IT IS POSSIBLE TO WATCH A MEMORY forming in the brain. Researcher Sheena Josselyn at the Hospital for Sick Children in Toronto, for instance, looks through an ultralight “mini endoscope” attached to a lens in the brain of a mouse. The animal’s brain cells have been tagged with substances that make them light up when they become active, and she waits to see exactly which neurons “switch on” in response to an event she thinks the animal will remember—most often, a mild electric shock.

Josselyn is fascinated by fundamental questions about memory, including how individual neurons “decide” to become part of a particular memory, called an engram. In a landmark experiment, published in 2009 in Science, she noted the role of neurons’ “excitability.” At any given time, certain neurons show an increased readiness to pass along electrical signals and connect with other neurons. Using a genetic tweak in mice to increase levels of a protein called CREB, she was able to increase the excitability of particular neurons, and found that thisinfluenced where in the brain a memory was encoded. She showed that the CREB-boosted cells were nearly four times as likely as neighboring neurons to become part of the engrams that her team prompted.

The memory in that experiment was an association between a shock and a musical tone. To prove that the neurons actually encoded this, she turned off the brain cells that had lit up during memory formation, selectively killing them with a toxin. When exposed to the sound again, the mouse appeared to forget what it had learned, no longer freezing in anticipation of a shock.

Josselyn has conducted a host of similar experiments, both refining her team’s ability to map engrams and devising new techniques to switch them off and on again. Learning how to manipulate specific engrams in this way could yield revolutionary applications not only for memory disorders but also depression—for example, by restoring positive memories—and addiction, by eliminating pleasurable memories of using alcohol or drugs. And Josselyn says she thinks a similar approach could, if scientists can figure out how to map the complex engrams of human trauma, provide relief from a condition such as post-traumatic stress disorder.

Josselyn’s work is part of a new wave of research in neuroscience, aided by a convergence of increasingly powerful technologies, that seeks to reveal how memories are made, lost and might be restored. It could hardly be more timely, as the world’s population rapidly ages and Alzheimer’s disease and other forms of dementia threaten to erode the memories of millions of people. The National Institutes of Health is expected to spend more than $2 billion on Alzheimer’s research in 2019. But most of that work remains focused on the plaques and neurofibrillary tangles that are known to cause destruction in the brain as the disease progresses. Finding how to preserve and retrieve memories that may be hidden but not yet destroyed could prove to be an important complement to the work on plaques and tangles—and an essential step in helping people avoid or mitigate the loss of an essential part of who they are. “Memory is a fundamental building block of what constitutes humanity,” says Alcino Silva, professor of psychology at UCLA’s Brain Research Institute. “Without memory, how much of you is left?”


THERE ARE MANY TYPES of memory. Implicit memories, such as the motor memory of how to ride a bike, are acquired and used without conscious thought. They reside in the cerebellum and the basal ganglia, deep in the brain’s core. Short-term working memory, a sort of mental scratch pad that retains a small number of items for less than a minute, uses the prefrontal cortex. But most research that looks at memory—and the fear of losing it—studies declarative memory. This includes both memory of events (episodic, or autobiographical memory) and more general facts (semantic memory). It relies on three principal areas of the brain: the hippocampus, the neocortex and the amygdala.

The hippocampus, tucked deep inside the brain’s temporal lobes, is where such memories are formed and indexed. Through an ongoing process called memory consolidation—which happens over the course of weeks in humans, as they sleep and dream—memories initially encoded in the hippocampus are transferred for long-term storage to the wrinkly neural tissue of the neocortex, on the brain’s surface. The amygdala, an almond-shaped structure in the brain’s temporal lobe, attaches emotional content to memories, making them more durable. The amygdala plays a key role in forming new fear-related memories, which tend to form quickly—making mouse experiments that use electric shocks or loud noises a popular way to investigate the mechanisms of memory formation and retrieval.

Wherever they reside in the brain, all memories, at the most fundamental level, rely on what happens in synapses, the junctions between neurons. The theoretical idea that “neurons that fire together, wire together”—that memories are formed through the strengthening of connections that happen in these synaptic junctions—was first proposed by Donald Hebb, a Canadian psychologist in 1949. What came to be known as Hebb’s rule has been the fundamental insight driving the neuroscience of memory ever since, confirmed experimentally across many organisms.

