A combined study of whole-exome sequencing and structural neuroimaging markers in bipolar disorder

Abstract

Bipolar disorder (BD) is a highly heritable psychiatric disorder. An increasing number of whole-exome sequencing (WES) studies have been conducted on psychiatric research. However, few studies have investigated the association between single nucleotide variants (SNVs) in WES and structural brain changes in patients with BD. We conducted WES on 93 patients with BD and 161 healthy controls (HCs) to detect germline variants and copy number alterations (CNAs), as well as the gene-based effects of rare variants on BD. The impact of genetic factors on structural neuroimaging markers, including cortical thickness and white matter tract integrity, was examined using neuroimaging–genetic association analysis. We identified 382 BD-related genes, including SYNE1, PKHD1L1, FRAS1, and ZFHX4, from a systematic review, and the most frequently mutated genes. We identified a new gene, GNB3, which harbors a significant burden of rare nonsynonymous variants that affect BD. By integrating WES and structural neuroimaging data, we found that the single nucleotide polymorphism (SNP) rs56850341 in KMT2C was substantially correlated with structural neuroimaging measures, including cortical thickness and white matter tract integrity. In the present study, we suggest that rare nonsynonymous variants in GNB3 may be involved in the pathophysiology of BD, and that highly contributing heritable genetic variant in KMT2C may influence brain structural changes.

Introduction

Bipolar disorder (BD) is a complex and severe mental disorder characterized by mood changes ranging from severe depression to extreme mania [1]. BD has a worldwide lifetime prevalence of 1–3% and has a substantial impact on society and the economy because of its damaging effects on the quality of life and suicidality of an individual [2–4]. The World Health Organization (WHO) considers BD a major contributor to the global disease burden [4]. BD is a highly heritable psychiatric disorder, with an estimated heritability of approximately 60–85% [5]. Compelling evidence from genetic studies on families, twins, and adoption indicates considerable genetic contribution to the development of BD [1].

Genome-wide association studies (GWAS) have provided preliminary information on the genetic architecture and quantitative traits of complex diseases by evaluating common variants (minor allele frequency (MAF) > 5%) [6, 7]. GWAS has revealed several common single-nucleotide variants (SNVs) that contribute to the risk of BD [8–10]. However, the heritability of SNV captured by GWAS was estimated to be approximately 40% for BD [11]. This indicates that the genetic risk of BD cannot be assessed entirely using the combined effects of common SNVs.

Investigations into rare variants and copy number alterations (CNAs) may further elucidate the unknown genetic etiology of BD [12]. Advancements in next-generation sequencing methods have enabled the identification of rare variants (MAF < 1%) and CNAs associated with BD [13–17]. Whole-exome sequencing (WES) and whole-genome sequencing (WGS) have revealed an increasing number of specific rare variants in protein-coding or non-coding regions associated with BD [18]. However, few studies have used gene-based rare variant methods to identify genes associated with BD [19, 20]. From a mechanistic perspective, G-protein–mediated neurotransmitter signaling has been implicated in mood disorders; GNB3 encodes the β3 subunit of heterotrimeric G proteins that transduce signals from multiple receptors to intracellular cascades (including cAMP-related pathways) [21, 22]. In parallel, epigenetic/chromatin regulatory mechanisms are increasingly recognized as relevant to psychiatric illness and brain development; KMT2C encodes an enhancer-directed histone H3K4 methyltransferase involved in transcriptional regulation and neurodevelopmental/circuit maturation processes [23, 24].

The pathophysiology of BD involves structural alterations in brain networks that are crucial for emotional, reward, and cognitive processing [25]. Microstructural abnormalities of the white matter tracts involved in BD can be examined using diffusion tensor imaging (DTI) [18, 26]. For instance, lower fractional anisotropy (FA) has been reported in several brain circuits in patients with BD than in healthy controls (HCs) [26, 27]. BD is also associated with abnormalities in cortical morphology and white matter tracts [28]. Previous studies have indicated that individuals with BD exhibit a substantial reduction in cortical thicknesses (CTs) in multiple brain regions, including the frontal and temporal cortices, compared to HCs [29, 30]. CTs and white matter integrity generally exhibit high genetic heritability [31–34]. Recently, advances in neuroimaging techniques and combined neuroimaging-genetic approaches have enabled the analysis of genetic influences on neural phenotypes, emotions, behavior, and cognition in patients with psychiatric disorders [35]. Large-scale GWAS have identified numerous common BD risk loci with small effect sizes, yet common variants explain only a portion of the heritability and leave substantial ‘missing heritability’ [13, 15, 17]. This motivates complementary approaches that interrogate rare coding variation with potentially larger functional impact [36]. In parallel, neuroimaging studies have consistently reported structural brain abnormalities in BD, and imaging–genetics work has supported the concept that brain structural measures are heritable intermediate phenotypes [34, 35]. However, most imaging–genetics studies have primarily focused on common variants or polygenic scores [31, 37, 38], and relatively few have examined whether rare coding variants are associated with brain structural alterations in BD [36]. Therefore, integrating WES-based rare-variant analyses with structural MRI/DTI may provide novel insights by prioritizing biologically interpretable genes/pathways and by directly linking rare genetic liability to specific neuroanatomical phenotypes. Our previous WES studies identified several genetic variants associated with neuroimaging phenotypes, including major depressive disorder (MDD) and BD [36, 39]. Additionally, Kim et al. reported genes associated with Alzheimer’s disease-related neuroimaging endophenotypes using WGS [40]. Moreover, a GWAS reported correlations between polygenic risk scores for BD and changes in CT [41].

The integration of WES and neuroimaging phenotypes may provide additional information on heritable neural correlates associated with BD and related genetic variants [42]. We were the first to employ a combination of genetic and neuroimaging methods for BD [36]. In the present study, we combined WES and MRI data from patients with BD to investigate the association between heritable genetic variants and abnormalities in brain structure, including CT and white matter tract integrity, by recruiting a larger number of patients than in our previous WES study on BD [36].

In this study, we hypothesized that relatively rare variants with moderate effects would contribute to the complex inheritance patterns of BD. Furthermore, we anticipated that heritable genetic variants might induce structural alterations in the brains of patients with BD. To investigate these hypotheses, we sequenced whole exomes of 93 patients with BD and 161 HCs to analyze rare variants at the gene level and evaluated the effects of single nucleotide polymorphisms (SNPs) on CTs and white matter tract integrity.

