Memory, p.9
Memory, page 9
For a neurobiologist, the crucial questions are about how these forms of memory are instantiated in the brain. Do the categories reflect the engagement of different brain regions, and different molecular processes, or are they higher-level distinctions, without matching brain correlates? Until recent decades, the only effective way of addressing these questions in humans was by observing the effects of various forms of brain lesion and disease on memory. Classical disease-induced losses of memory, notably from what used to be called senile dementia but more frequently now Alzheimer’s disease, can’t answer these questions because the brain damage they cause is both progressive and very general. But some consequences of stroke, or accident – or surgically induced damage – can be instructive. The most famous case of iatrogenic memory loss is of an epileptic patient, known to every neuroscientist by his initials, HM, who was operated on in the 1950s to remove regions of his temporal cortex and hippocampus so as to eliminate the epileptic focus. The result was a catastrophic loss in his ability to transfer memory from short to long term. HM, who has continued to be a subject of research for the subsequent half-century, can remember events up to the time at which he was operated on, but forgets any new experience within minutes. Although he can show some procedural learning of new skills, he cannot retain declarative – especially episodic, autobiographical – memory. Events, as he himself puts it, simply fade away; he says ‘every day is by itself’. This observation, soon matched by studies in animals, suggests that the hippocampus has a crucial role to play in the registration of new experience, and that without it and adjacent brain regions items can no longer be transferred into longer-term memory.
Within the last decades, the possibility of studying ongoing memory processes in the living human brain has been transformed by the advent of new technologies, notably the windows opened by functional magnetic resonance imaging (fMRI) and magnetoencephalography (MEG). The former makes possible the measurement of changes in blood flow to small regions of the brain, the assumption being that the higher the rate of blood flow the more active that region is under any particular circumstance, such as performing a learning or memory task. The latter takes advantage of the fact that signalling within the brain is primarily electrical, and electrical current flow is accompanied by tiny changes in the magnetic field surrounding the current. Both techniques require rather formidable instrumentation; fMRI is better at localising sites of change, MEG is most helpful in charting the temporal dynamics, making it possible to plot changes in brain activity millisecond by millisecond.
Two examples reveal what can be learned from such techniques. Eleanor Maguire and her colleagues studied London taxi drivers asked, whilst under fMRI, to recall a complex journey within the city. The act of recalling the route activated their hippocampi. In our own experiments, using MEG, we took subjects on a virtual supermarket tour, asking them to make choices of items to purchase based on their past experience and preferences. Faced with a choice, say of three brands of coffee, subjects took about two seconds to press a key indicating their choice. But in those two seconds, there was a flurry of brain activity. Within 80 milliseconds, the visual cortex became active; by 300 milliseconds, the left inferotemporal cortex, assumed to be a site of memory storage. At 500 milliseconds, Broca’s area, a region associated with speech, was engaged as the subjects silently vocalised the range of choice items – and at 800 milliseconds, as they made their final decision as to which item they preferred – assuming they preferred any – the right parietal cortex, associated with affect-laden decisions, was active. These dynamics reveal the many regions of the brain involved in even a simple act of episodic and semantic memory; even primary sensory regions like the visual cortex are more active when people are performing a memory-related task than when they see the same images but are asked simply to make a cognitive choice – for instance of which item is the shortest of those displayed. Thus Baddeley’s working memory does not seem to be simply localisable to one brain region.
Revealing though such studies are – and they are as yet in their infancy as the techniques and instrumentation mature – there are limits to the types of answer they can provide. If learning and the making of memory demand cellular and molecular changes, these cannot be studied except in animals. To do this demands developing models of learning and memory in these animals that may serve in some sense as a surrogate for the same processes in humans. The doors to such an approach were opened early in the last century by Ivan Pavlov’s well-known experiments with dogs. Pavlov trained them to associate the ringing of a bell with the arrival of food, and hence to salivate (a learned or, in the jargon, a conditioned reflex). In the 1930s B.F. Skinner developed a different learning model (operant conditioning) in which animals had to perform some act, such as pressing a lever to obtain food, or to escape an electric shock. If after one or more trials the animal’s behaviour changes appropriately – for instance, by salivating to the bell or by pressing the lever sooner in response to a signal or running a maze faster and with fewer errors – the animal is said to have learned from the experience. And, when it performs the learned task in an errorfree way, it is said to be remembering the experience. The unspoken assumption is that whatever the brain processes that are involved in such changes in the animal’s behaviour may be, they will be similar to those occurring in the human brain when we learn and remember. The Skinnerian view was that all creatures learn and remember in the same way: that there are general laws of learning that are as universally applicable as the gas laws or gravitation in physics.
