¿ìè¶ÌÊÓƵ

So casually do we use memory that we tend to overlook its power. But
in case you’re in any doubt a child learns new words at an average rate
of more than 10 each day and may eventually build a vocabulary of 100 000
words, the best storytellers among the ancient Celts knew some 350 epics
by heart, and the neurologist Oliver Sacks once treated a patient who had
memorised all nine volumes of the 1954 edition of Grove’s Dictionary of
Music and Musicians, not to mention the music of 2000 operas.

How does a ball of soft tissue weighing no more than 1.5 kilograms manage
to store such vast amounts of information? How do memories form inside the
brain’s jungle of neurons, fibres and connections (synapses)?

Earlier this year the dogged quest for clues that keeps hundreds of
neuroscience laboratories busy all over the world was interrupted by a flurry
of excitement. The news was that researchers at the Massachusetts Institute
of Technology had used genetic techniques to produce a mouse with a defective
memory. Following a standard recipe, the MIT team had taken a mouse embryo,
‘knocked out’ one of its thousands of genes-a gene coding for an enzyme
known to be active in the brain-and then returned it to the womb. The result
was a mouse that appeared normal in every respect bar one: it was unusually
bad at learning its way around a pool of murky water a so-called ‘water
³¾²¹³ú±ð’.

At the same time it emerged that another American team, at Columbia
University, New York, had produced genetically engineered mice lacking a
similar type of enzyme. The behavioural tests were still in progress, but
the signs were that some of these animals were also dunces in the water
maze.

Here, it seemed, was powerful evidence linking a specific type of learning
to specific genes. But that was not the only reason for the excitement.
For when the researchers probed the electrical behaviour of nerve pathways
in the brains of their knockout mice, they discovered something else was
missing, or at least much reduced: long-term potentiation (LTP), a curious
strengthening of synapses which takes place when certain neural pathways
are artificially stimulated.

LTP has long intrigued neuroscientists as a possible mechanism for the
cellular basis of learning. The transgenic mice, combining as they do a
slow wit with defective LTP, have been hailed as the best evidence so far
(although there are complications; of which more later). Moreover as the
first transgenic mice to be used in research into mental functions such
as learning, the animals mark the beginning of a new era in neurobiology.
They are a sign that researchers are beginning to apply the genetic tools
that have revolutionised immunology and developmental biology to studies
of mammalian behaviour.

Yet advances in molecular genetics are not the only reason why neurobiologists
believe they are close to unravelling the neural basis of learning. Eric
Kandel, who heads the team studying the knockout mice at Columbia University,
is just as excited about recent discoveries suggesting that the cellular
mechanisms involved in forming new memories may be closely related to those
that underpin the brain’s growth in the early years of life. ‘One can think
of learning as a developmental process,’ he says. With many other neurobiologists,
Kandel now believes that uncovering the links between learning, synapses
and development will ultimately help to demystify complex mental functions
such as memory

To trace the roots of this optimism, one must take a broad look at recent
developments in research into learning and memory. While clues are coming
thick and fast from cell and molecular biology, the less reductionist disciplines
of neuropsychology and neuroanatomy are still producing fundamental insights.

Internal filing system

By studying people suffering from amnesia caused by brain damage, neurologists
can start to piece together the way the brain handles memories. The approach
is not without problems: brain damage incurred from head injuries, strokes
and tumours is often widespread, and damage to any one area of the brain
may affect different people in different ways. Even so, it has yielded some
striking results. One of the most important is that the brain organises
its knowledge of the world-semantic memory-in categories, almost as if it
operates an internal filing system.

The evidence for this notion comes from the highly selective nature
of many memory impairments. Patients with damage to the brain’s temporal
lobes (the parts of the cerebral cortex that lie behind the temples) may
find themselves able to recognise tools and utensils, say, but not animals.
While the physical basis of these categories is still a mystery one theory
suggests a link with the highly segregated, or ‘modular’, neural pathways
which the cortex uses to handle information from the body’s senses. ‘Sensory
channels in the cortex are the foundation of the neural differentiation
of memory’ says Rosaleen McCarthy a neuropsychologist at the University
of Cambridge.

