Only pairs of neurons from a single electrode that showed clearly separate clusters in the first three principle components of the spike waveform were included in the sample. Since the exact distance between neurons recorded from a single electrode was unknown, we arbitrarily assigned it to be 50 μm. Although

noise correlations were slightly greater for pairs of neurons recorded from a single electrode (0.042 ± 0.02) than for pairs recorded from different electrodes (0.033 ± 0.015), this difference was modest and not significant (p > 0.7, t test). Thus, data collected with single and multiple electrodes were pooled for analysis, yielding 179 cell pairs from a total of 270 neurons (maximum of 5 pairs in an experiment). Area MSTd was located ∼15 mm lateral to Selleckchem Adriamycin the midline and ∼2–6 mm Cobimetinib chemical structure posterior to the interaural plane, and was identified using both MRI scans and neurophysiological response properties (see Gu et al., 2006 for details). MSTd neurons had large receptive fields that typically occupied a quadrant or a hemifield on the display screen and were often centered in the contralateral visual field but could extend well into the ipsilateral field. Once the electrodes were targeted to MSTd, we recorded from any neuron that was spontaneously active or could be activated

by patches of flickering dots. Noise correlation (rnoise) was computed as the Pearson correlation coefficient (ranging between −1 and 1) of the trial-by-trial responses from a pair of neurons driven

by the same stimulus (Bair et al., 2001 and Zohary et al., 1994b). The response in each trial was taken as the number of spikes during the middle 1 s of the stimulus period (Gu et al., 2006). For each heading direction, responses were z-scored by subtracting the mean response and dividing by the standard deviation. This operation removed the effect of heading on the responses, such that the measured noise correlation reflected trial-to-trial variability. To avoid artificial correlations caused by outliers, we removed data points with z-scores larger than 3 (Zohary et al., 1994b). We then pooled data across headings to compute rnoise; the corresponding p value was used to assess the significance of correlation for each pair of neurons. Because there was no significant difference in rnoise between visual and vestibular stimulus conditions (Figure 1F), we pooled responses across conditions to too gain statistical power. To remove slow fluctuations in responsiveness that could result from changes in cognitive state over time (e.g., arousal), we renormalized the z-scored responses in blocks of 20 trials, as described by Zohary et al. (1994b). This additional normalization had no significant effect on rnoise (p > 0.3, paired t test; R = 0.9, p < < 0.001, Spearman rank correlation, n = 127, Figure S8). More importantly, the effect of renormalization on noise correlations was similar in naive and trained animals (p = 0.7, interaction effect, p = 0.9, group effect, ANCOVA, Figure S8).