NEUROBIOLOGY OF LEARNING AND MEMORY
I. Capacity for memory
II. Search for the engram
III. Distributed vs. nondistributed memory
IV. Memory Systems
Medial temporal lobe
Case of H.M. and other amnesics
Declarative vs. nondeclarative memory
What part of the medial temporal lobe is important for formation of declarative
memories? (Amygdala vs. hippocampus)
Skills and habits
V. Working memory
To put it simply, if we do not have the capacity to learn, we can not survive in this ever-
changing world. And in order for us to learn, experience has to modify the activity of the brain.
We actually have a great capacity to learn and remember. Consider that the average college
student recognizes tens of thousands of words, and even more if they are multilingual. Or even
the capacity of birds to remember hundreds of cache sites where food is hidden. The Canadian
psychologist, Lionel Standing, wanted to determine the capacity of human memory and found
that humans could recognize over thousands of slides they had seen only once. But be careful
of what you wish for. I also mentioned the case of Sherashevsky, the individual with an
amazing memory, who eventually became a stage performer. His ability to forget actually
complicated his life.
So how exactly does the brain change when we learn something new?
It seems that we like to conceive of memory as something stored away in the brain, in an
appropriate filing cabinet. But we need to realize that learning involves the changing of
circuits actively involved in perception, motor control, emotion, and cognition. So memory
really reflects changes in neuronal activity within these brain circuits.
But where do these changes take place?
There is a long history associated with the search for the engram, the physical representation of
a memory. One influential investigator in this area was Karl Lashley. Lashley hypothesized
that memories were stored in the cortex. He believed that when associations are formed
between 2 stimuli, connections between the cortical areas that encode those stimuli are
strengthened. He therefore hypothesized that if he made a cut in the cortex where the
connection resided, he would destroy that association, that memory. So he trained rats on a
series of mazes. After they learned to successfully complete the maze, he made a series of
knife cuts in the cortex, assuming that the knife cut that disrupted performance would illustrate
the location of the association. But knife cuts did not disrupt performance. He then made
ablations, removing large chunks of cortex. He found that the severity of memory loss was
related NOT to the location of the ablation, but rather with the amount of cortical tissue
removed. Based on his findings, he concluded that all parts of the cortex contribute to memory
equally (EQUIPOTENTIALITY) and that memories are stored in a distributed fashion (MASS
Lashley was incorrect in his assumption that all parts of the cortex contribute equally to
memory. However, the idea that memories may be stored in a DISTRIBUTED fashion was
highly influential. And for years there has been debate about whether memories are stored in a
highly localized fashion or in a distributed fashion. We now know that it depends. In some
cases, the locus of neuronal changes associated with learning is highly localized, and in other
cases, it is distributed throughout neuronal circuits.
How can memory be stored in a distributed fashion?
Consider what happens when you see a picture of a friend. You probably remember the way
they look, the sound of their voice, an experience that you shared together. One stimulus can
excite all types of associations. What does this mean in terms of how memories are encoded, or
rather, how learning changes the brain?
Memories are probably encoded by the relative activity of many neurons. How these memories
are stored in a distributed fashion can be illustrated in a simplified manner shown below:
Prior to learning, each neuron responds equally to each stimulus. Learning changes the
response of each neuron. Panel A demonstrates memory stored in a non-distributed fashion.
Each neuron or circuit encodes for an individual face. Panel B illustrates how memories are
stored in a distributed fashion. Each face is encoded by the relative activity of a combination of
neurons or circuits.
1 2 3
Neuron 1 2 3
1 2 3
Neuron 1 2 3
If a memory is stored in a non-distributed fashion, one can imagine that there would be a limit
to the number of memories one could form and that if a cell or circuit is lost, the memory is
lost. If memories are stored in a distributed fashion, it may explain how one memory can lead
to other associations. In addition, consider what may happen during a neurodegenerative
disease, like Alzheimer’s. If portions of the neuronal network, it may be more difficult to
access the memory.
MULTIPLE MEMORY SYSTEMS
Given that there is a lot of heterogeneity in the forms of learning and the types of things
learned, it comes as no surprise that there are multiple memory systems.