With Elizabeth Warrington of the National Hospital for Nervous Diseases,
London, McCarthy has chronicled a particularly intriguing case of selective
memory loss in a patient with a damaged left temporal lobe. When they showed
him a picture of a rhinoceros, he produced a perfectly serviceable reply:
‘Enormous, weighs over one ton, lives in Africa.’ But when they asked him
to define the spoken word ‘rhinoceros’, he simply said, ‘Animal, can’t give
you any functions.’ This mental deficit extended to many other living things
but not to inanimate objects; he could define those equally well regardless
of how he was asked.

The implication of such findings is that the brain places knowledge
in different ‘stores’ according to how it was obtained, verbally or visually.
Tests on other patients support these ideas. McCarthy is now working with
a patient who does not seem to recognise animals visually but can offer
many details about them in response to the spoken word. ‘The research is
at too early a stage to say exactly where these stores are,’ says McCarthy
but the brain seems to hold visual information received in the form of words
and pictures in different parts of the temporal lobe

Another type of lesion has been vital to research into memory: damage
to a structure deep within the brain called the hippocampus. The most celebrated
patient of this type is HM, an American assembly-line worker who had surgery
in 1953 to relieve him of epilepsy. During the operation surgeons removed
the hippocampal region on both sides of his brain, together with nearby
tissue. Although the operation had the intended effect, and apparently boosted
HM’s IQ into the bargain, it severely damaged his memory. He had good recall
of events before his operation, but could not retain new information, such
as the faces of people he met. He was able to learn new skills, such as
drawing while watching his hand in a mirror (a standard neurological test),
but forgot the experience of learning them.

The snag was that HM’s brain lesion spread into the cortex, so it was
impossible to be sure that his amnesia resulted solely from damage to the
hippocampus. More recently, however, attention has turned to people with
lesions confined to the hippocampus. During the 1980s Larry Squire and his
colleagues at the Veterans’ Affairs Medical Center, San Diego, studied a
patient with damage to the CA1 region of the hippocampus. His amnesia was
not as severe as HM’s, but the trend was similar. He had difficulty remembering
events that took place after his illness but had good recall of events before
the illness.

Such findings have dominated the genesis of theories about the hippocampus.
Researchers such as Squire see two ideas as central. The first is that we
need the hippocampus for creating ‘declarative’ memory-facts, events, names
and so on-but not for ‘procedural’ memory-skills such as playing tennis
or riding a bicycle. While procedural memories are usually formed through
repetition, declarative memories, such as learning someone’s name, often
appear to require only a ‘one shot’ learning experience. Second, although
the hippocampus helps us to form new memories, it cannot be the final storage
site, for many old memories survive its destruction.

Neuroscientists have sought to confirm these ideas with animal experiments.
During the 1980s researchers such as Mortimer Mishkin at the US National
Institute of Mental Health, Bethesda, and Squire, Stuart Zola-Morgan and
their colleagues in San Diego, investigated the effects of brain lesions
on monkeys. Normal monkeys, like humans, tend to forget things with time.
So if over a period of weeks, they learn a difficult task-choosing a particular
object from each of 100 pairs of objects-and then take a memory test, the
results are fairly predictable: their memory for recently learned objects
tends to be better than their memory for objects learned earlier. When Zola-Morgan
and Squire trained monkeys and then removed their hippocampuses, however
the opposite trend emerged-confirming, they say, that the hippocampus is
required for memory storage for only a limited period, and that over time
a more permanent memory trace must form in other parts of the brain, presumably
the cortex.

