Supplemental Figures

2 figures in JPEG format

Files in this Data Supplement:

• Figure S1 - Comparison of canonical and individually defined hemodynamic response functions (HRFs). Each graph shows two HRFs for a single subject (red, canonical HRF; blue, individually defined HRF). Next to each panel is the Pearson’s correlation coefficient of the two responses. The two HRFs were highly correlated in all of the subjects (mean r = 0.87 averaged across subjects and ROIs, SEM = 0.03). To assess whether the results of the ROI analysis were dependent on the choice of a canonical HRF, we estimated an HRF individually for each subject and tested how similar it was to the canonical one. We began by averaging data from all of the voxels in the ten ROIs (see Methods) so as to generate a single time-course for each subject in each of the RPS games (runs). The four measured time-courses corresponding to the four RPS games were concatenated into a single long time-course and a “deconvolution” analysis was performed to estimate an HRF for each subject. This analysis relied on linear regression, solving an equation of the form y = Ax , where vector y was the measured fMRI time-course and vector x was the estimated individual-subject HRF containing 13 values. The model matrix A had 13 columns (corresponding to the number of time points in the estimated HRF). The first column of A contained a value of 1 at indices corresponding to the onset of movement trials, the second column contained a 1 at indices corresponding to the second time point, and so on. Thus the model was made up of diagonals of 13 ones corresponding to every trial where a movement was observed/executed and zeros everywhere else. The result of this analysis yielded an individual HRF for each subject (Supplementary Figure 2, blue curves). The individual-subject HRFs were then compared with the canonical HRF. To make a fair comparison, we fit the fMRI time-courses using a canonical HRF. The model contained a row for every time point (2 per trial) and a single column. Neural activity was modeled as either “on” = 1 for all trials where a movement was executed/observed or “off” = 0 for all the blank trials (rest). The modeled neural activity was convolved with a canonical HRF (Boynton et al. 1996) and an estimate of a single response amplitude (beta value) was calculated for each subject using linear regression (solving an equation of the form y = Ax, where vector y was the measured fMRI time-course, x was the single response amplitude value, and A was the model containing a single column). Finally, we multiplied the canonical HRF with the computed response amplitude for each of the subjects to generate the estimated canonical response (Supplementary Figure 1, red curves). Technically, rather than simply scaling the canonical HRF, we performed a deconvolution analysis, as detailed above, on the model fits, to take into account influences of trial sequencing on the shape of the resulting HRF. Note that for a large enough number of trials this deconvolution analysis yields a copy of the canonical HRF scaled by the estimated response amplitude.
• Figure S2 - Region of interest (ROI) analysis using hemodynamic response functions (HRF) estimated for each subject individually. We reanalyzed the rock-paper-scissors data in the six ROIs that exhibited movement selectivity (Figure 5) using each subject’s HRF (see Supplementary Figure 1). Top row. Visual and motor adaptation. Comparison of fMRI response amplitudes in four conditions: observed non-repeat (dark green), observed repeat (light green), executed non-repeat (dark orange), executed repeat (light orange). Bottom row. Cross-modal interactions. Comparison of fMRI response amplitudes in four conditions: observed-then-executed non-repeat (dark blue), observed-then-executed repeat (light blue), executed-then-observed non-repeat (dark purple), and executed-then-observed repeat (light purple). Error bars, SEM across subjects. Asterisks, statistically significant difference (p < 0.05, paired t-test). The results of this analysis are almost identical to those computed with the canonical HRF (Figure 5) demonstrating that the visual and motor adaptation effects were robust in all but one ROI. The only exception was LO where the visual adaptation effect was lost.