Introduction

The corpus callosum (CC) is the largest white matter tract in the human brain, facilitating higher order functions of the cerebral cortex by allowing the two hemispheres of the brain to communicate1,2. This connection is essential for coordinating sensorimotor responses, performing associative and executive functions, and representing information in multiple dimensions3,4. Most CC fibers connect corresponding left and right cortical regions of the brain, with the organization, development of axonal elongation, and myelination of callosal fibers being correlated with the rostro-caudal (front-to-back) distribution of functional areas5,6. Regional alterations in CC shape are easily assessed with neuroimaging studies, which have found local callosal abnormalities in complex neurodevelopmental and neuropsychiatric disorders6,7,8,9,10,11, such as, on average, lower anterior volumes in people with autism spectral disorder12 and lower posterior thickness in individuals with bipolar disorder13. Twin studies show up to 66% heritability for CC area14,15, and previous single-cohort studies of genetic influences on CC volume and its relationship to neuropsychiatric disorders have found heritability estimates between 22–39%16,17. Yet, the interplay between genetic variants influencing CC morphometry, the cerebral cortex, and associated neuropsychiatric disorders is not well understood.

Three-dimensional (3D) magnetic resonance imaging (MRI) provides a non-invasive approach to quantify individual variations in brain regions and their connections6, including the morphometry of the CC, and how they are associated with brain-based traits and diseases. The midsagittal section of an anatomical brain MRI scan is able to capture the entire rostro-caudal formation of the CC, which is almost always in the field of view of 2D clinical and 3D research MRI scans alike. This 2D midsagittal representation can be segmented to offer a lower dimensional projection of the anatomical intricacies of the CC, allowing for structural measures of CC area and thickness to be computed18,19,20. We developed and validated a fully automated artificial intelligence based CC feature extraction tool, Segment, Measure, and AutoQC the midsagittal CC (SMACC), which we make publicly available at smacc20.

Using data from the UK Biobank21 (UKB) and Adolescent Brain Cognitive Development22 (ABCD) studies, here we present results from a genome-wide association study (GWAS) meta-analysis of total area and mean thickness of the CC derived using SMACC. We also present the results for five differentiated areas based on distinguishable projections to (1) prefrontal, premotor, and supplementary motor, (2) motor, (3) somatosensory, (4) posterior parietal and superior temporal, and (5) inferior temporal and occipital cortical brain regions23,24. These regions are believed to represent structural-functional coherence6. We performed a GWAS meta-analysis using two population-based cohorts, one of adolescents and another of older adults, to examine distinct genetic influences on CC area and thickness25,26. The principal analyses were in individuals of European ancestry and the same analyses were then repeated using the data from non-European participants to assess consistency in the magnitude and direction of effect sizes. Downstream post-GWAS analyses investigated the enrichment of genetic association signals in tissue types, cell types, brain regions, and biological pathways. We examined the genetic overlap at the global and local level, using LD Score regression (LDSC)27 and Local Analysis of Variant Association (LAVA)28, respectively, and the causal genetic relationships between CC phenotypes, cortical morphometry, and related neuropsychiatric conditions.

Results

Characterization of corpus callosum shape associated loci

We conducted a GWAS of area and mean thickness of the whole CC, and five regions of the Witelson parcellation scheme (Fig. 1)23,24, using data from participants of European ancestry from the UKB (N = 41,979) and ABCD cohorts (N = 4706). A meta-analysis of GWAS summary statistics of all CC derived metrics in UKB and ABCD was performed using METAL and the random-metal extension29,[30](https://www.nature.com/articles/s41467-025-64791-3#ref-CR30 “Hemani, G. Explodecomputer/random-Metal: Adding Random Effects Model. (Zenodo, 2022). https://doi.org/10.5281/ZENODO.6974695

“), based on the DerSimonian-Laird random-effects model (Methods). To examine the generalizability of single-nucleotide polymorphism (SNP) effects across ancestries, these same analyses were run using data from non-European participants (total N = 7040).

Fig. 1: Regions of the midsagittal corpus callosum and associated genomic loci.

An ideogram representing loci that influence total CC area, its mean thickness, and area and thickness of individual parcellations determined by the Witelson parcellation scheme in a rostral-caudal gradient (1–5). Results shown are from an inverse-weighted random-effects meta-analysis (DerSimonian-Laird method). Reported p-values are two-sided. All loci are significant at the Bonferroni corrected, experiment-wide threshold of p < 6.13 × 10−9. Created in part by using Biorender.com (agreement number UX28RS3P2L).

