of memory storage ¢rst came from studies of the
marine snail Aplysia. Aplysia lends itself to the study of
implicit memory storage. First, the animal can learn in a
number of di¡erent tasks and has both short- and long-
term memory (as we will discuss further below). Second,
the animal has a relatively simple central nervous system,
consisting of only about 20 000 neurons. Third, the
neurons of Aplysia are particularly large, which allows
them to be uniquely identi¢ed so that one can return to
the same cell in every animal of the species. Fourth, it is Figure 8. (a) The gill withdrawal re£ex in Aplysia is mediated
possible to map in detail the synaptic connections by a simple circuit. Approximately 50% of the learning
between individual cells and between a given cell and the observed in sensitization of this re£ex by tail shock results
sensory and motor periphery. As a result of these advan- from potentiation by serotonin of the direct synapse between
the sensory neuron innervating the siphon and the motor
tages, it is possible to work out signi¢cant parts of the
neuron innervating the gill. Individual identi¢ed neurons are
neural circuitry of a given behaviour ö such as the gill indicated. (b) The circuit controlling this re£ex can now be
and siphon withdrawal re£exesöin terms of uniquely studied in reconstituted cell culture. (SN, sensory neuron;
identi¢able cells and their pattern of interconnections 5-HT FN, facilitatory interneuron; MN, motor neuron.)
(Kandel 1976). One can culture Aplysia neurons and
construct with them in vitro in ways that are not yet
possible in other systems.
5. SHORT-TERM STORAGE INVOLVES FUNCTIONAL
Studies of memory in Aplysia also ¢rst illustrated the
CHANGES IN THE STRENGTH OF PRE-EXISTING
advantages that accrue from using elementary forms of
SYNAPTIC CONNECTIONS
non-associative and associative learning for studying
memory storage. These simple forms of memory consist of In Aplysia, most work has been carried out on the with-
a modi¢cation of the response to a speci¢c test stimulus drawal re£ex of the gill and the siphon to a weak tactile
(the CS) and can be examined at any time after learning stimulus applied to the siphon. This withdrawal re£ex is
by simply examining the time-locked re£ex response to mediated by both monosynaptic and polysynaptic connec-
that stimulus (Kandel et al. 1995). The key requirement tions. The sensory neuron that innervates the siphon
for a neurobiological analysis of memory is to work out, makes direct monosynaptic connections on the motor
in cellular detail, the neural circuitry mediating the beha- neuron that withdraws the gill. In addition, the sensory
viour and how it is modi¢ed by learning. One needs in neurons make polysynaptic connections to motor neurons
particular to work out the re£ex pathway whereby the via interneurons (¢gure 8). While the polysynaptic
sensory stimulus leads to a behavioural response. In connections contribute importantly to both the basal
learning tasks in which there is a clearly de¢ned altera- re£ex and to learning, most studies have concentrated on
tion in the behavioural response to the CS, all the impor- the monosynaptic connections. These connections form a
tant learning-related changes are contained within this signi¢cant component of the behaviour, their electro-
circuit. physiological properties recapitulate the basic properties

Phil. Trans. R. Soc. Lond. B (1999)

 2034 E. R. Kandel and C. Pittenger The past, the future and the biology of memory storage

of both short- and long-term memory for various forms Under normal circumstances and with non-associative
of learning, and they can be reconstituted in culture and learning such as habituation and sensitization, only the
therefore studied in great detail morphologically and AMPA receptors are activated, because the mouth of the
biochemically. NMDA receptor channel is blocked by the ion Mg2+. To
Although the gill and siphon withdrawal re£ex is quite remove the Mg2+ block and activate this channel, two
simple, it exhibits all three simple forms of learning. In events need to happen simultaneously: glutamate needs to
each, the animal learns to alter its behavioural response to bind to the receptor; and the postsynaptic membrane
the tactile stimulus to the siphon (which is the CS for these needs to be depolarized substantially so as to expel Mg2+
forms of learning), and in each case repetition converts a from the NMDA receptor channel mouth. Coincident
short-term form to a long-term form of implicit memory. binding of glutamate to the NMDA receptor and post-
With habituation of the gill and siphon withdrawal synaptic depolarization only occur when the weak siphon
re£ex, the animal learns about the properties of a single, stimulus (CS) and the strong tail shock (US) are paired.
novel stimulus, a weak tactile stimulus to the siphon. Only then is each of the postsynaptic motor cells su¤-
When this stimulus is ¢rst presented, the animal perceives ciently depolarized to allow activation of the NMDA
it as novel and responds to it with a brisk re£ex response. receptors (Lin & Glanzman 1994). As was shown earlier
But when the same weak stimulus is repeated, the animal for sensitization and habituation (Kandel 1976), changes
Downloaded from https://royalsocietypublishing.org/ on 26 December 2024

comes to recognize the stimulus as trivial and learns to at this synapse have now been observed after learning in
stop responding to it. As a result, the weak siphon the intact animal (Murphy & Glanzman 1997). This
stimulus that once produced a brisk response now compellingly supports the notion that changes in this
produces little or no response at all. This progressive synapse are importantly, causally involved in the learning
decrease in the response to the weak siphon stimulus is of new information.
re£ected in re£ex circuit as a weakening of the synaptic The NMDA receptor is particularly permeable to the ion
connections between the siphon sensory neurons and Ca2+, so when it is activated, Ca2+ enters the cell. The Ca2+
their central target cells ö the interneurons and motor in£ux in turn activates a signalling cascade in the post-
neurons (¢gure 8). This weakening results from a decrease synaptic cell. One component of long-term potentiation in
in the amount of the transmitter glutamate released from Aplysia is a retrograde signal that is generated in the post-
the presynaptic terminals of the sensory neurons. synaptic cell and sent back from it to the presynaptic neuron.
With sensitization, Aplysia learns about the properties This signal enhances transmitter release to an even greater
of an important aversive stimulus, a noxious shock to the extent than occurs in sensitization. Thus, in Aplysia, facilita-
tail (an unconditioned stimulus). The animal recognizes tion of the connections between the sensory and motor
the stimulus as aversive and learns to enhance its gill and neurons that occurs with classical conditioning has an addi-
siphon withdrawal responses to the CS, the weak touch to tional, associative component superimposed on the facilita-
the siphon. Sensitization is re£ected in the neural circuit tion produced by sensitization (Squire & Kandel 1999).
as an increase in the synaptic strength in the input These several studies in Aplysia generate three insights
connections of the re£ex, between the siphon sensory into short-term memory storage that have proven quite
neurons and their target cells (the CS pathway). This general and apply to explicit as well as to implicit
strengthening is due to an increase in the release of memory storage. First, these studies show that learning
glutamate from the terminals of the sensory neurons can lead to alterations in synaptic strength and that the
(Kandel 1976; Byrne & Kandel 1996; Carew & Sahley persistence of these synaptic changes in both the mono-
1986; Hawkins et al. 1993). and polysynaptic component of the neural circuit of the
Aplysia also can learn to associate these two stimuli; it re£ex circuit represents the cellular storage mechanisms
can learn classical conditioning. When a weak (CS) for memory. Second, a single synaptic connection can
stimulus to siphon is repeatedly paired with a shock to the participate in, and be modi¢ed by, di¡erent forms of
tail (the US), the re£ex response to the siphon stimulus is learning and participate in di¡erent types of short-term
enhanced. The enhancement of the response to the CS in memory storage. Finally, each of these three simple forms
classical conditioning is substantially greater than with of learning ö habituation, sensitization and classical
sensitization, when the weak siphon stimulus (CS) and conditioning ö can give rise to either a short- or a long-
the tail shock (US) are not paired. This classical term memory, depending on number of training repeti-
conditioning is re£ected in the neural circuitry as a tions, as we will discuss further below. Each of the long-
greatly enhanced strengthening of the input connections term forms of memory is associated with a long-term
of the sensory neurons to their target cells, an enhance- change in synaptic strength in the monosynaptic connec-
ment that is greater than that of sensitization. Whereas in tion between the sensory and motor neuron of the re£ex.
the increase in synaptic strength with sensitization the Thus, not only can a single synaptic connection partici-
principle change is presynaptic (an increased release of pate in di¡erent types of short-term memory storage, but
glutamate from the sensory neuron), in classical the same connection can also be the site of both short-
conditioning both presynaptic and postsynaptic mechan- and long-term memory storage.
isms participate (Squire & Kandel 1999).
There is now a reasonably good understanding of how
6. LONG-TERM STORAGE INVOLVES THE SYNTHESIS
the association in between CS and US is achieved on the
OF NEW PROTEIN AND THE GROWTH OF NEW
cellular level (Glanzman 1995). As we have seen, the
CONNECTIONS
sensory neurons release glutamate from their presynaptic
terminals. Glutamate acts on two types of postsynaptic Given that a single connection can participate in both
receptors: an AMPA receptor and an NMDA receptor. short- and long-term memory, what are the molecular

Phil. Trans. R. Soc. Lond. B (1999)

 The past, the future and the biology of memory storage E. R. Kandel and C. Pittenger 2035

mechanisms for these di¡erent phases of memory the memory for sensitization. Thus, one tail shock
storage ? In particular, how is short-term memory produces an enhancement of the withdrawal response
converted to long-term memory? Can molecular biology that lasts a few minutes. Five shocks produce an enhance-
reveal similarities in the mechanisms of storage of explicit ment that lasts several days, and further training gives
and implicit memory? rise to memory that lasts weeks. This long-lasting sensiti-
The ¢rst insight into the molecular mechanisms of zation requires protein synthesis during a critical time-
memory storage came from the discovery that there are window, whereas short-term sensitization does not
phases of memory storage. The study of memory phases (Montarolo et al. 1986). Here, then, is a very simple
dates to 1885 and the work of Herman Ebbinghaus. By model system of long-term memory that shares some of
forcing himself to memorize lists of nonsense syllabus and the key mechanistic properties of more complex verte-
then testing his own recall, Ebbinghaus determined that brate systems.
there are at least two phases to memory: a short-term Sensitizing tail stimuli activate three di¡erent classes of
phase which contains much information but is transient, modulatory interneurons that synapse on the axon term-
lasting minutes; and a long-term phase which is far more inals of the siphon sensory neurons; all three have similar
stable (Ebbinghaus 1913). This is consistent with our actions. Of the three, the interneurons that release sero-
everyday experience: we have access to far more informa- tonin (or 5-hydroxytryptamine, 5-HT) are thought to be
Downloaded from https://royalsocietypublishing.org/ on 26 December 2024

tion from the recent past (say, the last few minutes) than particularly important (Kandel 1976). In the intact animal
we will be able to remember a few hours hence. Ebbin- and in reduced preparations, serotonin acts on the sensory
ghaus's distinction is also consistent with observations of neuron by increasing the intracellular concentration of the
victims of injury: a person who is struck on the head or second messenger cAMP and by activating the cAMP-
shocked will typically lose memory for events that dependent protein kinase (PKA) and protein kinase C
occurred shortly before the insult but not for more remote (Braha et al. 1990; Brunelli et al. 1976). Transient activation
events (e.g. Zubin & Barrera 1941). of these intracellular signalling cascades with one tail
In the 1960s, Louis Flexner and his colleagues ¢rst shock or one pulse of 5-HT leads to a transient facilitation
found that short- and long-term memory are not only of the synapse between the sensory and motor neurons
distinguished by their time-course but also by their (Castellucci & Kandel 1976). Repeated training or ¢ve
biochemical mechanism. Long-term memory di¡ers from pulses of 5-HT produce long-lasting facilitation that can
short-term memory in requiring the synthesis of new persist for 72 h or more and is accompanied by the growth
protein (Flexner et al. 1965; Agrano¡ 1976). This require- of new synaptic connections (Castellucci et al. 1986;
ment for protein synthesis for long-term memory has a Montarolo et al. 1986; Bailey & Kandel 1993).
speci¢c time-window, during and shortly after training, Whereas a single pulse of 5-HT activates the kinases
which is called the consolidation phase. Blockade of PKA and PKC transiently, ¢ve pulses lead to a persistent
protein synthesis during the consolidation phase will activation of PKA and to the recruitment of the MAP
disrupt long-term memory, but blockade before or after kinase signal transduction pathway (Michael et al. 1998).
will have no e¡ect (Bourtchouladze et al. 1998; Freeman et Both PKA (Bacskai et al. 1993) and MAP kinase (Martin
al. 1995). et al. 1997b) then translocate to the nucleus, where they
The requirement for protein synthesis for long-term activate the transcriptional activator CREB1a (Bartsch et
but not short-term memory, and the existence of a conso- al. 1998) and inactivate the transcriptional repressor
lidation window during which memories are sensitive to CREB2 (Bartsch et al. 1995). The CREB family of tran-
disruption, has proven to be very general. It has been scriptional activators has been implicated in plasticity in
demonstrated in explicit as well as implicit storage and in many systems, as we will discuss below, and may be one
di¡erent vertebrates as well as in invertebrates. This of the most conserved molecular components of the
conservation in turn suggests that the proteins involved in mechanisms for switching on long-term plasticity. Once
the switch to long-term memory may also be conserved. CREB1a is activated and the repressive action of CREB2
If that were true, then a detailed study of the molecules is removed, a set of immediate-early genes is activated,
involved in the switch in any given memory storage including the transcriptional activator ApC/EBP
process in any animal is likely to yield proteins that are of (Alberini et al. 1994). ApC/EBP forms both homodimers
general importance. Moreover, molecular study of several and heterodimers with another factor (activating factor 1).
di¡erent instances of memory storage is likely to reveal The homodimers and heterodimers act di¡erently on
the general nature of a cognitive process: the switch downstream genes that initiate the growth of new
whereby a transient short-term memory is converted to a synaptic connections. This structural change, which
persistent, self-maintained long-term memory (Pittenger represents the long-term, stable form of memory, is
& Kandel 1998). During the last decade, studies in associated with a rearrangement of structural proteins
Aplysia, Drosophila and mice have begun to reveal some of such as the cell adhesion molecule ApCAM (Bailey &
the proteins essential for this switch. Kandel 1993). This internalization of ApCAM is thought
to be a necessary prerequisite for the growth of neuronal
(a) Aplysia and Drosophila use some of the same processes (¢gure 9).
genes and proteins for converting short- to The requirement for transcription in long-term facilita-
long-term memory tion in Aplysia explains why long-term memory requires
The initial molecular insights into long-term storage of the synthesis of new protein. However, this requirement
implicit memory came from studies of sensitization in poses a cell biological puzzle: since long-term plasticity
Aplysia. As with other forms of learning, repetition of the relies on the activation of genes in the nucleus, one might
sensitizing protocol in Aplysia increases the duration of expect that long-lasting changes in the connectivity of the

