Synapse-specific representation of the identity of overlapping memory engrams

See allHide authors and affiliations

Science  15 Jun 2018:
Vol. 360, Issue 6394, pp. 1227-1231
DOI: 10.1126/science.aat3810

Disentangling specific memories

Each memory is stored in a distinct memory trace in the brain, in a specific population of neurons called engram cells. How does the brain store and define the identity of a specific memory when two memories interact and are encoded in a shared engram? Abdou et al. used optogenetic reactivation coupled with manipulations of long-term potentiation to analyze engrams that share neurons in the lateral amygdala (see the Perspective by Ramirez). Synapse-specific plasticity guaranteed the storage and the identity of individual memories in a shared engram. Moreover, synaptic plasticity between specific engram assemblies was necessary and sufficient for memory engram formation.

Science, this issue p. 1227; see also p. 1182


Memories are integrated into interconnected networks; nevertheless, each memory has its own identity. How the brain defines specific memory identity out of intermingled memories stored in a shared cell ensemble has remained elusive. We found that after complete retrograde amnesia of auditory fear conditioning in mice, optogenetic stimulation of the auditory inputs to the lateral amygdala failed to induce memory recall, implying that the memory engram no longer existed in that circuit. Complete amnesia of a given fear memory did not affect another linked fear memory encoded in the shared ensemble. Optogenetic potentiation or depotentiation of the plasticity at synapses specific to one memory affected the recall of only that memory. Thus, the sharing of engram cells underlies the linkage between memories, whereas synapse-specific plasticity guarantees the identity and storage of individual memories.

Memories are formed through long-term changes in synaptic efficacy, a process known as synaptic plasticity (17), and are stored in the brain in specific neuronal ensembles called engram cells, which are reactivated during memory retrieval (813). When two memories are associated, cell ensembles corresponding to each memory overlap (1419) and are responsible for the association (18). Although multiple associated memories can be encoded in the overlapping population of cells, each memory has its own identity (14, 18). Synaptic plasticity is essential for the retrieval, but not the storage, of associative fear memories (5, 20, 21). However, how the brain defines the identity of a particular memory amid the many memories stored in the same ensemble has been elusive.

We asked whether individual memories stored in a shared neuronal ensemble would maintain their identities and have a different fate if one memory was erased by complete retrograde amnesia. We subjected mice to auditory fear conditioning (AFC), in which a tone was associated with a foot shock. This association is mediated by synaptic plasticity between neuron terminals of the auditory cortex (AC) and the medial part of the medial geniculate nucleus (MGm) and neurons of the lateral amygdala (LA) (22). Two different tones, at 2 and 7 kHz, were used. Mice discriminated between the two tones and showed a freezing response only to the 7-kHz tone that was paired with shock (figs. S1 and S2). To completely erase memories, we used autophagy, which is a major protein degradation pathway wherein the autophagosome sequesters a small portion of the cytoplasm and fuses with the endosome-lysosome system to degrade the entrapped contents. Autophagy contributes to synaptic plasticity (23, 24), and its induction by the peptide tat-beclin enhances destabilization of synaptic efficacy after reactivation of these synapses through the degradation of endocytosed α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptors (25, 26). When tat-beclin is combined with inhibition of protein synthesis after memory retrieval, complete retrograde amnesia is induced through enhanced memory destabilization and reconsolidation inhibition (25).

To optogenetically manipulate specific memories, we used c-Fos::tTA transgenic mice; we injected adeno-associated virus (AAV) expressing Cre recombinase under the control of tetracycline-responsive element (TRE) in combination with AAV encoding DIO-oChIEF-citrine, downstream of the human synapsin (hSyn) 1 promoter, into the AC and MGm (both of which relay auditory information to the LA) to label the activated ensemble with a channelrhodopsin variant, oChIEF (Fig. 1, A to D, and fig. S2). Mice were trained with AFC (7-kHz tone plus shock) 2 days after doxycycline withdrawal (OFF DOX). One day later, under the ON DOX condition, the LAs of these mice were infused with phosphate-buffered saline (PBS), anisomycin, or anisomycin combined with tat-beclin (Ani+tBC) immediately after the test session (day 5). The anisomycin infusion induced partial retrograde amnesia, whereas Ani+tBC accomplished complete amnesia, with the freezing level comparable to that of nonshocked and unpaired control groups (Fig. 1E). Optogenetic activation of the axonal terminals of the AC and MGm engram cells in the LA induced fear memory recall in the PBS and anisomycin groups, which is consistent with a previous study (21), whereas it failed to do this in the Ani+tBC-treated mice (Fig. 1F).

Fig. 1 Complete and long-lasting erasure of fear memory trace from AC-LA and MGm-LA engram circuits.

