Williams syndrome hemideletion and LIMK1 variation both affect dorsal stream functional connectivity

In Williams syndrome, a condition marked by hypersociability and visuospatial impairment, Gregory et al. show that intraparietal sulcus functional connectivity is increased with social processing regions and decreased with visual processing regions. Variation in LIMK1, which is hemideleted in Williams syndrome, is also associated with functional connectivity patterns in healthy controls.

Abstract

Williams syndrome is a rare genetic disorder caused by hemizygous deletion of ∼1.6 Mb affecting 26 genes on chromosome 7 (7q11.23) and is clinically typified by two cognitive/behavioural hallmarks: marked visuospatial deficits relative to verbal and non-verbal reasoning abilities and hypersocial personality. Clear knowledge of the circumscribed set of genes that are affected in Williams syndrome, along with the well-characterized neurobehavioural phenotype, offers the potential to elucidate neurogenetic principles that may apply in genetically and clinically more complex settings. The intraparietal sulcus, in the dorsal visual processing stream, has been shown to be structurally and functionally altered in Williams syndrome, providing a target for investigating resting-state functional connectivity and effects of specific genes hemideleted in Williams syndrome. Here, we tested for effects of the LIMK1 gene, deleted in Williams syndrome and important for neuronal maturation and migration, on intraparietal sulcus functional connectivity. We first defined a target brain phenotype by comparing intraparietal sulcus resting functional connectivity in individuals with Williams syndrome, in whom LIMK1 is hemideleted, with typically developing children. Then in two separate cohorts from the general population, we asked whether intraparietal sulcus functional connectivity patterns similar to those found in Williams syndrome were associated with sequence variation of the LIMK1 gene. Four independent between-group comparisons of resting-state functional MRI data (total n = 510) were performed: (i) 20 children with Williams syndrome compared to 20 age- and sex-matched typically developing children; (ii) a discovery cohort of 99 healthy adults stratified by LIMK1 haplotype; (iii) a replication cohort of 32 healthy adults also stratified by LIMK1 haplotype; and (iv) 339 healthy adolescent children stratified by LIMK1 haplotype. For between-group analyses, differences in intraparietal sulcus resting-state functional connectivity were calculated comparing children with Williams syndrome to matched typically developing children and comparing LIMK1 haplotype groups in each of the three general population cohorts separately. Consistent with the visuospatial construction impairment and hypersocial personality that typify Williams syndrome, the Williams syndrome cohort exhibited opposite patterns of intraparietal sulcus functional connectivity with visual processing regions and social processing regions: decreased circuit function in the former and increased circuit function in the latter. All three general population groups also showed LIMK1 haplotype-related differences in intraparietal sulcus functional connectivity localized to the fusiform gyrus, a visual processing region also identified in the Williams syndrome-typically developing comparison. These results suggest a neurogenetic mechanism, in part involving LIMK1, that may bias neural circuit function in both the general population and individuals with Williams syndrome.

Introduction

Williams syndrome is a rare genetic disorder caused by a well-delineated hemideletion of ∼1.6 Mb on chromosome 7 (7q11.23). Individuals with Williams syndrome are distinguished by a highly specific and sensitive neurobehavioural phenotype, including a characteristically hypersocial personality combined with significant anxiety in non-social situations (Klein-Tasman and Mervis, 2003) and a cognitive profile of severe visuospatial construction impairment together with relative strengths in language and facial recognition (Mervis et al., 1999, 2000; Meyer-Lindenberg et al., 2006; Mervis and John, 2010). This pattern of relative strengths and weaknesses is frequently contrasted with that in autism, in which individuals typically have impaired language, social skills, and face processing, often along with sparing of visuospatial abilities (Fishman et al., 2011; Sanders et al., 2011). Because of the sensitive and specific personality and cognitive profiles in Williams syndrome, and because, in contrast to most complex neuropsychiatric disorders, the affected genes are known, understanding the neural substrate of this syndrome offers an unusual opportunity to provide key information about how gene effects are translated in the brain to produce complex human characteristics (Tager-Flusberg et al., 2006).

Past research has implicated the dorsal visual processing stream as having a central role in the visuospatial deficits seen in individuals with Williams syndrome. Perhaps the most replicated neuroimaging finding in Williams syndrome is a structural anomaly early in the dorsal stream, specifically in the intraparietal sulcus (IPS) (Kippenhan et al., 2005; Campbell et al., 2009; Jackowski et al., 2009; Morris, 2010; Menghini et al., 2011; Fahim et al., 2012). Later in the dorsal stream, downstream from the IPS, task-based functional neuroimaging studies have found that individuals with Williams syndrome show decreased activation in response to visual processing demands (Meyer-Lindenberg et al., 2004; Mobbs et al., 2007; Sarpal et al., 2008; Atkinson and Braddick, 2011). These task-based findings suggest that the IPS anomalies are functionally propagated throughout the visual processing system.

Three prior studies have found altered functional connectivity in Williams syndrome using resting functional MRI. Findings included lower within-network and increased between-network connectivity (Vega et al., 2015; Gagliardi et al., 2018) or decreased connectivity within the default mode network (Sampaio et al., 2016). However, despite the importance of the IPS region in Williams syndrome, functional connectivity of this important hub has not specifically been tested and results have not been followed up as a function of SNPs in the general population. We hypothesized that participants with Williams syndrome would have decreased functional connectivity between the IPS and other visual processing regions as a feature of their neurofunctional topography as measured with functional MRI, even at rest. Further, because any identified alteration in neurofunctional circuitry involving the IPS would provide a strong target for investigating neurogenetic mechanisms relevant to Williams syndrome, we sought to clarify the role of a specific gene in the Williams syndrome 7q11.23 region.