Phil. Trans. R. Soc. Lond. B (1999)

 2036 E. R. Kandel and C. Pittenger The past, the future and the biology of memory storage

neuron would be cell-wide. Recent experiments have CREB is also required for the induction of transcription-
revealed that each synapse or group of synapses can be dependent plasticity (Davis et al. 1996). However, whereas
modi¢ed independently. This spatially restricted plasticity CREB is required for functional plasticity, it is not su¤-
requires the activity of CREB1 in the nucleus as well as cient for morphological plasticity. For morphological
local protein synthesis in the processes that were modu- plasticity, the loss of the cell adhesion molecule Fas1, a
lated by 5-HT. The mechanism of this speci¢city is homologue of ApCAM, is required, much as the intern-
revealed by a second phenomenon: synaptic capture. alization of ApCAM is required for learning-related
Certain Aplysia sensory neurons have bifurcated axons growth in Aplysia (Schuster et al. 1996).
and can be cultured such that they synapse on two widely
separated motor neurons. When synapse-speci¢c, long-
7. IMPLICIT MEMORY STORAGE AND SYNAPTIC
term facilitation is initiated at one of the two branches, a
PLASTICITY IN THE MAMMALIAN BRAIN
single pulse of 5-HT ö which normally is able to induce
only transient facilitation ö can induce long-term plastic Implicit forms of memory in mammals, such as habi-
changes when applied to a second branch (¢gure 10) tuation, sensitization and classical conditioning, resemble
(Martin et al. 1997a). This phenomenon suggests that the learning in Aplysia. As in Aplysia, the most successful
new genes that are being activated in the nucleus have analyses of implicit memory storage in mammals have
Downloaded from https://royalsocietypublishing.org/ on 26 December 2024

their products distributed widely, but that the products been based on a delineation of the re£ex circuit of the
only persistently strengthen those synapses that have been learned response, so the synapses at which plasticity may
somehow marked by short-term facilitation. A similar be important can in principle be identi¢ed. This approach
phenomenon has been observed in the vertebrate brain has been particularly fruitful in classical conditioning of
(Frey & Morris 1997). the eyeblink re£ex to a pu¡ of air, which depends on the
A similar set of genes important in the switch from cerebellum, and in conditioned fear, which depends on
short- to long-term memory has emerged from studies of the amygdala.
Drosophila. As an experimental system, Drosophila is in
many ways the complement of Aplysia. The great advan- (a) The cerebellum and motor memory
tages of Aplysia are that it is tractable for cell biological The cerebellum is necessary for a form of implicit
studies and that synaptic circuits can be reconstituted in memory: it is required for the learning of coordinated
cell culture. Genetics, however, is currently impossible in motor skills, and in particular for the acquisition of clas-
Aplysia. In Drosophila, on the other hand, cell biological sical conditioning of motor re£exes such as eyeblink in
and electrophysiological studies are di¤cult due to the the rabbit (Thompson et al. 1998). The cerebellum has
tiny size of the neurons, but the genetics are extremely an elegantly simple circuitry. Inputs to the cerebellar
tractable and mature. The great strength of the Drosophila cortex are of two sorts: the mossy ¢bres, which carry the
work in learning is therefore in its genetic analysis. Both information relating to the conditioned stimulus, and the
genetic screens and reverse genetic analysis, in which climbing ¢bres, which represent the unconditioned
speci¢c genes have been disrupted to investigate their stimulus. The mossy ¢bres converge on the large cortical
function, have been fruitful. Purkinje cells. (They do this indirectly by synapsing on
The pioneering work of Seymour Benzer, and the small granule cells, so it is the granule cell axons, the
subsequent studies of his students Quinn, Tully and Davis, parallel ¢bres, which actually synapse on the Purkinje
have led to the identi¢cation of a number of genes cell.) Whereas each climbing ¢bre synapses (approxi-
required for memory storage (Weiner 1999). Many of the mately) one to a Purkinje cell, granule cells are
genes identi¢ed in this way are the same as those impli- massively convergent, with as many as 100 000 parallel
cated in plasticity in Aplysia. For example, the Drosophila ¢bres synapsing on a single Purkinje cell (¢gure 12)
genes dunce, rutabaga and amnesiac all encode components (Llinäs & Walton 1998). The Purkinje cell is ideally situ-
of the cAMP-PKA cascade (¢gure 11) (Davis 1996). ated to form the connections required for classical condi-
Other genes identi¢ed encode participants in other signal tioning, between arbitrary sensory conditioned stimuli
transduction cascades (for example, the gene leonardo; represented by the mossy ¢bres and ¢xed responses
Skoulakis & Davis 1998) or cell^cell adhesion molecules represented by the climbing ¢bres (Marr 1969). Indeed,
similar to ApCAM (Grotewiel et al. 1998). Moreover, a when a single climbing ¢bre is stimulated it will some-
protein synthesis-dependent phase of learning has been times produce a muscle contraction, and when this
described by Tully and his colleagues, and Yin and Tully stimulation is repeatedly paired with an arbitrary
have shown that, as in Aplysia, CREB has a critical role neutral stimulus, that neutral stimulus will come to elicit
in the induction of long-term memory. They found that a the contraction. (That is, direct stimulation of the
dominant-negative CREB allele blocked the formation of climbing ¢bre can substitute for the US in classical
long-term memories (Yin et al. 1994), while overexpres- conditioning to a CS.) (Brogden & Gantt 1937; Swain
sion of the wild-type allele enhances long-term memory et al. 1992.)
stabilization (Yin et al. 1995). This work shows that CREB In this stereotyped circuit, one site of plasticity critical
has a role in learning in Drosophila that is similar or iden- for learning appears to be at the parallel ¢bre^Purkinje
tical to its role in Aplysia, demonstrating striking evolu- cell synapse (Thompson et al. 1998). This synapse exhibits
tionary conservation. long-term depression, or LTD. At the parallel ¢bre
Finally, Corey Goodman and his colleagues have used synapse, induction or LTD appears to require simulta-
the neuromuscular junction of Drosophila to examine the neous depolarization of the Purkinje cell and activation of
developmental plasticity of nerve^muscle synapse. the metabotropic glutamate receptor mGluR1 (a gluta-
Goodman and colleagues have found that at this synapse mate receptor which, rather than passing an ion, initiates

Phil. Trans. R. Soc. Lond. B (1999)

 The past, the future and the biology of memory storage E. R. Kandel and C. Pittenger 2037

sensory neuron

nucleus
CREB-1b CREB-1a

CREB-2 CRE CRE CAAT TAAC

early early late late

P
C/EBP
+ AF-1
CREB-1a AF-1
?

MAP K
Downloaded from https://royalsocietypublishing.org/ on 26 December 2024

C/EBP

adenylyl cAMP ubiquitin

cyclase hydrolase
growth

5-HT Persistent
tail kinase clathrin

apCAM
apCAM
short internalization
term
Figure 9. Multiple molecular
pathways are involved both in
short-term facilitation and in
long-term facilitation of the
sensory neuron-motor neuron
motor neuron synapse in Aplysia.

critical for LTD, is an isoform of the kinase PKC; inacti-

vation of PKC speci¢cally in the cerebellum in a trans-
genic mouse interferes with both parallel ¢bre LTD and
cerebellum-dependent learning (De Zeeuw et al. 1998).
Recently, in another exciting convergence of molecular
mechanisms, a transcription-dependent late phase of
cerebellar LTD has been described, which requires the
kinase CaMKIV and, once again, the transcription factor
CREB (Ahn et al. 1999).

(b) The amygdala and emotional memory

Various aspects of emotional and motivational beha-
viour require the amygdala, a complex, multifunctional
structure (Alheid et al. 1995). In the late 1950s, Lawrence
Weiskrantz ¢rst demonstrated that lesions of the amyg-
Figure 10. The ability to study two synapses from the same
dala impair learned fear (Weiskrantz 1956). Subsequent
sensory neuron on to widely separated motor neurons in
Aplysia cell culture allows synaptic tagging and synaptic work by M. S. Fanselow, J. E. LeDoux and their colla-
capture to be explored for the ¢rst time at the cellular level. borators indicated that one subnucleus of the amygdala,
the lateral nucleus, is required for implicit memory for
fear conditioning to a neutral tone (LeDoux 1998). Infor-
a cascade of intracellular events) (Linden & Connor mation about the CS, the tone, is carried to the lateral
1995). This requirement for two simultaneous activating nucleus via two pathways: the thalamo-amygdala
phenomena is analogous to the associative properties of pathway from the auditory thalamus and the cortico-
the NMDA receptor in Aplysia classical conditioning, as amygdala pathway from the auditory cortex. Long-term
we have discussed above. Downstream of mGluR1, and potentiation has been observed in vivo in the projection

Phil. Trans. R. Soc. Lond. B (1999)

 2038 E. R. Kandel and C. Pittenger The past, the future and the biology of memory storage

from the thalamus to the amygdala after tetanization of memory is likely to involve ongoing activity in frontal
the thalamic input (Clugnet & LeDoux 1990). Impor- lobe circuits (e.g. Lisman et al. 1998). In addition, humans
tantly, synaptic change resembling LTP has been exhibit a cortex-based implicit memory e¡ect known as
observed in this pathway after naturally occurring fear priming, in which sensory stimuli bias subsequent inter-
conditioning (Rogan et al. 1997). pretation of ambiguous stimuli without conscious aware-
Because of the anatomical complexity of the amygdala ness (Schacter 1996). Priming and working memory are
(especially when compared with more regular structures current topics for systems-level investigations in memory;
like the hippocampus and the cerebellum), precise mol- we will return to them brie£y below.
ecular characterization of the plasticity at de¢ned synapses
has been di¤cult. However, blockade of noradrenergic b
8. EXPLICIT MEMORY STORAGE IN THE MAMMALIAN
receptors (which, like 5-HT receptor in Aplysia, are
BRAIN: SPATIAL MEMORY, LONG-TERM
coupled intracellularly to the cAMP pathway) interferes
POTENTIATION AND THE ROLE OF CREB IN MICE
with the formation of emotional memory in humans,
suggesting that the cAMP pathway in the amygdala may What about explicit memory storage ? As we have seen,
be required (Cahill et al. 1994). More recently, LTP has explicit memory involves the conscious recall of facts and
been described at the synapse from cortex to lateral events. It is therefore more complex than implicit
Downloaded from https://royalsocietypublishing.org/ on 26 December 2024

amygdala, and, like facilitation in Aplysia, it has an early memory storage, for two reasons. First, explicit memory
phase and a protein synthesis-dependent late phase. The involves conscious participation in memory recall, while
induction of the early phase requires the activation of implicit recall is unconscious. Second, in the case of
postsynaptic NMDA receptors (like Aplysia classical explicit memory one cannot readily de¢ne a simple CS
conditioning and like the Scha¡er collateral LTP in the pathway; rather, explicit memory involves the integration
hippocampus we shall consider below) but also the action of multiple sensory cues. Nevertheless, the central
of presynaptic PKA, like Aplysia conditioning and sensiti- importance of explicit memory in everyday experience
zation and like mossy ¢bre LTP in the hippocampus has motivated strong interest in plastic mechanisms
(Huang & Kandel 1998). related to it.
As technology for manipulating the mouse genome
(c) The striatum and the memory for habits improved, it became clear that mice are likely to be
In certain other forms of implicit learning in verte- powerful systems for the study of explicit forms of
brates, the CS pathway is less well de¢ned than in the memory. Mice and other rodents, like humans, require
cerebellum and amygdala, and the analysis of the rela- the hippocampus for spatial memory and navigation
tionship of plasticity to learning is correspondingly less (O'Keefe & Nadel 1978) as well as for object recognition
well de¢ned. However, plasticity has been observed in (Ga¡an 1998). Spatial memory in mice seems a particu-
association with such more complex forms of implicit larly good model for explicit memory, because it requires
learning, and in many cases it shares molecular compo- for its formation not a simple CS but an arbitrary asso-
nents with better-studied systems. ciation between several di¡erent sensory cues. Finally, the
While it has been viewed as primarily a motor struc- hippocampus, which is critically involved in spatial and
ture, recent work has shown that the striatum (caudate other forms of explicit memory, exhibits several well-
and putamen) also has an important role in the learning studied forms of synaptic plasticity.
of habits (Graybiel 1995). Long-term potentiation has The basic hippocampal circuit consists of a three-
been described at projections from the cortex to the synapse loop ö the perforant pathway, the mossy ¢bre
dorsal striatum, but it is not as yet known whether this pathway and the Scha¡er collateral pathway ö that runs
long-term potentiation is correlated to learning. This from the entorhinal cortex to the CA1 region of the hippo-
corticostriatal LTP requires the NMDA receptor and campus (¢gure 13). In groundbreaking work, Bliss &
involves intracellular cAMP (Calabresi et al. 1997). CREB LÖmo (1973) found that when the input from the entorh-
is once again implicated in the plasticity: stimulation of inal cortex to the dentate gyrus (the ¢rst cell ¢eld of the
corticostriatal projections activates CREB (Sgambato et hippocampal formation) is stimulated repetitively at high
al. 1998), and interference with CREB attenuates learning frequency, the synapse is persistently strengthened. They
of a striatum-based task in transgenic mice (Pittenger et termed this phenomenon long-term potentiation, or LTP.
al. 1999). Furthermore, CREB is upregulated in the long- Subsequent work has shown that each of the three synapses
term plasticity induced in the ventral striatum (or nucleus in the hippocampal loop exhibits long-term potentiation,
accumbens, an area implicated in depression and in drug and that at each synapse LTP has an early and a late
addiction) by chronic antidepressant treatment (Nibuya et phase, which can be induced by di¡erent stimulus proto-
al. 1996). While the mechanisms and role of corticostriatal cols (¢gure 13). In each case, the late phase di¡ers from the
plasticity remain unclear, it appears possible that some of early phase in that it is blocked by drugs that inhibit
the same molecular pathways are involved. protein or mRNA synthesis. Moreover, the induction of
this late phase requires cAMP and PKA at all three
(d) The cerebral cortex, working memory and synapses. The similarities to the molecular mechanisms
priming involved in Aplysia provided the ¢rst suggestion of
The prefrontal cortex is involved in working memory, conserved mechanisms for converting short-term memory
an important on-line memory that has a restricted to long-term across phylogeny (Abel & Kandel 1998;
capacity for a limited number of items and that allows Huang et al. 1994).
continuous engagement in complex tasks (Smith & Most work has focused on the third synapse in the loop
Jonides 1999). Rather than plastic changes, working from the entorhinal cortex through the hippocampal

