Encoding, Storing, Retrieving: How Memory Works

Lauren Aguirre on Recovering Past Experiences and Forming Fake Ones

Memory has three stages: encoding, storing, and retrieving. If any one of them fails, an episode from three days or three decades ago will be missing from your personal story. In an obscure natural experiment that began on an operating table three quarters of a century ago, hints emerged that vanished memories could be found. Inside a turreted gray limestone building on a hilltop in Montreal, Dr. Wilder Graves Penfield gently lowers an electrode onto the glistening surface of the patient’s right temporal lobe. She is a 32-year-old woman with severe epilepsy. Suddenly, she speaks.

“I hear singing.”

Penfield lifts the electrode briefly, then lowers it again. “Yes, it is ‘White Christmas,’” she says.

Penfield waits for his assistant to place a small white paper square with the number 15 to mark the spot before moving the tip of the electrode a quarter of an inch or so away.

“That is different, a voice—talking—a man.” Penfield lifts the electrode, and the assistant marks the spot with a paper square numbered 16. Penfield discovers eight locations that each evoke different auditory memories—a radio program, a play, a violin. Twenty-six minutes after the patient hears the song “White Christmas,” Penfield returns to location 15 and lowers the electrode again. Again, she tells him the orchestra is playing “White Christmas.”

Penfield’s plan was to find the source of this woman’s seizures so he could remove the diseased brain tissue. He wasn’t looking for memories—but he appeared to have found them. Between the 1930s and the early 1960s, he documented 40 examples of what seemed like fragments of memories activated by brain stimulation. A 26-year-old woman says, “I hear voices. It’s late at night, around the carnival somewhere, some sort of traveling circus. I just saw lots of big wagons they use to haul animals in.”

“I hear my mother singing,” says another young woman.

“My mother is telling my brother he has got his coat on backward,” says a 12-year-old boy. A 21-year-old man who had recently traveled to Montreal from his home in South Africa describes hearing his cousins Bessie and Ann Wheliaw laughing, probably at some joke. Penfield wrote that “it was at least as clear to him as it would have been had he closed his eyes and ears 30 seconds after the event and rehearsed the whole scene ‘from memory.’ Sight and sound and personal interpretation—all were re-created for him by the electrode.”

The temporal lobe, home to the hippocampus, appeared to contain what Penfield later called keys of access to these memories.

Penfield used a stenographer in the operating theater to document what the patients described. He and other surgeons would often ask clarifying questions. Some patients said they experienced the sights or sounds not so much as if they remembered a long-ago event but as if the event was happening in the present moment. And yet they were also aware that they were lying on an operating table, under bright lights, their brains exposed, sharing what they felt. Penfield was struck by how vivid these recollections were, which he took as evidence that the memories were real and unchanged. “It is a hearing-again and seeing-again—a living through moments of past time.” Penfield called these phenomena recordings.

In the early 1950s, just a few years before a surgeon removed patient H.M.’s hippocampus, Penfield gave several lectures about the storehouse of memories in the brain. The temporal lobe, home to the hippocampus, appeared to contain what Penfield later called keys of access to these memories. The fact that the content of these experiential phenomena seemed trivial—standing on a street corner, hearing a mother call her child—supported Penfield’s belief that the brain permanently stored every experience, no matter how inconsequential.

His interpretation may not be correct, but what he found in the operating room was no fluke. In the decades since, neurosurgeons from many countries have described what appeared to be fragments of past events elicited by brain stimulation. Some date back decades. A 34-year-old man hears the theme song from The Flintstones, a TV cartoon he’d last seen at the age of 15. Another hears “Wish You Were Here” by Pink Floyd. A woman smells burnt wood, which reminds her of an evening in Brittany sitting around a campfire when she was a teenager.

Penfield stumbled upon traces of memories, yet his accidental discovery could have been predicted by an evolutionary biologist named Richard Semon. Semon’s ideas were shockingly before their time and are only now proving prescient. Back then, Semon’s peers were focused mostly on how memories are encoded and stored. He was just as interested in the process of recall. Unlike Penfield, who believed that memories are permanently and immutably stored as if on a strip of film, Semon suspected that a memory changes every time it’s replayed. He also thought that there might be multiple versions, or traces, of the original experience. Each one could be stored in slightly different connection patterns distributed across the brain, and these versions could interact with each other. He named these traces the “engram.” Semon made no bones about the fact that he could only theorize about the nature of engrams, leaving it to future scientists and better tools to figure out if he was correct. He was also a century ahead of his time, and his ideas were ignored. At the age of 59, shortly after the end of World War I and his wife’s death, Semon draped himself in a German flag, lay down on his bed, and shot himself in the heart.

