DISSECTING CONTRIBUTIONS TO SHARED SUBSPACES IN MULTI- REGIONAL NEURONAL POPULATION ACTIVITY DURING GOAL-DIRECTED REACHING

Neural computations underlying goal-directed behavior emerge from coordinated activity across distributed brain circuits.
However, how inter-regional communication supports the encoding of motor and non-motor variables remains unclear.
To investigate this, we analyzed multi-region Neuropixels recordings from two datasets with directional licking task (Svoboda et al., Cell 2024) and forelimb reaching task (Peng et al., BioRxiv 2025).
Using tensor component analysis, we identified a conserved dominant shared subspace at the brain-wide level that primarily encodes reward. Despite this brain-wide similarity in shared latent dynamics for each paired region, neural populations responding to different movement types could still be distinguished within this subspace.​
We further examined whether contribution to the shared subspaces is distributed across many neurons or concentrated in a small population across pairs of brain regions. Our results reveal an encoding concentration gradient from cortex to​​​​​ sub-cortical regions that is task-specific.
These findings suggest that goal-directed behavior relies on both brain-wide shared dynamics and region-specific computational gradients that systematically adapt to task requirement. Overall, our work will help to better understand how distributed neural circuits across regions are coordinated in action-specific subspaces.