Learning to detect direct electrical microstimulation is associated with the appearance of cells responding with changes with firing rate and cells responding with burst fraction. A Schematic of the experimental setup. B Progression of average response accuracy through training days, averaged across individual animals. Training lasted 1 to 2 days depending on the animal. After training, Pearson correlation of accuracy vs days after training was 0.86 (P =.085, bootstrap test). C Schematic of the separation of spike trains into trains of bursts and events based on an interspike interval (ISI) threshold. D Schematic of the calculation of burst fraction from the trial averaged burst and event rates. E Trial-averaged firing rate (black) and burst fraction (red) averaged across all cells recorded in the naive (top) or trained (bottom) condition. F Total response relative to the baseline FR (left) and BF (right) for naive (N) and trained (T) conditions. G Histogram of the number of action potentials per bursts in naive (black) and trained (purple) condition. Inset shows the average burst length for naive and trained condition (P =.001, rank sum test). H Flowchart for the classification of cells according to responses in either FR or BF. Cells showing a significant response in FR are FR-ON cells and cells showing a significant response in BF are BF-ON cells. I Pie chart of cell response class in trained animals. J The fraction of FR-ON cells (left) and BF-ON cells (right) in naive (N, black bar) and trained (T, purple bar) conditions (FR-ON P =.011, BF-ON P =.002; proportion z-test). Shaded area along lines in E represents ± 1 s.e.m. Response to both hits and miss trials are pooled.

Burst fraction reveals information independent from that in the event rate. A For an exemplar cell, the raster of spikes (top, black dots) and first spike in a burst (orange dots) is shown (top). The trial-averaged firing rate (FR, black line), event rate (ER, blue line) and burst fraction (BF, red line, right axis) are shown as a function of time from stimulation (bottom). B Heat map of the trial-averaged firing rate response to microstimulation (z-scored) ordered with most firing cells on top. C Heat map of the trial-averaged burst fraction (z-scored) keeping the ordering used in B. D Scatter plot of post-stimulus z-score of FR against BF and ER against BF. E Average FR and BF of all BF-ON cells (n =70 out of 153). F A Poisson model matching the firing rate of each BF-ON cell recovers the ensemble FR but not the ensemble BF. G Average BF-ON cell response from BF-ON cells (n =13 out of 279) found by searching through a surrogate Poisson population matching the FR of each cell in the experiment. Shaded area along lines represents ± 1 s.e.m. Response to both hits and miss trials are pooled.

Delayed representation of errors in S1. A Raster plot for an exemplar cells (top) separating Hits (right) from misses (left). Bottom panels shows associated trialaveraged firing rate (FR, black lines), event rate (ER, blue lines) and burst fraction (BF, red lines, right axis). B Population averaged burst fraction (top), event rate (middle) and firing rate (bottom) for hits (full lines) and misses (dashed lines). The black bar indicates the time bins for which a significant difference was observed (P < .05, corrected for multiple comparison at every time bin). The vertical dotted line indicates the end of the response window. C Trial and population averaged lick rates for hits (full line) and misses (dashed line). D Population averaged firing rate (black) and burst fraction (red, right axis) for catch trials (no stimulation, but rewards are given on licks) separating hits (full lines) from misses (dashed lines). Shaded area along lines represents ± 1 s.e.m.

The gradual temporal alignment of a global burst fraction modulation with the focal increase in event rate correlates with behavioral accuracy. A Trial- and population-averaged burst fraction and firing rate for FR-NR cells for hits only, separating recordings into subsequent days of recording and training. B The firing moment (see Methods) is shown as a function of training days for FR (left), BF (middle) and ER (right), for FR-NR cells. The line indicates a significant correlation (Pearson correlation -0.75 (P =.12), -0.99, (P <.0001), -0.9 (P =.042; bootstrap test), respectively). C Schematic: representations progress in later training days mainly by having an earlier modulation of burst fraction. D-F Trial- and population-averaged event rate (D), burst fraction (E) and firing rate (F) for FR-ON cells relative to baseline and separated according to training days (legend in panel A). Inset shows relationship between representation in the response period and training days (Pearson correlation -0.59 (P = .23), 0.83 (P = .84), 0.67 (P = .23) for ER, BF and FR, respectively). E, right Shows relationship between burst fraction relative to baseline and accuracy. Pearson correlation was significant for the response period (0.89, P =.042, bootstrap test). F, right Shows relationship between firing rate relative to baseline and accuracy. Pearson correlation was not significant (0.87, P =.050, bootstrap test). Responses are for hit trials only.