Deep-learning segmentation and multi-ancestry GWAS enhance genetic discovery of the cerebellum

Deep-learning segmentation and multi-ancestry GWAS enhance genetic discovery of the cerebellum | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Deep-learning segmentation and multi-ancestry GWAS enhance genetic discovery of the cerebellum Chunshui Yu, Hui Xue, Jilian Fu, Hao Ding, Sijia Wang, Jingliang Cheng, and 44 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6694135/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted You are reading this latest preprint version Abstract The genetic architecture of cerebellar substructure volumes is crucial for understanding their functions in cognition, emotion, and neuropsychiatric disorders. We utilized deep-learning segmentation and multi-ancestry analysis to identify genetic associations with 31 cerebellar volumetric traits in 57,071 participants. We found deep learning superior to atlas-based segmentation in heritability estimation, genetic discovery, and polygenic prediction. We identified 407 new loci for these traits (241 from univariate, four from sex-stratified, and 162 from multivariate analyses), along with 19 ancestry-specific and eight sex-specific associations. We prioritized 453 causal variants and 71 relevant genes, categorizing these cerebellar substructures into nine genetically informed clusters. We linked cerebellar volumes to 20 cognitive and mental phenotypes and seven brain disorders. These findings provide an overview of the genetic architecture of cerebellar volumes. Biological sciences/Genetics/Genetic association study/Genome-wide association studies Health sciences/Anatomy Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Main The cerebellum plays an essential role in our daily lives, housing over half of the brain's neurons while occupying just 10% of its total volume 1 . The human cerebellum has historically been attributed exclusively to the planning and execution of movements 2 . However, recent evidence has highlighted its relevance in cognition and emotion 3 – 5 . Although the cerebellum is one of the first brain structures to differentiate, it is among the last structures to achieve maturity. The prolonged developmental timeline makes the cerebellum especially vulnerability to neurodevelopmental disorders such as autism spectrum disorder (ASD) 6 , attention-deficit hyperactivity disorder (ADHD) 7 , and schizophrenia (SCZ) 8 . The human cerebellum is not a homogeneous structure consisting of substructures that exhibit distinct cytoarchitecture and functions. The volumes of the cerebellum and its substructures can be measured using structural magnetic resonance imaging (MRI). For instance, the spatially unbiased infra-tentorial (SUIT) template 9 is a widely used atlas-based method for segmenting cerebellar substructures, while FreeSurfer (FS) 10 can estimate the volumes of left and right white matter and cortex of the cerebellum. In recent decades, notable volumetric changes in the cerebellum and its substructures have been identified in various neuropsychiatric disorders (NPDs), including posttraumatic stress disorder (PTSD) 11 , depression 12 , ADHD 13 , Parkinson's disease (PD) 14 , Alzheimer’s disease (AD) 15 , and amyotrophic lateral sclerosis (ALS) 16 . As these cerebellar volumetric traits demonstrate high heritability 17 – 21 , several genome-wide association studies (GWASs) have explored the genetic architecture of total cerebellar volume 18 , 19 and the volumes of cerebellar substructures 20 , 21 . Although these GWASs improve our understanding of the genetic architecture of cerebellar volumetric traits, several limitations persist in these studies 18 – 21 . First, most previous studies 18 – 20 have focused on individuals of European ancestry (EUR), leaving other ancestral populations under-represented, although one study did include thousands of individuals of East Asian ancestry (EAS) 21 . Second, the SUIT template 9 was used for segmenting cerebellar substructures in these studie 20 , 21 . The volumes of cerebellar substructures obtained from the atlas-based segmentation may not be as precise as those derived from more advanced methods. For example, CerebNet uses a deep-learning model for cerebellar segmentation 22 , demonstrating superior accuracy, test-retest reliability, and sensitivity compared to the SUIT method. Finally, the GWASs on the 28 SUIT-derived cerebellar substructure volumes in the prior two studies 20 , 21 were performed as a minor component of the GWASs for more than 3,400 brain imaging phenotypes. The post-GWAS analyses overlooked the uniqueness of cerebellar volumetric traits, such as the biological processes involved in regulating these traits and the genetically informed subregions of the cerebellum. The first purpose of this study is to investigate whether the deep learning method for cerebellar segmentation (CerebNet) outperforms the atlas-based method (SUIT + FS) in genetic analyses (heritability estimation, genetic discovery, and polygenic prediction) of cerebellar substructure volumes. The second purpose is to perform multi-ancestry GWASs for 31 cerebellar volumetric traits in 57,071 participants, including 7,083 EAS participants from the Chinese Imaging Genetics (CHIMGEN) study 23 , 40,826 EUR participants from the UK Biobank (UKBB) study 24 , and 3,354 African (AFR) and 5,808 EUR participants from the Adolescent Brain Cognitive Development (ABCD) study 25 . The third purpose is to conduct a series of post-GWAS analyses to identify genetically informed cerebellar subregions, causal variants and enrichment pathways of cerebellar volumetric traits, along with their genetic correlation and colocalization with cerebellar gene expression, cognitive and mental health phenotypes, and NPDs. The study design is shown in Fig. S1 . Results Participants and data preparation We included 57,071 participants with qualified genomic and structural MRI data from three datasets and four subgroups, including UKBB EUR (n = 40,826), CHIMGEN EAS (n = 7,083), ABCD EUR (n = 5,808), and ABCD AFR (n = 3,354) ( table S1 ). The specific procedures for participant selection and quality control are detailed in Fig. S2 . We used CerebNet to segment the cerebellum to obtain the volumes of cerebellar substructures. To assess the reliability of CerebNet segmentation, we calculated intraclass correlation coefficients (ICCs) of cerebellar substructure volumes derived from structural MRI data obtained at two separate time points from the same participants (CHIMGEN: n = 26; UKBB: n = 2,944; and ABCD: n = 4,076). We found high reliability in the three datasets (CHIMGEN: ICC = 0.62–0.97; UKBB: ICC = 0.88–0.96; ABCD: ICC = 0.87–0.98; table S2 ). GWASs were finally conducted on 10,001,636 genetic variants for UKBB EUR , 9,017,308 for CHIMGEN EAS , 9,343,804 for ABCD EUR , and 7,395,374 for ABCD AFR participants, respectively ( Fig. S2, table S3 ). Comparison in genetic findings between cerebellar segmentation methods To explore whether CerebNet outperforms SUIT + FS in genetic analyses of cerebellar substructure volumes, we used the two segmentation methods to calculate the volumes of 30 cerebellar substructures, and then compared their performance in the heritability estimation, genetic discovery, and polygenic score (PGS) prediction. Heritability estimation. We applied GCTA-GREML 26 to estimate single nucleotide polymorphism (SNP)-based heritability (h 2 ) of the 30 cerebellar volumetric traits in the four subgroups, respectively. Although these traits obtained from both CerebNet and SUIT + FS were heritable ( P < 0.05; table S4 ), cerebellar substructure volumes obtained from CerebNet showed higher heritability (Wilcoxon rank-sum test, CHIMGEN EAS : P = 4.53 × 10 − 3 , UKBB EUR : P = 2.42 × 10 − 4 , ABCD EUR : P = 1.01 × 10 − 5 , ABCD AFR : P = 8.08 × 10 − 3 , Fig. 1 A) compared to those derived from SUIT + FS. GWAS discovery. We used the mixed linear model in fastGWA 27 to perform GWASs on 30 cerebellar substructure volumes from CerebNet and SUIT + FS in 40,826 UKBB participants. We used the Wilcoxon rank-sum test to compare the number, P -value, and power of significant variant-trait associations from both methods. We identified 2,159 and 1,169 significant variant-trait associations ( P < 5 × 10 − 8 ) for cerebellar substructure volumes from CerebNet and SUIT + FS ( table S5 ), confirming that CerebNet (median = 64) enhanced GWAS discovery (Wilcoxon rank-sum test, P = 1.11 × 10 − 5 , Fig. 1 B) compared to SUIT + FS (median = 39). In the 346 significant variant-trait associations identified by both methods, CerebNet (median: -log10( P ) = 13.01; power = 0.83) showed increased significance (Wilcoxon rank-sum test, P = 1.04 × 10 − 13 , Fig. 1 C) and power (Wilcoxon rank-sum test, P = 2.91 × 10 − 8 , Fig. 1 D) compared to SUIT + FS (median: -log10( P ) = 9.86; power = 0.61) ( table S5 ). PGS prediction. Using UKBB EUR -GWAS data as the base dataset and CHIMGEN EAS , ABCD EUR , and ABCD AFR raw data as three distinct target datasets, we used PRSice-2 28 to construct the best-fit PGS models for the 30 cerebellar substructure volumes obtained from CerebNet and SUIT + FS and to calculate the explained variance (R 2 ) of PGS for each trait in each target dataset. In ABCD EUR , CerebNet (median R 2 = 0.047) improved the PGS predictive performance (Wilcoxon rank-sum test, P = 3.58 × 10 − 11 ) in all cerebellar substructure volumes compared to SUIT + FS (median R 2 = 0.0027). In cross-ancestry prediction, CerebNet (EAS: median R 2 = 0.019; AFR: median R 2 = 0.0037) also improved the performance (Fig. 1 E, table S6 ) in 28/30 and 24/30 traits in CHIMGEN EAS (Wilcoxon rank-sum test, P = 1.49 × 10 − 6 ) and ABCD AFR (Wilcoxon rank-sum test, P = 9.67 × 10 − 3 ) compared to SUIT + FS (EAS: median R 2 = 0.0068; AFR: median R 2 = 0.0018). Overview of formal GWASs After verifying that CerebNet improves the genetic discovery of cerebellar substructure volumes, we performed comprehensive GWASs on the autosomal and X-chromosomal variants to investigate the genetic architecture of 31 cerebellar volumetric traits derived from CerebNet by further including the total cerebellar volume (Fig. 2 A). GWASs were divided into sex-combined and sex-stratified (males and females) analyses, which were further split into univariate and multivariate analyses. In each subcategory (e.g., sex-combined univariate), we conducted four single-dataset GWASs (UKBB EUR -GWAS, ABCD EUR -GWAS, CHIMGEN EAS -GWAS, and ABCD AFR -GWAS), an EUR-GWAS meta-analysis, and two cross-ancestry GWAS meta-analyses (EUR-EAS-GWAS and EUR-EAS-AFR-GWAS). We did not find any population stratification ( table S7 ) based on the genomic control inflation factor (λ GC ) and linkage disequilibrium (LD) score regression (LDSC) intercepts 29 . We regarded the sex-combined univariate GWASs as the primary results, while conducting other GWASs for supplementary analyses. We reported LD-independent associations, signals, and loci throughout the GWASs, and reported study-wide significant associations ( P < 1.61 × 10 − 9 ) in univariate GWASs and genome-wide significant associations ( P < 5 × 10 − 8 ) in multivariate GWASs. Genetic discovery in sex-combined univariate GWASs Single-dataset GWASs. We used the mixed linear model in fastGWA 27 to perform the sex-combined univariate GWASs ( P < 1.61 × 10 − 9 ) for the 31 cerebellar volumetric traits in 40,826 UKBB EUR , 5,808 ABCD EUR , 7,083 CHIMGEN EAS , and 3,354 ABCD AFR participants, respectively. We identified 1,289/948 variant/locus-trait associations and 447/252 signals/loci in UKBB EUR -GWASs; 39/35 associations and 21/18 signals/loci in ABCD EUR -GWASs; 31/31 associations and 17/16 signals/loci in CHIMGEN EAS -GWASs; and 5/5 associations and 3/3 signals/loci in ABCD AFR -GWASs ( Fig. S3, table S8, table S9 ). We noted that small non-EUR datasets could also yield new findings. For example, the locus-trait association between rs4752582 (10q26.13) and right cerebellar lobule VI volume was significant ( P = 1.46 × 10 − 9 ) in ABCD AFR -GWAS rather than in other three GWASs. The SNP is an expression quantitative trait locus (eQTL) of FGFR2 . FGFR2 signaling in cerebellar Purkinje neurons is vital to motor learning 30 . EUR-GWAS meta-analyses. Based on the summary statistics from UKBB EUR -GWASs and ABCD EUR -GWASs, we utilized the inverse variance weighted (IVW) fixed effect model implemented in METAL 31 to conduct the EUR-GWAS meta-analyses ( P < 1.61 × 10 − 9 ) for the 31 cerebellar volumetric traits. We found 1,602/1,144 variant/locus-trait associations and 558/288 signals/loci in EUR-GWASs ( Fig. S3, table S10, table S11 ). Cross-ancestry GWAS meta-analyses. We also used the IVW fixed effect model in METAL 31 to conduct the cross-ancestry GWAS meta-analyses ( P < 1.61 × 10 − 9 ) for the 31 cerebellar volumetric traits. Based on the summary statistics from EUR-GWASs and CHIMGEN EAS -GWASs, we performed the cross-ancestry EUR-EAS-GWASs. We identified 1,719/1,273 variant/locus-trait associations and 563/323 signals/loci ( Fig. S3, table S12, table S13) . By further including ABCD AFR -GWAS summary statistics, we performed the cross-ancestry EUR-EAS-AFR-GWAS meta-analyses ( P < 1.61 × 10 − 9 ). We found 1,681/1,266 variant/locus-trait associations and 553/316 signals/loci ( Fig. S3, table S14, table S15) . Pooling significant associations from sex-combined univariate GWASs. As most participants in our GWASs were of European ancestry and the majority of previous GWASs for cerebellar volumetric traits 18 – 20 were conducted on EUR individuals, we used the LD EUR reference to pool the results from sex-combined univariate GWASs and identify new findings. We identified 2,091/1,484 variant/locus-trait associations (Fig. 2 B) and 692/353 signals/loci ( table S16 , table S17 ). Compared to the 614/447 known variant/locus-trait associations and 217/135 signals/loci at P < 1.61 × 10 − 9 identified by all previous studies 18 – 21 , we found 1,641 new variant-trait associations, 1,102 locus-trait associations, 522 signals, and 241 loci ( table S16 , table S17 ). We provided evidence for improved genetic discovery in cross-ancestry GWASs. As an example, we identified a significant association ( P = 2.19 × 10 − 10 ) between rs79966207 (22q13.33) and the volume of right cerebellar lobule VI in the EUR-EAS cross-ancestry GWAS (Fig. 2 C), while the association was not significant in single-dataset GWASs or in the EUR-GWAS. The variant is a missense variant of PLXNB2 , regulating the timing of differentiation and the motility of cerebellar granule neurons 32 . Another example was the association between rs61742642 (14q24.3) and left cerebellar lobule X volume, which was only significant ( P = 3.91 × 10 − 10 ) in EUR-EAS-AFR cross-ancestry GWAS (Fig. 2 D). The variant is a missense variant of ESRRB , serving as a marker of Purkinje cells in the cerebellum 33 . Genetic discovery in sex-stratified univariate GWASs Male-specific GWASs. We performed the four single-dataset univariate GWASs for the 31 cerebellar volumetric traits in males, based on which we conducted EUR and cross-ancestry GWAS meta-analyses ( Fig. S4 ). Using the male-specific LD EUR reference, we pooled them into 523/429 variant/locus-trait associations and 201/142 signals/ loci. We additionally identified 8/5 new variant/locus-trait associations and 5/3 new signals/loci ( table S18, table S19 ). Female-specific GWASs. We also performed the four single-dataset univariate GWASs and three GWAS meta-analyses for these 31 cerebellar volumetric traits in females ( Fig. S5 ), and employed the female-specific LD EUR reference to pool the results into 689/556 variant/locus-trait associations and 251/174 signals/loci. We additionally found 18/15 new variant/locus-trait associations and 4/1 new signals/locus ( table S20, table S21 ). Genetic discovery in multivariate GWASs We applied C-GWAS 34 to conduct sex-combined multivariate GWASs ( P < 5 × 10 − 8 ) based on the summary data of sex-combined univariate GWASs for 27 non-overlapping cerebellar volumetric traits. We identified 386, 19, 34, 4, 428, 443, and 439 loci from UKBB EUR -GWASs, ABCD EUR -GWASs, CHIMGEN EAS -GWASs, ABCD AFR -GWASs, EUR-GWASs, EUR-EAS-GWASs, and EUR-EAS-AFR-GWASs, respectively ( Fig. S6 ). These loci were pooled into 484 LD-independent loci, including 153 additionally new loci ( table S22 ). We also performed the male- and female-specific C-GWASs for the 27 cerebellar volumetric traits ( Fig. S6 ). Using the sex-specific LD EUR reference, we pooled the results into 225 loci for males and 262 for females. We additionally discovered 9 new loci ( table S23, table S24 ). Allele-effect heterogeneity across ancestries Cochran's Q test (CQ-test) was performed to identify the allele-effect heterogeneity across ancestries (EUR, EAS, and AFR) for the pooled variant-trait associations ( P < 1.61 × 10 − 9 ) from sex-combined univariate GWASs (Fig. 3 A). In the 1,606 associations included in all ancestry-specific GWASs, we found 1,321 (82.25%) ancestry-shared associations (CQ-test: P ≥ 0.05, table S25 ). One ancestry-shared association (CQ-test: P = 0.33, Fig. 3 B) was found between rs2350079 (4q13.2) and cerebellar vermis VII volume. This SNP is an eQTL of EPHA5 , regulating the development of neuronal cytoarchitecture 35 . We identified 19 (1.83%) ancestry-specific associations (CQ-test: P < 3.11 × 10 − 5 , table S25 ). For example, we discovered one EAS-specific association (CQ-test: P = 2.63 × 10 − 8 , Fig. 3 C) between rs72838327 (17p13.2) and left cerebellar white matter volume. The variant is mapped to K1F1C , causing cerebellar dysfunction and cerebellar ataxia 36 , 37 . We also found an EUR-specific association (CQ-test: P = 9.36 × 10 − 6 , Fig. 3 D) between rs2217466 (2q36.1) and right cerebellar lobule V volume. The SNP is mapped to PAX3 , a marker of cerebellar development 38 . Allele effect heterogeneity between sexes To exclude the potential bias from inter-ancestry differences, we only used the CQ-test to identify the allelic-effect heterogeneity between sexes for the pooled variant-trait associations ( P < 1.61 × 10 − 9 ) obtained from the sex-stratified univariate GWASs in the EUR (n = 620), CHIMGEN EAS (n = 6), and ABCD AFR (n = 4) using