(A) Model showing the ensemble responsive to the 2- and 7-kHz tones in the AC and the fear-responsive ensemble in the LA. (B) Labeling strategy for the AFC-responsive ensemble in the AC and MGm, using the c-Fos::tTA transgenic mice. (C) Expression of oChIEF in AC and MGm neurons and their axonal terminals in the LA. Dashed lines show the borders of the MGm and LA. Scale bars, 100 μm. (D) Design of memory engram erasure experiment. Different chambers were used for each session. (E to I) Freezing levels (percent of time) before and after drug injection (E), during 10-Hz optical stimulation (F), in response to the conditioned and neutral tones after optical LTP (G), at a remote time point (H), and during 10-Hz stimulation at a remote time point (I). n = 20 mice per group in (E) and (F) and 10 mice per group in (G) to (I). Bottom panels of (E) to (G) show statistical significance between groups during test 2 (E), during light-off and light-on epochs (F), and during test 5 (G). The right panel of (I) shows statistical significance within and between groups during test 9. Statistical comparisons were performed using one-way analysis of variance (ANOVA) [(E), (G), and (H)] and two-way ANOVA [(F) and (I)]. *P < 0.05; **P < 0.01; ns, not significant. In the bottom panels of (F), (G), and (I), the colors of the upper asterisks indicate the comparison (e.g., blue asterisks indicate a comparison with the Ani+tBC group). Data are represented as mean ± SEM. Ani, anisomycin; tBC, tat-beclin.

To further confirm memory erasure, we tried to recover the erased memories by using optical long-term potentiation (LTP). High-frequency optical stimulation of the terminals of AC and MGm engram cells to the LA led to long-lasting potentiated field responses (fig. S3, A and B). Optical LTP allowed anisomycin-treated mice to completely recover from amnesia to the PBS group’s freezing level, which was specific to the 7-kHz conditioned tone (i.e., it did not generalize to the 2-kHz tone) (Fig. 1G). In the Ani+tBC-treated mice, optical LTP failed to completely recover the fear memory; these mice showed only a slight increase in the freezing level, which was similar to that which occurred in the unpaired control group (Fig. 1G). Because the unpaired conditioning did not form an associative fear memory, this slight increase in the freezing response might be attributed to the formation of a new artificial associative memory, rather than restoration of a previously stored associative memory.

In a remote memory test, the Ani+tBC group displayed significantly lower freezing than the anisomycin or PBS groups in both natural cue and optogenetic tests, indicating that memory erasure was long-lasting and that the memory did not undergo spontaneous recovery over time (Fig. 1, H and I). The Ani+tBC-treated mice that received LTP showed light-induced freezing comparable to that of the PBS group (test 9), excluding the possibility of LA damage from Ani+tBC treatment. Furthermore, similar results were obtained when engram cells in the LA were labeled and manipulated similarly but optical (instead of tone) recall was used (fig. S4).

To examine the synaptic mechanism underlying the complete retrograde amnesia, we conducted a LTP occlusion experiment, in which artificial induction of LTP was occluded in circuits with potentiated synapses, whereas it was facilitated in circuits with unpotentiated synapses (2729). High-frequency optical stimulation 1 day after test 1 induced LTP in the Ani+tBC group that was comparable to that in the nonshock group but significantly higher than that in the PBS and anisomycin groups (Fig. 2, A to D). Thus, synaptic plasticity was totally reset and returned to nonshock levels after complete amnesia.

Fig. 2 Resetting of synaptic plasticity and functional connectivity between engram cell assemblies as neural correlates of complete amnesia.

(A) Left, labeling strategy. Right, experimental design for the LTP occlusion experiment. (B) Freezing level during test 1. (C) Average of in vivo field excitatory postsynaptic potential slope (normalized to baseline) before and after LTP induction (two-way repeated-measures ANOVA; n = 4 mice per group). (D) Traces before (black) and after (red) optical LTP induction. (E) Left, labeling of engram cell assemblies in the AC, MGm, and LA using double transgenic mice (c-Fos::tTA/R26R::H2B-mCherry) (18). Right, experimental design. (F) Freezing levels during tests 1 and 2 (one-way ANOVA). (G) Freezing levels during test 3 (one-way ANOVA). (H) Representative images showing c-Fos+–mCherry+ overlap in the LA, indicated by arrowheads. Blue, 4′,6-diamidino-2-phenylindole (DAPI) staining. Scale bars, 50 μm. (I) c-Fos+–mCherry+ overlap cell counts (one-way ANOVA; n = 4 mice per group). Yellow lines represent chance level for each group. *P < 0.05; **P < 0.01. Data are represented as mean ± SEM.

