Forming and recalling memories. Artificially.

Memory manipulation has become one of the most hotly pursued topics in neuroscience. After all, much or of who are is based on what we’ve learned, including memories that we can consciously recall as well as acquired desires and habits that can lead to problems like addiction. In rodents, we’ve known for decades that damage to the hippocampus can erase recently-formed memories. Studies of reconsolidation have shown us that when a memory is retrieved it becomes labile and allows for new information to be added, thereby creating an updated version. More recently we (humans) have been able to identify the neurons involved in memory formation and show that killing them, and only them, results in memory erasure. Bringing us even closer to the stuff of movies, studies by Garner et al. in Science and Liu et al. in Nature have now artificially controlled memory formation and recall. We’re essentially talking about reactivating memory by pushing a button. Yes – you can say “dude, whoah” now.

Artificially activating neurons

The two studies use contextual fear conditioning as their test of learning and memory. In this test mice freeze in a place where they previously received a footshock. In order to remember the context where they received a footshock certain genes are expressed in the neurons that encode the memory. Both groups hijacked one of these genes, c-fos, using it to drive expression of either a DREADD or channelrhodopsin, for pharmacological or optogenetic control of neural activity, respectively. The DREADD is a receptor for a molecule that doesn’t exist naturally in rodents, but when the molecule is injected into the mouse by a mad scientist it binds to the DREADD and causes neurons to fire action potentials kind of like would happen when a neuron is involved in forming or retrieving a memory. Channelrhodopsin also allows a neuron to be selectively activated but it is activated by light. In both studies, these tools served somewhat similar roles – when mice formed a memory the neurons that were involved in encoding that memory (neurons that express c-fos) became labelled in such a way that they could be selectively activated, artificially, by the mad scientist.

Creating a “synthetic” memory

In the Garner et al. study, when a mouse explored a new context (A) the DREADD was expressed presumably only in those neurons that were involved in forming a memory of context A. The next day they put the mice into a different context, B, and delivered footshocks while also activating those neurons previously involved in forming a memory for context A. This is where things get synthetic. Until now, scientists have played with relationships between memories by getting mice to naturally learn or recall two memories at once. But here Garner et al. are teaching the mice one thing naturally (that context B is scary) while simultaneously, artificially, reactivating a neural circuit that represents a previous memory (of context A).

They note that several outcomes could be predicted: 1) By activating context A neurons when the animals form a fearful memory, the mice could form an association such that they freeze whenever those context A neurons are activated in the future. However, this didn’t happen – they did several experiments to show that simply activating context A neurons, or even context B neurons, was not sufficient to evoke fear behavior. 2) Maybe the mice would learn to fear context B normally, since the sensory environment would be the same during both memory formation and retrieval. Again, no – mice were severely impaired at forming fear memories for context B, indicating that activity in context A neurons either interfered with memory formation or… 3) The artificial activation of context A neurons become a requisite component of the fearful memory for context B. Indeed, only when the mice were put back in context B (where they were shocked) AND those same context A neurons were activated (just as they were during training) the mice showed high levels of freezing.

Memory recall, push of a button style

The other study, by Liu et al., is a bit more straightforward to understand and the findings are a bit different. Instead of creating a hybrid memory by co-activating a previous memory (artificially) and a new memory (via experience) as in the Garner study, they used light-activation of channelrhodopsin-expressing neurons to reactivate a context fear memory. By doing this they could reliably cause the mice to freeze in a novel context, suggesting that they could artificially induce memory retrieval.

Differences in artificial memory recall

The Garner study is interesting proof that neuronal populations that are active during learning, even artificially, become essential for memory recall. The fact that only exposure to the shock context or only artificial activation of the context A neurons was insufficient to induce freezing suggests, to me, that these two components of the memory were somewhat distinct. The dogma that “neurons that fire together wire together” would lead one to expect that activation of only a subset of the memory trace would lead to more complete reactivation of the full memory. But perhaps the artificially activated neurons did not fire at the right frequencies or intensities to enable them to become completely integrated.

In contrast, with the optogenetic techniques employed by Liu et al., they were able to sufficiently activate neurons to induce memory recall. This was remarkable because, unlike the DREADD approach used by Garner where presumably all the relevant neurons in the brain could be targeted for reactivation, Liu et al. were probably only able to activate tens or hundreds of neurons due to limited expression of channelrhodopsin in the dentate gyrus of the hippocampus and limited ability of light to penetrate brain tissue and activate those neurons. It could be that CA3, which is downstream of the dentate gyrus, did its job and was able to reactivate the full memory trace from only partial inputs.

Clues about the dentate gyrus

There are a few other things about the Liu et al. study that are particularly noteworthy to me. The first is that they were able to induce recall of a specific experience by activating dentate gyrus neurons. In their control experiments they showed that freezing could only be induced when activating dentate gyrus neurons that were involved in encoding the fear memory but not neurons involved in encoding an unrelated memory. But some of the leading work in the field has shown that the dentate gyrus uses the same cells to encode many different types of experiences (via differences in firing rate). So, if it’s always the same population of active neurons in the dentate gyrus, how could they show this specificity? Wouldn’t activation of neurons for the unrelated memory be the same as activating neurons encoding the fear memory? Well, their c-fos-driven channelrhodopsin/YFP turned out to be a useful tool because it’s expressed in neurons for a long time after memory formation, unlike the endogenous c-fos whose expression decays to baseline levels after a couple of hours. They were then able to exploit these differential timecourses to show that two distinct experiences were in fact encoded by two distinct populations of neurons. This was not at all the take home message of the paper but, as an aside, provides some strong evidence that our knowledge about how information is processed by the dentate gyrus incomplete.

Optogenetic controls. Thank you.

