How is a distinct memory formed and stored in the brain? Recent studies indicate that defined populations of neurons correspond to a specific memory trace1
, suggesting a cellular correlate of a memory engram. Selective ablation or inhibition of such neuronal populations erased the fear memory response5,6
, indicating that these cells are necessary for fear memory expression. However, to prove that a cell population is a cellular basis of a specific fear memory engram, one has to conduct a mimicry experiment to show that direct activation of such a population is sufficient for inducing the associated behavioral output9,10
The hippocampus is thought to be critical in the formation of the contextual component of fear memories11,12,13,14
studies have demonstrated an essential role of the dentate gyrus (DG) of the hippocampus in discriminating between similar contexts. Cellular studies of immediately early gene (IEG) expression showed that sparse populations of DG granule cells (2–4%) are activated in a given context18
. Moreover, although the same population of DG granule cells is activated repeatedly in the same environment, different environments19
or different tasks20
activate different populations of DG granule cells. These lines of evidence point to the DG as an ideal target for the formation of contextual memory engrams that represent discrete environments.
To label and reactivate a subpopulation of DG cells active during the encoding of a memory, we targeted the DG of c-fos-tTA transgenic mice3
with the AAV9
-TRE-ChR2-EYFP virus and an optical fiber implant (). This approach directly couples the promoter of c-fos
, an IEG often used as a marker of recent neuronal activity21
, to the tetracycline transactivator (tTA), a key component of the doxycycline (Dox) system for inducible expression of a gene of interest22
. In our system, the presence of Dox inhibits c-fos
-promoter-driven-tTA from binding to its target tetracycline-responsive element (TRE) site, which in turn prevents it from driving ChR2-EYFP expression. In the absence of Dox, training-induced neuronal activity selectively labels active c-fos-expressing DG neurons with ChR2-EYFP, which can then be reactivated by light stimulation during testing (). We confirmed that our manipulation restricts the expression of ChR2-EYFP largely to the DG area of the hippocampus ().
Basic experimental protocols and selective labeling of the DG cells by ChR2-EYFP
First, to characterize the inducible and activity-dependent expression of ChR2-EYFP, we examined its expression timeline under various treatments (). We observed a complete absence of ChR2-EYFP expression in DG neurons while mice were on Dox (). Two days off Dox was sufficient to induce ChR2-EYFP expression in home-caged mice (). The number of ChR2-EYFP-positive cells increased significantly in response to two days off Dox followed by fear conditioning (FC; ). We found that the vast majority of c-fos-positive cells were also ChR2-EYFP-positive (Supplementary Fig. 1
), confirming that activity-dependent labeling with ChR2-EYFP recapitulated the induction of endogenous c-fos. A similar increase in ChR2-EYFP expression was seen in a group of mice that was exposed to the same context and tone as the FC group but had no shocks delivered (NS; ). ChR2-EYFP expression lasted at least five days () and was gone by 30 days (). Kainic-acid–induced seizures resulted in complete labeling of DG cells with ChR2-EYFP (), indicating that the relatively sparse labeling in the FC or NS groups was not due to the low infection rate of the virus, but reflected the natural low activity of DG neurons during the training sessions18,23
. Notably, NS and FC treatments resulted in similar proportions of ChR2-EYFP-positive cells (). ChR2-EYFP expression following FC seemed to be restricted to the excitatory neurons, as no overlap was detected between ChR2-EYFP-positive neurons and GABA-positive inhibitory neurons (Supplementary Fig. 2
Activity dependent expression and stimulation of ChR2-EYFP
We injected c-fos-tTA mice with either AAV9
-TRE-ChR2-EYFP or AAV9
-TRE-EYFP, subjected them to fear conditioning while off Dox, and then put them back on Dox to test for light-evoked responses from DG cells the following day. The mice were anesthetized for in vivo
recordings and blue light pulses (473 nm, 0.1 Hz, 15 ms pulse duration) were delivered to the DG. Consistent with the sparse labeling of DG neurons (), we identified only 10 DG neurons that responded to light stimulation from nine c-fos-tTA mice injected with AAV9
-TRE-ChR2-EYFP (the ChR2 group). In these neurons, we detected a reliable increase of spike probability precisely time-locked to the onset of light pulses (). These cells also showed robust responses to trains of 20 Hz light stimulation with a slight decrease in spike probability over time that remained significantly higher above baseline (). We did not find any light-responsive cells in the 10 c-fos-tTA mice injected with AAV9
