Synapse-specific representation of the identity of overlapping memory engrams

What was read

• Full supplementary materials from CaltechAUTHORS (Materials and Methods, Figs S1–S7 legends, 31 references). Read in full.
• Abstract and erratum record (PubMed 29903972, 30072514).
• JST/University of Toyama press release (ScienceDaily, 14 July 2018), which paraphrases the four main results.
• NOT read: the main text and Figs 1–4 legends. Science 2018 is paywalled; no repository copy exists (Unpaywall, OpenAlex, CORE, Semantic Scholar, Toyama and NII repositories all checked). The erratum's content is unknown. Claims below are labelled by source.

Question

Two memories learned close in time are encoded in overlapping neuronal ensembles (memory linking). If the cells are shared, what keeps each memory's identity separate, and where is each memory actually stored?

Preparation (verified, SM)

• c-Fos::tTA mice, off doxycycline for 2 days around conditioning, so neurons active during a conditioning event express Cre from a TRE3G promoter and switch on a Cre-dependent oChIEF-Citrine channelrhodopsin. Labelled cells are the "engram" for that event. Time window closed by 1000 mg/kg DOX chow 6 h after conditioning (1 h plus a DOX injection in the two-memory design).
• Virus injected into auditory cortex (AC), medial geniculate (MGm) and lateral amygdala (LA). Optical fibres over LA stimulate the tagged AC/MGm axon terminals synapsing onto LA neurons: the tone pathway of auditory fear conditioning.
• Auditory fear conditioning: 7 kHz tone co-terminating with a 0.4 mA foot shock; 2 kHz tone as neutral tone or as a second conditioned tone. Fig S1: naïve mice discriminate 2 kHz from 7 kHz. Fig S2: the two tones activate different AC ensembles.
• Optical plasticity protocols: LTP = 10 trains of 100 pulses at 100 Hz, 90 s apart; LTD = 900 pulses at 1 Hz. Fig S3: in vivo field EPSP recordings confirm the protocols potentiate and depotentiate the AC/MGm-to-LA synapses.
• Amnesia agent: anisomycin (protein synthesis inhibitor) plus tat-beclin 1 (autophagy inducer), infused into LA immediately after a retrieval test. This is the lab's "complete amnesia" protocol (Shehata et al. 2018, J Neurosci, "Autophagy enhances memory erasure through synaptic destabilization").
• Two-memory design: 7 kHz conditioning, then 2 kHz conditioning 5 h later (overlapping ensembles) or 24 h later (separate ensembles). Fig S5: the 5 h group shows enhanced freezing, the behavioural signature of linking. Fig S6: 5 h on DOX suffices to stop labelling, so the second memory's cells are not tagged.
• Freezing scored by video tracking; n = 4–11 mice per group across experiments.

Results (abstract and press release; figure numbering from SM cross-references)

1. After complete retrograde amnesia of the 7 kHz memory (anisomycin + tat-beclin after retrieval), optogenetic 10 Hz and 20 Hz stimulation of the tagged AC/MGm terminals in LA no longer produced freezing, and the tone did not either, including at a remote time point (Fig S4 C–H). Press release: this "correlated with the resetting of synaptic strength and functional connectivity between engram assemblies". Interpretation in the abstract: the engram "no longer existed in that circuit".
2. Fig S4 and Fig 1 also used optical LTP to restore or occlude: an LTP occlusion experiment (5 trains at 3 min intervals, then in vivo recording) tests whether conditioning had already saturated the synapses.
3. With two memories 5 h apart in a shared LA ensemble, complete amnesia of the 7 kHz memory left the linked 2 kHz memory intact (Figs 3, S5–S7). Cell-count controls (Fig S7) show labelling efficiency did not differ between groups.
4. Optical LTD at the synapses tagged for one memory abolished recall of only that memory; optical LTP at those synapses affected only that memory (Fig 4). The other memory, held by the same cells, was untouched.

Conclusion in the authors' words: "the sharing of engram cells underlies the linkage between memories, whereas synapse-specific plasticity guarantees the identity and storage of individual memories."

Limits

• Mouse auditory fear conditioning, one synapse population (AC/MGm terminals onto LA), one-trial tone-shock memories. Nothing about episodic or declarative content.
• "Engram no longer exists" is inferred from failure of optogenetic recall after an aggressive two-drug protocol; it is the opposite outcome from Ryan et al. 2015 [7], where anisomycin alone left an optogenetically recoverable trace. The two papers together say: the trace lives in synaptic potentiation, cells can survive with the trace erased, and whether "amnesia" is retrieval failure or trace loss depends on how far the synapses were reset.
• Only the tagged input synapses are manipulated; the paper does not claim the whole memory is at those synapses, only that its identity within the shared ensemble is.
• Main text unread; effect sizes and exact statistics for Figs 1–4 are not verified here.

What the brief uses it for, and whether it holds

Brief section 2: "in mice, the distinct identities of overlapping memories have been tied to memory-specific synaptic plasticity rather than to which neurons participate [1]", supporting the claim that a bare connectivity map is not a description of what the system encodes.

• The citation is accurate to the paper's central claim: identity is carried by which synapses are potentiated, not by cell membership, and two memories in identical cells were separately erasable.
• One caveat for the brief's argument: the paper is about synaptic efficacy, which has structural correlates (spine volume, PSD size) that an EM connectome with synapse sizes would partly capture. The paper says nothing about whether structure could or could not reveal content. So [1] supports "cell list is not content" cleanly, and supports "wiring diagram is not content" only for a map without synaptic weights. The brief's word "bare" is doing the work; keep it.
• The paper is also relevant to brief section 6.3 (extraction can change the source), which cites only Suzuki [8]: here, retrieval followed by a pharmacological intervention produced complete erasure of one memory while sparing a linked one. Retrieval opens the window; whether anything happens depends on what follows.

Cross-references inside the evidence base

• Cites Ryan et al. 2015 [7] as ref 21 and is the lab's counterpoint to it (complete vs incomplete amnesia).
• Builds on Yokose et al. 2017 (Science 355:398) and Cai et al. 2016 (Nature 534:115) for memory linking via shared ensembles; Nabavi et al. 2014 (Nature 511:348) for optical LTD/LTP erasure and reinstatement of a fear memory.

Open items

• Obtain the main text if a licensed copy becomes available; check the erratum (Science 361:eaau8829, 3 Aug 2018) for which figure or statistic was corrected.