Intrainsular connectivity and somatosensory responsiveness in young children with ASD

Twenty-nine children with ASD and 26 typically developing (TD) children between the ages of 5 and 8 years were recruited into the study. This sample is the same as that reported by Pryweller et al. [12]. After excluding participants with poor image quality resulting from excessive motion or scanner/acquisition errors (ASD n = 5; TD n = 1), the final sample included 23 children with ASD (6.61 years ± 0.89, 2 females) and 24 children with TD (6.58 years ± 1.13, 4 females). Participants in the ASD group were recruited from the university medical center and surrounding community, and a diagnosis of ASD was confirmed with research-reliable administration of the Autism Diagnostic Observation Schedule (ADOS) [50], the algorithm items of the Autism Diagnostic Interview-Revised (ADI-R) [51], and the judgment of a licensed clinical psychologist based on DSM (4th ed.; DSM-IV) [52] criteria. Participants in the TD group were excluded if they had a diagnosed psychiatric or learning disorder or a first-degree relative with ASD. Additionally, TD participants were screened using the Social Communication Questionnaire (SCQ) [53] and the Child Behavior Checklist (CBCL) [54] to rule out risk for ASD and other psychiatric conditions. All participants were screened and excluded for any genetic and neurological problems, previous head injuries, and MRI contraindications.

Participants’ cognitive ability was assessed by trained research assistants using the Kaufman Brief Intelligence Test, Second Edition [55] (n = 47; KBIT-2) or WISC-IV [56] (n = 1; Weschsler et al. 2003), or estimated (n = 1) with the Mullen Scales of Early Learning (MSEL) [57], dependent on the language level of the participant. Although the groups did not differ on sex (χ ² (1) = .145, p = .70) or chronological age (t(43.3) = .085, p = .932), Full Scale IQ estimate was significantly higher in the TD group (t(28.8) = −4.01, p = .0004). See Table 1 for a summary of participant characteristics. Because IQ differed between groups, it was considered as a potential covariate; however, IQ was not significantly correlated with any diffusion variables.

Each child was administered the Tactile Defensiveness and Discrimination Test-Revised [12] (Baranek, 2010, unpublished manual). The Tactile Defensiveness and Discrimination Test-Revised (TDDT-R) is a structured observational assessment including internally controlled (active) touch (e.g., using the palm to rub chalk drawings from a square of carpet, digging pennies out of a box of sand) and externally controlled (experimenter-administered, passive) touch (e.g., light touch to the arm, face, or hand with a cotton swab, air puffs administered to the nape of the neck). Affective behavioral responses are scored for (1) tactile defensiveness both to passive, externally controlled/experimenter-administered touch and to active, internally controlled touch (i.e., haptic exploration of materials such as fabric, sand, or putty) and (2) sensory seeking responses (e.g., squealing with delight while enthusiastically playing with sand, repetitively touching fabric), reflecting unusual interest in tactile stimuli. Defensiveness to externally controlled items is scored using a Likert scale from 0–3, with 0 representing no aversive reaction, and 3 representing negative affect that includes crying or retreating from the stimulus. Defensiveness to internally controlled items is a composite of two scores, one assessing approach or avoidance of the stimulus (0–2 scale) and the other a binary score assessing aversive reaction. Total defensiveness is calculated as a composite (sum) variable collapsing across defensiveness scores for both internally and externally controlled stimuli. For internally controlled items, sensory seeking is scored using a binary present/absent score. Discriminative behavioral responses are scored for passive, externally controlled touch (i.e., localization of the body part touched by experimenter) and active, internally controlled touch (i.e., correct haptic identification of shapes or forms handled by the participant).

