Schizophrenia Research Forum - A Catalyst for Creative Thinking

What (Engineered) Memories Are Made Of

26 March 2012. Sometimes the best way to understand how something works is to try to build it—and researchers have taken this approach to create synthetic memories in a study published 23 March in Science. Using a designer receptor and ligand system to artificially evoke activity in a specific ensemble of neurons during fear conditioning in mice, Mark Mayford of The Scripps Research Institute in La Jolla, California, and colleagues detected a hybrid memory that mixed the perceived environment with artificially generated activity. A similar optogenetics study appearing online 22 March in Nature from the laboratory of Susumu Tonegawa at the Massachusetts Institute of Technology in Cambridge found that light-induced activation of hippocampal cells involved in laying down a memory is sufficient for recall. Together, the findings argue that distinct ensembles of neurons encode memory for different environments.

Though not directly relevant to schizophrenia, both studies introduce a valuable way to dissect the neural circuitry underlying behavior by simulating more naturalistically neural responses to the real world. This contrasts with past studies, in which electrical or optogenetic stimulation evokes activity in specific brain regions or subtypes of cells (see SRF related news story). Both new studies transiently couple expression of an introduced receptor—a G protein-coupled receptor dubbed hM3Dq in Mayford’s study, or channelrhodopsin-2 (ChR2) in Tonegawa’s study—to c-fos, an immediate early gene driven by electrical activity in neurons. This gave them access to more realistic ensembles of cells involved in encoding the different environments used in fear-conditioning paradigms.

Memory by design
The hM3Dq receptors in Mayford and colleagues’ study made a splash in 2009 when researchers used them to remotely control hippocampal cells (Alexander et al., 2009). Developed by Bryan Roth at the University of North Carolina, Chapel Hill, these receptors were evolved to fit an otherwise inert ligand called clozapine-N-oxide (CNO), and named DREADD for “designer receptor exclusively activated by designer drug.” Injecting CNO to activate hM3Dq-expressing neurons in the brain is less invasive and may reach a wider distribution of neurons in the brain than optogenetics techniques, which implant a light source into the brain that has a limited range. However, this spatial advantage of the DREADD system is undercut by the inability to precisely control the temporal patterns of activity, as pointed out by Richard Morris and Tomonori Takeuchi of the University of Edinburgh in Scotland in a commentary accompanying the Science study.

First author Aleena Garner and colleagues used a doxycycline-regulated system to permit hM3Dq expression when driven by c-fos. Removing doxycycline, then placing the mice in a specific environment allowed the researchers to specifically label the neurons activated in that context with hM3Dq receptors. The researchers could then later reactivate this ensemble by simply giving the mice CNO. The researchers recorded nearly a fivefold increase in neural activity in CA1 of the hippocampus 30 to 40 minutes after CNO delivery.

The researchers then used a fear-conditioning protocol to train the mice to fear an environment in which they receive a foot shock. On the first day, they took mice off doxycycline and exposed them to a new environment (context A)—thus driving the neurons activated by this environment to express hM3Dq. On day 2, doxycycline resumed, preventing any new hM3Dq expression. The researchers then exposed the mice to a different environment (context B), in which they also received a foot shock. This would normally teach the mice to fear context B—measured by the amount of freezing behavior they show when later placed into the same environment. But the researchers added a twist to this scenario by also giving CNO just before training the mice in context B.

Which would win out when testing for memory retrieval 24 hours later? Would the artificial activation of context A prevail and elicit more freezing, or would the naturalistic experience of context B dominate? It was a tie. When placed into context A, hM3Dq-expressing mice showed low levels of freezing compared to controls, with or without CNO on board. This suggested that natural or artificial activation of context A neurons was not sufficient for recall. Similarly, when placed into context B, hM3Dq-expressing mice spent 10 percent of the time freezing, whereas controls spent 40 percent, a significant decrease that suggests that the physical environment of context B alone was not enough for reliable recall.

