New Theory Suggests the Brain Forms Many Copies of the Same Memory

An MIT study of the neural circuits that underlie memory process reveals, for the first time, that memories are formed simultaneously in the hippocampus and the long-term storage location in the brain’s cortex. This image shows memory engram cells (green and red) which are crucial for permanent memory storage in the prefrontal cortex. Image: Takashi Kitamura.

I keep five copies of my photos—on my hard drive, external hard drive and three different cloud storage services. It’s excessive, I know. But I don’t want to risk the only lifeline to my memories when they begin fading away.

As it turns out, the brain’s memory storage system may also operate with a “backup”. By studying H. M. and other patients who can’t encode new memories or forgot old ones, scientists have long pinned down the hippocampus as the center of one of our most prominent memory networks.

The hippocampus encodes the stories of our lives—the whats, whos, whens and wheres, called episodic memories. It gives our daily morning coffees separate contexts, so we remember two different Starbucks runs as two distinct events.

But memories don’t settle down within hippocampal circuits. The seahorse-shaped structure acts more like a grand central station, eventually shuttling memories out into various regions of the cortex, the outermost layer of the brain. There, memories semi-permanently etch themselves into the connections—the synapses—between neurons and settle for the long haul.

Or so the story goes.

In a surprising study this month, Dr. Susumu Tonegawa (of memory-inception fame) and team at MIT uprooted the classical theory of memory with another narrative.

Rather than encoding a single "memory trace" that's transferred from hippocampus to cortex, the brain may be making two copies simultaneously in both regions. However, the cortical copy remains “silent” until nearly two weeks after learning—only then do they come online, ready for retrieval.

In other words, the brain may be encoding a ninja backup memory trace that gets unmasked as time passes.

So how did the scientists figure this out?

Memory Engram

Tonegawa has spent the last decade studying where memories are physically stored in the brain. Not every neuron is activated when we learn, so it makes intuitive sense that the memory is only stored in the set of firing neurons at the time of encoding—something that Tonegawa calls a “memory engram”.

A few years ago, his team figured out a way to “tag” these memory engram cells while the mouse learns.

Here’s how it goes: a session generally starts with an innocent-looking box, sprayed with flowery perfume and painted bright colors. Put the mouse in the box, and give its paws a series of electrical shocks, and the mouse quickly learns to associate that particular box with danger. Put a smart mouse back in the same place a few days later, and it’ll freeze in fear. This test, contextual fear conditioning, offers a way for scientists to test a mouse’s memory.

The team first traced the brain regions downstream of the hippocampus by using a type of virus (expressing a fluorescent tag) that jumps from neuron to neuron. Hippocampal neurons, through a part of the brain called the MEC, extensively projects to cells the in prefrontal cortex (PFC), a brain region often involved in flexible reasoning, and cells in the amygdala, the “emotion center” of the brain.

Mapping Fear

But are all these regions involved in fear learning? To answer this, the team genetically inserted a light-sensitive protein into neurons in the downstream regions and implanted thin optic fibers. By shining light on these proteins (called “rhodopsins”), scientists can control their activity, either turning them on or off—a technique called optogenetics.

Here’s the crucial part: the mice are given a drug that normally blocks the production of these proteins. No rhodopsins, no response to light.

In this way, once scientists take away the drug and let the mice explore the fear conditioning box, only cells involved in fear learning are “tagged” with the light-sensitive proteins. This allows scientists to manipulate entire memory engrams with light later on--activate engram, activate fear memory; inhibit engram, inhibit memory and bam--no more freezing!

Green shades are when the animal is on the drug. Take the drug away, and some cells (red) becomes labeled with the light-sensitive protein. During fear conditioning in the red triangle room the memory is stored in the red cells (B), not the grey ones that encode for a different place. When scientists shine light on the red cells (B), the memory activates, even when the mouse is in a different room (grey square). Putting the mouse back into the original triangle room (A') also retrieves the memory, whereas putting the mouse into a third room (white hexagon) lables a different population of cells, and the mouse doesn't freeze.