An American zoologist named Karl Lashley was next to take up the search. He would devote more than thirty fruitless years to the hunt. Lashley had a practical, evidence-based method for finding the engram. He trained rats to navigate a maze and then systematically went about slicing through snippets of tissue in different parts of the brain. The rats’ memory declined in proportion to how much tissue he damaged with his scalpel, but where he sliced at it didn’t seem to matter much. Just as Semon had imagined, Lashley’s evidence suggested that any given memory is distributed across many brain areas. But there was a problem. No matter how many sections of brain Lashley damaged, he could never fully abolish a memory. Memory seemed to be at once everywhere and nowhere. At the end of his career, he quipped, “I sometimes feel, in reviewing the evidence of the localization of the memory trace, that the necessary conclusion is that learning is just not possible.” Lashley abandoned the search for the engram just as Penfield appeared to have brought one to life by electrically stimulating the temporal lobe, the home of the hippocampus.

“What Penfield did was proof of principle that this is not crazy. This can actually happen,” says Steve Ramirez at Boston University’s Center for Memory and Brain. He has decorated his light-filled office with a large blow-up T. Rex dinosaur, and his Twitter handle is @okaysteve, but despite his fun-loving, low-key attitude, Ramirez has the same lofty goal pursued by Semon and Lashley. “Can we go in and create maps for memories in the brain? Like, what does a memory even look like? Memory has sights and sounds and smells and emotions associated with it. Can we find those elements and say this is what the totality of one particular memory physically looks like?”

About a decade ago, inspired in part by a painful breakup with his girlfriend, Ramirez and his collaborator at MIT, Xu Liu, set out to answer this question by trying to find a fear memory in the hippocampus of a mouse. They seized on a newly invented tool called optogenetics, which lets scientists turn neurons on by shining light through a fiberoptic wire into the brain of a genetically engineered mouse. But which neurons? Out of millions of cells in the hippocampus, Ramirez and Liu only wanted to see the ones involved in making a specific memory. In a feat of precision engineering, they adapted the tool so the light-activated switch only worked in neurons that had recently fired—in other words, the neurons involved in learning something new. In this experiment, they put the mouse in a box with an electrified floor, where it quickly learned to be afraid—and to freeze.

To see if they could artificially reactivate that memory, the scientists put the mouse in a different box without an electrified floor, sent light down an optical fiber into the hippocampus, and turned on the same collection of neurons that had fired in the first box. The mouse froze in fear even though there was no reason to freeze. This was a new environment; a space the animal should have assumed was safe. But by activating the tagged neurons, Ramirez and Liu had found an engram—or if not the totality of it, at least a sufficient number of neurons to turn it back on.

As if that hadn’t been enough, they immediately took on what seemed like the next logical challenge; over three long days in the short life of a mouse, could they edit the memory of being in a safe space and turn it into a fear memory? Ramirez and Liu dubbed this experiment Project Inception. They put the same type of genetically engineered mouse in a small box that smelled of overripe fruit in a dimly lit room under warm red lights. The mouse had 12 minutes to roam about, learning that there was nothing to fear in this box. At the same time, Ramirez labeled only those neurons that were firing. He placed the same mouse into a different environment on day two, a well-lit, almond-scented box. The mouse shouldn’t have mistaken it for the first box. But while it roamed around, Ramirez used light to reactivate the original memory. At the same time, he gave the mouse a gentle foot shock.

On day three, he returned the mouse to box number one, the safe space. The familiar smell of overripe fruit and the warm red lights should have cued the mouse’s brain to turn on the safe memory. But instead, the mouse instantly froze. It appeared that the scientists had edited the foot shock memory from box number two into the first memory. Ramirez and Liu proved what neuroscientists, psychologists, and anyone who’s disagreed with a friend about what actually happened have long believed to be true. Memories change. They’re updated—or reconsolidated—with new information that can be true or false. In 2012 and 2013, Ramirez and Liu announced the results of these experiments. Tragically, Liu passed away a few years later, but Ramirez continues his hunt for the engram.

So if a mouse’s brain can’t tell the difference between an edited engram and a real one, what about one created out of whole cloth? A few years ago, a team at the Hospital for Sick Children in Toronto led by Paul Frankland and Sheena Josselyn implanted an engram for a completely fake memory that linked the scent of oranges to either a rewarding or an aversive sensation. Mice have little bundles of neurons in the olfactory region that each code for specific smells. One of those smells is orange, an odor that the average mouse doesn’t seem to care about one way or the other. In this experiment, scientists optogenetically stimulated that bundle of neurons simultaneously with another brain region that processes the aversive aspect of an experience.

Afterward, when they put the mouse into a cage with a piece of filter paper infused with the orange scent, the mouse stayed away—just as it would have if the scientists had paired orange with a foot shock in the real world. Frankland and Josselyn’s team tried it the other way around too, stimulating the orange-scent-detecting bundle of neurons at the same time as another brain region that processes a rewarding experience. When this mouse went into the cage, it approached the orange-scented paper as if there had been something good about it. Both memories were fake, created from scratch just by stimulating different brain regions at the same time. But from the mouse’s perspective—or at least judging from its behavior—this engram was indistinguishable from reality.