Of course, there are problems with such an assumption. What and how an animal will learn is species-specific. Some food-storing birds, such as scrub jays, can recall during winter the many thousands of sites at which they cached edible seeds the previous summer. Others – songbirds like zebra finches – cannot learn such tasks but readily acquire new songs. Further, an animal can only inform a human experimenter that it is learning or remembering by way of some change in its performance of a task. It may ‘remember’ its previous experience but choose not to perform the task appropriately – a point made in the 1950s in a famous critique of Skinnerian approaches, a paper called simply ‘The Misbehaviour of Animals’. Despite heroic attempts at complex experimental designs, the taxonomic distinctions between procedural and declarative learning and memory are always going to be confounded in animal studies.
Yet, if learning from some new experience results subsequently in a change in the behaviour of the animal when presented with a similar situation, one must assume that something has changed in the brain to support the changed behaviour. This inferred intervening variable is regarded as a memory ‘store,’ ‘trace’ or ‘engram’, which is formed when learning is taking place, and reactivated when that learning is later recalled. The challenge for neurobiologists then became that of identifying the anatomical, cellular, molecular or physiological nature of the trace. The temporal distinction between short- and long-term memory, the evidence that short-term memory is labile and easily disrupted, whereas long-term memory seems relatively protected, suggested that it must depend on some structural remodelling of the patterns of neural connection within the brain, engraving the memory in the brain in a manner analogous to that of inscribing a magnetic trace on a tape or a CD which can subsequently be replayed, invoking the original material. The seductive metaphorical power of computer ‘memory’ has been very influential in shaping thought on this question.
The myriad nerve cells in the brain (a hundred billion in the human cortex alone) communicate by way of up to ten-thousand-fold (a hundred trillion) more junctions, known as synapses. It is at the synapses that electrical signals travelling down one nerve axon trigger the release of chemical signals – neurotransmitters – that in turn carry the message across a small gap to an adjacent nerve cell, stimulating a response in the second cell. Maybe learning results in some change in synaptic connections, so as to create novel signalling pathways? In 1948 the Canadian psychologist Donald Hebb framed the hypothesis that has shaped all subsequent biochemical and physiological research in the field, that learning involves the remodelling of such synaptic junctions. In his own words (and indeed his own italics):-
Let us assume then that the persistence or repetition of a reverberatory activity (or ‘trace’) tends to induce lasting cellular changes that add to its stability. The assumption can be precisely stated as follows: When an axon of cell A is near enough to excite a cell B and repeatedly or persistently takes part in firing it, some growth process or metabolic change takes place in one or both cells such that A’s efficiency, as one of the cells firing B, is increased.
The most obvious and I believe much the most probable suggestion concerning the way in which one cell could become more capable of firing another is that synaptic knobs develop and increase the area of contact between the afferent axon and efferent [cell body]. There is certainly no direct evidence that this is so … There are several considerations, however, that make the growth of synaptic knobs a plausible perception.
By the 1960s, neuroscientists had become sufficiently confident in the power of their technologies to attempt to verify Hebb’s hypothesis experimentally. But after an initial burst of enthusiasm the field became mired in controversy. Claims that training rats on some simple task resulted in increases in RNA [ribonucleic acid] and protein synthesis in their brains and, even more extravagantly, that when the RNA was extracted and injected into the brain of a recipient the memory was transferred too, achieved great publicity but were technically flawed. Research funds dried up and even to suggest that one was working on the biochemistry of memory became somewhat disreputable.
More patient experiments in the 1970s began to revive confidence, and Hebb’s plausible perception became tangible evidence. Two disparate approaches helped. One seems very remote from memory as we might understand it. The physiologists Tim Bliss and Terje Lomo placed stimulating and recording electrodes into cells in the rat hippocampus, and found that if they fired a train of electrical pulses into the cells, their output properties were permanently modified; the cells showed a ‘memory’ of their past experience. The phenomenon, called long-term potentiation, became intensely studied as either a mechanism or a model for memory over the succeeding decades. At around the same time, the psychiatrist turned neuroscientist Eric Kandel began exploring the physiological properties of the neurons in the giant sea slug Aplysia californica. Aplysia has two useful properties. One is that it can be trained on a simple task, to contract its gills (rather as a land slug rolls into a ball) in response to a jet of water being applied to its tail. The second is that many of its nerve cells are giant and the ‘same’ cell is easily identifiable from slug to slug. Kandel was able to map the neural circuitry involved in the withdrawal reflex, and to identify some specific synapses whose electrical properties and biochemistry changed as the slug learned. With a reductionist rhetorical flourish, Kandel offered the research community ‘memory in a dish’.