Phil. Trans. R. Soc. Lond. B (1999)

 The past, the future and the biology of memory storage E. R. Kandel and C. Pittenger 2039

formation, the Scha¡er collateral synapse, for several CREB-regulated transcription in mammals. Indeed, a
reasons. First, it is technically a particularly easy synapse complete knockout of all mouse CREB isoforms is embry-
to study. Second, studies in humans by Squire and collea- onically lethal (Rudolph et al. 1998). Furthermore, genes
gues have found that a patient with a hypoxic lesion related to CREB are altered in the partial CREB
limited to the CA1 ¢eld (the location of the Scha¡er knockout mice in compensation for the missing CREB
collateral synapse) had a signi¢cant amnesia, suggesting (Hummler et al. 1994).
that damage restricted to this area is su¤cient to disrupt Despite these complications, recent work from Daniel
memory formation (Zola-Morgan et al. 1986). Third, Storm and his colleagues has provided strong, albeit still
whereas dissociations between synaptic plasticity and circumstantial, evidence that CREB is indeed involved in
behaviour in mice have been described at other synapses mouse plasticity. They produced a transgenic mouse
in the hippocampal formation (see, for example, Huang et in which a lacZ reporter gene is activated by a
al. 1995), the e¡orts of many laboratories have built up a CREB-responsive promoter, and they found that this
long list of correlations between disruptions of Scha¡er reporter is activated both by L-LTP in vitro (Impey et al.
collateral LTP and disruptions of hippocampus-depen- 1996) and by certain forms of hippocampus-dependent
dent memory. Although there are occasional dissociations learning in vivo (Impey et al. 1998b). This demonstrates
even at this synapse (see, for example, Zamanillo et al. that CREB or CREB-like transcription factors are in fact
Downloaded from https://royalsocietypublishing.org/ on 26 December 2024

1999), this synapse seems a good place to start disentan- activated under circumstances that lead to plasticity and
gling the role of plasticity in the complex hippocampal suggests a causal role similar to that seen in Aplysia.
circuit. As we have noted above, the transcriptional regulation
There is consensus that induction of LTP in the of genes required for LTF in Aplysia is more complicated
Scha¡er collateral pathway involves activation of NMDA than the regulation of a single transcription factor, invol-
receptors (Nicoll & Malenka 1999). As we have seen, ving several other elements. Substantially more complex
unlike other glutamate receptors, which are activated interactions are likely to be encountered in mice. To work
whenever the presynaptic neuron releases glutamate on to out in detail the molecular switch for converting short-
the postsynaptic membrane, the NMDA receptor is only term to long-term facilitation in the Scha¡er collateral
activated when the release of glutamate by the presy- pathway, much more work will be required.
naptic cell is accompanied by substantial depolarization
of the postsynaptic cells. The NMDA receptor is therefore (a) Hippocampal plasticity beyond the Scha¡er
ideally suited to a role in initiating plasticity, where collateral pathway
synaptic strengthening can result from coincident ¢ring of While a great deal of work has focused on the Scha¡er
the pre- and postsynaptic neurons (Hebb 1949). Activa- collateral synapse, neuronal plasticity also has been
tion of the NMDA receptor leads to in£ux of Ca2+ into described elsewhere in the hippocampus. As we note
the postsynaptic cell and activation of CaMKII. Despite above, LTP is readily demonstrated in the two other
much work, there continues to be signi¢cant controversy major hippocampal ¢bre tracks, the perforant path from
as to subsequent events. Whether the expression of early the entorhinal cortex to the dentate gyrus and the mossy
LTP is post- or presynaptic or both remains unclear. We ¢bre pathway from the dentate gyrus to the CA3 cell ¢eld
will bypass this controversy and again focus on the late, (see ¢gure 13). In both of these pathways LTP has an
transcription-dependent phase. early, protein synthesis-independent phase and a late,
Whereas a single tetanus activates CaMKII, which protein synthesis-dependent phase, just as in the Scha¡er
induces early-phase LTP, multiple tetani increase intra- collateral synapse (Huang et al. 1994; Nguyen & Kandel
cellular cAMP and thus also recruit PKA and MAP 1996). In addition, dense collateral projects from CA3
kinase, which induce nuclear events (Impey et al. 1998a). cells to other CA3 cells may exhibit plasticity; these
Parallelling the ¢nding that MAPK is required for long- projections, while doubtless important for the network
term facilitation in Aplysia, David Sweatt and his collea- properties of the hippocampus, are problematical to study
gues have found that MAPK is activated both in vitro on (Rolls & Treves 1998). Despite an overall similarity in
induction of LTP (English & Sweatt 1996) and in vivo LTP in these pathways, the molecular speci¢cs di¡er. For
after certain forms of learning (Atkins et al. 1998). example, at the mossy ¢bre synapse PKA also is required
Although the mechanism remains obscure, there is for the early phase, which is similar to Aplysia but
currently a consensus that PKA activates the MAPK contrasts with CA1 plasticity.
cascade in neurons and that MAPK is particularly
important in the activation of nuclear factors required for
9. STRUCTURAL CHANGES AND THE BIOLOGICAL
the late phase.
BASIS OF INDIVIDUALITY
The sine qua non of L-LTP in mice is, of course, the
requirement of transcription and induction of genes in the As we have seen, structural changes activated by
potentiated cell. In another example of evolutionary CREB-1 are the de¢ning features of long-term memory
conservation, CREB is once again implicated. This was storage in invertebrates. There is beginning to be evidence
¢rst suggested in a study by A. J. Silva and colleagues, as well for structural changes with LTP in the hippo-
who found L-LTP and hippocampus-dependent learning campus. Thus, several studies suggest that a stably poten-
to be disrupted in a mouse with a knockout of the two tiated synapse can release multiple packets of
most prevalent forms of CREB (Bourtchuladze et al. neurotransmitter (or quanta), whereas an unpotentiated
1994). However, this study is complicated by the nature synapse releases either zero or one (Bolshakov et al. 1997;
of the knockout, which only eliminated some CREB Bolshakov & Siegelbaum 1995). Consistent with this,
alleles (Blendy et al. 1996), and by the complex nature of some ultrastructural studies have found that in tissue in

Phil. Trans. R. Soc. Lond. B (1999)

 2040 E. R. Kandel and C. Pittenger The past, the future and the biology of memory storage

CRE
dCREB-repressor
P dCREB-activator
CREB

Cat. subunit
Ca/CaM Reg. subunit
Downloaded from https://royalsocietypublishing.org/ on 26 December 2024

ATP
rutabaga
PKA
cAMP

L
PDE dunce
amnesiac
AMP Figure 11. Many of the learning-related
genes identi¢ed by screens in Drosophila,
Gs
including dunce, rutabaga and amnesiac,
participate in the cyclic AMP^PKA
pathway that is also implicated in Aplysia.
From Milner et al. (1998).

which potentiation has been induced, synapses can have Cortical plasticity in adults has been demonstrated in
multiple active release zones, or even be split into two both sensory and motor cortices in several systems. The
(Muller 1997). Given the widely held belief that each clearest demonstrations come from work by M. M.
active zone can release at most one quantum of neuro- Merzenich and J. H. Kaas in the somatosensory cortex of
transmitter, this sort of splitting is precisely what may be the monkey. When somatosensory input is altered by
needed to explain the observed release of multiple quanta nerve section (Kaas et al. 1983) or by amputation (Merze-
(however, see Sorra & Harris 1998). Recent studies indi- nich et al. 1984), a portion of the corresponding primary
cate that synaptic activation can lead to the outgrowth of somatosensory cortex is deprived of its input. Over time,
dendritic processes that appear to be percursors of the surrounding regions, which still receive normal input,
dendritic spines, providing a possible mechanism for such expand into the deprived area, so that the somatotopic
morphological change (Maletic-Savatic et al. 1999). map over the surface of the cortex is altered. In the
Anatomical plasticity has also been studied in cerebral converse experiment, Merzenich and colleagues showed
cortex during development and in sensory rewiring in the that when a normal monkey is trained to preferentially
adult. While ultrastructural change of the type described use only some ¢ngers, the cortical representation of those
above may well occur after plasticity in the cortex, the ¢ngers expands (Jenkins et al. 1990).
changes that have been well characterized are rearrange- Dramatic evidence for cortical reorganization in normal
ments of connections over larger populations of neurons. humans has been provided by studies from E. Taub and
Developmental plasticity has been studied in the visual colleagues. They scanned the brains of string instrument
system of frogs, cats, ferrets and monkeys, and in the players. During performance, string players are continu-
whisker-dedicated portion of the somatosensory cortex, or ously engaged in skilful hand movement. The second to
barrel cortex, in rodents. Most often studies involve ¢fth ¢ngers of the left hand, which contact the strings, are
systemically perturbing the relevant sensory input, such manipulated individually, while the ¢ngers of the right
as by blinding one eye or trimming a subset of whiskers, hand, which move the bow, do not express as much
and observing the results. Such studies in kittens have patterned, di¡erentiated movement. Brain images of these
revealed that plasticity in the developing visual system, musicians revealed that their brains were di¡erent from
like adult plasticity at the Scha¡er collateral synapse, the brains of non-musicians: the cortical representation of
requires activation of the NMDA receptor (Gu et al. the ¢ngers of the left hand, but not of the right, was larger
1989). Furthermore, in the mouse, monocular removal of in the musicians (Elbert et al. 1995). Such structural
visual input stimulates expression of a CREB-driven changes are more readily achieved in the early years of
reporter gene in the portion of visual cortex where life. Indeed, Taub and his colleagues found that musicians
synaptic plasticity is occurring (Pham et al. 1999). who learned to play their instruments by the age of 12

Phil. Trans. R. Soc. Lond. B (1999)

 The past, the future and the biology of memory storage E. R. Kandel and C. Pittenger 2041

parallel fibre–
Purkinje cell synapse
parallel fibre molecular layer

Purkinje
cell Purkinje cell layer

granular layer
Figure 12. The elegant
granule cell
circuitry of the cerebellum is
Purkinje cell
involved in motor
coordination, the learning of
white matter motor skills, certain classical
climbing fibre
conditioning tasks and
oculomotor learning. Long-
term depression has been
Downloaded from https://royalsocietypublishing.org/ on 26 December 2024

intensively studied at the

synapse between parallel
Purkinje cell axon ¢bres and Purkinje cells and
appears to be involved in the
learning of cerebellum-based
implicit tasks. Adapted from
mossy fibre Kandel et al. (1991).

years had a larger representation of the ¢ngers of the left and by Fanselow and colleagues in rats (Anagnostaras et
hand than did those who started later in life. al. 1999), suggests that whereas the hippocampus is essen-
These several studies suggest that long-term memory tial for the formation of new explicit memories and
storage leads to anatomical changes in the mammalian possibly for their early storage and recall, it is not the
and even the human brain much as it does in Aplysia. These ¢nal storage location. An elegant recent study in mice
anatomical changes are potentially signi¢cant for under- con¢rms this interpretation. R. Ja¡ard and colleagues
standing the biological basis of individuality, for they have used a radioactive metabolic labelling technique to
suggest that even identical twins, who share a genome, are show that recall of a spatial memory a short time (¢ve
likely to have somewhat di¡erent brains because they are days) after learning recruits the hippocampus and the
certain to have somewhat di¡erent life experiences. associated posterior cingulate gyrus, while recall of the
same memory a long time (25 days) after learning acti-
vates the frontal cortex and associated anterior cingulate
10. THE QUESTIONS CONFRONTING THE SYSTEMS
gyrus but not the hippocampus or posterior cingulate
COMPONENT OF MEMORY STORAGE ARE
(Bontempi et al. 1999). This study dramatically demon-
LIKELY TO OCCUPY OUR ATTENTION EVEN FOR
strates the change over time of the structures required for
THE DISTANT FUTURE
recall of an explicit memory; it is also consistent with
At the end of the decade of the brain and the beginning functional imaging ¢ndings in humans, to which we will
of the new millennium, it is appropriate to contemplate return below, that indicate a role for frontal cortex in the
possible future directions of memory research. The most recall of explicit memories.
di¤cult questions of memory storage are the systems The ¢nal locus of storage of memory is widely assumed
questions. These are likely to occupy us into the distant to be the cerebral cortex, though this is a di¤cult asser-
future. Here we discuss a few of the important systems- tion to prove. How, then, does the role of the hippo-
level questions and discuss how they may be addressed in campus relate to that of the cortex in memory formation,
the coming century. storage and retrieval ? If explicit memories are initially
stored in the hippocampus, how and when are they trans-
(a) What is the relationship of medial temporal lobe ferred to cortex or another repository?
to permanent memory storage? There are two major hypotheses on this issue. The ¢rst
One of the most fascinating systems issues in the study is that memories are never in fact stored in the hippo-
of memory is the question of the how the hippocampus campus but that its role is to form associations between
interacts with other structures that are the ¢nal repository representations of di¡erent aspects of a memory in
of stored memories of semantic and episodic knowledge. di¡erent regions of cortex ö a form of binding. According
As we note above, Milner's studies of H.M. revealed that to this theory, connections between the di¡erent compo-
damage to the hippocampus disrupt the formation of new nents of a memory in di¡erent regions of cortex are initi-
memories but leave more remote established memories ally weak, so the hippocampus continues to be necessary
largely intact. This observation, which has more recently to bind together the components of recent memories. As
been substantiated by L. R. Squire and others in other time passes, however, the connections between represen-
patients and in non-human primates (Squire & Kandel tations of the components strengthen, so more remote
1999; Rempel-Clower et al. 1996; Teng & Squire 1999) explicit memories can be recalled independently of the

Phil. Trans. R. Soc. Lond. B (1999)

 2042 E. R. Kandel and C. Pittenger The past, the future and the biology of memory storage

(a) has been taken as support for this notion (Skaggs &
Schaffer collateral fibre pathway McNaughton 1996). (It should be noted that recapitula-
(associative LTP) recording tion of hippocampal ¢ring patterns during sleep might
also be consistent with the ¢rst model for the relationship
CA1 of hippocampus to cortex.) Distinguishing between these
stimulus two possible models for the relationship of the hippo-
3 campus to the cortex will be one of the major long-term
tasks of the future. All studies of plasticity in the hippo-
campus implicitly suppose that some aspect of memory is
2 1 at least transiently stored there; otherwise there is no
CA3
reason to suppose that plasticity is related to memory in
any way. The resolution of this systems problem will
dentate
region therefore have profound implications for molecular inves-
tigations of memory.
mossy fibre pathway perforant fibre pathway
(non-associative LTP) (associative LTP)
(b) Do the subregions of the medial temporal lobe
Downloaded from https://royalsocietypublishing.org/ on 26 December 2024

serve distinctive functions?