“I’m really interested in understanding what’s real, in a very, very deep way,” says André Fenton from New York University. “I follow the line of reasoning that I am a biological object, and so for me to have experiences that I think are mine, I need to figure out the biology that gives rise to my experiences.” If memories are contained in the collections of neurons, Fenton studies the molecules that hold those collections together. Or rather one particular molecule, a protein called PKMzeta, which is concentrated in the synapses between neurons. There’s just one problem. Proteins only last about a week in the body before they get degraded and recycled.

“So how can you build something permanent—or permanent-ish—with something that’s very transient?” Fenton asks. As he and his collaborator, Todd Sacktor, discovered, PKMzeta has an unusual property that gets around that problem; once made, it can be replenished on site, generating a rotating cast of PKMzeta proteins that glue a particular synapse together indefinitely. There’s at least one other leading candidate for a protein that holds synapses together, and likely many other molecular players that help turn a single experience into a lasting memory. But Fenton and his collaborators have shown that just by blocking the activity of PKMzeta, they can unglue synapses and remove a memory.

Ramirez’s former professor at MIT, Susumu Tonegawa, looks at the question of what makes memories fail from the same perspective as Richard Semon—it could just as easily have to do with faulty retrieval as faulty encoding and storage. In one test of this idea, Tonegawa’s team discovered that even if they blocked the formation of proteins like PKMzeta that strengthen the connections between synapses, they could still artificially reactivate memories optogenetically. This also suggests that an even more transient, or “permanent-ish,” event takes place before proteins can be made. A recent experiment in Li-Huei Tsai’s lab revealed that the very first thing to happen after neurons are activated is that their DNA is primed—open for business—and can later begin to churn out proteins if the neurons are reactivated again during a very specific time window.

Tonegawa later found evidence that two copies of the memory engram are created at once—one in the hippocampus and the second in other brain areas that store all the individual fragments. This experiment made headlines because it ran counter to one long-held theory that memories are laid down in the hippocampus and then gradually transferred and strengthened elsewhere for long-term storage. Protein synthesis that strengthens connections made during the first few minutes after an experience, Tonegawa suggested, could be what makes it possible to retrieve a memory with natural cues, like a few notes from a familiar melody or the smell of burning wood. But that doesn’t mean that that original engram has disappeared. It may be silent, capable of being reawakened only with a more powerful cue, like the stimulating tip of an electrode or a beam of colored light traveling down a fiberoptic wire into the hippocampus.

Neuroscientist György Buzsáki at the New York University School of Medicine prefers to steer clear of human constructs like engrams. He thinks they can get in the way of seeing things from the brain’s perspective. “I don’t know what an engram is, and I don’t even try to understand it,” he says. “Everything is a relationship between the brain and what happens outside.” In his view, the brain can’t possibly be a blank slate waiting for experiences to etch new pictures onto it. Instead, it comes equipped with a huge reserve of built-in patterns, each one created by a connected group of neurons. Memory formation is a game of matching those patterns with meaningful experiences so as to better predict the future and the consequences of its own actions.

This dynamic system is enormously flexible because neurons can be swapped in and out of any group and still generate very similar patterns. Buzsáki argues that you could get rid of every neuron involved in recognizing a specific person and still retain that memory. “You are welcome to erase all my Jennifer Aniston cells in my brain with a magic laser, and I guarantee you that my memory of Jennifer Aniston will stay. Because the runner-ups will occupy the space right away.” Experiments by Ramirez and others support this dynamic view of memory, showing that the overlap between the neurons involved in the making of a memory and the neurons involved in reactivating it can be very small.

New frameworks for thinking about memory and new tools—like holographic optogenetics that can “see” in three dimensions, or noninvasive wireless optogenetics—may bring neuroscientists closer to understanding the fundamental underpinnings of this essential human faculty. But at the very least, such basic neuroscience research in rodents offers some distant vision of how to help people with memory disorders. Ramirez, for one, hopes there will be ways to translate memory-manipulation insights into therapies for people with mental illnesses like depression, anxiety, or PTSD. “More sensible and practical scientists work on how to do good things with this knowledge,” says André Fenton. “I work on things like PKMzeta and mechanisms with the assumption that when we understand that sufficiently, we will figure out how to apply it usefully.”


The Memory Thief, Lauren Aguirre

Excerpted from The Memory Thief by Lauren Aguirre. Reprinted with permission of the publisher, Pegasus Books. Copyright © 2021. 

Lauren Aguirre
Lauren Aguirre
Lauren Aguirre is an award-winning science journalist with experience in multiple formats; documentaries, podcasts, short-form video series, interactive games, and blogs. She built her career at the PBS series NOVA after graduating from M.I.T. Aguirre’s reporting on memory has appeared in The Atlantic, Undark Magazine, and the Boston Globe’s STAT. The Memory Thief is her first book.

More Story
How Mean Girls Taught Me to Fear Lesbians and Love Pink "Going into freshman year, I half expected high school to be like Mean Girls,” I told a church full of my classmates and their...