Extended Data Fig. 3: Correlation and potentiation index results are robust to changes in time windows and cell populations used for calculations.

(a) Recalculating noise correlations using a shorter 50 ms window, rather than a 200 ms window used in Fig. 1, showed a similar pattern of dynamic correlation reversal over trial blocks (indicated by the red arrow). Early versus late correlations on laser trials were significantly different from each other (** P = 0.001). Dashed vertical lines show time of laser onset and offset. (b) Correlation dynamics were similar if we separately considered pairs directly activated by the light (left-side panel), or indirectly via network connections (right-side panel). (c) We found similar effects when subselecting only significantly correlated pairs. To identify significantly correlated pairs, we shuffled control trials, and calculated correlations 1000 times. This created a distribution of chance level correlations. We next found pairs that had correlations on the first control trial block that were absolutely higher than the shuffled correlations > 95% of the chance correlations. The time- course of correlations on laser (blue) and control (gray) for these pairs is similar to the whole population of pairs (Fig. 1g). (d) Time course of correlation difference during rest looking only at the subset of light-responsive pairs on channels also identified as light-responsive during the awake condition (n = 264). (e) Recalculating the potentiation index using a larger 150 ms window around the CCG peak (rather than the 15 ms window used in Fig. 4) resulted in a similar significant increase (right-ward shift) in potentiation index distribution for narrow and broad cell pairs (purple, right side plot). This longer time window resulted in greater potentiation indices compared to the shorter time window (PI distribution means 1.68 versus 1.14, for long and short windows, respectively). Putative excitatory pairs (green, left side plot) did not exhibit a shift in the potentiation index using this longer time window (P = 0.056 Wilcoxon signed rank test). The distributions were significantly different from each other (P = 2.94E10-4, Wilcoxon ranked sum test, I-E vs. E-E, n = 1476 pairs). Arrowheads above distributions show the means. These results suggest that excitatory-excitatory and inhibitory-excitatory interactions can occur on different timescales.