(b) Studies of humans with lesions in the medial temporal
unimodal and polymodal association areas lobe structure suggest an equipotential view of hippo-
(frontal, temporal, and parietal lobes) campal function. Patients with lesions in di¡erent areas of
the hippocampal formation have similar de¢cits, and
perirhinal parahippocampal de¢cits in patients with large lesions are more severe but
cortex cortex qualitatively similar to those in patients with small lesions
(but see Vargha-Khadem et al. (1997) for a recent depar-
DG ture from such a view). Such an equipotential view is
CA3 reminiscent of that of Flourens in the 19th century and of
EC Lashley in the ¢rst half of this century and, like theirs,
CA1
S may arise from inadequate experimental tools.
The appearance of equipotentiality in patients with
lesions of the subregions of the medial temporal lobe
(c)
system may result from the generally disruptive e¡ects of
CA1
400 brain lesions. It therefore will be of interest to use
EPSP slope (% of pre)

L-LTP genetic manipulation in experimental animals and to

300 focus on speci¢c subregions of the medial temporal lobe.
Recent advances in regional control of gene expression
200 (Mayford et al. 1997) should permit one to selectively
E-LTP
disrupt individual components of synaptic plasticity in a
100
subregion of the hippocampus without interfering with
0 other components or with basal synaptic transmission
–60 0 60 120 180 through the circuit.
time (min) It will be important to use regional restriction to
explore which aspects of memory map onto which
Figure 13. (a) In the mammalian brain, plasticity has been portions of the medial temporal lobe system, if indeed
most intensively studied in the hippocampus, and particularly
any straightforward mapping is possible. A mapping of
at the Scha¡er collateral synapse between CA3 and CA1
pyramidal cell ¢elds. (b) Systems properties of memory storage.
memory onto structure could take one of at least two
The hippocampus consists of a series of interconnected cell forms: a mapping of di¡erent types of memory tasks or of
¢elds. Plasticity has been described at all synapses in this di¡erent components of a given memory task. Since the
circuit. (c) LTP at the Scha¡er collateral synapse has an early hippocampus is concerned with storing information about
and late phase. The early phase, induced by a single high- places, objects, people and other living things, di¡erent
frequency train of stimuli, lasts 60^90 min and is independent types of information could be distributed such that that
of macromolecular synthesis. The late phase, induced by some parts of the system are more concerned with place
repeated high-frequency trains, lasts many hours and depends and spatial recognition while others are more concerned
on gene induction and the synthesis of new proteins. with object recognition. Alternatively, di¡erent regions of
the systems could be concerned with one or another
component or operation of a given memory: (i) encoding
hippocampus (Marr 1971). The second hypothesis is that of information that can be maintained; (ii) consolidation
explicit memories are originally stored in their entirety in and storage of that information over time; and (iii)
the hippocampus but that they are somehow transferred retrieval of stored information at a later point.
at a later time to cortical storage. The leading idea as to
the mechanism of this transfer is that it occurs during (c) How is human memory best studied?
sleep (Crick & Mitchison 1983); the ¢nding of recapitula- Most of the detailed work in memory has of necessity
tion during rapid eye movement (REM) sleep of hippo- been performed in experimental animal systems. While
campal ¢ring patterns observed during waking behaviour human memory, both explicit and implicit, is likely to

Phil. Trans. R. Soc. Lond. B (1999)

 The past, the future and the biology of memory storage E. R. Kandel and C. Pittenger 2043

employ similar basic mechanisms as that of simpler (a) The search for a mechanistic framework: do
animal systems, it is also doubtlessly unique in other implicit and explicit storage share a common set of
respects. As we note above, important insights from H.M. molecular mechanisms?
and other patients have shed a great deal of light on the It is already clear that the mechanisms of synaptic plas-
mechanisms of memory, but until recently the resolution ticity involved in di¡erent forms of explicit and implicit
of human studies was limited. With advancing technology memory storage di¡er in detail. This is perhaps not
it will become possible to examine human memory in a surprising given the fact that implicit and explicit
more and more rigorous way. For example, imaging memory di¡er in at least three ways: (i) they use very
studies have shown that the frontal lobe is critically di¡erent logical strategies for recallöconscious aware-
involved in human explicit memory formation in a ness for explicit knowledge and unconscious recall for
modality-dependent way (Buckner et al. 1999), and implicit knowledge; (ii) they store di¡erent types of infor-
imaging of the hippocampus has con¢rmed that in mation; and (iii) they use very di¡erent neuronal
humans, as in rodents, this structure is involved in spatial systems. Despite these di¡erences, di¡erent systems for
navigation (Maguire et al. 1997). Imaging technology is memory storage seem to use an overlapping set of cellular
improving rapidly. In particular, spatial and temporal and molecular components.
resolution are increasing, although still far away from the Studies thus far available indicate that both implicit and
Downloaded from https://royalsocietypublishing.org/ on 26 December 2024

ideal of single-neuron visualization (Posner & Raichle explicit memory storage have phases, a short-term phase
1998). lasting minutes and a long-term phase lasting days or
In the long run, some insight into these issues will longer. In both implicit and explicit storage, repetition
come from studying similar tasks in monkeys and converts the short-term to the long-term form. For both
humans and then examining them by a combination of types of storage, the short-term form involves covalent
interventional studies in monkeys and non-invasive modi¢cations of pre-existing proteins mediated by one or
studies in humans. However, because of the brain's extra- another second messenger kinase (usually CaMKII, PKC,
ordinary complexity and the sheer number of nerve cells PKA and MAP kinase) leading to an alteration of pre-
it contains, we will never be able to achieve a complete existing connections. By contrast, long-term memory
understanding of its information-processing functions requires PKA, MAPK and CREB-mediated transcrip-
without a computational theory. Only convergence of tion, and anatomical changes in the number and size of the
these di¡erent modes of analysis will lead to a satisfactory synaptic connections involving cell adhesion molecules
understanding of human memory. such as ApCAM and Fas1. It seems likely that di¡erent
synapses involved in both implicit and explicit memory
(d) Consciousness and the recall of explicit memory storage achieve their plastic capabilities by using subsets of
An overarching question in human memory research a fairly small family of molecular transducers and e¡ec-
will be to what extent mechanisms of human memory are tors, though the precise members of this group employed
unique. Important changes in the mechanisms of memory will di¡er from synapse to synapse. This overlap is reminis-
storage and recall may be among the evolutionary adap- cent of a central ¢nding in developmental biology: related
tations that set humans apart. Addressing this question molecules or even functional cassettes of interacting mol-
will require signi¢cant advances both in memory research ecules are used in rather di¡erent contexts to do di¡erent
and in other aspects of cognitive neuroscience, such as the jobs.
study of a language, attention, imagery and self-
awareness. The de¢ning feature of explicit memory (b) What is the nature of the synaptic changes
storage is that recall requires a conscious attention. Here responsible for the initial expression of plasticity?
memory research touches on one of the deepest problems Although the forms of synaptic plasticity that contribute
in biological science, and one before which we stand in to short-term memory all involve covalent modi¢cations of
relatively complete ignorance. However, this problem pre-existing proteins, di¡erent forms of plasticity recruit
highlights the key issue to emerge from a consideration of di¡erent signalling systems. LTP is not a singular process
the systems problem of memory: it is impossible to but a family of processes, and slight variations in the indu-
dissociate the system problems of memory research from cing stimulus or learning protocol may alter the signalling
the general agenda of neuroscience to understand percep- pathways recruited. For example, at the Scha¡er collateral
tion, action and conscious awareness. pathway, high-frequency stimulation (100 Hz) triggers
Ca2+ in£ux through the NMDA receptor; Ca2+ activates
CaMKII, and this activation is critical for the initiation of
11. SOME OF THE QUESTIONS CONFRONTING THE
LTP (Giese et al. 1998; Lisman & Goldring 1988). PKA
MOLECULAR COMPONENT OF MEMORY
has no role in this form of LTP. By contrast, when this
STORAGE SHOULD BE SOLVABLE IN THE NEAR
same synapse is stimulated at low frequency (5 Hz), it
FUTURE
induces a form of LTP that also is NMDA receptor depen-
What is in store for the molecular biology of memory dent but now requires PKA (Winder et al. 1999). Thus,
in the new century? As the discussion above makes clear, even the NMDA-dependent form of LTP at the Scha¡er
although the molecular biology of memory is in its collateral synapse has several forms. It therefore seems
infancy, intellectually satisfying insights into the family of possible that di¡erent learning processes recruit di¡erent
mechanisms that contribute to di¡erent forms of combinations of programs for LTP.
memory-related plasticity are in sight. Here we outline How is LTP expressed? In the Schae¡er collateral
some of the most pressing issues confronting the pathway, there is evidence for both presynaptic and post-
immediate future. synaptic mechanisms in the expression of the early phase

Phil. Trans. R. Soc. Lond. B (1999)

 2044 E. R. Kandel and C. Pittenger The past, the future and the biology of memory storage

Figure 14. E¡ect of NMDA antagonist

day 1 day 2
on place cells. Place cells in the
hippocampus represent the most
advanced current example of a fusion of
systems and molecular investigations of
memory. Firing rates of two place cells
are shown here. Each circle corresponds
control
to an environment the animal explored;
coloured pixels correspond to ¢ring rate
at di¡erent points in the environment.
Firing rate is highest in a particular
region of the environment, the place
¢eld. In the upper cell, from a control
rat, the place ¢eld is stable across three
recording sessions on two days. In the
NMDA lower cell, from a rat treated with the
blocker NMDA blocker CPP, the place ¢eld is
stable over two training sessions on the
Downloaded from https://royalsocietypublishing.org/ on 26 December 2024

¢rst day (separated by 1 h) but is

dramatically altered 24 h later. Adapted
from Kentros et al. (1998).

of LTP, but the issue remains highly controversial, as we identi¢ed the neurotrophins BDNF and NT3 (Patterson et
note above. Good electrophysiological evidence suggests a al. 1992) and the secreted protease tPA (Qian et al. 1993);
presynaptic mechanism (Bolshakov et al. 1997; Bolshakov pharmacologically induced seizure, which is a still less
& Siegelbaum 1995). The molecule nitric oxide seems to re¢ned way of turning on depolarization-induced genes,
be capable of serving under some circumstances as a induces a larger number (Hevroni et al. 1998). These
retrograde messenger from the postsynaptic site of induc- examples suggest what some of the categories of induced
tion of LTP to initiate presynaptic change, much as genes will be. Some, such as ubiquitin hydrolase and
occurs in NMDA-dependent facilitation in Aplysia C/EBP, will function to extend or modulate the intra-
(Arancio et al. 1996). However, there also is very good cellular signalling events required for L-LTP. The protease
evidence that with 100 Hz stimulation there is a critical tPA also serves as a step in a signalling cascade (Mars et al.
postsynaptic contribution through activation of previously 1993). Proteases, cell adhesion molecules and neurotro-
silent AMPA receptors (Liao et al. 1995). With recent phins may prove important for the morphological plasti-
advances in imaging (Denk et al. 1996), we can expect city that may occur in L-LTP. Indeed, the Aplysia cell
new light to be shed on this issue in the near future. adhesion molecule gene ApCAM is transcriptionally regu-
Given the high quality of the data on both sides, it seems lated in the induction of LTF, though it is downregulated
likely that the ¢nal picture will contain elements of both rather than increased (Mayford et al. 1992). A related cate-
pre- and postsynaptic expression, perhaps with di¡erent gory might be the cytoskeletal molecules and that also
elements being invoked under di¡erent circumstances. participate in morphological plasticity at the cellular level.
We anticipate that a major thrust of work in the near
(c) How is the late phase of LTP induced and future will be the identi¢cation of more downstream
maintained? genes important for the stabilization of memory. Searches
While PKA seems to be required for L -LTP in all for these genes could take several di¡erent approaches.
regions of the hippocampus as well as in the lateral First, candidate genes can be screened for induction in in
nucleus of the amygdala, PKA does not act alone. For vitro potentiated hippocampal slices or in other models of
example, PKA activates the MAP kinase cascade in plasticity (such as kindling or seizure) (Hevroni et al.
neurons, and both PKA and MAPK are required for 1998). Second, genetically altered animals with speci¢c
CREB activation. There is increasing reason to believe defects in memory storage and speci¢c alterations in the
that the neurotrophin BDNF is involved in various transcriptional apparatus known to be induced in LTP
aspects of LTP (Kang & Schuman 1995; Patterson et al. and restricted in expression to some part of the medial
1996), both as a possible retrograde signal and as a temporal lobe region can be screened against controls to
contributor to growth, but its role has yet be integrated identify novel genes that may be induced. Finally, cells
with other mechanistic details (Patterson et al. 2000). from models of plasticity such as LTP can be speci¢cally
Much evidence points to CREB-1 as a key, conserved screened against controls to identify in a wholly unbiased
regulatory element in the activation of the genes required manner the genes that are induced. The last two of these
for implicit memory in Aplysia and Drosophila, but its role possible approaches will be greatly facilitated by recent
in L-LTP in the mouse is on much less ¢rm ground. advances in technology. In particular, gene chip tech-
Presumably, genes activated by CREB or related nology that has recently become available allows for the
factors are involved in the e¡ector mechanisms of late ¢rst time the genes expressed by two cell populations to
phase plasticity. Some downstream genes have been iden- be rapidly and quantitatively compared (Editorial 1999).
ti¢ed in Aplysia: C/EBP (Alberini et al. 1994), ubiquitin Comparison of cells in which LTP has been induced
hydrolase (Hegde et al. 1997), and elongation factor 1 against controls using this sort of technology should allow
alpha (Hegde et al. 1999). Work in hippocampal LTP has the exhaustive identi¢cation of induced genes; these genes

