KurisutinaGray Umbach, Pranish Kantak, Joshua Jacobs, Michael Kahana, Brad E. Pfeiffer, Michael Sperling, and Bradley Lega. Time cells in the human hippocampus and entorhinal cortex support episodic memory. PNAS 117(45), 28463–28474. Published online 27 October 2020; issue date 10 November 2020. Peer-reviewed human intracranial study. PMC, DOI: 10.1073/pnas.2013250117.
Evidence from this source
Twenty-seven epilepsy patients contributed 40 sessions of word-list encoding, distraction, and free recall. Among 509 selected hippocampal/entorhinal putative pyramidal neurons, 79 showed encoding time fields and 60 retrieval time fields. Sixteen belonged to both groups; their encoding/retrieval peak times did not significantly correlate. Greater stability of encoding time fields was associated with stronger temporal clustering of recalled words, but not significantly with overall recall. Fields described current task-epoch time, not recovered dates of remembered events. Ramping activity also tracked longer intervals. These are neural correlates of temporal organization, not memory transfer or demonstrated causal necessity. Electrode placement was clinical; word-list behavior may not generalize to autobiographical events. Authors treated sessions as independent, while some behavioral comparisons used cells sharing session-level outcomes. The study does not fully separate elapsed time, serial position, task expectation, and other changing context.
Original interpretation for memory extraction and functional transfer
The engineering target should separate what is available from how retrieval moves among available items. A system may list every event correctly but choose the wrong next recollection because its retrieval process lacks the person's temporal and contextual organization. Timestamps in a database can provide useful evidence, but their presence alone does not show that the model uses temporal context as the person does.
This evidence supplies a reason to test retrieval trajectories, not a ready-made decoder of chronology. In particular, neurons sensitive to time during retrieval need not replay the original sequence of experience. The relevant latent state may partly concern the current search process. Calling both kinds of signal “memory time” would obscure the extraction problem.
The statistical dependency also limits interpretation: multiple neurons and sessions from a participant cannot automatically stand for independent people. Before treating the reported association as a precise person-level effect, a reanalysis should account for participant/session nesting. This concern qualifies the behavioral association without making the existence of temporally modulated cells disappear.
For functional transfer, preserving a prediction of temporal contiguity could be useful without reproducing any particular neuron. Whether that representation remains useful across hours, years, emotional events, or deliberate narrative reconstruction is a separate empirical question.
Proposed discriminating experiment
Collect free-recall sequences for prospectively recorded events, retaining timing, cue, context changes, omissions, and intrusions. Compare a generic semantic retriever, recency/serial-position models, a temporal-context model, and an equal-information LLM. Score held-out next-recollection distributions and lag-dependent transitions while controlling for which items remain available. Hold out complete sessions and cue families.
For controlled episodes, vary presentation spacing independently of order. Separately change the retrieval context while preserving the encoded inventory. These manipulations could distinguish a useful temporal representation from simple ordinal position or semantic similarity. Neural recording would be an optional incremental observation, not required to establish this behavioral target. A copied event list and a copied retrieval process should receive different success labels.
Reading and acquisition record
Read the full main introduction, results, discussion, conclusion, materials and methods, figure captions, and data statement from the public Europe PMC XML. Supplementary figures/tables and code were not read; no raw-data or statistical reanalysis was performed. The title's causal-sounding language is stronger than the observational design alone warrants.
Local sources: Europe PMC XML, readable text, provenance and SHA-256 hashes. The public full-text XML is the retained primary source; attempted university PDF downloads failed and are not counted as acquired. The source states CC BY-NC-ND 4.0. The readable text is a local access copy derived from the XML; original structure and rights statement remain in the XML.