Extended Data Fig. 7: Additional support for the successor model and a comparison of different successor representation metrics.

(a) An important prediction of the successor representation, but not the geodesic representation, is that direct links between inner nodes (inner-inner) should be shorter than direct links between each outer node and the adjacent inner nodes (outer-inner). We calculated pairwise distances between spiking responses of all selective hippocampal-entorhinal neurons grouped by their preferred node. Then, we compared the mean distances of inner-inner links (2-3, 3-5, 5-2) to the mean distance of outer-inner links (4-2, 4-5, 6-5, 6-3, 1-2, 1-3). As predicted, the former were significantly shorter than the latter (P values from Wilcoxon rank-sum tests; two-sided; FDR-corrected). The plot shows mean distances from all 263 (outer: n = 119; inner: n = 144) selective hippocampal-entorhinal neurons ± s.e.m. (b) There were no significant differences between the two measures of the successor representation in terms of similarity to the neuronal data. The plot shows distance matrixes and graphs corresponding to the two measures (see Methods). Each graph shows the most faithful 2D representations of the respective distance matrix obtained from the multidimensional scaling analysis. The right panel shows the degree of similarity between data (546 hippocampal-entorhinal neurons) and each template throughout the study (Spearman’s correlation coefficients; Fisher-transformed; change from PRE). P values (FDR-corrected) were calculated as the number of permutations with a higher difference between the templates than the one actually detected, divided by the total number of permutations (10,000).