The GWAS meta-analysis identified 48 independent significant SNPs for total area and 18 independent SNPs for total mean thickness. Independent significant SNPs were determined in FUMA using the default threshold of r**2 = 0.6, and genomic loci were determined at r**2 = 0.1. This identified 28 genomic loci for total cross-sectional area, and 11 genomic loci for total mean thickness. All significant loci for total area and mean thickness showed concordance in the direction of effect between the two cohorts. There were 5 loci, all in intronic regions, each positionally mapped to genes31 that overlapped between area and mean thickness. These included IQCJ-SHIP1 (multimolecular complexes of initial axon segments and nodes of Ranvier, and calcium mediated responses)32, FIP1L1 (RNA binding and protein kinase activity)33, HBEGF (growth factor activity and epidermal growth factor receptor binding)34, CDKN2B-AS1 (involved in the NF-κB signaling pathway with diverse roles in the nervous system)35,36, and FAM107B (cytoskeletal reorganization in neural cells and cell migration/expansion)37. The genomic locus mapped to IQCJ-SHIP1 had a positive effect for total area (rs11717303, effect allele: C, effect allele frequency (EAF): 0.689, β = 4.28, s.e. = 0.51, p = 4.54 × 10−17). The same locus showed a negative effect for a different SNP on total thickness (rs12632564, effect allele: T, EAF: 0.305, β = −0.042, s.e. = 0.006, p = 2.59 × 10−12). The strongest locus for total area (rs7561572, effect allele: A, EAF: 0.532, β = −4.13, s.e. = 0.46, p = 1.98 × 10−18) was positionally mapped to the STRN gene. The strongest locus for mean thickness (rs4150211, effect allele: A, EAF: 0.265, β = −0.05, s.e. = 0.006, p = 8.20 × 10−18) was mapped to the HBEGF gene.

Loci for area overlapped between parcellations in a rostral-caudal gradient (1–5), such that: rs1122688 on the SHTN1 (or KIAA1598) gene (involved in positive regulation of neuron migration) overlapped between the genu (1) and anterior body (2); rs1268163 near the FOXO3 gene (involved in IL-9 signaling and FOXO-mediated transcription) overlapped between the posterior body (3) and isthmus (4); and rs11717303 on the IQCJ-SCHIP1 gene overlapped between the isthmus (4) and splenium (5). This gradient pattern was not observed for mean thickness. The strongest regional association was observed with splenium area (rs10901814, effect allele: C, EAF: 0.584, β = −1.69, s.e. = 0.16 p = 2.02 × 10−24) and thickness (rs11245344, effect allele: T, EAF: 0.570, β = −0.11, s.e. = 0.11, p = 6.28 × 10−22), both on the FAM53B gene. FAM53B is involved in the positive regulation of the canonical Wnt signaling pathway. We observed overlaps in the direction and magnitude of effects between the main European analyses with the results from the non-European participants. 124 (82%) out of the 152 significant loci identified across CC phenotypes in this study had effect sizes in European participants falling within the 95% confidence interval of those seen in the non-European participants. Furthermore, 78 loci demonstrated a consistent direction of effect across all four cohorts (2 European from UKB and ABCD, and 2 non-European from UK Biobank and ABCD, respectively). Detailed annotations and regional association plots of all genomic loci, independent significant SNPs and genes are in Supplementary Data 1–4 and Supplementary Data 42.

In order to test for the influence of total intracranial volume (ICV) on the GWAS, another set of GWASes were run controlling for ICV (Methods, Supplementary Data 37). The overlap coefficients and genetic correlations, respectively, were 0.64 and 0.75 (s.e.= 0.03) for total area and 1 and 1.02 (s.e. = 0.008) for total thickness, indicating a high degree of overlap for both analyses. Near perfect genetic correlations were observed comparing across all CC traits, except splenium area (r**g = 0.08, s.e. = 0.05), which may be driven by collider bias38,39 (Supplementary Data 38). Overall, the strongest enrichment signals were observed among genes that were significant in both the ICV-adjusted and non-ICV GWAS, as well as genes uniquely significant in the non-ICV GWAS - compared to the GWAS that included ICV as a covariate alone. Overall, greater enrichment was observed with genes common between ICV and specific to no ICV, compared to the GWAS which included ICV as a covariate. Genes mapped to significant loci common to both GWAS sets, were consistently mapped to canonical signaling pathways, including PI3K/AKT, PDGFR, and estrogen signaling, suggesting ICV-insensitive mechanisms involved in CC development and maintenance. Genes unique to the no ICV GWAS showed enrichment for mitochondrial respiration, oxidative stress response, and integrin signaling. Genes specific to the ICV-controlled GWAS exhibited much lower enrichment. Regionally specific enrichments were observed for area traits, including WDR5-mediated epigenetic regulation in the genu and apoptosis, as implicated by previous analyses, in the isthmus (Supplementary Data 39).

