Ensembles of human MTL neurons “jump back in time” in response to a repeated stimulus

Ensembles of human MTL neurons “jump back in time” in response to a repeated stimulus Howard Viskontas Shankar Fried

Tag layer of the DHSS Meta-Tagging Project. The article text is reproduced from the staged copy named in sourceDesc and is not altered; the tags are held separately in standOff so that overlapping and nested spans are preserved.

Ensembles of human MTL neurons “jump back in time” in response to a repeated stimulus 2012 Hippocampus 10.1002/hipo.22018 Howard2012-ensembles-of-human-mtl-neurons-jump-back-in Howard2012-ensembles-of-human-mtl-neurons-jump-back-in.txt, the project's staged copy e6ed050a3ecdca9af6b79a174ad46f89639241fe not-redistributable neuroscience

the paper's own words

the tag layer; a sentence written by the tagger about the paper, always anchored to a quotation but not itself a quotation

0 spans located by searching the stored text for the quote's exact words, allowing any whitespace between them whether a repeated stimulus makes a neural ensemble return to the state it was in BEFORE that stimulus was first seen Quotation withheld because this record is not cleared for redistribution. whether cognitive models of retrieved temporal context can be told apart from short-term memory models at the level of human MTL units Quotation withheld because this record is not cleared for redistribution. repetition of a stimulus elicited an ensemble pattern resembling the pattern present before that stimulus was first presented, confirming a direct prediction of temporal context models Quotation withheld because this record is not cleared for redistribution. existence Quotation withheld because this record is not cleared for redistribution. neural contiguity effect in episodic memory ("jump back in time") Quotation withheld because this record is not cleared for redistribution. electrophysiology electrophysiology Quotation withheld because this record is not cleared for redistribution. mechanisms A repeated stimulus triggers pattern completion of the gradually-changing context state that was active when it was first experienced, so the recovered state resembles states both before and after the original presentation Quotation withheld because this record is not cleared for redistribution. mechanisms The contiguity effect is produced by a jump back in time in which the repeated item recovers the pre-presentation state of temporal context Quotation withheld because this record is not cleared for redistribution. mechanisms Long-timescale temporal autocorrelation is generated by network properties or intrinsic cellular mechanisms that sustain firing after a transient input Quotation withheld because this record is not cleared for redistribution. intrinsic cellular or network properties could produce persistent firing after a transient input, which is what a slowly changing representation would need Quotation withheld because this record is not cleared for redistribution. recurrent CA3 connectivity could complete a gradually-changing pattern from a partial cue, which would recover the earlier context Quotation withheld because this record is not cleared for redistribution. a repeated stimulus recovers a state that is a mixture of earlier inputs weighted by temporal proximity, rather than reinstating its own original input Quotation withheld because this record is not cleared for redistribution. The data cannot establish whether the ensemble change caused by a stimulus's first presentation was specific to that stimulus. what this limits: what the ensemble change means: WE CANNOT DETERMINE WITH CERTAINTY FROM THESE DATA whether the change was specific to the stimulus or general Quotation withheld because this record is not cleared for redistribution. The experiment cannot bound the longest time scale over which the ensemble representation is autocorrelated. what this limits: the upper end of the timescale: THE RESULTS FROM THIS EXPERIMENT DO NOT PLACE A STRONG CONSTRAINT ON THE UPPER LIMIT OF THE TIME-SCALE OF THE AUTOCORRELATION Quotation withheld because this record is not cleared for redistribution. Averaging firing rates over 3 s windows makes any faster regional differences undetectable. what this limits: any comparison between regions: WE AVERAGED OVER 3 s INTERVALS SO THAT DIFFERENCES BETWEEN REGIONS THAT MANIFEST ON SHORTER TIME SCALES WOULD BE INVISIBLE TO US Quotation withheld because this record is not cleared for redistribution. The blocked design leaves some variables (position in block, distribution of recencies, new/old ordering) uncontrolled. what this limits: every contrast in the blocked design: IT WOULD BE IMPOSSIBLE TO COMPLETELY CONTROL ALL POSSIBLE RELEVANT VARIABLES, so position in block, recency distribution and new/old ordering ride along Quotation withheld because this record is not cleared for redistribution. The analyses do not discriminate among mathematical forms of the gradual change. what this limits: the functional form of the change: OUR DATA ANALYSES ASSUME THAT THE ENSEMBLE CHANGES OVER TIME BUT THEY DO NOT UNIQUELY SUPPORT A PARTICULAR MATHEMATICAL FORM OF THE CHANGE Quotation withheld because this record is not cleared for redistribution. Recency and contiguity had to be disentangled statistically rather than by design. what this limits: the recency-versus-contiguity separation, done statistically rather than by design: FUTURE RESEARCH SHOULD DISENTANGLE RECENCY FROM CONTIGUITY BY, FOR INSTANCE, PRESENTING REPEATED STIMULI AFTER A LONG DELAY Quotation withheld because this