Associated Genetics and Connectomic Circuitry in Schizophrenia and Bipolar Disorder

BACKGROUND
Schizophrenia (SCZ) and bipolar disorder (BD) are severe psychiatric conditions that can involve symptoms of psychosis and cognitive dysfunction. The 2 conditions share symptomatology and genetic etiology and are regularly hypothesized to share underlying neuropathology. Here, we examined how genetic liability to SCZ and BD shapes normative variations in brain connectivity.


METHODS
We examined the effect of the combined genetic liability for SCZ and BD on brain connectivity from two perspectives. First, we examined the association between polygenic scores for SCZ and BD for 19,778 healthy subjects from the UK Biobank and individual variation in brain structural connectivity reconstructed by means of diffusion weighted imaging data. Second, we conducted genome-wide association studies using genotypic and imaging data from the UK Biobank, taking SCZ-/BD-involved brain circuits as phenotypes of interest.


RESULTS
Our findings showed brain circuits of superior parietal and posterior cingulate regions to be associated with polygenic liability for SCZ and BD, circuitry that overlaps with brain networks involved in disease conditions (r = 0.239, p < .001). Genome-wide association study analysis showed 9 significant genomic loci associated with SCZ-involved circuits and 14 loci associated with BD-involved circuits. Genes related to SCZ-/BD-involved circuits were significantly enriched in gene sets previously reported in genome-wide association studies for SCZ and BD.


CONCLUSIONS
Our findings suggest that polygenic liability of SCZ and BD is associated with normative individual variation in brain circuitry.

Their shared genetic background suggests the involvement of common biological mechanisms. Neuropathological and neuroimaging studies have pointed out that SCZ and BD share major pathophysiological changes, such as a loss in dendritic spines of pyramidal neurons in the prefrontal cortex (12), decreased density of interneurons in the parahippocampal cortex (13), and gray matter volume disruptions (14). The 2 conditions further show overlapping pathology in terms of abnormalities in white matter tracts such as the uncinate fasciculus (15) and short-range connections among brain regions relevant to language processing, mood regulation, and working memory (16). Specifying to what extent individual variation in brain connectivity is directly associated with underlying unique and shared polygenic liability can improve our knowledge of the combined genetic and connectomic etiology of psychiatric conditions (17, 18).
We first examined polygenic scores (PGSs) that quantitatively estimate an individual's genetic predisposition for SCZ and BD based on genomic variants (19). Combining genetic and neuroimaging data of the UK Biobank (UKB) (20) with neuroimaging data from disease cohorts (part I) (21,22), we examined structural brain circuits related to combined polygenic effects for SCZ and BD in the healthy brain and demonstrated their relationship to connectivity-based pathology in disease conditions ( Figure 1). Second, we examined the genetic-connectomic association by conducting genome-wide association studies (GWASs) on SCZ-and BD-involved brain circuits (part II), stressing that genomic variants associated with brain circuits play a role in psychiatric conditions.

METHODS AND MATERIALS Compliance With Ethical Standards
The UKB study protocol was approved by the National Research Ethics Service Committee North West Haydock (reference 11/ NW/0382), and all procedures were conducted following the ethical principles for medical research declared in the World Medical Association Declaration of Helsinki. The COBRE SCZ study protocol was approved by the institutional review board of the University of New Mexico. The MACS BD study was approved by the ethics committees of the medical faculties of the University of Marburg and the University of Münster.  Table S1).
Genotype Data. Imputed genotype data of 9,203,453 genetic variants from the UKB were studied (26) [details described in (26,27)]. Subjects of European ancestry were association analysis was performed within the healthy sample to examine associations between structural connectivity and polygenic scores (PGSs) for schizophrenia (SCZ) and bipolar disorder (BD). PGSs were computed using genotype data from the UK Biobank and genome-wide association study (GWAS) summary statistics from a previous GWAS study on SCZ and BD from the Psychiatric Genomics Consortium (PGC). Between-group connectivity differences were also examined between healthy individuals and individuals with SCZ and/or BD. (C) Spatial correlation analysis was performed between the spatial pattern of PGS-connectivity associations and the pattern of case-control connectivity differences. (D) Two independent datasets, namely the Center for Biomedical Research Excellence (COBRE) SCZ dataset and the Marburg-Münster Affective Disorders Cohort Study (MACS) BD dataset, were used to study connectivity differences between healthy control subjects and patients with SCZ or BD. Spatial correlations between the spatial pattern of PGS-connectivity associations and the pattern of case-control connectivity differences were then validated. Connections showing deficits in disease conditions were identified and were taken as the phenotypes of interest in GWAS analysis, which was conducted using genotype data from the UK Biobank.