SNP heritability and genetic correlation between cohorts

Moderate to high genetic correlations were seen across CC phenotypes between cohorts using LDSC, with r**g ranging from 0.54 (s.e. = 0.27) and 0.92 (s.e. = 0.63) for area metrics, and 0.30 (s.e. = 0.16) and 0.99 (s.e. = 0.69) for thickness metrics. To complement the LDSC approach with an approach using individual level data, we used the bivariate GREML in GCTA40. Moderate genetic correlations between cohorts were seen using bivariate GCTA with r**g ranging between 0.40 (s.e. = 0.04) and 0.49 (s.e. = 0.03) across all traits. Age-related variability in white matter likely contributes to some of the lower correlation agreements between cohorts, as white matter volume tends to increase through childhood and adolescence, peak in early adulthood, and then gradually decline from middle age onward41. For instance, certain genetic variants might exert a stronger influence on CC structure during periods of white matter growth (as in the younger ABCD cohort) compared to periods of white matter decline (as in the older UKB cohort). The smaller sample size of the ABCD cohort may limit LDSC’s ability to detect polygenic effects, capturing primarily the strongest genetic signals[42](https://www.nature.com/articles/s41467-025-64791-3#ref-CR42 “Rkwalters & Palmer, D. Nealelab/UKBB_ldsc: v2.0.0 (Round 2 GWAS Update). https://doi.org/10.5281/zenodo.7186871

(2022).“). However, strong cross-cohort correlations for total area and isthmus thickness phenotypes suggest that genetic variants affecting these traits are likely consistent across developmental stages43,44 to estimate SNP heritability (h2SNP) and generic correlations between each cohort. Within the UKB, heritability values ranged for different CC phenotypes from 0.42 to 0.71, with similar results seen in the ABCD cohort (Supplementary Data 5–8). Total area (UKB h2SNP = 0.72, s.e. = 0.01; ABCD h2SNP = 0.74, s.e. = 0.03) and mean thickness (UKB h2SNP = 0.61, s.e. = 0.02; ABCD h2SNP = 0.78, s.e. = 0.02) showed the highest h2SNP across both cohorts. LDSC27 h2SNP estimates from the meta-analysis ranged between 0.10 (s.e. = 0.01) and 0.18 (s.e. = 0.05) for area, and 0.12 (s.e. = 0.01) and 0.16 (s.e. = 0.02) for thickness, with the area of the genu showing the highest, and area of the splenium showing the lowest h2SNP estimates. As shown in Supplementary Data 5–8, all LDSC and GCTA rG estimates between meta-analyzed CC phenotypes were significant.

Gene-mapping and gene-set enrichment analyses

Gene-based association analysis in MAGMA45 identified 30 genes for the total area, and 34 genes for total mean thickness of the CC, with 5 genes overlapping between area and thickness (IQCJ-SCHIP1, IQCJ, BPTF, PADI2, CHIC2). The strongest association seen with area was AC007382.1 and the strongest association with mean thickness was HBEGF (Fig. 2a). There were between 15 and 31 genes for area, and between 7 and 25 genes for thickness identified within regions of the CC. Notably, IQCJ, IQCJ-SCHIP1, and STRN overlapped for all parcellations of CC area. AC007382.1 overlapped for four out of five parcellations, and STRN and PARP10 overlapped for three out of five parcellations of CC thickness (Fig. 2b, Supplementary Data 1–4). Enrichment of SNP heritability in 53 functional categories for each trait was determined via LDSC46. The majority of enrichment and the strongest effects across parcellations of the CC were observed in categories related to gene regulation/transcription in chromatin (Fig. 3a, b).

