A mouse finds itself in a box it’s never seen before. The walls are striped on one side, dotted on the other. The orange-like odor of acetophenone wafts from one end of the box, the spiced smell of carvone from the other. The mouse remembers that the orange smell is associated with something good. Although it may not recall the exact nature of the reward, the mouse heads toward the scent.
Except this mouse has never smelled acetophenone in its life. Rather, the animal is responding to a false memory, implanted in its brain by neuroscientists at the Hospital for Sick Children in Toronto.
Sheena Josselyn, a coauthor on a 2019 Nature Neuroscience study reporting the results of the project, says the goal was not to confuse the rodent, but for the scientists to confirm their understanding of mouse memory. “If we really...
Sheena Josselyn, a coauthor on a 2019 Nature Neuroscience study reporting the results of the project, says the goal was not to confuse the rodent, but for the scientists to confirm their understanding of mouse memory. “If we really understand memory, we should be able to trick the brain into remembering something that never happened at all,” she explains. By simultaneously activating the neurons that sense acetophenone and those associated with reward, the researchers created the “memory” that the orange-y scent heralded good things.
Thanks to optogenetics, which uses a pulse of light to activate or deactivate neurons, Josselyn and other scientists are manipulating animal memories in all kinds of ways. Even before the Toronto team implanted false memories into mice, researchers were making rodents forget or recall an event with the flick of a molecular light switch. With every flash of light, they test their hypotheses about how these animals—and by extension, people—collect, store, and access past experiences. Scientists are also examining how memory formation and retrieval change with age, how those processes are altered in animal models of Alzheimer’s disease, and how accessing memories can influence an animal’s emotional state.
Life, in the real world, is an accumulation of an almost infinite number of memories across a lifetime.—Denise Cai, Icahn School of Medicine at Mount Sinai
“Almost every neurological disease or psychiatric disease—everything from autism to stress to PTSD to Alzheimer’s to epilepsy—they all affect the memory system,” says Denise Cai, a neuroscientist at the Icahn School of Medicine at Mount Sinai.
A little more than a decade ago, such memory manipulations might have seemed like science fiction—and in terms of applying them to people, they mostly still are. But in two watershed papers, published in 2009 and 2012, researchers blew open the doors to memory control in lab animals. In addition to optogenetic control over neuronal firing, transgenic techniques allowed scientists to prime or modify the specific cells that were activated when the animals first stored a new memory. That collection of neurons, called a memory trace, will fire again during recollection. In the first of these two seminal papers, Josselyn’s team showed they could control, ahead of time, which neurons would join a trace, and then kill those cells to eliminate the memory. Shortly thereafter, Susumu Tonegawa’s group at MIT presented techniques to identify and reactivate memory traces.
Since then, “it’s reached a fever pitch,” says Boston University neuroscientist Steve Ramirez, a Tonegawa lab alum and coauthor on the 2012 paper.
Forgetting and remembering
In Toronto, Josselyn is one-half of a memory-manipulating duo with her husband, Paul Frankland. In their 2009 Science study, the pair aimed to make mice forget a specific memory: that in a particular chamber, the sound of a tone preceded a foot shock. Fear conditioning is a common technique in the field because the memory forms quickly, within a few trials. Rodents that recall the experience freeze in fear upon hearing the tone again, while those that forget are more likely to carry on exploring their cage as normal.
Memory traces incorporate neurons throughout the brain, touching parts that process sights, sounds, smells, and emotions. For simplicity, researchers usually focus their studies on one brain area of interest. In this case, the Toronto team zeroed in on the amygdala, which processes emotions such as fear. The amygdala also takes part in storing memories of events and emotions.
Rather than wait to see which neurons would join the memory trace and identify those, the researchers primed certain neurons to join that trace. To do so, they took advantage of the fact that the production of the transcription factor CREB makes neurons excitable, and that excitability makes them more likely to take part in memory storage. Using a virus-delivered genetic construct, the researchers randomly overexpressed CREB in a subset of neurons in the amygdala. When the team trained the mice to link the tone with a shock, those high-CREB neurons were three times more likely to join the trace than unaltered cells.
The genetic construct Josselyn, Frankland, and their colleagues used also made the engineered neurons—many of which joined the memory trace—vulnerable to a toxin produced by the bacterium that causes diphtheria. Once the team injected the toxin to kill those trace neurons, the mice didn’t freeze in response to the tone. “The memory was gone,” says Josselyn.
That study showed that memory scientists could sit in the driver’s seat, bending traces and their associated memories to their will. Meanwhile, the MIT team set out to force mice to recollect a fearful experience whenever the scientists wanted them to. Doing so would prove that the cells in the trace they manipulated were indeed behind the memory.
