Supplementary Figure 5: Performance does not depend on baseline activity in E1 cells.

a. Behavioral performance plotted against the makeup of E1. In 40 sessions, E1 was composed of a mix of more- and less-active cells (defined as cells in the lowest quintile of spontaneous activity rates). In 6 training sessions, E1 was solely comprised of low-active cells, and in 4 cases all cells in E1 had zero baseline activity (filled circles); in three cases the animal still performed the task above chance. Performance in both conditions was significantly better than chance (t-test, high- and low-active ensembles, p=2.2e-15, t(38)=11.69; low-active ensembles, p=0.02, t(5)=3.7). b. Analysis of the contribution of individual E1 cells to learned modulation of E1 ensemble activity. Fluorescence modulations around hits were calculated for each E1 cell, and normalized to the sum of modulations in the entire ensemble. Each cell's relative contribution to these modulations was then averaged over the entire training session, and plotted against the number of cells in the E1 ensemble. A bimodal distribution of contributions around 0 and 1 would suggest the task were carried solely by one cell in an ensemble. If all cells contributed equally, the distribution would peak around 1/N (red curve), where N = number of cells in the ensemble. The animals appear to use a combination of both strategies. Less-active neurons (in blue) contributed to hits in a manner indistinguishable from more-active neurons.