-TRE-EYFP (the EYFP group; data not shown). Most of the ChR2-EYFP-positive cells in the ChR2 group of mice were also positive for endogenous c-fos after optical stimulation, although not all c-fos-positive cells expressed ChR2-EYFP. Very few neurons expressing EYFP in the EYFP group of mice were c-fos-positive ( and Supplementary Fig. 3
). The proportion of c-fos-positive cells in the downstream CA3 region was greater in the ChR2 group compared with the EYFP group after optical stimulation of DG neurons, and this number was comparable to the proportion of CA3 c-fos-positive cells obtained by exposing a separate group of fear conditioned mice to the conditioned context (Supplementary Fig. 4
Next, we tested whether activating a population of DG neurons labeled by ChR2-EYFP during the encoding of a fear memory was sufficient for memory recall. The experimental group (Exp) consisted of c-fos-tTA mice unilaterally injected with AAV9
-TRE-ChR2-EYFP and implanted with an optical fiber targeting the DG (). Mice were kept on Dox and underwent five days of habituation to record their basal level of freezing in one context (context A) during both light-off and light-on epochs. Next, they were taken off Dox and underwent fear conditioning in a distinct chamber (context B) in which a tone was paired with shock. The mice were then subjected to five days of testing with light-off and light-on epochs in context A while on Dox (). During the habituation sessions, the Exp mice showed very little freezing during either light-off or light-on epochs. In contrast, after fear conditioning, freezing levels during light-on epochs were significantly higher compared with light-off epochs, which indicated light-induced fear memory recall (). Increased freezing during light-on epochs was observed across all five days of test sessions with no discernible day-dependent difference (Supplementary Fig. 5g
). These data suggest that DG cells that express endogenous c-fos during training, and therefore become labeled by ChR2-EYFP, define an active neural population that is sufficient for memory recall upon subsequent reactivation.
Optical stimulation of engram-bearing cells induces post-training freezing
To rule out the possibility that post-training freezing by optical stimulation was due to the activation of DG cells unrelated to fear learning, we injected another group of mice (NS) with AAV9
-TRE-ChR2-EYFP and administered the same habituation, training, and test sessions as the Exp group, except that no shock was delivered during the training session. Despite the fact that a similar level of ChR2-EYFP expression was detected in the NS group compared with the Exp group, both in terms of proportion of cells labeled () and ChR2-EYFP fluorescence intensity per cell (Supplementary Fig. 6
), light did not induce post-training freezing in the NS group (). This indicates that the freezing observed in the Exp group requires optical activation of a specific subset of ChR2-EYFP-positive DG cells that are associated with FC and that activating a population of DG cells not associated with FC does not induce freezing. Yet another group of mice (EYFP) were injected with AAV9
-TRE-EYFP and underwent identical habituation, training, and testing sessions as the Exp group. The proportion of cells expressing EYFP was comparable to that seen in the Exp group expressing ChR2-EYFP (Supplementary Fig. 7
). However, the EYFP group did not show increased post-training freezing (). This result rules out the possibility that increased freezing in the Exp group was due to any non-specific effects of post-training optical stimulation.
The light-induced freezing levels of the Exp group were relatively low (~15%) compared with those typically reported from exposure to a conditioned context (~60%)3
. One possibility is that light activation of background-activity–induced ChR2-EYFP () interfered with the expression of the specific fear memory. We confirmed that limiting the off Dox period from two days to one day reduced the background expression of ChR2-EYFP by at least twofold (compare Supplementary Fig. 8a
Home cage with Home cage). A group of mice (Exp-1day) that went through the same design outlined in but with this modification showed greater freezing levels (~25%) during the light-on epoch of test sessions compared to the Exp group (). Another possible factor contributing to the modest light-induced freezing in the Exp group may be the limited number of cells optically stimulated. To test this possibility, we bilaterally injected a group of mice (Exp-Bi) with AAV9
-TRE-ChR2-EYFP and bilaterally implanted optical fibers targeting the DG, and then subjected these mice to the same scheme as that shown in . During the light-on epochs of the test sessions, the Exp-Bi group exhibited levels of freezing (~35%) that were almost as high as those induced by the conditioned context (, Supplementary Fig. 9
, and Supplementary movies).