the ancestry-specific LD reference. These associations were categorized into 490 sex-shared (CQ-test: P ≥ 0.05) and 8 sex-specific (CQ-test: P < 0.05/630 = 7.94 × 10 − 5 , Bonferroni corrected) associations (Fig. 3 E, table S26 ). For example, we found a female-specific association between rs6819982 (4q31.21) and right cerebellar white matter volume (CQ-test: P = 5.19 × 10 − 7 , Fig. 3 F). The variant is an eQTL of HHIP , involved in hedgehog signaling that orchestrates cerebellar development 39 . We found a male-specific association between rs6060308 (20q11.22) and right cerebellar lobule VIIb volume (CQ-test: P = 3.92 × 10 − 6 , Fig. 3 G). The variant is an eQTL of EDEM2 , involved in endoplasmic reticulum-associated degradation and linked to early-onset cerebellar ataxia 40 . Statistical fine-mapping We performed statistical fine-mapping based on the summary statistics of EUR-GWASs, EUR-EAS-GWASs, and EUR-EAS-AFR-GWASs on the 31 cerebellar volumetric traits, respectively. For each locus of the significant locus-trait associations, we estimated its 95% credible set and identified causal variants with posterior probability (PP) above 0.8 based on the matched LD reference and suitable approaches (Fig. 4 A). SuSiE-R 41 , 42 was used to EUR-GWASs with LD EUR reference, while SuSiEx 43 was applied to EUR-EAS-GWASs with LD EUR and LD EAS references and EUR-EAS-AFR-GWASs with LD EUR , LD EAS , and LD AFR references. We successfully identified 941 95% credible sets for 1,144 locus-trait associations from EUR-GWASs ( table S27 ), 1,660 for 1,273 locus-trait associations from EUR-EAS-GWASs ( table S28 ), and 2,073 for 1,266 locus-trait associations from EUR-EAS-AFR-GWASs ( table S29 ). We also found 53 causal variants (PP > 0.8) from EUR-GWASs ( table S30 ), 142 from EUR-EAS-GWASs ( table S31 ), and 356 from EUR-EAS-AFR-GWASs ( table S32 ) (Fig. 4 B), generating 453 unique causal variants across these GWASs. We aligned the 95% credible sets with one causal variant assumption across these fine-mapping analyses, generating 467 aligned 95% credible sets ( table S33 ). We used Wilcoxon rank-sum test to compare the size and maximal PP of these 95% credible sets. The median size of these 95% credible sets was 33, 14, and 10 for one, two, and three-ancestry fine-mapping, demonstrating significant differences (Wilcoxon rank-sum test: P = 2.25 × 10 − 47 for EUR vs EUR-EAS; P = 8.83 × 10 − 28 for EUR vs EUR-EAS-AFR; P = 5.97 × 10 − 6 for EUR-EAS vs EUR-EAS-AFR; Fig. 4 C). The median maximal PP values in these 95% credible sets was 0.09, 0.19, and 0.28 for one, two, and three-ancestry fine-mapping, also demonstrating significant differences (Wilcoxon rank-sum test: P = 1.62 × 10 − 35 for EUR vs EUR-EAS; P = 3.80 × 10 − 19 for EUR vs EUR-EAS-AFR; P = 2.33 × 10 − 5 for EUR-EAS vs EUR-EAS-AFR; Fig. 4 C). Functional annotations We used three approaches to perform functional annotations for the 453 unique causal variants (PP > 0.8) from fine-mapping. We used ANNOVAR 44 to annotate functional consequences, and found 226 (49.9%) variants in the intergenic region, 164 (36.2%) in the intron region, 6 (1.3%) in the UTR region, and 6 (1.3%) missense variants. We then assessed deleteriousness using the combined annotation-dependent depletion (CADD) score 45 , identifying 13 (2.9%) pathogenic variants (CADD score > 20). Finally, we evaluated the regulatory function of causal variants by RegulomeDB (RDB) 46 . We found 200 (44.2%) variants with regulatory potential (RDB < 4). These results ( tables S30-32 ) improve the understanding of genetic mechanisms of cerebellar structure. For example, as a causal variant of cerebellar vermis X volume (PP = 0.87), rs145919520 (17p13.1) is a missense variant of CLUH . CLUH plays a crucial role in maintaining functional mitochondria in axons 47 . As a fine-mapped causal variant of cerebellar vermis VIII volume (Fig. 3 D), rs9956387 (18q21.1) is a missense variant of SKOR2 , serving as a transcriptional regulator in Purkinje cells during cerebellum development 48 . The lead variant rs7540842 (1q25.2) mapped to ASTN1 was identified as a fine-mapped causal variant of right cerebellar lobule IX volume (Fig. 3 E). ASTN1 affects the cerebellar volume, neuronal migration, and development of Purkinje cells 49 . Colocalization with gene expression For loci of 1,144 locus-trait associations from EUR-GWASs for cerebellar volumetric traits, we used Coloc 50 to perform the Bayesian colocalization to identify the shared causal variants with eQTLs ( P < 5 × 10 − 8 ) of human cerebellar tissue from MetaBrain ( https://metabrain.nl/ ) 51 . Colocalization was defined as the posterior probability of shared causal variant (PP.H4) over 0.8. We found 160 colocalizations between 71 genes and 30 cerebellar volumetric traits ( Fig. S7A ; table S34 ). Among them, 36 genes had colocalization with at least two cerebellar volumetric traits, such as SLC44A5 , PTK2 , and ZFHX4 colocalized with 14, 13, and 6 traits, respectively. SLC44A5 is involved in lipid metabolism, and its SNPs have been associated with education attainment 52 and major depressive disorder 53 . PTK2 encodes FAK, a cell adhesion tyrosine kinase, implicating in synaptic branching 54 and linking to congenital cerebellar hypoplasia 55 . ZFHX4 acts as a marker of GABAergic interneurons in the forming deep nuclei of the cerebellum 33 . Pathway enrichment Based on 71 prioritized genes whose cerebellar expression colocalized with locus-trait associations from EUR-GWASs for cerebellar volumetric traits, we input these genes to g:Profiler ( https://biit.cs.ut.ee/gprofiler/gost ), a web server for functional enrichment analysis 56 , to perform the pathway enrichment analysis based on the pre-specified pathways in GO (15,472 biological processes) database 57 . We found 12 enrichment pathways ( P c < 0.05, Benjamini-Hochberg FDR corrected), demonstrating the nervous system development, neurogenesis, and system development ( Fig. S7B ; table S35 ) as the top three biological processes. Genetically informed cerebellar subregions The cerebellum consists of two hemispheres (including both cortex and white matter) connected by a central structure called the vermis. Based on its fissures, the cerebellum is categorized into three lobes: the anterior (lobules I-V), posterior (lobules VI-IX), and flocculonodular (lobule X) lobes. Nevertheless, the genetically informed subregions of the cerebellum remain unclear. Here, we utilized the high-definition likelihood (HDL) method 58 to estimate bivariate genetic correlations across the volumes of 27 distinct cerebellar substructures. Based on the genetic correlation matrix (Fig. 5 , table S36 ), we used genomic structural equation modelling (gSEM) 59 to perform genetic clustering analyses, and defined an acceptable fit of the model as the comparative fit index (CFI) > 0.90 and standardized root-mean-squared residual (SRMR) < 0.10 59 . We assessed a common factor model, yielding a poor model fit (CFI = 0.51, SRMR = 0.15). We then evaluated the anatomical scheme (anterior lobe, posterior lobe, flocculonodular lobe, vermis, and white matter), which also resulted a poor fit (CFI = 0.79, SRMR = 0.12). To determine the best-fit model, we conducted exploratory factor analysis (EFA) based on the genetic correlation matrix of cerebellar substructure volumes. We identified a nine-factor model that could explain 78.1% of the total genetic variance ( table S37 ). The subsequent confirmatory factor analysis (CFA) confirmed the nine-factor model by exhibiting an almost perfect fit (CFI = 0.99, SRMR = 0.058; Fig. 5 ). These results indicate that the cerebellum may be divided into nine subregions, each with a unique genetic architecture. The vermis substructures were clustered into a single subregion. In cerebellar hemispheres, identical substructures are grouped into the same subregions, such as the white matter subregion. Consistent with anatomical divisions, the bilateral anterior lobes (lobules I-V) and the bilateral flocculonodular lobes (lobule X) were also clustered into two distinct subregions, respectively. However, the cerebellar posterior lobe had complex genetic architecture and was divided into five subregions. Although lobules VI and IX were two distinct subregions, lobules CrusI, CrusII, VIIb, VIIIa, and VIIIb were clustered into three subregions that differ from the anatomical subdivisions. The lobules CrusII, VIIb, and VIIIa were included in one subregion, while lobules CrusI and VIIIb were clustered into two distinct subregions. Genetic associations with cognitive and mental health phenotypes The extensive cognitive and mental health assessments in the UKBB dataset provided us a unique opportunity to investigate the associations of PGSs of cerebellar volumetric traits with these cognitive and mental health phenotypes. We utilized summary statistics from EUR-GWASs for the 31 cerebellar volumetric traits to construct the PGS models, which were then applied to estimate the PGS scores of 371,558 unrelated EUR-UKBB participants not included in EUR-GWASs. In these participants, we used PHESANT 60 to conduct phenome-wide association studies (PheWASs) to identify associations between 31 PGSs and 198 cognitive and mental health phenotypes ( table S38 ). We found 141 significant associations ( P < 0.05/31/198 = 8.15 × 10 − 6 , Fig. 6 A, table S39 ) between 31 cerebellar volumetric traits and 20 behavioral phenotypes (two cognitive and 18 mental health phenotypes). For example, the reaction time in pairs matching was associated with PGSs of six cerebellar volumetric traits (Fig. 6 B), consistent with the longer reaction time in patients with cerebellar lesions 61 . The associations of PGS of cerebellar vermis volume with tenseness/restlessness ( β = -0.007, P = 2.78 × 10 − 16 ), happiness ( β = 0.015, P = 6.38 × 10 − 10 ), and morning drink of alcohol ( β = 0.036, P = 3.26 × 10 − 8 , Fig. 6 C) were also in line with previous observations 62 , 63 . These results offered further support for the cerebellum's involvement in the regulation of cognition and emotion. Genetic associations with neuropsychiatric disorders (NPDs) Genetic correlation. Based on the EUR-GWAS summary statistics for both cerebellar volumetric traits from this study and NPDs from prior studies ( table S40 ), we used HDL 58 to explore genetic correlations between 31 cerebellar volumetric traits and ten NPDs. We found 11 significant genetic correlations ( P < 1.61 × 10 − 4 , Bonferroni corrected) between nine cerebellar volumetric traits and four NPDs (Fig. 6 D, table S41 ), including genetic correlations between eight cerebellar volumetric traits and PTSD (all P < 2.96 × 10 − 5 ), total cerebellar volume and ALS (r = 0.076, P = 1.00 × 10 − 4 ), left lobules I-IV volume and BD (r = -0.075, P = 8.52 × 10 − 5 ), and left lobule VIIIa volume and depression (r = -0.070, P = 4.42 × 10 − 5 ). Genetic colocalization. Based on the EUR-GWAS summary statistics for 31 cerebellar volumetric traits and ten NPDs, we employed Coloc 50 to explore genetic loci that are shared between cerebellar volumetric traits and NPDs. Among the 1,144 locus-trait associations from EUR-GWASs, we identified colocalization (PP.H4 > 0.8; Fig. 6 E and table S42 ) between 17 cerebellar volumetric traits and six NPDs (AD, BD, depression, PD, PTSD, and SCZ) at 11 loci. We further performed multi-trait colocalization using HyPrColoc 64 , to identify the loci shared by cerebellar gene expression, cerebellar volumetric traits, and NPDs. We found nine multi-trait colocalizations (PP > 0.8, Fig. 6 E and table S43 ) at two loci between cerebellar expression of two genes ( SNX31 , ARHGAP27 ), five cerebellar volumetric traits, and three NPDs (AD, PD, and PTSD). One locus (17q21.31) was shared by cerebellar ARHGAP27 expression, left and right cerebellar white matter volumes, and three NPDs (AD, PD, and PTSD) (PP = 0.80–0.93, table S43 ). Another locus (8q22.3) was shared by cerebellar SNX31 expression, three cerebellar volumetric traits (left cerebellar cortex volume, total cerebellar volume, and cerebellar vermis IX volume), and AD (PP = 0.95–0.96, table S43 ). SNX31 is a member of the sorting nexin family, which is essential for synaptic function, and dysregulation of sorting nexin is observed in patients with AD 65 . Discussion In this study, we performed GWASs on 31 cerebellar volumetric traits obtained from deep-learning segmentation in 57,071 participants across three ancestries (EUR, EAS and AFR). We discovered that deep-learning outperformed atlas-based segmentation in heritability estimation, genetic discovery, and polygenic prediction. We identified 353 independent loci in sex-combined univariate GWASs, including 241 new loci. We then performed sex-stratified univariate GWASs and additionally discovered four new loci. We finally conducted sex-combined and sex-stratified multivariate GWASs and further found 162 new loci. Although most associations were shared by ancestries and sexes, we still found 19 ancestry-specific and eight sex-specific associations. We prioritized 453 causal variants and 71 genes and clustered 27 cerebellar anatomical substructures into nine subregions based on their genetic architecture. PGSs of cerebellar volumetric traits were correlated with 20 cognitive and mental health phenotypes. These cerebellar volumetric traits showed genetic correlations with four and genetic colocalizations with six neuropsychiatric disorders. A distinctive contribution of our study to the field of neuroimaging genetics is the demonstration that deep-learning-based cerebellar segmentation surpassed atlas-based cerebellar segmentation in heritability estimation, genetic discovery, and genetic risk prediction. The SUIT atlas is the most commonly used probabilistic atlas of the human cerebellum 9 . The cerebellar substructure volumes of each participant can be obtained by non-linearly registering structural magnetic resonance images to the atlas known as SUIT-based cerebellar segmentation. As the most reliable atlas-based segmentation for the human cerebellum, it has been applied to calculate cerebellar substructure volumes in all previous GWASs on cerebellar volumetric traits 18 – 21 . As a probabilistic atlas, SUIT cannot precisely estimate the boundary of each cerebellar substructure for each individual. The imprecise phenotyping may undermine heritability estimation, GWAS discovery, and PGS prediction 66 . Recently, deep learning methods have been applied to cerebellar segmentation, achieving superior performance compared to atlas-based segmentation. For instance, CerebNet is proposed as a fast and reliable deep-learning pipeline for detailed cerebellar segmentation and showed higher precision, reliability, and sensitivity than other methods including SUIT segmentation 9 . The improved heritability estimation, GWAS discovery, and PGS prediction for cerebellar volumetric traits obtained from CerebNet indicate that more precise phenotyping is essential for genetic analyses of human phenotypes. A significant contribution of our study to the field of neuroimaging genetics is the discovery of new genetic associations with cerebellar volumetric traits through precise phenotyping and comprehensive multi-ancestry GWASs. For instance, we found 1,484 locus-trait associations ( P < 1.61 × 10 − 9 ) in sex-combined univariate GWASs for 31 cerebellar volumetric traits, comprising 1,102 new associations, which is 3.3 times the 447 associations reported in previous GWASs for these traits 18 – 21 . A new association was observed between rs17379472 (16q21) and bilateral lobule VI volumes. The lead variant is a missense variant of ADGRG1 (also called as GPR56 ), leading to cerebellar hypoplasia 67 . We found that increased ancestry diversity in statistical fine-mapping could enhance the identification of causal variants and decrease the average size of 95% credible sets, emphasizing the importance of including ancestrally diverse populations in GWASs 68 . Although most genetic associations were shared by ancestries and sexes, we also found 19 ancestry-specific and eight sex-specific associations, providing the potential genetic substrates for the observed differences in cerebellar volumetric traits between ancestries and sexes 69 , 70 . By leveraging the distributed genetic influences across cerebellar substructure volumes, we performed multivariate GWASs for cerebellar volumetric traits and additionally discovered 162 new loci. For example, we found a locus (lead variant rs111511908) that was only significant in multivariate GWASs. The variant is an eQTL of CACNB4 , linked to neurodevelopmental disorder including cerebellar atrophy 71 . Another valuable aspect of this study is the identification of genetically informed subregions of the human cerebellum. Although correspondence was observed between the majority of anatomical and genetic subdivisions, notable differences were identified in the affiliation of the Crus and VIII sub-lobules within the posterior lobe. Based on the genetic architecture, CrusII, VIIb, and VIIIa were grouped into a single subregion, while CrusI and VIIIb were categorized into two separate subregions. The genetically informed classification of these sub-lobules is partially aligned with the functional and pathological distinctions between Crus I and Crus II, as well as between VIIIa and VIIIb, along with the observed coactivation and joint damage between VIIb and VIIIa. For instance, in language processing, CrusI was activated during the syntactic task, while CrusII was activated during the semantic task 72 . In patients with bipolar disorder, CrusI exhibited increased functional connectivity with other brain regions, while CrusII showed decreased connectivity 73 . In long-term plasticity, drum training resulted in increased volume in VIIIa but decreased volume in VIIIb 74 . Additionally, synergy among the CrusII, VIIb, and VIIIa has been frequently documented in earlier studies. As examples, VIIb and VIIIa were coactivated in both working memory and attention tasks 75 and jointly damaged in schizophrenia patients with persistent auditory verbal hallucinations 76 ; CrusII and VIIb exhibited volume reduction in patients with ASD 77 , PTSD 11 , and schizophrenia 78 ; and CrusII and VIIIa encoded anticipatory mechanisms