record is not cleared for redistribution. the mechanistic account of how the jump back in time is accomplished is labelled necessarily speculative what this limits: the mechanistic account that follows: WHILE NECESSARILY SPECULATIVE, IT IS WORTH THINKING ABOUT HOW THESE TWO FUNCTIONS MIGHT BE ACCOMPLISHED Quotation withheld because this record is not cleared for redistribution. the data cannot settle what exactly is retrieved when the ensemble state is recovered what this limits: the same scope as limitations[0] - what exactly is retrieved is undetermined Quotation withheld because this record is not cleared for redistribution. firing was averaged over three-second intervals, so regional differences on shorter timescales could not be seen what this limits: the same scope as limitations[2] - the 3 s averaging hides faster regional differences Quotation withheld because this record is not cleared for redistribution. some design variables cannot be controlled at all: the first stimulus of a block is necessarily new and the last necessarily old what this limits: the blocked design: IT IS NECESSARILY THE CASE THAT THE FIRST STIMULUS SHOWN IN A BLOCK IS NEW AND THE LAST STIMULUS IN THE BLOCK IS OLD Quotation withheld because this record is not cleared for redistribution. autocorrelated units did not show robust stimulus coding by this method, only category specificity what this limits: the stimulus-coding claim: ROBUST STIMULUS CODING WAS NOT OBSERVED USING THIS METHOD IN THE TEMPORALLY-AUTOCORRELATED PART OF THE ENSEMBLE, t(15) = 0.46 - only category specificity was Quotation withheld because this record is not cleared for redistribution. retrieved temporal context, brought from list-memory modelling to a prediction about single units Quotation withheld because this record is not cleared for redistribution. recency and contiguity as the two effects any account must produce Quotation withheld because this record is not cleared for redistribution. the analysis restricted to pairs of events WITHIN THE SAME BLOCK, which were necessarily from the same category eliminates: category membership as an artefactual source of contiguity - events from one category cluster in time by construction, so a temporal-contiguity effect could be category structure wearing a clock · outcome (stated by the tagger, no quotation): held Quotation withheld because this record is not cleared for redistribution. contiguity effect contiguity effect Quotation withheld because this record is not cleared for redistribution. category → event number THE TWO CANNOT BE SEPARATED IN THIS STUDY - the DESIGN form of the negative relation / No edge type can express it: all eight assert that a relation HOLDS. Recorded here because a confounding statement BOUNDS WHAT THE PAPER'S OWN RESULTS CAN MEAN, and it lives in a methods paragraph where nothing downstream will find it. A design confound stated plainly in a methods sentence. Quotation withheld because this record is not cleared for redistribution. THE TWO RIVALS ARE MADE TO AGREE EVERYWHERE BUT ONE PLACE: both predict a gradually changing state caused by the stimuli, and the paper says so before testing anything Quotation withheld because this record is not cleared for redistribution. and the discriminating prediction is an ASYMMETRY: short-term memory predicts resemblance only to states AFTER the original presentation, a jump back in time predicts resemblance in both directions Quotation withheld because this record is not cleared for redistribution. a jump back in time is defined as recovering the context that existed BEFORE the stimulus was first presented, not as reinstating the stimulus's own input Quotation withheld because this record is not cleared for redistribution. temporally-autocorrelated is an operational label: units whose lag-one regression coefficient on the previous event's firing rate is significantly above zero, about a quarter of all units Quotation withheld because this record is not cleared for redistribution. each unit is treated as an independent observation and as a one-cell ensemble, because unit counts vary across sessions and sessions were at least 48 hours apart Quotation withheld because this record is not cleared for redistribution. the design was chosen over free recall precisely because repetition order is under experimenter control rather than following the subject's own retrieval Quotation withheld because this record is not cleared for redistribution. claim_type: existence a second reading reaching the same value. Recorded: existence — note: a second reading reaching the same value from a different sentence. Not applied - a person decides. Quotation withheld because this record is not cleared for redistribution. method: microwire single-unit recording during a continuous recognition task, with ensemble similarity analysed as a function of recency and of lag from the original presentation a second reading DIFFERING from the recorded value. Recorded: electrophysiology — note: a second reading, verbatim-verified, disagreeing with the value on the record. Not applied - a person decides. Quotation withheld because this record is not cleared for redistribution. phenomenon: temporal autocorrelation of MTL ensemble firing over tens of seconds, with a contiguity effect superimposed on it a second reading DIFFERING from the recorded value. Recorded: neural contiguity effect in episodic memory ("jump back in time") — note: a second reading, verbatim-verified, disagreeing with the value on the record. Not applied - a person decides. Quotation withheld because this record is not