The other aspect of the Liu et al. paper I really liked was the controls. Optogenetics is new, powerful and has already produced many major discoveries. But many are outside of my area of expertise and most have not yet been replicated and so I can’t help but be a little cautious of the validity of these findings. This was how I felt upon seeing similar findings presented at SfN way back in 2009 – cool, but really? The specificity of the induced recall (discussed above), the requisite lack of effect of light in the absence of channelrhodopsin, the fact that activating more neurons with bilateral light delivery enhanced recall, and the fact that activating fewer neurons (but neurons more closely linked to the fear memory) also enhanced artificial recall all bolster the claims and the validity of the method, for me. Cool stuff.

What’s next?

Can we implant a memory that doesn’t exist? Could a population of hippocampal neurons be identified, a priori, that represent a context to be learned in the future? And if you pair activity in those neurons with activity in the amygdala could you elicit fearful behaviour of a context even when nothing dangerous has ever happened there? The answer is probably yes. But it will still be pretty cool (for our grandchildren) to see it demonstrated.


Postscript – be sure to check out coverage of the Liu study over at Nature’s Action Potential blog. Interesting because of the science but also because it gives a glimpse of the publication process – what did the reviewers think of the paper? How is it that the Garner and Liu studies were published in Science and Nature on the same day? Do Science and Nature….talk to each other?? They have the answer.


Aleena R. Garner, David C. Rowland, Sang Youl Hwang, Karsten Baumgaertel, Bryan L. Roth, Cliff Kentros, & Mark Mayford (2012). Generation of a Synthetic Memory Trace Science, 1513-1516 : 10.1126/science.1214985

Xu Liu, Steve Ramirez, Petti T. Pang, Corey B. Puryear, Arvind Govindarajan, Karl Deisseroth, & Susumu Tonegawa (2012). Optogenetic stimulation of a hippocampal engram activates fear memory recall Nature : 10.1038/nature11028

13 thoughts on “Forming and recalling memories. Artificially.”

  1. Nice write up, glad to just be a commentor. What does it say about the specificity of cfos and the “memory trace” if your biggest freezing effect occurs when 85% of the cells in ca1 showed cfos labeling?

    1. Thanks and nice to hear from you! I’ve aways enjoyed your blog, sparse as it may be.

      You’re talking about Fig 2D right, the correlation between freezing and cfos expression? That was just one animal that showed such high levels of cfos, the rest were in a normal range. A bit more of a concern is that 46% of dentate granule cells were activated by the DREADDs – a very non-physiological number. Their Fig 4C does control for the possibility that any DREADD-mediated activation in context B will increase freezing though the design was a little different than their main finding in Fig 2C…

      1. What’s a ‘normal’ level of cfos expression? The relationship between neural activity, cell type, and cfos expression is poorly understood and the timescale of dox-on dox-off is long. This makes interpretation murky.

        It is clearly possible to trigger a crude expression of behavioral modification (freezing) via activating a subset of neurons that were paired with a fearful stimulus. Its quite another thing to say “Our results show that memories really do reside in very specific brain cells,” Liu says, “and simply by reactivating these cells by physical means, such as light, an entire memory can be recalled.”

  2. Poor mice… Victims of interest. You’re right that it is like the movies. The horror movies. I hope scientists who use mice in their research will stop seeing them as objects and begin seeing them for their personalities and intelligence. And also spend their time doing some research that will actually help people, using human cells, and human volunteers, instead of cutting up rodents and shocking them.

    1. Hi Alix – I tried, albeit briefly, to emphasize at the beginning that understanding the neural basis of memory has HUGE implications for human health. That is indeed the primary goal of this research and a very big part of why most scientists do research. This research can help people but it cannot be done on people, except in very rare occasions where patients give us a window into the workings of their brain during medical treatment. I agree with you that, at least once we get the answers we need, we can wean ourselves off of animal research.

  3. Still need to do a lot of reading before I can fully grasp how all this ties together, but my initial thought is the difference between how the DG differentiates similar and non-similar contexts can account for some of these distinct cell populations being summoned in the DG. For example Sahay et al. ( showed how a different context, context C, was not treated the same way by the DG as context B, a similar context, in comparison to the original context A (same idea in Nakashiba et al with only new neurons involved). Then the overlap amongst cells is seen in the more cell (and association) rich CA3 ({8334BE29-8911-4991-8C31-32B32DD5E6C8} ).

  4. I think it would be quite interesting to test them on a simple t-maze task to asymptotic criterion where they must go left if they smell a neutral smell (say “wood”) or right if they smell “grass”. Find those populations for each of those memories (presumably, smelling one thing will cue the rest of the associative memory which would be distinct, possibly, from the other associative memory), and see if, by activating the reverse population, they can change the mouse’s behavior.

    1. That does sound like a cool direction – and it adds a new level of specificity, by activation one memory vs another. Also adds a level of complexity because you’d have to differentially label and activate two distinct populations. But I think such an experiment will eventually be done.

  5. Excellent comments! I am just thinking about if granule cells can work for different behaviors such as pattern separation and completion. One paper recently published on Cell also from Tonegawa lab said that there is a possibility that young and old granule cells might be responsible for oppositing behaviors. In fact, researchers have already found the similar results from piriform cortex, which driving the same or different ensembles of piriform cortex will induce opposite behaviors (aversive or appetitive) by using ChR2.

  6. Hi! can you help me with my thesis? I’m handling animal’s or our pet’s memory. I want to know if you have any idea than can be of my use. Proper credits and copyright’s shall be imposed. by the way, after reading your post, a question came across my mind, if through researches, it was found out that there’s a higher probability that memories can be manipulated in some way, i wonder if in cases of loss memories – i mean like amnesias – will it be a bit of help to the family of the patient if these strats and findings be used?

Comments are closed.