All images were acquired during a single scan session on a 3 Tesla Philips Achieva MRI scanner (Philips Healthcare, Inc., Best, Netherlands). During scanning procedures, participants wore foam earplugs in both ears and Philips headphones to attenuate noise and watched a video of their choice for the duration of the scan. A high-resolution T1-weighted anatomical volume (TR = 9 ms, TE = 4.6 ms, FOV = 256 mm [2], 1 mm isotropic voxels, 170 sagittal slices, 6 min 30 s duration) was collected to provide a template for image registration. Diffusion-weighted data were acquired using a high angular resolution diffusion imaging (HARDI) sequence (2.5 mm [2] isotropic voxels, 50 axial slices, 14 min 34 s duration). Ninety-two diffusion directions (b = 1600 s/mm [2]) and one T2-weighted volume (b = 0 s/mm [2]) were collected.

Images were visually inspected for common artifacts such as fat shift and ghosting and underwent standard preprocessing and quality assurance procedures that incorporated head motion, artifact propensity, variance, and bias of estimated measures [58]. A QA rating that weighted each of these measures was assigned, with values between 1 and 5. Only scans with ratings equal to or higher than 3 were included in the analysis (as in (12)). The final sample included in the analysis did not significantly differ in this comprehensive QA metric by group (t(45) = −1.79, p = .08). HARDI data were eddy current-corrected, motion-corrected, and skull-stripped using FMRIB Software Library (FSL) [59, 60]. Fractional anisotropy (FA) maps, eigenvectors, and eigenvalues were created through the DTIFIT module within FSL.

Diffusion parameters were estimated at each voxel through the Bayesian Estimation of Diffusion Parameters Obtained using Sampling Techniques (BEDPOSTX) tool in FSL, which uses Markov chain Monte Carlo sampling techniques to estimate diffusion parameters [61]. Within the right hemisphere, two tracts were defined: an intrainsular tract between anterior and posterior insula and a thalamocortical tract between the VPL nucleus of the thalamus and SI (Fig. 1). For the intrainsular tract, anterior and posterior insular seeds (Fig. 1) were traced with a protocol similar to Farb et al. [38] on a pediatric template derived from 324 children ages 4.5–8.5 [62, 63]. The insula was divided into distinct regions based on cytoarchitectonically defined subdivisions among the insular gyri. The anterior seed mapped onto the anterior accessory gyrus, and the posterior seed mapped onto the posterior long gyrus of the insula. For the thalamocortical tract, VPL was defined by the Oxford Thalamic Connectivity Atlas [64], and SI was defined as the post-central gyrus from the Laboratory of Neuroimaging (LONI) probabilistic atlas gray matter tissue map, excluding voxels below a threshold of 15% probability [65]. VPL and SI ROIs were then registered to the pediatric template, and all ROIs subsequently were registered into native space for each subject using a combination of FSL’s linear image registration tool and linear and non-linear Advanced Normalization Tools (ANTs) [66, 67]. Using these seeds and diffusion parameters generated by BEDPOSTX, probabilistic tracking was conducted using FSL’s ProbTrackX (curvature threshold = 0.2; 5000 samples) one-way condition with VPL and posterior insula as seeds and SI and anterior insula as respective waypoints and termination masks. For the larger thalamocortical tract, low-probability voxels were rejected at a threshold of 1%, similar to previous studies [68, 69]. Tracts were visually inspected to confirm that these thresholds reduced noise while producing viable tracts for analysis. Mean fractional anisotropy (FA), mean diffusivity (MD), and volume of each resultant tract were calculated for each thresholded tract. Tract volumes were normalized for each participant by dividing the value by the total brain volume (total white matter, gray matter, and cerebrospinal fluid on skull-stripped T1-weighted anatomical images).

Our primary questions were (1) whether the integrity of white matter in the intrainsular and thalamocortical tracts differs between children with ASD and TD and (2) whether white matter integrity differences in one or both tracts are associated with aberrant affective and discriminative behavioral responses to touch in individuals with ASD, as measured by the TDDT-R. For all diffusion tensor imaging (DTI) variables (FA, MD, and volume in each of the two tracts), we removed outliers (0–2 datapoints per variable) based on values that were extreme (3 times interquartile range). To address question 1, we assessed group differences in DTI variables (FA, MD, volume) for each tract using two sample t tests. Although it was not a primary research question, we also assessed group differences in variables derived from the TDDT-R: tactile haptic form perception (discriminative touch), tactile defensiveness (negative affective response to touch) to both internally and externally controlled touch, and tactile seeking (presumed positive affective response to touch). Mann–Whitney tests were used for these comparisons because TDDT-R variables were non-normally distributed (Table 2).

