We took four experimental approaches to test the hypothesis that variation in LIMK1 affects the IPS in WS. First, we confirmed that the IPS structural and functional anomalies, previously reported largely in adults, are present in children with WS. Second, with longitudinal imaging (structural brain scans acquired at 76 visits over an age range of 5–22 years), we showed that these anomalies are stable from early childhood to adulthood, consistent with an enduring genetic mechanism. Third, to more specifically link these structural and functional features to WS genes, we studied extraordinarily rare individuals having 7q11.23 hemideletions that do not include more telomerically located genes within the WS 7q11.23 locus that are typically hemideleted in WS (see supplementary Fig. S2). Importantly, these “short deletions” (SDs) have in common a hemideletion of LIMK1 (and elastin, but elastin has minimal expression in human brain parenchyma and is specifically not associated with the WS cognitive phenotype) [3]. Fourth, we tested for associations between a LIMK1 haplotype previously associated with gene expression and IPS functional connectivity (based on the three LIMK1 SNPs described above) [14, 34] and cortical organization in a discovery cohort of healthy, well-screened, typically developed members of the general population (GP), and we replicated those findings in a second such cohort. We hypothesized that we would find variations in IPS gray matter structure and/or function associated with LIMK1 hemideletions (in the SD study) and with LIMK1 haplotype variation (in the two general population cohorts). Such associations demonstrate the functional relevance of LIMK1 and might result from developmental alterations in LIMK1 protein availability via hemideletion or SNP-dependent changes in regulated processes such as alternative transcripts and/or splicing.
Cross-sectional study of structural and functional anomalies in children with Williams syndrome
In our initial cross-sectional investigation, we studied 31 children with WS (mean age 9.2 ± 3.2, 21 females) with typical WSCR deletions who had IQs in the normal to low-normal range. A summary of participant demographics for our initial structural study of children is shown in Table 1. For each participant visit, we acquired three 3-Tesla structural scans (GE MR-750, MEMPRAGE, 124 axial slices, TR = 10.5 ms, TE = 1.8 ms, resolution 1 × 1 × 1 mm). These images were intensity normalized [37] and then registered and averaged with AFNI [38] tools to improve signal-to-noise ratios. We used SPM12’s tissue segmentation and applied SPM12 [39] tools to perform diffeomorphic warping to a common space (based on a template balanced for age, sex, and group — WS vs. TD) and generate Jacobian-modulated maps of gray matter in the resulting standard space. After smoothing at 8-mm FWHM, we used AFNI’s 3dttest++ to search for voxel-wise gray-matter differences between the children with WS and typically developing children, covarying for age, sex, and total brain volume.
We performed functional MRI imaging on a subset (N = 12, mean age 11.3 ± 2.3, 11 females) of the children with WS above, on the same 3-Tesla scanner (GE MR-750). During these sessions, participants played a customized version of the video game Tetris, during which the children tried to fit a puzzle piece into a puzzle “landscape” as the piece descended down the screen. The level of difficulty was parameterized to allow for varying abilities and for flexibility in analyses. Following denoising, slice-timing correction, motion correction, warping to a study-specific template, artifact removal [40], and smoothing at 8-mm FWHM, SPM12 was used to perform general linear modeling to localize regions differentially active across the relevant contrast (“difficult” trials vs. “easy” trials) and between the two participant groups. We performed several post hoc analyses that successfully ruled out potentially problematic influences of group differences in sex ratios, task performance, motor responses, and motion.
