Inside the Memory Machine

Sam Deadwyler and Robert Hampson had spent the 1980s and early 1990s studying how neurons behaved in the rat brain while they performed a simple memory task. Two levers sat side-by-side on one wall of the rat’s cage. After the rat pressed one of the levers, it had to run over to the other side of the cage and stick its nose into a tiny opening. Then, to get its reward, the rat had to go back to the levers and press the other lever. In all of this running back and forth, the rat had to remember both which lever it had pressed and which lever it still needed to press.

While the rats performed this task, Deadwyler and Hampson, both neuroscientists at Wake Forest Baptist Medical Center in North Carolina, recorded neural activity in the hippocampus, a seahorse-shaped structure deep in the brain that is the headquarters of learning and memory. They had designed a device containing 16 tiny electrodes and implanted it into the rat’s brain to record electrical activity. Up until that point, the best scientists could do was measure single neurons, but this device could record the activity of a whole group of neurons, giving Deadwyler and Hampson a much more detailed and accurate picture.

A rat hippocampus as seen through an ultra-widefield high-speed multiphoton laser microscope

They focused on two regions of the hippocampus: CA3, which showed the highest activity when the rat pressed the first lever, and CA1, which showed highest activity when the rat had to decide which lever to push at the end of the task. After hundreds of repetitions, Deadwyler and Hampson noticed that when the rats pressed the first lever, they found that a group of neurons in CA3 fired in a specific pattern. Then, when the rats had to decide which lever to press the second time, they also found that CA1 patterns fired in a specific pattern.

“This pattern was the code of the memory, and it was nearly identical from time to time to time—and the system worked from rat to rat to rat,” Hampson says. Based on the activity they observed, they could even tell when the rats were going to make an error. “The rats are not making mistakes randomly. They’re responding the way they are because the hippocampus encoded the wrong information,” he adds.

Humans make similar errors when they try to look for their parked cars, Deadwyler says. Imagine if yesterday, you parked your car in the second row near the door, but today, you parked in the fourth row near the back. But today you were also preoccupied by a 9 a.m. meeting, and so you weren’t paying close attention to where you parked. If you tried to find your car in the second row of the lot, you would be making the same mistake as the rats because your hippocampus didn’t encode this morning’s parking spot very well.

After they reported these results in the Proceedings of the National Academy of Sciences in 1996, they got a phone call from Ted Berger, a neuroscientist at the University of Southern California who was creating mathematical models of hippocampal activity. Berger wanted to test his models against data collected from a live organism. Deadwyler and Hampson, for their part, wanted to create mathematical representations of how the neurons were behaving. The trio decided to collaborate.

Soon, Berger had created a mathematical model that predicted later CA1 activity based on the initial activity recorded in the CA3 region of the hippocampus. The results weren’t perfect, but they were right an astonishing 90% of the time, according to Deadwyler and Hampson’s rat data.

Shortly after they started to work together, the three began asking each other hypothetical questions. Mostly, it was a way of thinking through potential experiments. But they couldn’t get one “what if” question out of their minds. What if they could use this device not just to record memories, but replace them?

Into the Mainstream

Today, we have advanced prosthetics that can replace limbs with devices of uncanny agility, but when it comes to traumatic brain injuries, scientists and physicians have few options. A memory prosthetic would change that. Deadwyler and Hampson believe it’s possible to create a device that will help individuals with brain injuries and memory loss from Alzheimer’s and other dementias improve their ability to learn and remember. Although they have spent the last decade testing it in rats and monkeys, they hope to test it in humans in the near future.

The idea wasn’t always popular. “It sounded too much like science fiction. People told us they wanted to believe that we could do this, but they just didn’t think we could,” Deadwyler says.

That’s certainly changed in recent years. Researchers at the University of Pennsylvania and UCLA are also working on a prosthetic memory device, which works by boosting our ability to store a memory rather than recall it later, as Deadwyler and Hapson’s device does. Regardless of which aspects of memory affected by the prosthetics, several researchers say it’s an idea whose time has come. DARPA, the Defense Advanced Research Projects Agency, agrees and has provided $37.5 million to fund the new Restoring Active Memory (RAM) Project.

But memory prosthetics have also caught the attention of ethicists. Helping those with brain injuries is a noble quest, they argue, but altering memory could fundamentally change who a person is. Who should be helped first? What kind of injuries would benefit most? And where do we draw the line?

Nonetheless, even if these prototypes never proceed beyond animal testing, they have yielded important insight into how we learn and how we recall the information that makes us who we are.

Decoding the Hippocampus

Deadwyler and Hampson’s insight stemmed from their realization that they could predict when the rats would make a mistake. When the rats pressed the first lever, instead of a strong signal from the CA3 neurons, the device only recorded a weak signal, indicating that the memory was partly missing in the brain. They knew from their previous work that the activity of the CA1 neurons depended on the right signal from the CA3 neurons. (When one neuron repeatedly signals a neighboring cell across a synapse, the tiny space that separates the two cells, the link between them is strengthened, such that the second cell is much more likely to fire when the first cell is active. Young neuroscience students learn this as “cells that fire together, wire together.” CA3 and CA1 are definitely wired together.)