To address question 2, we performed nonparametric (Spearman) correlations only between variables (both DTI and behavioral) that had significant differences by group (MD in both tracts and all three TDDT-R variables, see results below). Limiting comparisons to those for which the groups differed significantly allowed us to address the contingency of question 2 on question 1 and to reduce the number of tests performed. To understand how the group interacted with MD in each tract to influence behavior, we conducted a multiple linear regression with the relevant behavioral response score from the TDDT-R as the dependent variable.

To address the possibility of type I error due to multiple comparisons, we conducted permutation testing for significant group comparisons of DTI and TDDT variables as well as TDDT and DTI correlations [70]. A null distribution was calculated for each test statistic across each permutation (n = 5000). A p value was determined based on the proportion of more extreme test statistics compared to the observed test statistic (α = 0.05). The second p value reported for the statistically significant findings reported below reflect this corrected value.

For both the intrainsular and thalamocortical tracts, MD was significantly higher in the ASD group (intrainsular MD t = 2.14, p = .039 (permuted t distribution (n = 5000), p = 0.031)); thalamocortical MD t = 2.3, p = .026 (permuted t distribution (n = 5000), p = 0.024); Fig. 2b, e). Intrainsular volume was marginally lower in the ASD group (t = −1.77, p = .084), and both intrainsular (t = −1.61, p = .116) and thalamocortical (t = −1.35, p = .183) FA showed slight trends to be lower in the ASD group (Fig. 2), consistent with a pattern of reduced tract integrity in ASD. There were no significant group differences or trends in thalamocortical volume.

As measured by the TDDT-R, the ASD group exhibited significantly poorer internally controlled (form) haptic discrimination (W = 384.5, p = .0215 (permuted W distribution (n = 5000), p = 0.017)) relative to the TD group. In ASD relative to TD, defensiveness to internally controlled tactile stimuli (haptic exploration of materials such as sand or putty) was significantly higher (W = 375, p = .0339, (permuted W distribution (n = 5000), p = 0.028)), and defensiveness to externally controlled touch (e.g., experimenter touch with a cotton swab) was also higher (W = 378.5, p = .0257 (permuted W distribution (n = 5000), p = 0.026)). The ASD group exhibited significantly greater tactile seeking behaviors (W = 438.5, p = .0003 (permuted W distribution (n = 5000), p < 0.001)) relative to the TD group.

A significant positive association was observed between intrainsular MD and tactile seeking behavior in the TD group (ρ(22) = 0.49, p = 0.0387 (permuted S distribution (n = 5000), p = 0.016)), whereas this association was negative in the ASD group (ρ(20) = −0.50, p = 0.0185 (permuted S distribution (n = 5000), p = 0.016, Fig. 3)). A Fisher r-to-z transformation confirmed that these correlation coefficients were significantly different (z = 3.47, p = .0005). Given the opposite directions of these associations, we then tested for a group*intrainsular MD interaction using a multiple linear regression with tactile seeking as the dependent variable. Full Scale IQ was also included in the model given the group differences in IQ (see Table 1). The model verified significant main effects of group on seeking and showed a main effect of insula MD (t (36) = −3.676, p = .0008, adjusted total model r ² = 0.4647), but not thalamocortical MD or Full Scale IQ (FSIQ), on tactile seeking (see Table 3). There was a significant interaction between group and insula MD (t (36) = 3.470, p = .0014, adjusted total model r ² = 0.4647).

Neither group exhibited significant associations between tactile discrimination scores and MD in either tract. The TD group showed a negative association between tactile defensiveness and MD in the intrainsular tract (ρ (22) = 0.42, p = 0.0402 (permuted S distribution (n = 5000), p = 0.042)); this association was absent in the ASD group.