An engram is born, in other words, when electrochemical signals pass through a synapse and this strengthens the connection between the adjacent neurons. With repeated stimulation during learning and recalling these memories later, the synaptic connections form an even stronger bond, in a process called long-term potentiation. Silva calls Hebb’s rule one of the most significant discoveries in neuroscience. “In heredity, the fundamental discovery is DNA,” he says. “In memory, it is knowing that changes in synapses encode memories.”


BUT MEMORIES DON’T EXIST in isolation. Rather, individual engrams overlap and link to each other, forming neurological webs that allow events to be anchored in time and place, and experiences to be cross-referenced with other occurrences. Silva has shown that some of the same processes involved in creating single memories are also essential in forming linked memories.

Silva hypothesized that after encoding a first memory, neurons stay in an excited state for a brief window of time, making it more likely they will participate in forming a second. Two memories created at about the same time thus would tend to use overlapping populations of neurons, increasing the chance that recall of one memory triggers recall of the other. Silva calls this the allocate-to-link hypothesis.

In a series of experiments reported in Nature in May 2016, researchers in Silva’s lab, led by Denise Cai, now an assistant professor of neuroscience at Mount Sinai in New York, tested the allocate-to-link hypothesis in mice. First, they showed that memories linked closely in time were more likely to be recalled together. The scientists did this by placing individual mice in two different cages—first, one in which nothing happened, and then about five hours later, to a second cage in which they got a mild foot shock. After training, all of the mice froze, as expected, in the “shock” cage. But surprisingly, they also froze in the neutral cage. The memories of the two cages—and their association with a shock—appeared to be linked. Repeating the experiment, but leaving more time—longer than a day—between exposure to the two cages, the researchers found no overlap between memories. Rather, the mice behaved “normally,” freezing only in the shock cage.

Next, the researchers looked at what was happening inside the mice’s brains. Employing brain-imaging techniques similar to those used by Sheena Josselyn, they saw that when mice were exposed to the different cages within a few hours, they did, in fact, form memories in overlapping clusters of neurons. When the time between cages was increased, the engrams didn’t overlap. Silva’s group concluded that neurons, after encoding an initial memory, stayed more excitable, or plastic, for up to five hours, creating a window in which they were more likely to be folded into a second memory.

These findings about neuron excitability might also explain other kinds of links among memories, such as those that connect objects to words, or experiences to places. They also have implications for what may happen during aging. Repeating their experiments in older mice, for example, Silva’s group found that while they could remember single memories as well as younger mice did, they weren’t able to link memories of closely related events. The researchers suspected that this was because of a decline in neuron excitability. So they used specially targeted drugs to boost the excitability of specific neurons in older mice, which indeed restored their ability to form overlapping engrams and to link memories behaviorally. Therapies based on these findings are already under development, and could help treat not only age-related cognitive decline but also disorders such as schizophrenia, depression and bipolar disorder, in which problems with neuronal excitability are known to play a role.


THE ANATOMY OF THE neuron itself may offer clues to why memory fails over time. Neuroscientists are increasingly focused on dendrites, branch-like extensions of nerve cells in the brain that probe the spaces between cells, picking up signals and communicating them to the cell body. Jutting out from these dendrites are thousands of tiny, bud-like appendages called dendritic spines, which serve as the actual points of synaptic contact with other neurons. The number and density of these spines is dynamic, fluctuating with time, and they appear to be particularly crucial in learning and memory formation.

For example, in juvenile zebra finches, commonly used in the lab to study vocal behaviors, a higher level of spine turnover—with new spines forming to replace the old—seems to be related to an increased ability to learn songs. As rodents age, however, dendritic spine formation in their brains appears to ebb, a condition that has also been observed in mice genetically engineered to have symptoms of Alzheimer’s and in some forms of autism in humans.

In a 2018 study that tracked dendritic spine dynamics in mice, Silva’s lab demonstrated for the first time in adult mammals that enhanced spine turnover and clustering indeed have a correlation with improved learning and memory. Mice engineered with a genetic mutation that reduces the production of the protein CCR5, which inhibits dendritic spine formation, did better on memory tests, and had more spine turnover before training, compared with non-mutated control mice. Silva says he believes that increased turnover of dendritic spines helps neurons increase synaptic connections during learning. He also thinks that, after learning, clustered spines may help to stabilize these new synapses.

