Neuropsychiatric mutations delineate functional brain connectivity dimensions contributing to autism and schizophrenia

16p11.2 and 22q11.2 Copy Number Variants (CNVs) confer high risk for Autism Spectrum Disorder (ASD), schizophrenia (SZ), and Attention-Deficit-Hyperactivity-Disorder (ADHD), but their impact on functional connectivity (FC) remains unclear. We analyzed resting-state functional magnetic resonance imaging data from 101 CNV carriers, 755 individuals with idiopathic ASD, SZ, or ADHD and 1,072 controls. We used CNV FC-signatures to identify dimensions contributing to complex idiopathic conditions. CNVs had large mirror effects on FC at the global and regional level. Thalamus, somatomotor, and posterior insula regions played a critical role in dysconnectivity shared across deletions, duplications, idiopathic ASD, SZ but not ADHD. Individuals with higher similarity to deletion FC-signatures exhibited worse cognitive and behavioral symptoms. Deletion similarities identified at the connectivity level could be related to the redundant associations observed genome-wide between gene expression spatial patterns and FC-signatures. Results may explain why many CNVs affect a similar range of neuropsychiatric symptoms.


Introduction
Copy number variants (CNVs) are deletions or duplications of DNA segments and represent an important source of genetic variation. An increase in rare CNV burden has been linked to a range of neurodevelopmental and psychiatric conditions 1,2. Twelve recurrent CNVs have been individually associated with autism spectrum disorder (ASD) 3, eight with schizophrenia (SZ) 4, and eight with attention deficit hyperactivity disorder (ADHD) 5 but the mechanisms by which they lead to neuropsychiatric disorders remain unclear. Although they have large impacts on neurodevelopment, their effect alone does not lead to a psychiatric diagnosis. CNVs could, therefore, be leveraged to identify major dimensions contributing to complex idiopathic conditions. CNVs at the proximal 16p11.2 and 22q11.2 genomic loci are among the most frequent large effect-size genomic variants and alter the dosage of 29 and 50 genes, respectively 6,7. They confer high risk for ASD (10-fold increase for the 16p11.2 deletion and duplication) 3, SZ (>10-fold increase for the 22q11.2 deletion and 16p11.2 duplication) 4, and ADHD 8-12. Gene dosage (deletions and duplications) affect the same neuroimaging measures in opposite directions (mirror effect). Structural alterations of the cingulate, insula, precuneus and superior temporal gyrus overlap with those observed in meta-analytical maps of idiopathic psychiatric conditions including ASD and SZ. 10,11 Large effect-size mutations can shed light on pathways connecting genetic risk to brain endophenotypes, such as functional connectivity (FC). FC represents the intrinsic low-frequency synchronization between different neuroanatomical regions. It is measured by means of resting-state functional magnetic resonance imaging (rs-fMRI) which captures fluctuations of blood oxygenation as an indirect measure of neural activity across brain areas when no explicit task is performed 13,14. Robust functional brain networks measured by rs-fMRI are also recapitulated by spatial patterns of gene expression in the adult brain 15,16. Few studies have investigated the effect of 'neuropsychiatric' CNVs on FC. Dysconnectivity of thalamichippocampal circuitry 17 has been reported in 22q11.2 deletion carriers, with prominent under-connectivity of the default mode network (DMN), which was predictive of prodromal psychotic symptomatology 18,19. Impaired connectivity of long-range connections within the DMN has also been reported by other studies 3 20. A single 16p11.2 study has shown a decrease in connectivity of frontotemporal and -parietal connections in deletion carriers 21. These initial studies have focussed on regions of interest but connectome-wide association studies (CWAS) analysing all connections without a priori hypotheses have not yet been performed in CNV carriers. Furthermore, their relation to idiopathic conditions has not been investigated.
Brain intermediate phenotypes of psychiatric conditions have mainly been studied by adopting a top-down approach, starting with a clinical diagnosis and moving to underlying neural substrates and further down to genetic factors 22. Studies applying this analytical strategy in ASD have repeatedly shown patterns of widespread under-connectivity with the exception of overconnectivity in cortico-subcortical connections, particularly involving the thalamus 23,24. SZ also exhibits a general under-connectivity profile, mainly involving the medial prefrontal cortex, the cingulate and the temporal lobe 25, with over-connectivity of the thalamus 26. These altered networks do not appear to be disorder-specific and have been reported across several disorders, including ASD, ADHD, and SZ 27. These similarities seem to be distributed across several continuous dimensions 28 which may be related to shared genetic contribution across diagnoses, which is documented for common 29 and rare 30 variants, including the 16p11.2 and 22q11.2 CNVs.