Materials and methods

Sample subjects

This study involved 93 patients with bipolar I or II and 161 HCs from an outpatient psychiatric clinic at Korea University Anam Hospital, Seoul, Republic of Korea, from January 2015 to August 2021. The patients with BD included in the study were aged ≥ 19 years, and their diagnoses of bipolar I or II disorder were determined by board-certified psychiatrists (B.-J.H. and K.-M.H.) through the Structured Clinical Interview from the fifth edition of the Diagnostic and Statistical Manual of Mental Disorders (DSM-5) Axis I disorders [43]. The inclusion criteria for the BD group were as follows: (i) patients with BD I or II aged 19–69 years and (ii) current euthymic or depressive state, operationally defined as a score of ≤ 12 on the 11-item Young Mania Rating Scale (YMRS) [44], indicating the absence of (hypo)manic symptoms [45, 46]. Patients meeting criteria for (hypo)mania were explicitly excluded. Depressive symptom severity was assessed using the 17-item Hamilton Depression Rating Scale (HDRS) for characterization of the sample [47], but no HDRS cut-off was applied because eligibility was determined by the absence of (hypo)manic symptoms rather than by a depressive symptom threshold. We excluded patients according to the following criteria: (i) comorbid status of any other major psychiatric disorders (including personality and substance use disorders); (ii) severe suicidal risk requiring prompt inpatient treatment; (iii) history of severe or unstable medical illness; (iv) primary neurological illness (e.g., Alzheimer’s Disease, Parkinson’s disease, encephalitis, or cerebrovascular disease); and (v) any restrictions for MRI scanning [36]. A total of 161 HCs aged ≥ 19 years were selected from the community using advertisements. For HCs, the presence of current or past psychiatric history was used as an additional criterion. Among the entire sample, 81 patients with BD and 154 HCs underwent MRI and were included in neuroimaging analyses. The detailed demographic and clinical characteristics of the patients included in the neuroimaging analyses are shown in Table 1. The severity of depressive symptoms was evaluated using the 17-item Hamilton Depression Rating Scale (HDRS) [47]. The severity of manic symptoms in patients with BD was assessed using the 11-item Young Mania Rating Scale (YMRS) [44]. The duration of the disease in patients with BD was estimated as the cumulative number of months with episode(s) of depressive and (hypo) manic episodes over a lifetime using the life-chart methodology. Medication status was obtained from medical records and participant interviews at the time of MRI acquisition. All BD participants included in the neuroimaging sample were receiving psychotropic medication, and current medications were categorized into major classes (e.g., antidepressants, antipsychotics, and mood stabilizers including lithium and antiepileptic drugs); the detailed distribution is summarized in Table 1. The study protocol was approved by the Institutional Review Board (IRB) of the Korea University Anam Hospital (2017AN0185). Written informed consent was obtained from all participants prior to inclusion in the study, which was conducted in accordance with the Declaration of Helsinki.

Table 1.

Demographics and clinical characteristics of participants.

Total sample	BD	HCs	P-value (t, χ2)
N	93	161	NA
Age	33.01 ± 10.70	38.41 ± 13.94	<0.001 (t = 3.46)
Sex (Female/Male)	61 (65.6%) / 32 (34.4%)	104 (64.6%) / 57 (35.4%)	0.981 (χ2 < 0.001)
Education years	13.82 ± 2.72	15.04 ± 2.26	<0.001 (t = 3.65)
HDRS−17 score	8.03 ± 5.67	0.99 ± 1.74	<0.001 (t = −11.66)
Remission state/depressive state	49 (52.69%) / 44 (47.31%)	NA	NA
Illness duration (months)	27.48 ± 21.03	NA	NA
Neuroimaging sample	BD	HCs	P-value (t, χ2)
N	82	154	NA
Age	32.78 ± 10.96	38.52 ± 14.04	<0.001 (t = 3.47)
Sex (Female/Male)	53 (64.63%) / 29 (35.36%)	98 (63.64%) / 56 (36.36%)	0.99 (χ2 < 0.001)
Education years	13.65 ± 2.38	15.01 ± 2.27	<0.001 (t = 4.24)
HDRS−17 score	8.12 ± 5.58	1.01 ± 1.76	<0.001 (t = −11.24)
Remission state/depressive state	43 (52.44%) / 39 (47.56%)	NA	NA
Illness duration (months)	27.19 ± 20.85	NA	NA
TICV (cm3)	1422.07 ± 159.08	1460.69 ± 148.89	0.07 (t = 1.81)
Drug-naive/drug-treated patients (n)	0 / 82 (100%)	NA	NA
Medication, n			NA
Antidepressants
None	48 (58.54%)	NA
SSRI	14 (17.07%)
SNRI	6 (7.32%)
NDRI	9 (10.97%)
NaSSA	2 (2.44%)
Others	2 (2.44%)
Combination of AD	1 (1.22%)
Antipsychotics
None	12 (14.63%)
AP	44 (53.66%)
Combination of AP	26 (31.71%)
Mood stabilizers
None	7 (8.54%)
Lithium	15 (18.29%)
AED	55 (67.07%)
Combination of lithium and AED	3 (3.66%)
Double AED	2 (2.44%)

Data are presented as mean ± standard deviation for age, years of education, HDRS−17 scores, illness duration, and TICV.

The P-values of the comparisons of age, years of education, HDRS−17 scores, and TICV are calculated using independent t-tests.

The P-values of the sex distribution are calculated using the chi-square test.

BD bipolar disorder, HCs healthy controls, HDRS−17, 17-item Hamilton Depression Rating Scale, TICV total intracranial cavity volume, SSRI selective serotonin reuptake inhibitor, SNRI serotonin and norepinephrine reuptake inhibitor, NDRI norepinephrine-dopamine reuptake inhibitor, NaSSA noradrenergic and specific serotonergic antidepressant, AD combination of two or more types of antidepressants, AP combination of two or more types of antipsychotics.

Processing of WES data

Genomic DNA was acquired from the peripheral blood of 93 patients with BD and 161 HCs, and the whole exome was captured using an Agilent SureSelect Human All Exome V5 kit, according to the manufacturer’s instructions (Agilent, Santa Clara, CA, USA). 101 bp paired-end sequencing was performed using HiSeq2000, HiSeq2500, or HiSeq4000 (Illumina, San Diego, CA, USA). The quality of the raw FASTQ data was evaluated using FastQC (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/; v0.11.9). Low-quality reads and adapter sequences were removed using Trimmomatic (v0.36) [48]. The reads were aligned to the GRCh38 human reference genome using the Burrows−Wheeler Aligner-Maximal Exact Match (BWA-MEM, v0.7.17-r1188) [49]. Using the Genome Analysis Toolkit (GATK, v4.2.0.0), duplicate reads were marked, and base quality scores were recalibrated [50]. Aligned reads with a mapping quality < 20 were removed using Samtools (v1.10) [51].

Germline variants detection and selection of BD-related genes

Germline variants were identified using GATK Best Practices (v4.2.0.0) [50]. Briefly, SNVs, insertions, and deletions were detected using HaplotypeCaller in GVCF mode. Subsequently, the GVCF files were merged using GenomicsDBImport and joint genotyping was performed using GenotypeGVCFs to detect potential variants at the per-sample level. Variants were filtered using SelectVariants and VariantFiltration based on the following criteria: QualByDepth (QD) < 2.0, FisherStrand (FS) > 200.0, and ReadPosRankSumTest (ReadPosRankSum) < −20.0. The effects of variants within the coding regions were annotated by ANNOVAR (ANNOtate VARiation) [52]. To determine BD-related genes, a systematic literature review was conducted using the following terms ((“WES” OR “WGS”) AND (“bipolar disorder” OR “mood disorders” OR “manic depress” OR “bipolar”)) in English-language peer-reviewed papers published up to March 2024 in the PubMed database. A total of 440 BD-related genes from 26 previous WGS or WES studies were selected and mapped to 93 BD genomes for further analysis.