Fig. 2: GWAS meta-analysis of midsagittal corpus callosum area and thickness.

a Miami plot for SNPs (top) and genes (bottom) based on MAGMA gene analysis for total area and total mean thickness. b Miami plot for SNPs (top) and genes (bottom) based on MAGMA gene analysis for area of thickness of the CC split by the Witelson parcellation scheme23. Results shown on the upper panels of (a) and (b) are from an inverse-weighted random-effects meta-analysis (DerSimonian-Laird method). Reported -log10(p-values) are two-sided. All loci are significant at the Bonferroni corrected, experiment-wide threshold of p < 6.13 × 10−9. Results shown on the lower panels of (a) and (b) are from the MAGMA gene-based analysis. Reported -log10(p-values) are two sided from the Z-statistic. All significant genes are shown at the Bonferroni corrected threshold of p < 2.74 × 10−6. Significant SNPs and genes are color-coded by CC traits. Created in part by using Biorender.com (agreement number PT28RS3SIJ).

Fig. 3: Partitioned heritability, functional annotation and enrichment of gene-sets of CC morphology associated genetic variants.

a Significant enrichment of SNP heritability across 53 functional categories computed by LD Score regression for area (left) and mean thickness (right). Analyses were completed using the meta-analyzed GWAS summary statistics (N = 46,485). Data are presented as mean values +/− s.e. b Proportion of GWAS SNPs in each functional category from ANNOVAR across each CC phenotype. c Significant gene-sets across CC phenotypes computed via MAGMA gene-set analysis using the equivalent of a one-sided two-sample t-test at the Bonferroni corrected threshold of 3.23 × 10−6. GOBP Gene-ontology biological processes, GOCC Gene-Ontology Cellular Components.

Gene-set enrichment analyses were also completed in MAGMA (Fig. 3c). The strongest effects of significant gene sets included those involved in postsynaptic specialization for total CC area, including GO:009901 (postsynaptic specialization, intracellular component) and GO:009902 (postsynaptic density, intracellular component). A theme of signal transduction-related pathways was observed for the splenium area, including R-HSA-6785631 (ERBB2 regulates cell motility) and R-HSA-8857538 (PTK6 promotes HIF1A stabilization). Enrichment of the “CARM1 and regulation of the estrogen receptor” was found for the posterior body thickness and is implicated transcriptional regulation via histone modifications. Enrichment of GO:1904714 (regulation of chaperone-mediated autophagy) was found for the isthmus area, which is implicated in lysosomal-mediated protein degradation. All significant results across all CC phenotypes are in Supplementary Data 18.

Tissue-specific and cell-type-specific expression of corpus callosum associated genes

Gene-property enrichment analyses were completed in MAGMA with 54 tissue types from GTEx v8 and BrainSpan47,48, which includes 29 samples from individuals representing 29 different ages, as well as 11 general developmental stages. An enrichment of genes associated with isthmus thickness were expressed in the cerebellum (p*(Bon)* = 0.017). Area and thickness across parcellations of the CC showed an enrichment of expression of genes in the brain from early prenatal to late mid-prenatal developmental stages. An enrichment of expression of genes associated with area and thickness of the anterior body of the CC was observed in brain tissue prenatally, 9–24 weeks post conception. Enrichment of expression of genes associated with area of the genu was observed in brain tissue 19 weeks post conception. Enrichment of expression of genes associated with the total mean thickness of the CC was observed in brain tissue 19 weeks post conception. All results are shown in Supplementary Data 19–21. These results, along with the gene-sets involved in histone modifications, were supported by LDSC-SEG analyses using chromatin-based annotations from narrow peaks49, which showed a significant enrichment in the heritability by variants located in genes specifically expressed in DNase in the female fetal brain for total CC thickness (p*(Bon)* = 0.0105). Chromatin annotations showed a consistent and significant enrichment of splenium area and thickness-associated variants in histone marks of the fetal brain and neurospheres (Supplementary Data 25).

Using microarray data from 292 immune cell types, the area of the posterior body showed a significant enrichment in the heritability by variants located in genes specifically expressed in multiple types of myeloid cells (p*(Bon)* < 0.05), and the area of the isthmus showed enrichment in innate lymphocytes (p*(Bon)* = 0.047). This further validates the aforementioned significant locus on gene FOXO3, which overlapped between the posterior body and isthmus (Supplementary Data 26).