We posit that seemingly distinct genetic variants and idiopathic disorders have overlapping patterns of dysconnectivity, which may help identify FC dimensions, providing insight into the complex connectivity architecture involved in psychiatric conditions. We aimed to 1) characterize the FC-signatures of four high-risk neurodevelopmental CNVs, 2) explore whether FC-signatures of CNVs represent dimensions observed in idiopathic ASD, SZ, or ADHD and 3) investigate the relationship between deletions at the FC and gene expression level.
To this end, we performed CWAS studies on 101 carriers of a 16p11.2 or 22q11.2 CNV, 122 of their respective controls, 751 individuals with idiopathic ASD, SZ, or ADHD and 948 of their respective controls.
To our knowledge, this is the first connectome-wide study to compare rare genomic variants and idiopathic psychiatric conditions. 4 Results 16p11.2 and 22q11.2 CNVs have large effects on functional connectivity at the global and regional level.   Figure 1b, d) when compared with control subjects. Underconnectivity predominantly involved the anterior and lateral DMN, and limbic network (Figure 1d). The temporal pole, the ventral anterior insula and peri-insular sulcus, the amygdala-hippocampal complex, the dorsal anterior cingulate cortex, and perigenual anterior cingulate cortex showed the strongest changes in connectivity (see Supplementary Table S1.8).
For 16p11.2 duplication carriers, none of the individual connections survived FDR correction ( Figure 1a and Supplementary Table S1.2) relative to controls and none of the individual connections survived FDR correction. For the 22q11.2 duplications, only 16 connections survived FDR and these included overconnectivity in the posterior medial and lateral visual network, the cerebellum I-V, and the lateral fusiform gyrus (Figure 1c and Supplementary Table S1.4 and S1.8).
Deletions and duplications at both loci showed a mirror effect at the global connectivity level. 16p11.2 deletions and duplications also showed mirror effects at the network level (p = 0.006, two-sided). This was not the case for 22q11.2 (Supplementary Results).
A sensitivity analysis showed that results are unaffected by differences in age distribution between deletions and control groups as well as the number of remaining frames available after scrubbing (see Supplementary   Results).
The effect sizes of deletions and duplications are twice as large as the effects of idiopathic SZ, ASD, or ADHD We performed three independent CWAS, comparing FC between patients with ASD, SZ, ADHD, and their respective controls. Idiopathic SZ showed overall underconnectivity affecting 835 connections, in line with previous reports 26,31 (Figures 3a, 3c, Supplemental Results and Supplemental Figure 4, Tables S1.6 and S1.8). Over-connectivity was restricted to 24 connections (FDR, q < 0.05).
For ADHD, none of the individual connections survived FDR correction (Supplemental Results and Tables S1.7 and S1.8). Sensitivity analyses excluding females from the SZ and ADHD cohorts showed identical results (Supplementary Results).
Among idiopathic conditions, the effect size of connectivity alteration was the highest in SZ (largest beta value = -0.56 std of the control group), followed by autism (largest beta value = -0.46), and ADHD (largest beta value = +0.26). Effect sizes observed for both deletions were approximately two-fold larger (beta values = +1.34 and -1.64 for 16p11.2 and 22q11.2 respectively) than those observed in idiopathic SZ, ASD, and ADHD ( Figure 3c). The largest effect size among the 16 connections surviving FDR for the 22q11.2 duplication was Cohen's d = 1.87. 6 Individuals with ASD and SZ relative to controls, show similarities with whole-brain

FC-signatures of CNVs
We tested the spatial similarity between whole-brain FC-signatures across CNVs and idiopathic psychiatric conditions. To this mean we computed the similarity (Pearson R) between group-level FC-signatures and the individual connectomes of either cases or controls from another group ( Figure 2). This was repeated 42 times between all CNVs and conditions and in both directions. Most of the significant whole-brain FC similarities were observed between individuals with either idiopathic ASD, SZ and 4 CNVs (Figure 3d).
ADHD did not show any significant similarities with any other group.
Thalamus and somatomotor regions play a critical role in dysconnectivity observed across CNVs and idiopathic psychiatric conditions We asked if whole-brain FC similarities between individuals with ASD, SZ and CNVs may be driven by particular regions. We thus repeated the same similarity analysis presented above at the level of the FC signatures of each of the 64 seed regions. Individuals with SZ showed increased similarity with 28 out of the 64 regional FC-signatures of the 16p11.2 deletion than controls (FDR, q < 0.05). They also showed Individuals with autism showed greater similarity with six regional FC-signatures of the 16p11. Individuals with ADHD did not show higher similarities with the regional FC-signatures of any CNVs except for 2 regional FC-signatures of the 16p11.2 duplication (Supplementary Tables S2.5 and S2.6).