Gene-based association test for rare variants

Using the FastPCA algorithm, we performed principal component analysis to summarize information on the genetic burden associated with the sequenced genomic regions [53]. Encompassing age, sex, and the two principal components as covariates, the association between the diagnosis of BD and rare variants was investigated using Rvtests (v20190205) with 1,000,000 permutations [54]. Only rare coding variants with three or more minor allele counts were included in the analysis under two MAF thresholds (MAF ≤ 0.1% and MAF ≤ 1%) and two functional categories (nonsynonymous and damaging) annotated by ANNOVAR [52]. Four types of statistical tests were performed: combined multivariate and collapsing (CMC) [55], sequencing kernel association test (SKAT) [56], optimal sequencing kernel association test (SKATO) [57], and the variable threshold model by permutation (VT) [58]. Subsequently, the aggregated Cauchy association test (ACAT) was used to improve the statistical power of the analysis by combining all four test results, as reported by Liu et al. [59].

$$P_{\mathit{ACAT}}=\frac{1}{4}\sum _{i=1}^4\tan \left\{\left(0.5-p_i\right)\pi \right\},$$

where p_i is the P-value of each of the four tests (CMC, VT, SKAT, or SKATO). The four tests were considered equal when combined.

Recurrent germline CNAs

Germline CNAs were defined for 93 patients with BD using the CNVkit (v0.9.9) and a hidden Markov model approach [60]. CNAs regarded as false positives were filtered using the built-in quality control procedures of CNVkit, which excludes low-confidence segments based on coverage and segmentation noise. Significant and recurrent CNAs were identified using GISTIC2.0 [61] on the segmentation data generated by the CNVkit. In GISTIC2.0, peak strength is quantified using G-scores, calculated as the product of CNA frequency and amplitude, while statistical significance is evaluated independently using permutation-based q-values (default threshold q < 0.25).

Case–control association test using common SNPs

Germline variants were filtered using the following criteria: SNPs with P-value < 0.001 in the exact Hardy–Weinberg equilibrium test, samples with call rates < 95%, and SNPs with call rates < 95%. The remaining variants were tested to identify common SNPs associated with BD using a chi-squared test in PLINK (v1.07) [62].

Obtaining and processing of neuroimaging data

Eighty-two patients with BD and 154 HCs from the entire sample underwent T1-weighted and DTI using a 3.0-Tesla TrioTM whole-body MRI scanner (Siemens Healthcare GmbH, Erlangen, Germany). The detailed imaging parameters are presented in the Supplementary Material. To calculate CTs in the brain, T1 image data were used with the automated processes of FreeSurfer v7.2 (Laboratory for Computational Neuroimaging, Athinoula A. Martinos Center for Biomedical Imaging, Charlestown, MA, USA; http://surfer.nmr.mgh.harvard.edu), which reconstructed a 3D model of the cortical surface calculated from T1 images. The detailed process of cortical region reconstruction was described in our previous studies [63, 64]. The thicknesses of the 74 cortical gyri and sulci in each hemisphere were calculated using FreeSurfer, according to the Destrieux atlas [65]. The thicknesses of the 38 cortical gyri in each hemisphere were included in the neuroimaging analysis. To measure white matter tract integrity, the Tracts Constrained by Underlying Anatomy (TRACULA), developed by Yendiki et al. [66] and implemented in FreeSurfer, was used to automatically compute four DTI parameters from the 18 major white matter tracts: FA, radial diffusivity (RD), mean diffusivity (MD), and axial diffusivity (AD). FA can summarize the measure of white matter tract integrity and reflect the density of axon fibers and white matter maturation [67, 68]. AD reflects the size of axons and may be sensitive to axonal damage associated with the microstructure and connectivity of the white matter tracts [67, 68]. In contrast, MD is considered a marker of cellularity, edema, and necrosis [67, 68]. RD is considered a sensitive marker of axonal degeneration demyelination [67, 68]. Using DTI data and automated global probabilistic tractography processes, TRACULA reconstructs the integrity of 18 major white matter tracts in both hemispheres [69]. In this study, we used all four complementary parameters of white matter integrity in the neuroimaging analysis. In a previous study, detailed information on 18 white matter tracts was described [66]. Briefly, the information was as follows: forceps minor and forceps major of the corpus callosum, anterior thalamic radiation (ATR), cingulum-angular bundle (CAB), cingulum-cingulate gyrus bundle (CCG), corticospinal tract (CST), inferior longitudinal fasciculus (ILF), superior longitudinal fasciculus-parietal bundle (SLFp), superior longitudinal fasciculus-temporal bundle (SLFt), and uncinate fasciculus (UNC).

Statistical analysis for neuroimaging–genetic association

We evaluated the association between neuroimaging markers (i.e., DTI parameters and CTs) and SNPs in patients with BD and HCs. The SNPs were selected based on three criteria: (i) significant results (P < 0.001) in the case–control association test; (ii) variants with a high mutation frequency (≥ 20%) in our samples from BD-related genes; and (iii) genes with P < 0.05 in the gene-based rare variant association analysis. To examine differences in neuroimaging markers between patients with BD and HCs, a one-way analysis of covariance (ANCOVA) was conducted with the following variables: (i) diagnosis (BD vs. HCs) as the independent variable; (ii) extracted neuroimaging markers as dependent variables; and (iii) age, sex, and years of education as covariates. The effects of SNPs were analyzed using a one-way ANCOVA with the following variables: (i) genotype under the dominant model (reference allele homozygotes vs. alternative allele carriers) as the independent variable; (ii) extracted neuroimaging markers as dependent variables; and (iii) age, sex, and years of education as covariates. Additionally, a two-way ANCOVA was performed to examine genotype-by-diagnosis interactions on neuroimaging markers, with genotypes and diagnosis (BD vs. HCs) as independent variables, extracted neuroimaging markers as dependent variables, and age, sex, and years of education as covariates. All analyses of cortical thickness were additionally adjusted for total intracranial volume (cm³). Effect sizes were calculated as Cohen’s d based on adjusted means and pooled residual standard deviation from the model. For the genotype effects, standardized coefficients were calculated using the ratio of the standard deviation of the independent variable to that of the MRI measure. Partial R² was calculated using the difference in explained variance between the full model and a reduced model excluding the genotype term, defined as (R²_full − R²_reduced)/(1 − R²_reduced). Bonferroni correction was applied to adjust for multiple comparisons and control the risk of type I errors. Sensitivity analyses were conducted for SNPs showing significant associations in the primary analyses to assess the robustness of the findings against potential confounding caused by medication use. Medication exposure was defined according to major medication groups, with detailed definitions provided in the Supplementary Methods.