While 26 genes in the 7q11.23 region are hemideleted in Williams syndrome, findings from exceedingly rare individuals with smaller hemideletions of this region shed light on which of these genes may be especially important in the Williams syndrome phenotype. Although not all studies of individuals with partial deletions of the Williams syndrome region agree (Gray et al., 2006), data from individuals with only the LIM domain kinase 1 (LIMK1) and elastin (ELN) genes affected are informative. Several such individuals have been found to have visuospatial construction problems, and since ELN has minimal expression in human brain parenchyma (Frangiskakis et al., 1996; Petryszak et al., 2016), LIMK1 may be particularly important in the visuospatial deficits that characterize individuals who have classic Williams syndrome (full hemideletion of the Williams syndrome region of 7q11.23). A mechanistic role for LIMK1 is biologically plausible because its molecular function is involved in regulating actin stabilization, hence regulating neuronal and axonal migration (Scott and Olson, 2007; Dong et al., 2012). Additionally, mice with selective knockout of the Limk1 gene show impaired spatial learning, consistent with the visuospatial construction problems in Williams syndrome (Meng et al., 2002). A haplotype in the promoter region of LIMK1, consisting of three single nucleotide polymorphisms (SNPs)—one −961 bp from the initiation site (rs6460071), another −428 bp from the initiation site (rs710968), and a third −187 bp from the initiation site (rs146777179)—has been associated with LIMK1 transcription by in vitro functional experiments (Akagawa et al., 2006), providing a potential genetic marker to test in vivo. Given these past findings, we tested for effects of LIMK1 variation in four cohorts. First, we characterized resting state functional connectivity of the IPS in children with Williams syndrome and the classic 7q11.23 copy number variation (CNV) as compared to matched typically developing children. Then, we tested for the effects of the LIMK1 promoter haplotype on IPS functional connectivity in two independent cohorts of healthy adults in the general population and one publicly-available paediatric cohort, each stratified by haplotype. We hypothesized that this sequence variation would be associated with differential IPS functional connectivity patterns similar to those seen with the full Williams syndrome hemideletion (which includes LIMK1), particularly in visuospatial processing regions, thus providing convergent evidence for effects of the LIMK1 gene.

Materials and methods

Participants and functional MRI data acquisition

Participants in this study constituted four separate samples. Three cohorts were studied at the National Institutes of Health (NIH) Clinical Center: (i) 20 children with Williams syndrome compared to 20 age- and sex-matched typically developing children; (ii) a discovery cohort of 99 healthy adults stratified by LIMK1 haplotype; (iii) a replication cohort of 32 healthy adults also stratified by LIMK1 haplotype; and (iv) data from a cohort of 339 healthy adolescent children from the Philadelphia Neurodevelopment Cohort (PNC) stratified by LIMK1 haplotype that is publicly available on dgGaP (accession number phs000607). Study procedures were approved by the NIH Combined Neuroscience IRB. Adults and parents of minor participants provided written informed consent and children provided assent. Demographic characteristics of the four samples are reported in Table 1.

Table 1.

Demographic characteristics of participants in each group

	n	Age, years	Sex M/F	LIMK1 Genotype	KBIT IQ
Williams syndrome	20	12.4 (±4.8)	6/14	LIMK1 CNV (1 copy)	91 (±11)
Typically developing	20	12.2 (±4.3)	6/14	2 LIMK1 copies	117 (±13)
PNC adolescent cohort	339	15.6 (±3.3)	167/172	239 CC	N/A
				100 TC/TT
Adult discovery cohort	99	32.6 (±9.8)	43/56	74 LIMK1 Hap1	N/A
				25 LIMK1 Hap2
Adult replication cohort	32	39.5 (±13.0)	16/16	24 LIMK1 Hap1	N/A
				8 LIMK1 Hap2

NIMH Williams syndrome sample and matched control group

The National Institute of Mental Health (NIMH) Williams syndrome sample and a matched control group, studied at the Intramural Research Program from 2010 to 2016, consisted of 20 children with genetically confirmed classic Williams syndrome hemideletions (mean age 12.4 ± 4.8 years, range 5.5–20.3 years, 14 females) and 20 typically developing children (mean age 12.2 ± 4.3 years, range 5.4–20.2 years; 14 females) individually matched for age and sex. All participants were in good physical health, based on physical examination and a review of available medical records, had IQs in the low-normal to normal range, and had intact cerebral vasculature based on a radiologist’s reading of magnetic resonance angiography. Each participant underwent two 6.13-min resting-state 3 T blood oxygenation level-dependent (BOLD) echo planar imaging (EPI) scans (12.26 min total; repetition/echo time: 2000/24 ms, flip angle: 77°, voxel size: 1.875 × 1.875 × 3 mm, 184 volumes each). Participants were instructed to lie still in the scanner and to keep their eyes open, and they were monitored for wakefulness throughout the scan. A T₁-weighted multi-echo MPRAGE structural scan (repetition/echo time: 10.5/1.8 ms, flip angle: 7°, voxel size: 1 × 1 × 1 mm, 176 slices) was also collected for image registration purposes.