for dexterous object manipulation 79 . Therefore, the genetic parcellation of the posterior cerebellar lobe may provide a new framework for understanding the function and damage of its sub-lobules. In PheWAS, we identified significant associations between PGSs of all cerebellar volumetric traits and mental health phenotypes, indicating that all cerebellar subregions may play a role in emotion processing. These findings confirmed and extended previous observations of the involvement of several cerebellar subregions in emotion processing 80 , 81 . Consistent with earlier studies 79 , 82 , we found that left CrusII and bilateral VIIIa were associated with cognitive performance. Therefore, these results verified the role of the cerebellum in cognitive and emotional processing. We also identified genetic correlations or colocalization of cerebellar volumetric traits with AD, PD, ALS, PTSD, BD, SCZ, and depression, offering potential genetic substrates underlying the observed cerebellar impairments in these disorders 11 , 12 , 14 – 16 , 73 , 76 . In the multi-trait colocalization analyses, we found two gene ( ARHGAP27 and SNX31 ) whose cerebellar expression showed colocalization with both cerebellar volumetric traits and NPDs. For instance, the expression of ARHGAP27 was colocalized with cerebellar white matter volumes and three NPDs (AD, PD, and PTSD). ARHGAP27 is a Rho GTPase-activating protein, acting as a key player in neurodegeneration 83 . Several limitations should be noted when interpreting our findings. First, the study population primarily comprised EUR individuals, resulting in unstable evaluation of heritability in individuals of EAS and AFR. Additionally, a lenient threshold was used to select AFR participants, which does not accurately represent the whole population of African ancestry. Second, although we have harmonized the cerebellar volumetric traits to minimize the influence of imaging scanners, confounding factors may still exist and limit the interpretation of our results. Finally, even if we have used the sample-weighted method to obtain the cross-ancestry LD reference, we cannot precisely estimate the LD structures in cross-ancestry population, which may influence analyses that require LD information. In conclusion, we confirmed that precise phenotyping is essential in neuroimaging genetic analyses. Through precise phenotyping and extensive multi-ancestry analyses, we identified 407 new loci associated with cerebellar volumetric traits, including sex-specific and ancestry-specific associations. We developed the first genetically informed atlas of the cerebellum and offered genetic insights into the associations between the cerebellum and cognition, emotion, and neuropsychiatric disorders. These findings may enhance our understanding of the genetic architecture of the cerebellum and its links to brain functions and disorders. Methods Participants The participants included in this study were sourced from three distinct datasets: the UK Biobank (UKBB; https://www.ukbiobank.ac.uk/ ) study 24 , the Chinese Imaging Genetics (CHIMGEN; http://chimgen.tmu.edu.cn/ ) study 23 , and the Adolescent Brain Cognitive Development (ABCD; https://abcdstudy.org/ ) study 25 . The UKBB study enrolled approximately 500,000 participants aged 40 to 69 years from 22 research centers throughout the United Kingdom 84 . This study was approved by the National Health Service Research Ethics Service (21/NW/0157), and written informed consent was obtained from each participant. We accessed to the data under application number 75556. The CHIMGEN study recruited 7,306 healthy Chinese Han participants aged 18 to 30 years from 32 research centers, which was approved by the Medical Research Ethics Committees of all institutions, and all participants provided written informed consent 23 . The ABCD study is a longitudinal cohort comprising over 10,000 children aged 9 to 10 years at their baseline assessment from 21 research centers. This study received approval from a central Institutional Review Board (IRB) at the University of California, San Diego for the majority of research centers, or from a local IRB for a few research centers. All parents or caregivers provided the written informed consent and all children provided the written assent 85 . We accessed to the data under application ID 17607. For each dataset, participants with qualified structural magnetic resonance imaging (MRI) and genomic data were included in this study. Data preparation UKBB. Of 487,207 participants who passed the initial quality control (QC) of genetic data 84 , we excluded 651 participants with sex chromosome aneuploidy and 186 with sex mismatch, remaining 486,370 participants with qualified genomic data, including 44,179 with structural MRI data. We employed CerebNet to segment the cerebellum into 30 substructures 22 . After excluding 919 participants with brain tumors, imaging artifacts, incomplete cerebellum coverage, or segmentation error, the volume of each cerebellar substructure was obtained for the 43,260 remaining participants. Following the removal of 125 participants whose substructure volume was outside five times the median absolute deviation (MAD) from the median, 43,135 participants were retained. As most participants reported European ancestry (EUR), we utilized SNPweights 86 to verify the ancestry of each participant against the EUR reference panel from the 1000 Genomes Project phase 3 (1KGP), ensuring that the EUR proportion was greater than 80% 87 . Ultimately, we included 40,826 EUR participants from the UKBB dataset. Using the criteria of minor allele frequency (MAF) ≥ 0.005, imputation information score (info) ≥ 0.6, and P ≥ 1 × 10 − 7 in Hardy-Weinberg equilibrium (HWE), we finally included 9,683,756 autosomal and 317,880 X-chromosomal bi-allelic variants. CHIMGEN. Among the 7,306 CHIMGEN participants, 7,195 with DNA samples were genotyped by Illumina ASA-750K (Asian Screening Array) that was specially designed for Asians. Details for the sample- and variant-level QC, genetic principal component analysis (PCA), and genetic data imputation are provided in our previous study 21 . All 7,163 participants with qualified genetic data also had structural MRI data. After excluding 61 participants without qualified structural MRI data, we used CerebNet 22 to calculate the cerebellar substructure volumes for the remaining 7,102 participants. Following the removal of 19 participants whose subregion volume exceeded five times the MAD from the median, we included 7,083 East Asian (EAS) participants from the CHIMGEN dataset in the genetic analyses. The analyses were performed for 8,790,144 imputed autosomal and 227,164 X-chromosomal bi-allelic variants (MAF ≥ 0.005, info ≥ 0.6, and P HWE ≥ 1 × 10 − 7 ). ABCD. Among the 11,760 participants from the ABCD dataset, 11,101 remained after genetic QC, including 10,660 with qualified structural MRI data. We utilized CerebNet 22 to calculate cerebellar substructure volumes and excluded 264 participants whose substructure volume was more than five times the MAD from the median, resulting in 10,396 participants. We also utilized SNPweights 86 to identify EUR participants with an EUR proportion over 80% 87 . As African descent (AFR) was also commonly seen in ABCD participants, we employed the criteria from a prior study 88 to identify the participants with African descent using the reference panels for African, European, and East Asian from 1KGP. A participant was classified as having African descent if AFR proportion exceeded 5%, along with EUR proportion < 80% and EAS proportion < 5%. We finally included 5,808 EUR and 3,354 AFR participants from the ABCD dataset. To maintain consistency with the genomic build of the UKBB and CHIMGEN datasets, we converted variants of the ABCD dataset from GRCh38/hg38 to GRCh37/hg19. With the same criteria (MAF ≥ 0.005, info ≥ 0.6, and P HWE ≥ 1 × 10 − 7 ), we finally included 9,064,819 autosomal and 278,985 X-chromosomal bi-allelic variants for the 5,808 EUR participants, and 7,162,933 autosomal and 232,441 X-chromosomal bi-allelic variants for the 3,354 AFR participants. Cerebellar segmentation In this study, we employed CerebNet 22 , a fast and reliable deep-learning pipeline, to automatically segment the cerebellum into 30 substructures using brain structural MRI data without any preprocessing. CerebNet uses a FastSurferCNN deep-learning model tailored for cerebellar segmentation, showing superior accuracy, test-retest reliability, extensibility, and sensitivity compared to other segmentation methods 22 . These 30 cerebellar substructures comprised five for the vermis (VI, VII, VIII, IX, and X), 20 for the hemisphere (left and right I-IV, V, VI, CrusI, CrusII, VIIb, VIIIa, VIIIb, IX, and X), along with five global ones (left white matter, right white matter, left cortex, right cortex, and vermis). CerebNet can automatically calculate the volumes of these 30 cerebellar substructures. To test whether CerebNet outperforms traditional segmentation methods in genetic analyses of cerebellar volumetric traits, we also computed the volumes of these cerebellar substructures by integrating the FreeSurfer automatic segmentation 10 and the spatially unbiased infra-tentorial template (SUIT) 9 , referred to as SUIT + FS. SUIT is the most detailed and widely used cerebellar atlas, subdividing the cerebellum into eight vermis and 20 hemisphere substructures. By combining three substructures of vermis VII into a single substructure and two substructures of vermis VIII into one substructure, we derived 25 cerebellar substructures similar to those from CerebNet. The vermis volume can be calculated by summing all the vermis substructures, whereas FreeSurfer-ASEG can estimate the volumes of bilateral white matter and cortex. Thus, we established the correspondence in 30 cerebellar volumetric traits between CerebNet and SUIT + FS. For UKBB participants, we downloaded the volume data (SUIT + FS) for cerebellar substructures from the database 89 . For CHIMGEN participants, we used the volume data for cerebellar substructures from SUIT + FS that have been utilized in a prior study 21 . For ABCD participants, we used the CHIMGEN pipeline (SUIT + FS) to obtain the volume data for cerebellar substructures. Specifically, we used computational anatomy toolbox (CAT 12, version r1364, http://dbm.neuro.uni-jena.de/cat ) to preprocess brain structural MRI data. After correcting for imaging inhomogeneity due to B1-field bias, brain structural images were segmented into gray matter (GM), white matter, and cerebrospinal fluid using an adaptive Maximum A Posterior (MAP) technique 90 . Based on the 10,294 ABCD participants, the Diffeomorphic Anatomical Registration Through Exponentiated Lie Algebra (DARTEL) algorithm 91 was used to create a population-specific GM template in the Montreal Neurological Institute (MNI) space. The segmented GM images were normalized to the population-specific GM template using the DARTEL algorithm and resampled into cubic voxels of 1.5 mm. Modulation was applied to the GM images to preserve absolute GM volume (GMV). Based on the voxel-wise GMV map, the SUIT approach in combination with FreeSurfer-ASEG was used to calculate the volumes of cerebellar substructures. To evaluate the reliability of CerebNet segmentation, we utilized this method to compute the volumes of cerebellar substructures based on the brain structural MRI data obtained at two separate time points from the same participants (CHIMGEN: n = 26; UKBB: n = 2,944; and ABCD: n = 4,076). For each test-retest dataset, we calculated the intraclass correlation coefficient (ICC) of the volume data of each substructure obtained at the two time points across all participants. We found high reliability in the three datasets (CHIMGEN: ICC = 0.62–0.97; UKBB: ICC = 0.88–0.96; ABCD: ICC = 0.87–0.98; table S2 ). Brain structural MRI data from the UKBB, CHIMGEN, and ABCD datasets were acquired using various scanners, which inevitably introduces bias into genetic analyses. Compared to including imaging sites as a covariate in the regression model, ComBat harmonization is a more effective method for eliminating the MRI scanner effects while preserving biological variability 92 . We tested the impact of ComBat harmonization in two participants who visited different centers and were scanned at 28 MRI scanners utilized for the CHIMGEN data acquisition. In each participant, we utilized CerebNet to calculate the volumes of 30 cerebellar substructures based on the MRI data obtained from each scanner, and used the coefficient of variation (CV) to evaluate the between-scanner variations of these traits. In both participants, CVs of these volumetric traits prior to harmonization were significantly reduced (Wilcoxon rank-sum test; Participant 1: P = 8.32 × 10 − 10 ; Participant 2: P = 1.40 × 10 − 5 ) after harmonization ( Fig. S8 ). Additionally, as the skewed data distribution would violate the assumption of normality when using a linear regression model for genetic analyses, quantile normalization was applied to these cerebellar volumetric traits. Comparing genetic findings between cerebellar segmentation methods Although CerebNet demonstrates greater accuracy, test-retest reliability, and sensitivity compared to other segmentation methods, including SUIT + FS 22 , its advantages in genetic analyses are still unclear. Here, we investigated whether cerebellar substructure volumes obtained from CerebNet exhibit higher heritability, more genetic associations, and improved predictive performance of polygenic scores (PGS) compared to those derived from SUIT + FS. Heritability. We used GCTA-GREML 26 to estimate single nucleotide polymorphism (SNP)-based heritability of each cerebellar substructure volume obtained from the two segmentation approaches in 40,826 UKBB EUR , 7,083 CHIMGEN EAS , 5,808 ABCD EUR , and 3,354 ABCD AFR participants, respectively. In these analyses, we controlled for age at scanning, genetically determined sex, age × sex, total intracranial volume (TIV), genetic principal components (PCs; top 40 for UKBB, 10 for CHIMGEN, and 32 for ABCD), and genetic batch (only for UKBB and ABCD). After calculating the genetic relationship matrix (GRM) for autosomal SNPs included in each GWAS, we estimated the variance of the trait explained by these SNPs, referred to as SNP-based heritability. The Wilcoxon rank-sum test was used to compare the difference in heritability of these traits obtained from CerebNet and SUIT + FS. Genetic discovery. To test whether CerebNet improves genetic discovery, we performed GWASs for 30 cerebellar substructure volumes obtained from CerebNet and SUIT + FS in 40,826 UKBB EUR participants. We generated a sparse GRM based on the full-dense GRM from the heritability estimation using a cutoff of 0.05. Based on the sparse GRM ( table S3 ), we employed the mixed linear model (MLM) implemented in fastGWA 27 to conduct GWAS (additive effect) with the same covariates as heritability estimate. We used the Wilcoxon rank-sum test to compare the number, P -value, and power of variant-trait associations ( P < 5 × 10 − 8 ) derived from the two segmentation methods. PGS prediction. To test whether CerebNet can improve PGS prediction compared to SUIT + FS, we employed the UKBB EUR data as the base dataset and the CHIMGEN EAS , ABCD EUR , and ABCD AFR data as three distinct target datasets. PRSice-2 28 was used to obtain the best-fit PGS model at the optimal P -value threshold for each cerebellar volumetric trait obtained from each segmentation method based on the UKBB-GWAS summary statistics for the trait. The best-fit model was defined as the PGS model with the largest explained variance (R 2 ) for the volumetric trait in the base dataset. The best-fit PGS model was then applied to each target dataset to calculate the R 2 of PGS for the trait in the target dataset. Then we used Wilcoxon rank-sum test to compare R 2 values of the 30 cerebellar volumetric traits obtained from CerebNet and SUIT + FS in each target dataset. Comprehensive GWASs After showing that CerebNet improves the genetic discovery of cerebellar substructure volumes, we performed comprehensive GWASs on the autosomal and X-chromosomal variants to investigate the genetic architecture of 31 cerebellar volumetric traits derived from CerebNet by further including the total cerebellar volume. GWASs included sex-combined and sex-stratified analyses, further dividing into univariate and multivariate analyses. In each subcategory, we conducted four single-dataset GWASs (UKBB EUR -GWAS, ABCD EUR -GWAS, CHIMGEN EAS -GWAS, and ABCD AFR -GWAS), one EUR-GWAS meta-analysis, and two cross-ancestry (EUR-EAS and EUR-EAS-AFR) GWAS meta-analyses using the same covariates as heritability estimate. We considered sex-combined univariate GWASs as the main results, while conducting other GWASs for supplementary insights. We reported the study-wide significant associations ( P < 1.61 × 10 − 9 ) in univariate GWASs and the genome-wide significant associations ( P < 5 × 10 − 8 ) in multivariate GWASs. The participants, genetic variants, and GWAS parameters are presented in table S3 . Sex-combined single-dataset univariate GWAS. In each of the sex-combined single-ancestry dataset (UKBB EUR , CHIMGEN EAS , ABCD EUR , and ABCD AFR ), we used MLM in fastGWA 27 to conduct univariate GWASs for the 31 cerebellar volumetric traits while controlling for the predefined covariates. Sex-combined univariate EUR-GWASs. Based on the summary data of sex-combined UKBB EUR -GWASs and ABCD EUR -GWASs for cerebellar volumetric traits, we utilized the inverse-variance-weighted (IVW) fixed effect model implemented in METAL 31 to conduct EUR-GWAS meta-analyses. Sex-combined univariate cross-ancestry GWASs. For each cerebellar volumetric trait, we used the IVW fixed effect model in METAL 31 to conduct EUR-EAS-GWAS meta-analysis based on the summary data of EUR-GWAS and CHIMGEN EAS -GWAS. We also used the same model to conduct EUR-EAS-AFR-GWAS meta-analysis based on the summary data of EUR-GWAS, CHIMGEN EAS -GWAS and ABCD AFR -GWAS. Sex-stratified GWASs. We repeated the above-mentioned GWASs in females and males, while controlling for the predefined covariates except for genetically determined sex and