Longitudinal study of structural and functional anomalies in children with Williams syndrome
In our expanded, longitudinal structural study, we scanned 33 children with WS (76 visits, mean visit age 12.0 ± 4.4, sex distribution of visits 23 M/53 F), acquiring (including TD participants) a total of 1056 MEMPRAGE structural MRI scans from 356 visits. Table 2 summarizes participant demographics for our longitudinal study of children with WS, and Fig. S1 depicts the timelines of visits for participants with WS and typically developing participants. Spatial normalization was performed within individuals and then on a group basis. Specifically, scans for all of an individual’s visits were spatially normalized to create a mid-time-point average image for that participant using SPM12’s longitudinal normalization tool [39]. Advanced Normalization Tools (ANTS) software [41] was used to derive a group template that was balanced for age and sex, and then the mid-time-point average image for each participant was warped to this template. Next, for each participant, ANTS was used to concatenate the deformation fields from the two spatial normalization stages, enabling the transformation to be performed with a single interpolation. Gray-matter images were Jacobian modulated based on the composite deformation field from each visit into the common group template space, followed by smoothing at 8-mm FWHM. From these data, voxel-wise penalized-spline models of longitudinal gray-matter trajectories across the brain were created using a generalized additive mixed-model (GAMM) approach (specifying participant as a random effect) as implemented in R’s gamm4 package [42] and AFNI’s 3dMSS tool [43]. We also performed linear mixed-effects cross-sectional analyses with AFNI’s 3dLME [44] to assess voxel-wise gray-matter differences between the groups, covarying for age, sex, and total brain volume.
For the longitudinal fMRI study, the contrast images (“difficult” trials vs. “easy” trials) from each visit (preprocessed as described above for the cross-sectional study) were used to model voxel-wise penalized-spline-based trajectories and determine group differences as described above for the longitudinal structural study.
All participants in the cross-sectional and longitudinal studies gave informed consent (written consent by parents of minor children and assent by children) according to National Institutes of Health Institutional Review Board guidelines.
Structural and functional anomalies in adults with short deletions
For the short deletion study, we acquired and analyzed structural and functional MRI scans for 12 individuals (mean age = 35.4 ± 13.7, nine females) ascertained by FISH screening of individuals who were referred with suspected WS and cascade screening of siblings and parents of probands [23]. Deletion breakpoints were localized using SNP copy number analyses with Affymetrix 500 K SNP microarrays and/or real-time PCR as described previously [3, 45], and breakpoints were confirmed by PCR amplification and sequencing of deletion junction fragments when possible. Each participant was determined to have one of five hemideletions that included LIMK1 (see Fig. S2). We compared their gray matter patterns and task-based fMRI responses to those of carefully matched controls (see Table 3). Gray matter volume was examined using a voxel-based morphometry (VBM) approach, employing diffeomorphic spatial normalization tools [46] to analyze structural MRI scans. For each participant, we acquired six 1.5 Tesla structural scans (SPGR, 124 axial slices, TR = 12 ms, TE = 5.2 ms, resolution 0.9375 × 0.9375 × 1.2 mm). These images were intensity normalized [37] and then registered and averaged with AFNI [38] tools to improve signal-to-noise ratios. We used SPM8’s tissue segmentation and DARTEL [46] tools to perform diffeomorphic intersubject alignment and generate Jacobian-modulated maps of gray matter in the resulting group-specific space. After smoothing at 6-mm FWHM, we performed an ANCOVA with SPM5 (http://www.fil.ion.ucl.ac.uk/spm/.SPM) to search for voxel-wise gray-matter differences between the groups, covarying for age, sex, and total brain volume.
The fMRI portion of the study challenged the visuospatial system by asking the participant to determine whether two shapes presented on a screen could be assembled to form a square [6]. This “square completion” task block was contrasted with a “shape-matching” task block, in which participants indicated whether two shapes were identical. Blood oxygen level-dependent T2*-weighted gradient-echo echo-planar images (TR = 3 s, TE = 30 ms, FOV = 24 cm, 90° flip, 64 × 64 matrix, 36 contiguous and sequential slices, voxel size 3.75 × 3.75 × 4 mm) were acquired on a 3-Tesla GE scanner with whole-head coil. Stimuli during scanning consisted of pairs of black shapes presented to the left and right of a fixation cross, with task instruction present throughout the block under the cross. For motor control blocks, each presentation was of a matching pair, and the same button was pressed. During “match” blocks, a shape was presented with either an identical copy or its mirror image. Participants pressed one of two buttons depending on whether or not they were the same. During “square completion,” participants were instructed to determine whether or not the two shapes could be assembled to form a square without flipping them over. For sensorimotor “control” trials, participants were shown two matching forms, repeated across all trials, and were asked to simply press a button at presentation. The three task conditions were presented in 16-s-long blocks, with a 2.8-s inter-trial interval. The task was self-paced, with a maximum response window of 7 s per trial. Performance data are summarized in Table 4. Phase-shifted, motion-corrected functional images were aligned to each individual’s structural images and affine transformed into a study-specific standard space averaged across all healthy controls and SD participants. Event covariates for each condition type (square, match, and control) were entered into a general linear model along with motion and drift covariates of no interest. Random effects analyses were used to localize regions differentially active across the different task conditions and between the two participant groups. All participants gave written informed consent according to National Institutes of Health Institutional Review Board guidelines.