Discovery cohort

The discovery cohort from the general adult population consisted of healthy Caucasian individuals of European descent (n = 99; mean age 32.6 ± 9.8 years, range 19–57 years; 56 females), who were also studied at NIMH from 2005 to 2015, independently from the other samples. In this group, we also tested resting-state connectivity with functional MRI, this time as a function of a haplotype spanning the LIMK1 promotor region, to determine whether LIMK1 sequence variation was associated with connectivity changes resembling those seen in the Williams syndrome CNV. These participants underwent one run of 7.98 min of resting BOLD EPI scanning at 3 T (repetition/echo time: 1596/28 ms, flip angle: 90°, voxel size: 4 × 4 × 5 mm, 300 volumes) and were also instructed to lie still and keep their eyes open during the resting scan.

Replication cohort

The replication cohort, also from the general population, consisted of healthy Caucasian adults of European descent (n = 32; mean age 39.5 ± 13.0 years, range 22-61 years; 16 females), studied at NIMH from 2010 to 2016 independently from the other samples. This slightly older group was studied to test for converging evidence of LIMK1 promotor haplotype effects found in the discovery sample. As in the Williams syndrome cohort, these participants also underwent two runs of 6.13 min of resting BOLD EPI scanning at 3 T (12.26 total minutes; repetition/echo time: 2000/24 ms, flip angle: 77°, voxel size: 1.875 × 1.875 × 3 mm, 184 volumes), during which they were instructed to lie still in the scanner and to keep their eyes open. A T₁-weighted multi-echo MPRAGE sequence (repetition/echo time: 10.5/1.8 ms, flip angle: 7°, voxel size: 1 × 1 × 1mm, 176 slices) was also collected for co-registration purposes. These data were collected on the same scanner and with the same acquisition parameters as the Williams syndrome-typically developing cohort.

Philadelphia neurodevelopmental cohort

Finally, the publicly-available PNC, obtained from dgGaP (accession number phs000607) (Satterthwaite et al., 2014), provided a third general population sample to test for evidence of LIMK1 haplotype effects in a developing sample of 339 healthy adolescent children (15.6 ± 3.3 years, 172 females) using resting-state functional MRI data and genetic information. Each participant in the PNC sample underwent 6.2 min of resting BOLD EPI scanning (repetition/echo time: 3000/32 ms, flip angle: 90°, voxel size: 3 × 3 × 3 mm, 124 volumes), as well as a T₁-weighted MPRAGE structural scan (repetition/echo time: 1810/3.51 ms, flip angle: 9°, voxel size: 0.9375 × 0.9375 × 1 mm), which was used for registration purposes. Participants were included in this analysis if they had: (i) a high-quality structural scan without evidence of significant artefacts based on visual inspection; (ii) high-quality resting functional MRI data not corrupted by artefacts, based on Artifact Detection Tools (ART; www.nitrc.org/projects/artifact_detect); (iii) 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 (iv) no significant past medical or neurological history.

LIMK1 genetic analyses in the adult samples

In the PNC sample, genetic data were acquired on multiple different SNP chips. Only data from the Illumina SNP chips were used in this analysis. Participants from both NIMH adult samples were genotyped on Illumina SNP chips (550K-2.5M SNP chips). Genotype information for rs6460071, rs710968, and rs146777170 for each participant was obtained from an imputed genome created for the NIMH and PNC samples using parallel methods. Pre-imputation QC procedures for each SNP chip were performed separately, based on previously reported methods (Anderson et al., 2010). Only participants found to cluster with HapMap3 CEU and TSI populations were retained for further analysis to ensure results were not related to ancestry. Prior to imputation, phasing was performed using Shapeit version 2.2 (Delaneau et al., 2013), and then imputation was performed on each chip separately using IMPUTE2 (Howie et al., 2009). For the largest, densest chip in each sample, imputation was performed using the 1000 Genomes Phase 3 data as a reference panel (Genomes Project et al., 2015). For all other chips, imputation was performed using the imputed result of the Omni2.5M chip (NIMH sample) or OmniExpress chip (PNC sample) as a reference panel. SNP concordance rates for all chips imputed were 98%. Individual chips were then combined to yield a final imputed genome for each sample, such that only SNPs with high quality imputation (INFO > 0.9 and Certainty >0.9) on all chips were retained. All three SNPs forming the haplotype associated with LIMK1 transcription (rs6460071, rs710968, and rs146777179) satisfied these criteria and were extracted. PHASE v2.0 was used to calculate haplotype groups for all participants using these three SNPs (Stephens and Donnelly, 2003). Two main haplotype groups were identified: those homozygous for all major alleles (Hap1; GGC homozygotes) and those carrying a minor allele for at least one of the three SNPs (Hap2).

Image preprocessing

Neuroimaging data from the NIMH Williams syndrome sample, the PNC cohort, and the adult replication cohort were processed with the same analytic pipeline. Anatomical scans were corrected for intensity non-uniformity using n3 normalization (Sled et al., 1998). For resting-state functional MRI, the first five images were discarded and the remaining images were slice-time corrected and motion corrected using AFNI (Cox, 1996). Each participant’s structural scan was co-registered to his or her functional images. Co-registered structural scans were nonlinearly warped to a structural template in MNI space using Advanced Normalization Tools (ANTs) (Avants et al., 2011).

