kayakstarsΤεχνίτη Νοημοσύνη και Ρομποτική

15 Νοε 2013 (πριν από 3 χρόνια και 6 μήνες)

144 εμφανίσεις





David R. Vago

A paper submitted to the faculty of

The University of Utah

in partial fulfillment of the requirements for the degree of

Doctor of Philosophy

Department of Psychology

University of Utah

April, 2003




LIST OF FIGURES………………………………………………………………










THE MEDIAL TEMPORAL LOBE…………………………………………..….




THE AMYGDALA……………………………………………………………….



MORY STORE……………………………………………………………….







A COGNITIVE PERSPECTIVE………………………………………………….












Multiple retrograde gradients………………………………………………… 12


Projections from BLA to other areas involved in consolidation

………….…. 25

3a. The standard model of co
nsolidation………………………………………… 30

Performance of the standard model in a RA experiment

……………………. 30


The role of MTL across multiple consolidation theories..…………………… 34


A prepositional interleaved semantic network..……………………………… 36


/Molecular signaling cascade……………………………………. 41


Shiffrin buffer model……………….……………………………… 48


Integrative model of consolidation…………………………………………… 62




I would like to thank Ray Kesner for his patience an

guidance towards

ight at the end of the tunnel, even though at times our perception of that light was slightly
different. I express sincere gratitude

Sarah Cream and Charlie Shimp for serving on my thesis
for their
time and
Big smiles
to Sonia Matwin and
“Hey, thanks dude” to
David Adler for their feedback and contributions to this manuscript.

“They say the best way to cook small fry is not to stir too much”

Chuang Tzu




“The stream of thought flows on; but most of its segments fall into

the bottomless abyss of oblivion. Of some, no memory survives the

instant of their passage. Of others, it is confined to a few moments,

hours, or days. O
thers, again, leave vestiges which are indestructible,

and by means of which they may be recalled as long as life endures.

(James, 1890, p.643).

All organisms have some fundamental form of memory that can be conceived of as a
“natural outcome of the
brain’s various information processing activities” (Eichenbaum, 2002).
Through an onslaught of perceptual, sensorimotor, and cognitive processing, collectively
referred to as experience, memory appears to arise as an essential means of filtering and
izing the necessary information for survival. Through an onslaught of multimodal sensory
and perceptual processing, we appear to be able to selectively gather information from our
experiences, permanently store them, and in a sense immortalize the sights,

sounds, smells, and
feelings that embrace the perceptual coloring that is characteristic of the original experience.
The selectivity of our memory is…“the very keel on which our mental ship is built. If we
remembered everything, we should on most occasi
ons be as ill off as if we remembered nothing.
It would take as long for us to recall a space of time as it took the original time to elapse, and we
should never get ahead with our thinking” (James, 1890, p. 680).

Another evolutionary outcome of our mem
ory system is its ability to maintain fluidity of
our experiences of moment after moment, and in that sense, memory maintains fluidity in its
construction or encoding. People may not always be aware of the succession of their experiences
as events, but if

asked to do so, they can report events in the natural progression or flow of their


personal happenings (Tulving, 1983). William James (1890) originally distinguished between
two forms of memory: 1) primary memory, which is made up our current conscious

state of
mind, referred to today as working or short
term memory (STM) and 2) secondary memory,
which is made up of knowledge from some previous state of mind, referred to today as long
memory (LTM). Since the conception of primary and secondary mem
ory (James, 1890), there
has been ample evidence demonstrating the existence of at least two distinctly different means of
representing acquired information in the central nervous system (see Gerard, 1949; Hebb, 1949).
Multiple issues have arisen and per
sisted since the conception of two
stage models of memory.
Specifically, what are the
, and/or
cognitive processes

involved in the inherent selectivity of memory? Under what circumstances
is mem
ory transformed into LTM? Are these circumstances an automatic biological process that
occurs by means of exposure itself or is there some active psychological phenomenon contingent
upon evolutionary pressures that must accompany an experience (i.e., rehe
arsal, attention, etc.)?
In either case, memory has been proposed to manifest itself in multiple ways by multiple
functionally and anatomically distinct brain systems (Eichenbaum, 2002; McGaugh, 2000;
Polster, Nadel, & Schacter, 1991; Squire & Alvarez, 19
95). Within the conceptual framework of
each emerging theory, there has been an underlying theme, in which there is a temporal cascade
of events organized intrinsically by evolutionary drives that describe the transformation process
of the experience into
an indelible trace at either a microscopic (molecular/cellular), or
macroscopic (cognitive systems) level. It is clear that there is an adaptive significance to the
existence of the selective nature of our memory as well as to its inherent ability to enco
de that
which is important to us; however, it is currently debated as to what cognitive and physiological


mechanisms are behind the selective pressures, as well as the manner in which particular
information from the experience at hand is stored.

A little

more than a century ago, at the University of Gottingen, Germany, Georg Elias
Müeller and his student, Alfons Pilzecker (1900) proposed that the neural processes underlying
this evolutionary important learning process first persist or perseverate in a lab
ile or short
modifiable state, which in time is then transferred into a relatively long
lasting, stable state. In
other words, STM is converted to LTM through a neurally
driven process of perseveration. This
process of transference has become known

as consolidation, for which multiple mechanisms
have been proposed at each of the multiple levels previously mentioned (i.e.,
, and/or
cognitive processes)
. Throughout the
cognitive neuroscience

literature, the process of consolidation has been described in terms that
appear to overlap with those semantic descriptions of the processes of acquisition, encoding
and/or storage and as may become apparent, this review refrains from making a semantic
istinction due to the mechanistic similarities between the referenced concepts. However, the
process of consolidation can be partitioned into multiple stages differing in the proposed
mechanisms underlying consolidation and in the time frame (seconds to h
ours and hours to
years) involved in each of the stages. This review intends to not only describe consolidation
theory from multiple perspectives, but also to consider the possible interactions between the
collateral approaches, as well as to elucidate po
ssible theoretical and/or analytical associations
between them.

Establishment of Consolidation Theory

Clinical and experimental evidence in both primates and other animals strongly supports
the hypothesis that memory storage is time
dependent. The time
course for consolidation has


been thought to extend like a gradient in which a memory for a recent event has been found to be
more susceptible to disruption by some type of brain trauma, electroconvulsive shock (ECS), or
pharmacological manipulation, than
more remote memories that were formed long before the
disruption (McGaugh and Herz, 1972). The interupting of brain activity shortly after learning
has been shown to impair the formation of LTM for that which was learned; however, the
duration of the susc
eptibility to disruption has been found to be brief (i.e., seconds, minutes, or
hours). This initial, fragile period in the life of a memory is assumed to be a particularly
sensitive and essential step in the process of consolidation. Prone to interferen
ce, the memory
trace has been shown to remain fragile until it has been forged by progressive stabilization
processes and matured into a retrievable trace that is resistant to disruption. It should be noted
that the stabilization process does not necessar
ily reflect any conscious or active cognitive
processes; however, it is not limited to passive or automatic biological processes. In either case,
these processes may take place in parallel with other behavioral and/or biological operations
(Weingartner an
d Parker, 1984).

The “perseveration
consolidation hypothesis” put forth by Müeller and Pilzecker
suggested that once perseverating neural activity (representing an ongoing learning experience)
has been transformed into a more stable, memory trace, disrup
tion would no longer be expected.
More recently, it has been suggested that the gradient for disruption varies greatly with the extent
of the damage and/or type of memory in question (Hodges, 1994; Squire & Alvarez, 1995; Nadel
& Moscovitch, 1997). Addit
ionally, evidence which will be elaborated upon later supports the
notion that a stable memory trace may be susceptible to disruption and reorganization depending
on future retrieval processes and subsequent reconsolidation (see McClelland, McNaughton, &


’Reilly, 1995; Misanin, Miller & Lewis, 1968; Nadel & Land, 2000; Nader, Schafe, & Ledoux,

Müeller and Pilzecker (1900) demonstrated that consolidation was due to persistent
physiological activity that “persists for some minutes in the nervous trac
ts concerned, and this …
increases the fixity of the associations” (cited in McDougal, 1901, p. 393 & Polster et al., 1991,
p. 97). In their pursuit to develop the perseveration
consolidation hypothesis, Müeller and
Pilzecker used lists of nonsense syllab
les that had been introduced earlier by Ebbinghaus (1885).
Müeller and Pilzecker presented these nonsense syllables as paired associates and later cued
subjects with the first syllable of each pair in random order and subjects were instructed to report
e second syllable from each pair. The subjects in these experiments reported a strong tendency
for the syllable pairs to come to mind between training sessions, which suggested some sort of
perseverative process. Perseveration, in this case, can be exten
ded to both the cognitive and
neurobiological domains of processing underlying the establishment of the particular
associations between syllable pairs. Müeller and Pilzecker proposed that perseveration of
syllable pairs was occurring during the time betwe
en initial presentation of the pairs and the task
of recall. Perseveration was necessary for not only establishing the representations of the
syllables in memory, but also for strengthening the associations between the syllable pairs.

