Marilyn Monroe and Jane Russell appeared outside Grauman’s Chinese Theatre to write their names and leave imprints of their hands and high heels in the wet concrete. Down on their knees, supported by a velvet-covered pillow for their elbows, they wrote “Gentlemen Prefer Blondes” in looping script, followed by their signatures and the date, 6-26-53. But how did those watching the events of that day manage to imprint a memory trace of it, etching the details with neurons and synapses in the soft cement of the brain? Where and how are those memories written, and what is the molecular alphabet that spells out the rich recollections of color, smell, and sound?
After more than a century of searching, an answer was recently found, strangely enough, just eight miles from Grauman’s. Although not located on any tourist map, the scene of the discovery can be reached easily from Hollywood Boulevard by heading west on Sunset to the campus of UCLA. There, amid one of the densest clusters of neuroscience research facilities in the world, stands the Gonda (Goldschmied) Neuroscience and Genetics Research Center. And sitting at a table in the building’s first-floor restaurant, the Café Synapse, is the neuroscientist who has come closer than anyone ever thought possible to finding the place where memories are written in the brain.
That spot, the physical substrate of a particular memory, has long been known in brain research as an engram. Decades of scientific dogma asserted that engrams exist only in vast webs of connections, not in a particular place but in distributed neural networks running widely through the brain. Yet a series of pioneering studies have demonstrated that it is possible to lure specific memories into particular neurons, at least in mice. If those neurons are killed or temporarily inactivated, the memories vanish. If the neurons are reactivated, the memories return. These same studies have also begun to explain how and why the brain allocates each memory to a particular group of cells and how it links them together and organizes them—the physical means by which the scent of a madeleine, the legendary confection that sparked Marcel Proust’s memory stream, leads to remembrance of things past.
“It’s amazing,” says neurobiologist Alcino Silva, codirector of the UCLA Integrative Center for Learning and Memory. “For the last hundred years, scientists have been looking for the engram in the brain. We have now gotten to the point that we know enough about memory and how memories are formed that we can actually find the engram, and by finding it, we can manipulate it.”
The work recalls the movie Eternal Sunshine of the Spotless Mind. But unlike that film, in which the characters played by Jim Carrey and Kate Winslet have their memories of each other fully erased, Silva and other researchers involved in the manipulation of engrams do not believe that they can wipe clean every bit of a memory, only certain parts. Targeting the amygdala, the almond-shaped region in all mammals where fear memories are stored, Silva and his colleagues have shown that they can largely eliminate a mouse’s fearful response to a tone the animal had previously learned to associate with an unpleasant electric stimulus. Whether the mouse still remembers having heard the sound is uncertain, but the animal demonstrably no longer remembers the lesson it learned—that this particular tone was a prelude to getting zapped.
The implications of that finding hold promise for the treatment of human memory disorders. On the one hand, it points the way toward the selective targeting of neurons that hold memories of events so traumatic that people are disabled by them. That violent attack that you cannot get over? Deactivate those neurons in the amygdala that are linked to it, and you might still remember the attack but be freed from the unbearable pall of fear. With 3.5 percent of U.S. adults estimated to be suffering from post-traumatic stress disorder (PTSD) over the course of a given year, an effective new treatment would mark a mental-health milestone.
While PTSD sufferers remember too well, those with Alzheimer’s disease and other forms of dementia suffer the opposite problem. And just as Silva and others studying engrams have demonstrated the ability to delete memories, they have also shown they can strengthen them. This past July Silva’s colleague Sheena Josselyn, a neurobiologist at the University of Toronto’s Hospital for Sick Children, reported that her lab improved memory in mice bred to have the equivalent of Alzheimer’s. Using the same tools effective in creating and then purging fear, she boosted an entire brain region, the hippocampus, known to be critical for forming long-term memories.
Offering up what he concedes is a “science fiction kind of idea,” Silva wonders if physicians treating patients with Alzheimer’s “could direct memories to those regions of the brain that remain strong. Especially in neurodegenerative disorders, you have parts of the brain that are healthy and others that are not. If we find strategies to funnel memories to those parts that are still intact, we may be able to extend function longer.”