For the NIMH Williams syndrome sample and matched typically developing controls, a template was constructed to be equally-representative of typically developing children and children with 7q11.23 CNV to minimize potential differences in the degree of spatial warping between groups (Buckner et al., 2004; Huang et al., 2010; Weng et al., 2015). For the PNC cohort, a template was created in-house from the PNC structural images. For the replication cohort, the template was created in-house from scans of 240 healthy adults. ART was used to exclude time points in the functional scans with a global signal more than 3 standard deviations (SD) from the mean across all time points, or an absolute movement between images >0.5 mm (Whitfield-Gabrieli and Nieto-Castanon, 2012). Functional MRI data were residualized with respect to signals of no-interest, censored for excluded time points, bandpass filtered (0.008 Hz < f < 0.1 Hz) and spatially smoothed using a Gaussian kernel of 6 mm full-width at half-maximum using AFNI tools. Signals of no-interest included those based on anatomical component-based noise correction (aCompCor) (Behzadi et al., 2007; Chai et al., 2012), as well as the six-direction residuals of motion.

Neuroimaging data from the discovery cohort were analysed with the same procedures as the other three cohorts, except for normalization procedures. As T₁ anatomical scans were not available for all participants in this cohort, EPI images were normalized directly to the MNI-EPI template using SPM5.

Functional connectivity analysis

To calculate IPS functional connectivity throughout the brain we used a 6-mm radius spherical seed region placed at the peak of a structural anomaly previously reported in an independent group of individuals with Williams syndrome (Meyer-Lindenberg et al., 2004), warped to the respective template (7q11.23 CNV: MNI_x,y,z = 30, −64, 34; healthy adult template: MNI_x,y,z = 30, −68, 32; PNC: MNI_x,y,z = 30, −63, 34). Using AFNI, functional connectivity z-score maps were created for each participant to determine the strength of the correlation of the IPS BOLD signal time course with the time course of each voxel in the brain. In the Williams syndrome versus typically developing sample, linear regression was performed to test for differences between diagnostic groups, while controlling for effects due to age and sex. In each healthy sample, linear regression was performed between LIMK1 haplotype groups (Hap1 versus Hap2), also while controlling for effects due to age and sex. For each of the three between-group analyses, the data were first thresholded at a voxelwise P < 0.005, uncorrected, and then cluster-corrected for multiple comparisons to a family-wise error corrected P-value < 0.05, using 3dClustSim to compute a cluster threshold based on 100 000 Monte Carlo simulations of synthesized white Gaussian noise, taking into account the smoothing and resampling parameters of the functional connectivity analyses using the ACF method (Cox et al., 2017).

Assessment of the effects of movement on resting state functional MRI

Recent work has demonstrated that resting connectivity is particularly susceptible to artefacts stemming from subject motion (Power et al., 2012; Satterthwaite et al., 2012; Van Dijk et al., 2012) and that this relationship can spuriously increase the apparent connectivity of short-range, local connections and decrease connectivity of long-range connections (Power et al., 2012; Satterthwaite et al., 2012; Ciric et al., 2017). Because effects of participant movement may be particularly important for studies of paediatric and clinical groups, and because there may have been differential movement between our participant cohorts, we used four approaches to ensure the results were not related to motion.

First, we calculated the mean framewise displacement and the scaled delta variation of signal (DVARS) for each individual’s resting scan and compared these measures across groups. Second, we examined the effects of scrubbing (i) by comparing the number of volumes scrubbed across groups; and also (ii) by comparing the voxel-wise distribution of the T-statistics from the Williams syndrome versus typically developing IPS connectivity analyses before scrubbing to the distribution after scrubbing. Third, as motion has been shown to relate to functional connectivity in a distance-dependent manner, we examined the distance dependence of IPS connectivity in our sample by performing a linear regression to model the association of IPS connectivity with framewise displacement at each voxel; we then plotted the distribution of the associations for all voxels as a function of their distances from the IPS seed region. We also compared this IPS connectivity-framewise displacement relationship across groups (Williams syndrome versus typically developing) using a voxel-wise linear regression. Fourth, from each group we identified 10 participants who had the least head movement, thus creating subgroups who did not differ in motion parameters; then, to determine whether the Williams syndrome versus typically developing IPS connectivity results in the full groups would also be found in a smaller group where movement was not a confound, we performed the same voxel-wise, between-group comparison as had been done across the entire cohorts.

Data availability

Raw data from the PNC are publicly available through dgGaP (accession number phs000607). Derived data are available within the article.

Results

The NIMH Williams syndrome sample versus typically developing children

We first calculated functional connectivity of the IPS in the NIMH typically developing children and then determined whether individuals with Williams syndrome exhibit an alteration in this connectivity pattern. As expected and consistent with previous reports from the general population (Uddin et al., 2010), the IPS seed region (Fig. 1A, shown in red) had robust positive functional connectivity with the dorsal and ventral visual processing streams and the frontal cortex in typically developing children (Fig. 1B). There were no regions that showed significant negative functional connectivity with the dorsal stream IPS seed.

Figure 1.