Conceptually, “str
engthening” of the memory trace emerges across all levels of
consolidation theory as being phenomenologically necessary for long
term storage. Precisely
what is strengthened in the multimodal mosaic of contextual attributes is itself debatable.
Müeller a
nd Pilzecker suggested that by disrupting the perseverative process, they could interfere
with the formation or consolidation of the associative memories. They then gave subjects the
disruptive task of reading another list of non
sense syllables or of des
cribing the landscape of a


picture between the time of studying the original list of non
sense syllables and being tested on
their memory for them. It was demonstrated that the subjects performed better if there was no
intervening disruptive task, regardl
ess of whether the intervening task was related to the learned
material (Lechner et al., 1999; Polster et al., 1991). The effect of the intrusive activity was called
retroactive inhibition because “this type of inhibition acts on the effect of an ostensib
completed process…” (Müeller and Pilzecker, 1900, p. 179). Müeller and Pilzecker also varied
the time course between the presentation of the original list and the intervening task, noting that
delaying the intervening task by more than a few minutes le
ssened its interfering effect. They
suggested that the time course for consolidation was approximately 10 minutes; although, the
longest interval between the presentation of the original list and the intervening task was 6
minutes and retroactive inhibiti
on was not observed with intervals longer than 1 minute (Lechner
et al., 1999). Burnham (1903) later emphasized that consolidation requires the repetition of
information over time, along with physiological processes involving the establishment of
ions in order for memories to be organized for stability. The idea was that once the
memory was stable, it was less susceptible to damage and was assumed stored for the long term.
This assumption further implied that interpolated activity may compete for

the cognitive and
physiological resources necessary for forming a stable memory trace and thus, experiencing the
event alone was not enough for consolidation.

Neuropsychological evidence

Given the difficultly in operationalizing the nature of conso
lidation, the behavioral
substrates for consolidation have been equally difficult to quantify. If we are interested in
determining the length of the consolidation process, we would hope that there would be some
extrinsic, observable behavior that correlat
es with what may be a perseverative process.


However, it has been proposed that consolidation processes for a particular experience continue
well into the time in which subjects are attending to completely different stimuli (Weingartner &
Parker, 1984).
Russel (1971) asks, “If the memory of how to swim involves circuits which never
rest, why are we not constantly waving our arms and legs around in response to this constant
activity?” (p. 67). The focus of much clinical research has been the cognitive dys
(especially memory impairments) associated with brain damage or electroconvulsive shock
therapy (ECS), which produced an amnestic effect. Human and animal models emerged
depicting anterograde memory impairments (failure to form new memories) and
a time
dependent retrograde amnesia (RA), whereby more recent information compared to more remote
information acquired prior to a cerebral insult is lost. The slope of the RA gradient describes the
relationship between the likelihood of retrieval and the
length of the interval between the
subject’s exposure to a to
remembered experience and an amnestic (e.g., cerebral insult,
pharmacological manipulation) agent. The phenomenon of RA gave insight into the length of
the consolidation process, the structu
res that were important in the process and the particular
types of memory more susceptible to damage. It is important to note that consolidation theories
based on retrograde gradients have not sufficiently accounted for some observed accounts of an
ive flat RA representing events decades prior to the amnestic event (see Fujii, Moscovitch
& Nadel, 2000; Hodges, 1995). Additionally, it is important to consider that these retrograde
gradients are assessed by performance and that the observed deficits m
ay be related to disruption
in retrieval mechanisms. However variable, the gradient is believed to demonstrate what has
been learned is not instantly made permanent (Nadel & Moscovitch, 1997).

Even before Müeller and Pilzecker’s perseveration
ion hypothesis, Ribot
(1882, 1887, 1892) referred to the temporal gradient of consolidation in his clinical assessments


of patients with acute brain trauma. Ribot found patients not only lost recollection of the
accident, but also the ability to recall a
long period of his/her life before the accident (Ribot,
1892). In most of the cases Ribot reported, amnesia was found to affect memory more for the
events that occurred minutes before the trauma. “The most recent formations are first destroyed,
and the d
estructive work goes on, descending, so to speak, from layer to layer, until it reaches the
oldest acquisitions

that is to say, the most stable . . .” (Ribot, 1887, pp. 183). Ribot (1887)
also stated that degeneration was greatest in certain types of m
emory (i.e., complex memories
were more affected than simple memories) because “they have not been repeated so often in
experience” (p. 204). He also noted a gradient in the preserved ability to recall automatic types
of memory compared to the degeneratio
n observed in voluntary types. More recent descriptions
of amnesia have suggested that skill
related or motor memories as well as overlearned factual
knowledge are spared (Zola
Morgan, Squire, & Amaral, 1986). What emerged from these early
clinical cases
was Ribot’s “law of regression”, which argues that recent memories are more
susceptible to disruption than more remote memories. In support of the existence of a temporal
gradient for consolidation, Ribot also argued that remote memories become more resis
tant to
disruption as time passes after learning (Albert, 1984). The scientific literature soon provided
ample evidence supporting Ribot in that a
cute brain trauma would consistently produce a
selective loss of memory for recent experiences (Burnham 1903;

McGaugh, 1966; Russell and
Nathan 1946; Scoville & Milner, 1957).
Russel and Nathan (1946) described a 22
old man
who, a week after a motorcycle accident, demonstrated a disruption to his remote memories that
spanned at least seven years. Two weeks

after the injury, the patient remembered the five years
in the remote past; however, the past two years were a “complete blank as far as his memory was
concerned” (p. 291).



In order to thoroughly examine Müeller and Pilzecker’s proposed time
consolidation processes, it was necessary to have an experimental model of RA similar to the one
observed by Ribot in a clinical setting. It was reported by Zubin & Barrera (1941) that when
electroconvulsive shock (ECS) was administered shortly after pair
ed associate learning, there
was a disruption on memory performance, but as the time interval increased between learning
and presentation of ECS, the intensity of the disruption decreased. Similar studies continued to
be conducted extensively using both h
uman and animal subjects. Because the findings paralleled
those observed in clinical patients, the retrograde effects due to “shock amnesia”, as it was
termed, were commonly cited as evidence for why ECS was an experimentally controlled
procedure that int
erfered with consolidation processes.

Duncan (1949) then demonstrated experimentally
induced RA in rodents on an avoidance
learning task in which amnesic treatments (e.g., ECS) were given at varying intervals post
training. He observed what appeared to
be a consolidation gradient when he administered ECS
at varying intervals ranging from 20 sec. to 14 hrs. after the active avoidance task, in which he
conditioned rats to associate a light with a foot shock. Duncan observed that the ECS disrupted
the rat’
s ability to remember the association when tested post
treatment. He noted that the
disruptive effect more robustly the sooner the ECS was administered after the light
pairing. A retroactive amnesia was observed in the conditioned rats when ECS was

given up to
15 minutes after conditioning and no disruptive effects were found if the ECS was given more
than 60 minutes after conditioning (see Duncan, 1949). This suggested that consolidation lasted
for up to an hour. This time estimate generally supp
orted Müeller and Pilzecker’s proposed time
course of 10 minutes in that they both emphasized a finite amount of time post
experience in
which consolidation processes were occurring. Many similar experiments followed in the


attempt to determine the precis
e time course for the consolidation process; however, they were
not limited to the use of ECS. Other methods have included, but have not been limited to, using
anoxia (e.g., Hayes, 1953), temperature changes (Gerard, 1955), brain stimulation (e.g.
n, 1958), and protein synthesis inhibition (Davis & Squire, 1984).

The observed temporal gradient implied that there was a limited amount of time in which
memories remained fragile or susceptible to disruption and therefore consolidation processes
were a
ssumed to be limited by a certain amount of time. However, it remained unclear as to
whether this time
frame in which consolidation took place was universal for all types of
memory, nor was it clear as to where in the brain consolidation processes were oc
curring, or
where the memories were stored, if they were at all. Additionally, it was not apparent whether
the observed deficits were due to the failure of consolidation processes or retrieval dysfunction.