By upending the long-standing notion of what a memory is, Silva and Josselyn have wrangled a new view of memory from the realm of science fiction, where the individual engram has long been relegated by mainstream scientists as the neurological equivalent of unicorns and UFOs.
The word engram was coined in 1904 by the German evolutionary biologist Richard Semon in a book entitled The Mneme. The great puzzle of memory’s physical embodiment in the brain, Semon asserted, is that whereas an ordinary object reacts to a force only while that force acts upon it, the nervous system is somehow permanently altered and able to show a unique reaction to a long-ago stimulation years later. Consider, for instance, a baby who grabs a dog by the tail and gets bitten: The bite marks heal, but the painful lesson about not grabbing doggy’s tail remains. The capacity for such learning constituted what Semon called the mneme, borrowing the name of the Greek goddess of memory. The practical result of that capacity is what he called the engram, “this permanent change wrought by a stimulus.” If the mneme can be thought of as Semon’s conception of a biological hard drive, the engram is the byte (or, in that baby’s case, bite) of information written onto it.
But finding the engram proved much harder than naming it. In 1950 the eminent neurologist Karl Lashley described his exhaustive efforts in a famous paper entitled “In Search of the Engram.” In hundreds of experiments conducted over the course of 30 years, Lashley trained laboratory animals in routine memory tasks, such as opening a latched box, and then carefully made small surgical lesions in their brains to see the effects. No single cut in the brain, he found, ever obliterated a rat’s recollection of a maze path. Even two or three cuts made little difference. In fact, he could sometimes remove entire portions of the brain, and the animal would still remember the association between, say, a black floor tile and an electric shock. The only consistent effect he could find was that the animals’ overall capacity for memory fell in proportion to the number of neurons destroyed: The more brain cells lost, the more memories—all memories—weakened.
“All of the cells of the brain are constantly active and are participating, by a sort of algebraic summation, in every activity. There are no special cells reserved for special memories,” Lashley wrote. “Limited regions may be essential for learning or retention of a particular activity, but within such regions the parts are functionally equivalent. The engram is represented throughout the region.”
Lashley’s view that memories are encoded everywhere in general but nowhere in particular reigned supreme for more than 30 years. And then it came undone in the blink of an eye—actually the blink of a rabbit’s eye. Richard Thompson, now professor emeritus of psychology, biological sciences, and neuroscience at the University of Southern California, trained rabbits in what is called eyeblink conditioning, in which the sound of a musical tone is paired with a puff of air to the eye. (Pavlovian Conditioning 101: The puff of air is the “unconditioned” stimulus, because it requires no experimental conditions to produce a behavioral response, the blink. The tone is the “conditioned” stimulus, because only when it is paired with the puff of air will the animal learn to associate the two, producing what then becomes known as the “conditioned response”: a reflexive blink in response to the tone alone that has been produced through the conditions of the experiment.)
In a landmark paper published in Science in 1984, Thompson demonstrated that after he trained rabbits and then surgically removed just a few hundred neurons from the interpositus nucleus (a section of the cerebellum, located near the brain’s base), the animals no longer blinked in response to the tone.
The meaning of Thompson’s finding was clear: He had found an engram encoding the association between the puff of air, the tone, and the eyeblink, showing for the first time that the destruction of one particular set of neurons could wipe out one particular memory. “The whole point,” Thompson told me, “is that the memory is localized. The eyeblink conditioning is stored in a small number of cells in a particular region of the cerebellum.”
But Thompson was left wondering. Was the phenomenon in the interpositus nucleus just an exception to Lashley’s rule? And what changes were going on in those cells to permit them to act as an engram? For that matter, why were those particular cells, and not some others nearby, drafted into storage duty? Answering those questions would have to await a younger generation of neuroscientists.
My wife says I have a selective memory—I only remember good things.” Alcino Silva sits laughing, something he does frequently, over lunch at the Café Synapse, where madeleines are always served. Alcino, who is Portuguese by birth and spent his childhood in Angola, looks like the Hispanic half sibling of the actor Bill Murray, exuding a humor and ebullience that seem breezily at odds with the lofty work he conducts on the floors above. In his recollection, even the hunting trip his father took him on in the jungles of Angola, where they came under attack by anticolonialist rebels, was a good time.