IPS functional connectivity in typically developing children. (A) Axial view of the IPS seed region, shown in red, used in the analyses, displayed on a template equally-representative of children with 7q11.23 copy-number variation and typically developing children. (B) Functional connectivity of the IPS in the 20 NIMH typically developing child participants, displayed on the brain surface, showing regions where the IPS seed showed positive functional connectivity (oranges and yellows); there were no regions where the IPS was negatively connected. Results are displayed at P < 0.05 after familywise error correction.

Compared to this matched typically developing group, children with Williams syndrome showed decreased IPS functional connectivity with several visual processing regions, including the right and left fusiform and occipital gyri and the right superior parietal lobule (Fig. 2A, blue). Surprisingly, as seen in Fig. 2, in the Williams syndrome group we also observed robustly increased connectivity of the IPS dorsal visual processing stream seed with regions not thought to be related to visual processing and not part of the typical IPS connectivity network that has been consistently documented in the literature (Uddin et al., 2010; Kravitz et al., 2013). These included the posterior cingulate (PCC), bilateral medial prefrontal cortex (MPFC), bilateral temporo-parietal junction (TPJ), right orbitofrontal cortex, bilateral superior frontal gyri and right inferior temporal gyrus (Fig. 2A and Table 2). Interestingly, all of the regions showing increased IPS connectivity in Williams syndrome have been previously implicated in the neurobiology of social processing (Frith and Frith, 2007; Blakemore, 2008; Gotts et al., 2012), whereas all of the regions showing decreased IPS connectivity in Williams syndrome are known to subserve visual processing functions (Ungerleider and Mishkin, 1982; Ungerleider and Haxby, 1994; Kravitz et al., 2013). Follow-up examination revealed that these socially-related regions had minimal IPS functional connectivity in the typically developing group, but positive connectivity to the IPS in Williams syndrome, whereas the opposite relationship was found for visual processing regions (Fig. 2B). Additionally, we found that 95% of the voxels showing increased IPS connectivity in Williams syndrome fell outside the IPS connectivity network found in the typically developing cohort, whereas 99.5% of the voxels showing decreased IPS connectivity in Williams syndrome resided within the typical IPS-connected network.

Figure 2.

Differences in IPS functional connectivity in children with Williams syndrome versus typically developing children. (A) Surface views of contrasting IPS functional connectivity in Williams syndrome versus typically developing children. Orange colours represent increased connectivity in children with Williams syndrome compared with typically developing children; blues represent decreased connectivity in children with Williams syndrome. Results are shown at P < 0.05 after familywise error correction. (B) Comparison of IPS functional connectivity in visuospatial and social regions in Williams syndrome and typically developing groups. Y-axis shows the average residualized z-scores of voxels within significant visuospatial and social regions separately.

Table 2.

Brain regions showing differential IPS connectivity with LIMK1 variation

Region	Cluster size, mm3	x	y	z	Peak t-score
Williams syndrome versus typically developing (LIMK1 hemideletion)
Posterior cingulate cortex (BA 23)	18 088	15	−53	30	5.5
Right superior frontal gyrus (BA 8)	13 816	15	37	51	6.9
Left superior frontal gyrus (BA 8)	12 603	−24	28	54	5.4
Left medial prefrontal cortex (BA 10)	11 264	−9	58	0	7.0
Right medial prefrontal cortex (BA 10)	10 314	11	63	9	6.7
Right superior occipital gyrus (BA 18/19)	8216	32	−92	21	−5.2
Right temporo-parietal junction (BA 39)	7109	54	−66	35	6.1
Left middle occipital gyrus (BA 18/19)	6844	−41	−96	9	−5.5
Right inferior temporal gyrus (BA 20)	4862	60	−17	−36	5.4
Left temporo-parietal junction (BA 39)	4366	−45	−77	39	5.5
Right orbito-frontal cortex (BA 47)	3185	47	47	−18	5.6
Right superior parietal lobule (BA 7)	2457	23	−60	54	−4.4
Right medial prefrontal cortex (BA 11)	2384	11	43	−12	4.4
Left orbito-frontal cortex (BA 47)	2257	−41	35	−18	5.8
Right fusiform gyrus (BA 37)	1561	49	−59	−18	−4.3
PNC cohort (Hap1 versus Hap2)
Posterior cingulate cortex	945	−9	−66	18	3.5
Right fusiform gyrus (BA37)	864	42	−63	−33	−3.8
Adult discovery cohort – LIMK1 haplotype (Hap1 versus Hap2)
Right fusiform gyrus (BA 37)	918	54	−60	−9	−3.8
Adult replication cohort – LIMK1 haplotype (Hap1 versus Hap2)
Right fusiform gyrus (BA 37)	1075	53	−57	−6	−5.9

Effects of movement on the Williams syndrome versus typically developing analysis

As head movement has been a significant confound in the resting state functional MRI literature, we performed multiple analyses to ensure our results were not driven by motion. First, we tested for between-group differences in framewise displacement and DVARS. We did not find significant differences in motion, as measured by framewise displacement, between groups (P > 0.12), though the group of children with Williams syndrome nominally had more head motion than the typically developing cohort (mean framewise displacement: Williams syndrome = 0.1 ± 0.07 mm, typically developing 0.08 ± 0.04 mm). Additionally, there was no significant difference in the mean DVARS between groups (P > 0.22).