It was first believed that the temporal
gradient reflected in the RA of these animals was
indicative of a time course for memory consolidation; however, there were inconsistent results
that depended on the methodological differences of each experiment (see figure 1 adapted from
Chorover, 1976).

Russel (1959) reported “islands” of preserved memory during recall for
remote events in trauma patients. Some studies suggested the processes underlying
consolidation take place within 500 msec of learning (e.g., Lewis, Miller & Misanin, 1969), and

in which there was a flat gradient or an ungraded RA, indicated that the process extended
approximately 40 years (Sanders & Warrington, 1971), suggesting consolidation lasted close to
an entire lifetime. Chorover (1976) noted that most of these differenc
es could be attributed to
the strain, sex, and previous experience of the subject; the nature and complexity of the learning
task; the kind of reinforcement used; the physical parameters of the stimulus and/or the mode of


treatment administration (Albert,
1984). Additionally, the observed gradient differences may
be linked to the differences in allocation of attentional resources during learning. Most research
tends to support that the attentional resources we allocate to a particular experience are
portional to the emotional intensity or vividness associated with it. By recruiting attentional
and/or emotional brain mechanisms, the likelihood of creating a more robust LTM from the
experience is great, so much that James (1890) suggested, “an impressi
on may be so exciting
emotionally as almost to leave a scar upon the cerebral tissues” (p. 670). If less attentional
resources were allocated, it would foreseeably make the learned material or those particular
experiences more likely to be lost as observe
d in examples of RA following ECS . Furthermore,
McGaugh and Gold (1976) suggested that memory disruption experiments do not provide any
information about the underlying memory process itself, nor do they specify the amount of time
required of long
term m
emory consolidation, but rather address the susceptibility of the memory
to disruption.

The perseveration
consolidation hypothesis, and specifically retroactive inhibition
became obsolete as they were replaced by interference theory in which learning of
material too close in time will compete for representational space (McGeoch & Irion, 1952).
Notably, the idea of consolidation and its requisite temporal gradient suggested by Ribot,
Müeller and Pilzecker, was revitalized by Donald Hebb’s (1949) d
trace hypothesis. Hebb
proposed a clear neurological distinction or fractionation of the information at hand, with the
information that was necessary for storage. The fractionation clearly defined two stages of
memory formation, a labile short
term s
tore (STS) and a fixed, long
term store (LTS). Hebb
suggested, as did Gerard (1955), that a pattern of simultaneously activated, interconnected cells,


termed a “cell assembly”, represented the temporary STS of a particular experience. If the
g activity within the cellular assembly persisted for a sufficiently long amount of

Figure 1 (adapted from Chorover, 1976): Multiple retrograde gradients.

A wide range of “consolidation curves” were obtained using ECS in a variety

of tasks, species
, and methodology. PA: passive avoidance



time, reciprocal connections would then become stronger and more efficient, creating lasting
changes in synaptic connections, and thus providing a cellular mechanism for creating an
enduring memory trace
or LTS. Gerard (1955) used the metaphor of a single water drop that
leaves a sandhill undisturbed to demonstrate the passage of an impulse through a synapse. He
further suggested a series of drops produces a runoff channel as repeated circulating activit
through cellular assemblies may represent the production of the enduring memory trace (Gerard,
1955). Gerard also theorized on the localization of the memory trace. Expanding on the
observation that extensive localized lesions were unlikely to abolish
memories, he suggested that
cellular assemblies were not represented by a “single loop, but a multiply linked network of
loops…[and] a memory would not be entirely lost unless all parallel paths of the net were
destroyed” (Gerard, 1955, p. 229). Gerard wa
s arguing that 1) memories are represented by
multiple pathways in a distributed network of cell assemblies and 2) memories can be elicited by
exciting either one of the particular pathways alone or exciting a degraded version or subset of
the cell assembl
y. This completion of the entire cellular assembly later became known as
“pattern completion” (see Marr, 1971; McNaughton & Morris, 1987). Both Gerard and Hebb
influenced the future conception of the cellular component to memory, namely long
ation (LTP). LTP and the associated forms of synaptic plasticity currently provide the
most comprehensive framework for the study of cellular consolidation theory. However, at the
time, their conception of reverberating circuitry was limited to short ter
m consolidation; it
remained unclear how the dual
trace hypothesis dealt with longer consolidation gradients


extending many years. Similarly, the disruption experiments mentioned previously were not able
to clarify an agreed upon time frame for consolidat
ion. Such limitations gave rise to a manifold
of neurobiological approaches to examining the underlying neuroanatomical, cellular and
molecular components that were responsible for memory consolidation.

The medial temporal lobe

In 1958, Wilder Penfiel
d stimulated particular locations of epileptic patients’ brains prior
to ablation of seizure
prone areas. He found that electrical stimulation of the temporal lobe
produced complex sensations that resembled hallucinations or vivid recollections of experie
Penfield’s studies suggested that particular memories may be stored in the temporal lobe. Long
before Penfield, around the time of Müeller and Pilzecker, Von Bechterew (1900) suggested that
memory impairment could be result from damage to the media
l temporal lobe (MTL). He
demonstrated that a 60
old patient with bilateral damage to the piriform and periamygdaloid
cortices as well as the underlying amygdaloid and hippocampal complex showed memory
problems that were limited to the last 20 years
of his life (Zola
Morgan, Squire, and Amaral,
1986). Glees and Griffith (1952) anecdotally reported considerable anterograde and RA that
extended many years in patients with bilateral damage to the hippocampus and surrounding
cortex (Zola
Morgan et al., 1
986). The MTL has usually been reported to comprise the
hippocampus proper (CA subfields and dentate gyrus) as well as the subiculum, surrounding
neocortical regions (entorhinal cortex, prahippocampal gyrus, perirhinal cortex), and amygdala
(Eichenbaum, 2
002; Haist, Gore, & Mao, 2001; Squire, 1992; Shimamura, 2002).

The importance of the MTL in consolidation theory became widely accepted following a
seminal case study of a patient, H.M. who had a bilateral temporal lobectomy (~ 5 cm.
rostrocaudal) due to i
ntractable seizures (see Corkin, Amaral, Johnson, & Hyman, 1997).


Scoville and Milner’s reports on H.M. described a severe anterograde and temporally graded RA
(see Corkin, 1984, 2002; Scoville, 1954; Scoville and Milner, 1957). Particular experiences
cently formed prior to surgery were more severely disrupted than remote memories H.M. had
formed years earlier. Similarly to Ribot’s original findings, H.M. demonstrated a retrograde
deficit in memory for public and personal events that extended back almo
st 11 years prior to
surgery. The original paper (Scoville & Milner, 1957) described the findings for 9 schizophrenic
patients along with H.M., all of whom received types of medial temporal lobe ablations that
included most notably, the rostrocaudal porti
ons of the hippocampus. Based on their findings,
the authors concluded that the memory deficit and associated consolidation disruption was due to
the bilateral removal of the hippocampus proper and parahippocampal gyrus (Corkin et al.,

The portio
n of brain more recently determined to be removed from H.M. included the
amygdala, the anterior two
thirds of the hippocampus, and the surrounding cortex, including the
piriform gyrus, uncus, and parahippocampal cortex. Recent neuroimaging findings of H.M
demonstrate that the most dorsal portion of the amygdala (including the central nucleus) remains
intact as do parts of the posterior hippocampus; however, virtually all of the entorhinal cortex
(bilaterally) was removed (Corkin et al., 1997). Those port
ions of the temporal lobe that had
been removed produced little if any effect on H.M.’s perceptual, motor, and cognitive functions,
in fact, he performed similarly to controls on such tasks (Eichenbaum, 2002; James & MacKay,
2000). Further, his intellige
nce, as measured by the Wechler
Bellevue scales, increased slightly
surgery. The increase has been proposed to be due to the alleviation of his seizures.
Although his spelling is poor and he has slight deficits in verbal fluency, H.M.’s language ski
remain preserved as does his ability to recognize and name common objects. Additionally, H.M.


maintained the ability to perform perceptual priming tasks as well as demonstrating intact
immediate or short
term working memory ability. For example, he c
ould reproduce a list of
numbers as well as controls as long as he was allowed to actively rehearse. The memory deficit
would become evident as soon as there was a delay in rehearsal and his attention was distracted
or his memory span was exceeded (see Sc
hacter, 1987; Squire, 1987, 1992, Corkin et al., 1997).
He also sustained a deficit in odor identification and discrimination, preserving his ability to
discriminate intensity differences. It was suggested by Eichenbaum, Morton, Potter, & Corkin
(1983) t
hat the olfactory deficits were due to ancillary piriform cortex damage.