Cell-type-specific analyses were performed in FUMA using data from 13 single-cell RNA sequencing datasets from the human brain. This tests the relationship between cell-specific gene expression profiles and phenotype-gene associations50. Of the 12 phenotypes tested, only total CC thickness showed significant results after going through the 3-step process using conditional analyses to avoid bias from batch effects from multiple scRNA-seq datasets. The most significant association was seen with oligodendrocytes located in the middle temporal gyrus (MTG, p*(Bon)* = 0.001) from the Allen Human Brain Atlas (AHBA). Oligodendrocytes (p*(Bon)* = 0.03) and non-neuronal cells (p*(Bon)* = 0.03) located in the lateral geniculate nucleus (LGN) from the AHBA also showed significant associations but were collinear (Supplementary Data 22).

LAVA-TWAS analyses28,51 (Fig. 4) of expression quantitative trait loci (eQTLs) and splicing quantitative trait loci (sQTLs) of protein-coding genes in 16 different brain, cell type, and whole blood tissues revealed the strongest eQTL associations of area and thickness with CROCC expression in whole blood for the isthmus (ρ = −0.53, p = 1.29 × 10−10). Other notable eQTL (Supplementary Data 29) findings included total CC area and isthmus area and thickness being positively associated with ATP13A2 expression in fibroblasts (ρ = 0.48, p = 1.58 × 10−7). The strongest sQTL association was a positive association observed with KANSL1 (cluster 11710) in fibroblasts for genu area (ρ = 0.83, p = 1.46 × 10−14), which was the tissue type where most observed associations occurred across CC phenotypes (Supplementary Data 30). Moreover, a negative association was observed in a KANSL1 (cluster 11707) in fibroblasts for the genu area (ρ = 0.82, p = 3.11 × 10−7). An sQTL in MFSD13A (cluster 7894) in the anterior cingulate showed very strong yet opposite associations for total CC thickness (ρ = 0.42, p = 1.12 × 10−13) and total CC area (ρ = −0.44, p = 2.98 × 10−11). Other notable findings across tissue types included CRHR1 in the cortex, nucleus accumbens, and putamen, as well as UGP2 in fibroblasts, whole blood, and the putamen. No significant results from LAVA-TWAS gene-set enrichment analyses were observed after Bonferroni correction (Supplementary Data 31–32).

Fig. 4: LAVA-TWAS analyses of corpus callosum traits with gene-expression (eQTLs) and splicing (sQTLs).

Results of local genetic correlations between CC traits and eQTLs and sQTLs from GTEx v8 using the LAVA-TWAS framework. All significant points are colored by tissue type and labeled by CC trait. Significance was tested as a two-sided t-test statistic. Significance thresholds for eQTLs (p < 2.01 × 10−6) and sQTLs (p < 5.45 × 10−7) were determined by Bonferroni correction. Associations between (a) CC area and eQTLs, (b) CC thickness and eQTLs, (c) CC area and sQTLs, and (d) CC thickness and sQTLs are shown via -log10p values scaled by the direction of association (y-axis) and chromosomal location (x-axis).

Genetic overlap of corpus callosum and cerebral cortex architecture

Broadly, we observed a pattern of negative genetic correlations with area and thickness of the CC with cortical thickness across regions of the cingulate cortex, but positive genetic correlations with regions’ cortical thickness across the neocortex (Fig. 5a). Specifically, we observed a significant negative genetic correlation between total area with cortical thickness of the rostral anterior cingulate (r**g = −0.35, s.e. = 0.06) and posterior cingulate (r**g = −0.28, s.e. = 0.06). Mean thickness had a negative genetic correlation with cortical thickness of the rostral anterior cingulate (r**g = −0.29, s.e. = 0.06) and posterior cingulate (r**g = −0.23, s.e. = 0.05). Positive genetic correlations were observed with cortical thickness of the lingual gyrus (r**g = 0.26, s.e. = 0.05) and cuneus (r**g = 0.27, s.e. = 0.06). When parcellating by the Witelson scheme23, negative genetic correlations were observed for area and mean thickness with cortical thickness of regions across the cortex and the cingulate, but positive genetic correlations with regions in the occipital lobe. We also observed a significant negative genetic correlation between total area of the CC with surface area of the precuneus (r**g = −0.20, s.e. = 0.04). (Supplementary Data 9–10).