Over the decades that followed, evidence from a variety of labs, including my own, showed that indeed, when an animal – in my case a young chick – is trained on some novel task, there are increases in the size and strength of specific synaptic connections in particular brain regions. Under the microscope, the connections are structurally larger and the efficacy of the neurotransmitters within them is enhanced. The experimental problem was to prove that these changes were in some way associated with the storage of the putative memory trace, rather than a consequence of other aspects of the task and its performance. For example, in the task we use, young chicks are offered a small bright bead. Almost invariably, they will peck at such a bead within a few seconds of seeing it. If the bead is made to taste unpleasant (we dip the bead in a rather bitter, curryish tasting liquid), the chick will peck once, and then demonstrate its distaste by shaking its head energetically and wiping its bill on the floor of its pen. If it is subsequently – any time up to several days later – offered a similar but dry bead, the chick will not peck it, but back away, sometimes replicating the earlier pattern of head shaking and bill wiping. We infer that the chick has learned, after a single experience, that this particular colour, shape and size of bead tastes unpleasant – at least in this specific context – and that when the bead is presented once more, this memory is reactivated. For the initiates, this is described as one-trial passive avoidance learning.
The passive avoidance task has a number of experimental advantages. It is quick and reproducible, and builds upon a normal aspect of the young chick’s behavioural development – that is, to spontaneously explore its environment by pecking at small objects. Because the training event – the peck at the bead – takes only a few seconds, one can readily separate the immediate consequences of the bitter taste from the subsequent cascade of events during the transitions between shorter- and longer-term memory. Advantages have corresponding disadvantages. Is what we discover about learning in such a young animal, where the brain is developing rapidly, relevant to learning in adulthood? Do the molecular events involved when a chick learns in a single trial to peck a bead in any way correspond to those during the many trials a rat needs to learn to run a maze – still less those when a child learns the names of the days of the week or what to expect on its birthday?
Even setting these queries aside, can we be sure, even for the chick, that the change we find in the synapses is actually some form of memory trace? That is, that it is a necessary, sufficient and exclusive change in the brain which in some way ‘represents’ the memory, enabling it later to be recalled? Could the change not have occurred simply as a result of some aspect of the initial experience, such as the taste or sight of the bead, or the learned motor activity of pecking? Or, as we cannot know whether the chick has learned the task without testing it, maybe it is a consequence of the recall experience rather than the learning itself? I don’t intend here to reprise the decade-long series of control experiments that enabled us to distinguish between general experience-induced and learning-induced changes. I have discussed these at some length in my book The Making of Memory. But it may be of more than merely technical interest to outline the sorts of approach one can use.
There are broadly two approaches to identifying the molecular processes that occur in the minutes to hours following training on a simple task such as passive avoidance and which are presumed to be required for the maintenance of short-term memory and to underpin the transition to long-term memory, a process called memory consolidation. One can train the animal on the task and look for changes in some putative biochemical measure – the activity of an enzyme, the concentration or rate of synthesis of a molecule. Or one can attempt to disrupt the consolidation process by administering an inhibitor – some drug or anti-metabolite, known to block a specific biochemical process believed to be necessary for consolidation. If the drug blocks such a process, then the animal should subsequently not recall the task; that is, it should show a specific amnesia. Observing the changes in the suspected biochemical measure over time, or the time window during which the administered amnesic agent is effective, makes it possible to plot a temporal sequence of molecular events – a biochemical cascade – occurring over the hours following training and which seem to culminate in the lasting modulation of synaptic strengths. Since the 1980s I have used both methods in tandem in elucidating this cascade in my chicks.
Within the minutes following the onset of the training experience, there are changes in the release of neurotransmitters at the synapses in specific brain regions. As well as activating the post-synaptic nerve cell to fire, these increases also stimulate a wave of biochemical activity in the cell, which in due course results in the synthesis of a family of proteins, called cell adhesion molecules, destined to be transported to the synapses. Cell adhesion molecules are a bit like Velcro. They are located in the cell membrane, for instance at the synapse, with one end (the Velcro end) sticking out into the space between one nerve cell and the next, holding the two sides, pre- and post-synaptic, together. The newly synthesised adhesion molecules that are produced as a result of the training experience are dispatched to the activated synapses (a process that takes some 4–6 hours in chicks and rats), and inserted into their membranes, altering the strength of connections between the two sides of the synapse. This would seem precisely to confirm Hebb’s hypothesis for how memory might be coded and stored in the brain.
The distinguished Nobel prize winning biochemist, Hans Krebs, in whose Oxford lab I was based during a post-doctoral period in the early 1960s, once told me that for every biological problem, God had chosen an appropriate organism in which to tackle the problem. I have argued that, for the study of the molecular processes involved in memory formation, the chick is indeed God’s organism. Others have made different choices, ranging from fruit flies and sea slugs to the more familiar laboratory rats and mice. In 2001, the Nobel Committee opted for the slug – although the prize they gave its developer, Eric Kandel, was not so much for his memory work with the slug (Aplysia californica) as for his studies of its neurotransmitters. What is interesting and encouraging is that despite the differences and learning paradigms, a sequence of broadly similar molecular processes has been shown to occur in the brains or nervous systems of these varying species during and following the training experience.