Similarity to deletion FC-signatures is associated with symptom severity
We investigated whether regional FC similarities with deletions described above are associated with symptom severity among individuals in idiopathic psychiatric cohorts. Symptom severity was assessed using the Autism Diagnostic Observation Schedule (ADOS, 32), in ASD, Positive and Negative Syndrome Scale (PANSS, 33) in SZ, and Full Scale Intelligence Quotient (FSIQ) in ASD. The 10 seed regions with significant FC similarity between ASD and either deletion were those showing the strongest association with the ADOS symptom-severity score (two regions passed FDR correction q < 0.05: the caudate nucleus and temporal pole) and FSIQ ( Figure 6 and Supplementary Tables S3.1-3.4). Among the seed regions contributing to the similarity between SZ individuals and deletions, none were significantly associated with PANSS measures after FDR correction. FSIQ data was not available in the SZ cohorts.

Discussion
This proof of concept study provides the first connectome-wide characterization of four CNVs that confer high risk for psychiatric disorders. Deletions and duplications at the 16p11.2 and, to a lesser extent, the 22q11.2 locus were associated with mirror effects at the connectome-wide level. Overconnectivity in the 16p11.2 deletion predominantly involved the ventral attention, motor, and frontoparietal networks.
Underconnectivity in the 22q11.2 deletion involved the anterior and lateral DMN and the limbic network.
Regional FC-signatures of deletions and duplications, in particular, those implicating the thalamus, somatomotor, posterior insula and cingulate showed significant similarities with the complex architecture of idiopathic ASD, SZ but not ADHD. Seemingly distinct, rare neuropsychiatric mutations may converge on dimensions representing mechanistic building blocks shared across idiopathic conditions. The spatial expression patterns of genes encompassed in both genomic loci were associated with FC-signatures of the corresponding deletion but many genes outside these 2 loci also show similar levels of association. This redundancy may represent a factor underlying shared FC signatures between both deletions. 22q11.2 and 16p11.2 CNVs showed large effect sizes on FC that are similar to those previously reported for structural neuroimaging measures, cognition, and behaviour 8,10,11. In sharp contrast, there is significant discordance between the severe clinical manifestations observed in idiopathic ASD and SZ, and the small effect size observed in case-control studies at the FC level. Previous structural neuroimaging studies of the same idiopathic psychiatric conditions have also reported small effect sizes 34,35. This discordance may be due to the heterogeneity of these idiopathic conditions and hints at the presence of subgroups or latent dimensions associated with larger effect sizes 28. The FC-signatures of both deletions (and to a lesser extent those of both duplications) showed similarities with autism and schizophrenia, but not ADHD. Regions contributing to these similarities were also those with the highest number of connections altered by each deletion individually. The FC-signature of the same seed regions also showed the highest association with ASD severity scores and general intelligence in the idiopathic autism sample.
Among the regions, results highlighted overconnectivity between the thalamus and sensory-motor, auditory and visual networks as a common alteration across CNVs and individuals with idiopathic autism or schizophrenia who do not carry CNVs. This is in line with recent rs-fMRI studies performed across psychiatric illnesses 28 Functional connectivity studies using a top-down case-control approach (eg. autism versus control) have characterized large-scale brain network changes associated with diseases, but this framework is unable to describe the directionality of this relationship 49. FC-changes may not necessarily represent an intermediate brain phenotype but rather a secondary impact of psychiatric illnesses. Our strategy integrating top-down and bottom-up approaches shows that individuals with idiopathic ASD or SZ as well as CNV carriers who do not meet diagnostic criteria for these conditions share regional FC alterations. This suggests that the risk conferred by genetic variants and the associated FC-patterns represent important dimensions that are necessary but insufficient to cause disease. Additional factors and associated FC-patterns are required (incomplete penetrance 50). Bottom-up approaches studying rare variants have almost exclusively been performed individually. Our results suggest, however, that they likely converge on overlapping intermediate brain phenotypes, consistent with a recent study showing overlapping effects on subcortical structures across 12 different CNVs conferring risk to SZ 51.

