HOW do you know who you are? Will you still be the same person tomorrow? How about in 30 years’ time? In a lifetime the brain squirrels away a vast stash of facts, names, faces, sights, sounds and smells, records scores of events and their associated emotions, and helps us acquire countless skills, from tying shoelaces to making popcorn to driving a car. All these memories define who we are. They shape our personalities, anchor us in time and give us a sense of reality. And somehow, even though the molecules from which our brains are built are replaced many times in a lifetime, our memories and sense of self persist. Sometimes for more than 80 years.
Despite the persistence of some memories, others are all too fallible. Caught up in the bustle of daily life, you forget somebody’s birthday or worry that you didn’t turn the hob off after making that popcorn. There are even times when people insist they can remember something that can’t possibly have happened. At its best, however, the capacity of human memory is mind-boggling. In 1973, a Canadian psychologist called Lionel Standing showed volunteers a series of photographs of objects for about 5 seconds each. Three days later, the volunteers were shown paired photographs, one that they had seen before and the other new, and were asked to say which was familiar. Standing increased the number of photographs shown to each person to an astonishing 10,000 and still they managed to identify the ones they’d seen before, with very few mistakes. Although this experiment tested whether they recognised something put in front of them, which is much less challenging than recalling something without any external cue, the results suggest that some aspects of human memory are effectively limitless.
The puzzle of how our brains perform such feats can be investigated at many different levels. These include studies of how behaviour changes with experience and investigations of neural activity, in particular, changes in the synapses that electrically connect nerve cells in the brain. In the past few years, we have even begun to unravel the complex molecular changes that take place within cells as we learn.
Advertisement
It turns out that there are several memory systems, each serving a different purpose. The two major categories are short and long-term memory. In humans, short-term memory is often referred to as working memory, and it holds new information in the mind for a matter of seconds in an active, conscious state. Several parts of the brain work together, at least partly under the supervision of the frontal cortex, to produce working memory. It is reliable, but its capacity is very limited – most people can hold between five and nine items in it, such as the digits of a telephone number. It has three main components: the phonological loop, the visuospatial sketchpad and the central executive system.
You use the phonological loop to talk silently to yourself, and it’s also important for learning new words. By contrast, you use the visuospatial sketchpad to manipulate images in your mind’s eye. The central executive system controls awareness of items held in working memory, and coordinates information from the other two components. It is vital for activities involving planning and decision-making, such as playing chess or organising a timetable. The central executive system is severely damaged in people with Alzheimer’s disease.
What about long-term memory? There are various types, each serving very different functions and located in different areas of the brain. Two major categories are memories involving “knowing that” and those telling us “how to” (see Figure). The first is called declarative memory and includes remembering facts, places, events and so on. The second, procedural memory, allows us to develop skills like riding a bicycle or playing the guitar. Learning these skills requires extensive repetition, and once learned they are rarely forgotten – an adult who has not ridden a bicycle since childhood can usually do so again without a second thought. By contrast, as every student knows, facts are all too easily forgotten.
Declarative memory is subdivided into semantic and episodic memory. Semantic memory is a giant storehouse of items such as words, objects, concepts, places and people. It is fundamental to producing and understanding language, as well as reading and writing. Patients who show a selective loss of their semantic memory, known as semantic dementia, have provided important insights into the way this kind of memory is organised. The types of mistake made by them show that semantic memory is a bit like a filing system that allows us to retrieve information efficiently. For example, patients might call a mouse a dog, but are unlikely to call it a car or a mobile phone – even though in some ways, such as size, a mouse might look more like a cellphone than a dog. This shows that animate and inanimate objects are stored in separate categories, but the contents of each can get jumbled.
Episodic memory concerns events in your personal life history that you can consciously recall and replay in your mind. It is often described as a snapshot system, because events happen only once and the memory has to be created in one go. Facts, on the other hand, can be hammered in by repetition. Episodic memories stretch back through most of your life. Studies of patients with dementia show that recent events are probably not stored in the same area of the brain as distant ones. For example, people with Alzheimer’s tend to forget recent events, while their childhood memories are relatively well preserved. Patients with semantic dementia also suffer damage to their episodic memories, but the pattern is reversed: they generally remember events from the past couple of years but lose their childhood memories.