Their work depended on the optogenetic tool channel-rhodopsin, a protein that spans cell membranes and responds to light. Blue light, delivered via an implanted optic fiber, causes the channel to open, letting in positive ions and making neurons fire. The researchers developed a system to express channelrhodopsin in mouse neurons involved in a particular memory trace in the dentate gyrus, a brain structure closely associated with the hippocampus. The hippocampus is involved in learning and memory as well as emotion and motivation; the dentate gyrus integrates sensory inputs during memory formation. Like Josselyn and Frankland, the MIT group used fear conditioning, training mice to associate a specific place and tone with a shock.
The team designed their transgenic mice so that memory traces would not be labeled if the animals’ diets included the antibiotic doxycycline. The researchers removed doxycycline from the mice’s meals for a short time to create a memory with channelrhodopsin expressed in the relevant trace cells, then reinstated the doxycycline treatment to avoid labeling any other memory traces. When they later shone blue light into the brain to activate the target trace while the mice were in a different place, the animals would recall the foot-shock situation and freeze. The team concluded in their 2012 Nature study that those cells truly represented the memory in the brain. (See illustration below.)
Those early days of memory manipulation were “incredibly exciting,” recalls neuroscientist Tomás Ryan, who trained at MIT with Tonegawa during that time before starting his own lab at Trinity College Dublin. The new techniques “entirely changed how we can do things.”
Since then, many researchers have adopted Tonegawa’s system, or created variations, to ask their own questions about memory. For example, rather than force a recollection, a parallel technique can make a mouse forget. Christine Ann Denny, a neuroscientist at Columbia University Irving Medical Center, and colleagues bred mice to produce archaerhodopsin, a protein that pumps protons out of the cell in response to yellow light, silencing neurons. The researchers manipulated a group of hippocampal neurons that linked a lemon-scented chamber to a foot shock. When they silenced those neurons with yellow light, the mice forgot their fear.
Perhaps the ultimate test for scientists’ understanding of memory is to make one, from scratch, as Frankland, Josselyn, and colleagues did last year. To succeed, they needed to do two things: First, fake some cue—the neural equivalent to the real-life sensation, such as a tone, that mice are normally exposed to in conditioning experiments. And second, falsify the mouse’s associated expectations—the good or bad outcome that the animal would anticipate when it sensed that cue.
For the cue, the team chose smell because the neurons in the olfactory system are understood in detail. Olfactory neurons with a receptor called M72 are activated by orange-scented acetophenone. Using mice that produce channelrhodopsin in every M72 sensory neuron, the team could shine blue light in this part of the brain to trigger the sensation of a whiff of orange.
To set the animals’ expectations, the researchers tapped into one of two known pathways into the midbrain’s ventral tegmental area, which is involved in behavior reinforcement. One of the pathways is linked to reward, the other to aversion. By pairing the optogenetic stimulation of one of those pathways with optogenetic stimulation of M72, the team could link the scent cue to a good or bad “memory.”
Control mice didn’t particularly prefer one side or the other of the striped and dotted box, despite the different wallpaper and scents. But mice that had been optogenetically trained to associate M72 activation with a reward spent more time near the end smelling like oranges. If they were conditioned to link the orange smell with an unpleasant sensation, they avoided it. The mice showed no preference for or aversion to carvone, the control scent.
It was exactly as the team predicted, demonstrating that the researchers understood the rudiments of the underlying memory systems. When the scientists examined the neurons activated in the animals’ brains, there was significant overlap between the memory traces of mice with the artificial aversion memory and those of mice that had actually experienced a foot shock while smelling acetophenone, further validating the results.
“This shows that we are beginning to have a much deeper understanding of how memories are made,” says Josselyn, “so much so that we can mimic the process and create an artificial memory using only optogenetics.”
Methods of Memory Manipulation
As a memory forms, certain neurons are incorporated into a memory trace, a neural network associated with a particular experience that is active when the memory is recalled. By permanently altering those neurons in mice, researchers can control their activity. Neurons are engineered to produce channelrhodopsin (Chr), a light-sensitive ion channel, once they’re recruited into a specific memory trace. From memory formation onward, blue light can activate them, triggering the animal to act as if it is recalling the previous experience.
© lucy conklin
Scientists engineer mice such that neurons will produce channelrhodopsin once recruited into a memory trace. The mouse’s diet determines when the neurons are vulnerable to this effect.