We next examined whether the light-induced fear memory recall was context-specific. First, to test whether two different contexts activate similar or distinct populations of DG cells, we took the mice off Dox for two days and then exposed them to a novel context (context C, an open field) to label the active DG cells with ChR2-EYFP. After being put back on Dox, the mice were fear conditioned in a different context (context B) and sacrificed 1.5 hours later (). The expression of ChR2-EYFP was used to identify cells previously activated in context C whereas endogenous c-fos was used to identify cells recently activated in context B. Immunohistochemical analyses revealed a chance level of overlap between ChR2-EYFP–positive and c-fos–positive cells, suggesting that two independent DG cell populations were recruited for the representation of the two distinct contexts (). To test the context-specificity of light-induced recall of a fear memory, we subjected a new group of mice (an open field-fear conditioned group, OF-FC) to habituation sessions in context A, followed by two days off Dox and context C exposure to label neurons active in context C with ChR2-EYFP. Next, we put the mice back on Dox and performed fear conditioning in context B (). These mice were then placed back in context A and tested for light-induced freezing. Light failed to evoke a significant increase in freezing responses (). Similarly low levels of freezing were observed in another group of mice (FC-OF) in which fear conditioning in context B while on Dox preceded exposure to context C while off Dox (Supplementary Fig. 10
). Together, these results indicate that light reactivation of cells labeled in context C did not induce fear memory recall associated with context B.
Labeling and stimulation of independent DG cell populations
Here we have shown that optical activation of hippocampal cells that were active during fear conditioning elicits freezing behavior. To our knowledge, this is the first demonstration that directly activating a subset of cells involved in the formation of a memory is sufficient to induce the behavioral expression of that memory. Our results and previous studies that addressed the necessity of similarly sparse cell populations in the amygdala5,6
argue that defined cell populations can form a cellular basis for fear memory engrams. The memory engram that we selectively labeled and manipulated is likely contextual in nature, as previous studies have demonstrated that hippocampal interventions affect conditioned freezing responses to a context but not a tone12,13,24
. Indeed, recent findings show that optogenetic inhibition of the hippocampal CA1 region during training or testing both inhibited the recall of contextual fear memory, while leaving auditory-cued fear memory recall intact25
. However, we cannot completely rule out the possibility that the fear memory recalled in our experiments may have some tone memory component.
Our observation that freezing responses were elicited by optical stimulation in the experimental groups (Exp, Exp-1day, and Exp-Bi), but not in the OF-FC or FC-OF group, strongly supports a dual memory engram hypothesis of contextual fear conditioning26,27,28
. In this hypothesis, hippocampal cells are recruited to form contextual memory engrams, but these contextual engrams alone do not represent a complete fear memory. For a fear memory to be formed, the information from the contextual memory engram must be transferred to the basolateral amygdala (BLA) coincidentally with the information representing a foot shock. In the OF-FC or FC-OF scheme, two distinct contextual memory engrams were formed in the DG, which were represented by two distinct sets of DG cells. One of these two contextual engrams (the one for context B) was associated with the representation of the shock, but not the other engram (the one for context C). Since only the latter, but not the former, was labeled by ChR2, optical stimulation could not elicit fear memory expression.
Although we have demonstrated the “sufficiency” of a DG memory engram for the behavioral expression of fear memory, it does not necessarily mean that this engram is “necessary” for behavioral recall. During contextual fear conditioning, it is likely that multiple contextual memory engrams are formed in a series of hippocampal regions. Each of these engrams may contribute to the formation of the complete fear memory in the BLA and may also be capable of reactivating it independently as we observed in the case of the DG engrams. Since the hippocampus is not a linear feed-forward network but contains several parallel circuits, inhibiting the formation or activation of contextual engrams in one region may not necessarily block the expression of the fear memory. For instance, disruption of contextual memory engrams in the DG could be circumvented by CA1 engrams, which could be generated through the direct input from the entorhinal cortex and may be sufficient to activate the fear memory engram in the BLA. Indeed, we recently generated a mouse mutant, which permitted us to demonstrate that the DG input to the CA3 is dispensable in the formation and retrieval of contextual fear memory17
The approach and methods described in this work will be a powerful tool for mapping multiple component engrams, each contributing to an overall memory. A multifaceted analysis of these engrams and their interplay will reveal the nature of the overall memory engram.