H.M. has now been studied extensively for almost 50 years, along with other patients
with less extensive bilateral medial temporal lobe damage. All patients have been assessed on a

multitude of memory tests that have elucidated the particular role each structure within the MTL
has on memory consolidation as well as the type of memory specific to each memory system.

About 25 years after H.M. was first studied, the successful dev
elopment of animal models of
human amnesia were able to control for discrepancies in the human MTL amnesics, in which
there was sparing of hippocampal tissue as well as damage that extended well into the amygdala
and surrounding parahippocampal, and rhinal

cortices (Mishkin, 1978; Mishkin & Murray, 1994;
Morgan & Squire, 1993). It was also very difficult to assess definitive memory
impairments due to specific MTL damage, in that the memory content being assessed involved
experiences that happened prio
r to brain damage. There was no distinct way to know which
memories were stored and to what extent; thus, it was difficult to assess what had been retained
and what had been lost. Demonstrating similar deficits to those observed in H.M., rats,
monkeys, a
nd humans with damage restricted to the hippocampus (Zola
Morgan et al., 1986;
Clower et al., 1996) suggested that the hippocampus proper is responsible for much of


the mnemonic processing of the MTL. Functional imaging has confirmed this view (see

et al., 1992) in addition to the supportive findings in which patients with damage limited to the
amygdala failed to show any amnesic qualities (Zola
Morgan, et al., 1986). The hippocampus
has also been associated with memory impairments that are
a result of viral encephalitis
(Damasio et al., 1985; Rose and Symonds, 1960), posterior cerebral artery occlusion (Benson et
al., 1974), and hypoxic ischemia (Volpe and Hirst, 1983). However, H.M’s deficits along with
other amnesics have not been limited

to the hippocampus proper, which has left open for
speculation the precise contribution of the hippocampus as well as the length of time it is
involved in the consolidation process. It is apparent that the hippocampus is involved in
consolidation; howeve
r, related structures within the medial temporal lobe (e.g., entorhinal
cortex perirhinal cortex, parahippocampal gyrus, the amygdala, and medial thalamus) may
complement the hippocampal formation in the consolidation process.

The use of animal models not

only allowed restricted lesions to the areas of interest, but
has also controlled for the particular learning experience prior to the lesions. An entire array of
dissociations between spared and impaired aspects of memory became more apparent in the
sia models. The initial question that the animal models attempted to answer was whether or
not a RA gradient that was similar to the temporal gradient seen in H.M. and subsequent other
clinical cases, would be apparent. In humans, RA may be short, only a
ffecting months to a few
years before damage. This pattern was demonstrated by H.M. as well as in studies that
administered ECS (Squire, Slater & Chace, 1975; Cohen & Squire, 1981). Other gradients were
flat and affected recall of events that extended fo
r decades without a temporal gradient. This
type of gradient was shown in a variety of neurodegenerative disorders including multiple
sclerosis (Beatty et al., 1988), Huntington’s disease (Albert, Butters, & Brandt, 1981),


Parkinson’s disease (Sagar, Sull
ivan, Gabrieli, Corkin & Growdon, 1988), frontal lobe dementia
(Hodges & Gurd, 1994), as well as Korsakoff’s syndrome as a result of chronic alcholism.
Hodges (1995) presents evidence that Korsakoff’s patient’s have demonstrated extensive,
temporally grad
ed RA, but that it may be secondary to the primary defect in laying down
(encoding) new memories during the extensive period of alcohol abuse that preceded the
amnestic syndrome. Accompanied by a severe anterograde amnesia, minimal, ungraded RA was
ed in a hypoxic patient, R.B., who attained a discrete lesion to the CA1 region of the
hippocampus (Hodges, 1995; Zola
Morgan, Squire, Amaral, 1986). Interestingly, R.B. showed
deficits on tests of verbal and nonverbal memory function (i.e., prose passag
e recall with delay;
Osterreith figure copy with delay) with spared abilities in autobiographical episodes (i.e.,
public events; famous faces). It appeared that his impairments were very limited to recall after a
significant delay (see Zola
Morgan et
al., 1986). R.B.’s lesion may have been confined to the
CA1 region with no cortical abnormalities; however, it blocked a major output of the tri
syanaptic loop, from the hippocampus to the subiculum and entorhinal cortex, and therefore may
have contribute
d to the significant anterograde amnestic deficits. Two other patients, A.B. and
L.J., with lesions limited to the entire hippocampal formation demonstrated an anterograde
semantic or factual knowledge impairment for vocabulary, recognition of famous peop
le, and
public events with a relative sparing of such information retrograde (Reed & Squire, 1998).
With more extensive damage to the hippocampus and portions of the entorhinal cortex, W.H. and
L.M. demonstrated extensive temporally
graded RA extending 15
25 years (Squire & Alvarez,
1995; Squire et al., 2001). Similarly to H.M., with damage involving the entire MTL (including
lesions that extend laterally to the fuisform gyrus), E.P. had extensive RA that covered 40


years with a relative sparing of fac
tual information and autobiographical episodic memories from
early on in life.

Temporally graded amnesia has been observed in multiple animal models including rats,
mice, and non
human primates (see table 2 in Brown, 2002; table 1 in Murray & Bussey, 200
table 2 in Nadel & Moscovitch) and the findings have shown to be similar to the human data in
that temporally graded amnesia becomes more extensive as damage extends beyond the
hippocampus proper (Squire et al., 2001). In 13 experimental animal studie
s that involved either
lesions of the hippocampus alone or the entorhinal cortex, the extent of RA was found to range
from about 5 days to a month or more (see Squire et al., 2001). In a study that lesioned both the
perirhinal and entorhinal cortex, no te
mporal gradient was observed; monkeys were impaired in
the retention of object discrimination whether they learned 1 or 16 weeks prior to surgery
(Thornton et al., 1997). Additionally, Salmon, Zola
Morgan & Squire, (1987) demonstrated that
monkeys with la
rge MTL lesions (including the amygdala, hippocampus, perirhinal and
parahippocampal cortices) were impaired in retention of 100 object discriminations learned 2
weeks prior to surgery. Such data agrees with Squire’s (1992) argument that the length of
gradient can be expected to vary depending on the extent to which the medial temporal lobe is
damaged and on the normal rate of forgetting for the material being tested.

Following from the multitude of neuropsychological evidence surrounding the MTL,

more specific memory function for the hippocampus proper began to emerge which included
descriptions such as conscious, episodic, declarative, explicit, configural, or relational
(Eichenbaum, 1997; O’Keefe & Nadel, 1978; Schacter & Tulving, 1994; Squire
, 1992;
Sutherland & Rudy, 1989; Tulving, 1972). The types of information represented by these
descriptions include a conscious recollection or knowledge about personal events, facts about the


world, contextual associations (including spatial and temporal

attributes) and the flexible
expression between these types of information (Cohen, 1984; Eichenbaum, 2002; Squire, 1992).
The type of memory that is spared in the majority of neuropsycholgocial cases has been
described as implict, unconscious, non
ative, procedural or memory that represents
motor skill alone (Mishkin & Petri, 1984; Schacter & Tulving, 1994; Squire, 1992).
Given that these varying types of memory processing have each been shown to be dependent on
different brain systems (
see Cohen & Squire, 1980; Eichenbaum, 2002), it could be argued that
the varying retrograde gradients are due to the different memory processes being assessed
depending upon the extent of damage found in the MTL (see Tulving & Markowitsch, 1998).
More spe
cifically, deficits for autobiographical episodes containing rich contextual information
about one’s life (e.g., wedding, birthday, graduation) have appeared to be most affected by
damage to the hippocampal region in humans. Thus, with large hippocampal l
esions, the
gradient of RA for these particular autobiographical episodes is relatively flat or non
extending as much as 25
40 years, and in some cases encompassing the subject’s entire lifetime
(see table 1 in Nadel & Moscovitch, 1997; table 2 i
n Brown, 2002). Memory for public, or
popularized information in the media (e.g., famous faces), often exhibited a graded retrograde
deficit and was less affected than autobiographical information. The least affected memories
have appeared to be related
to personal semantics or knowledge not tied to any specific episode
(e.g., familiar places, names of friends and family) followed by overlearned semantic memories
(e.g., vocabulary, grammar, factual knowledge from textbooks) (Shimamura, 2002). One major
roblem with the assumption that retrograde gradients are indicative of the length of time
associated with consolidation is that most people would not consolidate any autobiographical
information before they die (Nadel & Moscovitch, 1997). Thus, some model
s of consolidation


that use observed retrograde gradients as evidence for support (see Alvarez & Squire, 1994;
Squire & Zola
Morgan, 1991; Squire, 1992) fail to account for such a discrepancy. The
hippocampus as a unitary structure is currently not though
t to be responsible for all consolidation
processing, nor is it a permanent repository for past experiences. Another discrepancy is the
variability associated with hippocampal damage observed in anterograde deficits assessed after
classical fear condition
ing (see Maren, Aharonov & Fanselow, 1997 and Fanselow, 2000). It
was suggested (see Rudy & O’Reilly, 2001) that there might be some form of hierarchical
processing where an intact rat is biased to use a hippocampal system; however, in cases in which
hippocampal system is damaged, extrahippocampal structures within the MTL may be
sufficient to encode particular contextual features associated with an aversive stimulus. It is
clear that several other kinds of learning and memory are possible in, and may

compensate for,
the absence of the hippocampus (see Eichenbaum, 2002; Squire, 1992).