As the mouse experiences a foot shock, delivered in a specific enclosure and accompanied by a tone, the neurons recruited to that memory trace are altered and begin to make channelrhodopsin.
Later, scientists can use blue light to activate the trace neurons, causing the cells to fire and the mouse to freeze in fear, as it learned to do when presented with the tone that heralded a foot shock.
Scientists engineer M72 olfactory receptor neurons, which sense orange-scented acetophenone, to respond to blue light.
They do the same with neurons that control aversion to unpleasant stimuli such as a foot shock.
© lucy conklin
Stimulation of both areas simultaneously results in the formation of a false memory, linking the acetophenone odor to unpleasantness.
© lucy conklin
A mouse that has never experienced the smell of acetophenone will avoid the orange-like odor.
Most neuroscientists in this field work in rodents and study episodic memories—memories of experiences that an animal has lived through. In contrast, University of Texas Southwestern Medical Center neuroscientist Todd Roberts has successfully implanted procedural memories, which encode how to do something, in the brains of birds. Young male songbirds must learn their father’s song in order to woo mates when they grow up. For zebra finches, Roberts says, just a few seconds of dad’s tune—a repetition of three to six unique elements, about 100 milliseconds each—is enough to seed the young bird’s memory. Male chicks then spend months practicing until their songs match the melodies they remember. “They will develop a perfect copy,” says Roberts.
In 2014, Roberts and then-graduate student Wenchan Zhao set out to do something much simpler than implant a memory: they wanted to disrupt the song-learning circuit in the bird’s nidopallium, a brain region that serves similar top-level functions to those of the mammalian cortex. They assumed that if they did so while the bird was listening to an adult’s song, the memory of the song would be scrambled. But a control experiment yielded unexpected results. Zhao used channelrhodopsin to stimulate the learning circuit of a young bird raised without a father figure, before the animal was transferred to the company of an adult male tutor. Zhao expected that when the baby interacted with the male tutor, it would learn that male’s song. It didn’t. “This bird, when it grew up, had a really weird song,” Roberts says. Zhao’s brief pulse of light had set the bird’s song memory, implanting an artificial sense of how the tune should sound. The bird then spent its youth striving to measure up to that fake memory.
Intrigued, Zhao experimented by exposing young, untutored birds with channelrhodopsin-carrying neurons in their learning circuit to the blue light for different periods of time, then raising them without tutors. If Zhao flashed the light for 50 milliseconds, the chicks grew up to sing songs with shorter-than-normal elements, producing a melody more like the quick trills of a canary. If she lit up the brain for 300 milliseconds, the sound elements were too long. “It sounds like they’re just yelling one pulse,” says Roberts. “It’s really quite bizarre.”
Good memories, bad memories
Many memories aren’t neutral, but are charged with emotion. Ramirez recalls enduring a breakup that took place over a large iced coffee at Crema Café in Harvard Square in 2012 when he was in graduate school. “In the immediate aftermath, going past Crema was a painful reminder . . . an emotional kick to the gut,” he recalls. The cafe became linked to the unpleasant memory, he says.
But with time, as Ramirez came to terms with the breakup and continued to frequent Harvard Square, that emotional tinge faded. He says he could visit the cafe for his favorite peanut butter–banana sandwich without distress by the time Crema closed last year, seven years later. His experience illustrates the theory behind exposure therapy for negative memories, which involves re-experiencing the real situation linked to trauma or anxiety, or an imagined version or virtual reality simulation. For example, therapists have used virtual reality scenes featuring jungles and helicopters to treat PTSD in Vietnam veterans. The hypothesis holds that repeated exposure to a memory can drain it of emotional power.
Ramirez wondered if he could do the same thing, optogenetically, in mice—if repeatedly activating the memory of something scary could diminish the associated freezing behavior.
His team focused their attention, and light beams, on the top of the dentate gyrus, where contextual information such as place and time is recorded. (See “Memories of Time” on page 32.) They trained mice to associate a particular chamber with a shock, and tagged the corresponding memory trace in the dentate gyrus with channelrhodopsin. Then they reactivated that trace with light for 10 minutes, twice daily for five days, forcing the mice to recall the experience while in a novel, shock-free zone. Mice that were returned to the original, shock-linked chamber were less likely to freeze than mice who had not been subject to memory reactivation. In the treated mice, fewer neurons from the original trace were active the second time in the chamber. “We think, in this case, it’s that particular fearful memory that we were able to turn the volume down on,” says Ramirez.
The researchers discovered that the location of the stimulation mattered. When they reactivated neurons from the same trace, but in the bottom of the dentate gyrus—associated with responses to stress and anxiety—they got the opposite results. Mice activated more of the neurons associated with the original fear memory when returned to the original enclosure, and the animals were more likely to freeze, as if the volume of their fearful memory had been turned up.
