Fig. 2: Estimation of coupling filters and their remapping with optic flow density.

A P-GAM model. The encoding model (main text, see Fig. S9 for another variant) included 17 analog or digital task-variables, as well as spike history and coupling filters (in orange for emphasis). Figure adapted from REF³⁵ B Tuning function gain. Gain in tuning functions in the low-density condition relative to the high-density optic flow condition. A value under ‘1’ (dashed black line) shows that the amplitude of the tuning function was reduced in the low-density condition relative to the high-density condition. Error bars are ± 95%CI. C Coupling filter strength. For all pairs of neurons that were coupled both in the high- and low-density conditions, we quantified the strength of their coupling and averaged within sessions. Coupling strength increased in the low-density condition in MSTd (top; green) but was unchanged in area 7a (middle; blue) and dlPFC (bottom, red). Each dot is a session. D Coupling filter shape stability. Cumulative density functions (i.e., cumulative probability) of the coupling filters correlation coefficients (r) between low- and high-density conditions for all three brain regions (see Fig. S9 for the same data plotted as boxplot). Chance and ceiling-levels were determined by permutation, respectively correlating coupling filters recorded on separate sessions, channels, and animals (i.e., chance level), as well as correlating the same coupling pair (same session, animal, density, sender and receiver units) on odd and even trials (i.e., maximum level). MSTd (green) being the closest to “max” indicates that this region had the most stable (i.e., most correlated) coupling filters across density conditions. Source data are provided as a Source data file.