Abrupt hippocampal remapping signals resolution of memory interference

Abstract

Remapping refers to a decorrelation of hippocampal representations of similar spatial environments. While it has been speculated that remapping may contribute to the resolution of episodic memory interference in humans, direct evidence is surprisingly limited. We tested this idea using high-resolution, pattern-based fMRI analyses. Here we show that activity patterns in human CA3/dentate gyrus exhibit an abrupt, temporally-specific decorrelation of highly similar memory representations that is precisely coupled with behavioral expressions of successful learning. The magnitude of this learning-related decorrelation was predicted by the amount of pattern overlap during initial stages of learning, with greater initial overlap leading to stronger decorrelation. Finally, we show that remapped activity patterns carry relatively more information about learned episodic associations compared to competing associations, further validating the learning-related significance of remapping. Collectively, these findings establish a critical link between hippocampal remapping and episodic memory interference and provide insight into why remapping occurs.

Fig. 1. Experimental design and behavior.

a Participants learned 36 scene-object associations. The 36 scenes comprised 18 scene pairmates which consisted of highly similar image pairs (e.g., “lighthouse 1” and “lighthouse 2”). Scene pairmates were also associated with similar objects (e.g., “guitar 1” and “guitar 2”). b Participants completed six rounds of study, test, and exposure phases. During the study, participants viewed scenes and associated objects. During the test, participants were presented with scenes and had to select the associated object from a set of two choices, followed by a confidence rating (high or low confidence; not shown). During exposure, scenes (rounds 1–6) or objects (rounds 1 and 6) were presented and participants made an old/new judgment. fMRI data were only collected during the scene and object exposure phases. c Mean percentage of high confidence correct responses for each test round. d Data from a representative participant showing the “inflection point” in learning (red horizontal line), for each pairmate. The inflection point was defined as the point at which participants transitioned to high confidence correct retrieval for both scenes within a pairmate—a transition from “pre-learned” (black) to “learned”(aqua). e Mean number of scene pairmates that transition to a learned state at each round. N.L. indicates pairmates that were never learned. Notes: Data were presented as mean values ± SEM, n = 31 independent participants. Source data are provided as a Source Data file.

Fig. 2. Pairmate similarity scores change at the behavioral inflection point.

a Regions of interest included CA3/dentate gyrus (CA3/DG, pink) and CA1 (blue) in the hippocampus, the parahippocampal place area (PPA, yellow), and early visual cortex (EVC, green). b Correlation matrix illustrating how pairmate similarity scores were computed at the behavioral inflection point. See Methods for details. c Pairmate similarity scores at the behavioral inflection point (IP) and just prior to the inflection point (pre-IP) across different regions of interest (ROIs). Pairmate similarity scores significantly varied by ROI (p = 0.009, repeated measures ANOVA) and there was a significant interaction between ROIs and behavioral state (p = 0.037, repeated measures ANOVA). In CA3/DG, pairmate similarity scores at the IP were significantly lower than 0 (p = 0.025, two-tailed one-sample t-test) and significantly lower than the pre-IP state (p = 0.033, two-tailed paired samples t-test). In PPA, pairmate similarity scores decreased from pre-IP to IP (p = 0.030, two-tailed paired samples t-test). d A permutation test (1000 iterations) was performed by shuffling, within participants, the mapping between the behavioral inflection point and scene pairmates. In CA3/dentate gyrus the actual mean group-level pairmate similarity score at the IP was lower than 98.70% of the permuted mean similarity scores (p = 0.013, one-tailed permutation test). e Pairmate similarity scores calculated by correlating the learned round (LR) with each of the three preceding rounds (– distance to LR) and each of the three succeeding rounds (+ distance to LR). [Note: the inflection point was defined as the correlation between the LR and the immediately preceding round (LR − 1); the inflection points are depicted by filled circles and are the same values as in c]. In CA3/dentate gyrus, pairmate similarity scores were significantly lower when the LR was correlated with preceding rounds compared to succeeding rounds (p = 0.006, two-tailed paired samples t-test). The difference was not significant for any other ROIs (CA1: p = 0.435; PPA: p = 0.955; EVC: p = 0.760; two-tailed paired sample t-tests). f Conceptual illustration of a decrease in pairmate similarity scores from pre-IP to IP. In the pre-IP state (top panel), A₁ and A₂ are nearby in representational space. In the IP state (bottom panel), the representational distance between A₁ and A₂ has been exaggerated. When pairmates (e.g., A₁ and A₂) are farther apart in representational space than non-pairmates (e.g., A₁ and B₂) the pairmate similarity score will be negative (i.e., pairmate similarity < non-pairmate similarity), consistent with a repulsion of competing representations. Notes: *p < 0.05, **p < 0.01. No correction for multiple comparisons was applied given the a priori predictions for CA3/dentate gyrus. Data were presented as mean ± SEM and all data reflect n = 31 independent participants. Source data are provided as a Source Data file.

