Neuropsychiatric copy number variants exert shared effects on human brain structure

Abstract

. However, CNV studies have been conducted one mutation at a time, which hinders our understanding of any potential general mechanism linking CNVs to effects on brain architecture.Indeed, several lines of evidence suggest shared mechanism across CNVs: 1) Neuropsychiatric CNVs have been associated with a similar range of adverse effects on childhood neurodevelopment with only subtle quantitative and qualitative differences 20 , 2) CNVs affecting coding genes decrease IQ and increase risk for ASD across a large proportion of the genome 21,22 , 3) Schizophrenia is associated with an increase in overall CNV burden 2 .
The large number of genes encompassed in CNVs has also limited the study of mechanisms associating CNVs to brain morphometry.Studies of the 16p11.2locus have suggested that TAOK2 and KCTD13 within the 16p11.2locus are implicated in embryonic neurogenesis and are candidate genes for differences in total brain volume in animal models 6,23 .Similarly, DGCR8 and TBX1 within the 22q11.2locus were found to be involved in neurogenesis 24,25 .The effects of these genes on regional volumes have not yet been studied in humans or animal models.
The body of literature on CNVs raises several questions: What is the relationship between global and regional alterations associated with CNVs?Does gene dosage at distinct genomic loci converge on a common set of brain alterations, or do genes lead to specific effects?More broadly, the field is lacking a conceptual framework to identify general principles linking CNVs to effects on brain architecture and risk for psychiatric conditions.
In this study, we aimed to characterize shared and specific neuroanatomical variations across multiple CNVs by capitalizing on multivariate and univariate methods.
. We analyzed T1 weighted MRIs of the largest multi-site dataset of CNV carriers (n=484, of which 87 have not yet been published) and controls (n=1296).Voxel and surfacebased methods were performed in parallel.First, a multi-view pattern-learning algorithm (Canonical Correlation Analysis, CCA) was used to identify latent brain morphometry dimensions explaining the effects of CNVs across genomic loci.Second, we investigated brain alterations shared across deletions and duplications using univariate linear models.CNVs assessed in non-clinical populations: Genetic and neuroimaging data from nonclinical population were obtained from the UK Biobank dataset.PennCNV and QuantiSNP were used, using standard quality control metrics, to identify CNVs 21,22 .

Methods
Deletions and duplications with neuroimaging data included in the study were selected on the following breakpoints: 16p11.

Multi-view pattern-learning analysis:
We re-purposed canonical correlation analysis (CCA) to interrogate the shared and distinct impact on brain morphometry (130 grey matter regions) across CNVs 31,32 .This principled multivariate approach allowed investigating the underlying relationship between two sets of variables, and has been widely used in neuroimaging studies 31,32 .In our study, CCA identified modes of coherent co-variation that jointly characterize how CNVs and patterns of regional volumes systematically co-occur across subjects.We refer to these modes of co-variation as 'CCA dimensions' or 'gene-morphology dimensions'.Details are provided in supplementary eMethod 3.

Voxel-based measures and statistical analyses:
To complement and illustrate the CCA analysis, we performed a whole-brain voxel-based approach.Analyses tested voxel-wise differences in volume using a mass-univariate analysis framework implemented in SPM.

Surface-based measures and statistical analyses:
In parallel to VBM, we used surfacebased GLM-based analysis to test differences in CT and SA (SurfStat toolbox 36 ).Each GLM used the surface feature as the dependent variable, the groups -as independent variable, which were adjusted for age, age 2 , sex, site, and Total-SA/Mean-CT.Post-hoc contrasts compared each CNV group against controls, including estimation of Cohen's d effect size estimates from t-values 35 .False Discovery Rate (FDR), with p-value at 0.05 [12][13][14][15] was applied to control for false positive errors due to multiple comparisons.

Comparison across VBM, SA, and CT:
To compare spatial pattern of voxel-and surfacebased results, we projected VBM results on fsaverage, using Freesurfer's vol2surf function.
Null hypotheses using spin permutations and label shuffling: We used spin permutation and label shuffling to calculate empirical p-values for 1) the deletion and duplication convergence pattern and 2) the correlation/dice-index between two maps.