Associated Genetics and Connectomics in Psychiatry
Biological Psychiatry July 15, 2023; 94:174-183 www.sobp.org/journal included in the current study (27) (Supplemental Methods). Genetic principal components (PCs) were computed within the full UKB sample, based on a set of 145,432 independent (r 2 , 0.1) autosomal single nucleotide polymorphisms (SNPs) using FlashPCA (28). The first 20 PCs were used as covariates to correct for population stratification (29).
PGS Calculation. PGSs were calculated on imputed genotype data for each individual from the UKB, based on summary statistics from GWASs on SCZ and BD (10) (Table S2). PGS regarding combined polygenic effects for SCZ and BD (referred to as SCZ1BD) was computed using the GWAS comparing combined patients with SCZ and BD with HCs (10). PGS for the differentiated polygenic effects between SCZ and BD was also calculated (referred to as SCZ2BD) (10). Summary statistics from the 2 most recent GWASs for SCZ (30) and BD (31) were included for validation purposes. Eleven GWASs on other psychiatric and neurological conditions were used to examine to what extent results relevant to SCZ and BD are specific and/or are generalizable to other conditions ( Magnetic Resonance Imaging Data. T1-weighted magnetic resonance imaging (MRI) and diffusion weighted imaging data were used for reconstruction of brain connectivity circuits. Scanning parameters and data processing are summarized in (33,34) and Supplemental Methods. A 114 3 114 connectivity matrix describing all reconstructed region-to-region connections was formed for each subject using FreeSurfer (version 6.0) and CATO (version 3.1.2) (35-37) (for details, see Supplemental Methods). Considering a high heritability as evidenced by both twin MRI studies (38) and GWAS (39), mean fractional anisotropy (FA) of reconstructed tractography streamlines was taken as a metric of the strength of connections. Results of streamline density weighted connectivity are summarized in Supplemental Results.
Linear Regression Analysis. Linear regression analysis was used on the discovery dataset to identify relationships between polygenic effects for the different contrasts (SCZ, BD, SCZ1BD, and SCZ2BD) and brain connectivity (global-, regional-, and connectionwise). Global connectivity strength (mean strength across connections), regional connectivity strength (mean strength of connections of a region), and strength of single connections were used, respectively, for global-, regional-, and connectionwise analysis. For connectionwise analysis, group thresholding was applied by selecting consistent connections (N = 1311) that were mapped in .60% of the subjects (40). PGS-connectivity association was examined using the following formula: where x i indicates the standardized PGS of subject i, y i the standardized connectivity strength, c ip the standardized pth covariate, and ε i the residual. Age, sex, genotyping array, assessment center, and 20 ancestry PCs were included as the p covariates in the model. The standardized regression coefficient b indicates the effect size, with t tests performed to express the corresponding p value. Networkbased statistic (NBS) analysis (41) was used to control familywise error rate and identify subnetworks showing significant PGS-connectivity associations (Supplemental Methods).
Cross-reference to Disease Conditions. Connectivity maps were similarly formed for the 26 SCZ and 124 BD cases in the UKB dataset. Two-tailed two-sample t tests were performed on the 1311 connections in the connectivity matrix to assess connectionwise differences for SCZ1BD compared with matched HCs (150 randomly selected HCs, matched for age and sex of the SCZ and BD groups, respectively). Resulting t scores were correlated to the standardized regression coefficient b obtained in the PGS-connectivity association analysis across all 1311 connections, with permutation testing performed to rule out the spatial autocorrelation effects (Supplemental Methods). Similar analyses were conducted for the COBRE SCZ and the MACS BD datasets for validation.