Results

Germline variant profiles of BD-related genes

In the present study, 93 patients with BD and 161 HCs were analyzed for germline variants. The average sequencing depth was 81.91X (58.71–140.53X) for all samples. For patients with BD, an average of 43,955 variants per sample (median, 43,979; range, 43,094–44,771) was obtained using WES analysis. Specifically, 9554 non-silent (median 9543; range 9365–9771) and 10,139 silent (median 10,137; range 9949–10,329) variants were detected. The most common point mutations were C > T (53%) and T > C (20%). After a comprehensive systematic review of the literature, 440 BD-related genes were identified, 382 of which 440 genes were identified in our study from 93 patients with BD (Supplementary Table S1). Rare missense variants in coding regions with MAF < 1% from these genes were filtered using the Genome Aggregation Database, 1000 Genomes Project, and the Korean Variant Archive [70]. All 93 patients with BD had 1145 rare coding missense variants in 382 BD-related genes. The five most frequently mutated genes were SYNE1 (28%), PKHD1L1 (23%), FRAS1 (22%), ZFHX4 (22%), and AHNAK (18%) (Fig. 1).

Fig. 1. Mutational landscape of 93 patients with BD.

The top 15 most frequently mutated genes are also identified. The left-side label of the plot shows mutated genes related to BD (rows) in patients with BD (columns). The bar plot on the right side indicates the number of mutated patients with BD. The percentages indicate the proportion of mutated patients with BD for each gene. On the bottom, the stacked bar plot displays the proportion of mutation events classified into six transversions and transitions in each sample. “Multi-Hit” indicates that more than one mutation in a gene is detected from one patient.

Gene-based association test for rare variants

Four tests, including CMC, SKAT, SKATO, and VT, were performed to detect genes associated with the diagnosis of BD using two functional categories (“damaging” or “nonsynonymous”) and two MAF thresholds (“MAF ≤ 1%” or “MAF ≤ 0.1%”). To enhance the statistical power of our analysis, the P-values of the four tests were combined using ACAT [59] which is powerful even if the proportion of causal variants is small or if some uninformative functional annotations are included. For the “nonsynonymous” category, STARD8 showed nominally significant associations with BD under two MAF thresholds (“MAF ≤ 1%” and “MAF ≤ 0.1%”) (Table 2). OR2T11 and KMT2C showed nominally significant associations with BD at “MAF ≤ 1%” and “MAF ≤ 0.1%” thresholds, respectively (Table 2). Interestingly, for the “damaging” category, GNB3 showed a significant association with BD at the “MAF ≤ 1%” threshold with the following Bonferroni thresholds: P-value < 2.45 × 10⁻⁵ (0.05 / 2041 genes) (Table 2).

Table 2.

Gene-based rare variant test results (BD vs. HCs).

Nonsynonymous
Gene	MAF ≤ 0.1%	MAF ≤ 1%
OR2T11	5.60 × 10−2	4.95 × 10−2
STARD8	1.91 × 10−2	4.27 × 10−2
KMT2C	4.58 × 10−2	5.53 × 10−2
Number of genes	4724	7223
P-value threshold	1.06 × 10−5	6.92 × 10−6
Damaging
GNB3	2.20 × 10−1	8.00 × 10−6
Number of genes	1182	2041
P-value threshold	4.23 × 10−5	2.45 × 10−5

After adjusting for the number of genes in each group, the P-values in bold indicate significance. We first use the ACAT method for P-values that converts p-values (CMC, SKAT, SKATO, and VT) to Cauchy variables and then use their weighted sum as the test statistic to analyze significance. Four categories resulting from filtering for minor allele frequency (MAF) and functional prediction were included.

BD bipolar disorder, HCs healthy controls, MAF minor allele frequency, ACAT aggregated Cauchy association test.

Recurrent germline CNAs in patients with BD

CNAs were detected in patients with BD and recurrent CNA regions were identified. A total of 29 recurrent focal amplifications and 30 recurrent focal deletions were detected in the BD genomes (Fig. 2). Copy number gains at 15q11.1 in 41 (44%) patients with BD and losses at 14q32.33 in 42 (45%) patients with BD were the most recurrent CNAs. Recurrent focal amplification and deletion of the BD-related genes were detected at 5q13.2 (BDP1). The sizes of the CNA regions associated with the BD-related genes are listed in Supplementary Table S2.

Fig. 2. Recurrent CNAs in patients with BD.

A Recurrent regions with focal amplification (red lines) and (B) deletion (blue lines) identified using GISTIC 2.0 analysis in 93 patients with BD and 161 HCs. Peak strength is shown on the top scale as G-scores (G = frequency × amplitude), whereas statistical significance is shown on the bottom scale as q-values. The green lines denote the cut-off points for significance (q-value < 0.25) and the vertical axis indicates the number of chromosomes. BD-related genes and their corresponding regions are shown in bold. CNAs copy number alterations, BD bipolar disorder, HCs healthy controls.

Analyses of structural neuroimaging markers and disease-related variables

Structural neuroimaging markers, including CTs and DTI parameters, were compared between the 82 patients with BD and 154 HCs (Supplementary Tables S3 and S4). Compared to HCs, patients with BD had significant thinning in 49 cortical regions, including the frontal pole, medial and lateral parts of the orbitofrontal cortex (OFC), dorsolateral and ventrolateral prefrontal cortex, subgenual, pregenual, and dorsal parts of the anterior cingulate cortex (ACC), posterior cingulate cortex, and several temporal and parietal regions in the bilateral hemispheres. CT was not substantially higher in the BD group than in the HC group. Detailed results comparing the CTs scans of patients with BD and HCs are presented in Supplementary Table S3. For DTI parameters, the BD group had significantly reduced FA in the left and right ILF, right CCG, FMajor, and FMinor, and increased AD, MD, and RD in 52 white matter tracts, including FMajor, FMinor, ATR, CAB, CCG, CST, ILF, SLFP, SLFT, and UNC, compared to HCs after Benjamini–Hochberg correction (Supplementary Table S4).

Pearson’s partial correlation between illness duration and neuroimaging markers was estimated. After Benjamini–Hochberg correction, illness duration was considerably and negatively correlated with the thicknesses of the four cortical regions, including the bilateral orbital gyri (i.e., lateral part of the OFC, left: r = −0.345, adjusted P-value = 1.53 × 10⁻²; right: r = −0.345, adjusted P-value = 1.53 × 10⁻²), right gyrus rectus (i.e., medial part of the OFC, r = −0.327, adjusted P-value = 2.34 × 10⁻²), and right triangular part of the inferior frontal gyrus (r = −0.314, adjusted P-value = 2.98 × 10⁻²) (Supplementary Table S5). The DTI parameters were not significantly correlated with illness duration after Benjamini–Hochberg correction (adjusted P-value > 0.05) (Supplementary Table S6).

