KurisutinaCitation: Sadegh Nabavi, Rocky Fox, Christophe D. Proulx, John Y. Lin, Roger Y. Tsien, Roberto Malinow. Engineering a memory with LTD and LTP. Nature 511, 348–352. DOI. Published online 1 June 2014; issue 17 July 2014. Peer-reviewed research article.
Reading and source scope: Read the complete main narrative, methods, main-figure legends, and Extended Data 1–10 legends in the PMC author manuscript, cross-checked against the author-hosted publisher-formatted PDF. The PDF has advance-online pagination and contains the methods and ten extended figures. The downloaded versions are distinguished in provenance. No separate supplementary file was linked on the inspected publisher page. No raw-data reanalysis or independent numerical digitization was performed. The article is publicly available for research access; no Creative Commons license was identified.
Evidence, concise paraphrase: Male rats learned to associate optical stimulation of auditory inputs to lateral amygdala with shock. Subsequent low-frequency stimulation suppressed the conditioned response; high-frequency stimulation restored it, including repeated reversals in the main twelve-rat experiment. Electrophysiology supported depression and potentiation of the targeted pathway. Crucially, potentiation did not produce the response in naïve animals, and it failed to reverse behavioral extinction. The primary behavioral measure was suppression of previously trained lever pressing, supported by freezing measurements. Advancement to later behavioral phases required a stable initial conditioned response. The authors explicitly allow additional learning-induced modifications beyond the manipulated synapses: their conclusion is necessity without sufficiency of potentiation at this input. This is causal evidence connecting pathway efficacy to learned-response expression; it does not decode event details, locate the complete information store, or transfer a memory to another animal or computational system. Original article.
Original scientific reasoning: A successful reversible intervention identifies a control point, but identifying a control point and identifying all information required to reconstruct a memory are different inverse problems. Consider two candidate mechanisms: changing associative content itself, or changing access from a cue to content retained elsewhere. If both predict the measured response sequence, that sequence alone does not identify which mechanism is correct. Additional outcomes and interventions must separate them. It is therefore premature to force a choice between a distributed network account and a synaptic efficacy account; they can describe different levels of one system.
For an Amadeus-like system, the transferable lesson is a requirement to distinguish stored evidence, cue access, and behavioral expression. It is not a requirement to reproduce the same synaptic implementation. Likewise, restoring one conditioned output is much weaker than restoring a person's counterfactual judgments or learning across new circumstances.
Experimental discrimination proposed here: In a computational memory system, separately perturb the content representation, cue-to-memory access, and output policy. Evaluate independently cued content, alternate behavioral reports, and downstream learning. In biological evidence reviews, favor experiments with multiple independent content-sensitive outcomes and crossed retrieval routes. These are design proposals; this paper did not run these comparisons.
Next comparison: Read alongside Roy 2022 and the existing Ryan 2015 and Abdou 2018 summaries. Ask which intervention changes a stored distinction, which blocks its expression, and which available observation actually separates those explanations.