Recent episodic memories are stored in the hippocampus, but they are rehearsed over time and are gradually transferred to the recently evolved, outer layers of the cortex collectively called the neocortex, possibly during sleep. The pattern of memory loss in different types of dementia fits this view: in Alzheimer’s, the hippocampus is progressively destroyed and with it the ability to store recent memories, whereas in semantic dementia, the neocortex is the main site of damage.
Early clues about the importance of the hippocampus in memory came from studying patients with clinical amnesia, in particular a patient known as HM. In 1953, in an attempt to relieve HM’s epilepsy, surgeons had removed some of his brain, including parts of the hippocampus, amygdala and temporal lobe. The importance of these structures in memory was unknown at the time and the consequences of the operation were catastrophic. HM became completely unable to commit new facts or events to long-term memory, although his motor skills were unaffected. Strikingly, his performance in tests relying on procedural memory – such as tracing lines while watching the process in a mirror (with left and right reversed) – improved with practice over several days. However, he would always deny that he had ever seen the test before. His intact procedural memory allowed him to improve his performance, but his severely disrupted declarative memory prevented his remembering the learning process.
Other functions such as language, reasoning ability and intelligence (as measured during tasks that did not require short-term memory) were unaffected by the operation. As well as showing the importance of the hippocampus, the tragedy of HM revealed that memory can be studied independently of other, higher cognitive abilities such as thought, language and consciousness.
Patients like HM provide invaluable insights, but scientists are wary of drawing conclusions exclusively from damaged brains. Brains are highly adaptable or “plastic” organs – if one part is damaged, in time other parts may compensate. Stroke patients, for instance, are sometimes able to relearn skills they have lost, either because their brains make new connections or because an undamaged region takes over the functions of a damaged part. An alternative approach to studying memory is to use techniques such as positron emission tomography (PET) or magnetic resonance imaging (MRI) to look at healthy brains. PET involves injecting short-lived radioisotopes into the bloodstream and detecting where they are in the brain up to half an hour later. If the person being studied is given a mental task, such as memorising poetry, active areas of the brain “light up”. MRI relies on the high-frequency radio waves emitted by water molecules in living tissue placed in a strong magnetic field, and allows very rapid mapping of active brain areas.
Scientists at the Institute of Neurology at University College London used PET and MRI to show that the hippocampus is used in spatial memory and navigation. Experienced London taxi drivers were asked to describe the route they would take between two points in London, a task that would require them to access a mental map encoded in semantic memory. The brain areas that became active during this task were compared with those that lit up when the drivers were asked to visualise a series of world-famous landmarks in cities they had never visited.
Although similar brain areas were used for both tasks, there was a striking difference: the right hippocampus was active when route planning, but not when visualising landmarks, showing that it is involved in complex navigation. What’s more, when the researchers compared the volume of taxi drivers’ hippocampi with those of other people, one particular region was significantly larger. This change in their brains, probably as a result of their work, is an example of brain plasticity. Conversely, people with a damaged hippocampus have great difficulty navigating.
As well as working out where different types of memory are encoded in the brain, scientists also aim to unravel the changes that take place between and within cells in the brain as we learn. Of course, practical and ethical constraints make it impossible to investigate such changes in the human brain. Instead, biologists have studied the brains of other animals.
First, they worked out how nerve cells, or neurons, fire electrical impulses by studying squid, whose neurons have giant axons nearly 1 millimetre in diameter. An axon is the biological equivalent of an electric cable. When it’s not firing off signals, a neuron has a difference in electrical charge across its cell membrane called a resting potential. The instant a nerve is stimulated, channels in its membrane open, allowing positively charged sodium ions to flood into the cell. The resulting rapid change in electrical charge across the membrane is called an action potential, and this passes in a wave down the axon.
The neuron transmits this signal to a target cell – another neuron, a muscle cell, or a gland – via synapses at the end of its axon. These are tiny clefts between 20 and 50 nanometres wide. The action potential causes the release of chemicals called neurotransmitters, which diffuse across the gap and bind to receptor molecules in the membrane of the target cell (see “Secret language of cells”, Inside Science No. 148). Depending on the target cell, the signal may make a muscle contract, a gland release a hormone, or another neuron fire – contributing to memory formation.
Neurons in different areas of the vertebrate brain, including the hippocampus, have very similar characteristics, so memory cannot be a property of the cells themselves. But each neuron connects with thousands of others in an intricate network, providing huge potential for encoding complex information by altering the pattern and strength of the synapses between them.