There are two important caveats to keep in mind:
Memory is not stored in “filing cabinet”, but rather, learning involves plasticity in circuits
actively involved in sensation, motor function, emotion, and cognition
Multiple memory systems in the brain can interact with one another
Much of what we know of the neurobiology of memory in humans comes from clinical cases of
amnesia, the most famous being H.M. H.M. sustained a head injury at the age of 11 that led to
the development of epileptic seizures. The severity of seizures increased, so by the time he hit
young adulthood, he lost control of the seizures and subsequently lost his job. In 1953, the
neurosurgeon William Scoville removed the foci of the seizures, the medial temporal lobes.
Prior surgeries had shown that this had little effect on an individual, but earlier surgeries
removed less tissue. The surgery was a success—he could now control his seizures with
medication. However, he also suffered from memory loss. He had some retrograde amnesia,
which is not all that uncommon following a trauma. More strikingly, he suffered from severe
So, H.M. had bilateral removal of the medial temporal lobes.
Recall that RETROGRADE AMNESIA refers to the loss of memories that were formed PRIOR
to the insult. This is really not that uncommon following head trauma. ANTEROGRADE
AMNESIA refers to the inability to form new memories.
However, H.M. was able to learn some new things, including perceptual learning, classical
eyeblink conditioning, and the learning of skills and habits. He simply had no conscious
recollection of learning these new things. Based on his case and others, memory systems were
divided into 2 categories.
Memory systems are usually divided into 2 broad categories:
1-DECLARATIVE of EXPLICIT (facts and events that can be declared)
For example, “I had Cheerios for breakfast” or the capital of Thailand is
Bangkok or the coyote will never catch the roadrunner. Although this was
originally believed to be verbal learning, it is not the same and explicit memory
can be studied in nonverbal animals and humans.
2-NONDECLARATIVE or IMPLICIT (skills and behaviors)
For example, riding your bike or playing a piano, classical conditioning, or
learning a skill
H.M. was no longer able to form declarative memories, but could form non-declarative
memories. In other words, structures in the medial temporal lobe were involved in the
consolidation of declarative memories.
But H.M. had bilateral removal of large portions of the medial temporal lobe, which contains
the hippocampus, amygdala, and cortical tissues. So, this led to the question of what role
structures contained in the medial temporal lobe play in the formation of memory.
Other amnesia cases, like R.B., suggested that the hippocampus is the critical structure. R.B.
suffered from ischemia, which more selectively damaged the CA1 pyramidal neurons of the
hippocampus. R.B. also suffered from anterograde amnesia like H.M., although it was not as
Hippocampus, Amygdala, and Limbic cortex: which is important for forming new
The best way to determine which structures are critical for memory formation is with animal
model systems. Many of these studies use the nonmatching to sample task described in class to
measure working memory, which requires the same circuits as declarative memory
Delay period (no
In this task, a sample stimulus is presented. Then the stimulus is removed for a delay period.
Finally, two choice stimuli are presented, the sample stimulus and another stimulus. In the
delayed matching to sample task, the subject gets a reward for choosing the sample. In the
delayed nonmatching to sample task, the subject gets a reward for choosing the choice stimulus
that does NOT match the sample. Thus, during each trial, the subject must keep the memory
for the sample stimulus in working memory. After the trial is over, the memory is no longer
In the 1980s and 1990s, Zola-Morgan and colleagues conducted a series of lesion studies in
monkeys to determine the relative roles of medial temporal lobe structure in memory.
For example, they compared learning performance in monkeys with lesions to the hippocampus
and surrounding cortical tissue (H+) to monkeys with lesions to the hippocampus, amygdala,
and surrounding cortical tissues (H+A+). They found the most severe explicit memory deficits
in those monkeys with lesions to both the hippocampus, amygdala, and respective surrounding
Since the H+A+ lesion produced more severe deficits compared to the H+, there are two
possible reasons. 1- the amygdala contributes to explicit memory formation or 2- the cortical
areas surrounding the amygdala contribute to explicit memory formation.