Fig. 5: The genetic overlap of the corpus callosum and cerebral cortex.

a Global genetic correlations (LDSC - rG) between CC phenotypes and cerebral cortex phenotypes. Significance was based on a two-sided Z-statistic. The Bonferroni significance threshold was set at p = 6.1 × 10−5. Surface area and cortical thickness of significant cortical regions with each CC phenotype are displayed on brain plots. b Of the significant global genetic correlations, significant Mendelian randomization (GSMR) results are displayed, representing the effect of CC phenotypes on cortical phenotypes free of non-genetic confounders. Significance was determined using the two-sided t-statistic calculated within GSMR. The number of SNPs used in GSMR were N = 26, N = 18, and N = 10 for the precuneus, rostral anterior cingulate, and posterior cingulate respectively. Data are presented as beta values +/- s.e. c Chord plot displaying the number of significant bivariate local genetic correlations (LAVA) between CC and cortical phenotypes. Underlined numbers represent the total number of genes shared with that phenotype. d Volcano plots showing degree (-log10 p-values) and direction (rG) of local genetic correlations (LAVA) between cortical and CC phenotypes. Significance was tested as a two-sided t-test statistic. Colors represent cortical regions labeled on the chord plot in section C. Significant genes (Bonferroni significance threshold was set at p = 2.18 × 10−6) across all phenotypes are labeled.

Genetic correlations can reflect direct causation, pleiotropy, or genetic mediation. To explore potential causal relationships between CC phenotypes and morphometry of the cerebral cortex, we ran Generalized Summary-data-based Mendelian Randomization (GSMR) analyses52 directional effect of CC phenotypes on morphometry of the cerebral cortex, but not vice-versa. (Fig. 5b, Supplementary Data 14). There was a strong negative unidirectional effect of total CC area on the precuneus surface area (b**xy = −0.50, s.e. = 0.13, p = 0.0002), implying a greater total area and thickness of the CC results in a lower surface area of the precuneus. There was also a negative unidirectional effect of total CC mean thickness and cortical thickness of the posterior cingulate (b**xy = −0.02, s.e. = 0.008, p = 0.02), but not vice versa. When using the Witelson parcellation scheme, there was a strong negative unidirectional effect on the area of the genu on the cortical thickness of the rostral anterior cingulate (b**xy = −0.001, s.e. = 0.0003, p = 0.003).

Local genetic correlations of area phenotypes of the CC and surface area of the cerebral cortex with LAVA28 showed many significant negative correlations in genes between the total area and posterior body and the precuneus SA along the 2p22.2 cytogenetic band (QPCT, PRKD3, SULT6B1, NDUFAF7, EIF2AK2, HEATR5B, GPATCH11, CEBPZ, CEBPZOS, CDC42EP3, STRN, VIT) (Fig. 5c, d). Negative genetic correlations between total CC area and caudal middle frontal gyrus SA in 5 genes along the 17q24.2 cytogenetic band (HELZ, PSMD12, PITPNC1, ARSG, BPTF) were also observed. Positive local genetic correlations along the 2p22.2 cytogenetic band were observed with the anterior body area and the surface area of the posterior cingulate (CDC42EP3, PRKD3), as well as the total area of the CC and precentral gyrus surface area (HEATR5B).

Many negative local genetic correlations were observed with mean thickness of the splenium and cortical thickness of the superior parietal gyrus (TEX36, EDRF1, UROS, BCCIP, DHX32) and the parahippocampal gyrus (ZNF879) along the 10q26.13–10q26.2 cytogenetic bands, while positive genetic correlations were observed with isthmus cingulate cortical thickness along the 10q26.13–10q26.2 cytogenetic bands (EDRF1, TEX36, UROS, BCCIP, DHX32, CTBP2, CPXM2, GPR26, ZRANB1, FAM53B).

The area of the posterior body showed a negative local genetic correlation with the cortical thickness of the pericalcarine gyrus (GPATCH11). The area of the isthmus showed positive local genetic correlations with the cortical thickness of the superior parietal gyrus (LRRC73), caudal middle frontal gyrus (GPATCH2L), and isthmus cingulate (PLPPR3, CFD, R3HDM4, PTBP1, ELANE, MED16, PALM) along the 19p13.3 cytogenetic band.