“I was about 8,” he says. “Rebels using whistle calls were surrounding us. My father wrapped me and my sister in heavy blankets in preparation for tossing us out of the moving vehicle, so we wouldn’t be captured if the rebels got to the jeep. Then by chance we came upon a camp of Portuguese army regulars, and the rebels disappeared. But I don’t remember that as something horrible. To me it was a grand adventure. I have only the best memories of Angola.” After moving back to Portugal to flee the revolution, Silva decided at age 16 that he wanted to attend college in the United States because, he says with another laugh, “there was no better place for science—and it was the farthest place I could go.”
Still the keeper of cherished memories, Silva ended up as a postdoc under Nobel laureate Susumu Tonegawa at MIT in 1989. There he studied the synapse, the gap between neurons, which are considered the cellular site of memory storage. Indeed, because the overwhelming preponderance of neurons in the adult brain have long since lost their capacity to divide and proliferate, memory can be etched into the neural network only if existing nerve cells expand, growing more branches connected to other nerve cells through ever more synapses. The shifting sea of synapses functions like a kind of computer code, allowing information to be stored within its spaces in different ways. Silva decided to study memory, seeking the cellular mechanisms at its root, because he couldn’t think of anything more important. “Memory is not all that we are, but almost,” he says. “We are the entire set of memories that we acquire. Every one of our memories changes who we are.”
By the time Silva took his first staff position at Cold Spring Harbor Laboratory in New York in 1992, Eric Kandel, the Columbia University neuroscientist who would win a Nobel for his work in this field, was already on the trail. To trace the molecular basis of memory, Kandel was using the sea slug Aplysia, a neurologically simple organism that contains only perhaps 20,000 neurons, many of them quite large. (The human brain contains an estimated 100 billion neurons.) Kandel’s elegant strategy was to trace the molecular cascades generated when the sea slug was shocked. Depending on the nature of the shock, its gill would move in a certain way, and a memory—short- or long-term—would form. A single, brief pulse of stimulation to a sea slug, for instance, caused its gill to withdraw for minutes. That brief reflex was due, invariably, to the release of the neurotransmitter serotonin from the stimulated neuron. The serotonin simply crossed the synapse and temporarily modified proteins in the membrane of the target cell, leading to production of messenger chemicals such as CAMP (cyclic adenosine monophosphate).
Five pulses of the same stimulation led to a long-term response in which the sea slug’s gill withdrew for weeks. Here, the cell produced a protein called protein kinase A in such abundance that it provoked production of another, similar molecule called mitogen-activated protein kinase, or MAPK. Together, these two molecules traveled to the nucleus of the cell and activated a third biomolecule, CAMP Response Element-Binding protein, or CREB. Kandel found that CREB works by binding to DNA in the nucleus of the cell and then switching on genes that produce a host of proteins involved in memory, even when no further stimulation occurred. Long-term memory, reflected in the extended flexion of the gill and the neural architecture of the sea slug, was the end result of the whole chemical parade. By 1991 CREB had also been shown to be essential for forming long-term memories in the fruit fly Drosophila.
“CREB is a master conductor regulating the expression of many genes inside cells,” Silva explains. “When CREB is activated, it turns on the manufacture of a lot of other proteins that are needed for all sorts of things, one of which is memory.”
In 1994 Silva took the next step, showing for the first time that CREB is also essential for mammals—in this case, mice—to maintain long-term memories. Mice bred to have a malfunctioning form of CREB, Silva’s study demonstrated, did not remember from one day to the next that they had already been taught to navigate a water maze or that they had previously been placed in a special cage where they received a mild electric shock. Without a functioning version of CREB, they might as well have been born yesterday.
“CREB changes how easily neurons fire,” Silva says. “When a memory is ready to be stored, the more CREB a neuron has, the more likely it is the neuron will go into long-term memory mode. It makes the neurons more alert.”