Association test between neuroimaging markers and genotypes

The associations between neuroimaging markers and SNPs were examined in patients with BD and HCs. A total of 189 non-synonymous SNPs were included in the analysis, selected based on the following criteria: (i) significant results (P < 0.001) from the case–control association test (14 SNPs); (ii) SNPs from BD-related genes with high mutation frequencies (≥ 20%) in our samples (134 SNPs); and (iii) genes with P < 0.05 in gene-based rare variant association test (41 SNPs). Detailed information on these SNPs is provided in Supplementary Table S7.

The tests were conducted under six conditions: (i) CTs in patients with BD; (ii) DTI parameters in patients with BD; (iii) CTs in HCs; (iv) DTI parameters in HCs; (v) CTs in combined patients with BD and HC samples; and (vi) DTI parameters in combined BD and HC samples. To control for type I errors, Bonferroni correction was applied separately for the two MRI modalities using the following thresholds: (i) P < 3.48 × 10⁻⁶ (0.05 / [76 brain regions × 189 SNPs]) for CT analyses; and (ii) P < 3.67 × 10⁻⁶ (0.05 / [4 DTI parameters × 18 white matter tracts × 189 SNPs]) for DTI analyses.

For the combined sample of patients with BD and HCs, SNP rs56850341 in KMT2C showed a significant genotype effect on CT of the right lingual gyrus after Bonferroni correction (P < 3.48 × 10⁻⁶) (Fig. 3A and B, Table 3; Supplementary Table S8). Homozygotes for the G allele of KMT2C rs56850341 exhibited a thinner CT in the brain region than A allele heterozygotes. In addition, this SNP showed a significant effect on DTI parameters following Bonferroni correction (P < 3.67 × 10⁻⁶) (Fig. 3C; Table 3; Supplementary Table S9). In the combined BD and HC sample, G allele homozygotes of KMT2C rs56850341 showed significantly higher MD, RD, and AD in the white matter tracts, including the FMajor, FMinor, ATR, CAB, CCG, CST, ILF, SLFP, SLFT, and UNC, than A allele heterozygotes. However, no significant SNPs exhibited genotype-by-diagnosis interactions. After additionally adjusting for major medication groups in sensitivity analyses, the genotype–imaging associations for KMT2C rs56850341 remained qualitatively unchanged, with consistent directions and comparable effect sizes (Supplementary Tables S10 and S11).

Fig. 3. Significant differences in cortical thickness and white matter tract integrity according to SNP genotypes in patients with BD and HCs.

A The cortical region (right lingual gyrus) showed a significant difference in thickness according to rs56850341 genotypes. B Boxplots indicate that G allele homozygotes of rs56850341 exhibited a thinner cortical thickness in the right lingual gyrus than A allele carriers. C Probabilistic reconstruction of 18 major white matter pathways generated by TRACULA using diffusion tensor imaging (DTI) from one study participant (superior, inferior, and left lateral views). D White matter tracts showing the most significant differences in DTI parameters (top 30% of 51 DTI parameters) between G allele homozygotes of rs56850341 and A allele carriers. Asterisks denote significance after Bonferroni correction, with corresponding P-values listed in Table 3. BD bipolar disorder, HC healthy control, SNP single nucleotide polymorphism, KMT2C lysine N-methyltransferase 2C.

Table 3.

Significant associations between KMT2C rs56850341 and structural neuroimaging markers.

SNP / Cortical thicknesses	Mean	SD	Mean	SD	Genotype					Genotype*Diagnosis
					Cohen’s d	Standardized β	Partial R2	F	P-value	F	P-value
KMT2C rs56850341 (chr7:152273846)*	GG (n = 116)		GA (n = 120)
Right lingual gyrus	1.86	0.12	1.94	0.13	−0.70	0.33	0.11	28.22	2.56 × 10−7	0.06	8.01E-01