Part II: GWAS Analysis on SCZ-/BD-Involved Brain Circuits
Abovementioned analyses focused on identifying brain connectivity in relation to PGSs for SCZ and BD in healthy subjects. We further examined the genetic-connectomic association using GWAS analysis on a phenotype capturing subnetworks of connections related to SCZ and BD. SCZinvolved connections and BD-involved connections were selected using the external COBRE SCZ and MACS BD datasets, computed by means of two-sample t tests on all connections between the patients and HCs in these datasets. Disease-involved connections were selected if two-sided p , .05 and t score , 0 (i.e., connectivity strength reduced in patients, resulting in 46 SCZ-involved connections and 100 BD-involved connections). Next, in the UKB sample (n = 22,799), including both healthy and disease samples (results of healthy samples only are shown in Supplemental Results), the mean strength of the selected SCZ-and BD-involved connections was computed and taken as the phenotype of interest in a following GWAS analysis. GWAS was conducted in PLINK version 2.00 (42), using an additive linear regression model controlling for covariates of age, sex, 20 European-based ancestry PCs, genotyping array, assessment center, and 2 quality control metrics of diffusion weighted imaging data (Supplemental Methods). Total brain volume was additionally taken as a covariate to rule out genetic effects that are generally related to the brain. Genetic correlation analysis was conducted between the resulting GWAS summary statistics and the summary statistics for SCZ1BD, SCZ, BD, and SCZ2BD, using linkage disequilibrium score regression (LDSC) (43,44).  Significant associations between SCZ1BD PGS and regional connectivity strength were observed for the left posterior cingulate and superior parietal regions and the right lateral occipital and anterior cingulate regions (q , .05, FDR corrected across 114 brain regions) (Figure 2A; Table S4). Similar results were observed for PGSs for SCZ and BD separately ( Figure S7). Connectionwise regression analyses revealed 2 subnetworks of a total of 72 connections that showed significant associations with SCZ1BD PGS (b = 20.020 to 20.039, p NBS = .001 and .006) ( Figure 2B). These connections linked intrahemisphere cortical regions including the inferior parietal cortex and insula/superior temporal cortex, supramarginal and lateral orbitofrontal cortex, etc. Post hoc examinations revealed longer connections to show stronger PGS-connectivity associations (r = -0.240, p , .001) (Supplemental Results). A stronger PGS-connectivity association was also observed for connections spanning between type 2 Economo cortical areas (i.e., homotypic frontal cortex; t 1309 = 23.555, p fdr = .002) and connections spanning between regions involved in cognitive domains of cognition and manipulation (t 1309 = 23.312, p fdr = .006 and t 1309 = 22.723, p fdr = .020, respectively) (Supplemental Results). Correlating SCZ PGS to connectivity strength showed significant associations in 3 subnetworks of a total of 47 connections (p NBS , .05; 18 connections nested within the above SCZ1BD PGS-identified network). PGS for BD was associated with a subnetwork of 8 connections (p NBS = .020; 6 connections nested within the SCZ1BD PGS-identified network). No specific effects were observed for PGS for SCZ2BD (p NBS = .478).
Post hoc analysis further revealed significant associations between SCZ1BD PGS and within-network connectivity strength of the default mode network (see Supplemental Methods) (45) (b = 20.020, p fdr = .032) ( Figure 2D). PGS for SCZ1BD was associated with between-network mean connectivity strength among examined networks ( Figure 2E), with the highest effect observed for connections spanning between the ventral attention network and dorsal attention network (b = 20.038, p fdr , .001). Results of SCZ PGS are shown in Figure S8. No significant The specificity of the association between SCZ1BD PGS and brain circuits was tested by examining PGSs for other mental conditions. The SCZ1BD PGS subnetwork significantly correlated to PGS for attention-deficit/hyperactivity disorder (46) (b = 20.026, p fdr = .009) and major depressive disorder (47) (b = 20.021, p fdr = .036), but not for other examined mental disorders (Supplemental Results).