Spin permutation testing:
The spin permutation test provides a null hypothesis quantifying the probability of observing by chance a dice index or correlation value, .
while controlling for spatial auto-correlations inherent in neuroimaging data.This method has been established previously in 37,38 , and detailed in supplementary eMethod 5.

Label shuffling:
We additionally tested the overlap significance by performing permutation of control and CNV labels, generating empirical null distributions and calculating dice index distribution with regard to the convergence pattern.Details are available in supplementary eMethod 6.
Overlap with cross-psychiatric disorders map: Dice-index was computed to estimate the overlap between deletion and duplication convergence maps and the statistical maps obtained from a large cross-disorder neuroimaging meta-analysis (http://anima.fzjuelich.de) 39.

Effects on global brain morphometry
Deletions and duplications of each genomic loci (except 15q11.2) showed opposing effects on total intracranial volume (TIV), total grey matter volume (GM) and total surface area (SA) (Figure 1a-c).For mean cortical thickness (CT), 22q11.2 and 15q11.2deletions and duplications showed opposing effects (Figure 1d).The directionality of global effects differed across loci: They were positively correlated with gene dosage at the 1q21.2a, supplementary eTable 2-4).While the two gene-morphology dimensions allowed for discrimination between genomic loci and between most CNV carriers and controls, they also pointed to brain regions contributing to variation shared across loci (Figure 2b).

Deletions at 4 genomic loci converge on shared neuroanatomical alterations
To further dissect the "gene-morphology dimensions" identified above, we performed univariate whole-brain VBM analyses contrasting each deletion and duplication group with controls.We first only considered clinically ascertained individuals to account for selection bias.Cohen's d maps from the 22q11.2and 16p11.2CNVs were consistent with previous studies (Figure 3) 10,15 .Findings for all 3 deletions including 1q21.1 are detailed in Figure 3 and supplementary eTable 2.
The conjunction analysis of 3 deletion vs control contrasts showed overlap in the left and right middle cingulate gyrus (Figure 4a), corresponding to the 8 th and 18 th highest loading regions on the first CCA dimension.However, conjunction analyses of FWE thresholded maps are extremely stringent and constrained by the group with the smallest effect and sample size.For example, in the 1q21.1 deletion group we could identify 1.5% of grey matter voxels that survived FWE correction, while the large subject sample and effectsize of the 22q11.2deletions allowed for the identification of alterations in 14.8% of grey matter (supplementary eTable 2).
Therefore, to identify shared patterns across CNVs we ranked Cohen's d maps and overlapped voxels with similar rankings.The conservative group of voxels with Cohen's d values <5 th and >95 th percentiles, showed spatial overlap across all deletions in the middle cingulate gyrus (p-value SHUFFLE <10e-4, Figure 4b).Using a lenient threshold for .voxels with Cohen's d <15 th and >85 th percentiles, we observed a broader overlap between deletions (p-value SHUFFLE <10e-4, Figure 4c).Volumes of the middle and anterior cingulate extending to the medial frontal cortex and supplementary motor cortex were decreased in all deletions while volumes were increased in the thalamus, ventral diencephalon, orbital gyrus and parahippocampal gyrus (Figure 4d).Supplementary motor cortex, posterior insula, orbital gyrus, precentral gyrus and thalamus were also found in the top 20 contributors of the CCA dimensions 1. Cerebellum and anterior insula were found in the top 20 contributors of CCA dimension 2 (Figure 2b, supplementary eTable 5).
Spatial convergence could be related to clinical ascertainment.We, therefore, recomputed the deletion convergence map by replacing the clinically ascertained 1q21.1 Cohen's d map by the one calculated using 1q21.1 deletion carriers from the UK Biobank (Table 1).
This new deletion convergence map was similar to the initial one presented above with a dice index of 39.4% (p-value SPIN < 10e-4).
Finally, we investigated the robustness of the deletion convergence map by intersecting it with a fourth map calculated with 72 carriers of 15q11.2deletions and 965 controls from the UK Biobank (supplementary eTable 2, Figure 4i).Voxels with Cohen's d values <15 th and >85 th percentiles significantly overlapped with the clinically ascertained deletion convergence pattern (p-value SPIN <10e-4, Figure 4e).We did not perform these analyses for 16p11.2 and 22q11.2due to the limited sample size in UKBB.
Comparable findings were observed in the analysis, performed in parallel, for Freesurfer derived SA and CT measures (supplementary eFigure 3-4; 6-10).

