Fig. 3: Hippocampal–entorhinal population activity remaps to structured experience.

a, The logic behind the population decoding analysis. b, Neuronal responses contained enough information to successfully decode the stimulus identity during PRE (chance level ≈ 17%; data from all identified neurons; n = 1,456). The plot shows mean decoding accuracy (±s.e.m.) from 100-ms bins averaged across all recording sessions (n = 21; time zero is the stimulus onset). The shaded grey area marks the time window used for further analyses. c, Results from the hippocampal–entorhinal neurons (n = 546). P values obtained from the Kolmogorov–Smirnov tests between cumulative distribution functions (CDFs) of posterior probabilities assigned by the decoder during PRE versus subsequent study phases (one-sided). d, The difference between CDFs for direct and indirect stimuli remained significant during POST (Kolmogorov–Smirnov test; one-sided). e, Top row, combined data from trials where the actually presented stimulus was at an outer node of the pyramid. P values represented by dotted or solid lines of different widths were obtained from Kolmogorov–Smirnov tests between each pair of nodes (one-sided; FDR-corrected). Colour intensities correspond to distances (Kolmogorov–Smirnov z-statistic) between the respective CDFs. The seed node is marked in orange. Bottom row, analogous results for trials where the stimulus actually presented was at an inner node. NS, not significant. f, Distance matrixes and graphs corresponding to the geodesic, Euclidean and successor templates. Each graph shows the most faithful 2D representation of the respective distance matrix using the multidimensional scaling analysis. Note that the matrix and graph obtained from the neuronal data (right) closely resemble the successor template (546 hippocampal–entorhinal neurons; E4–E6 data combined for illustration purposes). g, The degree of similarity between data and each template throughout the study. Spearman’s correlation coefficients (Fisher-transformed) between each template and neural data from respective phases (changes from PRE).