Mammalian brains contain about a trillion nerve cells, making it almost impossible to work out what happens to synapses during learning. Instead, scientists have turned to the sea slug Aplysia, whose simpler nervous system consists of a modest 20,000 nerve cells. Aplysia’s nerve cells are large and arranged in characteristic patterns that are the same in different slugs, allowing scientists to trace the effects of stimulating or manipulating corresponding cells in different animals. This approach was pioneered by Eric Kandel of Columbia University in New York, and in 2000 his contribution to our understanding of memory earned him the Nobel Prize in Physiology or Medicine.
Kandel’s team studied simple kinds of memory called habituation and sensitisation, which are seen in all animals with nervous systems. Habituation is learning to ignore a repeated, unimportant stimulus – for instance, you’re aware of the feeling of clothes on your skin when you get dressed in the morning, but you quickly stop noticing it. Conversely, during sensitisation an animal learns to react to a range of stimuli that are usually neutral, if they are accompanied, for example, by danger or food. So if you heard a gunshot uncomfortably close to you, the chances are you’d jump out of your skin if someone then tapped you on the shoulder.
Aplysia breathes through its gill, which is in a cavity on its dorsal (upper) side. The rear end of the cavity forms a fleshy siphon through which it expels water. Gently touch the siphon and the animal withdraws both gill and siphon in a simple defensive reflex. The strength of the reflex can be reduced by habituation, or increased by sensitisation. The sea slug can be trained to develop either a short or long-term memory for this. It has been found that short-term memory involves strengthening existing synapses, while proteins must be manufactured to build new synapses to store long-term memories (see “Learning in a dish”).
But does the procedural memory of a slug relating to a simple reflex have anything in common with how you recall the name of an old friend? Do similar molecular processes create the more sophisticated declarative memories that are stored in vertebrate brains?
Like memory in sea slugs, declarative memory in mammals has a short-term phase that does not involve protein synthesis, and a long-term phase that does. The main cellular model for memory in mammalian brains is an intriguing phenomenon called long-term potentiation (LTP), which was discovered nearly 30 years ago in anaesthetised rabbits. Repeated electrical stimulation of certain neurons in the hippocampus makes these cells more responsive to subsequent stimulation, and increases the spontaneous firing of neurons further along neural pathways. LTP can last for hours, days, or even weeks. It may be what happens when something is being stored in memory.
Several characteristics of LTP make it a convincing model for memory. The effect is confined to specific, activated synapses, and does not spread to other synapses on the same cell, so each neuron has the potential to store a huge amount of information. What’s more, like memory, LTP develops in phases. The transitions between phases are analogous to transitions between short and long-term memory.
Finally, LTP shows a property called associativity: weak stimulation that would normally not be enough to trigger LTP may do so if, at roughly the same time, strong stimulation triggers LTP at another set of synapses on the same cell (see Figure). In other words, the cell is forming a link between two stimuli. Pavlov’s classic experiments with dogs showed that one way in which animals learn is by associating two stimuli that occur at roughly the same time (see “Cells that fire together, wire together”). Unlike habituation and sensitisation, association involves the linking of highly specific stimuli.
The molecular details of LTP have been worked out using various approaches, including blocking parts of the pathway with drugs, and studying “knockout” mice that have been engineered to lack genes that code for key enzymes.
In a neural pathway, a cell that stimulates a neighbouring cell via a synapse is known as a presynaptic cell, and the stimulated cell is a postsynaptic cell. LTP begins with the release of a neurotransmitter, glutamate, from the presynaptic neuron in response to electrical stimulation (see Figure). Glutamate binds to two types of receptor on the postsynaptic neuron, called AMPA and NMDA. These are known as gated channels, because when glutamate binds to them they open up (see “Border control”, Inside Science No. 132).
When glutamate binds to AMPA receptors they allow sodium to flood into the postsynaptic neuron, triggering an action potential. This is what happens whenever one nerve transmits an impulse to another. The situation is more subtle with NMDA receptors because they will only open up in response to glutamate if the membrane is already slightly depolarised as a result of an earlier impulse from another nerve. This means they act as coincidence detectors, allowing the cell to respond more strongly to inputs at one set of synapses if it has just been stimulated by another set. In other words, the cell is learning to associate the two stimuli. It remembers.