Creating scenes

On the face of it, then, such experiments suggest that loss of the hippocampus
causes the same kind of amnesia in monkeys and humans alike. But other researchers
see complications. One concern, says David Gaffan of the University of Oxford,
is that the tests used to detect memory loss were based on the monkeys’
abilities to point to particular objects. ‘Humans suffering from amnesia
don’t forget their general knowledge of objects . . . they forget about
events, meeting people and the contexts in which they came across objects,’
he says.

In his own research on how hippocampal lesions affect monkeys, Gaffan
has begun to use a memory test based on organised scenes, rather than on
objects. He has just completed a study he says, in which he used scenes
from the movie ‘Raiders of the lost Ark’. Typically monkeys were shown one
set of scenes several months before surgery and a second set just beforehand.
The results, says Gaffan, run counter to the American findings: the hippocampal
lesions did not selectively destroy the monkeys’ memory of the most recently
learned scenes.

All this underscores the uncertainty surrounding the precise function
of the hippocampus during learning. We know that the hippocampus does not
function as a store for older memories; judging from studies of people with
amnesia, these appear to reside in the cortex. And we know from the neural
architecture of the temporal lobe that the hippocampus and cortex engage
in an intense dialogue. But the rest is speculation.

Among the thicket of theories, though, some common themes are discernible.
One intriguing theory championed by Antonio Damasio of the University of
Iowa College of Medicine, forges a link between memory and sensory perception.

On the basis of studies of amnesic patients, Damasio concludes that
when we recall a memory we probably activate the same neural systems in
the cortex that we use for perception. In other words, the various elements
of a past experience-its visual record, its sound, its smell-may reside
in sites within the cortex that specialise in the relevant sense. Recall
(the theory runs) involves activating these separate sites in unison, creating
an integrated memory from shards of experience. How this might happen is
still a mystery but it could involve the hippocampus. The structure might,
for instance, hold pointers to these areas, or bind the components of the
episode together allowing them to be recalled as a whole.

How cortical sites store memories is also a matter for conjecture. Patterns
of electrical activity within groups of neurons are the brain’s likely currency,
its coded way of representing the world. But how can such patterns be stored
and recalled?

Enter the synapse. For most theories about memory posit that groups
of neurons learn patterns of electrical activity by adusting the strengths
of their synapses. Confidence in this idea has been bolstered by research
on neural networks, computer devices designed to behave like collections
of neurons. Some of these so-called ‘connectionist’ models are made out
of electronic components, while others are simulated on conventional computers.

Typically, a simple network comprises several interconnected ‘neuron-like’
units, each of which can be either active or inactive depending on how many
signals it is receiving from its neighbours. These signals pass through
connections whose strength can be adjusted according to simple rules. The
end result is that the network can learn, and thence recall to order, any
pattern of activity.

Everyone agrees that this kind of learning addresses only a small part
of the biological puzzle of memory. Yet it is a start. Patricia Churchland,
a philosopher who specialises in neuroscience at the University of California
in San Diego, sees neural modelling as a powerful way to generate testable
theories. Recent studies of the nervous systems of invertebrates bear this
out. For example, by simulating the nervous system of the mollusc Aplysia,
Jack Byrne of the University of Texas has correctly predicted how certain
neurons behave as the animal learns.

The goal now says Churchland, is to make neural models, whether they
rely on hardware or software, more biologically realistic. The quest for
realism in hardware is in full swing at the University of Oxford, where
Misha Mahowald and Rodney Douglas are designing silicon chips that miinic
both the architecture and electrical behaviour of real neurons. Already
they have built a chip with five neuron-like units, each of whose activity
is controlled by electrical currents that mimic the tiny ion currents that
flow through channels in the membranes of neurons, inhibiting or stimulating
activity. At present the chip can store patterns of activity but cannot
learn like a neural network because the structures equivalent to dendrites
and synapses are not yet in place. Other proponents of biological realism
include Gerald Edelman and his colleagues at the Neurosciences Institute
in New York, who have designed a simulated nervous system called Darwin
IV which learns by selectively strengthening subsets of its synapses (see
photograph).