Once Silva had demonstrated CREB’s role in mammals, hundreds of other researchers around the world quickly joined the effort to decode the memory process. One of those was a young postdoctoral research fellow named Sheena Josselyn, a petite, red-haired native of Ohio with a passion for pop culture icons like Jerry Seinfeld and Tom Cruise. (A recent paper of hers in the Journal of Psychiatry and Neuroscience referenced both a best-selling song by Shakira, “Hips Don’t Lie,” and the title of the Steven Seagal film Hard to Kill.)
Working in the late 1990s in the laboratory of Yale neuroscientist Michael Davis, Josselyn decided to see what would happen if, instead of taking CREB away to weaken an animal’s memory, she added extra CREB to strengthen it.
The challenge was sneaking the CREB in. Luckily for Josselyn, another postdoc who worked across the hallway, Bill Carlezon (now conducting neuroscience research at Harvard University), had developed a neat strategy for doping neurons with extra CREB. His trick was to insert the gene coding for CREB into a version of the herpesvirus that can infect neurons without spreading to nearby cells. The virus, in other words, served as the perfect Trojan horse to slip the CREB through the cellular gates.
Borrowing Carlezon’s technique, Josselyn injected viruses loaded with the CREB gene directly into the lateral amygdala of laboratory rats’ brains—the region of the amygdala that works as “fear central” in all mammals. Once the rats had recovered from the procedure, Josselyn used standard Pavlovian conditioning to train both the treated rats and an untreated group to associate a light with a mildly unpleasant foot shock. Once trained, they would display an exaggerated startle response, jumping high into the air whenever they saw the light. To dilute the effect Josselyn used a so-called weak-training technique, in which training sessions are bunched together without an intervening rest period, an approach previously shown to produce only a weak memory. After the weak training, the CREB-enhanced rats jumped five times higher than normal rats after seeing the light.
Josselyn realized then that CREB is not only necessary for normal long-term memory; it can also act as a memory booster.
But there was a vexing problem with Josselyn’s data. Only 15 percent of the neurons in the lateral amygdala, the section she had targeted, had been infected with the herpesvirus that delivered the extra CREB. How could such a small number of neurons produce such a dramatic increase in memory? A peer reviewer of her paper found the discordance so glaring that he questioned whether the results were reliable and urged the journal to which she had submitted the paper to reject it.
Not long after that, Alcino Silva, who was about to move his lab from Cold Spring Harbor to UCLA, stopped by for a casual visit with Josselyn’s boss. Such drop-ins are a practice he’s made a habit of to further collaboration with leading colleagues in the field. “Neuroscience today is all about connections,” Silva says.
During that visit, the three sat down to see if they could figure out the discrepancy in the data. The “problem,” Silva felt, might in fact be an opportunity: a hint of how they could use CREB as a tool not merely to enhance or suppress memories but to explore each new memory’s precise location—to locate the engram. Maybe after all these years, it would be possible to find true tracks of memory in the brain. Perhaps it was actually necessary for only a small percentage of neurons to be involved in forming a memory. Maybe memory formation is a kind of competitive sport. CREB could play an essential role in recruiting the neurons lucky enough to underlie the memories we form.
By the time Josselyn’s study was published, in 2001, she had already accepted an invitation extended to her and her husband, Paul Frankland, also a postdoc in neuroscience, to join Silva at UCLA.
Once there, Josselyn dreamed up an ingeniously elaborate study with Silva to test their “neuronal competition” theory of memory formation. First they decided to use a mutant mouse engineered to have very low levels of CREB, so that any behavioral effects of injecting extra CREB into the amygdala would jump out. Then Josselyn and Silva devised a twist on the herpesvirus technique Josselyn had used at Yale. Rather than just adding the gene for CREB to the virus, this time they also inserted the gene for green fluorescent protein. That way, any neuron that had extra CREB by virtue of getting infected with the virus would also conveniently light up under the microscope.
But to know for certain if the neurons with the extra CREB—and only the neurons with the extra CREB—were the ones that stored the memory engram, the researchers had another trick up their sleeve: a means to show which neurons had been active in recollecting a memory.
