Extended Data Fig. 4. Symmetric bilateral activation during decision-making and unique explained variance.
(a) Trial-averaged response maps for all correct, leftward trials across different PyN types. Cortical maps are as shown in Fig. 3a and Fig. 7c. (b) Average activity in auditory, parietal, and frontal cortex on the left (black) and right hemisphere (green), which are contra- and ipsilateral to the chosen side, respectively. In all PyN types, different trial events, such as initiation, sensory stimulation and animal responses increased neural activity. However, surprisingly few differences were seen between cortical hemispheres. To resolve differences in inter-hemispheric activation for left- versus rightward choices we therefore employed the choice decoder analysis (Figs. 6 and 7). Note that low or negative weights from the choice decoder (as seen for IT and CStr neurons in frontal cortex) do not reflect a lack of choice-related activity activity but are rather based on small differences in the activation of hemispheres that are either ipsi- or contralateral to the chosen side. (c) To isolate unique contributions from movement or task variables, we computed averaged maps of the loss in predicted variance (ΔR2) by removing either group of variables from the full model. This allowed us to separately examine their respective impact on cortex-wide activity by determining, for each PyN type, where in the cortex predictive power was lost. While movement ΔR2 patterns were comparable across PyN types (top row), PyN-type-specific differences were uncovered when removing task variables: ΔR2 was highest in frontal cortex of EMX and PT mice, but more diffuse in IT mice with the highest ΔR2 in auditory cortex (bottom row). Note differences in scale between two rows. (d) Examination of ΔR2 for individual task variables further suggest distinct roles for each PyN type. Here, the ‘choice’ variable had the highest contributions in PT neurons but was overall weaker in IT neurons. Conversely, contributions from other task variables were higher in IT neurons. This dichotomy was not observed in EMX neurons, indicating that IT and PT neurons might have different functional roles that cannot be resolved without PyN-type specific measurements. Each row represent a mouse. (e) Comparison of ΔR2 for choice (top) and stimulus variables (bottom) between PyN types. IT mice had significantly lower ΔR2 for choice (pEMX = 1.4 × 10−22, pIT = 1.1 × 10−7; nEMX = 62, nIT = 71, nPT = 59 sessions) but higher ΔR 2 for the stimulus as EMX or PT mice (pEMX = 1.2 × 10−5, pIT = 4.9 × 10−14). Note that lower ΔR2 for choice in IT mice does not imply a lack of involvement in decision formation but rather that their population activity does not clearly differ for left versus right choices. Dots indicate individual sessions. Stars indicate significant differences across sessions (two-sided unpaired t-test, p < 0.01, bonferroni-corrected).