Limitations
Reproducibility of rs-fMRI in psychiatry has been challenging. However, when studies using similar analytical strategies are compared, there are consistent results. In SZ and ASD, global decrease in FC has been reported by most studies except for those adjusting for global signal 24,52. Increased thalamocortical connectivity is also repeatedly reported in both conditions 24,26,31. These previous findings are consistent with our results (see Supplementary Results). The 22q11.2 deletion FC signature is consistent with previous works on 22q11.2 FC alterations that showed 1) underconnectivity of the DMN 53,54, 2) thalamocortical overconnectivity and underconnectivity involving the hippocampus 17. The only rs-fMRI study previously published for the 16p11.2 deletion focused on the dmPFC 21. Using the same approach and regressing global signal, we also found underconnectivity of the dmPFC with the same set of regions. This highlights the fact that many seemingly discrepant results can be reconciled once methodologies are aligned.
There is no available genetic data for any of the three idiopathic cohorts. However, the frequency of 16p11.2 and 22q11.2 CNVs in ASD or SZ is < 1% 3,4. This suggests that the observed FC similarities between CNVs and ASD or SZ are driven by other factors.
Expression data were derived from 6 adult brains of the AHBA and results should be interpreted with caution.
The results on duplications should be interpreted with caution due to our limited power to detect changes in connectivity. The limited phenotypic data in the SZ group did not allow to investigate the relationship between deletion FC-signatures and cognitive traits in this sample. Lack of similarity observed for ADHD is line with the small association between 16p11.2 CNVs and ADHD but is discordant with the association reported for 22q11.2 5. ADHD has a smaller effect size than SZ and ASD, which may have limited our analysis 55. Several confounding factors may have influenced some of the results. Those include sex bias, which is present across all 3 psychiatric cohorts, age differences in the 16p11.2 deletion group, diagnosis of ASD and ADHD in 22q11.2 deletion carriers, and medication status in the idiopathic ASD and SZ groups.
However, carefully conducted sensitivity analyses, investigating all of these confounders did not change any of the results.

Conclusion
Deletion and duplication at several genomic loci result in mirror effects across many human traits 56-59, including brain connectivity. Haploinsufficiency may define functional connectivity dimensions that represent building blocks contributing to idiopathic psychiatric conditions. Thalamo-sensory disturbance may represent one dimension central across genomic mutations and psychiatric diagnoses. The redundant associations observed, genome-wide, between gene expression and connectivity may explain similarities across genomic variants and idiopathic conditions and the fact that many CNVs affect a similar range of neuropsychiatric symptoms. It is, therefore, becoming increasingly difficult to justify the study of psychiatric conditions or rare genetic variants in isolation. Large scale studies simultaneously integrating a top-down approach across diagnostic boundaries, and a bottom-up investigation across a broad set of genomic variants are required to improve our understanding of specific and common psychiatric outcomes associated with genetic variants and FC signatures.

Samples
We performed a series of CWAS using individuals from five data sets (Table 1 and  Imaging data were acquired with site-specific MRI sequences. Each cohort used in this study was approved by the research ethics review boards of the respective institutions. Signed informed consent was obtained from all participants or their legal guardian before participation. Secondary analyses of the listed datasets for the purpose of this project were approved by the research ethics review board at Sainte Justine Hospital. After data preprocessing and quality control, we included a total of 1,928 individuals (Table 1).
Preprocessing and quality control procedures.

Computing connectomes
We segmented the brain into 64 functional seed regions defined by the multi-resolution MIST brain parcellation 64.

Statistical analyses
All of the following analyses are summarised in Supplemental Materials and Methods.