Since this encoding error happened seconds before the rats actually had to recall the memory of which lever they had pressed, Deadwyler, Hampson, and Berger wondered whether they could override the weak pattern with the stronger, correct one. It would mean processing the recorded signal from the CA3 neurons at breakneck speed, but improvements in computing had made such math possible. If it worked, it would be the first bit of proof that we could replace missing memories.

They tested their hunch using the same lever-pressing task. They recorded the signal from the CA1 and CA3 neurons when the rats got the test right. Then, they tested the device not just as a sensor, but as a memory prosthetic, too. When the electrodes detected a weak signal from the CA3 neurons as the rat first pressed the lever, the researchers inserted the correct CA1 signal when it was time for the rats to decide which lever to press second.

When the team watched and didn’t activate the device, the rats performed the trial correctly 80% of the time. But when they switched on the device that provided the proper CA1 signal, the accuracy rate jumped to 95%. They also found that they could interfere with the memory, too, by stimulating the CA1 neurons with an incorrect signal. Then, accuracy dropped to 75%, according to results published in the Journal of Neural Engineering.

Deadwyler, Hampson, and Berger also tested the device on animals that had never completed the lever-pressing task. Without any input from the device, these naïve animals were correct about 60% of the time. With the device, however, the accuracy improved to just below 80%. A naïve rat with the device was about as good as a practiced rat without. “This was the ultimate example that such a device would be useful,” Deadwyler says.

To the team, these results were profoundly significant. But other scientists thought there must be some noise that was spoiling the data. “To publish anything, we had to do a ridiculous number of controls to rule this stuff out,” Deadwyler says, which dramatically slowed the pace of their research.

Although rats are good models for human brains, they’re are obviously not identical. One major difference is the location of the hippocampus in the brain. In rats, the hippocampus is only covered by a thin layer of cortex, that gray wrinkled layer of tissue that we frequently associate with the brain. In humans and other primates, however, the hippocampus is covered by a much thicker layer of cortex. Inserting an electrode in a primate brain would be much more difficult, since the probes would have to be much longer and they would need to be inserted without damaging any other tissue. It took Deadwyler and Hampson two years, and a collaboration with physiologist Greg Gerhardt at the University of Kentucky, before they had something that could be tested in rhesus macaques.

For the macaques, however, the scientists used a slightly more complicated test. First, they showed the monkey an image on a computer screen that was surrounded by a circle or a square. After viewing the image, the screen changed and revealed a group of five different images. If the macaques saw the circle on the first screen, they had to select the identical image from the group. But if it was the square, they had to select the image in the bottom left corner, no matter what it was. Similar to the rat test, the monkeys had to remember the image they had seen, the shape surrounding it, and what image to select next.

After Berger worked his mathematical magic on the output from the electrodes, the researchers found that they could, once again, predict CA1 activity based on input from CA3 neurons. The researchers also found that the monkeys made mistakes in very predictable ways. When they saw an image surrounded by a circle, they sometimes selected the image in the bottom left of the screen. They made a similar error when the square was shown first, picking the matching image instead of the one in the correct location. Neural activity during these mistakes revealed the same type of encoding errors they saw in the rats.

In a follow-up experiment with the monkeys, they used the device to override the incorrect signals from the CA3 neurons. Again, they found that the device significantly improved the accuracy of the macaques in the behavioral test. The results were published in the Journal of Neural Engineering in December 2013.

Finally, Berger, Deadwyler, and Hampson weren’t regarded with raised eyebrows at conferences or regarded as cranks, but were warmly welcomed by their peers. In February, Deadwyler was elected as a Fellow of the American Academy of Sciences.

A Generalizable Device

The beauty of their memory prosthetic, Deadwyler says, is that the patterns of activity in CA3 and CA1 are actually quite general, and the overall process is similar in rats, monkeys, and humans. They indicate a memory being stored and received, not its specific content. This means that the device can be used as a bridge for any type of memory, since it’s just boosting the recall ability of the hippocampus.

“The information that goes into the device would normally be processed by the hippocampus, but the device substitutes for that processing. We’re not putting in anything that isn’t already there, and we’re not telling the brain things like ‘remember an apple’ or ‘remember a face.’ It simply strengthens the normal memory processing that’s already there,” Hampson says.

However, it will be at least several more years until the device is ready to test in humans. For one, they have to create electrodes that can be used in humans and figure out how to insert them without damaging other parts of the brain. They also have to develop a reliable power source for the device. And before anyone can use it, researchers also have to record the activity of CA1 and CA3 neurons to be able to insert the correct signal. None of these are easy tasks.

