White Matter Abnormalities in Autism and Unaffected Siblings

The DTI data collected were part of a larger set for functional MRI studies in our group. No significant between-group differences were found for age, gender, or handedness (p≥0.05; Table 1). The study included a total of 50 participants: 19 with ASD, 20 US, and 11 TD controls. All participants were screened for neurological disease or insult that could be related to abnormal brain development, including seizures and head injury with loss of consciousness.

For the ASD group, diagnosis was based on parental information, clinical history according to criteria in DSM-IV-TR,⁶ and evaluation by licensed psychologists with expertise in autism. The evaluation consisted of gold standard assessments, including the Autism Diagnostic Interview–Revised and the Autism Diagnostic Observation Schedule administered by a licensed psychologist who is research reliable on both assessments.^7,8 All ASD participants were high functioning, with cognitive abilities group matched to TD participants using the Differential Ability Scale–II.⁹

For the US group, the ASD diagnosis of a sibling was confirmed using DSM-IV-TR.⁶ It was not a criterion that the affected sibling participated in this study, although some sibling pairs were used. An important characteristic of this study is that all US participants were screened and excluded for the broad autism phenotype using the Social Responsiveness Scale (<76), and US and TD participants were group matched based on their scale scores.¹⁰ In addition, a semistructured clinical interview was conducted with US participants to further confirm the absence of autistic traits and to rule out any history of other neurological or psychiatric conditions. This strict exclusion criterion ensures that any identified differences in white matter structure between US and TD groups are not the result of use-dependent processes and are not reflective of subtle differences in social ability. Most US participants’ cognitive abilities were evaluated using the Wechsler Abbreviated Scale of Intelligence.¹¹

TD participants were screened for the broad autism phenotype with the same semistructured clinical interview and Social Responsiveness Scale cutoff score as US participants. In addition, TD participants were screened for history of ASD in first- or second-degree relatives. All group characteristics along with relevant statistics are summarized in Table 1. The Yale Human Investigation Committee approved this study. Informed consent was obtained from a parent/guardian and each participant provided verbal assent.

Three-tesla MRI was performed with the following DTI parameters: directions=30, b0=5, repetition time=6200 ms, echo time=85 ms, field of view=240×240 mm², thickness=2.5 mm (isotropic), number of slices=55, averages=3, and total scan time=11 minutes. Whole-brain T1-weighted MRI was performed using a sagittal 1-mm³ magnetization-prepared rapid gradient echo sequence. All imaging was performed in the same session after completing mock MRI session(s).

The quality of diffusion MRI data was ensured through a multistep procedure that addressed eddy current distortion, head motion, and signal dropout. First, diffusion-weighted images were corrected for eddy current distortion and head motion. Second, all volumes (i.e., diffusion directions) were individually inspected for head motion and signal dropout by an experienced rater (R.J.J.) who performed DTI analyses in previous studies.^5,28 Volumes with excessive motion or dropout were spliced out and excluded from the analysis. If more than six volumes required exclusion, then the entire run was dropped from the analysis. Using this method, one participant from the TD group and one participant from the ASD group were excluded from the analysis; no participants were excluded from the US group. Next, multiple runs (per subject) of diffusion-weighted data were averaged to improve the signal-to-noise ratio. FA maps were individually assessed to confirm absence of artifact or distortion.

All data processing and local diffusion modeling were conducted using the FMRIB Software Library (FSL) (fsl.fmrib.ox.ac.uk/fsl/). The program was used to generate a mask to separate brain areas from nonbrain areas. Using the FMRIB’s Diffusion Toolbox (FDT), a diffusion tensor model was created and FA, axial diffusivity, and radial diffusivity were computed for each voxel.