cleared for redistribution. builds-on → retrieved temporal context models of episodic memory Quotation withheld because this record is not cleared for redistribution. extends → Manning et al. (2011) Quotation withheld because this record is not cleared for redistribution. contradicts → Norman and O'Reilly (2003) Quotation withheld because this record is not cleared for redistribution. contradicts → accounts identifying the MTL ensemble with persistent stimulus-specific activation in short-term memory Quotation withheld because this record is not cleared for redistribution. contradicts → short-term memory accounts in which contiguity comes from persistent stimulus activation and direct inter-item associations Quotation withheld because this record is not cleared for redistribution. supports → retrieved temporal context models of recency and contiguity Quotation withheld because this record is not cleared for redistribution. extends → Manning 2011's finding that recall reinstates the pattern present during study of preceding items status: the paper's own words: is not clear Quotation withheld because this record is not cleared for redistribution. contradicts → models in which the gradually changing memory representation is independent of the stimuli presented Quotation withheld because this record is not cleared for redistribution. builds-on → evidence for temporal autocorrelation and stimulus selectivity in MTL neurons Quotation withheld because this record is not cleared for redistribution. jump back in time —distinct-from→ short-term memory jump back in time short-term memory strength of the relation: 1 · note: AND THEY DIFFER IN EXACTLY ONE PLACE: short-term memory predicts resemblance only FORWARD of the original presentation, a jump back predicts it in both directions Quotation withheld because this record is not cleared for redistribution. jump back in time —part-of→ temporal context jump back in time temporal context strength of the relation: 1 · note: the recovered state is a mixture of earlier inputs weighted by proximity, NOT a reinstatement of the stimulus's own input Quotation withheld because this record is not cleared for redistribution. block —distinct-from→ temporal autocorrelation block temporal autocorrelation strength of the relation: 1 · note: and the block structure is removed by restricting every comparison to pairs of events inside one block Audit 2026-09-28: DOUBTFUL - read against its span and kept, but the span isolates autocorrelation from block effects; distinct-from is loose. Quotation withheld because this record is not cleared for redistribution. recency —causes→ ensemble similarity recency ensemble similarity strength of the relation: 1 · note: and the recency gradient is then ESTIMATED and SUBTRACTED, so what remains is the contiguity effect alone Audit 2026-09-28: DOUBTFUL - read against its span and kept, but the span is a counterfactual ('if recency were the only factor'); causes states it as a finding. Quotation withheld because this record is not cleared for redistribution. persistent firing —causes→ temporal autocorrelation persistent firing temporal autocorrelation strength of the relation: 0.9 Audit 2026-09-28: DOUBTFUL - read against its full span and kept, but the autocorrelation 'COULD RESULT FROM' properties leading to persistent firing; causes is hedged. Quotation withheld because this record is not cleared for redistribution. pattern completion —causes→ jump back in time pattern completion jump back in time strength of the relation: 0.8 · note: offered as the recovery mechanism, from the recurrent structure of CA3 Audit 2026-09-28: DOUBTFUL - read against its full span and kept, but CA3's structure 'COULD BE USED FOR' pattern completion to recover the state; causes is hedged. Quotation withheld because this record is not cleared for redistribution. hippocampus —contrasted-with→ amygdala hippocampus amygdala strength of the relation: 1 · note: A REGIONAL NULL: hippocampus, parahippocampal region and entorhinal cortex all far exceed the Type I error rate and do not differ from each other; the amygdala does not exceed it at all Audit 2026-09-28: DOUBTFUL - read against its span and kept, but the span is cut before the regional contrast is stated. Quotation withheld because this record is not cleared for redistribution. continuous recognition —operationalised-as→ jump back in time continuous recognition jump back in time strength of the relation: 1 · note: chosen over free recall BECAUSE the repetition order is set by the experimenter rather than by the subject's retrieval Audit 2026-09-28: DOUBTFUL - read against its full span and kept, but the continuous recognition experiment gives 'a better opportunity to assess' the jump back in time; operationalised-as is loose. Quotation withheld because this record is not cleared for redistribution. recency effect —contrasted-with→ contiguity effect recency effect contiguity effect strength of the relation: 1 · note: the two behavioural effects any account must produce, and here they appear superimposed in one curve Quotation withheld because this record is not cleared for redistribution. linear regression linear regression Quotation withheld because this record is not cleared for redistribution. t-test t-test Quotation withheld because this record is not cleared for redistribution. paired t-test paired t-test Quotation withheld because this record is not cleared for redistribution. chi-squared test chi-squared test Quotation withheld because this record is not cleared for redistribution. ANOVA ANOVA Quotation withheld because