The parahippocampal region

If there is an underlying assumption that MTL structures outside of the hippocampus
have multiple functions independently, but contribute colla
terally in all aspects of memory
consolidation, it is important to consider the contributions of each independently. Anatomical
and physiological studies have suggested that all sensory information enters the hippocampus
through the lateral entorhinal cor
tex via the lateral perforant path; whereas, other limbic
structures project information through the medial entorhinal cortex via the medial perforant path
(Riedel & Micheau, 2001). Eichenbaum (2002) proposes that the parahippocampal region
(including ent
orhinal, perirhinal, postrhinal and parahippocampal cortex) serves as a
convergence center for neocortical inputs and serves as a bridge mediating the flow of
information between cortex and hippocampus in times of reorganization or modification. The


ippocampal region has heavy projections to multiple cortical association areas (i.e.,
prefrontal, parietal, temporal) with back projections into the subdivisions of the hippocampus.
Squire et al. (2001) indicate that the hippocampus itself may be importan
t for a relatively short
amount of time after learning and that adjacent perirhinal and parahippocampal cortices may
remain important for longer periods. Young, Otto, Fox & Eichenbaum (1997) demonstrated that
neurons in the parahippocampal region may supp
ort recognition performance of recently
sampled odors, in that they maintain or regenerate their activity during an extended delay period
in which rats are required to detect match/mismatch odor presentations. Murray, Gaffan &
Mishkin (1993) demonstrated
that hippocampal lesions post
learning of visual stimulus
associations (90% criterion over 6 weeks) yielded no significant deficits in monkey’s retention;
however, other monkeys trained on the same paradigm showed extremely poor retention after
mbined lesions of the amygdala, hippocampus, and surrounding perirhinal, entorhinal, and
parahippocampal cortices. Tulving & Markowitsch (1997) reported that memory for objects
depends on the perirhinal cortex, and that memory for object
location relation
ships is mediated
by the parahippocampal region. Jarrard (2001) warns that observed functional differences
between the functional components of the MTL may be due to using variable, selective lesioning
techniques (i.e., electrolytic vs. ibotenic acid). D
ifferences have also been shown within the
same species using different learning tasks (radial arm maze vs. Morris water maze). In
particular, separate lesions of the hippocampus and pre

and parasubiculum impair only the
utilization of spatial informatio
n in a working memory task, whereas lesions of the perirhinal,
postrhinal, and entorhinal cortices impair only nonspatial, cue information in reference memory
processing (Jarrard, 2001). However, deficits outside the hippocampus are not always consistent
across species. Temporally graded retrograde deficits were observed in rats trained on visual


object discrimination tasks with lesions limited to perirhinal (Wiig , Cooper, & Bear, 1996) and
rhinal cortex (Kornecook, Anzarut, Pinel, 1999); whereas, with s
imilar lesions to the perirhinal
plus entorhinal cortex in monkeys, there was a flat, ungraded RA (Thornton, Rothblat, &
Murray, 1997). Cho & Kesner (1996) reported a temporally graded effect on retention of spatial
discrimination with electrolytic lesion
s to the entorhinal cortex, subiculum, and pre

parasubiculum. On the other hand, Bolhuis, Stewart, and Forrest (1994) showed a flat gradient
for spatial information with ibotenic lesions to the subiculum.

The amygdala

Another MTL structure that pla
ys a major role in modulation of memory is the amygdala,
which appears to be significantly involved in memory for emotionally arousing experiences
(McGaugh et al., 2000). More specifically, the basolateral amygdala (BLA) and associated
related horm
onal systems have been attributed to modulating the strength of the
consolidation process in the multiple memory systems that have been proposed (see figure 2
adapted from McGaugh, 2002) (Gerard, 1961; McGaugh, 1966, 2002). Kapur (2000) suggests
the initi
al strength of an experience determines the temporal extent and severity of the RA,
which, for example, can be attributed to conceptual familiarity or emotional significance. Stress
hormones (e.g., glucocorticoids, corticotropic releasing hormone) play a
n important,
driven role in modulating the activation of the sympathetic nervous system in order
to mobilize energy stores in response to a particularly threatening experience. Threatening
experiences tend to be avoided in the future; thus, a
n associated release of stress

hormones is part of the functional mechanism involved in consolidating the memory for that
particular experience strongly. Notably, the consolidation of such memories rarely requires an
explicitly conscious awareness or for
m of cognitive processing.



n humans as well as in animals (Soetens, Casaer, D’Hooge & Hueting, 1995), stimulant
drugs or drugs affecting hormonal systems (e.g. adrenocorticortical hormones, adrenaline)
administered shortly after learning, either periphe
rally or directly into the amygdala, have been
shown to improve long
term memory (McGaugh 1973, 1983), which has implicated the
amygdala in modulating emotional memories. Many types of learning (Izquierdo & Diaz, 1983;
Sternberg, Isaacs, Gold, McGaugh, 1
985) have found to be influenced by the use of drugs that
mimic the arousal, stress, and anxiety associated with intense emotional experiences. The
majority of investigations examining the influence of stress hormones in the amygdala have
focused on an in
hibitory avoidance task in which rats are initially placed in a well
lit chamber
attached to a larger dimly
lit area. Following the opening of a door that separates the chambers,
the rat will typically cross over into the larger area where the floor is el
ectrified. After a series of
brief foot shocks, the animal is allowed to escape back into the original chamber. Retention is
tested later on (immediately to 48 hours post
learning) by measuring the latency to step into the
larger chamber in which the ave
rsive stimulus was presented. A dose
dependent, time
enhancement of consolidation is observed when noradrenaline (norepinephrine), and other
glucocorticoid agonists are infused directly into the amygdala shortly after inhibitory avoidance

(McGaugh, 2002). In other words, rats that took longer to step into the chamber
associated with the aversive stimulus during the retention test demonstrated an enhanced or more
robust and long
term memory for the initial aversive event. Administration o

receptor antagonists directly into the amygdala have been shown to impair retention when
administered immediately after training, but have no disruptive effect when administered 6 hours
after training (Gallagher, Kapp, Musty & Driscoll, 1977,
1981). Retention after training has
been found to be enhanced by low
intensity stimulation and disrupted by high


stimulation of the amygdala (Kesner & Wilburn, 1974; McGaugh & Gold, 1976). McGaugh
(2000) suggests these stimulation studies indic
ate that mild activation of the amygdala in an

Figure 2: Projections from BLA to other areas involved in consolidation. (Adapted from
McGaugh, 2002).



endogenous manner may enhance memory consolidation similarly. Furthermore, it has been
demonstrated that lesions of the stria terminalis (a major amygdala pathway to multiple brain
regions), or the amygdala block the facilitating effects of adrenaline or glucocorticoids on
consolidation (Cahill & McGaugh, 1991; Liang & McGaugh, 1983). Ad
ditionally, Packard and
Chen (1999) demonstrated that the amygdala also blocked the glutamatergic enhancement of
consolidation occurring in the hippocampus by inactivating the amygdala with lidocaine.
Lesions of the BLA or nucleus accumbens (NAc) has also

been shown to disrupt the
enhancement of memory consolidation induced by infusions of glucorticoid
receptor agonist into
the hippocampus (Roozendall & McGaugh, 1997). Furthermore, infusions of lidocaine into the
caudate have been shown to prevent the amp
induced enhancement of caudate
dependent learning (Packard & Teather, 1998).