Three group-wise comparisons were made: 1) ASD versus TD, 2) US versus TD, and 3) ASD versus US. Voxel-wise analysis of multisubject diffusion data was conducted using tract-based spatial statistics (TBSS). The procedure consisted of the following steps: 1) application of nonlinear registration so that all images were in Montreal Neurological Institute space, 2) creation of a mean image, 3) skeletonization of the mean image, 4) projection of all subjects’ diffusion imaging data onto the mean skeleton, and 5) submission of the four-dimensional–projected data for statistical testing. Areas of significant difference were identified (p<0.05 corrected for multiple comparisons across space using threshold-free cluster enhancement or TFCE). Voxels with significant differences in FA values were labeled as “affected voxels,” and only these voxels were used in the comparison of FA, axial diffusivity, and radial diffusivity between groups. Using the T1-weighted MRI data, intracranial volume was calculated using the FreeSurfer image analysis suite (surfer.nmr.mgh.harvard.edu/).

Affected white matter structures (fiber tracts) were identified using the Johns Hopkins University White Matter Tractography Atlas. This atlas was not only useful for identifying potentially affected fiber tracts, but it was also useful for more precise characterization of potential abnormality by quantifying affected voxels based on their tract labels. The Montreal Neurological Institute coordinates of all affected and unaffected voxels were captured and intersected with the aforementioned atlas. These voxels were assigned to atlas labels linking them to their respective association, commissural, or projection tracts. For each participant, affected voxels were grouped by label with their corresponding metrics (FA, axial diffusivity, or radial diffusivity), and counts were generated for each fiber tract. Summary statistics for FA, axial diffusivity, and radial diffusivity were computed for each fiber tract, first by individual and then by group. Percent abnormality was calculated for each tract by dividing the number of affected voxels by the total number of voxels compared. This represents the degree of white matter abnormality by tract.

Post-hoc Pearson's correlation analyses were performed between mean FA within affected fibers and SRS total scores. Where appropriate, Bonferroni correction implemented for multiple comparisons.

The ASD group exhibited significant bilateral reductions in FA in all examined tracts, including association, commissural, and projection tracts (Figure 1 [A], Table 2). FA abnormalities were widespread; however, not all tracts were equally affected (Figure 2). Tracts with the highest percentage of affected voxels included association and commissure fibers. The most affected structures included the uncinate fasciculus, forceps minor, and inferior fronto-occipital fasciculus. The ASD group had significant bilateral increases in radial diffusivity involving the association, commissural, and projection tracts (Figure 1 [B]). Distribution of these abnormalities was similar to that observed for FA. There were no significant group differences in axial diffusivity. Notably, there were no voxels in which FA was significantly increased or radial diffusivity was significantly decreased in the ASD group. Exploratory Pearson’s correlation analyses did not yield significant relationships between mean FA within affected fibers and Social Responsiveness Scale total scores (Table 2).

The US group had significant bilateral FA reductions in all but two tracts (uncinate fasciculus and forceps minor), including the association, commissural, and projection tracts (Figure 1 [C], Table 3). Distribution of FA anomalies was widespread; however, not all tracts were equally affected (Figure 2). Those tracts with the highest percentage of affected voxels were different compared with the ASD and TD relationship described earlier. Specifically, abnormalities were skewed more toward projection and away from association fibers. Moreover, the severity by which each was affected was much lower. More severely affected structures for the US group included the corticospinal tract, forceps major, and anterior thalamic radiation. The US group had significant bilateral increases in radial diffusivity involving the association, commissural, and projection tracts. There were some differences in the distribution of these abnormalities compared with that observed for FA. For radial diffusivity, tracts with the highest number of affected voxels included the superior longitudinal fasciculus, forceps major, and inferior fronto-occipital fasciculus. There were no significant group differences in axial diffusivity. Notably, there were no voxels in which FA was significantly increased or radial diffusivity was significantly decreased in the US group. Exploratory Pearson’s correlation analyses did not yield any significant relationships between mean FA within affected fibers and Social Responsiveness Scale total scores after correcting for multiple comparisons (Table 3).

Differences between the ASD and US groups were not statistically significant with respect to FA, radial diffusivity, and axial diffusivity. However, as demonstrated in Figure 2, when each group was compared with the TD group, the degree to which each tract was affected differed between the ASD and US groups.