this record is not cleared for redistribution. z-score standardisation z-score standardisation Quotation withheld because this record is not cleared for redistribution. lag one regression coefficient lag one regression coefficient Quotation withheld because this record is not cleared for redistribution. Type I error rate Type I error rate Quotation withheld because this record is not cleared for redistribution. paired t-test paired t-test Quotation withheld because this record is not cleared for redistribution. MTL ensemble activity changed gradually over tens of seconds, and the change was causally affected by which stimulus was shown. Quotation withheld because this record is not cleared for redistribution. Repeating a stimulus recovered ensemble activity that had been present before its original presentation — a reliable backward contiguity effect. Quotation withheld because this record is not cleared for redistribution. Ensemble similarity also fell off in the forward direction, which is consistent with either short-term memory or retrieved-context accounts. Quotation withheld because this record is not cleared for redistribution. Ensemble similarity was greater for same-category than for different-category pairs. Quotation withheld because this record is not cleared for redistribution. The temporally autocorrelated part of the ensemble did not show robust stimulus coding. Quotation withheld because this record is not cleared for redistribution. The contiguity analysis gives very strong evidence against all of the temporal autocorrelation being artifactual. Quotation withheld because this record is not cleared for redistribution. THE RIVALS SHARE TWO PREDICTIONS BY CONSTRUCTION: both require a gradually changing state caused by the presented stimuli Quotation withheld because this record is not cleared for redistribution. AND DIFFER IN ONE ASYMMETRY: short-term memory predicts resemblance only to states following the original presentation, a jump back in time predicts both directions Quotation withheld because this record is not cleared for redistribution. so the signature sought is a repeated stimulus recovering activity present BEFORE its original presentation Quotation withheld because this record is not cleared for redistribution. lag one regression coefficients were above zero overall and about a quarter of units qualified as temporally autocorrelated Quotation withheld because this record is not cleared for redistribution. autocorrelated units fired above baseline BEFORE the stimulus appeared, so they were not simply responding to it Quotation withheld because this record is not cleared for redistribution. hippocampus, parahippocampal region and entorhinal cortex all far exceeded the Type I error rate; the amygdala did not Quotation withheld because this record is not cleared for redistribution. BLOCK STRUCTURE REMOVED BY RESTRICTION: all comparisons confined to pairs of events within one block, which are necessarily same-category Quotation withheld because this record is not cleared for redistribution. A LOWER BOUND ON THE TIMESCALE: excluding the eight nearest recencies still left a significant slope, so the ensemble kept changing past 27 seconds Quotation withheld because this record is not cleared for redistribution. CATEGORY AND RECENCY SEPARATED BY MATCHING: comparisons made across pairs of similarities at the same recency Quotation withheld because this record is not cleared for redistribution. so the category effect cannot be attributed to the confounding of recency with category structure Quotation withheld because this record is not cleared for redistribution. A DISSOCIATION INSIDE THE ENSEMBLE: the autocorrelated units showed category specificity but not robust stimulus coding, t = 0.46 Quotation withheld because this record is not cleared for redistribution. while the non-autocorrelated units showed both stimulus and category selectivity Quotation withheld because this record is not cleared for redistribution. THE ARTEFACT CLASS DEFINED BY WHAT IT CANNOT KNOW: no recording artefact would make spike shape depend on which picture was shown Quotation withheld because this record is not cleared for redistribution. ONE ANALYSIS SERVING TWO PURPOSES: the contiguity analysis both diagnoses a causal link to the stimulus and looks for the jump back in time Quotation withheld because this record is not cleared for redistribution. and the question is declared meaningless for units that are not autocorrelated in the first place Quotation withheld because this record is not cleared for redistribution. SORTING ERROR ONE ANSWERED: splitting one neuron across units could fake autocorrelation but not stimulus specificity, block sensitivity, or retained information about preceding stimuli Quotation withheld because this record is not cleared for redistribution. SORTING ERROR TWO ANSWERED: merging neurons could fake conjunctions, but not a response to the neighbours of a particular stimulus's original presentation Quotation withheld because this record is not cleared for redistribution. curves peaked at zero and falling off both ways would otherwise require the particular picture to change the shape of recorded spikes Quotation withheld because this record is not cleared for redistribution. AND AN ENVIRONMENTAL CLASS RULED OUT TOO: a fan or other sound present during part of a session cannot produce a contiguity effect Quotation withheld because this record is not cleared for redistribution. the artefactual and stimulus-independent accounts would both have predicted a flat curve Quotation withheld because