Dendritic spines might also bolster resilience against Alzheimer’s disease. In the October 2017 Annals of Neurology, scientists from the University of Alabama at Birmingham and Emory University School of Medicine compared dendritic structures in the autopsied brains of people divided into three groups: those with no signs of disease; those with some physiological indications of the disease but no symptoms of dementia; and those with full-blown Alzheimer’s. Brains in the control group and from people without clinical dementia had similar levels of spine density, but the brains of people with Alzheimer’s had a significantly reduced number. Variations in spine density could help explain why as many as half of older people with the pathology of Alzheimer’s—for example, the presence of amyloid beta, a protein fragment that forms amyloid plaques in the brain—never develop dementia.

Other research, from Susumu Tonegawa’s lab at the Picower Institute for Learning and Memory at MIT, suggests that enhancing the formation of dendritic spines could help recover certain kinds of “lost” memories. In studies of mouse models of Alzheimer’s disease, in 2016, and retrograde amnesia, in 2017, Tonegawa’s team found that a reduction in the number of dendritic spines contributed to the phenomenon of “silent” engrams, in which memories become inaccessible through normal cues, but can be reactivated artificially by stimulating neurons using optogenetics—targeting specially engineered cells with a light that makes them active—or treating them with a protein called PAK1 that increases spine density. In the late stages of Alzheimer’s disease, the death of neurons wipes away memories completely, but Tonegawa’s research has shown that some earlier memory failures could be a problem of retrieval related to reduced numbers of dendritic spines. That may offer hope that in those cases, memory loss could be reversed.

Right now, there’s no feasible way to shoot bursts of light into the brains of living humans, but PAK1 or other experimental drugs might lend themselves to human therapies. In 2016, for example, a group headed by Jerry Yang at the University of California, San Diego reported it had developed a set of molecules, called benzothiazole amphiphiles, that boosted the growth of new dendritic spines in mice. But Tonegawa’s research suggests that any such interventions would need to be highly targeted. Successfully retrieving silent memories in mice required increasing dendritic spines only in the affected memory cells, rather than in broader populations of neurons. This highlights the importance of the work done by Sheena Josselyn and her colleagues, because if researchers can learn how to map engrams with greater accuracy, it will allow those engrams to become better therapeutic targets.


WHILE AGING IMPAIRS MANY facets of memory, some of the most significant changes involve the inability to keep memories separate and distinct. “The best example of this is if you park your car in a different location in a parking lot every day. There will be many days when you can’t find your car because of interference from other similar memories,” says Amar Sahay, a neuroscientist at the Center for Regenerative Medicine at Massachusetts General Hospital and principal faculty member at the Harvard Stem Cell Institute. Older people often have trouble distinguishing between old and new memories, or may experience “false recognition”—such as when an older relative with dementia mistakes a new visitor for her long-dead husband.

A growing body of research is connecting these cognitive changes to age-related alterations in a tiny region of the hippocampus called the dentate gyrus. The dentate gyrus is where the heavy lifting of memory takes place. It’s where all sensory information entering the brain is initially encoded as representations of “what, when and where” in engrams, before being conveyed to the prefrontal cortex for long-term storage. As the years wear on, the dentate gyrus can gradually become “overloaded” and start to display noisy and unreliable recall, as individual synapses attempt to satisfy the requirements of too many memories.

Counteracting this tendency is the ability of the dentate gyrus to refresh itself by generating new neurons throughout a lifetime, a process called neurogenesis. Apart from the amygdala and the olfactory bulb, which is responsible for the sense of smell, the dentate gyrus is the only part of a mammal brain that continues to grow new cells into adulthood. By providing a steady supply of new cells, neurogenesis ensures that there will be a place for new memories to form, so that new experiences can continue to register clearly and distinctly and be recalled without confusion.

Yet while neurogenesis is an ongoing process, numerous studies in animals and humans have shown that the production of fresh cells ebbs with age. Researchers at the Autonomous University of Madrid measured roughly 42,000 new neurons per cubic millimeter in the hippocampus of a 43-year-old, while a 60-year-old had only between 30,000 and 40,000 per cubic millimeter. But Sahay’s research suggests that such declines may be, in part, reversible. In a study published in Neuron in 2016, scientists in Sahay’s lab, led by Kathleen McAvoy, effectively doubled the population of “fresh” neurons in mice and found that this led to improved memory precision in both middle-aged and older mice. The mice were better able to differentiate between their memories of getting a shock in similar-looking environments and did well in other tests of spatial and context discrimination. “When you improve neurogenesis, the animals are better at discriminating between similar experiences,” Sahay says.