Fig. 3. Representational structure across timepoints.

a Schematic illustration showing the rank order of scene pairmates based on pairmate similarity scores at various time points (N, N + 1, N + 2). If scene pairmates with relatively high pairmate similarity scores at a given timepoint are systematically associated with relatively low pairmate similarity scores at a succeeding time point (red arrows), this will produce a negative rank correlation. b Mean rank order correlations of pairmate similarity scores across timepoints for CA3/dentate gyrus (CA3/DG, pink) and CA1 (blue). Lag 1 correlations reflect correlations between a given timepoint and an immediate succeeding timepoint (e.g., timepoints 2 and 3). Lag 2 correlations reflect correlations between a given timepoint and a timepoint two steps away (e.g., timepoints 2 and 4). At lag 1, there was a negative correlation in CA3/dentate gyrus (p = 0.006, two-tailed one-sample t-test), but a positive correlation in CA1 (p = 0.043, two-tailed one-sample t-test). At lag 2, correlations were not significant in either CA3/dentate gyrus (p = 0.485, two-tailed one-sample t-test) or CA1 (p = 0.120, two-tailed one-sample t-test) indicating that correlations in the representational structure were specific to temporally adjacent rounds. c Pairmate similarity scores at the inflection point (IP) as a function of relative pairmate similarity scores in the pre-IP state (first quartile = lowest similarity, fourth quartile = highest similarity). Pairmate similarity scores in CA3/dentate gyrus were significantly lower than CA1 (p = 0.008, two-tailed paired samples t-test) and significantly below 0 (p = 0.017, two-tailed one-sample t-test) for pairmates with the highest pre-IP similarity (fourth quartile). See Supplementary Fig. 4 for the distributions of pre-IP pairmate similarity scores. Notes: *p < 0.05, **p < 0.01. No correction for multiple comparisons was applied given the a priori predictions for CA3/dentate gyrus. Data were presented as mean ± SEM and all data reflect n = 31 independent participants. Source data are provided as a Source Data file.

Fig. 4. Scene-object similarity as a function of behavioral state.

a Example associations between scene pairmates and objects. The scene-object similarity was calculated by correlating activity patterns evoked during the scene exposure phases (at different behavioral states) and the object exposure phases. Target similarity refers to correlations between a given scene and the object with which it was studied. Competitor similarity refers to correlations between a given scene and the object with which its pairmate was studied. b Scene-object similarity as a function of object relevance (target, competitor), ROI (CA3/dentate gyrus, pink; CA1, blue), and behavioral state (pre-learned round, learned round). Mean correlations between unrelated scenes and objects (across pairmate similarity; not shown) were subtracted from target and competitor similarity values. For CA3/dentate gyrus (CA3/DG), there was a significant interaction between behavioral state and object relevance (p = 0.002, repeated measures ANOVA). Note: **p < 0.01. No correction for multiple comparisons was applied given the a priori predictions for CA3/dentate gyrus. Data were presented as mean ± SEM and all data reflect n = 31 independent participants. Source data are provided as a Source Data file.