Phil. Trans. R. Soc. Lond. B (1999)

 The past, the future and the biology of memory storage E. R. Kandel and C. Pittenger 2045

can serve as candidates for disruption, and their identities outcomes. Recent work has begun to emphasize the
may provide immediate clues for the ultimate mechan- contributions of dendritic geometry, internal calcium
isms of late phase LTP. stores and active conductances to the induction of
synaptic plasticity (Linden 1999), but such work still tends
(d) What are the molecular mechanisms for to focus on potentiation at single synapses. A broader
stabilizing memory storage, for synaptic targeting, context of the relative contributions and behaviour of
synaptic tagging and synaptic growth? multiple synapses at di¡erent sites on a neuron will
In Aplysia, and in the hippocampus, the stable self- greatly aid our understanding of physiological plasticity
maintained forms of long-term plasticity seem to require in the future.
the induction of a cascade of genes and the growth of new
synaptic connections (Bailey & Kandel 1993). The ¢nding (f) The important distinction between models and
of a transcriptional switch in converting short- to long- mechanisms of memory
term memory explained why long-term memory requires A pressing question that needs to be addressed for all
new protein synthesis. As we have described above, the forms of plasticity is this: to what extent do the conclu-
products of transcription are targeted to the appropriate sions we draw about experimentally induced forms of
synapses by some sort of synaptic tag. Early evidence in synaptic plasticity studied in vivo or in vitro apply to the
Downloaded from https://royalsocietypublishing.org/ on 26 December 2024

Aplysia suggests the marking process for long-term growth forms of plasticity that underlie learning in an intact
has two components: a covalent, PKA-mediated marking brain? This question has complicated the analysis of the
signal for the growth of new connections, and a local role LTP in hippocampus-based explicit memory in parti-
protein synthesis-dependent stabilization signal (Martin cular because the cell physiological study of this type of
et al. 1997a). Determining the nature of this covalent storage has taken a bottom-up rather than a top-down
mark will necessarily overlap with studies of short-term approach.
synaptic plasticity such as E -LTP on the one hand In the cases of implicit memory for habituation, sensiti-
(because establishment of the tag is presumably one zation and classical conditioning in Aplysia, a top-down
aspect of early LTP), and with the mechanisms of late approach has been used. The analysis began with a
LTP on the other (because proteins bound for the synapse characterization of learning at the behavioural level, then
must interact with the tag to establish stable potentiation proceeded to the neural circuitry that mediates the
at the appropriate sites). behaviour, and only then focused on the cell and
molecular biological changes in the neural circuitry
(e) Are there non-synaptic mechanisms of neuronal produced by learning that serve as sites of memory
plasticity? storage. As we have emphasized above, many implicit
Ever since the time of Ramon y Cajal (Cajal 1893), forms of memory storage result from modi¢cations in the
investigators have focused on the synapse as the site of response to the input (or conditioned stimulus) pathway.
memory-related change. Su¤cient evidence has accumu- Implicit forms of memory, such as eyeblink conditioning
lated, both experimental and theoretical, that it is nearly and fear conditioning, therefore lend themselves to a top-
impossible to doubt that synaptic plasticity is important down approach: once the wiring diagram of the beha-
in memory storage; however, there is no strong reason to viour that is modi¢ed has been delineated, the sites
believe that the synapse is the only important locus of altered by learning, the pathway of the conditioned
change. In particular, modulation of the electrical proper- stimulus and response can be examined quite readily. In
ties of the dendrite proper could lead to plasticity over such cases, the task of matching synaptic change to beha-
large portions of the dendritic arbour. Likewise, modula- viour has been met, at least partially, by the analysis that
tion of the biochemical cascades involved in synaptic plas- has preceded it.
ticity could change the subsequent behaviour of the Studies of the hippocampus-based spatial memory
neuron on a large scale. Finally, any alterations in inte- have taken a reverse, bottom-up approach. There has
grative sites within the neuron, such as the sites of active been little progress, for example, in de¢ning the neural
conductances in the dendrites or the axon hillock, could circuitry for spatial navigation in the mouse. In addition,
obviously alter the input^output relationships of the we do not know whether the animal uses LTP in order to
neuron. The contributions, if any, of these potential alter- learn a new spatial task, or even at which synapses the
native sites of plasticity in neurons have scarcely been important changes occur. LTP was not found in any
investigated to date and may prove to be an important behavioural context but by a fairly arti¢cial examination
area of future research. of synaptic physiology (Bliss & LÖmo 1973). This bottom-
The research that we have emphasized is limited in up approach greatly complicates attempts to relate
another way: it focuses on plasticity at single synapses. synaptic physiology to behaviour.
Despite its advantages, such a focus neglects not only the LTP su¡ers from other weaknesses as an experimental
function of multiple neurons in a circuit but also the func- model. It is generally studied in a slice of tissue, not in
tion of single neurons as an integrated whole. Since the intact brain (though much in vivo LTP work has also been
discovery of active dendritic conductances (Llinäs & carried out). It typically involves an inducing stimulus so
Walton 1998; Spencer & Kandel 1961), it has been clear strong as to verge on the pathological. It is most
that the larger geometry and electrophysiological proper- frequently measured in populations of neurons, from
ties of neurons signi¢cantly in£uence the processing of which the behaviour of single synapses can be inferred
individual synaptic inputs. Therefore, identical plastic only with di¤culty. It evokes an extraordinary number of
processes operating on synapses at di¡erent points in a molecular signalling responses, making it di¤cult to
neuron's dendritic tree may have rather di¡erent discriminate mediators from modulators (Sanes &

Phil. Trans. R. Soc. Lond. B (1999)

 2046 E. R. Kandel and C. Pittenger The past, the future and the biology of memory storage

Lichtman 1999). A recent description of a knockout perspectives, both historically and in current work. These
mouse with little or no LTP (in at least one inducing perspectives on the study of memory di¡er in the questions
paradigm) and normal hippocampus-dependent learning they ask, their methodology and their conceptual frame-
(in at least one behavioural task) forces reconsideration of work. However, our understanding of the mechanisms of
the relevance of experimentally induced LTP to learning memory will not be complete until we can unite both
(Zamanillo et al. 1999). It seems very likely that perspectives into a single, uni¢ed framework. Early steps
prototypical LTP, produced by a 100 Hz train, is not an towards such a synthesis are already apparent in some
important physiological phenomenon underlying learning- current work. Completing this synthesis is one of the ¢nal
related plasticity, any more than the path¢nding of goals of memory research in the future. We illustrate an
developing axons in culture is identical to the path¢nding early attempt at this uni¢cation by considering two exam-
of axons in a developing nervous system. Rather, LTP is ples: (i) LTP, place cells and spatial memory, and (ii)
an experimental model that may share characteristics and molecular therapeutics. In considering the relationship of
mechanisms with the actual mechanisms used for LTP, place cells and spatial memory, we will also re-
learning. examine a question we considered earlier: How can we
Despite these and other weaknesses, recent work relate models of memory to mechanisms of memory?
suggests that some aspects of LTP are indeed likely to
Downloaded from https://royalsocietypublishing.org/ on 26 December 2024

represent a valid model for some aspects of memory (a) Place cells as a synthesis of molecular and
storage. We have already noted that both certain forms of spatial models of memory in the hippocampus
learning and LTP have a protein synthesis-independent How can we bridge the gap between the molecular
phase and a protein synthesis-dependent phase and that mechanisms of synaptic plasticity and the systems problems
both can be blocked by antagonists of the NMDA of spatial memory? In spatial learning, the animal is
receptor. In addition, many di¡erent genetically altered thought to use the sensory input ö visual, proprioceptive,
mice have been generated in the past decade, and vestibular, olfactory ö to develop an internal representa-
molecular de¢cits in Scha¡er collateral LTP correlate tion of its spatial environment and then to use that repre-
surprisingly well with de¢cits in hippocampus-dependent sentation for spatial navigation. Thus, one step towards
learning. Finally, recent studies have shown that learning understanding will be working out representation of space
of hippocampus-based task correlates with the activation in the hippocampus. This representation is thought to exist
of CamKII, PKA, and MAP kinase (Atkins et al. 1998), in the pyramidal cells of the hippocampus, which can form
with the turning o¡ of phosphatases, and with the induc- a spatial map of the environment.
tion of CRE-mediated gene transcription (Impey et al. Hippocampal pyramidal cells function as place cells:
1998b), all of which are predicted from using LTP as a they ¢re when an animal occupies a speci¢c area in its
model system. environment (called the place ¢eld for a speci¢c cell).
However, all of this evidence is correlative. A conclusive Any pyramidal cell can function as a place cell and, in
answer to the question of what aspects of memory storage any given environment, about half of the cells in the
in fact is being measured by LTP will require a better hippocampus function as place cells. These place cells
appreciation of the family of processes that LTP represents, form a `cognitive map' of the environment, and learning a
but also a better understanding of the functional circuitry new environment and storing that new representation in
of spatial memory where these processes are expressed. We the hippocampus involves the creation and stabilization
need a more detailed understanding of how sensory infor- of a new cognitive map (O'Keefe & Nadel 1978). In a new
mation about the spatial environment enters the medial environment, the ¢ring ¢eld of a place cell forms within a
temporal lobe system and how it modulates the naviga- period of minutes (comparable to the acquisition of a
tional system activated during spatial search. learning task), and once formed can be stable for months
Making direct mappings from behaviour to plasticity, (Muller et al. 1987), comparable to long-term memory.
and vice versa, is particularly challenging in the hippo- The map is environment speci¢c, in that a cell's place
campus because of the complex nature of the sensory ¢eld in one environment does not in any way predict the
input of explicit memory storage. In contrast to the place ¢eld in another environment (Muller & Kubie
siphon withdrawal re£ex in Aplysia and other simple 1987). A variety of manipulations of the animal can lead
re£ex systems, the neural input to the hippocampus is to a remapping, in which the place ¢elds of all cells
multimodal, highly processed and often not time-locked. change. Such a remapping may correspond to disorienta-
Spatial learning involves not a change in the response to a tion, in which the animal no longer recognizes that it is
de¢ned conditioned stimulus but rather a change in in a familiar environment.
strategy ö in how the animal processes a complex constel- Research on how the network properties of place cells in
lation of sensory events. Until we have a better under- the hippocampus are remapped in a new environment by
standing of the nature of these inputs and how they learning represents perhaps the best example of research
participate in spatial navigation, the relationship between on the plasticity in an important cellular network involved
hippocampal plasticity and spatial memory-based tasks is in explicit memory storage. An analysis of plasticity in a
likely to remain correlational. system of place cells during spatial learning represent an
important advance in the molecular biology of cognition
that we advocate for the next century. With such research
12. MOLECULAR COGNITION: COMBINING THE
the top-down and bottom-up modes of analysis begin to
SYSTEMS AND MOLECULAR PERSPECTIVES
converge (Rolls & Treves 1998).
As we have emphasized, studies in the biological basis of To bring this ¢ne-grained systems level analysis
memory can be divided into systems and molecular together with a molecular one requires investigating the

Phil. Trans. R. Soc. Lond. B (1999)

 The past, the future and the biology of memory storage E. R. Kandel and C. Pittenger 2047

mechanisms of plasticity in a population of place cells benign senescent forgetfulness. This memory loss is prob-
that may be involved in the learning of a new environ- ably the most bothersome and frequently mentioned
ment. There is now evidence based on such an approach complaint of the elderly (Adams et al. 1997). Age-related
that hippocampal LTP may be involved in the stabiliza- memory loss is typically more pronounced in explicit
tion of place ¢elds as an environment becomes familiar. than in implicit tasks, reminiscent of the de¢cits seen
When an NMDA blocker (the drug CPP) was given to after lesions to the hippocampal system. Indeed, we now
rats during exposure to a novel environment, the animals know that important aspects of age-related memory loss
formed normal place ¢elds, and (surprisingly) those ¢elds does involve a reduction in functioning of the hippo-
were stable for an hour. However, when the animals were campus (Uttl & Graf 1993).
returned to the environment after 24 h, a complete Age-related memory de¢cits are not limited to humans.
remapping occurred: the place cell map of the environ- Useful rat and mouse models have recently been
ment was not stable. Importantly, the place cells in a developed to study the e¡ects of normal ageing on explicit
control environment with which the animal had been and implicit forms of memory (Barnes 1979). As is the
familiarized before drug treatment remained stable case with humans, when tested in explicit spatial memory
(¢gure 14) (Kentros et al. 1998). This study demonstrates tasks, some aged animals perform as well as youthful
that pharmacological treatment that blocks hippocampal controls, but others are strikingly impaired (Gallagher &
Downloaded from https://royalsocietypublishing.org/ on 26 December 2024