At the simplest level, neural modelling suggests how real clusters of
neurons might learn. To act as a ‘memory’, a cluster must be able to ‘recall’
a specific pattern of electrical activity on demand. Now suppose that each
neuron in the cluster is connected to all the others by synapses that can
be strengthened according to a simple rule: become stronger if the neurons
you link are both active at the same time. Such a synapse is called a Hebbian
synapse, after the Canadian psychologist Donald Hebb who proposed the rule
in 1949. Once all the synapses have put that rule into effect, the memory
has been stored. Neurons that were once active together are now linked by
stronger synapses; during recall, they will tend to rouse one another and
help to recreate the original pattern.

In theory, a fully fledged cellular network of this type-an auto-association
memory-could hold many memories simultaneously with each synapse participating
in several memories and each memory being ‘encoded’ by several synapses.
Its powers of recall would be spectacular. If prompted with only a small
fragment of a memory its synapses would ensure that it regenerates that
memory in its entirety-a phenomenon known as completion. We experience (so
the argument goes) this effect whenever the sound of a voice conjures up
a name, a face or an episode from the past.

Armed with ideas like these, researchers have looked anew at the anatomy
of the hippocampus and other structures. Edmund Rolls of the University
of Oxford, for example, has studied the hippocampus, looking for evidence
of memory networks. The neural architecture of the CA3 region is particularly
suggestive. In rats each CA3 neuron receives inputs from 12 000 other CA3
neurons-exactly the kind of pattern one would expect, says Rolls, if the
area worked as an autoassociation memory

But a complex web of connections is far from proof that a cluster of
neurons functions as a memory network. A more fundamental question is whether
its neurons and synapses behave in a Hebbian manner. In the case of the
hippocampus, the beginnings of an answer emerged 20 years ago from animal
research by Terje Lomo at the University of Oslo, Tim Bliss, now at the
National Institute for Medical Research, London, and Tony Gardner-Medwin,
now at University College London.

The researchers were using microelectrodes to study electrical activity
in the brains of anaesthetised rabbits. The site of interest: a part of
the hippocampus called the dentate gyrus. Impulses enter the dentate gyrus
through an array of input fibres and then spread through synapses to a network
of neurons nearby. The researchers found that sending a sustained volley
of artificial impulses along these input fibres results in a dramatic and
long-lasting increase in the strengths of the synapses. In short, they had
discovered LTP.

Once LTP has been triggered, it can last for weeks or even months-enduring
enough to make it a likely candidate for the physical changes that underpin
the formation of recent memories. There are two further reasons why LTP
has become so central to theories about learning. First, it acts in an impeccably
Hebbian manner, only taking place at synapses connecting neurons that are
both active at the same time. Secondly it behaves like a cellular mimic
of classical conditioning. If a weak impulse arrives at a neuron that is
already responding to a strong impulse, the synapse transmitting the weak
impulse may grow stronger.

For more than a decade after its discovery, LTP’s Hebbian behaviour
baffled neuroscientists. It seemed that certain synapses had a mechanism
for detecting when both partners-the neuron sending the signal (the presynaptic
neuron) and the neuron receiving the signal (the postsynaptic neuron)-were
simultaneously active. But what was the mechanism? The breakthrough came
in the early 1980s with the discovery of a receptor known as the NMDA receptor.

When neurons in the hippocampus fire, they discharge a neurotransmitter
known as glutamate across their synapses. This stimulates receptor molecules
embedded in the membranes of postsynaptic neurons, which respond by opening
channels in the neuron’s membrane, admitting ions and so triggering an electrical
impulse. In the hippocampus the receptors are of two main types, called
AMPA and NMDA after chemicals that trigger them with particular vigour.