So, Zola-Morgan and colleagues created more specific lesions: amygdala only (A),
hippocampus plus surrounding cortex (H+), hippocampus and surround cortex plus amygdala
(H+A), and hippocampus and surrounding cortex and amygdala and surrounding cortex
(H+A+). In addition, an unoperated control was included. The results indicated that there was
no memory deficit in the A group, similar moderate deficits in the H+ and H+A groups, and the
most severe deficits in the H+A+ group. These results suggest that the amygdala does not
contribute to explicit memory, but the surrounding cortical tissue may.
This suggests that there is a dissociation between the hippocampus/ limbic cortical tissue and
the amygdala. The hippocampus and surrounding cortical tissue is important for explicit
memory formation. It turns out that the amygdala is also important for learning, but a different
type of learning. The amygdala seems to be critical for emotional conditioning (see below).
This has been examined with human subjects. Bechara et al (1995) conducted a study with 3
individuals who sustained brain damage and four controls. One individual had bilateral
hippocampal damage, another had bilateral amygdala damage, and the third had bilateral
damage to the medial temporal lobe, including both the hippocampus and amygdala.
Subjects were tested on a Pavlovian conditioning task, an emotional conditioning task. Lights
of various colors were presented (green, blue, red, white). Following the blue light, a load boat
horn was presented, which produces an unconditioned fear response. Thus, the boat horn is an
unconditioned stimulus (US) that elicits an unconditioned fear response (UR). They measured
a change in skin conductance, which serves as a measure of sympathetic nervous system
activation. The stimuli were repeatedly paired in a manner that would create a conditioned
response (CR) to the blue light (CS). After training was completed, subjects were asked the
following questions to test their declarative memory:
1. How many different colors did you see?
2. Tell me the names of the colors?
3. How many different colors were followed by the horn?
4. Tell me the name(s) of the colors that were followed by the horn?
+ BANG! FEAR!
Blue Light (CS) 100 db boat horn (US) UR (skin conductance response)
FEAR (CR) (skin conductance response (SCR))
They reported the following results:
Normal group conditioned SCR
good declarative memory
Amygdala lesion no conditioned SCR
normal declarative memory
Hippocampus lesion normal conditioned SCR
no declarative memory
Medial temporal lesion no conditioned SCR
no declarative memory
In other words, damage to the amygdala produces deficits in classical emotional conditioning,
whereas damage to the hippocampus and adjacent cortical areas produce deficits in declarative
Other Limbic Structures
BUT, anterograde amnesia can also be caused by damage to other limbic structures. For
example, N.A. was sitting in his barracks when he was accidentally pierced with a miniature
fencing foil. The foil entered his nostril and took a course into the brain, damaging the
mammillary bodies and thalamus. He suffered from some retrograde amnesia, but also
exhibited anterograde amnesia that was similar in nature to that of H.M., although less severe.
What does this mean?
It means that damage to a number of limbic structures may alter the formation of declarative
memories (except the amygdala). When we consider brain structures, they are not islands, but
rather, they form circuits. The hippocampus, adjacent cortical areas (perirhinal cortex,
entorhinal cortex, parahippocampal cortex), and other limbic structures (like the diencephalons)
form a circuit, so there is a high degree of interconnectedness and damage to various parts may
disrupt declarative memory.
But we still hear a lot about the hippocampus and memory—what exactly does the
We did not have time to go into details, but models of the function of the hippocampus and
related limbic structures have not yet provided a parsimonious “function” of the hippocampus.
Suffice it to say that it is involved in:
1- consolidation and possibly the retrieval of declarative memories
2- spatial memory
3- working memory
4- relational memory
5- contextual memory
So limbic structures are necessary for the consolidation of declarative memories, but declarative
memories are not “stored” in the limbic structures for long periods of time.
Long-term declarative memories are not “stored” in the hippocampus.
Moreover, declarative memories can be divided into 2 categories:
Episodic memory is autobiographical memory that pertains to a person’s particular
history. For example, you may recall of a specific episode or event when you saw a friend, or
recall your trip to Paris.
Semantic memory is generalized memory, such as knowing the meaning of a word
without knowing where or when you learned it, or knowing that Paris is the capital of France.
EPISODIC MEMORY depends on the integrity of the medial temporal lobes and hippocampus.
That is, the medial temporal lobe is important for the consolidation of episodic memories. In
addition, the prefrontal cortex also plays a key role in long-term episodic memory. Although
patients with selective damage to prefrontal regions do not develop a profound amnesia, they
have great difficulty remembering when and where recent event occurred—the defining
features of episodic memory.