This conclusion was further supported by analysis of functional connectivity. Using c-Fos::tTA/R26R::H2B-mCherry double transgenic mice (18), we measured the connectivity pattern between upstream and downstream engram cells after memory erasure. Engram cells in the LA were labeled with Cre-dependent mCherry, and the axonal terminals of the AC and MGm engram cells were labeled with oChIEF. The terminals were optogenetically stimulated at 10 Hz, and the number of cells that were double-positive for mCherry and c-Fos, which represented the degree of functional connectivity between upstream and downstream engram cells, was counted (Fig. 2, E to I, and fig. S7, A and B). Complete amnesia resulted in a significant decrease in the c-Fos+–mCherry+ overlap in the Ani+tBC group in comparison with the PBS and anisomycin groups, which is consistent with the behavioral data and the total resetting of synaptic efficacy.

Considering that memories are stored in interconnected networks, and the brain can store two memories in a shared ensemble (1418), we examined the effect of complete retrograde amnesia of one memory on another memory by using two different AFC events: a 7-kHz AFC (event 1) followed by a 2-kHz AFC (event 2) (Fig. 3). When these two events were separated by 5 hours, memory for event 2 was enhanced (fig. S5), indicating interaction between the memories (17). The majority of the LA engram cells for event 1 (mCherry+) also encoded event 2 (c-Fos+), whereas the memories were encoded in two distinct populations in the AC (Fig. 3, A to E, and fig. S7, C to F). When the two memories were separated by 24 hours, they were allocated to distinct populations in both the LA and the AC.

Fig. 3 Synapse-specific erasure of overlapping fear memories.

(A) Model for the neuronal ensemble in the LA and AC after two associative memories encoded with a 5-hour interval. Memories 1 and 2 respectively correspond to events 1 and 2. (B) Left, strategy to label engram cells in the AC and LA using double transgenic mice (c-Fos::tTA/R26R::H2B-mCherry) injected with AAV-TRE3G-Cre. Right, experimental design to check the overlapping ensembles between two associative memories that were encoded with different time intervals separating them. (C) Freezing level during the 7-kHz test session in the 7-kHz–7-kHz group. (D) Top, images for the overlapping ensembles, indicated by arrowheads, in the AC for different time intervals. Bottom, same as top but in the LA. Blue, DAPI staining. Scale bars, 50 μm. (E) Top, c-Fos+–mCherry+ overlap cell counts in the AC (one-way ANOVA; n = 4 mice per group). Bottom, same as in top but in the LA (unpaired t test; n = 4 mice per group). (F) Design for the selective memory erasure experiment. (G and H) Freezing levels for gp1 and gp2 during 7- and 2-kHz tones before and after drug injection (G) and during light-off and light-on epochs (H) (unpaired t test; n = 10 mice per group). T1, test 1; T2, test 2; and so forth. (I) Model for selective erasure of either 7-kHz-tone fear memory (red) or 2-kHz-tone fear memory (green). Overlapping ensembles are in yellow. *P < 0.05. Data are represented as mean ± SEM.

We then used the c-Fos::tTA transgenic mice to label the neural ensembles in the AC and MGm that were activated specifically during event 1 with oChIEF (Fig. 3F and fig. S6). After 5 hours ON DOX, mice were exposed to event 2 and then divided into two groups. The first group received PBS after event 1 memory retrieval and Ani+tBC after event 2 memory retrieval (gp1), whereas the second group received the opposite treatment (gp2). In gp1, memory of event 2 was erased by Ani+tBC (test 4), whereas memory of event 1 was not affected (test 2). In contrast, in gp2, event 1 memory was erased (test 2), whereas event 2 memory was not affected (test 4; Fig. 3G). Moreover, optogenetic stimulation of the presynaptic terminals of the AC and MGm engram cells corresponding to event 1 memory induced a freezing response in gp1, but not in gp2, although in both groups, the LA neurons storing both associative memories underwent Ani+tBC treatment (Fig. 3H). These results reveal synapse-specific engram erasure and indicate that memories stored in the shared engram cells are synapse-specific and have different fates (Fig. 3I).

We then addressed the question of how each memory reserves its individual identity within the shared ensemble. We carried out a loss-of-function experiment using optical long-term depression (LTD) to depotentiate the synaptic efficacy in synapses specific for event 1 memory (fig. S3, A and C, and Fig. 4, A and B). In comparison with a control group, mice that received LTD showed impairment in event 1 memory recall, but not in event 2 memory recall (Fig. 4C). Optogenetic stimulation to the terminals of the AC and MGm ensemble of event 1 memory triggered freezing in the control group, whereas it failed to trigger freezing in the LTD group, despite the fact that event 2 memory was intact (Fig. 4D). Thus, selective depotentiation of synaptic plasticity deconstructs the specific connectivity between engram assemblies, thereby erasing one memory without disrupting the other memory in the same population of neurons.

Fig. 4 Engram-specific synaptic plasticity is crucial and sufficient for information storage and keeps the identity of the overlapping memories distinct.