McGaugh (1966, 2000, 2002) and others (see Polster, Nadel, & Schacter, 1991)
have proposed that the nuclei of the amygdala are involved in multiple forms of learnin
g and
memory, including attention (Holland & Gallagher, 1999), cue
, place

and object
associations (Holland & Gallagher, 1999; White & McDonald, 2002; Easton & Gaffan, 2000;
Baxter & Murray, 2000; Salinas, Introini
Collison, Dalmaz, McGaugh, 1997),

conditioned taste
aversion (Lamprecht & Dudai, 2000), appetitive conditioning and drug addiction (Everitt,
Cardinal, Hall, Parkinson & Robbins, 2000). Even more specifically, McGaugh (2002) proposes
the BLA may be the locus of neuroplasticity underlying
consolidation of Pavlovian fear
conditioning. Post
training intra
amydala infusions of drugs affecting the GABA, opioid, as well
as muscarinic acetylcholine receptor systems in the amygdala have also demonstrated dose

dependent influences on mem
ory consolidation (McGaugh, 2002). It has been extensively


demonstrated that the BLA
NAc pathway is essential for the modulation of consolidation in
multiple memory systems in association with adrenocorticoid receptor activation. Although the
y of research on memory modulation is based on the inhibitory avoidance task, there have
been similar findings in discrimination and maze learning tasks (Eichenbaum, 2002).
Additionally, studies conducted on human subjects have provided evidence that emot
arousal can affect different forms of memory and that this effect can be mediated by the
amygdala (Eichenbaum, 2002). Although the evidence presented is not exhaustive by any
means, it is basis enough to suggest that through the multitude of conver
ging inputs, the
amygdala demonstrates abilities that facilitate the consolidation of memory along the dimensions
of emotional and physiological arousal as well as attachment of perceptions to emotions in
collaboration with the multiple memory systems in e
xistence. It is clear that the experiences that
form lasting, long
term memories occur gradually over time and after recruiting multiple areas of
the MTL, become diffusely represented in multiple areas of the brain.

The transition from a short
term to a

term memory store

The transition from a short
term to a long
term memory store assumes a particular
experience is consolidated or “fixed”, in which particular neural mechanisms are elicited to bind
information into a “cohesive” permanent memory trace

(Eichenbaum, 2002; Fujii, Moscovitch,
& Nadel, 2000; Nadel & Moscovitch, 1997). The logic of consolidation theory rests upon the
idea that the hippocampus is important for the initial registering or storage of information for a
limited amount of time.
If we continue with the assumption that memories are being formed
gradually over time, they must then be sent elsewhere for storage. Thus, the hippocampus,
amygdala, diencephalon and associated cortices play a time
limited role in the cohesion process,
til long
term consolidation is complete and a permanent memory is distributed throughout


neocortex. Bontempi, Laurent
Demir, Destrade & Jaffard, 1999) demonstrated in mice that
retention shortly after learning (5 days) depended on the hippocampus through
imaging. With time (25 days), it became dependent on other neocortical structures (frontal,
anterior cingulate and temporal). This assumption is further supported by the observation that
damage to the MTL, specifically the hippocampus pro
per, will typically impair recent memories
and leave more remote memories intact (Alvarez & Squire, 1994). This has been consistently
found across many species and multiple tasks. Additionally, there is the apparent time
enhancement of consolidat
ion when the MTL is infused with norepinephrine
like agonists post
learning and a time
limited disruption when ECS or protein synthesis inhibitors are administered
learning. It has been suggested that the surrounding cortices serve an intermediary ro
between the hippocampus proper and association areas of cortex (e.g., temporal and parietal
cortex), which are thought to be permanent repositories of LTM (Squire et al., 2001). How then
does consolidation theory describe the interaction between cortex

and the MTL in order for
information to be distributed widely as a more stable and permanent form of long
term memory?
Moreover, it was questioned whether the hippocampus directed the processes of consolidation or
whether there was some endogenous hippo
campal activity that enabled consolidation in a
nonspecific way.

Marr (1971) originally proposed that the hippocampus could act as a temporary store for
what he termed, “simple memories”. His computational theory predicted that the neocortex
served as a pe
rmanent store for which converging inputs to the hippocampus represented the
same information temporarily. It was later suggested that the hippocampus served to “imprint”
a memory that was activated or represented by higher order cortical association are
as by what
Marr (1971) referred to as an auto
associative memory mechanism. He explained imprinting in


terms of a strengthening or perseverative process using the activation of recurrent collaterals
existing between CA3 hippocampal pyramidal cells with ex
citatory feedback connections to
carry out what later surfaced in cognitive terms as “rehearsal”. Another suggested role of the
hippocampus has been one of an “index” (see Teyler & Discenna, 1986), in which a pattern of
cells that become coactivated with

neocortical cells in response to a particular event become a
retrieval index for the future recall of that particular event. The theory supposes a bi
connectivity between the hippocampus and neocortical representation and implies future
anization and strengthening of the stored memory with periodic reactivation.

The standard model of consolidation (see figure 3a) proposed by Alvarez & Squire
(1994) posited that MTL structures can initially mediate storage and retrieval of the memory
ce; however, as the consolidation process continues, the contribution of the MTL structures
are no longer involved and the neocortex alone can maintain the memory trace and mediate its
retrieval. It was anticipated that once the connections between the co
rtical association areas have
been strengthened to a particular threshold the cortical representation would be sufficient to
support the maintenance of the memory without the help of the hippocampus or the rest of the
MTL. The cerebral cortex is capable
of storing large amounts of information that is represented
by a diffuse pattern of overlapping multiple cortical areas. According to the standard model,
these multiple areas are interconnected with synapses that change their weights slowly and
lly. The changes in connections between the MTL and the association cortex were
designed to be rapid, but short
lived given that hippocampal synapses are “soft” and capable of
changing quickly (Alvarez & Squire, 1994, 1995). Because of the rapid synaptic

changes that
occur in the hippocampus, Alvarez & Squire (1994, 1995) proposed it would be able to serve as


a temporary memory store until the hippocampal system repeatedly activated the same
connections (representations) in neocortex, leading to a strong
interconnected network

Figure 3a. The standard model of consolidation. Areas cortex1 and cortex2 represent
association neocortex. Each unit represents a simplified neuron. Bidirectional cortico
connections change much faster than cortical
cal connections. Adapted from Alvarez &
Squire, 1994.



Figure 3b. Performance of the standard model in a RA experiment. The intact model and a
disconnected or “lesioned” MTL are both tested on their ability to reconstruct a pattern
depending on the am
ount of time that has passed between training and testing.

diffusely represented in multiple cortical sites. The model includes two distinct association
cortical regions (cortex 1 & 2) that communicate with one MTL region. Each region consists of

units (4 in MTL and 8 in cortex 1 & 2) that are reciprocally connected to all units in all
other areas and the connection strengths are modified by a use
dependent, Hebbian, competitive
learning rule. McNaughton (1989) describes the MTL representations
as a “summary sketch” or
as being compressed versions of the patterns in neocortex, not necessarily a direct copy of the
pattern of activation distributed over numerous regions of neocortex.

Similar to Teyler and DiScenna’s model (1986), the standard mod
el predicts that the
MTL acts as an “index”, in that the pattern of activation in the cortical
cortical connections
activates a group of hippocampal cells, which in turn become linked together. When the MTL
area is randomly stimulated it acts as a retriev
al index activating the original cortical areas.
Similar to typical retrograde gradients, the standard model succeeds at predicting the inability of
a lesioned MTL to complete a particular pattern from the presentation of a partial pattern shortly
after l
earning. Further, as demonstrated by figure 3b, after a few more pairings, a lesioned MTL
would no longer prevent the network from reconstructing the pattern. “Normal” networks can
reconstruct recently learned patterns better than ones that are learned m
ore remotely, whereas
“lesioned” networks perform better for patterns learned at a remote time period than for patterns
learned more recently (Alvarez & Squire, 1994, 1995). The length of time for the MTL
connections to be active before the trace