Connectome-wide association studies
We performed seven CWAS, comparing Functional Connectivity (FC) between cases and controls for four CNVs (16p11.2 and 22q11.2, either deletion or duplication) and three idiopathic psychiatric cohorts (ASD, SZ, and ADHD). Note that controls were not pooled across cohorts. Within each cohort, FC was standardized (z-scored) based on the variance of the respective control group. CWAS was conducted by linear regression at the connectome level, in which z-scored FC was the dependent variable and clinical status the explanatory variable. Models were adjusted for sex, site, head motion, and age. We determined whether a connection was significantly altered by the clinical status effect by testing whether the β value (regression coefficient associated with the clinical status variable) was significantly different from 0 using a two-tailed t-test. This regression test was applied independently for each of the 2,080 functional connections. We corrected for the number of tests (2,080) using the Benjamini-Hochberg correction for FDR at a threshold of q < 0.05 67, following the recommendations of Bellec et al. 2015 68. We defined the global FC shift as the average of the β values across all 2,080 connections and tested for significance using a permutation test. We performed 5000 random CWAS by contrasting CNV carriers and controls after shuffling the genetic status labels. For example, we randomly permuted the clinical status of  Table S1.9).
Similarity of whole-brain FC-signatures between idiopathic psychiatric conditions and CNVs.
We tested the similarity between dysconnectivity measured across idiopathic psychiatric conditions and CNV. This similarity was tested by correlating individual whole-brain connectomes of cases and controls of one group to the whole brain FC-signature (group level) of another group (Figure 2). The group-level FC-signature was defined as the 2,080 β values obtained from the contrast of cases vs. controls. This was repeated 21 times between all CNVs and conditions and in both directions (n=42 similarity tests).
Individual connectomes of cases and their respective controls were used after independently adjusting for sex, site, head motion, age, and average group connectivity for each of the datasets.
Similarity scores were derived by computing Pearson's correlations between the whole brain connectomes.
We asked whether cases compared to their respective controls had significantly higher (or lower) similarity to whole-brain FC-signature of another group using a Mann-Whitney U test. We reported significant group differences after FDR correction accounting for the 42 tests (q < 0.05).

Similarity of regional FC-signatures between idiopathic conditions and CNVs
The same approach described above was performed at the regional level. Each of the 1705 connectomes of individuals with idiopathic psychiatric conditions and their respective controls was independently adjusted for sex, site, head motion, age, and average group connectivity for each dataset. We calculated a similarity score between these individual connectomes and the FC-signatures of the 16p11.2 and 22q11.2 deletions and duplications. The FC-signatures were broken down into 64 region-level FC-signatures and similarity scores were derived by computing Pearson's correlations between the 64 β values associated with a particular region. For each region, we tested whether individuals with a psychiatric diagnosis had significantly higher (or lower) similarity to 16p11.2 or 22q11.2 deletion FC-signatures than their respective controls using a Mann-Whitney U test. We reported significant group differences after FDR correction (q < 0.05) for the number of regions (64).
We investigated the relationship between symptom severity and similarity with deletions. The similarity of individuals with deletion FC-signatures was correlated (Pearson's r) with cognitive and behavioral measures. Those included the ADOS and FSIQ in the autism sample and the PANSS in the SZ sample. The p-values associated with these correlations were corrected for multiple comparisons (FDR, q < 0.05).
Similarity between 16p11.2 and 22q11.2 deletions at the regional level We correlated the 22q11.2 group-level deletion-FC-signature with individual connectomes of 16p11.2 deletion carriers and their respective controls. We correlated as well the 16p11.2 group-level deletion-FCsignature with individual connectomes of 22q11.2 deletion carriers and their respective controls. For each region, we tested whether individuals with a deletion had significantly higher (or lower) similarity to the other deletion FC-signatures than their respective controls using a Mann-Whitney U test. We reported significant group differences after FDR correction (q < 0.05) for the number of regions (64).

Gene expression analyses
We aligned the gene expression maps from AHBA to the MIST64 functional parcellation following previously published guidelines 70  To investigate the association between FC alterations and expression patterns of individual genes, we computed Pearson correlations. The null distribution was defined by the same 5000 random FC-signatures described above.
To test the specificity of the relationship between gene expression and FC, we randomly sampled 10000 gene sets (n=24 for 16p11.2 genes and n=37 for 22q11.2 genes) from 15633 genes and re-computed the PLSR 10000 times. The explained variance (R-squared) was used as test-statistics for the null distribution, and the p-value was calculated as the number of times the explained variance of the random gene-set exceeded the variance explained by 16p11.2 or 22q11.2 genes. A similar approach was performed for the individual gene correlations using median correlation as test-statistics.

Data availability
Beta maps from all Connectome wide association studies (16p11.2 deletion and duplication, 22q11.2 deletion and duplication, ASD, SZ, and ADHD) performed in this study are available in the supplemental

Code availability
The processing scripts and custom analysis software used in this work are available in a publicly accessible  Table 1. (d) Low specificity for the relationship between spatial patterns of gene expression and regional FC deletion signatures. In red: the 16p11.2 regional FC associated with the expression patterns of both the 16p11.2 and the 22q11.2 genes. In blue, the 22q11.2 regional FC is associated with the expression patterns of genes in both genomic loci. In purple, 7 regions were found with both deletion FC-signatures, and the expression patterns of genes encompassed in both genomic loci.     Table S1. CWAS beta estimates, ranking, region and networks labels. •