Fig. 3: Functional stability of spatial coding influences cognitive outcome.

A VR performance after stroke relative to the healthy baseline revealing two different cognitive outcomes: A subset of mice showed a chronic cognitive deficit (No-Recovery group in red), while another group recovered from the initial cognitive decline within one week (Recovery group in blue). Sham-operated animals (grey) maintained their VR performance. A mixed-effects model with the Greenhouse-Geisser correction and Tukey-Kramer multiple comparisons test revealed significantly reduced task performances depending on the stroke outcome group (F_Group×Time(22, 180) = 5.380, p < 0.001). Asterisks denote significance levels between sham and Recovery (blue) or No-Recovery (red) groups, and between Recovery and No-Recovery groups (black). Data are presented as mean values +/- SD. B Percentages of stable place cells out of all imaged neurons in the three outcome groups (Sham, n = 11; Recovery, n = 5; No-Recovery, n = 4) across experimental phases. C Predicted corridor positions generated by a Bayesian decoder (orange) from neural ΔF/F traces and true position (grey) of a single exemplary trial. The background heatmap visualizes frame-wise decoder confidence (arbitrary units) for each VR position bin. The bin with the highest confidence is the prediction of the decoder for each frame. Prediction errors (e.g. around second 13) can often be attributed to the confusion of different reward zones with similar wall patterns. D Accuracy (percentage of frames with correctly predicted position bin) of the decoder indicating performance on data from the same session on which the decoder was trained (within-session decoder) in the three outcome groups (Sham, n = 11; Recovery, n = 4; No-Recovery, n = 4) across experimental phases. Dashed line indicates chance level (1.25%). Data from one mouse was excluded with a decoder accuracy not significantly different from the chance level in the healthy condition. E The sensitivity of the within-session decoder to detect reward zones (percentage of frames within reward zones correctly predicted as being in a reward zone) in the three outcome groups (Sham, n = 11; Recovery, n = 4; No-Recovery, n = 4) across experimental phases. Dashed line indicates chance level (43.75%). F Accuracy of the cross-session decoder (decoder trained on the last healthy session and applied to all other sessions) in the three outcome groups (Sham, n = 11; Recovery, n = 4; No-Recovery, n = 4) across the experiment. Hash symbols indicate significance (one-sample two-sided t-tests with Bonferroni correction) to chance level (dashed line, 1.25%). Boxplots are drawn with the box extending from the 25th to 75th percentiles, with the centre line at the median. Whiskers reach to the minimum and maximum values of the distribution. Group differences in B, D–F were evaluated with two-way repeated measures ANOVA with the Greenhouse-Geisser correction and Tukey-Kramer multiple comparisons tests. Asterisks and hash symbols indicate significances: */# p < 0.05, **/## p < 0.01, ***/### p < 0.001.