PGS-Connectomic Associations Overlapped With
Disconnectivity of Psychiatric Disorders. The spatial pattern of correlations between SCZ1BD PGS and connectivity strength significantly correlated to the pattern of connectivity differences between the groups of patients with SCZ and BD and HCs from the UKB (r 1,309 = 0.239, p permut , .001) (Figure 3). Similar spatial correlations were observed for the patterns of connectivity differences between patients with SCZ and HCs (r 1,309 = 0.214, p permut , .001) and between patients with BD and HCs (r 1,309 = 0.216, p permut = .002). These findings suggest that connections associated with a higher SCZ1BD PGS in the healthy population display larger changes in brain circuitry in clinical groups (SCZ and BD). The association for SCZ remained significant when controlling for polygenic effects of BD (r 1,309 = 0.186, p permut , .001), and vice versa (see Supplemental Results). Outgroup analyses correlating the spatial pattern of PGS-connectivity associations to the pattern of connectivity differences between HCs and non-SCZ and non-BD subjects (n = 6665) with other mental conditions showed a nonsignificant effect (r 1,309 = 0.028, p = .658). Analyses separately on 4 distinct mental disorders, including depression (n = 4731), anxiety (n = 34), autism (n = 31), and obsessivecompulsive disorder (n = 112), similarly resulted in nonsignificant effects (all p . .3), suggesting the observed spatial correlation between SCZ and BD PGSs and brain patterns to be relatively specific to SCZ and BD conditions. PGSs, Connectivity, and Cognition. SCZ and BD share a genetic background with genetics of intelligence and cognition (48). Thus, we further investigated potential interactions across PGSs, brain connectivity, and  (Figure 4), suggesting that higher connectivity strength covaried with lower polygenic risk for SCZ/BD and overall higher scores on fluid intelligence tests. Post hoc analysis using permutation testing showed that this observed effect was relatively specific to the SCZ1BD PGS network with effects significantly exceeding the null distribution of effect sizes obtained when we computed this correlation with randomly selected connections across the brain (p permut = .025, 1000 permutations). Mediation analysis showed connectivity strength of the SCZ1BD PGS-derived subnetworks to significantly mediate the relationship between SCZ1BD PGS and cognitive function (b = 20.002, p , .001), effect accounting for 2.5% of the total effect size for the PGS-intelligence association) (Figure 4). The SCZ1BD PGS subnetwork was also significantly associated with cognitive performance in reaction time (b = 20.062, p , .001), pairs matching (b = 20.033, p , .001), and numeric memory (b = 0.044, p , .001), with no effect observed for prospective memory (b = 0.004, p = .655).
Validation. The robustness of the observed connectomewide association with SCZ1BD PGS was tested using the hold-out dataset (n = 2589) (see Methods and Materials). First, the spatial pattern of PGS-connectivity association reported using the discovery sample again significantly correlated to the pattern of PGS-connectivity association observed in the holdout sample (r 1,309 = 0.150, p , .001). Second, the pattern of connectomewide correlations between PGS for SCZ1BD and connectivity strength similarly showed a positive correlation to the pattern of disconnectivity in the combined group of patients with SCZ and BD (r 1,309 = 0.160, p , .001), as well as to the pattern of disconnectivity in SCZ (r 1,309 = 0.167, p , .001) and BD (r 1,309 = 0.135, p , .001).
Using the independent COBRE SCZ dataset (21,22), we further validated the overlapping patterns of SCZ1BD PGSconnectivity correlations (as observed in the UKB healthy sample) and the connectivity differences between SCZ and HCs (r 1,309 = 0.18, p permut = .006) ( Figure S9). Analyzing the independent MACS BD dataset (24,25) (n = 84 patients, 346 HCs) validated the observation of the overlapping pattern between the PGS-connectivity correlations and connectivity difference between HC and BD groups (r 1,309 = 0.129, p permut = .014) ( Figure S9).  The spatial profile of between-group connectivity difference is correlated to the spatial profile of schizophrenia1bipolar disorder PGS-connectivity correlations across healthy subjects (r = 0.239). Permutation testing using randomly permuted group assignment confirms the observed association to be significant (two-sided p , .001, 1000 permutations) involved connections revealed 32 independent significant variants (p , 5 3 10 28 ), tagging 9 independent genomic loci ( Table S5). The SNP-based heritability (h 2 SNP ) estimated by LDSC was 24.0% (standard error = 2.9%). The LDSC intercept of 1.008 was close to 1 and the observed inflation level (l GC ) was 1.099, suggesting that the inflation of genetic signals is mostly due to polygenicity rather than population stratification (49). One of the observed genomic loci (rs3129171; chromosome 6; position 29155749) ( Figure 5B) was within the significant loci reported in a recent SCZ GWAS study (50). Five of the 9 observed loci overlapped with the loci reported in a recent GWAS on brainwide FA (39).
The identified SNPs were mapped to 261 genes using positional mapping, expression quantitative trait loci mapping, and chromatin interaction mapping implemented in FUMA (51). Gene enrichment analysis based on previously curated gene sets showed significant enrichment of the identified genes in the GWAS catalog-reported gene sets of autism spectrum disorder or schizophrenia (p fdr = 1.70 3 10 252 ), schizophrenia (p fdr = 4.05 3 10 211 ), bipolar I disorder (p fdr = 1.31 3 10 23 ), and 12 other traits related to sleep, lung cancer, social communication, and blood protein levels ( Figure 5D).
LDSC genetic correlation analysis showed a trend-level correlation between our GWAS on SCZ-involved connectivity and previous GWAS on SCZ (10) (r g = 20.111, nominal p = .015; p fdr = .058, corrected across 4 tests). Controlling for the effect of global FA (which showed phenotypic correlation with SCZ-involved circuits: r = 0.730), permutation testing (see Supplemental Methods) showed r g for SCZ and SCZ1BD (r g = 20.078, p fdr = .104) to significantly exceed the null distribution of r g yielded by GWAS analysis on same sized random connections (p permut = .030 and .045 for SCZ and SCZ1BD, respectively; 200 permutations) ( Figure 5C). Correlating to GWAS results of BD (r g = 20.052, p fdr = .208) and SCZ2BD (10) (r g = 20.109, p fdr = .104) revealed no additional significant effects.

