Engram cells retain memory under retrograde amnesia

What was read

• Published Science report (PDF from the Tonegawa lab site): abstract, main text, Figs 1–5 legends, references. Read in full.
• NOT read: supplementary materials (methods, Figs S1–S13).

Question

Does memory consolidation happen inside the engram cells, and when a protein-synthesis inhibitor causes retrograde amnesia, is the memory gone or merely unretrievable?

Method

• c-Fos-tTA mice: cells active during contextual fear conditioning (CFC) in context B, off doxycycline, express a TRE-driven label (mCherry, or ChR2-EYFP for optogenetics) in the dentate gyrus (DG). Anisomycin (ANI) or saline (SAL) injected systemically immediately after training. Cycloheximide replicates ANI.
• Ex vivo electrophysiology: paired recordings of labelled and unlabelled DG cells while optogenetically stimulating ChR2-labelled entorhinal (perforant path) axons; AMPA/NMDA ratios; dendritic spine density from biocytin fills; DG-engram to CA3-engram connection probability.
• Behaviour: natural recall (freezing in context B on days 1 and 3), optogenetic recall (blue light on DG engram cells in neutral context A on day 2, four 3-min epochs), optogenetic place avoidance (OptoPA), CA1 engram at 4 Hz, lateral amygdala engram after tone fear conditioning, reconsolidation-blockade amnesia, false-memory "inception" onto an amnesic context engram, longevity to 8 days.
• Controls: no-shock engrams; ANI given 24 h after training (outside the consolidation window); DREADD silencing of CA1 during encoding; c-Fos counts in amygdala after natural vs light-induced recall; Arc+ counts to confirm ANI blocked protein synthesis; ChR2 labelling counts to confirm ANI did not block the label.

Results

1. Consolidation is engram-cell-specific. 24 h after CFC, labelled DG engram cells in saline mice have higher synaptic strength (EPSC amplitude, AMPA/NMDA ratio) and higher spine density than neighbouring unlabelled cells. In ANI mice, engram cells are indistinguishable from non-engram cells on both measures. A context-only experience produces the same engram-specific increases. ANI at 24 h leaves them intact.
2. Amnesia. ANI mice freeze much less than saline mice in context B on day 1 (P < 0.005) and still on day 3.
3. The memory is still there. Blue light on DG engram cells in context A produces freezing in ANI mice equal to saline mice (Fig 2D). Same result for place avoidance, for the CA1 engram at 4 Hz, for the lateral amygdala engram after tone fear conditioning, for amnesia induced by blocking reconsolidation after a retrieval (Fig 4A), and 8 days later. Encoding-disrupted (DREADD) mice show no rescue, so the rescue depends on a memory having been formed.
4. Downstream reactivation. Natural cues reactivate fewer amygdala cells in ANI mice; light activation of DG engram cells reactivates CA3 and BLA engram cells equally in both groups. DG-engram to CA3-engram connection probability is above chance and unaffected by ANI: "preferential and protein synthesis–independent functional connectivity" between engram cells.
5. The amnesic engram can be used: pairing light activation of an ANI-treated context-A engram with shock creates a context-A-specific false fear memory in both groups ("inception").

Authors' interpretation: augmented synaptic strength "is not a crucial component of the stored memory"; the specific engram-to-engram connectivity pattern "may represent a fundamental mechanism of memory information storage"; the role of synaptic potentiation after consolidation may be "to provide natural recall cues with efficient access to the soma of engram cells". Amnesia here is a retrieval deficit, not trace loss.

Limits

• Mouse contextual and tone fear; one-trial memories; days-old.
• Cells were tagged during learning by an activity-dependent transgene. Rescue requires knowing and owning the engram cells in advance. No method here locates an untagged memory.
• Systemic anisomycin during the consolidation window; the same lab and others later showed stronger protocols can erase (see [1] Abdou 2018, anisomycin plus autophagy induction: no rescue).
• "Storage in connectivity" is inferred from connection probability between labelled populations, not from a complete wiring map, and is stated as a proposal.
• Supplementary methods and figures not read.

What the brief uses it for, and whether it holds

Brief section 4: "Direct activation of cells tagged during learning elicited the learned response after normal retrieval had been impaired: retrieval failure need not mean trace loss." Limit: "Cells were tagged during learning; no procedure finds an arbitrary unknown memory in an unfamiliar brain."

• Holds exactly, and the limit is the right one.
• Two things worth registering for the brief's argument:
  • Tension with [1]. Ryan et al. say augmented synaptic strength is not the storage substrate and the engram connectivity pattern may be; Abdou et al. say synapse-specific plasticity "guarantees the identity and storage of individual memories" and that with a stronger amnesia protocol the engram "no longer existed". The brief cites [1] for "content is in plasticity, not in which cells" and [7] for "retrieval failure is not trace loss". Both citations are accurate to their papers, but the two papers disagree about where the trace lives, and [7] is the most pro-connectivity statement in the evidence base. Section 2's "no decoder from structure to content" survives either way; "a bare connectivity map would not close the gap even in principle" is stronger than [7] supports, since [7] proposes that a specific connectivity pattern is the store.
  • Support for 6.3 in a form the brief does not use: Fig 4A shows that a retrieval followed by protein-synthesis inhibition produces amnesia that is still optogenetically rescuable. In this model, elicitation-induced disruption affects access, not the trace. That is a reason to frame the perturbation audit around retrievability and first-response drift, not around "erasure".
• Relevance to Variant A: the paper is sometimes read as showing memory survives in structure. What it shows is that a living engram, with intact connectivity, can be driven artificially. Nothing here supports reading a memory out of tissue without the tags and the light.

Cross-references

• [1] Abdou 2018 cites this paper (ref 21) and is designed as the complete-amnesia counterpart.
• [8] Suzuki 2004 is cited here (ref 24) for the reconsolidation window and for ANI at 24 h not causing amnesia.