However, when hM3Dq-expressing mice were placed into context B and given CNO, their freezing increased to about 25 percent, and did not differ from controls also given CNO. This suggested that during training, the mice formed a hybrid representation of the environment that mixed their experience of context B with the CNO activation of neurons associated with context A. Reactivating these ensembles together was key for retrieval, which is consistent with an earlier study from Mayford’s group, which found that learning and memory retrieval in fear conditioning activate the same neurons (Reijmers et al., 2007).

Specific and sufficient
The CNO-induced activity seemed to reflect activation of a specific ensemble of neurons encoding context A, rather than a generalized brain state, such as a mood, that also modulates memory. For example, the researchers followed the fear-conditioning protocol that resulted in the hybrid memory trace, but before testing retrieval, they allowed hM3Dq expression to fade away, then re-labeled neurons activated by a third environment (context C) with hM3Dq receptors. These animals were impaired in memory recall when placed in context B with or without CNO, which argues that the memory trace formed during conditioning incorporated activity specific to context A. In another experiment, the researchers found that CNO-activation of an ensemble related to one context did not enhance retrieval when the mice were placed in that same context, again suggesting that both the actual experience of the environment and CNO were acting on the same group of neurons.

The Nature paper from Tonegawa’s group shows that optogenetics is also up to the task of stimulating naturalistic groupings of neurons. Interested in the neural overlap between learning and memory retrieval, first authors Xu Liu and Steve Ramirez used c-fos-driven ChR2 expression to label hippocampal neurons that were activated during fear conditioning in a particular environment (context B). They found that simulating this activity later, through light-induced activation of these neurons, was sufficient to induce freezing in a different environment (context A), and increased with bilateral stimulation. They also found that the light-induced recall was context specific, and turned up anatomical evidence for sparse but distinct neural populations in the hippocampus that were activated by different contexts.

These techniques further refine the manipulations of neural activity, and will let researchers test more thoroughly their ideas about how the brain works. The studies also touch on how perception of the physical world can be jumbled by internally generated activity, an interaction that may eventually point the way to a deeper understanding of mental illness.—Michele Solis.

Garner AR, Rowland DC, Hwang SY, Baumgaertel K, Roth BL, Kentros C, Mayford M. Generation of a Synthetic Memory Trace. Science. 2012 Mar 23; 335: 1513-1516. Abstract

Morris RGM, Takeuchi T. The Imaginary Mind of a Mouse. Science. 2012 Mar 23; 335: 1455-1456. Abstract

Liu X, Ramirez S, Pang PT, Puryear CB, Govindarajan A, Deisseroth K, Tonegawa S. Optogenetic stimulation of a hippocampal engram activates fear memory recall. Nature . 2012 Mar 22. Abstract

Comments on News and Primary Papers

Primary Papers: Generation of a synthetic memory trace.

Comment by:  John Neumaier
Submitted 26 March 2012
Posted 26 March 2012

The study by Mayford's group is the most elegant application of DREADD receptors so far. While it is not clear what the minimal ensemble of neurons that can represent the memory trace are, the fact that the distributed set of neurons can be selectively activated so unobtrusively (with CNO) is a real innovation.

View all comments by John Neumaier

Primary Papers: Generation of a synthetic memory trace.

Comment by:  John Gray
Submitted 26 March 2012
Posted 26 March 2012

In this week’s issue of Science, Garner et al. take an ingenious leap forward in the use of the growing collection of tools for the remote activation of neurons. They utilize a genetic tool known as a DREADD receptor, a designer receptor exclusively activated by designer drug, pioneered in the laboratory of Bryan Roth (Alexander et al., 2009). DREADDs are G protein-coupled receptors that have been created to only be activated by clozapine-N-oxide (CNO), an analogue of clozapine that does not have a high-affinity binding site in the brain. Thus, when a DREADD, in this case, hM3Dq, is introduced into neurons in the brain, those cells can be activated by the administration of CNO. These exogenous receptors can be introduced into specific brain regions or cell types by viral injection or the use of transgenic mice. The conceptual advance that Garner et al. provide is the expression of hM3Dq only in neurons that were previously activated by experience. To accomplish this, they coupled the expression of hM3Dq to the endogenous expression of the immediate early gene c-fos, a transcription factor that is rapidly expressed when a neuron fires. Thus, in cells that have been sufficiently activated to express c-fos, the introduced gene for hM3Dq will turn on. Subsequent CNO administration will then cause the firing of only the specific network, or ensemble, of cells that were previously activated by an experience. This exciting new approach will allow researchers to address fundamental questions in systems neuroscience, and can be fine-tuned with the power of mouse genetics to look at the role of specific brain regions and subsets of neurons in neural representations.