was supported sufficiently by cortical


connections was unclear. It was suggested by Marr (1971) that the reactivation might occur
during the night. Both Buzsaki (1989) and Wilson and McNaughton (1994) elaborated on, and
have provided evidence t
hat, hippocampal place cell activity that fired together during the day
would again fire together during slow
wave sleep. Additionally, it has been suggested that initial
hippocampal activity necessary for encoding takes place during theta rhythm oscillat
ions, and the
subsequent transfer to neocortex during sharp waves observed in quiet waking as well as slow
wave sleep (Buzsaki, 1998; Hasselmo & McClelland, 1999). This type of spontaneous
reactivation of memory traces may prove to be a necessary step in
the gradual selection of
synaptic weight distribution that may incorporate recent experiences into a long
term storage
base (Sutherland & McNaughton, 2000). McClelland et al. (1995) have predicted that during
recall or during a similar experience, a part
of the pattern representing the original experience is
activated in the neocortical system leading to the activation of the initial or compressed pattern
of cells in the hippocampus and a subsequent convergence of the overlapping patterns followed
by assoc
iative plasticity. It is not clear whether this type of reactivation may act as
supplementary training or as part of a novel pattern of activation. The plastic changes associated
with reactivation would tend to make this pattern what McClelland et al. (1
995) refer to as an
“attractor”, or a pattern toward which neighboring patterns or incomplete versions of the pattern
will tend to converge (McClelland et al., 1995). The time in which a particular memory is
consolidated would then be determined by the de
gree of reactivation of the experiences as well
as the convergence of multiple overlapping patterns of neocortical activation. Notably, these
overlapping networks of MTL
cortical activity are defined intrinsically by the type of
information and its interr
elatedness to previous representations. Additionally, the modulation of


such connectivity is highly dependent upon the MTL structures involved and level of arousal
associated with the experience.

The initial fixation of information, or cohesion process, i
n which the standard model
refers can be applied fairly well to gradients of consolidation lasting seconds, minutes, or hours
until what may be considered a long
term stage of consolidation begins which involves gradients
that can last hours to days, month
s, or even years. As suggested previously, the extent of RA in
animals and humans is dependent on the extent of damage to the hippocampus proper and
associated MTL structures. When particular tasks absolutely require the hippocampus for
learning (e.g., s
patial/temporal tasks), there is most often a flat retrograde gradient in response to
a complete hippocampal lesion. When the task is one in which the hippocampus is not essential
(e.g., object discrimination), there is typically a temporally graded or ab
sent RA (Nadel &
Moscovitch, 1997). In order to account for the temporal gradients lasting more than 30 years for
particular types of autobiographical memory, as well as the common finding in which there is
preserved semantics (vocabulary) during periods
for which subjects have RA, the “Multiple
Trace Theory” (MTT) was proposed by Nadel and Moscovitch (1997, 1998). MTT proposes that
the establishment and maintenance of particular episodic memories involve both the
hippocampus and neocortex and that each r
eactivation leads to a new encoding event in the
hippocampus. MTT was formalized in a computational model introduced by Nadel,
Samsonovich, Ryan, and Moscovitch (2000). Unlike the Alvarez and Squire (1994) standard
model, MTT suggests that increased repr
esentations (multiple traces) take place in the MTL
rather than the neocortex and that general semantics are learned and represented outside the
hippocampal complex (or MTL) (Nadel & Moscovitch, 1997). Alternatively, “Relational
Binding Theory” (RBT) prop
osed by Shimamura (2002) suggests that the MTL, by way of


relational binding of cortico
hippocampal connections, facilitates both the access and retrieval of
recently bound associations or remote memories (see figure 4). Shimamura (2002) further
this process is not qualitatively different for semantic or episodic memories. Relational
memory implies that representations present at the time of encoding act as a context in

Figure 4. The role of MTL across multiple consolidation theories. Soli
d lines and filled circles
represent cortical networks that are activated during retrieval between association cortex and the
MTL. Standard consolidation theory supposes that remote memores are solely represented by
cortical connections. MTT sup
poses that remote (episodic) memories are established by
traces coded within the MTL and are always necessary for retrieval. According to RBT, remote
memories have fixed cortical
cortical connections, but retrieval is facilitated by connections to
the MTL

(represented by dashed lines). Adapted from Shimamura, 2002



which previous memories are associated (see Cohen & Eichenbaum, 1993 for similar
descriptions). In addition, RBT suggests that the ultimate formation of cortical
connections is ba
sed upon the frequency with which representations are reactivated. The MTT
implicates the MTL
cortical connections in the initial consolidation of episodic types of memory
and in their replication in the form of multiple traces. Each new encoding event
then causes a
creation of a new memory trace proliferating a highly distributed representation of a particular
event and making the memory particularly resistant to disruption following damage to the
hippocampal system. To account for semantic information

not represented in MTT, Moscovitch
and Nadel (1999) later applied MTT to semantic dementia, a degenerative disorder of the
posterior and lateral temporal lobes sparing the hippocampal complex. Damage to these areas is
reported to lead to semantic word lo
ss with relative sparing of episodic memory and of words
that are associated, and therefore dependent on, the reactivation of the related episodes (Nadel &
Moscovitch, 1997; Moscovitch & Nadel, 1999). Thus, if these areas are spared, RA for these
types of

information should be absent.

Regardless of whether or not MTT or RBT fit the data more accurately, it appears that
similar contexts with partial neocortical representation may reactivate the representative trace of
particular memories in which the hipp
ocampus played an active role in creating, thus
perpetuating a more diffusely represented organization of related semantic or episodic


characteristics in parallel multidimensional networks. McClelland et al. (1995) proposed a
connectionist representation
of interleaved semantic knowledge (see figure 5) similar to ones
proposed by Hinton (1989) and Rumelhart (1990). There were problems associated with the past
connectionist models in that weighted changes were necessary to add new items to an already
blished network. This form of catastrophic interference made it difficult for such a model to

Figure 5. A propositional interleaved semantic network. Adapted from McClelland et al., 1995.



correctly identify the old information. McClelland et

al. (1995) accounted for such a discrepancy
by proposing the hippocampus and associated MTL structures would create their own temporary
network representing the rapid learning and consolidation so as to avoid a major disruption of the
structured system (i
n cortical networks) of knowledge gradually built up through experience.
This smaller network would then gradually train the larger network through gradual changes.
Thus, Eichenbaum proposes (2001) that consolidation can be conceived of as a life
olution of cortical networks, one that only reaches some form of stability when the
interleaving of similar constructs does not alter the network substantially. The interrelatedness
of the semantic components represented in the large cortical networks is
designed to generalize
to many other domains (i.e., sensory or motor) and must learn these connections slowly through
experience and time. Such a model may be applicable to the slow learning rates proposed by the
standard model, MTT and RBT that occur at
cortical connectivity. If indeed there are
mulitiple traces formed for each new episode or semantic association, each new trace would
share some or all of the information about the initial episode (Nadel & Moscovitch, 1997) and
connection weights

would be adjusted accordingly to the amount of recurrent overlapping
exposure (see Rumelhart, 1990).



It is assumed that the hippocampus proper and neocortex consistently interact over time
in ways that can shape the revivification or necessary update o
f past experiences. Therefore, as
episodic memories age, they would either be subject to decay and forgetting or reactivation, and
new encoding as a different trace to ease retrieval later on. With multiple traces in existence,
retrieval would be easier
as the number of routes for access proliferate. Nadel & Moscovitch
(1997) explain, “differences in the extent and temporal gradient of RA among the different types
of memory would be determined by the complexity or richness of the trace that is to be
vered” (p. 223). MTT accounts well for the multiple gradients that exist from damage to the
MTL and Nadel et al. (2000) provide a good model for stabilization process based on increased
representations in the MTL. Some studies however, argue that tempora
l gradients exist in both
human and animals when damage is limited to the hippocampal complex (see Knowlton &
Fanselow, 1998) or even to the CA1 region (see Fujii et al., 2000), while others argue that the
demand that is placed on the working memory compon
ent of MTL in terms of retrieval
mechanisms should be attributed to the prefrontal cortex (Shimamura, 2000). Eichenbaum
(2001) suggests an extension to the propositional model in that a sequence of highly specific
associations between stimuli and actions
in particular episodes may be represented by a sequence
of cellular or nodal representations that may be common across multiple experiences. Activation
of these nodes may initially occur in the hippocampus and with the high level of
interconnectivity betw
een the hippocampus and associated cortex may then further involve the
formation of cortical nodal representations after repeated experience. Thus, the hippocampus
may provide a means for the interleaving processeses of consolidation underlying MTT or the

propositional model.