age × sex. M ultivariate GWASs. Considering the correlation structure of cerebellar volumetric traits, we employed C-GWAS 34 ( https://github.com/Fun-Gene/CGWAS ) to combine GWAS summary data of correlated traits to enhance the discovery of loci associated with cerebellar volumetric traits. C-GWAS was conducted on the volumes of 27 non-overlapping cerebellar substructures ( P < 5 × 10 − 8 ). Population stratification. We estimated the population stratification for each univariate GWAS using genomic control inflation factor (λ GC ) and linkage disequilibrium score regression (LDSC) intercept 29 . λ GC was calculated as the median of the resulting chi-squared (χ 2 ) test statistics (z scores) divided by 0.4549, which is the expected median of the χ 2 distribution with one degree of freedom. As high λ GC indicates either genomic inflation or polygenicity, we used LDSC intercepts to identify genomic inflation based on LD scores. We considered no population stratification if the LDSC intercept is close to one. Linkage disequilibrium (LD) references. We used imputed genotype data from 3,354 ABCD AFR , 7,083 CHIMGEN, and 40,826 UKBB participants with qualified genetic and imaging data to construct LD references for African ancestry (LD AFR ), East Asian ancestry (LD EAS ) and European ancestry (LD EUR ), respectively. We further constructed two merged LD references (LD EUR−EAS and LD EUR−EAS−AFR ) using the sample-weighted method based on the above-mentioned data. We also constructed female-specific and male-specific LD references using the same strategies. The LD references used in each analysis are presented in table S3 . Independent associations and loci. For each GWAS, the matched LD reference was used to identify independent variant-trait associations by PLINK clumping 93 with the following steps: (1) all significant variants were included in a list of prioritized variants; (2) the most significant one was defined as the first lead variant (independent variant), and variants within 1 Mb from or in LD with (r 2 > 0.1) the lead variant were clumped; (3) the remaining variants formed a new list, and then step (2) was repeated; and (4) the iterative process stopped until the list was empty. We identified independent locus-trait associations by: (1) creating loci for all independent variants by adding 1 Mb to both sides; (2) merging overlapping loci; (3) merging loci if an independent variant of one locus was in LD (r 2 > 0.1) with independent variant of another locus; and (4) merging loci overlapped with the major histocompatibility complex (MHC) or 8p23.1 region into one locus. The associations of the remaining loci with this trait were defined as independent locus-trait associations. For all variant-trait associations, we merged the lead variants with an LD r 2 > 0.1 or within 1 Mb. The remaining variants were defined as independent signals. We created loci for all locus-trait associations by merging 1 Mb to both sides, and then merged the overlapping loci to identify independent loci. Pooling significant associations. As the sample size of EUR participants was at least four times larger than those of other ancestries in this study, we used the LD EUR to pool the significant associations from all GWASs. We used the above-mentioned strategies to identify independent variant-trait associations, locus-trait associations, signals, and loci. Identifying new associations and loci. As prior GWASs for cerebellar volumetric traits 18 – 21 were conducted mainly in EUR participants, we utilized the same strategies and LD EUR reference to pool the GWAS results ( P < 1.61 × 10 − 9 ). We identified 614 known variant-trait associations, 447 locus-trait associations, 217 signals, and 135 loci. Based on these results, we defined a new variant-trait association if the variant was 1 Mb away from and not in LD (r 2 < 0.1) with any variants of the trait in the list of known variant-trait associations; a new locus-trait association when the locus was not overlapped with loci of known locus-trait associations and all lead variants in the locus were not in LD (r 2 < 0.1) with any lead variants in the loci of all known locus-trait associations; a new independent signals when the signals was 1 Mb away from and not in LD (r 2 < 0.1) with any known signals; and a new locus when the locus was not overlapped with any known loci and all lead variants in the locus were not in LD (r 2 < 0.1) with any lead variants in known loci. Allelic effect heterogeneity The Cochran's Q test (CQ-test) was employed to assess the allelic-effect heterogeneity of pooled variant-trait associations ( P < 1.61 × 10 − 9 ) among the sex-combined univariate EUR-GWAS, CHIIMGEN EAS -GWAS and ABCD AFR -GWAS. CQ-test was conducted for variants included in all three GWASs. We defined ancestry-shared associations as variant-trait associations with P ≥ 0.05 in CQ-test and ancestry-specific associations as those with P c < 0.05 (Bonferroni correction for the number of tested associations) in CQ-test. CQ-test was also used to identify sex-shared and sex-specific associations in EAS, AFR, and EUR, respectively. With LD EAS , we pooled the variant-trait associations from sex-stratified univariate CHIIMGEN EAS -GWASs. The CQ-test was used to test the allelic-effect heterogeneity for the pooled associations between females and males. The same procedures were then applied to sex-stratified univariate EUR-GWASs, and ABCD AFR -GWAS. We defined variant-trait associations with P ≥ 0.05 in CQ-test as sex-shared associations and those with P c < 0.05 (Bonferroni correction for the number of tested associations) in CQ-test as sex-specific associations. Functional annotations Genomic location and functional consequence. For each GWAS, we used ANNOVAR 44 to categorize the variants from significant variant-trait associations based on their genic position, such as exon, intron, untranslated region, and intergenic region. We used the combined annotation-dependent depletion (CADD) score to prioritize deleterious and pathogenic variants with scores above 20 45 . We also employed the RegulomeDB score to prioritize the variants in regulatory elements 46 . Statistical fine-mapping. We employed the matched LD reference to perform statistical fine-mapping to identify the 95% credible set and the causal variants with posterior probability (PP) above 0.8 for each locus of significant locus-trait associations ( P < 1.61 × 10 − 9 ). We employed SuSiE-R 41 , 42 to conduct single-ancestry fine-mapping based on EUR-GWASs, while we utilized SuSiEx 43 to performed two-ancestry fine-mapping based on EUR-EAS cross-ancestry GWASs and three-ancestry fine-mapping based on EUR-EAS-AFR cross-ancestry GWASs. To compare the performance of the three fine-mapping strategies, we applied these methods to all locus-trait associations identified by any of the three GWASs. We aligned the 95% credible sets with one causal variant assumption across the three fine-mapping strategies, and used the Wilcoxon rank-sum test to compare the size and max PP of 95% credible sets. Colocalization with gene expression. Coloc 50 ( https://chr1swallace.github.io/coloc/ ) was used to perform Bayesian colocalization to identify the shared loci between those associated with cerebellar volumetric traits in EUR-GWAS and expression quantitative trait loci (eQTLs; P < 5 × 10 − 8 ) of human cerebellar tissue. With the default priors ( P 1 = 1 × 10 − 4 , P 2 = 1 × 10 − 4 , and P 12 = 1 × 10 − 5 ), we considered genetic colocalization when PP.H4 (the posterior probability of shared causal variant) was greater than 0.8. The cis-eQTL data of cerebellar tissue (715 tissue samples from 492 EUR individuals) were downloaded from MetaBrain 51 ( https://metabrain.nl/ ). We defined prioritized genes as those colocalized with cerebellar volumetric traits (PP.H4 > 0.8). Pathway enrichment. We performed the pathway enrichment analyses by inputting the prioritized genes to g:Profiler website ( https://biit.cs.ut.ee/gprofiler/gost ) 56 . From the pre-specified GO pathways (15,472 biological processes) 57 , we identified the significant pathways at P c < 0.05 (Benjamini-Hochberg FDR corrected). Genetic clustering of cerebellar substructures To cluster the 27 non-overlapping cerebellar substructures, we used the high-definition likelihood (HDL) method 58 to compute genetic correlations of these substructures based on EUR-GWASs. HDL is an extension of LDSC that can effectively reduce the variance of estimates, thereby increasing precision. We used the imputed HapMap3 reference panel from UKBB after excluding the MHC region. Based on the genetic correlation matrix, we performed genomic structural equation modelling (gSEM) 59 . First, we used the common factor model to test whether a single factor was sufficient for genetic clustering. Second, we tried an anatomical parcellation scheme. If these two models cannot achieve an acceptable fit, we used the nScree function of the nFactor R package to conduct exploratory factor analyses (EFA) to identify the optimal number of factors. The resulting factor model from EFA was validated using confirmatory factor analysis (CFA). The model fit was assessed using comparative fit index (CFI) and standardized root-mean-squared residual (SRMR). An acceptable model fit is defined as CFI > 0.9 and SRMR 0.95 and SRMR < 0.05 59 . Genetic relationships with other phenotypes To investigate the genetic relationships between cerebellar volumetric traits and other cerebellum-related phenotypes, we performed three types of analyses: (1) we conducted the phenome-wide association study (PheWAS) to identify the associations of the PGS of each cerebellar volumetric trait with 198 cognitive and mental health phenotypes; (2) we investigated the genetic correlations between 31 cerebellar volumetric traits and ten neuropsychiatric disorders (NPDs); and (3) we calculated the genetic colocalizations between 31 cerebellar volumetric traits and ten NPDs. PheWAS. We used the PHESANT package 60 in R to conduct PheWAS to identify associations ( P < 0.05/31/198 = 8.15 × 10 − 6 , Bonferroni corrected) between PGSs of 31 cerebellar volumetric traits and 198 cognitive and mental health phenotypes ( table S38 ) 94 , while controlling for age, sex, the first 40 genetic PCs, genotype batch, and assessment centers. The base dataset comprised EUR participants included in EUR-GWASs for cerebellar volumetric traits, while the target dataset consisted of 371,558 unrelated UKBB Caucasians not included in EUR-GWASs. Based on the EUR-GWAS summary statistics of cerebellar volumetric traits and UKBB EUR -LD reference, we used PRS-CS 95 to calculate the PGSs of cerebellar volumetric traits for each participant in target datasets. PRS-CS used a Bayesian regression framework and used a continuous shrinkage (CS) prior on effect size of genetic variant, which is robust to varying genetic architectures. Genetic correlation. HDL was used to calculate genetic correlations ( P < 0.05/31/10 = 1.61 × 10 − 4 , Bonferroni corrected) between 31 cerebellar volumetric traits and 10 NPDs based on EUR-GWAS summary statistics ( table S40 ) and the UKBB reference panel (excluding the MHC region) provided by HDL 58 . Genetic colocalization. Coloc 50 ( https://chr1swallace.github.io/coloc/ ) was used to identify the loci shared by locus-trait associations ( P < 1.61 × 10 − 9 ) of each cerebellar volumetric trait and locus-disease associations ( P < 5 × 10 − 8 ) of each NPD based on the EUR-GWAS summary statistics. With the default priors ( P 1 = 1 × 10 − 4 , P 2 = 1 × 10 − 4 , and P 12 = 1 × 10 − 5 ), we considered colocalization if PP.H4 was greater than 0.8. For the locus-trait associations of cerebellar volumetric traits that were colocalized with both cerebellar gene expression and NPDs (PP.H4 > 0.8), we further performed multi-trait colocalization using hypothesis prioritization in multi-trait colocalization (HyPrColoc) 64 to identify the colocalization (PP > 0.8) of the three traits. Declarations Data availability: All GWAS summary statistics in this study are available. GWAS summary statistics will be deposited in the GWAS catalog. Code availability: This paper does not report original code. We made use of publicly available software and tools in this study. All relevant software and code are described in the text and can be found at references cited. All codes used to generate results are publicly available (https://github.com/xuehui2014/Deep-learning-segmentation-and-multi-ancestry-GWAS-enhance-genetic-discovery-of-the-cerebellum). Acknowledgements: We are grateful to participants and researchers of CHIMGEN, UKBB and ABCD, who generously donated their time to make this resource available. We acknowledge funding from the National Natural Science Foundation of China (82430063, 82030053 to C.Y., 82402253 to J. F., 8247052 to W. Q.). Author contributions C. Y., H. X., and J. F. designed the study and wrote the manuscript. H. X., J. F., H. D., S. W. analyzed the data. C. Y., W. Q., M. L. supervised this work. All authors critically reviewed the manuscript. Competing interests Authors declare that they have no competing interests. References Wang, V. Y. & Zoghbi, H. Y. Genetic regulation of cerebellar development. Nature Reviews Neuroscience 2 , 484-491, doi:Doi 10.1038/35081558 (2001). Sathyanesan, A. et al. Emerging connections between cerebellar development, behaviour and complex brain disorders. Nature Reviews Neuroscience 20 , 298-313, doi:10.1038/s41583-019-0152-2 (2019). Glickstein, M. What does the cerebellum really do? Curr Biol 17 , R824-R827, doi:DOI 10.1016/j.cub.2007.08.009 (2007). Strick, P. L., Dum, R. P. & Fiez, J. A. Cerebellum and Nonmotor Function. Annual Review of Neuroscience 32 , 413-434, doi:10.1146/annurev.neuro.31.060407.125606 (2009). Buckner, R. L. The Cerebellum and Cognitive Function: 25 Years of Insight from Anatomy and Neuroimaging. Neuron 80 , 807-815, doi:10.1016/j.neuron.2013.10.044 (2013). Stoodley, C. J. et al. Altered cerebellar connectivity in autism and cerebellar-mediated rescue of autism-related behaviors in mice. Nat Neurosci 20 , 1744-1751, doi:10.1038/s41593-017-0004-1 (2017). Seidman, L. J., Valera, E. M. & Makris, N. Structural brain imaging of attention-deficit/hyperactivity disorder. Biol Psychiatry 57 , 1263-1272, doi:10.1016/j.biopsych.2004.11.019 (2005). Picard, H., Amado, I., Mouchet-Mages, S., Olié, J. P. & Krebs, M. O. The role of the cerebellum in schizophrenia: an update of clinical, cognitive, and functional evidences. Schizophrenia Bull 34 , 155-172, doi:10.1093/schbul/sbm049 (2008). Diedrichsen, J., Balsters, J. H., Flavell, J., Cussans, E. & Ramnani, N. A probabilistic MR atlas of the human cerebellum. Neuroimage 46 , 39-46, doi:10.1016/j.neuroimage.2009.01.045 (2009). Fischl, B. et al. Whole brain segmentation: Automated labeling of neuroanatomical structures in the human brain. Neuron 33 , 341-355, doi:Doi 10.1016/S0896-6273(02)00569-X (2002). Huggins, A. A. et al. Smaller total and subregional cerebellar volumes in posttraumatic stress disorder: a mega-analysis by the ENIGMA-PGC PTSD workgroup. Molecular Psychiatry 29 , 611-623, doi:10.1038/s41380-023-02352-0 (2024). Fu, X., Chen, Y., Luo, X., Ide, J. S. & Li, C. R. Gray matter volumetric correlates of the polygenic risk of depression: A study of the Human Connectome Project data. Eur Neuropsychopharmacol 87 , 2-12, doi:10.1016/j.euroneuro.2024.06.004 (2024). Mooney, M. A. et al. Smaller total brain volume but not subcortical structure volume related to common genetic risk for ADHD. Psychol Med 51 , 1279-1288, doi:10.1017/s0033291719004148 (2021). Kerestes, R. et al. Cerebellar Volume and Disease Staging in Parkinson's Disease: An ENIGMA-PD Study. Movement Disord 38 , 2269-2281, doi:10.1002/mds.29611 (2023). Gellersen, H. M., Guell, X. & Sami, S. Differential vulnerability of the cerebellum in healthy ageing and Alzheimer's disease. Neuroimage Clin 30 , 102605, doi:10.1016/j.nicl.2021.102605 (2021). Qiu, T. et al. Precentral degeneration and cerebellar compensation in amyotrophic lateral sclerosis: A multimodal MRI analysis. Human Brain Mapping 40 , 3464-3474, doi:10.1002/hbm.24609 (2019). Strike, L. T. et al. Heritability of cerebellar subregion volumes in adolescent and young adult twins. Human Brain Mapping 45 , doi:10.1002/hbm.26717 (2024). Chambers, T. et al. Genetic common variants associated with cerebellar volume and their overlap with mental disorders: a study on 33,265 individuals from the UK-Biobank. Molecular Psychiatry , doi:10.1038/s41380-022-01443-8 (2022). Tissink, E. et al. Genome-wide association study of cerebellar volume provides insights into heritable mechanisms underlying brain development and mental health. Communications Biology 5 , 710, doi:10.1038/s42003-022-03672-7 (2022). Smith, S. M. et al. An expanded set of genome-wide association studies of brain imaging phenotypes in UK Biobank. Nature Neuroscience 24 , 737-745, doi:10.1038/s41593-021-00826-4 (2021). Fu, J. L. et al. Cross-ancestry genome-wide association studies of brain imaging phenotypes. Nature Genetics , doi:10.1038/s41588-024-01766-y (2024). Faber, J. et al. CerebNet: A fast and reliable deep-learning pipeline for detailed cerebellum sub-segmentation. Neuroimage 264 , doi:10.1016/j.neuroimage.2022.119703 (2022). Xu, Q. et al. CHIMGEN: a Chinese imaging genetics cohort to enhance cross-ethnic and cross-geographic brain research. Mol Psychiatry 25 , 517-529, doi:10.1038/s41380-019-0627-6 (2020). Miller, K. L. et al. Multimodal population brain imaging in the UK Biobank prospective epidemiological study. Nature Neuroscience 19 , 1523-1536, doi:10.1038/nn.4393 (2016). Casey, B. J. et al. The Adolescent Brain Cognitive Development (ABCD) study: Imaging acquisition across 21 sites. Dev Cogn Neuros-Neth 32 , 43-54, doi:10.1016/j.dcn.2018.03.001 (2018). Yang, J. et al. Genetic variance estimation with imputed variants finds negligible missing heritability for human height and body mass index. Nat Genet 47 , 1114-1120, doi:10.1038/ng.3390 (2015). Jiang, L. D. et al. A resource-efficient tool for mixed model association analysis of large-scale data. Nature Genetics 51 , 1749-+, doi:10.1038/s41588-019-0530-8 (2019). Choi, S. W. & O'Reilly, P. F. PRSice-2: Polygenic Risk Score software for biobank-scale