Data Analysis

Except for seed determination (see below), we used an adaptation of the processing pipeline of Fauvel et al. (2014) with SPM8 version 5236 (http://www.fil.ion.ucl.ac.uk/spm/; Wellcome Department of Imaging Neuroscience Group, London, UK).

Each functional volume series was automatically inspected for excessive head movements with the tsdiffana routine (http://imaging.mrc-cbu.cam.ac.uk/imaging/DataDiagnostics). No abnormal spike of variance, rotational (>1.5°) or translational (>3 mm) movement, was observed in time series. T1-weighted structural images were spatially normalized to the Montreal Neurological Institute (MNI) template (ICBM AVG152), segmented using VBM8 (http://dbm.neuro.uni-jena.de/vbm/), which is incorporated in the SPM8 software, and smoothed using an 8-mm full width at half maximum (FWHM) isotropic Gaussian kernel. EPI volumes were corrected for slice timing, realigned on the first volume, and coregistered to the T1 volume. The coregistered T1 and EPI volumes were normalized on the basis of the segmented gray matter, and 4-mm FWHM smoothing was applied to the EPI volumes. The signal was bandpass filtered (0.01–0.08 Hz). Finally, the individual segmented gray matter T1 volumes were averaged in the MNI space, and a binary mask was created including only voxels with values above 0.3 in the average gray matter image and with a higher probability to be gray matter than white matter or cerebrospinal fluid. This binary mask was used in all subsequent analyses.

As in the seminal study by Biswal et al. (1995), we used functionally defined seeds (see also, e.g., Dosenbach et al. 2007; Hampson et al. 2006; Vahdat et al. 2011), and we further checked these seeds to be congruent with anatomical data. The seed regions were 10-mm-diameter spheres in right and left Heschl's gyri, centered on the MNI coordinates [45,−19,6] and [−44,−18,5] observed in the magnetoencephalographic (MEG) data of Albouy et al. (2013a). The seeds correspond to the sources of the N100 responses for tone encoding, where significant differences of activity and of connectivity with frontal areas were observed between amusics and controls by Albouy et al. (2013a, 2015). Note that activity in these seeds was observed in an active music memory fMRI protocol in the same subjects (Albouy et al. 2014). The entire sphere was located within the most medial part of Heschl's gyri of each individual's anatomical MRI and did not overlap with other nonauditory regions (e.g., insula; Rademacher et al. 2001).

Furthermore, to exclude that anatomical differences within Heschl's gyri might drive functional differences between groups, a voxel-based morphometry analysis was run (using SPM8) on both seed regions of interest (ROIs) to compare gray or white matter concentrations based on the segmented T1-MRIs (see above). A two-sided, two-sample t-test on the mean value within the seed sphere for gray matter and for white matter did not reveal any significant difference between controls and amusics [t(24) = −0.97, P = 0.742 for gray matter; t(24) = −0.51, P = 0.433 for white matter]. As control seed regions, we also used 10-mm-diameter spheres in the right and left primary visual cortex (V1), centered on the MNI coordinates [16,−75,9] and [−19.5,−78,9] (Johnston et al. 2008).

For each participant, the time series were extracted and averaged across voxels within the seed regions with the MarsBaR toolbox (Brett et al. 2002), and the correlations between the seed time series and the time series of all other voxels of the entire brain gray matter mask were calculated, with motion parameters, white matter (WM), and cerebrospinal fluid (CSF) time series as regressors of noninterest. To extract the WM and CSF time series, WM and CSF masks were computed by thresholding the mean of the spatially normalized WM and CSF images (≥1) with ImCalc (SPM8). These masks were then eroded by three voxels along each of the three axes with Anatomist (http://brainvisa.info/). Individual connectivity maps were then transformed into Z-score maps, with connectivity defined as a pairwise correlation between the seed time series and the time series of other voxels.

At the second level, a one-sample t-test was used to establish the amusic and control connectivity maps for each seed, with an FWE-corrected threshold of P < 0.05. Two-tailed, two-sample t-tests were then used to compare Z-score connectivity maps of amusic and control participants. Because effect sizes were expected to be subtle for these between-group contrasts, a threshold of P < 0.001 (uncorrected) was chosen, with a minimum cluster size of 40 mm³.

Correlation analyses were also computed between participants' mean MBEA scores and the mean connectivity strength within a set of ROIs. These ROIs were 10-mm-diameter spheres, centered on the main peaks observed in the between-group connectivity contrast maps.

To attest that the differences of connectivity between amusic and control participants were specific to the Heschl's gyri and would not be observed for other nonauditory areas, we conducted additional control analyses with the two seeds placed in left and right V1, respectively (see Seed determination above). Individual Z-score connectivity maps were generated from these seeds using exactly the same method as for the Heschl's gyri. The main significant clusters obtained from the between-group comparisons of Heschl's gyri-based connectivity maps were isolated, and their connectivity values with V1 seeds were extracted for all participants in 10-mm-diameter spheres. For each ROI, for each seed (Heschl's gyri or V1), connectivity values of amusics and controls were compared with two-sample, two-sided t-tests. We did not expect any between-group difference for V1 connectivity values.