AMPA receptors are responsible for routine transmission across synapses,
but NMDA receptors have more exotic properties: they are the molecular machines
that put Hebb’s rule into effect. The reason? Unlike most receptors which
require only one trigger the NMDA receptor requires two before it will swing
open to allow ionic traffic. The presynaptic neuron must deliver glutamate
and the postsynaptic neuron must coincidentally dislodge a magnesium ions
blocking the NMDA channel. And, just as Hebb’s rule demands, both events
require neurons to be electrically active. When the NMDA receptors finally
swing open, they spark off a cascade of biochemical changes that bring about
LTP

If LTP is involved in learning, then the NMDA receptor lies at the molecular
heart of memory. Or, as Kandel puts it, ‘the core element of classical conditioning
is built into the architecture of a single receptor molecule.’

Breaking the rules

Over the past decade research into LTP has intensified as scientists
have fixed their sights on unravelling the chain of biochemical events that
causes it. Many of these events have now been identified, though the role
of some of the biochemical intermediaries involved are still controversial
. One interesting twist is the discovery of the need for a so-called retrograde
messenger-a substance that carries a message back across the synapse, from
the postsynaptic neuron to the presynaptic neuron. In making this short
journey it breaks one of the oldest rules in neuroscience, namely that information
flows across synapses in one direction only

The key remaining uncertainty is the relationship between LTP and behaviour.
‘It might have nothing to do with memory but it certainly obeys all the
rules,’ says Richard Morris of the University of Edinburgh, who has played
a leading role in designing experiments to test the behavioural implications
of LTP. His approach is to give rats a compound, AP5, that prevents LTP
by disabling the NMDA receptor and see if they can still learn simple tasks.

During the tests, rats have to learn the whereabouts of a submerged
platform hidden in a tank of water -the water maze. Milk is added to the
water to make it cloudy so that the rats cannot see the platform. Rats are
good swimmers but they show little inclination for bathing in milk and quickly
learn to find the platform.

The team soon established that AP5 was enough to turn a sharp-witted
rat into a slow learner. Could this disruption have been due to some general
impairment of the brain? This is unlikely, argues Morris, because the rats
were able to Perform another task-choosing between two raised platforms
in the tank on the basis of visual cues-with ease. This suggests that their
motivation, their senses and so on were all intact.

A tougher question is, were the rats’ problems directly due to a lack
of LTP in the hippocampus? Research in Morris’s laboratory suggests that
rats which receive larger doses of AP5 are poorer learners than rats which
receive smaller doses. They also suffer a greater disruption of LTP. This
does not clinch the issue, but everyone agrees that it is an impressive
correlation.

Knockout mice

It was partly to sidestep the potential problem of unwanted side effects
of LTP-blocking drugs that Kandel and his collaborators, and Susumu Tonegawa
with his group at MIT and Charles Stevens of the Salk Institute in San Diego,
decided to engineer transgenic mice with an in-built block on LTP. Each
team targeted a different gene. The MIT researchers chose alpha calcium/
calmodulin-dependent kinase II, the Columbia group a kinase called Fyn.
Both enzymes are believed to be required for LTP

Kandel’s group has yet to report its tests on behaviour but the mice
bred at MIT turned out to be dunces at the water maze test and underachievers
at LTP-a result that Morris calls ‘exciting’. But, as ever, there are hidden
complications. One is that both kinases are normally active over a broad
area of the brain, so there is no guarantee that the amnesia results solely
from the loss of LTP in the hippocampus. Another problem is developmental
abnormalities. According to Kandel, his team’s transgenic mice carry a surfeit
of neurons in the hippocampus.

Anatomical changes of this kind are a nuisance for researchers trying
to get to the bottom of LTP, but they hint at an interesting question. Are
subtle changes in the strengths of synapses sufficient to explain long-term
memory, or does learning lead to more drastic changes in the brain’s cellular
architecture? So far there is no clear verdict, but indirect evidence suggests
that to store long-term memories, the brain may have to make new synaptic
connections.

