Subject-specific

voxels of interest were defined by identifying all animal and tool picture selective voxels (p = 0.05, uncorrected) within each sphere for each individual. Finally, the BOLD-response to animal and tool words were extracted from these voxels and compared across age. Higher BOLD-related confounds in Y-27632 in vivo children can compromise the results of age-comparisons. As described in the previous section, harmful effects of motion artefacts were minimised by applying strict run exclusion criteria for overall motion, and by capturing signal changes resulting from small sudden movements in regressors of non-interest. To exclude the possibility that despite these procedures, age-differences in picture-like responses to printed

words could still be driven by larger BOLD-related confounds in children, we tested if age differences across all subjects persisted when the same comparisons were performed across sub-groups of adults and children matched on the following two noise indices: Because sudden movements can leave residual noise in the BOLD-signal after registration, scan-to-scan motion is a good indicator of motion-related variance in the signal after standard correction procedures are applied. The mean Euclidian translational movement distance ΔD from one volume to the next was calculated in millimetres and the mean absolute scan-to-scan rotational motion Δθ was calculated in Urease radians: ΔD=∑TR=1N-1(XN+1+XN)2+(YN+1+YN)2+(ZN+1+ZN)2N-1 Δθ=∑TR=1N-1abs(pitchN+1+pitchN)+abs(rollN+1+rollN)+abs(yawN+1+yawN)N-1 This reflects residual variance in the data unaccounted for R428 clinical trial after fitting

the full General Linear Model with regressors of interest and nuisance regressors. It is an inclusive measure of BOLD-related noise and goodness of model fit. For animal and tool picture category-selective voxels in each spherical region of interest, residual variance of the GLM was extracted from the subject/scan.feat/stats/sigma-squaredes.nii images in FSL that were first resampled to standard space and averaged across all scans. Using the formula reported in (Golarai et al., 2007), we then computed mean percentage of residual noise in the signal of each ROI: %Res=100×1Nvox∑i=1NvoxSigmasquareds(i)MeanAmp Mean Amp is the average BOLD signal across all scans within the relevant voxels of interest, extracted from the mean_func.nii.gz image in the second-level subject/allscans.gfeat folder in FSL. Finally, resulting %Res values were averaged across all ROIs to obtain one total value per subject. In the Appendix B, Table 1, these indices of noise in the data are reported for all age groups, and for two subgroups of 9 adults and 9 children matched on these BOLD-related confounds. Control analyses with these matched sub-groups are reported in the final section of Section 3.