Fig. 4: Assembly drift predicts successful memory and occurs more frequently in the posterior compared to the anterior hippocampus.

A Top, schematic of the anterior (AH) and posterior (PH) hippocampus, entorhinal cortex (EC), amygdala (AMY), and parahippocampal gyrus (PHG) with all recorded neurons from an example session superimposed. The exact neuron location is for illustration only. Both member (M) and non-member (NM) neurons are displayed for an example assembly and neurons demonstrating drift across activation events (+d) are noted. The drift fraction is displayed above the schematic. Bottom, the firing rate of four example neurons from the recording session across all assembly activation events, showing examples of neurons that decrease (red) and increase (blue) their contribution and that remain stable (gray) across the session. For display purposes only, the firing rate curve is smoothed with a 10 s window and normalized and set the value at time zero to 0. The anatomical region and category of neurons corresponding to the displayed data were shown to the left of the plots. B Comparison of the observed drift fraction (DF) across all assemblies to a null distribution obtained by shuffling the activation bin order for each neuron before re-classifying neurons (permutation test, n = 1000 shuffles, p = 0.0009). C Positive Spearman’s rank correlation between the drift fraction and percentage of words recalled across the experiment (n = 45, r = 0.470, p = 0.0011). D Comparison of the observed DF-memory correlation to a null distribution obtained by shuffling activation bin order within each cell (permutation test, n = 1000 shuffles, p = 0.0009). E Pairwise comparison of DF between the AH and PH with the overall DF for each region displayed as the bar height and each line representing a single assembly (paired t-test, t(n = 21) = −2.432, p = 0.0123). F Comparison of the effect size of the observed pairwise DF difference along the hippocampal longitudinal axis to a null distribution generated by shuffling anatomical labels (permutation test, n = 1000 shuffles, p = 0.007). G The same as (F) but with the null distribution created by shuffling activation bins (permutation test, n = 1000 shuffles, p = 0.0009). H An example of asymmetric drift between the AH and PH, following the same convention as outlined in (A). However, AH and PH DFs are displayed separately above the schematic. Source data are provided as a source data file.