SNP / White matter tracts	Mean	SD	Mean	SD	Genotype					Genotype*Diagnosis
					Cohen’s d	Standardized β	Partial R2	F	P-value	F	P-value
KMT2C rs56850341 (chr7:152273846)*	GG (n = 116)		GA (n = 120)
AD LH CST	1.27 × 10−3	6.79 × 10−5	1.21 × 10−3	4.77 × 10−5	1.16	−0.50	0.14	77.50	3.18 × 10−16	0.11	7.46 × 10−1
AD RH CST	1.25 × 10−3	8.22 × 10−5	1.17 × 10−3	5.52 × 10−5	1.08	−0.47	0.22	66.25	2.46 × 10−14	0.11	7.37 × 10−1
MD RH CST	7.55 × 10−4	6.50 × 10−5	6.99 × 10−4	4.32 × 10−5	1.06	−0.47	0.17	64.83	4.30 × 10−14	0.00	9.90 × 10−1
MD LH CST	7.70 × 10−4	5.36 × 10−5	7.25 × 10−4	3.54 × 10−5	1.06	−0.47	0.23	64.57	4.78 × 10−14	0.47	4.95 × 10−1
AD LH ATR	1.15 × 10−3	6.63 × 10−5	1.09 × 10−3	5.06 × 10−5	1.06	−0.47	0.20	64.11	5.73 × 10−14	0.00	9.93 × 10−1
AD LH SLFT	1.20 × 10−3	6.14 × 10−5	1.15 × 10−3	4.15 × 10−5	1.06	−0.47	0.19	63.92	6.18 × 10−14	0.03	8.69 × 10−1
AD RH ILF	1.30 × 10−3	7.59 × 10−5	1.23 × 10−3	6.32 × 10−5	1.05	−0.46	0.19	62.96	9.08 × 10−14	0.00	9.55 × 10−1
AD LH ILF	1.30 × 10−3	6.69 × 10−5	1.24 × 10−3	5.61 × 10−5	1.04	−0.46	0.13	62.09	1.28 × 10−13	0.06	8.03 × 10−1
AD RH SLFT	1.18 × 10−3	7.91 × 10−5	1.12 × 10−3	5.63 × 10−5	1.00	−0.45	0.18	57.51	8.14 × 10−13	0.40	5.29 × 10−1
AD RH ATR	1.15 × 10−3	7.26 × 10−5	1.09 × 10−3	5.27 × 10−5	1.00	−0.45	0.15	57.38	8.60 × 10−13	0.02	8.76 × 10−1
MD LH ATR	7.74 × 10−4	4.67 × 10−5	7.38 × 10−4	3.46 × 10−5	1.00	−0.44	0.19	56.79	1.09 × 10−12	1.64	2.02 × 10−1
MD RH SLFT	7.74 × 10−4	5.54 × 10−5	7.29 × 10−4	3.97 × 10−5	1.00	−0.45	0.18	56.78	1.10 × 10−12	0.06	8.04 × 10−1
MD RH ILF	8.22 × 10−4	5.23 × 10−5	7.79 × 10−4	3.96 × 10−5	0.99	−0.44	0.10	55.76	1.66 × 10−12	0.41	5.23 × 10−1
AD LH UNC	1.22 × 10−3	6.58 × 10−5	1.16 × 10−3	5.41 × 10−5	0.96	−0.44	0.15	52.85	5.52 × 10−12	0.00	9.95 × 10−1
MD RH ATR	7.70 × 10−4	4.95 × 10−5	7.33 × 10−4	3.63 × 10−5	0.96	−0.43	0.16	52.37	6.75 × 10−12	2.49	1.16 × 10−1
RD RH CST	5.10 × 10−4	6.05 × 10−5	4.62 × 10−4	4.34 × 10−5	0.96	−0.43	0.05	52.32	6.88 × 10−12	0.06	8.08 × 10−1
AD LH SLFP	1.16 × 10−3	6.90 × 10−5	1.11 × 10−3	4.66 × 10−5	0.95	−0.43	0.00	52.03	7.76 × 10−12	0.10	7.58 × 10−1
MD LH SLFP	7.86 × 10−4	4.85 × 10−5	7.49 × 10−4	3.49 × 10−5	0.95	−0.43	0.02	51.31	1.05 × 10−11	0.60	4.39 × 10−1
MD RH UNC	7.87 × 10−4	5.80 × 10−5	7.41 × 10−4	4.70 × 10−5	0.94	−0.43	0.12	51.10	1.14 × 10−11	0.40	5.30 × 10−1
MD LH SLFT	7.87 × 10−4	4.55 × 10−5	7.53 × 10−4	3.28 × 10−5	0.92	−0.42	0.04	48.93	2.83 × 10−11	0.64	4.26 × 10−1
MD LH UNC	8.17 × 10−4	4.46 × 10−5	7.83 × 10−4	3.64 × 10−5	0.92	−0.42	0.11	48.92	2.85 × 10−11	0.98	3.23 × 10−1
MD LH ILF	8.24 × 10−4	4.58 × 10−5	7.90 × 10−4	3.67 × 10−5	0.92	−0.42	0.16	48.74	3.08 × 10−11	0.81	3.70 × 10−1
AD FMajor	1.43 × 10−3	7.12 × 10−5	1.38 × 10−3	5.32 × 10−5	0.92	−0.42	0.18	48.67	3.16 × 10−11	0.59	4.44 × 10−1
RD RH SLFT	5.69 × 10−4	4.64 × 10−5	5.34 × 10−4	3.45 × 10−5	0.92	−0.42	0.02	48.67	3.17 × 10−11	0.93	3.37 × 10−1
MD RH SLFP	7.69 × 10−4	5.97 × 10−5	7.24 × 10−4	4.24 × 10−5	0.92	−0.42	0.17	48.48	3.43 × 10−11	0.00	9.89 × 10−1
AD RH UNC	1.19 × 10−3	7.28 × 10−5	1.13 × 10−3	6.60 × 10−5	0.92	−0.42	0.11	48.08	4.05 × 10−11	0.00	9.89 × 10−1
AD LH CCG	1.32 × 10−3	1.01 × 10−4	1.24 × 10−3	6.96 × 10−5	0.92	−0.42	0.15	48.03	4.15 × 10−11	0.41	5.21 × 10−1
MD LH CCG	7.72 × 10−4	5.77 × 10−5	7.30 × 10−4	4.28 × 10−5	0.91	−0.41	0.14	47.00	6.41 × 10−11	0.36	5.51 × 10−1
MD LH CAB	8.45 × 10−4	9.16 × 10−5	7.72 × 10−4	7.34 × 10−5	0.91	−0.41	0.11	46.97	6.49 × 10−11	1.06	3.05 × 10−1
RD RH SLFP	5.79 × 10−4	5.34 × 10−5	5.39 × 10−4	4.01 × 10−5	0.89	−0.41	−0.05	45.66	1.13 × 10−10	0.10	7.48 × 10−1
AD FMinor	1.28 × 10−3	7.84 × 10−5	1.23 × 10−3	6.02 × 10−5	0.88	−0.41	0.16	44.39	1.95 × 10−10	1.49	2.23 × 10−1
RD LH CAB	6.75 × 10−4	8.98 × 10−5	6.06 × 10−4	7.21 × 10−5	0.88	−0.40	0.18	44.08	2.23 × 10−10	0.30	5.85 × 10−1
AD RH SLFP	1.15 × 10−3	7.75 × 10−5	1.09 × 10−3	5.40 × 10−5	0.88	−0.40	0.08	44.07	2.24 × 10−10	0.18	6.70 × 10−1
MD RH CAB	8.03 × 10−4	9.67 × 10−5	7.29 × 10−4	7.94 × 10−5	0.87	−0.40	0.15	43.80	2.51 × 10−10	0.00	9.62 × 10−1
RD RH UNC	5.87 × 10−4	5.62 × 10−5	5.48 × 10−4	4.31 × 10−5	0.86	−0.39	0.00	41.88	5.74 × 10−10	1.01	3.17 × 10−1
RD LH SLFP	5.99 × 10−4	4.17 × 10−5	5.69 × 10−4	3.25 × 10−5	0.85	−0.40	−0.08	41.74	6.10 × 10−10	1.12	2.91 × 10−1
MD RH CCG	7.65 × 10−4	5.48 × 10−5	7.25 × 10−4	4.55 × 10−5	0.85	−0.39	0.12	41.08	8.12 × 10−10	0.00	9.84 × 10−1
RD RH CAB	6.30 × 10−4	9.04 × 10−5	5.66 × 10−4	7.34 × 10−5	0.85	−0.39	0.09	41.01	8.38 × 10−10	0.35	5.57 × 10−1
RD LH CST	5.19 × 10−4	5.30 × 10−5	4.84 × 10−4	3.67 × 10−5	0.84	−0.38	0.10	40.15	1.22 × 10−9	0.66	4.19 × 10−1
MD FMajor	8.06 × 10−4	4.84 × 10−5	7.73 × 10−4	3.56 × 10−5	0.83	−0.38	0.13	39.12	1.91 × 10−9	0.02	9.02 × 10−1
RD LH ATR	5.85 × 10−4	4.23 × 10−5	5.60 × 10−4	3.13 × 10−5	0.80	−0.36	0.20	36.85	5.18 × 10−9	4.51	3.49 × 10−2
AD RH CAB	1.15 × 10−3	1.20 × 10−4	1.06 × 10−3	1.15 × 10−4	0.80	−0.37	0.08	36.38	6.40 × 10−9	0.85	3.57 × 10−1
RD RH ATR	5.82 × 10−4	4.22 × 10−5	5.57 × 10−4	3.40 × 10−5	0.80	−0.36	0.14	36.33	6.52 × 10−9	6.58	1.09 × 10−2
MD FMinor	8.08 × 10−4	4.92 × 10−5	7.76 × 10−4	3.62 × 10−5	0.80	−0.37	0.07	36.32	6.57 × 10−9	0.57	4.50 × 10−1
AD LH CAB	1.19 × 10−3	1.10 × 10−4	1.10 × 10−3	9.75 × 10−5	0.79	−0.37	0.11	35.84	8.11 × 10−9	2.52	1.13 × 10−1
RD RH CCG	5.07 × 10−4	7.14 × 10−5	4.61 × 10−4	5.46 × 10−5	0.79	−0.37	0.13	35.66	8.78 × 10−9	0.22	6.42 × 10−1
RD RH ILF	5.81 × 10−4	4.70 × 10−5	5.52 × 10−4	3.44 × 10−5	0.78	−0.36	0.00	34.41	1.54 × 10−8	1.29	2.58 × 10−1
RD LH SLFT	5.78 × 10−4	4.14 × 10−5	5.54 × 10−4	3.21 × 10−5	0.74	−0.35	0.04	30.96	7.26 × 10−8	1.33	2.50 × 10−1
RD LH UNC	6.16 × 10−4	4.11 × 10−5	5.93 × 10−4	3.59 × 10−5	0.70	−0.32	0.09	28.31	2.45 × 10−7	2.39	1.24 × 10−1
RD LH CCG	4.98 × 10−4	4.70 × 10−5	4.73 × 10−4	4.05 × 10−5	0.66	−0.31	0.11	24.95	1.16 × 10−6	2.83	9.39 × 10−2
RD LH ILF	5.87 × 10−4	4.19 × 10−5	5.66 × 10−4	3.44 × 10−5	0.65	−0.31	0.11	24.29	1.58 × 10−6	1.59	2.08 × 10−1