Duplications at 4 genomic loci converge on common neuroanatomical alterations
Conjunction analysis of the FWE thresholded duplication maps at the 1q21.We carried out the same sensitivity analyses performed above for deletions.The new duplication convergence map including 1q21.1 duplications and controls from the UKBB showed significant overlap with the initial duplication convergence map (dice-index of 26%, p-value SPIN <10e-4).The intersection between the duplication convergence map above and a fourth map computed with 76 15q11.2duplication carriers and 965 controls from the UK Biobank (supplementary eTable 2, eFigure 5h) also showed significant overlap (p-value SPIN <10e-4, eFigure 5d).

Effect size ratio and mirror effects of deletions and duplications
Deletions had larger effect sizes on grey matter alterations compared to duplications.
Cohen's d distributions showed a 1.24 to 2-fold deletion/duplication ratio, F test, p<10e-16 (Figure 1, supplementary eTable 11).Similar effect-size ratio is also observed for SA alterations (supplementary eTable9).This was not the case for 15q11.2deletions and duplications, which showed equally small effect sizes.
We investigated the contrasting regional effects of deletion and duplication on brain morphometry.Deletion and duplication carriers showed brain-wide regional mirror effects for regional volumes and SA, but not CT, at all 4 genomic loci (Figure 5,

Relationships with previously defined cross-psychiatric disorder maps
We investigated similarities between our CNV convergence maps and alterations reported by a cross-disorder meta-analysis of case-control studies of schizophrenia, bipolar disorder, obsessive-compulsive disorder, substance-use disorder, major depression and anxiety 39 .Deletion convergence map showed alterations affecting the same regions (cingulate and insula) with contiguous, but only marginally overlapping .clusters (supplementary eFigure 11).The duplication convergence map also showed marginal overlaps with a contiguous cluster in the anterior insula.

Discussion
Our study demonstrates that deletions and duplications at 4 distinct genomic loci have independent effects on global and regional brain morphometry.The Eight CNVs, associated with ASD and SZ, affect regional brain volumes along two gene-morphometry dimensions that emerged from our CCA results.These dimensions included regions such as the temporal gyrus, calcarine cortex, supplementary motor cortex, insula, middle cingulate gyrus, accumbens, subcallosal area, cuneus, precuneus, brain stem and superior frontal gyrus.Deletions and duplications of the same genomic loci lie on opposite ends of these brain dimensions.Univariate analyses similarly point towards a pattern of differences involving anterior cingulate, posterior insula and supplementary motor cortex observed in deletions across 4 loci.For duplications a different pattern is observed involving middle cingulate, anterior insula and lingual gyrus.Mirror effects are observed at the global and regional levels across loci, for volume and SA but not CT.

Dissociation between global and regional effects
Systematic comparison across loci suggests that CNVs have independent effect sizes on global and regional brain morphometry.For example, the 1q21.1 deletions and duplications highlight the contrast between very large effects on global measures, with small regional effects adjusted for total GM.The same dissociation is observed between the directionalities of global and regional effects across the 4 genomic loci.As an .example, 1q21.1 and 22q11.2deletions have negative effects on TIV and GM, while 16p11.2deletions have positive effects, and 15q11.2deletions have no detectable effects.
In contrast, all four deletions show smaller regional volumes in the middle and anterior cingulate, medial frontal cortex and supplementary motor cortex, as well as larger volumes in the thalamus, orbital gyrus and parahippocampal gyrus.The same dissociation between global and regional effect sizes and between directionalities applies to duplications.
We posit that global and local effects may be mechanistically unrelated.Animal studies have proposed that CNV-related differences in global brain volume may be due to the modulation of embryonic neurogenesis 6 , dendrite growth or spine and synapse development 23 .Larger brain size in individuals diagnosed with autism-related genetic conditions such as tuberous sclerosis complex has been linked to cell body size and is sensitive to medication 40 .On the other hand, the mechanisms underlying local relative effects have not been investigated in animal models.Although both, global and regional brain alterations show mirror effects in deletions and duplications across most loci.