General population study: NIMH cohort
We studied a group of 255 healthy, right-handed Caucasian volunteers under the age of 50 years (mean age 33 ± 9.7 [std. dev.]; 113 males, 142 females), all of whom were given a Structured Clinical Interview for DSM-IV (SCID) to rule out the presence of any psychiatric illness, physical and neurological examinations, a battery of neuropsychological tests, and a screening MRI examination. Exclusion criteria included inability to give informed consent, learning disabilities, confounding medical illness, psychiatric diagnosis, or recent (within 3 months) psychotropic substance history, recent (within 1 year) head trauma with loss of consciousness or functional sequelae, and confounding indwelling metal or conditions that would increase MRI risk. Additionally, all subjects were fluent in English. No significant past substance abuse/use disorder history (< 5-year lifetime total) was permitted, and urine toxicology screen was performed at the time of the first study visit.
As previously reported [14], genotyping was performed on Illumina genome-wide SNP chips (550 K–2.5 M SNPs). After performing genotype quality control procedures [49], phasing and imputation were performed using SHAPEIT and Impute2, from which SNPs rs710968, rs146777179, and rs6460071 genotypes were determined for each individual. Additionally, each participant was genotyped with the TaqMan 5′-exonuclease assay for LIMK1 SNP rs710968, which showed 100% concordance with imputed genotypes for this SNP. PHASE software was used to determine 3-SNP haplotype groups. Individuals who were homozygous for all three major alleles (GGC 84.3%) were contrasted against individuals carrying a minor allele for any of these three SNPs. Predicted LIMK1 expression levels were computed using previously reported methods [50]. After LD pruning, we used the score function of plink (version 1.9, https://www.cog-genomics.org/plink2) to weight each LD-independent SNP by the beta value from the LIMK1 brain cortex cis-eQTL analysis from GTEx and create a transcription-based polygenic score for each individual. A t-test was then performed to determine whether the identified haplotype groups were associated with a difference in estimated LIMK1 brain expression.
Structural MRIs were collected on a 1.5 Tesla GE scanner, using a T1-weighted SPGR sequence (TR = 24 ms, TE = 5 ms, flip angle 45°, 0.9375 × 0.9375 × 1.5-mm sagittal acquisition). SPM12’s DARTEL tool was used to create a group-specific template, which was then affine transformed into MNI space. Jacobian modulation maps computed from the DARTEL deformation field were applied to gray matter maps to produce gray matter volume maps. SPM12 was used to perform ANCOVA analysis on the haplotype groups, controlling for age, sex, and total brain size. Results referring to small volume correction (SVC) were obtained by performing family-wise error correction within an a priori volume defined by the original VBM findings for full-deletion WS participants in the IPS region, thresholded at p = 0.001. All participants gave written informed consent as part of protocols approved by National Institutes of Health Institutional Review Boards.
General population study: PNC cohort
This replication cohort consisted of 255 healthy volunteers from the publicly available Philadelphia Neurodevelopment Cohort (PNC) [51] obtained from dbgap (accession number phs000607; mean age 16.0 ± 3.2 [std. dev.]; 125 males, 130 females). Participants were included in this analysis if they had the following: (i) a high-quality structural scan without evidence of significant artifacts based on visual inspection; (ii) high-quality genetic data from an Illumina SNP chip that clustered with the CEU and TSI HapMap3 populations, based on a principal components analysis of all genetic samples; and (iii) no significant past medical or neurological history. MRIs were collected on a 3-Tesla Siemens scanner as described elsewhere [51]. Genotyping, image processing and analysis were as described above for the NIMH GP cohort structural images. Polygenic-based scores for predicted LIMK1 expression were also computed as above for the NIMH GP cohort.
Add Comment