LTP can disrupt the stability of hippocampal place cells Pelleymounter 1988). These aged mice also show a decline
and supports the idea that an LTP-like phenomenon with age in Scha¡er collateral LTP. This decline selec-
underlies spatial learning in rodents. tively a¡ects the late, protein synthesis-dependent phase
Studies with genetically modi¢ed animals also support of LTP. This defect may re£ect a loss of dopaminergic
this general conclusion. Genetic manipulations of the modulation of the hippocampus with age, since dopamine
NMDA receptor (McHugh et al. 1996), CaMKII agonists, which increase intracellular cAMP, ameliorate
(Rotenberg et al. 1996) and PKA (Rotenberg et al. 1997) all both the defect in LTP and the spatial learning de¢cit in
show that place ¢elds form fairly normally without aged mice. Such pharmacological amelioration of an age-
NMDA-dependent plasticity, but are not stable in the long related defect may be the ¢rst step towards pharmaco-
term. These observations amount to signi¢cant evidence logical treatment of this widespread complaint in humans
(albeit still correlational) for a connection between the (Bach et al. 1999).
molecular mechanisms of synaptic plasticity, place cell Perhaps the most tragic disease of our time is Alzhei-
stability and spatial learning as assessed by behavioural mer's disease. This dramatic weakening of memory with
measures. If this correlation is substantiated it will repre- time is not simply an acceleration of benign senescent
sent a major step towards a uni¢cation of molecular and forgetfulness. Alzheimer's is associated with distinct
systems approaches to the study of spatial memory forma- pathological changes. While there are no e¡ective treat-
tion in rodents, and a major step towards the development ments for Alzheimer's disease at this time, a close exami-
of a meaningful molecular biology of cognition. nation of the molecular biology of the disease over the last
We have here only considered one point in the medial decade has led to an accelerating understanding of its
temporal lobe circuitry. To expand understanding of place mechanisms, and e¡ective therapies or prevention will
cells will require moving backwards from the CA1 region surely come from this understanding. Genetic screens of
to work out the nature of the sensory input into the families in which the disease occurs at an elevated
hippocampus that gives rise to a spatial map. How is the frequency have revealed alleles of ¢ve di¡erent genes ö the
relevant sensory information processed and encoded in amyloid precursor protein (APP); presenilins 1 and 2; the
the cortex ? How is it transformed at various relays in the Apo-E4 allele of Apo-E; and alpha 2 macroglobulin ö
medial temporal lobe, and how is that transformed infor- that are associated with an increased risk of Alzheimer's.
mation used by the motor system for spatial navigation? Examination of the products of these genes is leading to an
This analysis is extremely ambitious and will occupy us understanding of the etiology of the underlying patho-
into the distant future. logical lesion (Selkoe 1999). For example, there is
preliminary evidence to suggest that all ¢ve genes may
(b) Disorders of memory and steps toward a participate in a biochemical cascade concerned with the
molecular therapeutics degradation of the toxic beta amyloid peptide. Alzheimer's
A second e¡ort in molecular cognition is evident in disease therefore represents a prime example of how a
attempts to analyse and treat disorders of memory. molecular, mechanistic approach can lead to an under-
Throughout its century-long history, the investigation of standing of pathologies whose principal symptom is a
the biological basis of memory has been informed not behavioural degeneration. In the future, we can hope that
only by studies of normal memory storage processes in this mechanistic understanding of Alzheimer's will
humans and experimental animals, but also by studies of become more complete and lead to therapies, and that
disorders of memory. In the 21st century we anticipate other, less common, dementias will quickly follow.
that the study of memory will follow other biological We have speculated above that an increased under-
sciences in expanding into an applied as well as a basic standing of the molecular mechanisms of senescent
science and that one day our expanding understanding of decline and Alzheimer's may lead to pharmacological
the mechanisms of memory formation will lead to treat- treatment. One would hope that these therapeutic
ments for disorders of cognition. insights would extend to various forms of mental retarda-
One example of a memory disorders for which treat- tion. For example, Down's syndrome, autism and Fragile
ments may be available in the future is the gradual weak- X syndrome are often presumed to result from develop-
ening of memory with age, a di¤culty often referred to as mental errors but may in fact also result, at least in part,

Phil. Trans. R. Soc. Lond. B (1999)

 2048 E. R. Kandel and C. Pittenger The past, the future and the biology of memory storage

from acute abnormalities in the mechanisms of synaptic discovery in neuroscience. In the past ten years we have
plasticity (Bartsch et al. 1999). An increased under- indeed experienced an accelerating rate of discovery in
standing of the molecular mechanisms of such diseases many ¢elds of neuroscience, including memory research.
may allow for their pharmacological treatment as well. The gradual uni¢cation of neurobiology with cognitive
Alternatively, gene-based therapies, either through viral psychology and the subsequent emergence of a molecular
transfer or through recombination, may provide novel biology of cognition have been very fruitful and increased
modes of treatment for hitherto intractable diseases of substantially our understanding of the mechanisms of
memory, either in utero or in children or even adults. memory formation. We now have reached the end of both
the 20th century and of the decade of the brain. We have
at this point a clearer understanding of biologically
13. ETHICAL ISSUES IN THE STUDY OF MEMORY
meaningful subdivisions of memory storage and clearer
As our knowledge and interventional ability increase, understanding in outline of some molecular mechanisms
ethical issues will arise from advances in memory of storage relevant to each of these subdivisions. Most
research, as they have in so many other areas of the bio- impressive is the ¢nding that explicit and implicit storage
logical sciences. The most obvious of these is also the seems to use a common and limited set of mechanisms to
simplest to evaluate: the misuse of any advances in convert short- to long-term memory.
Downloaded from https://royalsocietypublishing.org/ on 26 December 2024

memory research to the detriment of patients or others, Whereas satisfactory insight into even the details of the
by damaging or manipulating their memories, is clearly storage mechanisms are in sight, the systems problems are
wrong. However, more subtle issues exist. much more di¤cult and will continue to occupy us for
Several products are currently on the market, which many decades. This is because the anatomical system that
claim to increase memory performance. While currently stores explicit memory is complex, as is the nature of the
available products have little if any proven e¤cacy, it is memory that is stored. Moreover, explicit memory is inti-
only a matter of time before bona ¢de memory-enhancing mately joined with conscious recollection, an area of
substances become available. The simplest issues raised by neuroscience into which we have little insight. Because the
such products are those of side e¡ects: If such drugs act complexity of explicit memory will take time to dissect
by increasing synaptic potentiation, will they increase the (probably another century) it will be advisable to continue
propensity for epilepsy, stressor-related depression or to analyse instances of implicit memory storage, including
post-traumatic stress disorder ? Will these drugs make it the simple implicit memory systems of vertebrates and
di¤cult to forget even unpleasant events? A more subtle invertebrates, and use them as prototypes for under-
societal issue will be more di¤cult to evaluate. If such standing more complex explicit systems. Because explicit
drugs are developed, they are most likely to be developed memory storage is so deeply embedded in perception,
by an established drug company, and once available they action and consciousness, its future is the future of
are likely to be costly. Since insurers seem unlikely neuroscience. As a result, insofar as understanding explicit
initially to deem memory enhancement a medical neces- memory will require understanding much of the brain, the
sity, this implies that memory-enhancing drugs will be mass action arguments of Flourens and Lashley and the
available primarily to the a¥uent. Since memory localization arguments of Gall, Pen¢eld and Milner will,
enhancement would have obvious advantages in educa- in the long run, be joined. We may realize that, even
tion and in almost any profession, such a development though memory is localized, in its explicit form it can have
might concretely enhance the gap in opportunity for an extremely distributed representation that, in the limit,
education that already exists between rich and poor. The may involve much of the brain.
societal implications are troubling.
Another ethical issue may arise in the context of drug The authors thank James H. Schwartz, Jennifer Turner and
addiction and treatment. As we brie£y note above, drug Arthur Pittenger for critical input, Charles Lam for assistance
addiction may involve plastic change in the reward circuits in preparation of the ¢gures, and Harriet Ayers and Millie
Pellan for assistance in typing the manuscript. E.R.K. is
of the ventral striatum. An enhanced understanding of the
supported by the Howard Hughes Medical Institute and the
mechanisms of such plastic change might lead eventually National Institutes of Health. C.P. is supported by the Columbia
to therapeutic techniques to reverse such change or to University Medical Scientist Training Program.
avoid it entirely. Such a development would be of obvious
societal bene¢t; the e¡ective treatment of addiction could
lead to a great reduction in violent crime. If addiction REFERENCES
becomes easily treatable or completely avoidable, much of Abel, T. & Kandel, E. 1998 Positive and negative regulatory
the motivation for the stigmatization and criminalization mechanisms that mediate long-term memory storage. Brain
of drugs in our society will be removed. If the addictive Res. Rev. 26, 360^378.
aspects of drugs of abuse were dissociated from their plea- Adams, R. D., Victor, M. & Ropper, A. H. 1997 Principles of
surable e¡ects, a serious reconsideration of their place in neurology. New York: McGraw-Hill.
our society might ensue. It is not clear what the ethical or Agrano¡, B. W. 1976 Biochemical concomitants of the storage of
behavioral information. Biochem. Sensory Funct. 597^623.
societal consequences of such a development would be.
Ahn, S., Ginty, D. D. & Linden, D. J. 1999 A late phase of cere-
bellar long-term depression requires activation of CaMKIV
14. CONCLUSION and CREB. Neuron 23, 559^568.
Alberini, C. M., Ghirardi, M., Metz, R. & Kandel, E. R.
The 1990s were proclaimed as the `decade of the brain' 1994 C/EBP is an immediate-early gene required for the
in acknowledgement of both the unique challenges posed consolidation of long-term facilitation in Aplysia. Cell 76,
by the human nervous system and the rapid pace of 1099^1114.

Phil. Trans. R. Soc. Lond. B (1999)

 The past, the future and the biology of memory storage E. R. Kandel and C. Pittenger 2049

Alheid, G. F., de Olmos, J. S. & Beltramino, C. A. 1995 Bourtchouladze, R., Abel, T., Berman, N., Gordon, R., Lapidus,
Amygdala and extended amygdala. In The rat nervous system K. & Kandel, E. R. 1998 Di¡erent training procedures
(ed. G. Paxinos). San Diego, CA: Academic Press. recruit one or two critical periods for context memory conso-
Anagnostaras, S. G., Maren, S. & Fanselow, M. S. 1999 lidation, each of which requires proteins synthesis and PKA.
Temporally graded retrograde amnesia of contextual fear Learn. Mem. 5, 365^374.
after hippocampal damage in rats: within-subjects examina- Braha, O., Dale, N., Hochner, B., Klein, M., Abrams, T. W. &
tion. J. Neurosci. 19, 1106^1114. Kandel, E. R. 1990 Second messengers involved in the two
Arancio, O., Kiebler, M., Lee, C. J., Lev-Ram, V., Tsien, R. Y., processes of presynaptic facilitation that contribute to sensiti-
Kandel, E. R. & Hawkins, R. D. 1996 Nitric oxide acts zation and dishabituation in Aplysia sensory neurons. Proc. Natl
directly in the presynaptic neuron to produce long-term Acad. Sci. USA 87, 2040^2044.
potentiation in cultured hippocampal neurons. Cell 87, Broca, P. 1865 Sur le siëge de la facultë du langage articulë. Bull.
1025^1035. Soc. Anthropol. 6, 377^393.
Atkins, C. M., Selcher, J. C., Petraitis, J. J., Trzaskos, J. M. & Brogden, W. J. & Gantt, W. H. 1937 Cerebellar conditioned
Sweatt, J. D. 1998 The MAPK cascade is required for re£exes. Am. J. Physiol. 119, 277^278.
mammalian associative learning. Nature Neurosci. 1, 602^609. Brunelli, M., Castellucci, V. & Kandel, E. R. 1976 Synaptic
Bach, M. E., Barad, M., Son, H., Zhuo, M., Lu, Y. F., Shih, facilitation and behavioral sensitization in Aplysia: possible
R., Mansuy, I., Hawkins, R. D. & Kandel, E. R. 1999 role of serotonin and cyclic AMP. Science 194, 1178^1181.
Age-related defects in spatial memory are correlated with Buckner, R. L., Kelley, W. M. & Petersen, S. E. 1999 Frontal
Downloaded from https://royalsocietypublishing.org/ on 26 December 2024

