Fig. 6. Local signals support a M2 to M1 hierarchy.

a Illustration of approach to identifying local activity using artificial data. Black line represents a population firing rate for a single time bin. Population firing was projected onto a shared axis defined by CCA (solid yellow line) to obtain the cross-area signal (dotted yellow line), and onto the hyperplane orthogonal to the CCA subspace (light red plane), to obtain the local signal (dotted red line). b Example animal local activity in early learning (z-scored). (Top) Single-trial local activity trajectories for M2. Black dots indicate reach onset and white dots indicate transition to grasp on each trial. (Middle) Single-trial local activity trajectories for M1. (Bottom) Mean local activity trajectories for M2 (blue) and M1 (red). Shaded regions show SEM, n = 211 trials. c As in b, but for late learning, n = 297 trials. d Distributions of timing differences between median timing of single-trial M2 and M1 local activity in early (brown) and late learning (gray) (from example animal in b, c). Black dotted line indicates zero lag in M2–M1 median timing of local activity. e Quantification of M2–M1 timing differences (n = 4 rats). In gray, permutation-based reference distribution of timing differences with data randomly assigned to M1 or M2 (10⁵ permutations). M2 local activity significantly preceded M1 activity in early (green) and late (orange) learning. f Quantification of tighter coupling between M2 and M1 local activity from early to late learning (n = 4 rats). In gray, permutation-based reference distribution of mean difference in M2–M1 timing coupling with M2–M1 timing differences randomly assigned to early or late learning (10⁵ permutations). The true difference in M2–M1 coupling between early and late learning (purple line) was more negative than any of the reference values, indicating that the timelag between M2 local activity and M1 local activity significantly decreased with learning.