That trick was a tough one to pull off. First Josselyn and Silva placed the mice inside a cage where they were shocked each time they heard a tone. Demonstrating the expected conditioned response, the mice would then freeze in fear whenever they heard the tone—even though shocks were now withheld. Next the researchers sacrificed the animals within five minutes of the time they heard the tone and froze in fear. Why five minutes? Because as soon as a memory forms, bits of RNA that produce a protein called Arc (activity-regulated cytoskeleton-associated protein) are activated inside neurons for precisely five minutes. Detecting Arc inside a neuron, in other words, shows that a memory had just been activated inside it less than five minutes before the animal’s death. To detect the presence of Arc, the scientists used a fluorescent tagging technique, as they had with CREB.
Ok, so they’ve injected their mice’s amygdalas with extra CREB, taught them to recognize a tone that they heard when they got shocked, played the tone again to see that they froze, and then immediately sedated and sacrificed them. Next the scientists removed the mice’s brains, separated out the amygdalas, sliced them thinly, and placed them on glass slides.
Under the microscope, the resulting slices of amygdala lit up like a Christmas tree: green if they had absorbed the extra CREB, red if positive for Arc. While the red and green did not overlap precisely, Josselyn and Silva did find that the neurons fluorescing red (active with Arc) were 3 to 10 times as likely as their neighbors to glow green (positive for CREB) as well. Makes sense, right? When baseball players take steroids, they hit more home runs. CREB, essentially a memory stimulant, is like steroids for neurons.
But this is where things got seriously strange. Imagine if, when home-run kings Barry Bonds, Alex Rodriguez, and Sammy Sosa allegedly used steroids and began hitting better, their teammates who did not take steroids actually started playing worse. Incredibly, that is exactly what happened on the neuronal playing field: Neurons that did not glow green with CREB were about one-twelfth as likely as their CREB-rich counterparts to shine red with Arc, meaning they had not been involved in the memory engram at all.
Silva and Josselyn finally concluded that memory storage among neurons is a zero-sum game. CREB helps form memories not by making all neurons stronger but by turning up the contrast between the haves and the have-nots.
“It is not enough to succeed; others must fail,” says Josselyn, quoting Gore Vidal.
Josselyn happens to like Tom Cruise. “His movie Risky Business changed my life,” she says. Wearing sneakers and a faded Ohio State University T-shirt, Josselyn sits in her small office at the University of Toronto’s Hospital for Sick Children, located kitty-corner to her husband’s office. (Eat your heart out, Mr. Cruise.) In July 2003, they left UCLA together for Toronto, where they were each given their own laboratories and named assistant professors of neuroscience. Next to her office door now stands a “Festivus pole,” which she used in December for an office party to celebrate the fictional secular holiday depicted on Seinfeld.
When they arrived in Toronto, Josselyn says, she and Frankland collaborated on a study aimed at taking the next step in manipulating the CREB-enriched memory engram: killing it. If the neurons with extra CREB were truly essential to maintaining a memory, they reasoned, eliminating those neurons should eliminate the memory as well. But unlike the neurons in the interpositus nucleus that Thompson had surgically removed back in the 1980s, the 15 percent of CREB-enriched neurons didn’t mass together in a huddle that could be targeted with a scalpel. Instead, they were evenly dispersed around the amygdala. The challenge: how to kill them without destroying surrounding neurons.
To do so, Josselyn and Frankland looked for something they could add to the herpesvirus, in addition to CREB, that would eventually cause the infected neurons to die. Finally a collaborator of theirs, the neurobiologist Steven Kushner, learned about a new technique for selectively assassinating neurons using diphtheria toxin. Mice do not naturally carry the receptor that permits the toxin to enter neurons, but monkeys do. By adding the simian gene for the diphtheria receptor to the herpesvirus that would also carry the CREB gene, the scientists now had the means to kill only those neurons with extra CREB.
With diphtheria as their time bomb, the team once again trained mice in auditory fear conditioning. As before, animals that had received injections of the virus carrying the gene for CREB (as well as the toxin receptor) learned better, demonstrating far more fear than those without it. And as predicted, when the husband-wife team infused the mice with diphtheria, selectively killing only those neurons with both extra CREB and the toxin receptor, the mice no longer froze when they heard the tone. Josselyn and Frankland had caused selective amnesia—a historic finding that brought them one step closer to finding a treatment for PTSD.