Next, we examined the effects of scrubbing and found that the number of volumes censored out because of motion was higher in children with Williams syndrome than in typically developing children (12% ± 7% scrubbed volumes versus 7% ± 5% scrubbed volumes, respectively, P = 0.022). Prior to scrubbing, the voxel-wise distribution of the T-values from the Williams syndrome versus typically developing connectivity analysis was shifted to the right, such that children with Williams syndrome had, on average across the entire brain, greater IPS connectivity than typically developing children (i.e. there was spuriously increased cross-time course covariance due to motion in the Williams syndrome cohort). However, after scrubbing, this effect was normalized, with the mean of the distribution becoming very close to zero (Supplementary Fig. 1).

Further, as motion has been shown to induce a distance-dependence on resting-state functional MRI connectivity measures, we examined the effect of distance from the IPS seed on the relationship between IPS connectivity and framewise displacement at each voxel in our entire sample, both before and after scrubbing. Before scrubbing, a distance-dependence was indeed present, resulting in a negative relationship between IPS-connectivity and framewise displacement that was substantially similar to prior reports (R² = 0.26, Supplementary Fig. 2) (Satterthwaite et al., 2012). However, after scrubbing, the correlation between distance and the connectivity-framewise displacement relationship was minimized (R² = 0.015, Supplementary Fig. 2); this correlation, if anything, was in the opposite direction to (i.e. a positive, not negative correlation) and more than an order of magnitude less than what has been reported previously in the literature [e.g. R² ∼0.25 in Satterthwaite et al., (2012)], suggesting that the effect in this sample after processing was, indeed, minimal. We also tested whether any brain regions exhibited a significant relationship between IPS connectivity and framewise displacement in a voxel-wise analysis, both within- and between-groups, and found none.

Finally, we sought additional evidence that our findings were not related to group differences in head motion by repeating our Williams syndrome-versus-typically developing IPS connectivity analysis using only the 10 children with the least motion from each cohort. There were no significant differences in mean framewise displacement or number of scrubbed volumes between these Williams syndrome and typically developing subgroups (both P’s > 0.35), and, importantly, the identified differential circuitry closely recapitulated the findings in the entire groups (Supplementary Fig. 3).

As there may be maturational changes that accompany typical and atypical development, we tested for effects of age on both the Williams syndrome versus typically developing IPS connectivity contrast and on IPS connectivity in each of our Williams syndrome and typically developing groups separately. IPS connectivity was not significantly associated with age in any of these analyses.

IPS functional connectivity related to LIMK1 promotor variation

Having identified IPS connectivity alterations related to hemideletion of the 7q11.23 Williams syndrome region (including LIMK1) in patients with Williams syndrome, we next sought to determine whether the LIMK1 three-SNP haplotype previously associated with transcription rate (Akagawa et al., 2006) was also associated with differential IPS functional connectivity in similar brain regions in participants from the general population. We first tested the connectivity pattern in a discovery cohort of 99 healthy adults. IPS connectivity in the entire group revealed a robust network comprising the dorsal and ventral visual processing streams and frontal cortex, similar to the pattern seen with typically developing children in the prior analysis. Genotyping of this initial sample revealed that all three LIMK1 SNPs were in Hardy-Weinberg equilibrium. Computation of haplotype groups with PHASE (Stephens and Donnelly, 2003), identified 74 participants with Hap1 (i.e. GGC homozygotes) and 25 participants with Hap2. Between-haplotype group analysis revealed that Hap1 participants, relative to Hap2, had decreased connectivity between the IPS and an area nearly identical to one with the same connectivity pattern in the Williams syndrome versus typically developing analysis: the right posterior fusiform gyrus (Fig. 3, top). No other regions showed differential connectivity related to haplotype groups.

Figure 3.

IPS connectivity in the discovery and replication cohorts based on LIMK1 haplotype. Results are shown at P < 0.05 after family-wise error correction. Note the high degree of anatomical overlap in the right fusiform gyrus (arrow) related to LIMK1 haplotype in the discovery and replication cohorts.

In our replication adult cohort, genotyping also revealed all three SNPs to be in Hardy-Weinberg equilibrium. Twenty-four participants were found to have Hap1, and eight were found to have Hap2. We again identified a region in the right posterior fusiform gyrus with significant differential connectivity as a function of haplotype group, and this region was nearly identical to that identified in the discovery cohort (Fig. 3, bottom). No other regions showed significant differential connectivity.

Sensitivity analyses

We conducted multiple analyses to confirm that our results were not driven by methodological decisions. First, to examine whether the identified results were lateralized to the right IPS, we repeated our analyses in the Williams syndrome and typically developing cohorts using a left IPS seed region also previously reported to have a structural anomaly in Williams syndrome (Meyer-Lindenberg et al., 2004) and found the pattern of connectivity differences to be nearly identical to the pattern of connectivity differences with the right IPS (Supplementary Fig. 4A). We also used this left IPS seed region in our healthy cohorts stratified by LIMK1 haplotype and again found differential connectivity with the right fusiform gyrus in the same direction as identified with the right IPS (IPS connectivity in Hap1 < Hap2; Supplementary Fig. 4B).

