KurisutinaCitation: Dheeraj S. Roy, Young-Gyun Park, Minyoung E. Kim, Ying Zhang, Sachie K. Ogawa and colleagues. Brain-wide mapping reveals that engrams for a single memory are distributed across multiple brain regions. Nature Communications 13, 1799. DOI. Version of record published 4 April 2022; peer-reviewed, CC BY 4.0.
Reading and source scope: Read complete main introduction, results, discussion, methods, and figure legends from the PMC full text, with the publisher PDF retained. Read every supplementary figure/table legend; visually inspected Supplementary Figures 4–6, including the random-label and combined activation/inhibition controls. Read all three reporting-summary pages, with OCR checked against rendered pages. Tables 1–3 are image-based: their captions and criteria were read, but individual regional rows were not exhaustively transcribed or reanalyzed. Supplementary movies, raw imaging, public analysis code, and the source-data workbook were not examined. Exact artifacts, hashes, and versions are in provenance.
Evidence, concise paraphrase: In male mice, contextual fear conditioning and recall were mapped across 247 brain regions. Activity-based screening yielded 117 candidate regions; a separate overlap assay identified 88 regions with above-chance reactivation. These counts are screening results, not independently validated content stores. Activating learning-tagged ensembles in nine tested regions induced freezing, unlike random-label controls. CA1 and basolateral amygdala inhibition reduced natural recall; supplementary crossed manipulation showed that inhibiting the latter reduced freezing induced by CA1 activation. Activating several tagged ensembles together increased freezing toward natural-recall levels. Broad activity tagging and a behavioral endpoint support distributed participation without resolving every region's informational role. The mapping cannot identify silent ensembles that fail natural reactivation, and the authors discuss possible arousal or motor contributions. No episode-content decoder, synapse-by-synapse storage map, or transfer between organisms was demonstrated. Original article.
Original scientific reasoning: Keep three claims distinct: a population is associated with learning; manipulating it changes a memory-dependent behavior; its local state contains a particular recoverable component of the remembered content. The first two can motivate the third but do not logically entail it. A distributed system may contain information stores, routing mechanisms, motivational state, and response machinery, all causally important for one observed behavior. This interpretation preserves the intervention evidence while avoiding the assumption that every causally effective site holds an independently readable copy of an episode.
For functional transfer, the key question becomes which jointly measured variables preserve distinctions between experiences under new probes. Anatomical coverage is not a measure of recovered information. Conversely, success at one behavioral output cannot specify how much of a distributed representation was restored. These distinctions also explain why Nabavi's pathway manipulation and distributed ensembles need not be competing accounts.
Experimental discrimination proposed here: A stronger content-transfer test would use several learned episodes sharing valence and motor response, then require intervention-specific predictions of episode identity, context, and new consequences. Compare selective restoration with matched nonspecific changes in arousal or response readiness. A computational recipient should generalize across those tests from the transferred representation, without receiving the answer labels at test time. This proposal is not an accomplished biological-to-digital transfer.
Statistical interpretation: Treat the regional screen as hypothesis generation. A follow-up should specify correction across all screened regions and stimulation settings, independently validate selection, and quantify uncertainty at the animal level. The paper reports within-region comparisons and selected behavioral validation; this note does not recompute its regional false-discovery rate.