In Aplysia, for example, researchers have found that artificial electrical
impulses which produce a form of long-term memory also cause synapse-like
structures to grow on the nerve fibres. And dramatic evidence that learning
can cause ‘rewiring’ in mammals has come from an experiment by Michael Merzenich
of the University of California at San Francisco. Merzenich encouraged a
monkey to touch a rotating disc repeatedly with only the three middle fingers
of its hand. As a result, the part of the animal’s cortex devoted to processing
messages from these fingers expanded at the expense of that devoted to the
other fingers.

Such discoveries help to explain why neuroscientists are looking for
links between learning and the mechanisms that control the brain’s growth
in the embryo. One powerful-but still unproven-idea is that LTP, or closely
related mechanisms, are involved in fine-tuning connections between neurons
during late stages of the brain’s growth. If this is true, argue researchers
such as Kandel, then one can think of the synaptic mechanisms that underlie
learning as ‘relics’ from the embryo, a sign that the brain never quite
stops growing.

Next week: Sexuality and the brain

* * *

1: Memory in artion

Memory is a diverse collection of talents, each of which may depend
on different learning mechanisms and involve different areas of the brain.
Imagine that you are talking with a friend about the day you first learned
to ride a bike. Your conversation involves: semantic memory, your knowledge
about language and the world, including the concept of a bike and what you
know about bikes and how to ride them; and episodic memory, your memory
of the day itself and all the surrounding circumstances, including the cut
you inflicted on your knee, or the feeling of elation when you succeeded
in staying upright. Researchers sometimes bracket together these two types
of memory, episodic and semantic, as declarative memory-memory that can
we can bring to mind and reflect upon. It is this kind of memory which is
affected in patients with damage to the hippocampus and nearby regions of
the brain.

Your conversation is about another type of memory: the skill of riding
a bike. Skills are classified as part of non-declarative memory along with
other phenomena, such as classical conditioning-the variety observed in
Pavlov’s famous dogs, which learnt to salivate in response to a bell that
had regularly been sounded just before they were fed. Experiments on conditioning
in rabbits have shown that this type of memory involves a part of the brain
called the cerebellum, which is located at the back of the brain.

Your conversation depends on another brand of memory: working memory,
which enables us to hold fleeting material in our heads so that we can build
and understand complex sentences. This type of memory would also come into
play if you were to read off the serial number of the bike in order to write
it down.

Neuropsychologists have identified three main mental components to working
memory. The phonological loop is the machinery that enables us to retain
a sequence of digits, letters or words. The visuo-spatial scratch pad is
a kind of inner eye that lets us do things like rotating shapes in our heads,
or remembering our place on a printed page, and the central executive helps
us to perform a host of tasks such as reasoning and mental arithmetic.

* * *

2: The supple synapse

The human brain has about 100 billion neurons and 100 000 billion synapses.
If, as neuroscientists believe, learning results from small adjustments
to the strengths of these synapses, then the origins of the brain’s immense
capacity are clear. Even storing information at the low average rate of
one bit per synapse, which would only require two levels of synaptic activity
(high and low), the structure as a whole would generate 10 sup 14 bits.
Today’s supercomputers, by comparison, command a memory of about 10 sup
9 bits.

Such calculations are all very well, but the key challenge for neuroscience
is to show how synapses store information. Twenty years after it was first
discovered, long-term potentiation (LTP)-in which synapses linking active
neurons appear to grow stronger through use-is still researchers’ best lead.
How does LTP work? Researchers agree that LTP is triggered when calcium
pours into postsynaptic neurons through special channels in neuron membranes.
This influx may activate enzymes such as protein kinases, which can add
phosphate groups to various proteins and so change their properties. Two
kinases have been implicated: calcium/calmodulin dependent kinase II (CaMKII)
and protein kinase C. Experiments on brain slices show that chemicals which
muzzle these enzymes block LTP.