Increasing neurogenesis might also play a role in combating the development of Alzheimer’s, which appears to be accompanied by a sharp decrease in the formation of new cells. In the Madrid brain study, older people with dementia had much lower rates of neurogenesis, with the brain of a 78-year-old who died with Alzheimer’s having only about 10,000 new neurons per cubic millimeter, compared with 23,000 in a healthy brain of the same age.

SAHAY SAYS THAT IN disorders such as Alzheimer’s disease, the connectivity between neurons in the hippocampus are disrupted, leading to inefficient consolidation and storage of memories in the prefrontal cortex, and a nearly universal problem of the aging brain—that memories tend to become less precise over time, losing important details.This generalization of memories is more pronounced in people with age-related cognitive impairments, and it’s connected to what happens in the brains of people with PTSD, in which specific fearful experiences cast a pall over wide swaths of memory.

According to Sahay, growing evidence suggests that these kinds of fuzzy memories stem from an imbalance of excitatory and inhibitory signals passing from the dentate gyrus to the CA3—a section of the nearby Cornu Ammonis region, also in the hippocampus. It serves as a way station where engrams are stabilized and promotes long-term storage of memories in the prefrontal cortex. With aging, the inhibitory signals to the CA3 decrease, thus letting the excitatory signals prevail and resulting in hyperactivity of that part of the brain. That’s a characteristic of mice models of Alzheimer’s and is also seen in humans with mild cognitive impairment.

In healthy brains, hyperactivity in the CA3 region is constrained in part with the help of special cells called inhibitory interneurons—neurons that sit between excitatory neurons and can boost, or weaken, the signals passing between them. In ongoing studies of neurogenesis, Sahay has observed that new, adult-born neurons are better at “recruiting” inhibitory interneurons than older cells, suggesting that any therapy that boosts neurogenesis might, in turn, also help dampen hyperactivity and improve long-term memory precision.

In research published in Nature Medicine in 2018, a team in Sahay’s lab, led by Nannan Guo, showed that it is possible to complement natural neurogenesis by making older cells perform like younger ones by rewiring their connectivity with inhibitory interneurons. The researchers’ method relied on a complex system of checks and counterchecks. Their principal target was abLIM3, a protein that works against the inhibitory, CA3-calming signals coming from the dentate gyrus. By using an engineered virus to decrease levels of abLIM3, they restored connections with inhibitory neurons, reducing hyperactivity in the CA3 and helping improve memory precision in middle-aged and older mice.

That work could relate to an emerging model of long-term memory called multiple trace theory, which suggests that recalling memories depends on ongoing connections between partial engrams that remain in the hippocampus and fuller versions that are stored in the cortex. Sahay says he believes that reducing hyperactivity in the CA3 helps to maintain those engrams in the hippocampus, which serve as a kind of index that helps make memories less vulnerable to erosion over time.

Sahay imagines that a variety of approaches aimed at stimulating neurogenesis—or that can mimic the effects of neurogenesis on brain circuitry—will ultimately play key roles in combating Alzheimer’s disease as well as milder forms of memory impairment. He and others see potential in metformin, a drug already approved by the U.S. Food and Drug Administration, which has been shown to increase neurogenesis in mice. Some antidepressants also boost the formation of new neurons. And aerobic exercise has a well-documented role in boosting neurogenesis. “Running is the best natural intervention,” Sahay says.

These approaches would complement existing efforts in Alzheimer’s research that focus on amyloid plaques and neurofibrillary tangles, offering opportunities for earlier interventions. Hyperactivity in the CA3 region, left unchecked, may contribute to the accumulation of beta-amyloid and tau proteins and the subsequent death of neurons. “The way to think about the problem,” Sahay says, “is disrupting the feedback loop.” In 2018, Sahay’s lab received an Alzheimer’s Association research grant to explore whether reducing abLIM3 might also help limit beta-amyloid accumulation and improve memory in people with Alzheimer’s.

Advances in the basic understanding of what memory is and how it functions “will have more impact on the species than anything else in science in the past 30 to 40 years,” suggests UCLA’s Alcino Silva. “It’s a million-part puzzle. But we have figured out an important corner of that puzzle.”