this record is not cleared for redistribution. THE DECIDING RESULT: the backward effect is not predicted by persistent stimulus activation but is predicted by a jump back in time Quotation withheld because this record is not cleared for redistribution. the recency gradient was estimated and subtracted so that what remained on the lag axis was contiguity alone Quotation withheld because this record is not cleared for redistribution. SOME CONFOUNDS ARE STRUCTURAL AND UNCONTROLLABLE: the first stimulus of a block is necessarily new and the last necessarily old Quotation withheld because this record is not cleared for redistribution. the prior free-recall result had two named limitations: it was not at unit level, and recall order was set by the subject Quotation withheld because this record is not cleared for redistribution. which is why repetition under experimenter control gives a better test of the causal relationship Quotation withheld because this record is not cleared for redistribution. UNIT CRITERIA STATED NUMERICALLY: clear refractory periods, amplitude over 100 microvolts, under 1 percent of spikes below 3 ms Quotation withheld because this record is not cleared for redistribution. with a power-spectrum check discarding units carrying significant 60 Hz line noise Quotation withheld because this record is not cleared for redistribution. and every unit inspected by a human observer as well as by the clustering algorithm Quotation withheld because this record is not cleared for redistribution. the three-second analysis window is shorter than the minimum inter-stimulus interval, so no spike is counted twice Quotation withheld because this record is not cleared for redistribution. patients discriminated new from old reliably with no bias, hit rate .82 and correct rejection .83 Quotation withheld because this record is not cleared for redistribution. recency effect recency effect Quotation withheld because this record is not cleared for redistribution. temporally-autocorrelated units temporally-autocorrelated units Quotation withheld because this record is not cleared for redistribution. lag one regression coefficient lag one regression coefficient Quotation withheld because this record is not cleared for redistribution. contiguity effect contiguity effect Quotation withheld because this record is not cleared for redistribution. jump back in time jump back in time Quotation withheld because this record is not cleared for redistribution. temporal autocorrelation temporal autocorrelation Quotation withheld because this record is not cleared for redistribution. ensemble similarity ensemble similarity Quotation withheld because this record is not cleared for redistribution. repetition of a stimulus repetition of a stimulus Quotation withheld because this record is not cleared for redistribution. average firing rate average firing rate Quotation withheld because this record is not cleared for redistribution. ensemble similarities of event vectors ensemble similarities of event vectors Quotation withheld because this record is not cleared for redistribution. recency recency Quotation withheld because this record is not cleared for redistribution. lag lag Quotation withheld because this record is not cleared for redistribution. temporal autocorrelation temporal autocorrelation Quotation withheld because this record is not cleared for redistribution. .248 — proportion of recorded units classified as temporally autocorrelated Quotation withheld because this record is not cleared for redistribution. 44/270 — number of units that were category-selective Quotation withheld because this record is not cleared for redistribution. mean .094 ± .01 — average lag one regression coefficient across units, significantly above zero Quotation withheld because this record is not cleared for redistribution. hit rate = .82 ± .07; correct rejection rate = .83 ± .07 — patients' recognition performance on the continuous recognition task Quotation withheld because this record is not cleared for redistribution. .0042 ± .0015, F (1, 9) = 7.48, p < .03 — backward-direction regression of ensemble similarity on |lag| — the key jump-back-in-time result Quotation withheld because this record is not cleared for redistribution. more than 27 s — lower bound on the time scale over which the ensemble kept changing reliably Quotation withheld because this record is not cleared for redistribution. four patients, eight recording sessions — size of the human single-unit dataset Quotation withheld because this record is not cleared for redistribution. .248 — share of units classified as temporally autocorrelated Quotation withheld because this record is not cleared for redistribution. more than 27 s — lower bound on the timescale over which the ensemble kept changing Quotation withheld because this record is not cleared for redistribution. > 100 microvolts and less than 1% — amplitude and refractory-violation criteria defining a single unit Quotation withheld because this record is not cleared for redistribution. 3.10 s — minimum inter-stimulus interval, which is what makes the 3 s window safe Quotation withheld because this record is not cleared for redistribution. .82 ± .07 and .83 ± .07 — hit rate and correct rejection rate, showing no response bias Quotation withheld because this record is not cleared for redistribution.

The full text of this article is not redistributable, so it is not included here. The evidence sentences are withheld with it; the tag layer and its grounding status remain. The article itself is at https://doi.org/10.1002/hipo.22018.