There is less agreement about the way LTP is put into effect and maintained
once it has been triggered. A synapse is a partnership between two neurons;
either, or both, could be responsible for increasing the efficiency of that
synapse. Some researchers look to the presynaptic half of the partnership.
They argue that the basis of LTP is an increase in the amount of neurotransmitter
(in this case the amino acid glutamate) released by the presynaptic neuron
whenever it is electrically active. Tim Bliss and his colleagues at the
National Institute for Medical Research, London, have measured the concentration
of glutamate around synapses in parts of the hippocampus and found that
it is higher after LTP.

Other researchers focus instead on the postsynaptic neuron. They propose
that LTP is sustained by an increase in either the number or the sensitivity
of receptors for glutamate catried by the postsynaptic neuron. Over the
past two years various research teams have tried to resolve the question
by scrutinising the synapses in question.

Neurons release neurotransmitters in small, uniformly sized packets,
with each nerve impulse prompting the discharge of a large number of packets.
In theory, it should be possible to use microelectrode techniques to discover
whether LTP is accompanied by release of more packets at a time-a presynaptic
effect-or by an increase in the effectiveness of the individual packets,
implying that the postsynaptic neuron has become more sensitive.

Neither side in the debate has been able to claim outright victory.
The impasse may simply reflect the intense difficulty of such research-which
is at the edge of what is technically possible-or it may mean that LTP is
a Janus-like phenomenon.

Yet one thing is certain. If LTP is triggered by calcium rushing into
the postsynaptic neuron but maintained, at least in part, by changes in
the presynaptic neuron, then some signal must pass back across the synapse,
informing the presynaptic neuron about its partner’s state.

The search is now on for this ‘retrograde messenger.’ One candidate,
according to research by Bliss and his colleagues, is a fatty substance
called arachidonic add. However, neurobiologists are becoming increasingly
excited by evidence pointing to another, faster-acting molecule, nitric
oxide. Some studies suggest that chemicals which block enzymes needed to
produce nitric oxide also block LTP. The results, though far from conclusive,
raise the prospect of a neat irony-namely, that one of our most complex
mental functions (learning) might ultimately depend on the simplest substance
in the brain.

* * *

3: Snails, slugs and slow-wiffed flies

Sea snails that behave like Pavlov’s dogs and fruit fly dunces that
forget their lessons: these are arnong the many invertebrate animals that
have bolstered the notion that simple forms of memory depend on specific
structural and molecular changes at synapses.

Slugs and snails are susceptible to classical conditioning. The sea
snail Hermissenda, for example, can learn to associate a flash of light-which
is normally attractive-with an unpleasant stimulus, such as rapid rotation
on a turntable. After a suitable period of training it shuns light. This
memory is the work of a biochemical mechanism that relies on enzymes called
protein kinases, which modify ion channels and change the electrical properties
of neurons.

Another darling of neuroscience laboratories is the sea hare Aplysia.
This creature reacts to touch by retracting its gill-a response that becomes
all the more vigorous when the animal is given an electric shock to the
tail.

Researchers have shown that this process of ‘sensitisation’ is the work
of a biochemical cascade which culminates in the release of larger amounts
of neurotransmitter at a certain type of synapse-and hence a more vigorous
retraction of the gill.

One of the players in the cascade is cylic adenosine monophosphate (cyclic
AMP), a powerful messenger molecule that activates many enzymes in cells.
Experiments with mutant strains of fruit fly have also implicated cyclic
AMP in learning. One strain, dunce, is hopelessly forgetful; it can learn
a task such as avoiding a smell linked to a shock, but cannot retain it
for more than a few moments. Dunce turns out to have a faulty gene that
causes it to accumulate too much cyclic AMP. Another stupid mutant fly,
rutabaga (the North American name for the swede) has unusually low levels
of cyclic AMP.