Gary Marcus

The Birth of the Mind

Regardless of which species we talk about, or which aspect of mental life we investigate, the ability to learn starts with the ability to remember. An organism can learn from experience only if it can rewire its nervous system in a lasting way; there can be no learning without memory. Most research on the biology of memory has focused on something I'll call 'synaptic strengthening.' Synapses, the connections between one neuron and the next, are thought to vary in strength, with strong connections between neurons that are in some way closely tied together. Let us suppose that a simple organism has one neuron for recognizing a special sound, call it the 'bell neuron,' and another for triggering the complex set of cells involved in eating, call it the 'munch neuron.' The bell neuron would fire whenever the simple creature heard the bell, the munch neuron whenever the creature began to eat. If the animal was consistently fed right after the bell rang, one might expect that, over time, the connection—the synapse—between the bell neuron and the munch neuron would get stronger, making the creature more likely to want to munch whenever it heard the bell. Indeed, Pavlov's famous experiments with dogs in the early twentieth century suggested exactly this.

And nature does seem to have a process that explains his findings. It is now known as LTP, which stands for 'long-term potentiation.' The idea is that certain kinds of learning might depend on 'potentiating'—strengthening—the synaptic connections between neurons. This process of strengthening a synapse is long and complex—more than a hundred different molecules may be involved, and there are at least fifteen distinct steps in the process, but they can be roughly divided into five basic stages. First, the brain notices that something interesting has happened and some neuron 'fires,' releasing neurotransmitters on the 'transmitting' side of the synapse. Next, the neurotransmitters that are released on the transmitting side bind to appropriate receptors on the receiving side of that synapse. Those receptors then allow charged atoms through. Once inside the cell, those charged atoms launch a biochemical cascade that ultimately switches on a set of early-response genes. Those early-response genes then ultimately launch a second round of gene expression, which in some way (still under investigation) physically strengthens the synapse, quite likely by using many of the same genes and proteins (such as cell adhesion molecules) that direct initial synapse formation endogenously, prior to experience.

Each stage in the process of memory formation has a genetic component. The receptors that respond to neurotransmitters, for example, are proteins, and interfering with them interferes with memory. Genetically engineered mice that have been designed to lack specific kinds of receptors have trouble with specific kinds of learning, and the same holds for mutants that lack other proteins that are important in the process of memory formation, such as CaM Kinase II (a calcium-activated enzyme for energy transfer and signaling). Indeed, the whole process of protein building is essential for long-term memory formation, and interfering with that process can lead to amnesia for specific events, and even preventing songbirds from learning new songs.

Through judicious genetic tinkering, memory can be, at least to some extent, improved. A 1999 study showed that mutant mice that have extra NMDA receptors—special receiving-end 'coincidence' receptors that appear to excel at noting when two things happen at the same time—have better memories than normal mice. The newspaper headlines—'Scientist Creates Smarter Mouse'—were, as usual, a bit of an exaggeration: There were no permanent gains, and no evidence that the mice really were smarter. Mice with extra NMDA receptors did better on some short-term memory measures, outperforming controls on tests that required them to recognize objects. But the gains were fleeting, lasting for a few days, and disappearing by the time the mice were tested a week later, suggesting that the extra receptors help with some intermediate process rather than with the ultimate consolidation into permanent memory. Mind you, even if the results were stronger, I wouldn't recommend that you try injecting NMDA receptors at home. It is a good bet that there is a reason nature hasn't endowed us with massive quantities of them, at least one of which became clear when later studies revealed that the mutants were also more sensitive to inflammatory pain. You probably don't want to remember everything better.

Roughly the same sets of molecules seem to play more or less the same roles in just about every organism that's been studied. As far we can tell, whether a chick is recording the appearance of its mother or a songbird is learning a new song, the mechanisms of information storage appear to be the same. Immediate early genes and NMDA receptors, for example, seem to contribute to the imprinting of a chick, the aversion to a taste that has induced nausea in a rat, and the songlearning ability of a sparrow. As psychologist Randy Gallistel has put it, 'Information is information': 'An important principle in modern computing and communication is that different kinds of information are equivalent and interchangeable when it comes to storage and conveyance; a mechanism suited to store or convey one kind of information is equally well suited to store or convey any other kind.' If memory is like a blackboard, it seems likely that most mental processes may use the same kind of chalk.

We know something about what that chalk might be, but there is plenty left to be discovered. We know little about the mechanisms by which memories are retrieved, and even less about the 'codes' the brain uses to store its memories; it is as if we understood the process by which chalk is applied to blackboards, but nothing of writing or how it is read. Even when it comes to the synapse-strengthening process that I described earlier, which is the best-understood of the neural processes related to memory and learning, there is much that has not yet been resolved. It is not yet clear whether the process of synaptic strengthening truly plays a role in long-term memory, or whether it plays a temporary role only in an intermediate consolidation from short-term memory to long-term memory. And although a great deal of evidence suggests that the genes involved in LTP are necessary for memory, there is as yet no demonstration that they are sufficient for memory. Other genes may well be involved, especially in the formation of permanent memory, which some researchers have suggested might rely not only on changes in synapses but on other mechanisms, such as changes in DNA itself.

Though there may be just one kind of chalk, there is surely more than one blackboard. Neural substrates for memory are found not just in one particular location in the brain, but spread throughout, with different circuits supporting different kinds of memory. Memory systems can be found not only in the hippocampus (which has some role in spatial memory) but also in the cortex, the amygdala, and in a variety of visual and motor areas. Although the same general cascades of biochemical processes—from the binding of receptors for neurotransmitters to the activity of early-response genes to the modification of synapses—take place in each memory system, each memory system has a different function. Memory in the hippocampus has to do with spatial locations, for example, whereas memory in the amygdala has to do with emotional events. Selectively lesioning the hippocampus in a rat selectively impairs the rat's spatial memory. Impairing its amygdala impairs its emotional memory.

Disruptions to memory systems can have very different effects depending on which memory system is disrupted. By selectively disrupting CaM Kinase II (that energy-transfer/signaling enzyme mentioned above) in two different brain locations, Nobel laureate Eric Kandel created two different types of mutant mice: 'hippocampal mutants' with impaired spatial memory and 'amygdala mutants' with impaired emotional memory. (The term 'such-and-such mutant' is laboratory shorthand for an animal, here a mouse, that has been genetically engineered to have a disruption in a particular region of the brain. A hippocampal mutant is an animal that has been genetically engineered to have a disruption in a particular gene normally expressed in the hippocampus, an amygdala mutant one that has been engineered for a disruption in the amygdala.)

The hippocampal mutants had little difficulty learning to fear a tone (or novel environment) that was paired with a foot shock, yet they couldn't find their way out of a circular maze that always had the same exit, even after forty days of practice. The amygdala mutants easily mastered the circular maze, but they never learned to fear the tone that warned of the shock. Imaging studies in humans show similar specialization. Although each memory system appears to use more or less the same set of molecular mechanisms, different cognitive systems store their memories in different places. Same chalk, different blackboards.