We didn’t have time to talk about this so much, but K.C. is a patient with damage to the frontal
cortex who exhibited fine nondeclarative memory and semantic memory, but poor episodic
memory. He could not acquire new knowledge regarding his own experiences.
SEMANTIC MEMORY depends on the medial temporal lobes. That is, the medial temporal
lobe is important for the consolidation of semantic memories. In addition, long-term semantic
memories are probably “stored” in cortical areas, including the temporal cortex.
NONDECLARATIVE TYPES OF MEMORIES
1- Nonassociative learning, or Reflex Modification
This refers to modification of reflexes, including habituation (a reduction in responsivity)
and sensitization (an increase in responsivity). For example, if you live by a railroad, you
may exhibit a startle response the first time a train passes, but you do not after many
presentations. This is habituation, a reduction in responsivity due to repeated presentations
of the stimulus. Or let’s say that your house regularly makes creaking, settling sounds.
You normally do not respond. But if you are home alone and hear a big crash outside, now
if the house creaks, you jump. This is sensitization. You respond to stimuli that normally
do not produce a response. The plasticity occurs in the reflex pathways.
This memory allows us to recognize stimuli that have been perceived before. This allows
us to identify and categorize objects and situations.
Simple perceptual learning appears to take place in appropriate regions of sensory
association cortex. So, recognition of auditory stimuli occurs in auditory association cortex.
Recognition of visual stimuli occurs in visual association cortex.
If you recall, different components of visual information are processed in different cortical
areas. So, object recognition occurs in the inferior temporal cortex, whereas information of
the location of objects and recognition of location depends on the posterior parietal cortex.
You may recall that damage to the inferior temporal cortex leads to difficulties in
recognizing visual objects. For example, in the inferior temporal cortex, there are cells that
are activated by faces. If damage occurs in this area, it can result in a specific type of
AGNOSIA (without knowledge), PROSOPAGNOSIA (the inability to recognize faces).
Remember this when we discussed vision?
Emotional conditioning requires the integrity of the amygdala. If a neutral stimulus is
paired with an emotional-eliciting stimulus, individuals with amygdala damage fail to
develop a conditioned emotional response, as discussed above.
Probably the best understood neural substrate for learning and memory is that involved in
classical eyeblink conditioning. Richard Thompson and his colleagues have discovered the
locus of synaptic changes that occurs during classical eyeblink conditioning.
The neuronal plasticity associated with eyeblink conditioning occurs in the cerebellum. The US
and CS send inputs into the CERBELLAR CORTEX (this is the outer layer of the cerebellum).
The cerebellar cortex ahs connections with a deep nucleus called the INTERPOSITUS
NUCLEUS, which sends output to the red nucleus which sends output to the motor neurons
involved with the eyeblink. Richard Thompson and colleagues have demonstrated that some of
the the plasticity associated with learning appears to occur at the interpositus nucleus.
Consistent with his experimental findings, imaging studies show that the cerebellum is active
during classical eyeblink conditioning, and individuals who sustain cerebellar damage show
eyeblink conditioning deficits.
**another thing to keep in mind is that these various memory systems can interact and when
you engage in a memory task, it may require the activation of several memory systems. For
example, during delay conditioning, where the CS and US overlap in time, only the cerebellum
is necessary. During trace conditioning, when there is a delay between the offset of the CS and
onset of the US, both the cerebellum and the hippocampus are necessary. It is called trace,
because performance depends on a memory trace.
So when you think of the memory tasks in which you engage day to day, many are
complex and may require the use of multiple systems.
Depends on cerebellum
Depends on cerebellum and hippocampus
4-Skill learning, or procedural learning
This is memory for skill or habits, “knowing how” rather than “knowing what.”
Procedural memories are acquired gradually through repetitive practice.
Two areas of the brain are important for this type of learning:
The cerebellum is particularly important for motor learning. For example, one task that
is disrupted by cerebellar damage is mirror learning--learning how to trace a star using only a
Similarly, the basal ganglia plays a role in the learning of new habits. Consider the
following task. Subjects must predict the weather based on presentation of cues. The trials
occurred very quickly and subjects reported that they did not know what the rules were, but did
show improved performance. Interestingly, Larry Squire’s group (Knowlton, et al, 1996)
demonstrated a dissociation between individuals with Parkinson’s disease (basal ganglia
damage) and those with hippocampal damage.