Bonferroni correction was used as follows: P < 0.05 / (76 cortical thicknesses × 189 SNPs) = 3.48 × 10⁻⁶

Age, sex, total intracranial volume, and years of education were included as covariates.

Bonferroni correction was used as follows: P < 0.05 / (4 DTI parameters × 18 white matter tracts × 189 SNPs) = 3.67 × 10⁻⁶

Cohen’s d values were calculated as the difference in adjusted means between groups, divided by the pooled residual standard deviation; positive values indicate higher values in the reference group.

Age, sex, and years of education were included as covariates.

BD bipolar disorder, HCs healthy controls, SNP single nucleotide polymorphisms, F F-statistic, SD standard deviation, KMT2C lysine N-methyltransferase 2C, FMajor forceps major, FMinor forceps minor of the corpus callosum, ATR the anterior thalamic radiation, CAB cingulum angular bundle, CCG cingulum cingulate gyrus bundle, CST corticospinal tract, ILF inferior longitudinal fasciculus, SLFP superior longitudinal fasciculus-parietal bundle, SLFT superior longitudinal fasciculus-temporal bundle, UNC uncinate fasciculus, FA fractional anisotropy, RD radial diffusivity, MD mean diffusivity, AD axial diffusivity, LH left hemisphere, RH right hemisphere.

* UCSC GRCh38/hg38.

Discussion

We identified rare coding variants in 382 BD-related genes in 93 patients with BD. Among these, SYNE1, PKHD1L1, FRAS1, and ZFHX4 were the most frequently mutated genes (>20%; non-silent mutations). Our results indicate that the burden of rare nonsynonymous variants of GNB3 is substantially associated with BD. For recurrent CNA, focal amplification and deletion at 5q13.2 were identified and annotated with BDP1, a BD-related gene. By integrating WES and structural neuroimaging data, we identified that KMT2C rs56850341 is significantly associated with structural brain changes in CT and white matter tracts integrity. These findings not only provide evidence of the link between genetic factors and BD but also suggest that they may influence brain structural changes in BD.

Gene-based rare variant association analysis revealed that GNB3 had a considerable damaging effect on patients with BD. GNB3 encodes the β3 subunit of G proteins, which are key regulators of cellular responses through the cyclic adenosine monophosphate pathways [71, 72]. G protein dysfunction is closely associated with the pathophysiology of numerous psychiatric disorders [73]. The protein levels of G-protein subunits, including β3, are elevated in the postmortem brains of patients with BD, indicating that BD may be associated with alterations in G-protein-mediated neurotransmitter systems [74]. Furthermore, several studies have suggested that GNB3 has notable effects on the efficacy of antidepressants [75–77]. In patients with BD or MDD, SNP rs5443 in GNB3 is associated with a response to selective serotonin reuptake inhibitors [76]. Our study suggests that rare nonsynonymous variants in GNB3 can affect the pathophysiology and development of BD by altering monoaminergic neurotransmission systems.

In the integrated analysis of MRI and SNP data, a significant genotype effect of KMT2C rs56850341 was observed on both CT and multiple DTI parameters. In detail, KMT2C rs56850341 was the only variant that met the Bonferroni threshold for cortical thickness, and this association was confined to a single cortical region, whereas multiple associations with DTI parameters remained significant after Bonferroni correction. In HCs, rs56850341 exhibited a minor allele frequency of approximately 0.47 with no missing genotypes; however, the genotype distribution deviated from Hardy–Weinberg equilibrium (exact test P = 2.3 × 10⁻⁵), with no AA homozygotes observed. This deviation may reflect sampling variability associated with a relatively small sample size, and the observed genotype–imaging associations should therefore be interpreted cautiously. Despite the need for cautious interpretation, our findings may suggest potential biological relevance. Among patients with BD, homozygosity for the KMT2C rs56850341 risk allele (GG) was associated with a brain-wide compromise of white matter integrity, reflected by concordant increases in AD, MD, and RD across commissural and long association pathways (e.g., forceps major/minor, ILF, SLF, ATR, uncinate), with no genotype-by-diagnosis interaction. This triple elevation is less compatible with selective axonal injury and more consistent with a generalized loosening of microstructural constraints—myelin thinning/demyelination contributing to RD (and MD) elevations–together with expansion of the extracellular/free-water compartment and reduced axonal/cellular packing that increase isotropic diffusion and can even produce apparent AD increases in crossing-fiber regions under modest b-values [78–80]. Converging with these distributed effects, the same genotype was linked to reduced cortical thickness in the right lingual gyrus, a hub within the posterior visual–association network interfacing with limbic and prefrontal systems. While the overall white matter pattern is widespread, the prominence of posterior callosal and occipitotemporal pathways (notably forceps major and ILF) provides an anatomical bridge to the lingual finding, suggesting a genetically modulated vulnerability of visual–association circuitry [26, 30, 81, 82]. Given that KMT2C encodes an enhancer-directed H3K4 methyltransferase involved in neurodevelopment and circuit maturation, these co-occurring cortical and white-matter alterations are consistent with a chromatin-regulated, endophenotypic structural liability relevant to BD pathophysiology [83].