data. Gigascience 8 , doi:10.1093/gigascience/giz082 (2019). Bulik-Sullivan, B. K. et al. LD Score regression distinguishes confounding from polygenicity in genome-wide association studies. Nature Genetics 47 , 291-+, doi:10.1038/ng.3211 (2015). Chen, X. et al. Transcriptomic mapping uncovers Purkinje neuron plasticity driving learning. Nature 605 , 722-727, doi:10.1038/s41586-022-04711-3 (2022). Willer, C. J., Li, Y. & Abecasis, G. R. METAL: fast and efficient meta-analysis of genomewide association scans. Bioinformatics 26 , 2190-2191, doi:10.1093/bioinformatics/btq340 (2010). Van Battum, E. et al. Plexin-B2 controls the timing of differentiation and the motility of cerebellar granule neurons. Elife 10 , doi:10.7554/eLife.60554 (2021). Sepp, M. et al. Cellular development and evolution of the mammalian cerebellum. Nature 625 , doi:10.1038/s41586-023-06884-x (2024). Xiong, Z. et al. Combining genome-wide association studies highlight novel loci involved in human facial variation. Nat Commun 13 , 7832, doi:10.1038/s41467-022-35328-9 (2022). Das, G. et al. EphA5 and EphA6: regulation of neuronal and spine morphology. Cell Biosci 6 , 48, doi:10.1186/s13578-016-0115-5 (2016). Caballero Oteyza, A. et al. Motor protein mutations cause a new form of hereditary spastic paraplegia. Neurology 82 , 2007-2016, doi:10.1212/wnl.0000000000000479 (2014). Siddiqui, N. et al. Force generation of KIF1C is impaired by pathogenic mutations. Curr Biol 32 , 3862-3870.e3866, doi:10.1016/j.cub.2022.07.029 (2022). Ramirez, M. et al. Temporal analysis of enhancers during mouse cerebellar development reveals dynamic and novel regulatory functions. Elife 11 , doi:10.7554/eLife.74207 (2022). De Luca, A. et al. Sonic hedgehog patterning during cerebellar development. Cell Mol Life Sci 73 , 291-303, doi:10.1007/s00018-015-2065-1 (2016). Kyöstilä, K. et al. A SEL1L mutation links a canine progressive early-onset cerebellar ataxia to the endoplasmic reticulum-associated protein degradation (ERAD) machinery. PLoS Genet 8 , e1002759, doi:10.1371/journal.pgen.1002759 (2012). Wang, G., Sarkar, A., Carbonetto, P. & Stephens, M. A simple new approach to variable selection in regression, with application to genetic fine mapping. J R Stat Soc Series B Stat Methodol 82 , 1273-1300, doi:10.1111/rssb.12388 (2020). Zou, Y., Carbonetto, P., Wang, G. & Stephens, M. Fine-mapping from summary data with the "Sum of Single Effects" model. Plos Genetics 18 , doi:10.1371/journal.pgen.1010299 (2022). Yuan, K. et al. Fine-mapping across diverse ancestries drives the discovery of putative causal variants underlying human complex traits and diseases. Nature Genetics , doi:10.1038/s41588-024-01870-z (2024). Wang, K., Li, M. & Hakonarson, H. ANNOVAR: functional annotation of genetic variants from high-throughput sequencing data. Nucleic Acids Res 38 , e164, doi:10.1093/nar/gkq603 (2010). Rentzsch, P., Witten, D., Cooper, G. M., Shendure, J. & Kircher, M. CADD: predicting the deleteriousness of variants throughout the human genome. Nucleic Acids Research 47 , D886-D894, doi:10.1093/nar/gky1016 (2019). Dong, S. C. et al. Annotating and prioritizing human non-coding variants with RegulomeDB v.2. Nature Genetics 55 , 724-726, doi:10.1038/s41588-023-01365-3 (2023). Zaninello, M. et al. CLUH maintains functional mitochondria and translation in motoneuronal axons and prevents peripheral neuropathy. Science Advances 10 , doi:10.1126/sciadv.adn2050 (2024). Wang, B. P., Harrison, W., Overbeek, P. A. & Zheng, H. Transposon mutagenesis with coat color genotyping identifies an essential role for Skor2 in sonic hedgehog signaling and cerebellum development. Development 138 , 4487-4497, doi:10.1242/dev.067264 (2011). Adams, N. C., Tomoda, T., Cooper, M., Dietz, G. & Hatten, M. E. Mice that lack astrotactin have slowed neuronal migration. Development 129 , 965-972, doi:10.1242/dev.129.4.965 (2002). Wallace, C. Eliciting priors and relaxing the single causal variant assumption in colocalisation analyses. Plos Genetics 16 , doi:10.1371/journal.pgen.1008720 (2020). de Klein, N. et al. Brain expression quantitative trait locus and network analyses reveal downstream effects and putative drivers for brain-related diseases. Nature Genetics 55 , 377-+, doi:10.1038/s41588-023-01300-6 (2023). Okbay, A. et al. Polygenic prediction of educational attainment within and between families from genome-wide association analyses in 3 million individuals. Nature Genetics 54 , 437-+, doi:10.1038/s41588-022-01016-z (2022). Dahl, A. et al. Phenotype integration improves power and preserves specificity in biobank-based genetic studies of major depressive disorder. Nature Genetics 55 , 2082-2093, doi:10.1038/s41588-023-01559-9 (2023). Rico, B. et al. Control of axonal branching and synapse formation by focal adhesion kinase. Nature Neuroscience 7 , 1059-1069, doi:10.1038/nn1317 (2004). Di Gregorio, E. et al. A de novo X;8 translocation creates a PTK2-THOC2 gene fusion with THOC2 expression knockdown in a patient with psychomotor retardation and congenital cerebellar hypoplasia. Journal of Medical Genetics 50 , 543-551, doi:10.1136/jmedgenet-2013-101542 (2013). Kolberg, L. et al. g:Profiler-interoperable web service for functional enrichment analysis and gene identifier mapping (2023 update). Nucleic Acids Research 51 , W207-W212, doi:10.1093/nar/gkad347 (2023). Ashburner, M. et al. Gene Ontology: tool for the unification of biology. Nature Genetics 25 , 25-29, doi:Doi 10.1038/75556 (2000). Ning, Z., Pawitan, Y. & Shen, X. High-definition likelihood inference of genetic correlations across human complex traits. Nat Genet 52 , 859-864, doi:10.1038/s41588-020-0653-y (2020). Grotzinger, A. D. et al. Genomic structural equation modelling provides insights into the multivariate genetic architecture of complex traits. Nat Hum Behav 3 , 513-525, doi:10.1038/s41562-019-0566-x (2019). Millard, L. A. C., Davies, N. M., Gaunt, T. R., Smith, G. D. & Tilling, K. Software Application Profile: PHESANT: a tool for performing automated phenome scans in UK Biobank. Int J Epidemiol 47 , 29-35, doi:10.1093/ije/dyx204 (2018). Molinari, M. et al. Cerebellum and procedural learning: evidence from focal cerebellar lesions. Brain 120 ( Pt 10) , 1753-1762, doi:10.1093/brain/120.10.1753 (1997). Levisohn, L., Cronin-Golomb, A. & Schmahmann, J. D. Neuropsychological consequences of cerebellar tumour resection in children - Cerebellar cognitive affective syndrome in a paediatric population. Brain 123 , 1041-1050, doi:DOI 10.1093/brain/123.5.1041 (2000). Sullivan, E. V. et al. Disturbed Cerebellar Growth Trajectories in Adolescents Who Initiate Alcohol Drinking. Biol Psychiatry 87 , 632-644, doi:10.1016/j.biopsych.2019.08.026 (2020). Foley, C. N. et al. A fast and efficient colocalization algorithm for identifying shared genetic risk factors across multiple traits. Nature Communications 12 , doi:10.1038/s41467-020-20885-8 (2021). Vieira, N., Rito, T., Correia-Neves, M. & Sousa, N. Sorting Out Sorting Nexins Functions in the Nervous System in Health and Disease. Mol Neurobiol 58 , 4070-4106, doi:10.1007/s12035-021-02388-9 (2021). Manolio, T. A. et al. Finding the missing heritability of complex diseases. Nature 461 , 747-753, doi:10.1038/nature08494 (2009). Bahi-Buisson, N. et al. GPR56-related bilateral frontoparietal polymicrogyria: further evidence for an overlap with the cobblestone complex. Brain 133 , 3194-3209, doi:10.1093/brain/awq259 (2010). Peterson, R. E. et al. Genome-wide Association Studies in Ancestrally Diverse Populations: Opportunities, Methods, Pitfalls, and Recommendations. Cell 179 , 589-603, doi:10.1016/j.cell.2019.08.051 (2019). Schöls, L., Bauer, P., Schmidt, T., Schulte, T. & Riess, O. Autosomal dominant cerebellar ataxias: clinical features, genetics, and pathogenesis. Lancet Neurol 3 , 291-304, doi:10.1016/s1474-4422(04)00737-9 (2004). Steele, C. J. & Chakravarty, M. M. Gray-matter structural variability in the human cerebellum: Lobule-specific differences across sex and hemisphere. Neuroimage 170 , 164-173, doi:10.1016/j.neuroimage.2017.04.066 (2018). Coste de Bagneaux, P. et al. A homozygous missense variant in CACNB4 encoding the auxiliary calcium channel beta4 subunit causes a severe neurodevelopmental disorder and impairs channel and non-channel functions. PLoS Genet 16 , e1008625, doi:10.1371/journal.pgen.1008625 (2020). Nakatani, H., Nakamura, Y. & Okanoya, K. Respective Involvement of the Right Cerebellar Crus I and II in Syntactic and Semantic Processing for Comprehension of Language. Cerebellum 22 , 739-755, doi:10.1007/s12311-022-01451-y (2023). Chen, G. et al. Abnormal cerebellum-DMN regions connectivity in unmedicated bipolar II disorder. J Affect Disord 243 , 441-447, doi:10.1016/j.jad.2018.09.076 (2019). Bruchhage, M. M. K. et al. Drum training induces long-term plasticity in the cerebellum and connected cortical thickness. Sci Rep 10 , 10116, doi:10.1038/s41598-020-65877-2 (2020). Brissenden, J. A., Tobyne, S. M., Halko, M. A. & Somers, D. C. Stimulus-Specific Visual Working Memory Representations in Human Cerebellar Lobule VIIb/VIIIa. J Neurosci 41 , 1033-1045, doi:10.1523/jneurosci.1253-20.2020 (2021). Cierpka, M. et al. Cerebellar Contributions to Persistent Auditory Verbal Hallucinations in Patients with Schizophrenia. Cerebellum 16 , 964-972, doi:10.1007/s12311-017-0874-5 (2017). Wang, Y. et al. Abnormalities in cerebellar subregions' volume and cerebellocerebral structural covariance in autism spectrum disorder. Autism Res 18 , 83-97, doi:10.1002/aur.3287 (2025). Laidi, C. et al. Cerebellar parcellation in schizophrenia and bipolar disorder. Acta Psychiatr Scand 140 , 468-476, doi:10.1111/acps.13087 (2019). Marneweck, M., Barany, D. A., Santello, M. & Grafton, S. T. Neural Representations of Sensorimotor Memory- and Digit Position-Based Load Force Adjustments Before the Onset of Dexterous Object Manipulation. J Neurosci 38 , 4724-4737, doi:10.1523/jneurosci.2588-17.2018 (2018). Wang, X. et al. Revisiting reward impairments in schizophrenia spectrum disorders: a systematic review and meta-analysis for neuroimaging findings. Psychol Med 53 , 7189-7202, doi:10.1017/s0033291723000703 (2023). Ernst, T. M. et al. The cerebellum is involved in processing of predictions and prediction errors in a fear conditioning paradigm. Elife 8 , doi:10.7554/eLife.46831 (2019). Peterburs, J., Blevins, L. C., Sheu, Y. S. & Desmond, J. E. Cerebellar contributions to sequence prediction in verbal working memory. Brain Struct Funct 224 , 485-499, doi:10.1007/s00429-018-1784-0 (2019). Stankiewicz, T. R. & Linseman, D. A. Rho family GTPases: key players in neuronal development neuronal survival, and neurodegeneration. Front Cell Neurosci 8 , doi:10.3389/fncel.2014.00314 (2014). Bycroft, C. et al. The UK Biobank resource with deep phenotyping and genomic data. Nature 562 , 203-+, doi:10.1038/s41586-018-0579-z (2018). Auchter, A. M. et al. A description of the ABCD organizational structure and communication framework. Dev Cogn Neuros-Neth 32 , 8-15, doi:10.1016/j.dcn.2018.04.003 (2018). Chen, C. Y. et al. Improved ancestry inference using weights from external reference panels. Bioinformatics 29 , 1399-1406, doi:10.1093/bioinformatics/btt144 (2013). Mukamel, R. E. et al. Repeat polymorphisms underlie top genetic risk loci for glaucoma and colorectal cancer. Cell 186 , doi:10.1016/j.cell.2023.07.002 (2023). Nievergelt, C. M. et al. Genome-wide association analyses identify 95 risk loci and provide insights into the neurobiology of post-traumatic stress disorder. Nature Genetics 56 , doi:10.1038/s41588-024-01707-9 (2024). Alfaro-Almagro, F. et al. Image processing and Quality Control for the first 10,000 brain imaging datasets from UK Biobank. Neuroimage 166 , 400-424, doi:10.1016/j.neuroimage.2017.10.034 (2018). Rajapakse, J. C., Giedd, J. N. & Rapoport, J. L. Statistical approach to segmentation of single-channel cerebral MR images. IEEE Trans Med Imaging 16 , 176-186, doi:10.1109/42.563663 (1997). Ashburner, J. A fast diffeomorphic image registration algorithm. Neuroimage 38 , 95-113, doi:10.1016/j.neuroimage.2007.07.007 (2007). Fortin, J. P. et al. Harmonization of cortical thickness measurements across scanners and sites. Neuroimage 167 , 104-120, doi:10.1016/j.neuroimage.2017.11.024 (2018). Purcell, S. et al. PLINK: A tool set for whole-genome association and population-based linkage analyses. American Journal of Human Genetics 81 , 559-575, doi:10.1086/519795 (2007). Shen, X. et al. A phenome-wide association and Mendelian Randomisation study of polygenic risk for depression in UK Biobank. Nat Commun 11 , 2301, doi:10.1038/s41467-020-16022-0 (2020). Ge, T., Chen, C. Y., Ni, Y., Feng, Y. C. A. & Smoller, J. W. Polygenic prediction via Bayesian regression and continuous shrinkage priors. Nature Communications 10 , doi:10.1038/s41467-019-09718-5 (2019). Additional Declarations There is NO Competing Interest. Supplementary Files SupplExcel.xlsx Supplementary Tables SupplementalMaterial.pdf Supplementary Figures Cite Share Download PDF Status: Under Review Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6694135","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":462729672,"identity":"79dd4828-3a81-4e6e-afb9-2d85ce07dd18","order_by":0,"name":"Chunshui Yu","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA00lEQVRIiWNgGAWjYDACCSBmbGDgYWBgPgAROUC8FrbEBpK0AAGPIXFa+Gf3mD38ueOwjMGNnO+PbrYxyPHdSGD8XIDPkjtnzI15zxzmMbiRu7E5t43BWPJGArP0DDxaDCRyzKQZ2xBaEjfcSGBj5iGgRfInWEvOQ5CWeqK0SPBCtDCCtCQYENIicSOtTJq3LZ1H8swzw9k55yQMZ5552CyNTwv/jORtQIdZ2/MdT37wOafMRh7IOPgZnxY4UDgAsZUBFk2EgTyR6kbBKBgFo2AEAgCXqEy+lZ+xdAAAAABJRU5ErkJggg==","orcid":"https://orcid.org/0000-0001-5648-5199","institution":"Tianjin Medical University General Hospital","correspondingAuthor":true,"prefix":"","firstName":"Chunshui","middleName":"","lastName":"Yu","suffix":""},{"id":462729673,"identity":"98c77d9a-6009-4d4d-aef6-8443e4ad6a46","order_by":1,"name":"Hui Xue","email":"","orcid":"","institution":"Tianjin Medical University General Hospital","correspondingAuthor":false,"prefix":"","firstName":"Hui","middleName":"","lastName":"Xue","suffix":""},{"id":462729674,"identity":"890fe806-4ea6-47f7-a805-9cbaaf7db24e","order_by":2,"name":"Jilian Fu","email":"","orcid":"","institution":"Tianjin Medical University General Hospital","correspondingAuthor":false,"prefix":"","firstName":"Jilian","middleName":"","lastName":"Fu","suffix":""},{"id":462729675,"identity":"7f4f814e-7b07-485b-8938-1f637d7d3967","order_by":3,"name":"Hao Ding","email":"","orcid":"","institution":"Tianjin Medical University","correspondingAuthor":false,"prefix":"","firstName":"Hao","middleName":"","lastName":"Ding","suffix":""},{"id":462729676,"identity":"0544b301-b752-4b0f-85d6-58ca5bc36ef7","order_by":4,"name":"Sijia Wang","email":"","orcid":"https://orcid.org/0000-0001-8949-303X","institution":"Tianjin Medical University General Hospital","correspondingAuthor":false,"prefix":"","firstName":"Sijia","middleName":"","lastName":"Wang","suffix":""},{"id":462729677,"identity":"9ef92c30-234f-4d73-aa7f-d369a8d72179","order_by":5,"name":"Jingliang Cheng","email":"","orcid":"https://orcid.org/0000-0001-9944-8419","institution":"The First Affiliated Hospital of Zhengzhou University","correspondingAuthor":false,"prefix":"","firstName":"Jingliang","middleName":"","lastName":"Cheng","suffix":""},{"id":462729678,"identity":"1ddc0cd8-d1fe-447f-9ca2-a7052024bf2d","order_by":6,"name":"Meiyun Wang","email":"","orcid":"https://orcid.org/0000-0003-4454-5005","institution":"Henan Provincial People's Hospital \u0026 the People's Hospital of Zhengzhou University","correspondingAuthor":false,"prefix":"","firstName":"Meiyun","middleName":"","lastName":"Wang","suffix":""},{"id":462729679,"identity":"a1796186-f8bd-4689-a148-6ae80ac481e7","order_by":7,"name":"Long Jiang Zhang","email":"","orcid":"https://orcid.org/0000-0002-6664-7224","institution":"Jinling hospital, Affiliated Hospital of Medical School, Nanjing University","correspondingAuthor":false,"prefix":"","firstName":"Long","middleName":"Jiang","lastName":"Zhang","suffix":""},{"id":462729680,"identity":"58c2461c-8e29-4b1e-b712-f56f9831fe31","order_by":8,"name":"Zuojun Geng","email":"","orcid":"","institution":"The Second Hospital of Hebei Medical University","correspondingAuthor":false,"prefix":"","firstName":"Zuojun","middleName":"","lastName":"Geng","suffix":""},{"id":462729681,"identity":"974d4a89-2587-43f8-93cd-ad6bac42b01f","order_by":9,"name":"Guang-Bin Cui","email":"","orcid":"","institution":"Tangdu Hospital, Fourth Military Medical University (Air Force Medical University)","correspondingAuthor":false,"prefix":"","firstName":"Guang-Bin","middleName":"","lastName":"Cui","suffix":""},{"id":462729682,"identity":"2fb92d58-6665-413e-b34b-df380d4c3441","order_by":10,"name":"Yongqiang Yu","email":"","orcid":"","institution":"The First Affiliated Hospital of Anhui Medical University","correspondingAuthor":false,"prefix":"","firstName":"Yongqiang","middleName":"","lastName":"Yu","suffix":""},{"id":462729683,"identity":"08ed9782-6438-432f-9575-4d587270c6e2","order_by":11,"name":"Weihua Liao","email":"","orcid":"","institution":"Xiangya Hospital, Central South University","correspondingAuthor":false,"prefix":"","firstName":"Weihua","middleName":"","lastName":"Liao","suffix":""},{"id":462729684,"identity":"50b42188-e4f2-48af-9846-9ef5f8aae500","order_by":12,"name":"Hui Zhang","email":"","orcid":"","institution":"The First Hospital of Shanxi Medical University","correspondingAuthor":false,"prefix":"","firstName":"Hui","middleName":"","lastName":"Zhang","suffix":""},{"id":462729685,"identity":"ec2f242f-6862-4e4b-b322-69784b288feb","order_by":13,"name":"Bo Gao","email":"","orcid":"","institution":"The Affiliated Hospital of Guizhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Bo","middleName":"","lastName":"Gao","suffix":""},{"id":462729686,"identity":"053f5105-58f1-47b4-8c43-9344187cd001","order_by":14,"name":"Xiaojun Xu","email":"","orcid":"https://orcid.org/0000-0002-0127-2812","institution":"The Second Affiliated Hospital of Zhejiang University, School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Xiaojun","middleName":"","lastName":"Xu","suffix":""},{"id":462729687,"identity":"bcb32745-b466-474d-a6e7-459d81060f17","order_by":15,"name":"Tong