In other experiments, Ramirez has examined the power of positive memories to alleviate depression-like symptoms in mice. To create that positive memory, the researchers let the mice spend time with a mouse of the opposite sex, while labeling active trace neurons in the dentate gyrus with channelrhodopsin. Then, they stressed the mice by immobilizing them in a cone-shaped device to produce a depression-like state. When the animals were then lifted by their tails, these mice spent less time struggling than non-stressed mice, and they showed little preference for sugar water, normally a desirable treat. But when the team stimulated the recollection of the earlier romantic interlude, the mice immediately acted like they felt better, choosing sugar water over plain water and spending more time trying to escape when dangled.
“It seemed to be a very effective way of reversing these depression-related behaviors,” says Ramirez. In contrast, mice that experienced reactivation of a neutral or negative memory didn’t show improvement of symptoms.
People with depression have difficulty recalling positive experiences, Ramirez notes, but if there were some way to promote those recollections, it might help. He wonders: “Could we almost view memory as a drug?”
Memories lost and found
Everyone’s memories can naturally fade. Memory loss can also be pathological, as in the case of Alzheimer’s disease or amnesia. But when memories disappear, are they gone for good? Or has the brain merely lost access to the trace?
At Columbia, Denny and colleagues tested whether they could give mice better access to lost memories, jump-starting the recollection with optogenetics. The researchers crossed mice modeling Alzheimer’s disease (AD) with ones that would allow them to label a memory trace in the dentate gyrus with channelrhodopsin. The team let the animals age until they started to show deficits in memory tests at six months (the equivalent of age around 30 in human years), then activated channelrhodopsin in the memory trace as the mice learned to anticipate a shock in a particular chamber. Five days later, when the animals were returned to that same chamber, the researchers stimulated the channelrhodopsin-labeled trace cells with light. With their memories reactivated, six-month-old AD mice froze as often as non-AD animals, indicating that the memory was still there. The effects wore off within a day of stimulation, though, suggesting more stimulation would be necessary to produce ongoing memory improvements.
These crazy things we can do in the lab are really important to back up our understanding of what the brain is doing.—Sheena Josselyn, Hospital for Sick Children
Ryan and Tonegawa saw similar results in tests of mice with amnesia. With stimulation of a trace, “the memory comes back,” Ryan says. “Even severe kinds of memory loss can be because the memory is locked in your brain, not destroyed.” That matches the tendency for most people with amnesia to recover.
Could such faded memories be restored in humans? Denny thinks that somehow stimulating the dentate gyrus in people with Alzheimer’s might help with memory loss. Of course, that’s easier said than done. “We’re not going to be sticking optic fibers into the human brain anytime soon,” says Ramirez. Clinical applications will require different tools, such as medications or psychotherapy.
In some cases, it’s simpler to stimulate a memory in a person than a mouse. Psychotherapists can bring up a past experience in conversation, or show a patient a picture. Just recalling a memory makes it malleable, vulnerable to being overwritten with a different emotional load. In other words, “face your fears,” says Johannes Gräff, a neuroscientist at the École Polytechnique Fédérale de Lausanne in Switzerland. Researchers are experimenting in clinical trials with drugs such as ketamine and MDMA (dubbed “ecstasy” by recreational users) that may help people change the emotional charge of certain memories as they reflect upon those episodes.
But a person who has experienced trauma or forgetfulness in a complex natural environment is hardly the same as a cloistered lab mouse worried about a foot shock. “Life, in the real world, is an accumulation of an almost infinite number of memories across a lifetime,” says Cai. And complete memory traces are not limited to the few thousand cells that scientists can access in a mouse brain using an optic fiber.
As a result, researchers are moving toward more-realistic interrogations of memory. Denny and Ramirez are building whole-mouse-brain, 3-D memory maps. The pair and others are investigating multiple memories, their interactions, and how the system changes with age. Experiments of this variety will provide deeper insights into the neuroscience of memory, which might eventually support the clinical use of memory manipulation.
While direct manipulations of human memory traces are a long way off, many neuroscientists remain in awe of what’s been achieved in animals after just a decade of using optogenetics to delete memories or implant false ones. “These crazy things we can do in the lab are really important to back up our understanding of what the brain is doing,” says Josselyn. Plus, she admits, “doing the science-fiction type things is really fun.”
Amber Dance is a freelance science journalist living in the Los Angeles area. Read her work or reach out at AmberLDance.com.