(A) Model for selective optogenetic targeting of synaptic plasticity with LTD. Memories 1 and 2 respectively correspond to events 1 and 2. (B) Design of the loss-of-function experiment. (C and D) Freezing levels in response to 7- and 2-kHz tones before and after optical LTD induction to event 1 memory–specific synapses (C) and in response to optical stimulation (D) (n = 10 mice per group). (E) Model for selective optogenetic targeting of synaptic plasticity with LTP. (F) Gain-of-function experiment. (G) Freezing levels before (tests 1 and 3) and after (tests 2 and 4) complete amnesia and after LTP induction to event 1 memory–specific synapses (tests 5 and 6) (n = 10 mice per group). Statistical comparisons were performed using two-way ANOVA. *P < 0.05. Data are represented as mean ± SEM.

Last, a gain-of-function experiment was performed in which both memories were erased with Ani+tBC and then optical LTP was induced in event 1 memory–specific synapses (Fig. 4, E and F). Mice that received the LTP protocol displayed higher freezing levels in response to the 7-kHz tone (test 5), whereas freezing responses to the 2-kHz tone (test 6) were unaffected (Fig. 4G).

Storing and distinguishing between several memories encoded in the same neurons are critically important for organizing unique memories. Our findings demonstrate that synapse-specific plasticity is necessary and sufficient for associative fear memory storage and that it guarantees uniqueness to the memory trace, pointing to plasticity as a substrate for the fear memory engram. This perspective is consistent with a recent observation that LTP is selectively induced in specific auditory pathways after fear memory formation (20).

Engram cells retain a memory after anisomycin-induced amnesia, and synaptic plasticity is dispensable for memory storage (21). However, synaptic plasticity and functional connectivity between engram cell assemblies are indispensable for fear memory storage, because after LTD induction, the depressed synapses might be nonfunctional. Therefore, not only the natural cue, but also the optical stimulation of synapses between the engram cell assemblies failed to retrieve the memory. Furthermore, the engram network no longer retained the associative fear memory after Ani+tBC-induced complete amnesia. The LTP occlusion experiment showed that synaptic potentiation persisted even 2 days after behavioral training in the PBS control group and that complete amnesia accompanied a reset of LTP. This further supports the idea that LTP is important for memory maintenance. The combined evidence suggests that synaptic plasticity can build a specific connectivity within the engram cell assemblies and that the functional connectivity is a simple reflection of the enhanced synaptic strength, rather than an independent mechanism for memory storage.

This study uncovered the mechanism by which the brain can maintain the uniqueness of a massive number of associated memories stored in shared cell ensembles. Furthermore, we achieved selective and total erasure of a fear memory from an engram network without affecting other memories stored in the shared ensemble by resetting the plasticity in a synapse-specific manner. These findings lead to a better understanding of the mechanisms underlying memory storage and may give insight into therapeutic approaches to treating post-traumatic stress disorder.

Supplementary Materials

Materials and Methods

Figs. S1 to S7

References (2931)

References and Notes

Acknowledgments: From the University of Toyama, we thank N. Ohkawa for his help in providing c-Fos::tTA mice, Y. Saitoh and M. Nomoto for their help with electrophysiology, and S. Tsujimura for maintenance of mice. We thank all members of the Inokuchi laboratory for discussion and suggestions. We also thank M. Ito and N. Takino (Jichi Medical University, Japan) for their help with production of the AAV vectors. Funding: This work was supported by a Grant-in-Aid for Scientific Research on Innovative Areas (“Memory dynamism”; JP25115002) from the Ministry of Education, Culture, Sports, Science, and Technology of Japan (MEXT); JSPS KAKENHI grant number 23220009; the Core Research for Evolutional Science and Technology (CREST) program (JPMJCR13W1) of the Japan Science and Technology Agency (JST); the Mitsubishi Foundation; the Uehara Memorial Foundation; and the Takeda Science Foundation (to K.I.). Additional support was provided by a Grant-in-Aid for young scientists from JSPS KAKENHI (grant number 25830007) to M.S. The Otsuka Toshimi Scholarship Foundation supported K.A. Author contributions: K.A., M.S., and K.I. designed the experiments. K.A., M.S., and K.I. wrote the manuscript. K.A., M.S., and K.C. performed the experiments. K.A., M.S., and K.I. analyzed the data. H.N. and M.M. produced and maintained transgenic mice. S.M. prepared AAVs. Competing interests: S.M. owns equity in a company, Gene Therapy Research Institution, that commercializes the use of AAV vectors for gene therapy applications. To the extent that the work in this manuscript increases the value of these commercial holdings, S.M. has a conflict of interest. Data and materials availability: All data are available in the main text or the supplementary materials.
View Abstract

Stay Connected to Science

Navigate This Article