Han","email":"","orcid":"","institution":"Department of Neurosurgery, Huanhu Hospital","correspondingAuthor":false,"prefix":"","firstName":"Tong","middleName":"","lastName":"Han","suffix":""},{"id":462729688,"identity":"7b0e4e37-029c-47b2-a48a-4b8c4322426a","order_by":16,"name":"Zhenwei Yao","email":"","orcid":"","institution":"Huashan Hospital, Fudan University","correspondingAuthor":false,"prefix":"","firstName":"Zhenwei","middleName":"","lastName":"Yao","suffix":""},{"id":462729689,"identity":"d14c95c0-f549-460a-ad45-a1c5a6ccb165","order_by":17,"name":"Quan Zhang","email":"","orcid":"","institution":"Characteristic Medical Center of Chinese People's Armed Police Force","correspondingAuthor":false,"prefix":"","firstName":"Quan","middleName":"","lastName":"Zhang","suffix":""},{"id":462729690,"identity":"c02a197a-757b-4e0a-b1a2-7651511f4819","order_by":18,"name":"Feng Liu","email":"","orcid":"https://orcid.org/0000-0002-3570-4222","institution":"Tianjin Medical University General Hospital","correspondingAuthor":false,"prefix":"","firstName":"Feng","middleName":"","lastName":"Liu","suffix":""},{"id":462729691,"identity":"9a5025d8-b17b-4f42-ae3c-dbccfce460bd","order_by":19,"name":"Nana Liu","email":"","orcid":"","institution":"Tianjin Medical University General Hospital","correspondingAuthor":false,"prefix":"","firstName":"Nana","middleName":"","lastName":"Liu","suffix":""},{"id":462729692,"identity":"a54cd115-7355-4c7f-b378-19d6c4cb5b38","order_by":20,"name":"Qiang Xu","email":"","orcid":"","institution":"Tianjin Medical University General Hospital","correspondingAuthor":false,"prefix":"","firstName":"Qiang","middleName":"","lastName":"Xu","suffix":""},{"id":462729693,"identity":"594cb104-29f1-4a8a-b4eb-dfd21ff1edf3","order_by":21,"name":"Jiayuan Xu","email":"","orcid":"https://orcid.org/0000-0001-9473-1047","institution":"Department of Radiology and Tianjin Key Laboratory of Functional Imaging, Tianjin Medical University General Hospital","correspondingAuthor":false,"prefix":"","firstName":"Jiayuan","middleName":"","lastName":"Xu","suffix":""},{"id":462729694,"identity":"ab036eeb-abf2-48b8-8496-6dd7dc3a75d8","order_by":22,"name":"Yuan Ji","email":"","orcid":"https://orcid.org/0000-0003-2874-5904","institution":"Tianjin Medical University General Hospital","correspondingAuthor":false,"prefix":"","firstName":"Yuan","middleName":"","lastName":"Ji","suffix":""},{"id":462729695,"identity":"c7f8e7f7-8ac2-4672-a26d-bbcedee3998e","order_by":23,"name":"Caihong Wang","email":"","orcid":"","institution":"The First Affiliated Hospital of Zhengzhou University","correspondingAuthor":false,"prefix":"","firstName":"Caihong","middleName":"","lastName":"Wang","suffix":""},{"id":462729696,"identity":"185289cc-1f9a-4ff4-a4b9-b0cf625e231d","order_by":24,"name":"Ping Wang","email":"","orcid":"","institution":"School of Medical Imaging and School of Medical Technology, Tianjin Medical University","correspondingAuthor":false,"prefix":"","firstName":"Ping","middleName":"","lastName":"Wang","suffix":""},{"id":462729697,"identity":"a2b13424-b46e-4a61-9ba5-576a785db0ab","order_by":25,"name":"Zhaoxiang Ye","email":"","orcid":"https://orcid.org/0000-0003-3157-8393","institution":"Tianjin Medical University Cancer Institute and Hospital","correspondingAuthor":false,"prefix":"","firstName":"Zhaoxiang","middleName":"","lastName":"Ye","suffix":""},{"id":462729698,"identity":"5c406f76-6276-495f-9c9b-75827ae113a2","order_by":26,"name":"Peng Zhang","email":"","orcid":"","institution":"Tianjin Medical University Cancer Institute and Hospital","correspondingAuthor":false,"prefix":"","firstName":"Peng","middleName":"","lastName":"Zhang","suffix":""},{"id":462729699,"identity":"8c2149bd-7c15-4446-8ebb-a173e9d9301d","order_by":27,"name":"Wei Li","email":"","orcid":"","institution":"Tianjin Medical University Cancer Institute and Hospital","correspondingAuthor":false,"prefix":"","firstName":"Wei","middleName":"","lastName":"Li","suffix":""},{"id":462729700,"identity":"105b2880-ac25-4973-a2a0-a777ce451714","order_by":28,"name":"Wei Wei","email":"","orcid":"","institution":"Department of Radiology, Henan Provincial People's Hospital \u0026 Zhengzhou University People's Hospital, 450003 Zhengzhou, China","correspondingAuthor":false,"prefix":"","firstName":"Wei","middleName":"","lastName":"Wei","suffix":""},{"id":462729701,"identity":"0196e209-6960-4358-9b07-127dc6531f9d","order_by":29,"name":"Dapeng Shi","email":"","orcid":"","institution":"Henan Provincial People's Hospital \u0026 Zhengzhou University People's Hospital","correspondingAuthor":false,"prefix":"","firstName":"Dapeng","middleName":"","lastName":"Shi","suffix":""},{"id":462729702,"identity":"70c69e4a-601f-447e-9c8e-d0efe78b3be8","order_by":30,"name":"Qiyong Gong","email":"","orcid":"","institution":"Huaxi MR Research Center (HMRRC)","correspondingAuthor":false,"prefix":"","firstName":"Qiyong","middleName":"","lastName":"Gong","suffix":""},{"id":462729703,"identity":"a53ef2db-7e73-4623-bbcb-48969a843ede","order_by":31,"name":"Zhihan Yan","email":"","orcid":"","institution":"The Second Affiliated Hospital and Yuying Children's Hospital of Wenzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Zhihan","middleName":"","lastName":"Yan","suffix":""},{"id":462729704,"identity":"8024e80e-a2fc-455c-9136-31abf0735c70","order_by":32,"name":"Feng Chen","email":"","orcid":"https://orcid.org/0000-0002-9129-7895","institution":"Hainan General Hospital","correspondingAuthor":false,"prefix":"","firstName":"Feng","middleName":"","lastName":"Chen","suffix":""},{"id":462729705,"identity":"3d66bbad-a75e-4bf4-8c2c-b7d150e332c2","order_by":33,"name":"Jing Zhang","email":"","orcid":"","institution":"Lanzhou University Second Hospital","correspondingAuthor":false,"prefix":"","firstName":"Jing","middleName":"","lastName":"Zhang","suffix":""},{"id":462729706,"identity":"99d68535-9089-44fa-bfa8-0a9331b7a12f","order_by":34,"name":"Wen Shen","email":"","orcid":"","institution":"Tianjin First Center Hospital","correspondingAuthor":false,"prefix":"","firstName":"Wen","middleName":"","lastName":"Shen","suffix":""},{"id":462729707,"identity":"9d1934e8-b66b-4411-b3b9-ea81711c3b2a","order_by":35,"name":"Yanwei Miao","email":"","orcid":"","institution":"The First Affiliated Hospital of Dalian Medical University","correspondingAuthor":false,"prefix":"","firstName":"Yanwei","middleName":"","lastName":"Miao","suffix":""},{"id":462729708,"identity":"61776227-d51a-459c-a577-40415d5defe6","order_by":36,"name":"Dawei Wang","email":"","orcid":"","institution":"Qilu Hospital, Shandong University","correspondingAuthor":false,"prefix":"","firstName":"Dawei","middleName":"","lastName":"Wang","suffix":""},{"id":462729709,"identity":"c8f0f08b-1c33-4588-a4eb-d5d7679236e1","order_by":37,"name":"Jia-Hong Gao","email":"","orcid":"https://orcid.org/0000-0002-9311-0297","institution":"National Biomedical Imaging Center","correspondingAuthor":false,"prefix":"","firstName":"Jia-Hong","middleName":"","lastName":"Gao","suffix":""},{"id":462729710,"identity":"9ced082e-e9eb-44a0-9ed5-d85719e67f00","order_by":38,"name":"Yunjun Yang","email":"","orcid":"","institution":"Department of Radiology, The First Affiliated Hospital of Wenzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Yunjun","middleName":"","lastName":"Yang","suffix":""},{"id":462729711,"identity":"c2acdd2f-6dbb-4e43-9bb0-512b96ef9e39","order_by":39,"name":"Junfang Xian","email":"","orcid":"https://orcid.org/0000-0003-2191-9393","institution":"Beijing Tongren Hospital, Capital Medical University","correspondingAuthor":false,"prefix":"","firstName":"Junfang","middleName":"","lastName":"Xian","suffix":""},{"id":462729712,"identity":"cfd4fe4d-c184-465c-919e-ffabc7ebac66","order_by":40,"name":"Xiaochu Zhang","email":"","orcid":"https://orcid.org/0000-0002-7541-0130","institution":"University of Science and Technology of China","correspondingAuthor":false,"prefix":"","firstName":"Xiaochu","middleName":"","lastName":"Zhang","suffix":""},{"id":462729713,"identity":"b5c0d6d2-96ea-4c72-a16a-fb820f48bc44","order_by":41,"name":"Xi-Nian Zuo","email":"","orcid":"https://orcid.org/0000-0001-9110-585X","institution":"Beijing Normal University","correspondingAuthor":false,"prefix":"","firstName":"Xi-Nian","middleName":"","lastName":"Zuo","suffix":""},{"id":462729714,"identity":"882cc8bd-3ac7-4896-9e03-495cb6c7b394","order_by":42,"name":"Mulin Jun Li","email":"","orcid":"https://orcid.org/0000-0003-3598-3679","institution":"School of Basic Medical Sciences, Tianjin Medical University","correspondingAuthor":false,"prefix":"","firstName":"Mulin","middleName":"Jun","lastName":"Li","suffix":""},{"id":462729715,"identity":"8e27e1ad-94ac-4e9f-b721-acb9e17c7baf","order_by":43,"name":"Shijun Qiu","email":"","orcid":"https://orcid.org/0000-0002-6739-4265","institution":"The First Affiliated Hospital of Guangzhou University of Traditional Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Shijun","middleName":"","lastName":"Qiu","suffix":""},{"id":462729716,"identity":"14358f8f-0a58-49f9-8f32-b1b4bc9194b2","order_by":44,"name":"Wenzhen Zhu","email":"","orcid":"https://orcid.org/0000-0001-6252-9450","institution":"Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology","correspondingAuthor":false,"prefix":"","firstName":"Wenzhen","middleName":"","lastName":"Zhu","suffix":""},{"id":462729717,"identity":"2b8ad3b1-59c6-4f61-9c53-91fcdeb35c0b","order_by":45,"name":"Kai Xu","email":"","orcid":"https://orcid.org/0000-0001-5407-5671","institution":"The Affiliated Hospital of Xuzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Kai","middleName":"","lastName":"Xu","suffix":""},{"id":462729718,"identity":"cc272582-8d7b-4b4c-a740-a2628f932ce0","order_by":46,"name":"Bing Zhang","email":"","orcid":"https://orcid.org/0000-0002-3953-0290","institution":"Nanjing Drum Tower Hospital, Affiliated Hospital of Medical School, Nanjing University","correspondingAuthor":false,"prefix":"","firstName":"Bing","middleName":"","lastName":"Zhang","suffix":""},{"id":462729719,"identity":"0d314cd4-1b96-47ba-883e-23ea818366d9","order_by":47,"name":"Meng Liang","email":"","orcid":"","institution":"School of Medical Imaging (School of Medical Technology) and Tianjin Key Laboratory of Functional Imaging, Tianjin Medical University, 300203 Tianjin, China","correspondingAuthor":false,"prefix":"","firstName":"Meng","middleName":"","lastName":"Liang","suffix":""},{"id":462729720,"identity":"5073b042-e5ee-46da-9c09-99d40700387d","order_by":48,"name":"Wen Qin","email":"","orcid":"https://orcid.org/0000-0002-9121-8296","institution":"Tianjin Medical University General Hospital","correspondingAuthor":false,"prefix":"","firstName":"Wen","middleName":"","lastName":"Qin","suffix":""},{"id":462729721,"identity":"cbff6247-e3f3-4950-ba4d-e32939de2bbe","order_by":49,"name":"the CHIMGEN Consortium","email":"","orcid":"","institution":"Tianjin Medical University General Hospital","correspondingAuthor":false,"prefix":"","firstName":"the","middleName":"CHIMGEN","lastName":"Consortium","suffix":""}],"badges":[],"createdAt":"2025-05-19 01:40:23","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6694135/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6694135/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":84198790,"identity":"537d9cf1-5841-4d6d-a851-3d41829b3d2b","added_by":"auto","created_at":"2025-06-09 08:09:31","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":1199327,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eComparison in genetic findings between cerebellar segmentation methods.\u003c/strong\u003e\u003cem\u003e\u003cstrong\u003e \u003c/strong\u003e\u003c/em\u003eWe compare the heritability estimates, genetic discoveries, and PGS predictions of 30 cerebellar substructure volumes derived from two segmentation methods. (\u003cstrong\u003eA\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003eThe scatter diagrams show that the heritability estimates of cerebellar substructure volumes obtained from CerebNet (red dots) are higher than those from SUIT+FS (blue dots) in the UKBB\u003csub\u003eEUR\u003c/sub\u003e, CHIMGEN\u003csub\u003eEAS\u003c/sub\u003e, ABCD\u003csub\u003eEUR\u003c/sub\u003e, and ABCD\u003csub\u003eAFR\u003c/sub\u003e participants, respectively. (\u003cstrong\u003eB\u003c/strong\u003e) The scatter diagram shows that the numbers of significant variant-trait associations (\u003cem\u003eP\u003c/em\u003e \u0026lt; 5 × 10\u003csup\u003e-8\u003c/sup\u003e) for these traits obtained from CerebNet (red dots) are higher than those from SUIT+FS (blue dots) in UKBB\u003csub\u003eEUR\u003c/sub\u003e-GWASs. (\u003cstrong\u003eC\u003c/strong\u003e) The scatter diagram shows that the \u003cem\u003eP\u003c/em\u003e-values of 346 shared variant-trait associations (\u003cem\u003eP\u003c/em\u003e \u0026lt; 5 × 10\u003csup\u003e-8\u003c/sup\u003e) obtained from CerebNet (red dots) are smaller than those from SUIT+FS (blue dots) in UKBB\u003csub\u003eEUR\u003c/sub\u003e-GWASs.\u003cstrong\u003e \u003c/strong\u003e(\u003cstrong\u003eD\u003c/strong\u003e) The scatter diagram shows that the statistical powers of the 346 shared associations (\u003cem\u003eP\u003c/em\u003e \u0026lt; 5 × 10\u003csup\u003e-8\u003c/sup\u003e) obtained from CerebNet (red dots) are higher than those from SUIT+FS (blue dots) in UKBB\u003csub\u003eEUR\u003c/sub\u003e-GWASs. In each diagram, the lines indicate the interquartile range and the upper edge of each box indicates the median value. (\u003cstrong\u003eE\u003c/strong\u003e) Using UKBB\u003csub\u003eEUR\u003c/sub\u003e as the base dataset and CHIMGEN\u003csub\u003eEAS\u003c/sub\u003e, ABCD\u003csub\u003eEUR\u003c/sub\u003e, and ABCD\u003csub\u003eAFR\u003c/sub\u003e as three target datasets, the PGS predictive performance (R\u003csup\u003e2\u003c/sup\u003e) for cerebellar substructure volumes obtained from CerebNet (red dots) are better than those from SUIT+FS (blue dots).\u003c/p\u003e","description":"","filename":"Figure1.png","url":"https://assets-eu.researchsquare.com/files/rs-6694135/v1/c78897d90d0ceeb7a0331502.png"},{"id":84197727,"identity":"7409da65-2d63-4161-b4ca-ca443b52abb5","added_by":"auto","created_at":"2025-06-09 08:01:31","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":4771271,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eGenetic discovery of sex-combined univariate GWASs for 31 cerebellar volumetric traits.\u003c/strong\u003e (\u003cstrong\u003eA\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003eThe spatial distribution of 31 cerebellar volumetric traits. (\u003cstrong\u003eB\u003c/strong\u003e) Ideograms demonstrate the pooled study-wide significant (\u003cem\u003eP \u003c/em\u003e\u0026lt; 1.61 × 10\u003csup\u003e−9\u003c/sup\u003e) locus-trait associations of cerebellar volumetric traits in sex-combined univariate GWASs. The rhombus and circle represent known and new locus-trait associations (\u003cem\u003eP \u003c/em\u003e\u0026lt; 1.61 × 10\u003csup\u003e−9\u003c/sup\u003e) compared to prior GWASs. (\u003cstrong\u003eC\u003c/strong\u003e) The regional plots exhibit a new locus-trait association at 22q13.33 between rs79966207 and right cerebellar lobule VI volume. (\u003cstrong\u003eD\u003c/strong\u003e) The regional plots show a new locus-trait association at 14q24.3 between rs61742642 and left cerebellar lobule X volume. The red and blue lines represent the genome-wide (\u003cem\u003eP \u003c/em\u003e= 5 × 10\u003csup\u003e-8\u003c/sup\u003e) and study-wide (\u003cem\u003eP \u003c/em\u003e= 1.61 × 10\u003csup\u003e-9\u003c/sup\u003e) significance thresholds, respectively.\u003c/p\u003e","description":"","filename":"Figure2.png","url":"https://assets-eu.researchsquare.com/files/rs-6694135/v1/ab54e101d11686ec85ddbc60.png"},{"id":84198791,"identity":"e269583d-1525-40a1-9b17-d309ae2f888c","added_by":"auto","created_at":"2025-06-09 08:09:31","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":1844366,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eAllele-effect heterogeneity between ancestries and sexes.\u003c/strong\u003e (\u003cstrong\u003eA\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003eManhattan plots illustrate the heterogeneity of allele effects in associations with cerebellar vermis VIII volume across the EUR-GWAS, CHIMGEN\u003csub\u003eEAS\u003c/sub\u003e-GWAS, and ABCD\u003csub\u003eAFR\u003c/sub\u003e-GWAS. (\u003cstrong\u003eB-D\u003c/strong\u003e) Regional plots illustrate examples of ancestry-shared association between rs2350079 (4q13.2) and cerebellar vermis VII volume (\u003cstrong\u003eB\u003c/strong\u003e), EAS-specific association between rs72838327 (17p13.2) and left cerebellar white matter volume (\u003cstrong\u003eC\u003c/strong\u003e), and EUR-specific association between rs2217466 (2q36.1) and right cerebellar hemispheric V volume (\u003cstrong\u003eD\u003c/strong\u003e). (\u003cstrong\u003eE\u003c/strong\u003e) Pie charts show the numbers of sex-shared (CQ-tests, \u003cem\u003eP\u003c/em\u003e ≥ 0.05), sex-specific (CQ-tests, \u003cem\u003eP\u003c/em\u003e \u0026lt; 7.94 × 10\u003csup\u003e-5\u003c/sup\u003e, Bonferroni corrected), and undefined associations among the 630 pooled associations. (\u003cstrong\u003eF-G\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003eRegional plots show a female-specific association between rs6819982 (4q31.21) and right cerebellar white matter volume (\u003cstrong\u003eF\u003c/strong\u003e) and a male-specific association between rs6060308 (20q11.22) and right cerebellar hemispheric VIIb volume (\u003cstrong\u003eG\u003c/strong\u003e) in EUR-GWAS. The red and blue dashed lines represent genome-wide (\u003cem\u003eP \u003c/em\u003e= 5 × 10\u003csup\u003e-8\u003c/sup\u003e) and study-wide (\u003cem\u003eP \u003c/em\u003e= 1.61 × 10\u003csup\u003e-9\u003c/sup\u003e) significance, respectively.