DISCUSSION
Our study provides biological insights into associated brain connectivity and polygenic liability for SCZ and BD. Combining PGS and neuroimaging shows that healthy individuals with a higher PGS for SCZ and BD display relatively lower levels of connectivity strength in brain circuits matching those involved in disease samples.
Our findings suggest that normative variations of macroscale brain circuitry are associated with combined polygenic effects of SCZ and BD, findings that are in support of a general relationship between polygenic liability and structural (52,53) and functional (54,55) brain organization. SCZ and BD are known to share a common genetic background and several molecular pathways (10,11,56). The observed association between polygenic liability and brain connectivity might be attributable to the role of disorder risk genes in white matter organization (52). Functional studies on disorder risk genes also pinpoint the role of genes in biological processes related to synaptic and oligodendrocytes (10,30), processes that are known to shape the cellular organization (57) and dynamics of brain connectivity (58,59). Our findings converge on crossscale interactions among genetic, cellular, and macroscale brain organization in the context of shared polygenic effects for SCZ and BD (58,60).
Association analysis on PGSs in a healthy population may identify new brain circuits that are potentially related to disease processes. This corroborates previous observations in functional connectivity, which have indicated that several functional networks are related to SCZ's PGSs (55). Structural circuits in this study involve the superior and inferior parietal cortex and the posterior cingulate regions of the default mode network, systems that have been broadly reported to be associated with both SCZ (16,(61)(62)(63) and BD (16,64,65). Combined, these and previous results suggest that accumulating liability for  Associated Genetics and Connectomics in Psychiatry psychiatric disorders in the healthy population can target particular brain substrates that are vulnerable to disease processes.
Neuroimaging and neurocircuitry analysis may provide valuable new endophenotypes to connect genetics and disease conditions (66). GWAS analyses on SCZ-and BDinvolved circuits point to a common genetic architecture of brain white matter integrity and mental health traits (39). Overlapping genes of SCZ-involved circuits and SCZ traits include, for example, ZSCAN31, a gene that regulates pivotal SCZ risk genes such as VIPR2 and NPY, as well as the PI3K-AKT and the NOTCH signaling pathways in SCZ (67), and XPNPEP3, PCDHA7, and PCDHA8, genes reported to show altered expressions in SCZ and BD (68).
Several remarks have to be considered when interpreting our results. First, PGSs explain only a small proportion of variance of case-control differences in SCZ (here, 3.2%-11.5%) and BD (2.3%-9.2%) (69), but it should be noted that PGSs explain an even smaller proportion of variance in brain connectivity (w1%), which is similar to previous literature examining other neuroimaging traits (52,70). Second, our samples are all from European ancestry, which limits the generalizability of the results to populations of different ancestry (32). Third, brain connectivity was reconstructed using tractography, which is known to have several limitations regarding the reconstruction of complex oriented white matter fibers, for example, fibers through the corpus callosum (71,72). This might explain why no interhemispheric connection was found to correlate to PGSs for SCZ1BD (12 of the 142 interhemispheric connections showed nominal p , .05, b = 20.017 to 20.025; p fdr . .05). Fourth, information on antipsychotic treatment and disease duration was not available in the studied data cohorts. Future study on the impact of antipsychotic medication dosage on the reported geneticconnectomic associations is warranted (73).