How do NMDA receptors do this? Normally, when the cell membrane is at its resting potential, a magnesium ion prevents the channel from opening in response to glutamate binding. But the ion is expelled if the membrane is slightly depolarised. Then, when glutamate binds to the receptors, they not only open the floodgates to sodium but also to calcium ions. Calcium activates an enzyme named CaMKII, which in turn acts on the AMPA receptors, making them more permeable to sodium ions. This means the cell is more sensitive to electrical impulses, but at the activated synapse only. This helps to explain the high specificity of LTP.
Even though calcium levels in a postsynaptic neuron quickly drop back to normal levels, LTP can persist for weeks. This happens because when calcium activates CaMKII, it flips a molecular switch in the enzyme, leaving it permanently “on”. And if the stimulation is repeated, the raised calcium levels recruit other enzymes that trigger the synthesis of new proteins and the building of new synapses.
Scientists are beginning to relate the molecular details of LTP to some aspects of memory. Like London taxi drivers, animals need to be good at finding their way around. They build up mental maps of their environment in the hippocampus which they use to find food, avoid predators, and so on. The maps are created by the activity of specialised hippocampal cells called place cells, which fire only when an animal is in a specific location. Put a rat or mouse somewhere unfamiliar and within minutes a group of place cells begins firing. As a result of LTP, synaptic connections are built up as the animal gets to know its new environment, generating a mental map that remains stable for days. But if the animals are given drugs that block NMDA receptors, LTP is prevented and the animals are unable to remember their new environment.
Research like this is revealing how and where memories are created and stored in the brain. Practical benefits could include developing drugs that halt or reverse the progression of dementias such as Alzheimer’s disease. But will healthy people ever be able to pop a pill that enhances their memory? Even if they could, it might have unforeseen consequences. A good memory is selective: a delicate balance between remembering the important things and forgetting the unimportant. Nevertheless, understanding how the brain works can certainly help us to hone our learning skills and make the most of its amazing, prodigious memory.
Cells that fire together, wire together
If you’re hungry, the mere sight of food makes your mouth water. This is an “unconditioned”, inbuilt reflex. At the end of the 19th century, the Russian physiologist Ivan Pavlov observed that if a bell is rung just before a dog is fed, it eventually learns to salivate at the sound of the bell alone. He called this a “conditioned” reflex. Pavlov also found that timing was crucial, because if he rang the bell too long before or after the food was brought, the dogs did not learn to salivate when they heard it.
In the 1940s, the American psychologist Donald Hebb tried to explain Pavlov’s results at a cellular level. Hebb proposed that when sensory receptors in a dog’s eyes or nose are activated by the sight or smell of food, they stimulate other neurons (nerve cells) via synaptic connections, and eventually the signal reaches neurons in the cortex. These are connected to neurons that trigger salivation by stimulating the salivary glands. Other neurons, for example those originating in the ears, make weak connections via synapses with the cortical neurons in this pathway. If the ring of a bell coincides with the sight of food, signals from both sets of sensory neurons arrive at the cortical neurons at about the same time. Eventually, the synapses from the auditory nerves are strengthened until the sound of the bell alone is enough to make the dog salivate. This is often summarised as “cells that fire together, wire together.”
Learning in a dish
In an unruffled sea slug, sensory neurons pick up a touch to its siphon and transmit a signal to motor neurons, which make the gill muscle contract. This pulls the gills out of harm’s way. Give the animal’s tail a mild electric shock, however, and it becomes “sensitised” – a gentle touch on its siphon makes it snatch back its gill for longer than usual.
A single shock results in sensitisation lasting only a matter of minutes – short-term memory. But even in slugs, it seems, practice makes perfect: after four or five shocks at intervals, it remembers for several days, and if the training is repeated over four days, the memory lasts for weeks.
During sensitisation, sensory neurons in the tail detect the shock and release a neurotransmitter, serotonin. This binds to receptors in the sensory neurons from the siphon. For short-term memory, a “molecular cascade” – a series of molecular changes inside the neurons – stimulates the release of more serotonin at the synapse between the sensory and the motor neuron for the gill muscle. This enhances the signal to the motor neurons, making the muscle contract for longer. For long-term memory, an extra step in the cascade triggers gene transcription, boosting levels of particular proteins that are used to build extra synapses between sensory and motor neurons.
- • Neuroscience edited by Dale Purves and others (Sinauer Associates, 2001)
- • Molecular Cell Biology by Harvey Lodish and others (W. H. Freeman, 1999)