A year and a half later, Josselyn described these and other findings at the enormous annual meeting of the Society for Neuroscience during a symposium she and Silva cochaired on “mechanisms of memory.” The standing-room-only crowd, comprising hundreds of neuroscientists, buzzed with excitement at the notion that they might finally have a better tool than talk therapy to treat the millions of veterans and others disabled by traumatic memories. A door had been opened: The fear engram had been found. It had been strengthened with CREB and obliterated with diphtheria—and a path now beckoned toward a practical treatment for people.
Not even Silva or Josselyn believe that human subjects will readily accept infusions of CREB loaded with diphtheria anytime soon. “It’s very interesting work but not practical for clinical treatment of patients,” says Denis Paré, a neuroscientist at Rutgers University in New Jersey who is also studying pharmacological approaches to manipulating fearful memories. “These are experimental manipulations that you couldn’t do in a human.”
Silva agrees but remains cautiously optimistic about the prospects for translating his and Josselyn’s research into useful treatments. “We can’t fix something that we don’t understand,” he says. “For the longest time we had no understanding of the physical representation of memory in the brain, of engrams. We now know that CREB has a role in determining where memories go. We can manipulate CREB to funnel memories into specific cells.”
A mouse’s memory of a single fearful event is one thing; the complex associations of human memory, powered by a dense network of neuronal connections, is quite another. “We’re studying a really simple, basic kind of memory,” Josselyn concedes. “With mice, you can’t ask them, ‘Do you remember that cage we put you in yesterday?’ All we can do is observe their behavior to see if they’ve learned something or not. More complex memories, like the recollection of an event that happened to you, are stored in many different areas of the brain. But even for treating PTSD, we wouldn’t want to take away the entire memory, just the part that leaves you disabled with fear. And that comes from the amygdala.”
So it turns out that Karl Lashley’s belief in memory as existing in a distributed network is still alive and well; Silva and Josselyn have not overturned it, only supplemented it, showing that some parts of some memories do exist in a discrete number of neurons.
As for the practical application of using CREB as a memory enhancer, Josselyn is exploring it—but warily. Back in the 1990s, when CREB research was first getting under way, a company called Memory Pharmaceuticals invested millions in an effort to develop a pill that would extend the action of CREB. But the company, along with others that had bet on so-called smart pills, hardly fared well.
“People thought we’d see a pill on the market by now,” Josselyn says. “Companies were set up and folded, trying to find this pill. The field was giddy in the initial stages. It turns out that memory is really tough to strengthen, especially in people.”
Yet cautious progress is being made. Perhaps Josselyn’s most thrilling finding so far, in research published this past July in Neuropsychopharmacology, involves mice with the equivalent of Alzheimer’s disease. When her group injected CREB into the animals’ hippocampus using engineered herpesvirus, the mice regained their ability to learn.
Could a similar process be used to deliver CREB into human brains? “It’s possible,” Josselyn says. “The principle is there, but we need much finer tools. The herpesvirus is probably not a good way to go in people. But we’re really not doing all this just to improve learning in mice. We want to figure out how people learn and remember and how we can help them when they don’t.”
For my middle-aged brain, a breakthrough cannot come too soon. At the conclusion of my meeting with Silva in Los Angeles, he offered to walk me to my car. But in an absurdly ironic twist, I couldn’t remember where I had parked it, on some back street in an underground lot. Determined to help, Silva accompanied me as we searched along Westwood Plaza, past the Semel Institute for Neuroscience and Human Behavior, past the Reed Neurological Research Center, past the Ahmanson Laboratory of Neurobiology—none of them doing me a damn speck of good. Any farther and we’d have ended up in front of Grauman’s, where the trace of Marilyn’s and Jane’s handprints were holding up a lot better than my memory.
And so, like two mice lost in a maze, we walked the streets, the journalist and one of the world’s leading memory researchers, as he tried to help me remember.
Dan Hurley is the author of Diabetes Rising: How a Rare Disease Became a Modern Pandemic and What to do About It. He lives in New Jersey.