Alexander GM, Rogan SC, Abbas AI, Armbruster BN, Pei Y, Allen JA, Nonneman RJ, Hartmann J, Moy SS, Nicolelis MA, McNamara JO, Roth BL. Remote control of neuronal activity in transgenic mice expressing evolved G protein-coupled receptors. Neuron . 2009 Jul 16 ; 63(1):27-39. Abstract

View all comments by John Gray

Primary Papers: Generation of a synthetic memory trace.

Comment by:  Christoph Kellendonk
Submitted 30 March 2012
Posted 3 April 2012
  I recommend this paper

One major limitation in the study of cognition has been that cognitive processes such as learning and memory may be mediated by distributed networks in the brain that engage only a low percentage of neurons in a given brain structure (e.g., the hippocampus). In consequence, widely used tools such as brain imaging, in-vivo electrophysiology, and postmortem analysis do not have the necessary resolution and signal-to-noise ratio to study the functioning of these distributed networks.

Recently, Mark Mayford used an elegant transgenic mouse strategy to label cells that were activated by a learning experience. Using these mice, they were able to show that those cells are also reactivated during memory retrieval, suggesting that they form part of the "memory trace" encoding this learning experience. Now, in two equally elegant studies, Mayford's group and a group led by Susumu Tonegawa have studied what happens if you reactivate these neurons at later time points.

The data from the Tonegawa group suggest that the cells activated during learning not only get reactivated during recall, but activation of these cells on its own is actually sufficient to retrieve memory—further confirming the idea that these neurons are part of the "memory trace."

The study from the Mayford lab addresses how activation of competing internal representations affects learning and recall. The data suggest that reactivating memory traces in a new context interferes with both learning and retrieval in the new context. However, activation of the competing representation is not just destructive; rather, it becomes part of a new learning experience leading to a hybrid representation that itself can be recalled later.

As someone interested in schizophrenia, I wonder about the consequences if a pathological induced internal representation due to a psychotic episode is competing with new learning. It may interfere with the learning process and could be one reason for disrupted cognitive functioning observed during acute psychosis. More speculatively, a hybrid representation may be formed between the newly learned information and the content of the psychosis which is later recalled as a perceived "real" memory. Unfortunately, since we can't (yet) artificially induce psychosis in mice using the DREADD system or optogenetics, we will have to wait until Mark Mayford comes up with another brilliant idea in order to test this idea.

View all comments by Christoph Kellendonk

Comments on Related News

Related News: SfN 2013—New Tools for Rational Drug Design

Comment by:  Hugo Geerts
Submitted 29 January 2014
Posted 5 February 2014

Multi-target drug discovery has typically been neglected in the world of genetics and high-throughput screening because of the difficulty of rationally defining a pharmacological profile, but it has major advantages for treating complex disorders such as schizophrenia. It is no wonder that the currently approved antipsychotics do have a rich pharmacology and substantially improve the clinical phenotype. With so many different genotypes defining individual patients, focusing on only one target is likely to have small effects that might disappear in clinical trials with larger patient populations. Even over all indications (not only CNS), more than half of the first-in-class medicines approved in the last decade have been found by using phenotypic assays and have typically multi-target pharmacology (Swinney and Anthony, 2011).

The approach presented here suggests a rational way to identify 1) a set of targets and 2) chemical structures that might serve as hits for further medical chemistry development. It might therefore alleviate the concerns of many medical chemistry departments in pharmaceutical companies.

Changing the mindset from developing the next extremely specific and potent inhibitor to pursuing multi-target pharmacology is urgently needed to break the deadlock of unsuccessful new drug development in schizophrenia.


Swinney DC, Anthony J. How were new medicines discovered? Nat Rev Drug Discov . 2011 Jul ; 10(7):507-19. Abstract

View all comments by Hugo Geerts