General effects of gene dosage on brain structure
Altering gene dosage at 4 distinct loci encompassing 60, 29, 12 and 4 genes lead to a degree of shared regional patterns.In line with this observation, a complex landscape of rare deleterious CNVs have been associated with cognition 21 , increased risk of neurodevelopmental and psychiatric disorders such as autism 1,20,22 and schizophrenia 2 .
Recent studies estimated that 71%-100% of any 1-MB window in the human genome contributes to increased risk for schizophrenia 41 .The same has been demonstrated for autism 22 .We therefore speculate that CNVs will lead to patterns of brain alterations similar to the one characterized in our study irrespective of their genomic location.
A plausible hypothesis for the pervasive effects of CNVs on cognition, behavior and brain architecture, is that these phenomena are related to emerging properties of the genome, rather than the function of individual genes 42 .In other words, gene dosage at any node of the genomic network will alter its efficiency leading to a measurable effect on brain organization and behavior.Gene dosage affecting diverse molecular functions may lead to a limited number of ways in which the brain reconfigures, compared to noncarriers.

Neuroanatomical patterns across brain measures
Systematic analysis through the two most widespread computational neuroanatomy frameworks (VBM and Freesurfer) shows that main grey matter volume findings are recapitulated by SA results.On the other hand, changes in CT seem to be distinct, which is in line with previous investigations of 16p11.2CNVs 15 as well as studies suggesting that apparent CT is mainly related to cortical myelination 43 .Volume is a product of SA and CT, measures that have been shown to be genetically unrelated 44 .

Limitations
Our study was focused on investigating shared features across genomic loci.The study of many more loci would be required to characterize specific effects.Integrating data from multiple study sites may have introduced noise and heterogeneity in our investigation.
We have previously shown that, although site effects on neuroimaging data are .measurable, they do not influence the neuroanatomical patterns associated with CNVs at the 16p11.2,22q11.2 and 15q11.2loci 10,15,45 .
We carried out the first whole-brain voxel-and vertex-wise analyses for 15q11.

Conclusions
In this proof of concept, our investigation demonstrates the relevance of simultaneously analysing the effect of several genomic variants on neuroimaging intermediate phenotypes.Extending our approach to the rapidly expanding number of rare genomic variants associated with psychiatric disorders should provide the field with a scheme to understand general principles linking deleterious genomic variants with their consequence on human brain architecture.These general neuroanatomical patterns may help understand the broad locus heterogeneity of psychiatric conditions. .
Clinically ascertained CNV carriers: Individuals carrying 1q21.1 (Class I & II), 22q11.2(BPA-D) or 16p11.2(BP4-5) CNVs, were assessed as either probands referred for genetic testing, or as relatives.Controls were either non-carriers within the same families or individuals from the general population.We pooled data from 5 cohorts: Cardiff University (UK), 16p11.2European Consortium (Lausanne, Switzerland), University of Montreal (Canada), UCLA (Los Angeles, USA) and the Variation in individuals Project (SVIP, USA) detailed in the supplementary files.
CNVs: To adjust for the unequal power to detect change across different CNV groups in the univariate analyses, which have different sample and effect sizes, we ranked Cohen's d distributions of all voxels (/vertices) from the estimated un-thresholded statistical maps.We then tested for spatial overlap between maps across CNVs after thresholding the tails of the distribution at: i) 5th & 95 th quantiles, and ii) 15th & 85 th quantiles.The dice-index was calculated using publicly available Matlab scripts and functions (https://github.com/rordenlab/spmScripts).

Figure 2 :
Figure 2: Co-analysis of shared brain changes due to 8 CNVs.

Figure 3 :
Figure 3: Cohen's d maps of VBM regional brain differences in deletion and duplication

Table 1
and eTable 1).Signed consents were obtained from all participants or legal representatives prior to the investigation.MRI data: Data sample included T1-weighted (T1w) images at 0.8 -1 mm isotropic resolution across all sites.The data distribution overview, as well as the population description is available in Table1and eTable 1. MRI protocol information is available in .(