We computed the probability of correct classification of trials in the preferred direction as fast or slow RT trials, based on firing rate in the eight cells with significant RT selectivity based on an ANOVA. Before coordinated movements to the preferred direction, coherent cells significantly

predicted whether a trial had a LY294002 fast or slow RRT (decode probability correct: 0.86; p < 0.001 binomial test) and a fast or slow SRT (decode probability correct: 0.72; p < 0.001). In contrast, not coherent cells did not significantly predict RRT (decode probability correct: 0.58, p = 0.16) or SRT (decode probability correct: 0.48; p = 0.56). Coherently active cells predicted RRT significantly better than cells that were not coherently active (p < AZD8055 order 0.005, two-sample binomial test). Coherently active cells also predicted SRT significantly better than not coherent cells (p < 0.05). Importantly, coherent cells encoded the speed of RTs only when movements were coordinated. When saccades were made alone, despite the fact that mean firing rate did not differ for reach and saccade versus saccade alone trials (Figures 5A and 5B), the decoder performed at chance (decode probability correct: 0.56,

p = 0.16). The not coherent cells also did not predict SRT (decode probability correct: 0.48, p = 0.56). The performance advantage of the coherent cell population in decoding RT was not due to the fact that there were more cells in the coherent population than the not coherent population. We repeated the analysis for increasing sizes of coherent and not coherent populations up to the number of available cells. The coherent population outperformed the not coherent population

for all cell subsets greater than two (Figure S2). Although we report here the results for eight cells, note that, for all numbers of coherent cells greater than three, the decoder performs best and above chance for RRT during coordinated movements. The decoder also performs well and usually above chance for SRT during coordinated movements at all numbers of cells but not for SRT during saccades made alone or for not coherent cells. We also examined whether the better decoding performance of the coherent cells could be due to their higher overall firing rate. When we subtracted the mean firing Suplatast tosilate rate from each cell before decoding the firing rate, we found that the results maintained the same pattern of significance. Coherently active cells predicted coordinated movement RT significantly better than cells that were not coherently active (RRT: p < 0.05. SRT: p < 0.01). Neither group of cells predicted SRT before saccades made alone (Coherent decode probability correct = 0.48. Not coherent decode probability correct = 0.46). Additionally, we decimated the firing rate of the significantly coherent units by 50% to match the firing rate of the not coherent units (see Experimental Procedures).