The mean thickness of the posterior body showed negative local genetic correlations with the surface area of the lingual gyrus (STC2, NKX2-5, 5q35.2) and pericalcarine gyrus (NKX2-5). Mean thickness of the isthmus showed negative local genetic correlations with the precuneus (EIF2AK2, GPATCH11, 2p22.2) and superior frontal gyrus (TBX19) surface area. Total mean thickness of the CC showed a positive genetic correlation with the surface area of the insula (PDZRN3). The mean thickness of the anterior body showed positive local genetic correlations with surface area of the superior parietal gyrus (RETN, FCER2). Splenium mean thickness showed positive genetic correlations with inferior temporal gyrus surface area (ZNF318, CRIP3, SLC22A7) along the 6p21.1 cytogenetic band.

Genetic overlap of corpus callosum and associated neuropsychiatric phenotypes

We observed a significant genetic correlation (Fig. 6a, Supplementary Data 11) between total CC area and ADHD (r**g = −0.11, s.e. = 0.03), bipolar disorder (BD, r**g = −0.10, s.e. = 0.03), and bipolar I disorder (BD-I, r**g = −0.10, s.e. = 0.03). Total mean thickness was genetically correlated with BD (r**g = −0.10, s.e. = 0.03) and BD-I (r**g = −0.10, s.e. = 0.03). When analyzing the regional Witelson parcellations23, the area of the genu was genetically correlated with ADHD risk (r**g = −0.13, s.e. = 0.03), and the mean thickness of the splenium was genetically correlated with risk for BD (r**g = −0.13, s.e. = 0.03) and BD-I (r**g = −0.12, s.e. = 0.03).

Fig. 6: The genetic overlap of the corpus callosum and neuropsychiatric phenotypes.

a Global genetic correlations between CC traits and neuropsychiatric phenotypes. Significance was based on a two-sided Z-statistic. Significant results are designated by the * at the Bonferroni significance threshold of p = 0.0015. Significant negative genetic correlations are observed between total and splenium thickness, and bipolar disorder (I). Significant negative genetic correlations are also observed with CC area phenotypes and ADHD. Of the significant global genetic correlations, significant Mendelian randomization (GSMR) results are displayed, representing the effect of CC phenotypes on neuropsychiatric phenotypes free of non-genetic confounders. The number of SNPs used in the GSMR analysis were N = 29 for BD on total mean thickness, N = 26 for BD-I on total mean thickness, N = 11 for total mean thickness on BD, N = 25 on BD-I on splenium mean thickness, and N = 11 on total mean thickness on BD-I. Data are presented as beta values +/− s.e. b Volcano plots showing degree (-log10 p-values) and direction (r**G) of local genetic correlations (LAVA) between neuropsychiatric and CC phenotypes. Significant local negative genetic correlations on the STH, KANSL1, SPPL2C, CRHR1, and MAPT genes are observed between the genu area and neuroticism. Significant local positive genetic correlations on the MAPT, SPPL2C, STH, CRHR1, ARHGAP27, KANSL1, PLEKHM1, MAP3K14, and DCAKD genes are observed between the genu area and Parkinson’s disease. A significant positive local genetic correlation is observed on the CEP170 gene between total mean thickness and Parkinson’s disease. Significance was tested as a two-sided t-test statistic. Phenotypes with significant associations are colored (IQ and bipolar II disorder). Significant genes (Bonferroni significance threshold was set at p = 2.79 × 10−6) across all neuropsychiatric phenotypes are shown. AD Alzheimer’s disease, ADHD attention deficit hyperactivity disorder, ASD autism spectrum disorder, BD bipolar disorder, BD-I bipolar I disorder, BD-II bipolar II disorder, COPC chronic overlapping pain conditions, IQ intelligence quotient, OCD obsessive-compulsive disorder, PD Parkinson’s disease, PTSD post-traumatic stress disorder, SCZ schizophrenia.

GSMR analyses showed causal bidirectionality of genetic liability of BD (b**xy = −0.06, s.e. = 0.02, p = 0.006) and BD-I (b**xy = −0.05, s.e. = 0.02, p = 0.003) on total mean thickness of the CC, and mean thickness of the CC on BD (b**xy = −0.19, s.e. = 0.08, p = 0.01) and BD-I (b**xy = −0.23, s.e. = 0.09, p = 0.02). When using the Witelson parcellation23, GSMR analyses showed causal directionality of genetic liability of BD-I on mean thickness of the splenium (b**xy = −0.09, s.e. = 0.04, p = 0.01), but not vice versa (Fig. 6a, Supplementary Data 15).

Local genetic correlations with LAVA[28](https://www.nature.com/articles/s41467-025-64791-3#ref-CR28 “Werme, J., van der Sluis, S., Po