Ramirez, S., Liu, X., Lin, P., Suh, J., Pignatelli, M., Redondo, R., . . . Tonegawa, S. (2013). Creating a False Memory in the Hippocampus. Science, 341(6144), 387-391. 

It worked. Focusing first on the PFC, the team activated its fear memory engram with light while the mice were in an otherwise "safe" environment. They froze, as if remembering the scary incident even though they weren't re-exposed to the fear box (which normally triggers the bad memory).

Similarly, the PFC engram sparked into action when the mice were placed back into the box--but there's a twist. The PFC only came online 12 days after the initial training. When tested only two days after, it remained silent.

What’s more, inhibiting the PFC engram two days after training had no effect on the mouse’s behavior—it still froze in fear inside the old box. But try the experiment on day 12 and it’s a different story: the mouse seemed to have forgotten the memory.

Finally, the team used a toxin to inhibit the activity of the hippocampal engram cells. Lo-and-behold, this also disrupted the reactivation of PFC engram cells. In other words, it seems that the hippocampus is somehow helping the PFC engram "mature"--without their help, the PFC long-term memory doesn't come online.

Putting it all together, it seems that while the cortex forms an engram right off the bat, it lays dormant until later stages—generally when scientists believe memories become long-term.

You may be wondering what’s happening within two weeks of learning. While the PFC engram remained silent, the hippocampal copy shouldered the brunt of the memory. Activating the hippocampal engram on day two resulted in the mouse freezing in place, even in a safe environment; but on day 15, the engram no longer seemed to be dominant. When researchers looked at the density of dendritic spines—little protrusions where synapses sit—they were dramatically reduced by day 15 compared to day 5 after training, suggesting that the neurons were no longer super important for memory recall.

Division of Labor

The study shows that the brain has two complimentary memory encoding systems, working simultaneously but with different strengths and weaknesses. According to Tonegawa, the hippocampus forms active memories quickly, whereas the cortex takes care of the long-term stability.

This division of labor offers the perfect safety net for our memories. “If you don’t need prolonged memory, the hippocampus is enough; if you don’t have to form active memory quickly, the cortex is enough; but we want both,” says Tonegawa.

But what are these different engram cells doing? Scientists don’t have the answer, but previous studies suggest that hippocampal neurons are likely encoding the “where”, or the environment of the box. The PFC neurons may help with choosing strategies or weighing options—fight or flight, so to speak—which may be more important once the memory becomes stable. 

Although the hippocampus and PFC are generally considered the sexier parts of the brain, the authors also stressed the importance of the amygdala, the “emotion center”, for fear memory. The almond-shaped amygdala connects with the hippocampus and the PFC. Block the amygdala during learning—bam—no engrams and no memory. During retrieval blocking hippocampal to amygdala connections two days (but not 15 days) after training impaired memory. The opposite timeline effect was true when researchers blocked the PFC-amygdala connection.

In other words, the amygdala engram is always present, but as the memory “matures”, a switch flips in the use of its connections. For example, says Tonegawa, at day 15 when the PFC engram is “online”, the connectivity between it and the amygdala engram allows the mouse to recall the fear memory.

Brain hurtin'? Me too! Here's a nifty summary from Fig 4 of the paper. The PFC engram forms during training based on inputs from the hippocampus (HPC) and amygdala (BLA). Recent recall only requires the connections between the hippocampus and amygdala. As the PFC engram matures (with some help from the hippocampus), the connection between the PFC and amygdala becomes more dominant in retrieving the memory.

The study is really humbling, in that it creatively takes aim at one of the most beloved theories of memory formation in neuroscience. It really shows how much we have yet to learn, and makes a great case for developing new technologies to manipulate neural circuits and memory engrams.

Whether similar “backups” occur for other types of memory (such as motor skill—learning to ride a bike) remains to be seen. It's true that these studies don’t have immediate applications to our daily lives. However, little by little, they do build upon our understanding of how memories work. And one day, they may lead us to new ways to safeguard or even boost the precious histories of our lives.