Second, because the Williams syndrome versus typically developing analysis was carried out with a developmental sample and the between haplotype discovery and replication analyses were carried out in adults, we repeated our LIMK1 haplotype analysis in the PNC, a sample of healthy adolescents (n = 339 after quality control and ancestry exclusions). As in the Williams syndrome versus typically developing analysis and the adult Discovery and Replication analyses, we again identified a nearly identical region of the right posterior fusiform gyrus to show significant differential connectivity as a function of haplotype group, in the same direction (Hap1 < Hap2) as seen in both adult samples (Supplementary Fig. 5). Additionally, we also found a region of the PCC that showed differential connectivity in the opposite direction (Hap1 > Hap2). This region overlapped with a PCC region showing increased IPS-connectivity in Williams syndrome.

Finally, as another gene in the 7q11.23 Williams syndrome critical region, specifically the GTF2I gene, has also been linked to brain development and intellectual abilities (Morris et al., 2003; Ghaffari et al., 2018; Deurloo et al., 2019), we repeated our analyses in the general population samples with a SNP in the GTF2I gene previously associated with brain functioning, rs2527367 (Jabbi et al., 2015). This SNP is in high linkage disequilibrium with a number of other GTF2I SNPs between chromosomal location 73706683 and 73777987, and thus obviates the need to investigate a number of related GTF2I SNPs and minimizes the number of statistical procedures necessary (Jabbi et al., 2015). Using this SNP, we did not identify significant differences in IPS connectivity with any brain region between genotype groups in either the discovery or replication cohorts.

Discussion

Here, we show that both the 7q11.23 hemideletion in Williams syndrome (which contains the LIMK1 gene) and LIMK1 sequence variation in the general population alter functional connectivity of the IPS and do so in similar ways. Specifically, individuals with Williams syndrome and individuals from the general population who have a haplotype previously associated with gene transcription show decreased IPS connectivity with visual processing regions. Additionally, individuals with Williams syndrome show increased functional connectivity with social processing regions, a finding that was only observed with LIMK1 variation in the developing sample, but not in adults. In this regard, the 7q11.23 hemideletion appears to be accompanied by a redistribution or de-differentiation of the IPS-centred network.

The functional connectivity findings in individuals with Williams syndrome echo the clinical picture of impaired visuospatial construction abilities and a hypersocial personality. Combining this result with prior work, the structure (Meyer-Lindenberg et al., 2004; Kippenhan et al., 2005), function (Sarpal et al., 2008), and now resting functional connectivity of the IPS, all are significantly altered in Williams syndrome and likely account for a considerable portion of the visuospatial processing deficits that typify the syndrome. Additionally, here we show that in Williams syndrome, instead of the usual robust functional connectivity with visual processing regions observed in control populations, the IPS shows increased connectivity with social processing regions. This finding suggests that in Williams syndrome there is a change in the interregional functioning of the IPS, diverting neural resources from dorsal visual processing stream functions to social processing functions. Though such a bias in social versus visuospatial processing networks is in line with the clinical picture of Williams syndrome, it was not completely predicted.

Nonetheless, although the brain changes in Williams syndrome most probably result from interactions of multiple genes, LIMK1 is likely to play an important role in IPS dynamics. Such a role is biologically plausible because the LIMK1 protein is known to subserve key functions in neuronal and axonal migration (Scott and Olson, 2007; Dong et al., 2012). This LIMK1 sphere of influence may be especially relevant for the IPS, as this highly evolved brain region is under particular mechanical tension during development (Armstrong et al., 1991; Kippenhan et al., 2005) and, in Williams syndrome, has white matter architecture changes consistent with altered migration during gestation (Hoeft et al., 2007; Marenco et al., 2007), providing a potential critical window in which altered LIMK1 function could affect IPS connectivity. Indeed, in addition to observing effects of hemideletion of the entire Williams syndrome region on IPS functional connectivity, we also observed similar changes in IPS connectivity to visual processing regions with LIMK1 sequence variation in the general population. Examining effects of a previously described functional haplotype in the promoter region of LIMK1, we found that, compared to individuals harbouring a minor allele (Hap2), those homozygous for the major alleles (Hap1) showed decreased connectivity with the fusiform gyrus, a visual processing region likely involved in object representation and visual motion (Tootell et al., 1995; Grill-Spector et al., 2001). In contrast, we tested for similar effects of another 7q11.23 gene that has been prominently linked to the neurobehavioural phenotype of Williams syndrome, GTF2I. In that analysis we did not observe GTF2I genotype-dependent changes in IPS connectivity in the general population, thus, supporting a particularly important role for LIMK1 in the pathophysiology of IPS connectivity.

Even though the LIMK1 haplotype we tested in individuals from the general population has been shown to affect transcription (Akagawa et al., 2006), the neural effects of this sequence variation are likely to be less robust than the effects of hemideletion of the entire LIMK1 gene and other genes in the Williams syndrome region. And indeed, as expected, the effects in individuals with Williams syndrome were more robust than those associated with LIMK1 variation in the general population, as the latter elicited lower average β effect estimates across the clusters (0.28 for the CNV analysis versus 0.19 for the haplotype analysis), which were also noted to be smaller in size. This is not surprising as differences between the Williams syndrome and typically developing groups involve hemideletion of multiple genes and an easily distinguishable clinical phenotype. In contrast, the genetic differences between LIMK1 groups encompassed just single base pair changes in three SNPs and no known clinical differences between the groups. Our findings of altered IPS connectivity related to both CNV and sequence variation of the LIMK1 gene support the notion that genes code for molecular processes that directly and/or indirectly mediate how neurons and neural circuits adapt to influence complex behaviours, even in the general population.