Data in Panel A show learning on the habit task. Control subjects (CON) show improvement,
reaching 70% correct. Amnesics (with hippocampally related damage) also show learning. But
Parkinson’s patients (PD), particularly those most severely affects (PD*) did not show learning.
In contrast, when tested on a declarative memory task, the PD patients did not differ from
controls, but, as expected, the medial temporal patients exhibited impairments.
Similarly, if you damage the basal ganglia, rats have difficulty learning a task that require they
enter arms with lights for a food rewards. Individuals with damage to the striatum have
difficulty learning rules.
In contrast, panel B shows performance on a declarative memory task. You can see that
Parkinson’s patients show no impairments, whereas the hippocampal damaged amnesics do.
One can also separate stages of memory processing, including short vs. long-term memory and
working memory. Often, the type of memory is operationally defined by a particular task.
One form of memory that has received a lot of attention is working memory. WORKING
MEMORY refers to the memory held “online” while you are working on a problem. For
example, if I ask you to add 52 and 137, you would hold that information in your memory while
you performed the addition. The information is important for the current task. It is akin to
short-term memory, however it is thought that working memory will not be transferred to long-
term memory. The delayed nonmatching to sample task described above is a working memory
task. The subject has to remember the sample stimulus, but after the trial is over, that
information is no longer useful.
Several areas of the brain are important for working memory
We did not have much time to go into this, but areas include:
1-hippocampus and related cortical areas
Damage to the hippocampus and related cortical areas disrupts Delayed-Nonmatching-
2-prefrontal cortex (we will talk more about this when we get into cognition)
Damage to the prefrontal cortex also disrupts performance on the Delayed-Nonmatching
to Sample. In addition, if you record from cells in the prefrontal cortex, you would find
that certain cells were activated during the delay, dependent on the stimulus. So, there
is evidence that some of those cells are involved in the memory.
Keep in mind that the frontal cortex is also connected to these medial temporal lobe
3-lateral intraparietal cortex is important for visual working memory
So, we see that there are multiple memory systems. What this means is that learning changes
the circuits actively involved in sensation, motor function, emotions and cognition. Memory is
not stored in little filing cabinets. Rather, learning changes the activity of sensory, motor,
emotional and cognitive systems in the brain. In other words, learning changes the functioning
of circuits in your brain so that it affects the way you perceive, move around, and think about
References, suggested readings:
Bear, M.F. & Malenka, R.C. 1994. Synaptic plasticity: LTP and LTP. Current Opinion in
Neurobiology, 4, 389-399.
Bechara, A., Tranel, D., Hanna, D., Adolphs, R. Rockland, C., & Damasio, A.R. 1995. Double
dissociation of conditioning and declarative knowledge related to the amygdala and
hipocampus in humans Science, 269, 1115-1118.
Byrne, J.H. 1987. Cellular analysis of associative learning. Physiology Review, 67, 329-439.
Knowlten, B.J., Mangels, J.A. & Squire, L.R. 196. A neostriatal habit learning system in
humans. Science, 273, 1399-1401.
Malenka, R.C. & Nicoll, R.A. 1999. Long-term potentiation: A decade of progress? Science,
Martin et al., 2000. Synaptic plasticity and memory: An evaluation of the hypothesis. Annual
Reviews in Neuroscience, 23, 649-711.
McGaugh, J.L. 2000. Memory—a century of consolidation. Science, 287, 248-251.
O’Keefe, J. 1976. Place units in the hippocampus of the freely moving rat. Experimental
Neurology, 51, 78-109.
Suzuki, W. & Clayton, N.S. 2001. The hippocampus and memory: A comparative and
ethological perspective. Current Opinion in Neurobiology, 10, 768-773.
Zola-Morgan, S., & Squire, L.R. 1985. Medial temporal lesions in monkeys impair memory on
a variety of tasks sensitive to human amnesia. Behavioral Neuroscience, 99, 22-34.