Rob Malenka, a psychiatrist and neurologist at Stanford University, thinks that memory prosthetics are a promising technology, but he’s not sure how the animal trials will translate into actual use in humans. “In terms of proof of principle, these experiments are very exciting, but translating them to the human brain is a huge leap. They might be able to help you remember your home address, but what about the rest of the information you need in order to be a productive member of society?” Malenka says.

Other scientists are also developing their own types of memory prosthetics. At the University of Pennsylvania, theoretical neuroscientist Michael Kahana is developing a device that would boost the signal in the hippocampus when the brain is trying to encode a memory. When researchers stimulated a region of the hippocampus called the entorhinal cortex in individuals undergoing surgery for epilepsy, the subjects’ memories improved significantly, according to results published in the New England Journal of Medicine.

“This shows that if you can stimulate the brain in the appropriate way, you can improve their memory performance,” says Josh Jacobs, a former postdoc in Kahana’s lab who now runs his own cognitive neuroscience lab at Columbia University where he continues to work on the subject. “We identified the brain signals that were correlated with successful memory encoding, and now we’re trying to devise stimulation protocols that will cause these signals to reappear.”

The two approaches—the one from Kahana and the other from Berger, Deadwyler, and Hampson—would help you remember where you parked your car. The latter prosthesis would be triggered when someone is walking back out to their car and trying to recall where they had left it, while the former, from Kahana’s lab, would be activated when you park, helping to stamp that location in your brain. These devices can’t erase memories nor can they implant false ones. Instead, they would serve to help those with memory difficulties function better in everyday life.

To Loren Frank, a neuroscientist at the University of California, San Francisco, the problem with these devices is their heavy focus on the hippocampus. Certainly, the hippocampus is crucial to memory formation, he says, but his own experiments show that memory formation, storage, and recall involves the whole brain.

“Everywhere that we have looked, we have seen activity that’s related to what’s going on in the hippocampus. If I ask you to think about what you had for breakfast, you can remember what the food tasted like, you can remember where you were and what it looked like. If you had a conversation, you can remember what was said. And we know that all these parts of this experience are processed and stored in various parts of the neocortex,” Frank says. It’s still not clear whether a memory prosthetic would be able to activate the whole, complex range of a memory, he points out.

While both the scientists and funding agencies envision these devices for use in individuals with brain damage from Alzheimer’s disease or a traumatic brain injury, they also have the potential to be used by healthy individuals. And for ethicists, that’s where things get sticky.

Ethical Uncertainties

For James Giordano, a neuroethicist at Georgetown University, the side effects of memory prosthetics are troublesome. He points out that emotion and memory are tightly linked. Remembering your grandmother’s cooking can evoke fond memories of togetherness, whereas other, unpleasant memories can trigger panic or rage. “What if a boost in memory also boosts the emotions related to that memory?” he asks. “Will we be trading a neurological disorder for a psychiatric one?”

Then there’s the bigger issue of who will be given access to memory prosthetics. Neurosurgery is time-consuming and expensive, and it’s not clear how this technology, if and when it appears, will be made accessible to all those who need it.

“How we decide this is a major question for the healthcare system. And this technology won’t just appear in the United States—it’s happening all over the globe. It could happen that someone with enough money will be able travel abroad for a device that can boost their memory, creating a further divide between the haves and the have nots,” Giordano says.

Arthur Caplan, a bioethicist at the New York University School of Medicine, points out that many are hoping these devices can give us back the people we used to know, whether they suffered a brain injury or have Alzheimer’s. The problem is, these memory prosthetics don’t work that way. They don’t rehabilitate lost memories, but rather restore the ability to form new ones. If someone with Alzheimer’s forgets who their child is, the memory prosthetic won’t make them suddenly remember. It will have to be re-learned.

Our memories and experiences, both what we remember and what we don’t, are the foundation of who we are, Caplan explains. Memory prosthetics have the potential to change all that, and it’s unlikely animal studies will help scientists grapple with the subject. “How many mouse studies will be enough? How will we know?” Caplan says. “Even if we could do all the mouse studies in the world, no one is going to wait for that. There’s a tremendous pressure to get these into people right away.”

Frank sees other, more subtle problems. “When you remember, you don’t want to get confused about your memory being real life,” he says. We can think back to where we parked our car while in the grocery store without actually thinking we’re currently parking our car. We know we’re in the produce aisle. For a memory prosthetic to work, Frank says, it will have to allow us to maintain this perspective and not try to put our car in reverse while picking out watermelons.

Nor, he says, is it clear how these devices will function when the circuitry of the brain as a whole is degrading. Will they still operate as expected? Maybe they will, he says, but then again, maybe they won’t.

What many researchers and clinicians do agree on is the need for such devices and the likelihood that they will emerge as a potential treatment for people suffering with severe memory impairments long before we’ve grappled with all of the questions they pose. Scientists have deciphered many aspects of memory, and advances in technology promise to decode even more. For those struggling with memory loss, a device to help them remember and remain the person they are can’t come soon enough.