\u003c/p\u003e","description":"","filename":"Figure3.png","url":"https://assets-eu.researchsquare.com/files/rs-6694135/v1/831285b9f0e67d1c515572b8.png"},{"id":84198793,"identity":"ee082de2-c1ee-4ab8-b834-a5e4bfb5bcb0","added_by":"auto","created_at":"2025-06-09 08:09:31","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":1640282,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eStatistical fine-mapping and functional annotation\u003c/strong\u003e. (\u003cstrong\u003eA\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003eA summary of the statistical fine-mapping methods and their matched LD references. (\u003cstrong\u003eB\u003c/strong\u003e) The histogram shows the distribution of numbers of unique causal variants (PP \u0026gt; 0.8) in statistical fine-mapping based on EUR-GWAS (yellow), EUR-EAS-GWAS (red), and EUR-EAS-AFR-GWAS (blue). (\u003cstrong\u003eC\u003c/strong\u003e) The scatter diagrams show the comparisons (Wilcoxon rank-sum test) in size (left) and max PP (right) of 95% credible sets derived from EUR-GWAS (yellow dots), EUR-EAS-GWAS (red dots), and EUR-EAS-AFR-GWAS (blue dots). \u003cstrong\u003e(D-E)\u003c/strong\u003e Regional plots illustrate the improved resolution by cross-ancestry fine-mapping in two locus-trait associations between 18q21.1and cerebellar vermis VIII volume \u003cstrong\u003e(D)\u003c/strong\u003e and between 1q25.2 and right cerebellar hemispheric IX volume \u003cstrong\u003e(E)\u003c/strong\u003e.\u003c/p\u003e","description":"","filename":"Figure4.png","url":"https://assets-eu.researchsquare.com/files/rs-6694135/v1/8e039cb5c76df66de60f0d5a.png"},{"id":84197732,"identity":"2142cf7c-35b7-4883-8546-9a2af24505e6","added_by":"auto","created_at":"2025-06-09 08:01:31","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":993792,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eGenetic clustering of 27 cerebellar substructure volumes\u003c/strong\u003e. Based on genetic correlation matrix between each pair of 27 cerebellar substructure volumes (right panel), we use gSEM package to perform genetic clustering analyses. The distribution of nine-factors is illustrated in left panel.\u003c/p\u003e","description":"","filename":"Figure5.png","url":"https://assets-eu.researchsquare.com/files/rs-6694135/v1/51273f27b3bad5bf55a43918.png"},{"id":84197731,"identity":"b8734c30-f359-453c-bb13-a58d90131334","added_by":"auto","created_at":"2025-06-09 08:01:31","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":1539928,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eGenetic relations between cerebellar volumetric traits and other phenotypes\u003c/strong\u003e. (\u003cstrong\u003eA\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003eThe histogram illustrates the numbers of significant associations (\u003cem\u003eP\u003c/em\u003e \u0026lt; 8.15 × 10\u003csup\u003e-6\u003c/sup\u003e, Bonferroni corrected) of PGSs of 31 cerebellar volumetric traits with 198 cognitive and mental health phenotypes. (\u003cstrong\u003eB\u003c/strong\u003e) The associations between PGSs of cerebellar volumetric traits and reaction time (mean time to correctly identify matched). The upper panel shows\u0026nbsp;the effect sizes (\u003cem\u003eβ\u003c/em\u003e) and 95% CIs\u0026nbsp;for significant associations between PGSs of six global cerebellar volumetric traits and reaction time. The red color means associations significant at \u003cem\u003eP \u003c/em\u003e\u0026lt; 8.15 × 10\u003csup\u003e-6\u003c/sup\u003e. The lower map shows the associations between PGSs of 25 cerebellar substructure volumes and reaction time. Asterisk (*) means associations significant at \u003cem\u003eP \u003c/em\u003e\u0026lt; 8.15 × 10\u003csup\u003e-6\u003c/sup\u003e. (\u003cstrong\u003eC\u003c/strong\u003e) The scatter plot shows the significant associations (dashed line, \u003cem\u003eP\u003c/em\u003e \u0026lt; 8.15 × 10\u003csup\u003e-6\u003c/sup\u003e) between PGS of cerebellar vermis volume and behavioral phenotypes. (\u003cstrong\u003eD\u003c/strong\u003e) The heat plot shows genetic correlations between cerebellar volumetric traits and NPDs. Asterisk (*) indicates significant genetic correlation (\u003cem\u003eP \u003c/em\u003e\u0026lt; 1.61 × 10\u003csup\u003e-4\u003c/sup\u003e, Bonferroni corrected). (\u003cstrong\u003eE\u003c/strong\u003e) The ideogram shows genetic colocalizations (PP.H4 \u0026gt; 0.8, rhombus) between cerebellar volumetric traits and NPDs and multi-trait colocalizations (PP \u0026gt; 0.8, circle) among cerebellar gene expression, cerebellar volumetric traits, and NPDs. Abbreviations: AD, Alzheimer's disease; ADHD, Attention deficit hyperactivity disorder; ALS, Amyotrophic lateral sclerosis; ASD, autism spectrum disorder; BD, Bipolar disorder; PD, Parkinson's disease; PTSD, Posttraumatic stress disorder.\u003c/p\u003e","description":"","filename":"Figure6.png","url":"https://assets-eu.researchsquare.com/files/rs-6694135/v1/9f34e40855e4f8aab452f63f.png"},{"id":84199571,"identity":"50458182-c7c8-4bd2-9d03-5c64b94b5524","added_by":"auto","created_at":"2025-06-09 08:17:39","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":15323087,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6694135/v1/1d4619af-e643-4191-a868-d5be4a1f59cd.pdf"},{"id":84197726,"identity":"50e4344f-723a-4b18-8f29-b41b4962d334","added_by":"auto","created_at":"2025-06-09 08:01:31","extension":"xlsx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":4245526,"visible":true,"origin":"","legend":"Supplementary Tables","description":"","filename":"SupplExcel.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-6694135/v1/1087fcaa018a24a10ac7c16f.xlsx"},{"id":84197739,"identity":"aa09ecbe-1a22-4385-ae07-a649b5f32ed0","added_by":"auto","created_at":"2025-06-09 08:01:32","extension":"pdf","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":55058713,"visible":true,"origin":"","legend":"Supplementary Figures","description":"","filename":"SupplementalMaterial.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6694135/v1/4e482f3791840911dff8cf2f.pdf"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"Deep-learning segmentation and multi-ancestry GWAS enhance genetic discovery of the cerebellum","fulltext":[{"header":"Main","content":"\u003cp\u003eThe cerebellum plays an essential role in our daily lives, housing over half of the brain's neurons while occupying just 10% of its total volume\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e. The human cerebellum has historically been attributed exclusively to the planning and execution of movements\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e. However, recent evidence has highlighted its relevance in cognition and emotion\u003csup\u003e\u003cspan additionalcitationids=\"CR4\" citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e. Although the cerebellum is one of the first brain structures to differentiate, it is among the last structures to achieve maturity. The prolonged developmental timeline makes the cerebellum especially vulnerability to neurodevelopmental disorders such as autism spectrum disorder (ASD)\u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e, attention-deficit hyperactivity disorder (ADHD)\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e, and schizophrenia (SCZ)\u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eThe human cerebellum is not a homogeneous structure consisting of substructures that exhibit distinct cytoarchitecture and functions. The volumes of the cerebellum and its substructures can be measured using structural magnetic resonance imaging (MRI). For instance, the spatially unbiased infra-tentorial (SUIT) template\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e is a widely used atlas-based method for segmenting cerebellar substructures, while FreeSurfer (FS)\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e can estimate the volumes of left and right white matter and cortex of the cerebellum. In recent decades, notable volumetric changes in the cerebellum and its substructures have been identified in various neuropsychiatric disorders (NPDs), including posttraumatic stress disorder (PTSD)\u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e, depression\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e, ADHD\u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e, Parkinson's disease (PD)\u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e, Alzheimer\u0026rsquo;s disease (AD)\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e, and amyotrophic lateral sclerosis (ALS)\u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e. As these cerebellar volumetric traits demonstrate high heritability\u003csup\u003e\u003cspan additionalcitationids=\"CR18 CR19 CR20\" citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e, several genome-wide association studies (GWASs) have explored the genetic architecture of total cerebellar volume\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e,\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u003c/sup\u003e and the volumes of cerebellar substructures\u003csup\u003e\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e,\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eAlthough these GWASs improve our understanding of the genetic architecture of cerebellar volumetric traits, several limitations persist in these studies\u003csup\u003e\u003cspan additionalcitationids=\"CR19 CR20\" citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e. First, most previous studies\u003csup\u003e\u003cspan additionalcitationids=\"CR19\" citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e have focused on individuals of European ancestry (EUR), leaving other ancestral populations under-represented, although one study did include thousands of individuals of East Asian ancestry (EAS)\u003csup\u003e\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e. Second, the SUIT template\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e was used for segmenting cerebellar substructures in these studie\u003csup\u003e\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e,\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e. The volumes of cerebellar substructures obtained from the atlas-based segmentation may not be as precise as those derived from more advanced methods. For example, CerebNet uses a deep-learning model for cerebellar segmentation\u003csup\u003e\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e, demonstrating superior accuracy, test-retest reliability, and sensitivity compared to the SUIT method. Finally, the GWASs on the 28 SUIT-derived cerebellar substructure volumes in the prior two studies\u003csup\u003e\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e,\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e were performed as a minor component of the GWASs for more than 3,400 brain imaging phenotypes. The post-GWAS analyses overlooked the uniqueness of cerebellar volumetric traits, such as the biological processes involved in regulating these traits and the genetically informed subregions of the cerebellum.\u003c/p\u003e \u003cp\u003eThe first purpose of this study is to investigate whether the deep learning method for cerebellar segmentation (CerebNet) outperforms the atlas-based method (SUIT\u0026thinsp;+\u0026thinsp;FS) in genetic analyses (heritability estimation, genetic discovery, and polygenic prediction) of cerebellar substructure volumes. The second purpose is to perform multi-ancestry GWASs for 31 cerebellar volumetric traits in 57,071 participants, including 7,083 EAS participants from the Chinese Imaging Genetics (CHIMGEN) study\u003csup\u003e\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u003c/sup\u003e, 40,826 EUR participants from the UK Biobank (UKBB) study\u003csup\u003e\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u003c/sup\u003e, and 3,354 African (AFR) and 5,808 EUR participants from the Adolescent Brain Cognitive Development (ABCD) study\u003csup\u003e\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e\u003c/sup\u003e. The third purpose is to conduct a series of post-GWAS analyses to identify genetically informed cerebellar subregions, causal variants and enrichment pathways of cerebellar volumetric traits, along with their genetic correlation and colocalization with cerebellar gene expression, cognitive and mental health phenotypes, and NPDs. The study design is shown in \u003cb\u003eFig. \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e\u003c/b\u003e.\u003c/p\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eParticipants and data preparation\u003c/h2\u003e \u003cp\u003eWe included 57,071 participants with qualified genomic and structural MRI data from three datasets and four subgroups, including UKBB\u003csub\u003eEUR\u003c/sub\u003e (n\u0026thinsp;=\u0026thinsp;40,826), CHIMGEN\u003csub\u003eEAS\u003c/sub\u003e (n\u0026thinsp;=\u0026thinsp;7,083), ABCD\u003csub\u003eEUR\u003c/sub\u003e (n\u0026thinsp;=\u0026thinsp;5,808), and ABCD\u003csub\u003eAFR\u003c/sub\u003e (n\u0026thinsp;=\u0026thinsp;3,354) (\u003cb\u003etable \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e\u003c/b\u003e). The specific procedures for participant selection and quality control are detailed in \u003cb\u003eFig. S2\u003c/b\u003e. We used CerebNet to segment the cerebellum to obtain the volumes of cerebellar substructures. To assess the reliability of CerebNet segmentation, we calculated intraclass correlation coefficients (ICCs) of cerebellar substructure volumes derived from structural MRI data obtained at two separate time points from the same participants (CHIMGEN: n\u0026thinsp;=\u0026thinsp;26; UKBB: n\u0026thinsp;=\u0026thinsp;2,944; and ABCD: n\u0026thinsp;=\u0026thinsp;4,076). We found high reliability in the three datasets (CHIMGEN: ICC\u0026thinsp;=\u0026thinsp;0.62\u0026ndash;0.97; UKBB: ICC\u0026thinsp;=\u0026thinsp;0.88\u0026ndash;0.96; ABCD: ICC\u0026thinsp;=\u0026thinsp;0.87\u0026ndash;0.98; \u003cb\u003etable S2\u003c/b\u003e). GWASs were finally conducted on 10,001,636 genetic variants for UKBB\u003csub\u003eEUR\u003c/sub\u003e, 9,017,308 for CHIMGEN\u003csub\u003eEAS\u003c/sub\u003e, 9,343,804 for ABCD\u003csub\u003eEUR\u003c/sub\u003e, and 7,395,374 for ABCD\u003csub\u003eAFR\u003c/sub\u003e participants, respectively (\u003cb\u003eFig. S2, table S3\u003c/b\u003e).\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eComparison in genetic findings between cerebellar segmentation methods\u003c/h3\u003e\n\u003cp\u003eTo explore whether CerebNet outperforms SUIT\u0026thinsp;+\u0026thinsp;FS in genetic analyses of cerebellar substructure volumes, we used the two segmentation methods to calculate the volumes of 30 cerebellar substructures, and then compared their performance in the heritability estimation, genetic discovery, and polygenic score (PGS) prediction.\u003c/p\u003e \u003cp\u003e \u003cb\u003eHeritability estimation.\u003c/b\u003e We applied GCTA-GREML\u003csup\u003e\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u003c/sup\u003e to estimate single nucleotide polymorphism (SNP)-based heritability (h\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e) of the 30 cerebellar volumetric traits in the four subgroups, respectively. Although these traits obtained from both CerebNet and SUIT\u0026thinsp;+\u0026thinsp;FS were heritable (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05; \u003cb\u003etable S4\u003c/b\u003e), cerebellar substructure volumes obtained from CerebNet showed higher heritability (Wilcoxon rank-sum test, CHIMGEN\u003csub\u003eEAS\u003c/sub\u003e: \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;4.53 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;3\u003c/sup\u003e, UKBB\u003csub\u003eEUR\u003c/sub\u003e: \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;2.42 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;4\u003c/sup\u003e, ABCD\u003csub\u003eEUR\u003c/sub\u003e: \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;1.01 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;5\u003c/sup\u003e, ABCD\u003csub\u003eAFR\u003c/sub\u003e: \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;8.08 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;3\u003c/sup\u003e, Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eA) compared to those derived from SUIT\u0026thinsp;+\u0026thinsp;FS.\u003c/p\u003e \u003cp\u003e \u003cb\u003eGWAS discovery.\u003c/b\u003e We used the mixed linear model in fastGWA\u003csup\u003e\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u003c/sup\u003e to perform GWASs on 30 cerebellar substructure volumes from CerebNet and SUIT\u0026thinsp;+\u0026thinsp;FS in 40,826 UKBB participants. We used the Wilcoxon rank-sum test to compare the number, \u003cem\u003eP\u003c/em\u003e-value, and power of significant variant-trait associations from both methods. We identified 2,159 and 1,169 significant variant-trait associations (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;5 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;8\u003c/sup\u003e) for cerebellar substructure volumes from CerebNet and SUIT\u0026thinsp;+\u0026thinsp;FS (\u003cb\u003etable S5\u003c/b\u003e), confirming that CerebNet (median\u0026thinsp;=\u0026thinsp;64) enhanced GWAS discovery (Wilcoxon rank-sum test, \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;1.11 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;5\u003c/sup\u003e, Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eB) compared to SUIT\u0026thinsp;+\u0026thinsp;FS (median\u0026thinsp;=\u0026thinsp;39). In the 346 significant variant-trait associations identified by both methods, CerebNet (median: -log10(\u003cem\u003eP\u003c/em\u003e)\u0026thinsp;=\u0026thinsp;13.01; power\u0026thinsp;=\u0026thinsp;0.83) showed increased significance (Wilcoxon rank-sum test, \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;1.04 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;13\u003c/sup\u003e, Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eC) and power (Wilcoxon rank-sum test, \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;2.91 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;8\u003c/sup\u003e, Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eD) compared to SUIT\u0026thinsp;+\u0026thinsp;FS (median: -log10(\u003cem\u003eP\u003c/em\u003e)\u0026thinsp;=\u0026thinsp;9.86; power\u0026thinsp;=\u0026thinsp;0.61) (\u003cb\u003etable S5\u003c/b\u003e).\u003c/p\u003e \u003cp\u003e \u003cb\u003ePGS prediction.