defects in the late phase of hippocampal long-term poten- cortex contributes to human memory formation. Nature
tiation in vitro and are attenuated by drugs that enhance Neurosci. 2, 311^314.
the cAMP signaling pathway. Proc. Natl Acad. Sci. USA 96, Byrne, J. H. & Kandel, E. R. 1996 Presynaptic facilitation revis-
5280^5285. ited: state and time dependence. J. Neurosci. 16, 425^435.
Bacskai, B. J., Hochner, B., Mahaut-Smith, M., Adams, S. R., Cahill, L., Prins, B., Weber, M. & McGaugh, J. L. 1994 Beta-
Kaang, B. K., Kandel, E. R. & Tsien, R. Y. 1993 Spatially adrenergic activation and memory for emotional events.
resolved dynamics of cAMP and protein kinase A subunits in Nature 371, 702^704.
Aplysia sensory neurons. Science 260, 222^226. Cajal, S. R. y 1893 Neue Darstellung vom Histologischen Bau
Bailey, C. H. & Kandel, E. R. 1993 Structural changes accom- des Centralnervensystem. Arch. Anat. Entwick. 319^428.
panying memory storage. A. Rev. Physiol. 55, 397^426. Calabresi, P., Pisani, A., Centonze, D. & Bernardi, G. 1997
Barnes, C. A. 1979 Memory de¢cits associated with senescence: Synaptic plasticity and physiological interactions between
a neurophysiological and behavioral study in the rat. J. Comp. dopamine and glutamate in the striatum. Neurosci. Biobehav.
Physiol. Psychol. 93, 74^104. Rev. 21, 519^523.
Bartsch, D., Ghirardi, M., Skehel, P. A., Karl, K. A., Herder, Carew, T. J. & Sahley, C. L. 1986 Invertebrate learning and
S. P., Chen, M., Bailey, C. H. & Kandel, E. R. 1995 Aplysia memory: from behavior to molecules. A. Rev. Neurosci. 9, 435^487.
CREB2 represses long-term facilitation: relief of repression Castellucci, V. & Kandel, E. R. 1976 Presynaptic facilitation as a
converts transient facilitation into long-term functional and mechanism for behavioral sensitization in Aplysia. Science 194,
structural change. Cell 83, 979^92. 1176^1178.
Bartsch, D., Casadio, A., Karl, K. A., Serodio, P. & Kandel, E. R. Castellucci, V. F., Frost, W. N., Goelet, P., Montarolo, P. G.,
1998 CREB1 encodes a nuclear activator, a repressor, and a Schacher, S., Morgan, J. A., Blumenfeld, H. & Kandel, E. R.
cytoplasmic modulator that form a regulatory unit critical for 1986 Cell and molecular analysis of long-term sensitization in
long-term facilitation. Cell 95, 211^23. Aplysia. J. Physiol. 81, 349^357.
Bartsch, D., Bourtchouladze, R., Patterson, S., Karl, K. A., Clugnet, M. C. & LeDoux, J. E. 1990 Synaptic plasticity in fear
Chen, A., Till, S., Gordon, R., Osman, M. & Kandel, E. R. conditioning circuits: induction of LTP in the lateral nucleus
1999 Toward a model of the learning and memory de¢cit in of the amygdala by stimulation of the medial geniculate body.
Down syndrome: transgenic mice with tetracycline regulated J. Neurosci. 10, 2818^2824.
minibrain kinase expression in the hippocampus. Soc. Neurosci. Cooter, R. 1984 The cultural meaning of popular science: phrenology
Abstr. (In the press.) and hte organization of consent in nineteenth-century Britain. New
Bergson, H. 1913 Matter and memory. New York: Macmillan. York: Cambridge University Press.
Blendy, J. A., Kaestner, K. H., Schmid, W., Gass, P. & Schutz, Copplestone, F. 1977 Modern philosophy. A history of philosophy. New
G. 1996 Targeting of the CREB gene leads to up-regulation of York: Doubleday.
a novel CREB mRNA isoform. EMBO J. 15, 1098^106. Corkin, S. 1965 Tactually guided maze learning in man: e¡ects
Bliss, T. V. & LÖmo, T. 1973 Long-lasting potentiation of synaptic of unilateral cortical excisions and bilateral hippocampal
transmission in the dentate area of the anaesthetized rabbit lesions. Neuropsychologia 3, 339^351.
following stimulation of the perforant path. J. Physiol. 232, Crick, F. & Mitchison, G. 1983 The function of dream sleep.
331^356. Nature 304, 111^114.
Bolshakov, V. Y. & Siegelbaum, S. A. 1995 Regulation of hippo- Davis, G. W., Schuster, C. M. & Goodman, C. S. 1996 Genetic
campal transmitter release during development and long- dissection of structural and functional components of synaptic
term potentiation. Science 269, 1730^1734. plasticity. III. CREB is necessary for presynaptic functional
Bolshakov, V. Y., Golan, H., Kandel, E. R. & Siegelbaum, S. A. plasticity. Neuron 17, 669^679.
1997 Recruitment of new sites of synaptic transmission during Davis, R. L. 1996 Physiology and biochemistry of Drosophila
the cAMP-dependent late phase of LTP at CA3-CA1 synapses learning mutants. Physiol. Rev. 76, 299^317.
in the hippocampus. Neuron 19, 635^651. De Zeeuw, C. I., Hansel, C., Bian, F., Koekkoek, S. K., Van
Bontempi, B., Laurent-Demir, C., Destrade, C. & Ja¡ard, R. Alphen, A. M., Linden, D. J. & Oberdick, J. 1998 Expression
1999 Time-dependent reorganization of brain circuitry of a protein kinase C inhibitor in Purkinje cells blocks cere-
underlying long-term memory storage. Nature 400, 671^675. bellar LTD and adaptation of the vestibulo-ocular re£ex.
Bourtchouladze, R., Frenguelli, B., Blendy, J., Cio¤, D., Schutz, Neuron 20, 495^508.
G. & Silva, A. J. 1994 De¢cient long-term memory in mice Denk, W., Yuste, R., Svoboda, K. & Tank, D. W. 1996 Imaging
with a targeted mutation of the cAMP-responsive element- calcium dynamics in dendritic spines. Curr. Opin. Neurobiol. 6,
binding protein. Cell 79, 59^68. 372^378.

Phil. Trans. R. Soc. Lond. B (1999)

 2050 E. R. Kandel and C. Pittenger The past, the future and the biology of memory storage

Ebbinghaus, H. 1913 Memory: a contribution to experimental Huang, Y. Y. & Kandel, E. R. 1998 Postsynaptic induction and
psychology. New York: Columbia University Teachers College. PKA-dependent expression of LTP in the lateral amygdala.
Editorial 1999 Probing the brain with DNA chips. Nature Neuron 21, 169^178.
Neurosci. 2, 1. Huang, Y. Y., Li, X. C. & Kandel, E. R. 1994 cAMP contributes
Elbert, T., Pantev, C., Wienbruch, C., Rockstroh, B. & Taub, E. to mossy ¢ber LTP by initiating both a covalently mediated
1995 Increased cortical representation of the ¢ngers of the left early phase and macromolecular synthesis-dependent late
hand in string players. Science 270, 305^307. phase. Cell 79, 69^79.
English, J. D. & Sweatt, J. D. 1996 Activation of p42 mitogen- Huang, Y. Y., Kandel, E. R., Varshavsky, L., Brandon, E. P., Qi,
activated protein kinase in hippocampal long term potentia- M., Idzerda, R. L., McKnight, G. S. & Bourtchouladze, R.
tion. J. Biol. Chem. 271, 24 329^24 332. 1995 A genetic test of the e¡ects of mutations in PKA on
Ferrier, D. 1890 The Croonian lectures on cerebral localization. mossy ¢ber LTP and its relation to spatial and contextual
London: Smith, Elder. learning. Cell 83, 1211^1222.
Flexner, L. B., Flexner, J. B., De La Haba, G. & Roberts, R. B. Hummler, E., Cole, T. J., Blendy, J. A., Ganss, R., Aguzzi, A.,
1965 Loss of memory as related to inhibition of cerebral Schmid, W., Beermann, F. & Schutz, G. 1994 Targeted muta-
protein synthesis. J. Neurochem. 12, 535^541. tion of the CREB gene: compensation within the CREB/ATF
Flourens, P. 1893 Rëcherches expërimentales. Archiv. Gën.de Mëd. family of transcription factors. Proc. Natl Acad. Sci. USA 91,
II, 321^370. 5647^5651.
Freeman, F. M., Rose, S. P. & Scholey, A. B. 1995 Two time Impey, S., Mark, M., Villacres, E. C., Poser, S., Chavkin, C. &
Downloaded from https://royalsocietypublishing.org/ on 26 December 2024

windows of anisomycin-induced amnesia for passive avoidance Storm, D. R. 1996 Induction of CRE-mediated gene expres-
training in the day-old chick. Neurobiol. Learn. Mem. 63, 291^295. sion by stimuli that generate long-lasting LTP in area CA1 of
Frey, U. & Morris, R. G. 1997 Synaptic tagging and long-term the hippocampus. Neuron 16, 973^982.
potentiation. Nature 385, 533^536. Impey, S., Obrietan, K., Wong, S. T., Poser, S., Yano, S.,
Ga¡an, D. 1998 Idiothetic input into object-place con¢guration Wayman, G., Deloulme, J. C., Chan, G. & Storm, D. R.
as the contribution to memory of the monkey and human 1998a Cross talk between ERK and PKA is required for
hippocampus: a review. Exp. Brain Res. 123, 201^209. Ca2+ stimulation of CREB-dependent transcription and ERK
Gall, F. J. & Spurzheim, G. 1810 Anatomie et physiologie due systeme nuclear translocation. Neuron 21, 869^883.
nerveux en general, et du cerveau en particulier, avec des observations Impey, S., Smith, D. M., Obrietan, K., Donahue, R., Wade, C.
sur la possibilite de reconnoitre plusieurs disopsitions intellectuelles et & Storm, D. R. 1998b Stimulation of cAMP response element
morales de l'homme et des animaux, par la con¢guration de leurs tetes. (CRE)-mediated transcription during contextual learning.
Paris: Schoell. Nature Neurosci. 1, 595^601.
Gallagher, M. & Pelleymounter, M. A. 1988 An age-related Jackson, J. H. 1884 The Croonian lectures on evolution and
spatial learning de¢cit: choline uptake distinguishes dissolution of the nervous system. Br. Med. J. 1, 591^593;
`impaired' and `unimpaired' rats. Neurobiol. Aging 9, 363^369. 660^663; 703^707.
Giese, K. P., Fedorov, N. B., Filipkowski, R. K. & Silva, A. J. 1998 James, W. 1890 The principles of psychology. New York: Henry Holt
Autophosphorylation at Thr286 of the alpha calcium- (1890). New York: Dover Publications (1950).
calmodulin kinase II in LTP and learning. Science 279, 870^873. Jenkins, W. M., Merzenich, M. M., Ochs, M. T., Allard, T. &
Glanzman, D. L. 1995 The cellular basis of classical condi- Guic-Robles, E. 1990 Functional reorganization of primary
tioning in Aplysia californica ö it's less simple than you think. somatosensory cortex in adult owl monkeys after behavior-
Trends Neurosci. 18, 30^36. ally controlled tactile stimulation. J. Neurophysiol. 63,
Graybiel, A. M. 1995 Building action repertoires: memory and 82^104.
learning functions of the basal ganglia. Curr. Opin. Neurobiol. 5, Kaas, J. H., Merzenich, M. M. & Killackey, H. P. 1983 The
733^741. reorganization of somatosensory cortex following peripheral
Graybiel, A. M. 1998 The basal ganglia and chunking of action nerve damage in adult and developing mammals. A. Rev.
repertoires. Neurobiol. Learn. Mem. 70, 119^136. Neurosci. 6, 325^356.
Grotewiel, M. S., Beck, C. D., Wu, K. H., Zhu, X. R. & Davis, Kandel, E. R. 1976 Cellular basis of behavior: an introduction to beha-
R. L. 1998 Integrin-mediated short-term memory in Drosophila. vioral biology. San Francisco, CA: W. H. Freeman & Co.
Nature 391, 455^460. Kandel, E. R., Schwartz, J. H. & Jessell, T. M. 1995 Essentials of
Gu, Q. A., Bear, M. F. & Singer, W. 1989 Blockade of NMDA- neural science and behavior. Stamford, CT: Appleton & Lange.
receptors prevents ocularity changes in kitten visual cortex Kang, H. & Schuman, E. M. 1995 Long-lasting neurotrophin-
after reversed monocular deprivation. Brain Res. Dev. Brain induced enhancement of synaptic transmission in the adult
Res. 47, 281^288. hippocampus. Science 267, 1658^1662.
Hawkins, R. D., Kandel, E. R. & Siegelbaum, S. A. 1993 Kentros, C., Hargreaves, E., Hawkins, R. D., Kandel, E. R.,
Learning to modulate transmitter release: themes and varia- Shapiro, M. & Muller, R. V. 1998 Abolition of long-term
tions in synaptic plasticity. A. Rev. Neurosci. 16, 625^665. stability of new hippocampal place cell maps by NMDA
Hebb, D. O. 1949 The organization of behavior. New York: Wiley. receptor blockade. Science 280, 2121^2126.
Hegde, A. N., Inokuchi, K., Pei, W., Casadio, A., Ghirardi, M., Lashley, K. S. 1929 Brain mechanisms and intelligence: a quantitative
Chain, D. G., Martin, K. C., Kandel, E. R. & Schwartz, study of injuries to the brain. Chicago University Press.
J. H. 1997 Ubiquitin C-terminal hydrolase is an immediate- LeDoux, J. E. 1998 Fear and the brain: where have we been,
early gene essential for long-term facilitation in Aplysia. Cell and where are we going? Biol. Psychiat. 44, 1229^1238.
89, 115^126. LeDoux, J. E. 1996 The emotional brain: the mysterious underpinnings
Hegde, A. N., Casadio, A., Martin, K., Inokuchi, K., Pei, W., of emotional life. New York: Simon & Schuster.
Giustetto, M., Kandel, E. R. & Schwartz, J. H. 1999 Induction Liao, D., Hessler, N. A. & Malinow, R. 1995 Activation of
of the polypeptide chain elongation factor EF1a is required for postsynaptically silent synapses during pairing-induced LTP
the late long-term facilitation in Aplysia. Soc. Neurosci. Abstr. (In in CA1 region of hippocampal slice. Nature 375, 400^404.
the press.) Lin, X. Y. & Glanzman, D. L. 1994 Hebbian induction of long-
Hevroni, D. (and 15 others) 1998 Hippocampal plasticity term potentiation of Aplysia sensorimotor synapses: partial
involves extensive gene induction and multiple cellular requirement for activation of an NMDA-related receptor.
mechanisms. J. Mol. Neurosci. 10, 75^98. Proc. R. Soc. Lond. B 255, 215^221.