This finding is similar to our previous WES–neuroimaging study, in which KMT2C rs4639425 showed a primary effect on DTI parameters without a genotype-by-diagnosis interaction, reflecting a similar direction of effect [36]. Although different SNPs were found in KMT2C due to differences in genome assembly, G allele homozygotes of KMT2C rs56850341 exhibited higher MD and RD values across several white matter tracts than A allele heterozygotes. Given these consistent findings across different SNPs within KMT2C, it is important to consider the broader role of this gene in neuropsychiatric disorders.

KMT2C, which modulates H3K4 methylation and chromatin remodeling, has been implicated in various neuropsychiatric disorders [83]. A WES study showed postzygotic de novo mutations of KMT2C were enriched among patients with BD [84]. Mosaic loss‑of‑function mutations of KMT2C were additionally documented in BD [85]. Furthermore, the KMT2C rs199796552 was found to co-segregate with schizophrenia in a multiplex family, implicating KMT2C in the genetic architecture of schizophrenia [86]. It has also been suggested that KMT2C-related neurodevelopmental disorders are associated with multiple, overlapping neurodevelopmental and psychiatric phenotypes [24]. Therefore, our findings suggest that SNP rs56850341 in KMT2C may be associated with the microstructure of CT and white matter tracts, potentially contributing to neurobiological variability observed in BD. However, these findings highlight the need for further research to explore the underlying mechanisms.

SYNE1, PKHD1L1, FRAS1, and ZFHX4 exhibit a high frequency of mutations (> 20%) in patients with BD. In this study, SYNE1 exhibited the highest number of rare variants, consistent with findings from several other WGS and WES studies in patients with BD [87, 88]. PKHD1L1 demonstrated numerous rare variants (mutation frequency > 20%) in our BD cohort. As a recently discovered gene, the subcellular localization and biological function of the PKHD1L1 gene product remain unknown [89, 90]. Several studies have investigated rare variants of PKHD1L1 in psychiatric disorders. A family-based WES study revealed that the rare segregating variant p.N4041Y in PKHD1L1 is associated with increased risk of BD in 27 families [87]. Another WES study suggested that the rare variant rs76957928 in PKHD1L1 is a potential risk factor for inheritance in schizophrenia families [91]. FRAS1 plays a crucial role in structural cohesion between the epithelium and mesenchyme during embryonic development, and the loss-of-function mutation c.9739delA in this gene can lead to an autosomal recessive disorder [92, 93]. In our previous study, the association between FRAS1 and BD was supported through a meta-analysis of a European case–control study of recurrent rare coding variants observed in patients with BD [15, 36]. ZFHX4 is prominently expressed and regulated during early muscle and brain development [94]. A WGS study revealed that a loss-of-function mutation (c.3646-1 G > A) in ZFHX4 resulted in developmental delays, including impaired motor and language skills [95]. Additionally, ZFHX4 in the 8q21.11 deletion region, has been associated with intellectual disability and specific facial features [96]. Although further studies are warranted, we suggest that the pathophysiology of BD may be explained by the substantial contribution of rare variants in these genes.

In patients with BD, we observed recurrent gain and deletion of 5q13.2, the BD-related gene, in the BDP1 locus. A recent study suggested that BDP1 upregulation is associated with post-traumatic stress disorder and comorbid excessive daytime sleepiness [97]. The gain and deletion of 5q13.2 were associated with individuals with alcohol dependence, which commonly shows comorbidity with BD [98–100]. In our study, the most recurrent CNAs were a gain at 15q11.1 and loss at 14q32.33. Several SNP array studies have implicated microduplications at 15q11-q13 in neurodevelopmental disorders and loss at 14q32.33 in schizophrenia [101, 102]. Furthermore, a comparative genomic hybridization study reported patients with schizophrenia with 15q11.1- q11.2 duplication in Asians, Caucasians, and Blacks [103].

This study had several limitations. First, the relatively small sample size may have limited the statistical power to detect the effects of heritable genetic variants on the structural neuroimaging markers. Substantial associations between rs56850341 and CT or white matter tract integrity were not confirmed in independent samples. Therefore, replication studies with larger sample sizes are required to validate our findings. Second, there was a lack of consideration of whether BD types I and II were divided or whether there was a previous history of psychotic symptoms. This may have been responsible for the increased heterogeneity of our samples, which may have affected our results. Third, only coding variants in patients with BD were examined in the current study. Previous studies have suggested that variants in non-coding regions may be involved in the pathophysiology of BD [104, 105]. Despite these limitations, to the best of our knowledge, this case-control study represents the largest BD sample size using WES to investigate the effects of rare genetic variants on the brain structural changes in Asians.

In conclusion, using WES, this study identified frequently mutated genes, including SYNE1, PKHD1L1, FRAS1, and ZFHX4, in 93 Korean patients with BD. Gene-level burden testing has identified GNB3 as a potentially damaging risk factor in patients with BD. For the neuroimaging–genetic association analysis, our study demonstrated that alterations in white matter tract integrity, along with changes in CT of the right lingual gyrus, may be associated with genetic variation in KMT2C. To the best of our knowledge, this is the first study to reveal an association between gene-based effects of rare variants and neuroimaging markers using WES in patients with BD. This study provides comprehensive information on the genetic effects of rare variants and the influence of these genetic factors on structural brain changes in BD.

Author contributions

S-WP and K-MH wrote the manuscript as co-first authors. S-WP and K-MH managed the literature search. S-WP, E-YO, K-MH, and M-RH statistically analyzed the data. K-MH, W-ST, AK, YK, JP, DS, and B-JH contributed to the analysis and interpretation of the MRI data. B-JH and M-RH conceived and designed the study and wrote the protocol as the co-corresponding authors. All authors contributed significantly to the study and approved the final manuscript.

Funding

This work was supported by the National Research Foundation of Korea (NRF) grants funded by the Korean government (MSIT) (RS-2025-02217919, RS-2025-00523110, PI: K.M.H.; RS-2025-24873317, PI: M.R.H.; NRF-2022R1A2C2093009, PI: B.J.H.).

Data availability

The whole-exome sequencing data and structural neuroimaging data generated and analyzed in the present study contain potentially identifiable human genetic and clinical information. According to the conditions of informed consent obtained from participants and the data-protection requirements of the Institutional Review Board of Korea University Anam Hospital, these individual-level data cannot be deposited in a fully open public repository. The processed datasets supporting the findings of this study are available from the corresponding authors (B.-J.H. and M.-R.H.) upon reasonable request and with appropriate data-use and ethical agreements. Reference databases used for variant filtering (Genome Aggregation Database, 1000 Genomes Project, Korean Variant Archive) are publicly available and cited in the Materials and Methods section.

Competing interests

The authors declare no competing interests.

Ethical standards

The authors assert that all procedures contributing to this study complied with the ethical standards of the relevant national and institutional committees on human experimentation and the 1975 Declaration of Helsinki, as revised in 2008. This study was approved by the Institutional Review Board of the Korea University Anam Hospital (protocol code: 2017AN0185, date of approval: June 5, 2017).

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41398-026-04291-0.