Conclusions
Our study shows a common genetic background for brain structural connectivity and SCZ and BD, with a combined polygenic liability for the 2 disorders playing a central role in key macroscale brain circuits. The integration of genetics and connectomics may pave an avenue for the transition of the diagnostic practice of psychiatric disorders from a traditional descriptive manner to diagnosis built upon the underlying biological systems of the brain. YW performed the analyses. MPvdH and YW conceived the idea of this study. MPvdH supervised this study. SCdL and ET preprocessed MRI data. JES prepared genetic data and genetic analysis pipeline. TQ performed the mediation analysis. JR and UD provided MACS BD data. DP supervised the genetic analysis pipeline. YW wrote the paper with contributions from all coauthors.

ACKNOWLEDGMENTS AND DISCLOSURES
Codes are available from the corresponding author on reasonable request. Data visualization uses the Gramm toolbox (74) and the Simple Brain Plot (75) implemented in MATLAB (version R2021a; The MathWorks, Inc.).
The UKB genotype data and MRI data that support the findings of this study are available in the UKB (accessed under application 16406; https:// www.ukbiobank.ac.uk). The Centers for Biomedical Research Excellence dataset that supports the findings of this study is available at http:// schizconnect.org. The Marburg-Münster Affective Disorders Cohort Study bipolar disorder data that support the findings of this study are available from the corresponding author on reasonable request.
The authors report no biomedical financial interests or potential conflicts of interest.