The publicly available GTEx database (http://gtexportal.org) allows predictions about the expected directionality of gene expression differences between LIMK1 Hap1 and Hap2 and their relation to the LIMK1 hemideletion in Williams syndrome. For two of the SNPs comprising our three-SNP haplotype (data for rs710968 were not available for brain cortex on GTEx portal), the major alleles were associated with decreased LIMK1 transcription in the cortex (Supplementary Fig. 4). Therefore, as Hap1 is the triple major allele genotype, one would expect the lower expression levels to be closer to those found with LIMK1 hemideletion in Williams syndrome.

In light of the changes seen here, individuals from the general population who carry LIMK1 Hap1 do indeed appear phenotypically closer to those with Williams syndrome than those with LIMK1 Hap2 in terms of visuospatial processing networks. This observation is particularly interesting because for two of three SNPs in this haplotype (rs6460071 and rs710968, but not rs146777179), the minor allele is also the ancestral allele and the derived allele is not present in other non-human primates (dbSNP; www.ncbi.nlm.nih.gov/projects/SNP). This suggests the influence of an evolutionary pressure away from the ancestral alleles, resulting in an IPS less connected to visual processing regions, closer to the Williams syndrome phenotype. Supporting this idea, all Neanderthals and Denisovans previously sequenced for both of these SNPs (examined via UCSC genome browser, genome.ucsc.edu and Max-Planck Department of Evolutionary Genetics, www.eva.mpg.de/neandertal), contain only the minor alleles at these sites. This observation is also consistent with archaeological evidence from Neanderthal skulls suggesting that, compared to modern humans, Neanderthal brains had increased resources devoted to visuospatial processing at the expense of social processing functions (Pearce et al., 2013). Further, modern, living humans who have higher proportions of Neanderthal-derived genetic polymorphisms show skull morphology more similar to Neanderthal fossils than those with lower proportions, and brain structure related to Neanderthal-derived polymorphisms shows structural changes in the same portion of the IPS used as a seed region in this analysis (Gregory et al., 2017). As the human brain has a limited supply of resources to utilize (Sokoloff et al., 1955), it is interesting to speculate that there may exist an evolutionary balance between social and visuospatial functioning, whereby increasing the neural bias toward one behaviour leads to a reciprocal decrease in the other. From an evolutionary and behavioural perspective, this could suggest that, for modern humans, forming larger, more complex social groups provides more survival benefit than does visuospatial skill. The results provided here suggest that LIMK1 may play a role in this balance.

It should be noted that we did not find age-related effects on IPS connectivity in our Williams syndrome and typically developing groups. However, analysis of the PNC developmental sample identified LIMK1 haplotype effects on connectivity of the IPS with the PCC, a brain hub of social processing. This finding is consistent with our paediatric Williams syndrome–typically developing cohorts but was not observed in either the discovery or replication adult general population cohorts. The possibility that the increased IPS-to-social-brain connectivity identified in children with Williams syndrome and associated with LIMK1 variation in typically developing children may be developmentally specific and modulated later in life will require further longitudinal investigation.

In sum, we show that the functional connectivity of the IPS is altered in Williams syndrome such that there is less connectivity with visual processing regions and increased connectivity to social processing regions. Similarly, sequence variation of LIMK1 gene also revealed differential IPS connectivity with visual processing regions. The results with sequence variation anatomically overlap with the regions showing differences in individuals with the Williams syndrome CNV and are found in two independent samples from the general population analysed with parallel methodology. Further, the results offer information not only about the neurobiology and specific genetic mechanisms responsible for the Williams syndrome phenotype but also about how genetic effects can produce complex human characteristics, providing insight into evolutionary genetic mechanisms that may have shaped the human brain.

Funding

This work was supported by the Intramural Research Program, National Institute of Mental Health, National Institutes of Health, Bethesda, MD (ZIAMH002863 and ZIAMH002942) and a 2010 NIH Bench-to-Bedside award and a 2014 Brain and a Behavior Research Foundation Distinguished Investigator Award to K.F.B. The data for the NIMH Williams syndrome/typically developing cohort were obtained under protocol 10M0112/NCT01132885. The data for the NIMH adult cohorts were obtained under protocols 00M0085/NCT00004571 and 95M-0150/NCT00001486. Some of this work utilized the computational resources of the NIH HPC Biowulf cluster (http://hpc.nih.gov). C.B.M’s participation in this project was supported by NICHD (R37 HD29957), NINDS (R01 NS35102), a 2010 NIH Bench-to-Bedside award, and the Williams Syndrome Association (WSA 0104, WSA 0111). Support for the collection of the data for Philadelphia Neurodevelopment Cohort (PNC) was provided by grant RC2MH089983 awarded to Raquel Gur and RC2MH089924 awarded to Hakon Hakonarson. Subjects were recruited and genotyped through the Center for Applied Genomics (CAG) at The Children's Hospital in Philadelphia (CHOP). Phenotypic data collection occurred at the CAG/CHOP and at the Brain Behavior Laboratory, University of Pennsylvania.

Competing interests

All authors report no competing interest.

Glossary

Abbreviations

CNV =

copy number variation

IPS =

intraparietal sulcus

NIMH =

National Institute of Mental Health

PNC =

Philadelphia Neurodevelopmental Cohort

SNP =

single nucleotide polymorphism