Phil. Trans. R. Soc. Lond. B (1999)

 The past, the future and the biology of memory storage E. R. Kandel and C. Pittenger 2051

Linden, D. J. 1999 The return of the spike: postsynaptic action Muller, R. U., Kubie, J. L. & Ranck Jr, J. B. 1987 Spatial ¢ring
potentials and the induction of LTP and LTD. Neuron 22, patterns of hippocampal complex-spike cells in a ¢xed envir-
661^666. onment. J. Neurosci. 7, 1935^1950.
Linden, D. J. & Connor, J. A. 1995 Long-term synaptic depres- Murphy, G. G. & Glanzman, D. L. 1997 Mediation of classical
sion. A. Rev. Neurosci. 18, 319^357. conditioning in Aplysia californica by long-term potentiation of
Lisman, J. E., Fellous, J. M. & Wang, X. J. 1998 A role for sensorimotor synapses. Science 278, 467^471.
NMDA-receptor channels in working memory. Nature Neurosci. Nguyen, P. V. & Kandel, E. R. 1996 A macromolecular synth-
1, 273^275. esis-dependent late phase of long-term potentiation requiring
Lisman, J. E. & Goldring, M. A. 1988 Feasibility of long-term cAMP in the medial perforant pathway of rat hippocampal
storage of graded information by the Ca2+/calmodulin- slices. J. Neurosci. 16, 3189^3198.
dependent protein kinase molecules of the postsynaptic Nibuya, M., Nestler, E. J. & Duman, R. S. 1996 Chronic antide-
density. Proc. Natl Acad. Sci. USA 85, 5320^5324. pressant administration increases the expression of cAMP
Llinäs, R. & Walton, K. D. 1998 Cerebellum. In The synaptic response element binding protein (CREB) in rat hippo-
organization of the brain (ed. G. Shepherd). New York: Oxford campus. J. Neurosci. 16, 2365^2372.
University Press. Nicoll, R. A. & Malenka, R. C. 1999 Expression mechanisms
McHugh, T. J., Blum, K. I., Tsien, J. Z., Tonegawa, S. & underlying NMDA receptor-dependent long-term potentia-
Wilson, M. A. 1996 Impaired hippocampal representation of tion. Ann. NY Acad. Sci. 868, 515^525.
space in CA1-speci¢c NMDAR1 knockout mice. Cell 87, O'Keefe, J. & Nadel, L. 1978 The hippocampus as a cognitive map.
Downloaded from https://royalsocietypublishing.org/ on 26 December 2024

1339^1349. Oxford, UK: Clarendon.

Maguire, E. A., Frackowiak, R. S. J. & Frith, C. D. 1997 Patterson, S. L., Grover, L. M., Schwartzkroin, P. A. &
Recalling routes around London: activation of the right Bothwell, M. 1992 Neurotrophin expression in rat hippo-
hippocampus in taxi drivers. J. Neurosci. 17, 7103^7110. campal slices: a stimulus paradigm inducing LTP in CA1
Maletic-Savatic, M., Malinow, R. & Svoboda, K. 1999 Rapid evokes increases in BDNF and NT-3 mRNAs. Neuron 9,
dendritic morphogenesis in CA1 hippocampal dendrites 1081^1088.
induced by synaptic activity. Science 283, 1923^1927. Patterson, S. L., Abel, T., Deuel, T. A., Martin, K. C., Rose,
Marr, D. 1969 A theory of cerebellar cortex. J. Physiol. 202, J. C. & Kandel, E. R. 1996 Recombinant BDNF rescues
437^470. de¢cits in basal synaptic transmission and hippocampal
Marr, D. 1971 Simple memory: a theory for archicortex. Phil. LTP in BDNF knockout mice. Neuron 16, 1137^1145.
Trans. R. Soc. Lond. B 262, 23^81. Patterson, S. L. et al. 2000 (In preparation.)
Mars, W. M., Zarnegar, R. & Michalopoulos, G. K. 1993 Pen¢eld, W. & Perot, P. 1963 The brain's record of auditory and
Activation of hepatocyte growth factor by the plasminogen visual experience. Brain 86, 595^696.
activators uPA and tPA. Am. J. Pathol. 143, 949^958. Pham, T. A., Impey, S., Storm, D. R. & Stryker, M. P. 1999
Martin, K. C., Casadio, A., Zhu, H., E, Y., Rose, J. C., Chen, CRE-mediated gene transcription in neocortical neuronal
M., Bailey, C. H. & Kandel, E. R. 1997a Synapse-speci¢c, plasticity during the developmental critical period. Neuron 22,
long-term facilitation of Aplysia sensory to motor synapses: a 63^72.
function for local protein synthesis in memory storage. Cell 91, Pittenger, C. & Kandel, E. 1998 A genetic switch for long-term
927^938. memory. C. R. Acad. Sci. III 321, 91^96.
Martin, K. C., Michael, D., Rose, J. C., Barad, M., Casadio, Pittenger, C., Scanlin, H., Winder, D. & Kandel, K. E. 1999
A., Zhu, H. & Kandel, E. R. 1997b MAP kinase translocates Expression of a dominant negative CREB mutant restricted
into the nucleus of the presynaptic cell and is required for to striatum and pyriform cortex reveals a role for CREB in
long-term facilitation in Aplysia. Neuron 18, 899^912. procedural memory for striatum-based tasks. Soc. Neurosci.
Mayford, M., Barzilai, A., Keller, F., Schacher, S. & Kandel, Abstr. (In the press.)
E. R. 1992 Modulation of an NCAM-related adhesion mole- Posner, M. I. & Raichle, M. E. 1998 The neuroimaging of
cule with long-term synaptic plasticity in Aplysia. Science 256, human brain function. Proc Natl Acad. Sci. USA 95, 763^764.
638^644. Qian, Z., Gilbert, M. E., Colicos, M. A., Kandel, E. R. &
Mayford, M., Mansuy, I. M., Muller, R. U. & Kandel, E. R. Kuhl, D. 1993 Tissue-plasminogen activator is induced as an
1997 Memory and behavior: a second generation of geneti- immediate-early gene during seizure, kindling and long-term
cally modi¢ed mice. Curr. Biol. 7, R580^ R589. potentiation. Nature 361, 453^457.
Merzenich, M. M., Nelson, R. J., Stryker, M. P., Cynader, Rempel-Clower, N. L., Zola, S. M., Squire, L. R. & Amaral,
M. S., Schoppmann, A. & Zook, J. M. 1984 Somatosensory D. G. 1996 Three cases of enduring memory impairment
cortical map changes following digit amputation in adult after bilateral damage limited to the hippocampal formation.
monkeys. J. Comp. Neurol. 224, 591^605. J. Neurosci. 16, 5233^5255.
Michael, D., Martin, K. C., Seger, R., Ning, M. M., Baston, R. Robbins, T. W. & Everitt, B. J. 1999 Drug addiction: bad habits
& Kandel, E. R. 1998 Repeated pulses of serotonin required add up. Nature 398, 567^570.
for long-term facilitation activate mitogen-activated protein Rogan, M. T., Staubli, U. V. & LeDoux, J. E. 1997 Fear condi-
kinase in sensory neurons of Aplysia. Proc. Natl Acad. Sci. USA tioning induces associative long-term potentiation in the
95, 1864^1869. amygdala. Nature 390, 604^607. (Erratum published in Nature
Milner, B., Squire, L. R. & Kandel, E. R. 1998 Cognitive 391, 818.)
neuroscience and the study of memory. Neuron 20, 445^468. Rolls, E. T. & Treves, A. 1998 Neural networks and brain function.
Montarolo, P. G., Goelet, P., Castellucci, V. F., Morgan, J., New York: Oxford.
Kandel, E. R. & Schacher, S. 1986 A critical period for Rotenberg, A., Mayford, M., Hawkins, R. D., Kandel, E. R. &
macromolecular synthesis in long-term heterosynaptic facili- Muller, R. U. 1996 Mice expressing activated CaMKII lack
tation in Aplysia. Science 234, 1249^1254. low frequency LTP and do not form stable place cells in the
Muller, D. 1997 Ultrastructural plasticity of excitatory synapses. CA1 region of the hippocampus. Cell 87, 1351^1361.
Rev. Neurosci. 8, 77^93. Rotenberg, A., Abel, T., Kandel, E. R. & Muller, R. U. 1997
Muller, R. U. & Kubie, J. L. 1987 The e¡ects of changes in the The ¢ring ¢elds of place cells in mice with impaired late-
environment on the spatial ¢ring of hippocampal complex- phase LTP are crisp but move to unpredictable locations. Soc.
spike cells. J. Neurosci. 7, 1951^1968. Neurosci. Abstr. (In the press.)

Phil. Trans. R. Soc. Lond. B (1999)

 2052 E. R. Kandel and C. Pittenger The past, the future and the biology of memory storage

Rudolph, D., Tafuri, A., Gass, P., Hammerling, G. J., Arnold, Tulving, E. 1972 Episodic and semantic memory. In Organization
B. & Schutz, G. 1998 Impaired fetal T cell development and of memory (ed. E. Tulving & W. Donaldson), pp. 607^628. New
perinatal lethality in mice lacking the cAMP response York: Academic Press.
element binding protein. Proc. Natl Acad. Sci. USA 95, Tulving, E., Schacter, D. L. & Stark, H. 1982 Priming e¡ects in
4481^4486. word-fragment completion are independent of recognition
Ryle, G. 1949 The concept of mind. University of Chicago Press. memory. J. Exp. Psychol.: Learn. Mem. Cogn 8, 336^342.
Sanes, J. R. & Lichtman, J. W. 1999 Can molecules explain Tulving, E., Schacter, D. L., McLachlan, D. R. & Moscovitch,
long-term potentiation? Nature Neurosci. 2, 597^604. M. 1988 Priming of semantic autobiographical knowledge: a
Schacter, D. L. 1996 Searching for memory: the brain, the mind, and case study of retrograde amnesia. Brain Cogn 8, 3^20.
the past. New York: BasicBooks. Uttl, B. & Graf, P. 1993 Episodic spatial memory in adulthood.
Schuster, C. M., Davis, G. W., Fetter, R. D. & Goodman, C. S. Psychol. Aging 8, 257^273.
1996 Genetic dissection of structural and functional compo- Vargha-Khadem, F., Gadian, D. G., Watkins, K. E., Connelly,
nents of synaptic plasticity. II. Fasciclin II controls presynaptic A., Van Paesschen, W. & Mishkin, M. 1997 Di¡erential e¡ects
structural plasticity. Neuron 17, 655^667. of early hippocampal pathology on episodic and semantic
Scoville, W. B. & Milner, B. 1957 Loss of recent memory after memory. Science 277, 376^380. (Erratum published in Science
bilateral hippocampal lesions. J. Neurol. Neurosurg. Psychiat. 20, 277, 1117.)
11^21. Weiner, J. 1999 Time, love, memory: a great biologist and his quest for
Selkoe, D. J. 1999 Translating cell biology into therapeutic the origins of behavior. New York: Knopf.
Downloaded from https://royalsocietypublishing.org/ on 26 December 2024

advances in Alzheimer's disease. Nature 399, A23^A31. Weiskrantz, L. 1956 Behavioral changes associated with ablation
Sgambato, V., Pages, C., Rogard, M., Besson, M. J. & Caboche, of teh amygdaloid complex in monkeys. J. Comp. Physiol.
J. 1998 Extracellular signal-regulated kinase (ERK) controls Psychol. 49, 381^391.
immediate early gene induction on corticostriatal stimulation. Wernicke, K. 1908 The symptom-complex of aphasia. In
J. Neurosci. 18, 8814^8825. Diseases of the nervous system (ed. A. Church), pp. 265^324. New
Skaggs, W. E. & McNaughton, B. L. 1996 Replay of neuronal York: Appleton.
¢ring sequences in rat hippocampus during sleep following Winder, D. G., Martin, K. C., Muzzo, I., Rohrer, D.,
spatial experience. Science 271, 1870^1873. Chruscinski, A., Kobilka, B. & Kandel, E. R. 1999 ERK
Skoulakis, E. M. & Davis, R. L. 1998 14^3-3 proteins in neuronal plays a regulatory role in the induction of LTP by theta
development and function. Mol. Neurobiol. 16, 269^284. frequency stimulation and its modulation by b-adrenergic
Smith, E. E. & Jonides, J. 1999 Storage and executive processes receptors. Neuron 24, 1^20.
in the frontal lobes. Science 283, 1657^1661. Yin, J. C., Wallach, J. S., Del Vecchio, M., Wilder, E. L., Zhou,
Sorra, K. E. & Harris, K. M. 1998 Stability in synapse number H., Quinn, W. G. & Tully, T. 1994 Induction of a dominant
and size at 2 hr after long-term potentiation in hippocampal negative CREB transgene speci¢cally blocks long-term
area CA1. J. Neurosci. 18, 658^671. memory in Drosophila. Cell 79, 49^58.
Spencer, W. A. & Kandel, E. R. 1961 Electrophysiology of Yin, J. C., Del Vecchio, M., Zhou, H. & Tully, T. 1995 CREB as
hippocampal neurons. IV. Fast prepotentials. J. Neurophysiol. a memory modulator: induced expression of a dCREB2 acti-
24, 260^271. vator isoform enhances long-term memory in Drosophila. Cell
Squire, L. R. & Kandel, E. R. 1999 Memory: from mind to 81, 107^115.
molecules. Scienti¢c American Library. New York: Scienti¢c Young, R. M. 1970 Mind, brain, and adaptation in the nineteenth
American. century. Oxford: Clarendon Press.
Squire, L. R. & Zola-Morgan, S. 1991 The medial temporal Zamanillo, D. (and 16 others) 1999 Importance of AMPA recep-
lobe memory system. Science 253, 1380^1386. tors for hippocampal synaptic plasticity but not for spatial
Swain, R. A. H., Shinkman, P. G., Nordholm, A. F. & Thompson, learning. Science 284, 1805^1811.
R. F. 1992 Cerebellar stimulation as an unconditioned stimulus Zeki, S. 1993 A vision of the brain. Boston, MA: Oxford.
in classical conditioning. Behav. Neurosci. 106, 739^750. Zola-Morgan, S., Squire, L. R. & Amaral, D. G. 1986 Human
Teng, E. & Squire, L. R. 1999 Memory for places learned long amnesia and the medial temporal region: enduring memory
ago is intact after hippocampal damage. Nature 400, 675^677. impairment following a bilateral lesion limited to ¢eld CA1 of
Thompson, R. F., Thompson, J. K., Kim, J. J., Krupa, D. J. & the hippocampus. J. Neurosci. 6, 2950^2967.
Shinkman, P. G. 1998 The nature of reinforcement in cere- Zubin, J. & Barrera, S. E. 1941 E¡ect of electric convulsive
bellar learning. Neurobiol. Learn. Mem. 70, 150^176. therapy on memory. Proc. Soc. Exp. Biol. Med. 48, 596^597.

Phil. Trans. R. Soc. Lond. B (1999)