A Cellular and Systems Reorganization of memory



If we view the consolidation process at a cellular level, we can observe the molecular
cascade that is involved in the fixation process on a relatively small time scale (seconds, minute
or hours). The cellular and molecular events associated with the fixation or cohesion process
have all been implicated as playing a role in inducing or maintaining LTP. The consolidation
process must then account for the gradients that last well beyon
d the time frame of LTP
mechanisms. Such processes involve the reactivation and subsequent reorganization of existing
memory traces apparent in reconsolidation theory. On a larger time scale (hours, months, years),
memory becomes susceptible to time
limited disruptive influences following retrieval.

Hebb (1904
1985) advanced the idea of perseveration (initially introduced by Muller and
Pilzecker in 1900) in the form of reverberation of local activity within the intrinsic neural
circuitry. He propos
ed that it was this reverberating activity that induced structural changes in
the network or pattern of synapses intrinsically activated that permits memory to be more
permanently stored (Hebb, 1949). Hebb’s ideas markedly shifted the approach to studying

memory consolidation. Instead of manipulating memory in intact animals, he inaugurated the
search for the biological events that underlie consolidation. The discovery of LTP, wherein brief
frequency stimulation of a neural pathway can induce long
asting changes in synaptic
efficacy (Bliss & Lomo, 1973), has become the quintessential cellular model for Hebbian
learning and has provided a tool to investigate the cellular and molecular cascade of events
necessary for long
term memory storage. Th
e study of the possible role of LTP, or LTP
changes, as well as synaptic changes associated with long
term depression (LTD), in memory
consolidation is supported and thoroughly documented by Grimwood, Martin, & Morris (2001)
as the synaptic plasticity

and memory (SPM) hypothesis. The hypothesis states, “activity
dependent synaptic plasticity is induced at appropriate synapses during memory formation and is


both necessary and sufficient for the information storage underlying the type of memory

by the brain area in which that plasticity is observed” (Grimwood et al., (2001), p.
521). Most studies on LTP have focused on the NMDA receptor (especially in the CA1 region
of the hippocampus) and the intracellular/molecular signaling that occur downs
tream of Ca

influx (figure 6). In brain slice experiments, it has been shown that stimulation of NMDA
glutamate receptors, as a result of postsynaptic depolarization of AMPA
type glutamate
receptors, allows Ca

to enter the postsynaptic neuron. F
ollowing Ca

influx, kinase [e.g.,
dependent protein kinase C (PKC) and calcium
dependent protein kinase II
(CaMKII)] activity may persist in order to maintain synaptic potentiation, for example by
continually phosphorylating postsynaptic
AMPA channels or modifying existing proteins. At
this point in the molecular cascade kinases have been shown to have the capacity for
autophosphorylation; they can be persistently active in the absence of Ca
for a period of time.
This phase of LTP has
been characterized as being protein
synthesis independent and has been
labeled the early phase of LTP (E
LTP), typically lasting about 1
3 hours (Frey, Huang &
Kandel, (1993). In contrast, a protein synthesis dependent, late phase of LTP (L
LTP), lasting
more than 7 hours has been observed after multiple high
frequency trains of stimulation (see
Silva, Kogan, Frankland, & Kida, 1998). This phase of plasticity has been shown to be
dependent upon cAMP
dependent protein kinase A (PKA) and the extracellular
activated protein kinase (ERK/MAPK). L
LTP causes long
enhancements in synaptic efficacy and demonstrates properties of LTM. It is currently thought
that both PKA and ERK/MAPK promote long
lasting changes by translocating
to the cell
nucleus and engaging activators of gene transcription (Alberini, Ghirardi, Huang, Nguyen &
Kandel, 1995; Schafe, Nader, Blair & Ledoux, 2001). One of the most studied transcription


factors, cAMP response
element binding protein (CREB) has bee
n shown, when phorphorylated,
to bind to DNA and modulate the expression of various genes necessary for long
term synaptic
plasticity (see Alberini, 1999; Kandel, 1997). The two phases of LTP have been suggested to fit

Figure 6. Intracellular/Molecula
r signaling cascade.



into short

and long
term plasticity models for memory consolidation, in that disruption along the
molecular cascade will induce respective short

and long
term memory deficits.

It has been widely demonstrated that either NM
DA antagonists (e.g., APV), CA1
gene knockouts, or the injection of Ca
chealators (which prevent the rise of intracellular Ca

into the postsynaptic neuron will prevent LTP induction in vitro, as well as prevent the formation
of hippocampus

and amygdala
dependent memories in vivo (Rotenberg, Abel, Hawkings,
Kandel & Muller, 2000; Wittenberg & Tsien, 2002). In hippocampal slice recordings,
pharmacological blockade or genetic knockouts of NMDA receptors has shown to block LTP
(Morris, Anderson
, Lynch & Baudry, 1986; Tsien, Huerta & Tonegawa, 1996). Additionally,
inhibiting both PKA and PKC have been shown to block L
LTP when infused during a critical
period following induction (60
120 min. post tetanus) (Thomas, Laroche, Errington, Bliss &
t, 1994). These data provide a link to behavioral learning and memory, in that the same
pharmacological and genetic manipulations have been found to cause deficits in behavioral
learning tasks.

NMDA receptor antagonists have been found to induce spatial

learning deficits (Morris,
1989), as well as a disruption in hippocampal theta rhythm (Leung & Desborough, 1988).
Similar to the inhibitory avoidance task originally used by Duncan (1949) to demonstrate the
disruptive effect of ECS on memory in rats, con
textual and auditory fear conditioning is a form
of Pavlovian conditioning that has received considerable attention in the last decade. It has


proven to be one of the most popular behavioral measures for the assessment of the function of
type mechanis
ms in making associations between certain environmental stimuli and
aversive events. APV has been shown to block the acquisition and expression of conditioned
fear when infused into the BLA prior to conditioning, but not immediately after training (Maren,

Aharonov, Stote & Fanselow, 1996). Through lesion studies, it has been established that
auditory and contextual fear conditioning are mediated by different neural subtrates (see Rudy,
Kuwagama & Pugh, 1999; Anagnostaras, Maren & Fanselow, 1999). The amy
gdala has shown
to be important in both conditioning a tone/light and/or context to an aversive stimulus, whereas
the hippocampus has been demonstrated to be particularly sensitive to context only.
Wallenstein, Vago & Walberer (2002) demonstrated that blo
cking both PKA and PKC (H7
dihyrdochloride) in the dorsal hippocampus 90 minutes, but not 1 minute nor 180 minutes post
conditioning, disrupted consolidation and hence expression of contextual memory 1 week post
conditioning. Additionally, it has been fou
nd that H7 infused directly into the BLA prior to
training attenuated long
term conditional fear, while at the same time sparing short
term fear
memory (Goosens, Holt & Maren, 2000). Several other recent studies have investigated the role
of both E
LTP an
d L
LTP in fear memory consolidation. These studies have shown that intra
amygdala infusion of protein
synthesis inhibitor impairs LTM of auditory and contextual fear
(Bailey, Kim, Sun, Thompson & Helmstetter, 1999). Similarly, infusion of either a prote
synthesis inhibitor or PKA in the amygdala impairs LTM for auditory fear conditioning, sparing
STM (Schafe & Ledoux, 2000). Although specific time courses have varied between in vitro
and in vivo methodology, in general, synaptic plasticity mechanisms
(LTP/LTD) and the
associated cascade of biochemical events in the MTL (dorsal hippocampus and BLA) have
demonstrated to be crucial in the underlying associative and contextual processing necessary in


memory consolidation. Future cellular and molecular mod
els of consolidation must take into
consideration these necessary molecular “stepping stones”, associated protein
synthesis and gene
expression when detecting or mimicking the activity
dependent synaptic plasticity associated
with memory consolidation.

ctivation and reconsolidation

One major assumption of consolidation theory has been that once a memory was
consolidated into LTM, it was no longer pervious to disruption or modulation. Evidence is
slowly surfacing that is challenging this fundamental line
ar formation of memory. After
amnestic treatments effective in inducing amnesia (e.g., ECS, hypoxia, protein synthesis
inhibition), it was understood that the subsequent disruptive effects were fixed as assessed by
retention; however, it has been found th
at retrieval of the memory could later be achieved by
exposing rats right before the retention test to cues associated with the original training (see Sara,
2000). Typical reminders that were effective in retrieving the memory for inhibitory avoidance
ining included a weak foot shock, brief exposure to the training context, or some combination
of the two (Lewis, Miller & Misanin, 1968; Sara, 2000). It was also found that stimulants such
as amphetamine or epinephrine would reverse the effect of the amne
stic treatment if
administered before the retention test. Warrington & Weiskrantz (1966, 1970) showed similar
effects in humans and suggested that at least some forms of amnesia may be due retrieval