\u003c/b\u003e Using UKBB\u003csub\u003eEUR\u003c/sub\u003e-GWAS data as the base dataset and CHIMGEN\u003csub\u003eEAS\u003c/sub\u003e, ABCD\u003csub\u003eEUR\u003c/sub\u003e, and ABCD\u003csub\u003eAFR\u003c/sub\u003e raw data as three distinct target datasets, we used PRSice-2\u003csup\u003e28\u003c/sup\u003e to construct the best-fit PGS models for the 30 cerebellar substructure volumes obtained from CerebNet and SUIT\u0026thinsp;+\u0026thinsp;FS and to calculate the explained variance (R\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e) of PGS for each trait in each target dataset. In ABCD\u003csub\u003eEUR\u003c/sub\u003e, CerebNet (median R\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.047) improved the PGS predictive performance (Wilcoxon rank-sum test, \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;3.58 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;11\u003c/sup\u003e) in all cerebellar substructure volumes compared to SUIT\u0026thinsp;+\u0026thinsp;FS (median R\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.0027). In cross-ancestry prediction, CerebNet (EAS: median R\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.019; AFR: median R\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.0037) also improved the performance (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eE, \u003cb\u003etable S6\u003c/b\u003e) in 28/30 and 24/30 traits in CHIMGEN\u003csub\u003eEAS\u003c/sub\u003e (Wilcoxon rank-sum test, \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;1.49 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;6\u003c/sup\u003e) and ABCD\u003csub\u003eAFR\u003c/sub\u003e (Wilcoxon rank-sum test, \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;9.67 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;3\u003c/sup\u003e) compared to SUIT\u0026thinsp;+\u0026thinsp;FS (EAS: median R\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.0068; AFR: median R\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.0018).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e\n\u003ch3\u003eOverview of formal GWASs\u003c/h3\u003e\n\u003cp\u003eAfter verifying that CerebNet improves the genetic discovery of cerebellar substructure volumes, we performed comprehensive GWASs on the autosomal and X-chromosomal variants to investigate the genetic architecture of 31 cerebellar volumetric traits derived from CerebNet by further including the total cerebellar volume (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eA). GWASs were divided into sex-combined and sex-stratified (males and females) analyses, which were further split into univariate and multivariate analyses. In each subcategory (e.g., sex-combined univariate), we conducted four single-dataset GWASs (UKBB\u003csub\u003eEUR\u003c/sub\u003e-GWAS, ABCD\u003csub\u003eEUR\u003c/sub\u003e-GWAS, CHIMGEN\u003csub\u003eEAS\u003c/sub\u003e-GWAS, and ABCD\u003csub\u003eAFR\u003c/sub\u003e-GWAS), an EUR-GWAS meta-analysis, and two cross-ancestry GWAS meta-analyses (EUR-EAS-GWAS and EUR-EAS-AFR-GWAS). We did not find any population stratification (\u003cb\u003etable S7\u003c/b\u003e) based on the genomic control inflation factor (λ\u003csub\u003eGC\u003c/sub\u003e) and linkage disequilibrium (LD) score regression (LDSC) intercepts\u003csup\u003e\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u003c/sup\u003e. We regarded the sex-combined univariate GWASs as the primary results, while conducting other GWASs for supplementary analyses. We reported LD-independent associations, signals, and loci throughout the GWASs, and reported study-wide significant associations (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;1.61 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;9\u003c/sup\u003e) in univariate GWASs and genome-wide significant associations (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;5 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;8\u003c/sup\u003e) in multivariate GWASs.\u003c/p\u003e\n\u003ch3\u003eGenetic discovery in sex-combined univariate GWASs\u003c/h3\u003e\n\u003cp\u003e \u003cb\u003eSingle-dataset GWASs.\u003c/b\u003e We used the mixed linear model in fastGWA\u003csup\u003e\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u003c/sup\u003e to perform the sex-combined univariate GWASs (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;1.61 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;9\u003c/sup\u003e) for the 31 cerebellar volumetric traits in 40,826 UKBB\u003csub\u003eEUR\u003c/sub\u003e, 5,808 ABCD\u003csub\u003eEUR\u003c/sub\u003e, 7,083 CHIMGEN\u003csub\u003eEAS\u003c/sub\u003e, and 3,354 ABCD\u003csub\u003eAFR\u003c/sub\u003e participants, respectively. We identified 1,289/948 variant/locus-trait associations and 447/252 signals/loci in UKBB\u003csub\u003eEUR\u003c/sub\u003e-GWASs; 39/35 associations and 21/18 signals/loci in ABCD\u003csub\u003eEUR\u003c/sub\u003e-GWASs; 31/31 associations and 17/16 signals/loci in CHIMGEN\u003csub\u003eEAS\u003c/sub\u003e-GWASs; and 5/5 associations and 3/3 signals/loci in ABCD\u003csub\u003eAFR\u003c/sub\u003e-GWASs (\u003cb\u003eFig. S3, table S8, table S9\u003c/b\u003e). We noted that small non-EUR datasets could also yield new findings. For example, the locus-trait association between rs4752582 (10q26.13) and right cerebellar lobule VI volume was significant (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;1.46 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;9\u003c/sup\u003e) in ABCD\u003csub\u003eAFR\u003c/sub\u003e-GWAS rather than in other three GWASs. The SNP is an expression quantitative trait locus (eQTL) of \u003cem\u003eFGFR2\u003c/em\u003e. \u003cem\u003eFGFR2\u003c/em\u003e signaling in cerebellar Purkinje neurons is vital to motor learning\u003csup\u003e\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003e \u003cb\u003eEUR-GWAS meta-analyses.\u003c/b\u003e Based on the summary statistics from UKBB\u003csub\u003eEUR\u003c/sub\u003e-GWASs and ABCD\u003csub\u003eEUR\u003c/sub\u003e-GWASs, we utilized the inverse variance weighted (IVW) fixed effect model implemented in METAL\u003csup\u003e\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e\u003c/sup\u003e to conduct the EUR-GWAS meta-analyses (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;1.61 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;9\u003c/sup\u003e) for the 31 cerebellar volumetric traits. We found 1,602/1,144 variant/locus-trait associations and 558/288 signals/loci in EUR-GWASs (\u003cb\u003eFig. S3, table S10, table S11\u003c/b\u003e).\u003c/p\u003e \u003cp\u003e \u003cb\u003eCross-ancestry GWAS meta-analyses.\u003c/b\u003e We also used the IVW fixed effect model in METAL\u003csup\u003e\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e\u003c/sup\u003e to conduct the cross-ancestry GWAS meta-analyses (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;1.61 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;9\u003c/sup\u003e) for the 31 cerebellar volumetric traits. Based on the summary statistics from EUR-GWASs and CHIMGEN\u003csub\u003eEAS\u003c/sub\u003e-GWASs, we performed the cross-ancestry EUR-EAS-GWASs. We identified 1,719/1,273 variant/locus-trait associations and 563/323 signals/loci (\u003cb\u003eFig. S3, table S12, table S13)\u003c/b\u003e. By further including ABCD\u003csub\u003eAFR\u003c/sub\u003e-GWAS summary statistics, we performed the cross-ancestry EUR-EAS-AFR-GWAS meta-analyses (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;1.61 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;9\u003c/sup\u003e). We found 1,681/1,266 variant/locus-trait associations and 553/316 signals/loci (\u003cb\u003eFig. S3, table S14, table S15)\u003c/b\u003e.\u003c/p\u003e \u003cp\u003e \u003cb\u003ePooling significant associations from sex-combined univariate GWASs.\u003c/b\u003e As most participants in our GWASs were of European ancestry and the majority of previous GWASs for cerebellar volumetric traits\u003csup\u003e\u003cspan additionalcitationids=\"CR19\" citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e were conducted on EUR individuals, we used the LD\u003csub\u003eEUR\u003c/sub\u003e reference to pool the results from sex-combined univariate GWASs and identify new findings. We identified 2,091/1,484 variant/locus-trait associations (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eB) and 692/353 signals/loci (\u003cb\u003etable S16\u003c/b\u003e, \u003cb\u003etable S17\u003c/b\u003e). Compared to the 614/447 known variant/locus-trait associations and 217/135 signals/loci at \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;1.61 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;9\u003c/sup\u003e identified by all previous studies\u003csup\u003e\u003cspan additionalcitationids=\"CR19 CR20\" citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e, we found 1,641 new variant-trait associations, 1,102 locus-trait associations, 522 signals, and 241 loci (\u003cb\u003etable S16\u003c/b\u003e, \u003cb\u003etable S17\u003c/b\u003e). We provided evidence for improved genetic discovery in cross-ancestry GWASs. As an example, we identified a significant association (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;2.19 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;10\u003c/sup\u003e) between rs79966207 (22q13.33) and the volume of right cerebellar lobule VI in the EUR-EAS cross-ancestry GWAS (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eC), while the association was not significant in single-dataset GWASs or in the EUR-GWAS. The variant is a missense variant of \u003cem\u003ePLXNB2\u003c/em\u003e, regulating the timing of differentiation and the motility of cerebellar granule neurons\u003csup\u003e\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e\u003c/sup\u003e. Another example was the association between rs61742642 (14q24.3) and left cerebellar lobule X volume, which was only significant (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;3.91 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;10\u003c/sup\u003e) in EUR-EAS-AFR cross-ancestry GWAS (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eD). The variant is a missense variant of \u003cem\u003eESRRB\u003c/em\u003e, serving as a marker of Purkinje cells in the cerebellum\u003csup\u003e\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e\n\u003ch3\u003eGenetic discovery in sex-stratified univariate GWASs\u003c/h3\u003e\n\u003cp\u003e \u003cb\u003eMale-specific GWASs.\u003c/b\u003e We performed the four single-dataset univariate GWASs for the 31 cerebellar volumetric traits in males, based on which we conducted EUR and cross-ancestry GWAS meta-analyses (\u003cb\u003eFig. S4\u003c/b\u003e). Using the male-specific LD\u003csub\u003eEUR\u003c/sub\u003e reference, we pooled them into 523/429 variant/locus-trait associations and 201/142 signals/ loci. We additionally identified 8/5 new variant/locus-trait associations and 5/3 new signals/loci (\u003cb\u003etable S18, table S19\u003c/b\u003e).\u003c/p\u003e \u003cp\u003e \u003cb\u003eFemale-specific GWASs.\u003c/b\u003e We also performed the four single-dataset univariate GWASs and three GWAS meta-analyses for these 31 cerebellar volumetric traits in females (\u003cb\u003eFig. S5\u003c/b\u003e), and employed the female-specific LD\u003csub\u003eEUR\u003c/sub\u003e reference to pool the results into 689/556 variant/locus-trait associations and 251/174 signals/loci. We additionally found 18/15 new variant/locus-trait associations and 4/1 new signals/locus (\u003cb\u003etable S20, table S21\u003c/b\u003e).\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eGenetic discovery in multivariate GWASs\u003c/h2\u003e \u003cp\u003eWe applied C-GWAS\u003csup\u003e\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e\u003c/sup\u003e to conduct sex-combined multivariate GWASs (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;5 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;8\u003c/sup\u003e) based on the summary data of sex-combined univariate GWASs for 27 non-overlapping cerebellar volumetric traits. We identified 386, 19, 34, 4, 428, 443, and 439 loci from UKBB\u003csub\u003eEUR\u003c/sub\u003e-GWASs, ABCD\u003csub\u003eEUR\u003c/sub\u003e-GWASs, CHIMGEN\u003csub\u003eEAS\u003c/sub\u003e-GWASs, ABCD\u003csub\u003eAFR\u003c/sub\u003e-GWASs, EUR-GWASs, EUR-EAS-GWASs, and EUR-EAS-AFR-GWASs, respectively (\u003cb\u003eFig. S6\u003c/b\u003e). These loci were pooled into 484 LD-independent loci, including 153 additionally new loci (\u003cb\u003etable S22\u003c/b\u003e). We also performed the male- and female-specific C-GWASs for the 27 cerebellar volumetric traits (\u003cb\u003eFig. S6\u003c/b\u003e). Using the sex-specific LD\u003csub\u003eEUR\u003c/sub\u003e reference, we pooled the results into 225 loci for males and 262 for females. We additionally discovered 9 new loci (\u003cb\u003etable S23, table S24\u003c/b\u003e).\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eAllele-effect heterogeneity across ancestries\u003c/h3\u003e\n\u003cp\u003eCochran's Q test (CQ-test) was performed to identify the allele-effect heterogeneity across ancestries (EUR, EAS, and AFR) for the pooled variant-trait associations (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;1.61 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;9\u003c/sup\u003e) from sex-combined univariate GWASs (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eA). In the 1,606 associations included in all ancestry-specific GWASs, we found 1,321 (82.25%) ancestry-shared associations (CQ-test: \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026ge;\u0026thinsp;0.05, \u003cb\u003etable S25\u003c/b\u003e). One ancestry-shared association (CQ-test: \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.33, Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eB) was found between rs2350079 (4q13.2) and cerebellar vermis VII volume. This SNP is an eQTL of \u003cem\u003eEPHA5\u003c/em\u003e, regulating the development of neuronal cytoarchitecture\u003csup\u003e\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e\u003c/sup\u003e. We identified 19 (1.83%) ancestry-specific associations (CQ-test: \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;3.11 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;5\u003c/sup\u003e, \u003cb\u003etable S25\u003c/b\u003e). For example, we discovered one EAS-specific association (CQ-test: \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;2.63 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;8\u003c/sup\u003e, Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eC) between rs72838327 (17p13.2) and left cerebellar white matter volume. The variant is mapped to \u003cem\u003eK1F1C\u003c/em\u003e, causing cerebellar dysfunction and cerebellar ataxia\u003csup\u003e\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e,\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e\u003c/sup\u003e. We also found an EUR-specific association (CQ-test: \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;9.36 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;6\u003c/sup\u003e, Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eD) between rs2217466 (2q36.1) and right cerebellar lobule V volume. The SNP is mapped to \u003cem\u003ePAX3\u003c/em\u003e, a marker of cerebellar development\u003csup\u003e\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\n\u003ch3\u003eAllele effect heterogeneity between sexes\u003c/h3\u003e\n\u003cp\u003eTo exclude the potential bias from inter-ancestry differences, we only used the CQ-test to identify the allelic-effect heterogeneity between sexes for the pooled variant-trait associations (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;1.61 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;9\u003c/sup\u003e) obtained from the sex-stratified univariate GWASs in the EUR (n\u0026thinsp;=\u0026thinsp;620), CHIMGEN\u003csub\u003eEAS\u003c/sub\u003e (n\u0026thinsp;=\u0026thinsp;6), and ABCD\u003csub\u003eAFR\u003c/sub\u003e (n\u0026thinsp;=\u0026thinsp;4) using the ancestry-specific LD reference. These associations were categorized into 490 sex-shared (CQ-test: \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026ge;\u0026thinsp;0.05) and 8 sex-specific (CQ-test: \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05/630\u0026thinsp;=\u0026thinsp;7.94 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;5\u003c/sup\u003e, Bonferroni corrected) associations (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eE, \u003cb\u003etable S26\u003c/b\u003e). For example, we found a female-specific association between rs6819982 (4q31.21) and right cerebellar white matter volume (CQ-test: \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;5.19 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;7\u003c/sup\u003e, Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eF). The variant is an eQTL of \u003cem\u003eHHIP\u003c/em\u003e, involved in hedgehog signaling that orchestrates cerebellar development\u003csup\u003e\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e\u003c/sup\u003e. We found a male-specific association between rs6060308 (20q11.22) and right cerebellar lobule VIIb volume (CQ-test: \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;3.92 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;6\u003c/sup\u003e, Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eG). The variant is an eQTL of \u003cem\u003eEDEM2\u003c/em\u003e, involved in endoplasmic reticulum-associated degradation and linked to early-onset cerebellar ataxia\u003csup\u003e\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eStatistical fine-mapping\u003c/h2\u003e \u003cp\u003eWe performed statistical fine-mapping based on the summary statistics of EUR-GWASs, EUR-EAS-GWASs, and EUR-EAS-AFR-GWASs on the 31 cerebellar volumetric traits, respectively. For each locus of the significant locus-trait associations, we estimated its 95% credible set and identified causal variants with posterior probability (PP) above 0.8 based on the matched LD reference and suitable approaches (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eA). SuSiE-R\u003csup\u003e\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e,\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e\u003c/sup\u003e was used to EUR-GWASs with LD\u003csub\u003eEUR\u003c/sub\u003e reference, while SuSiEx\u003csup\u003e\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e\u003c/sup\u003e was applied to EUR-EAS-GWASs with LD\u003csub\u003eEUR\u003c/sub\u003e and LD\u003csub\u003eEAS\u003c/sub\u003e references and EUR-EAS-AFR-GWASs with LD\u003csub\u003eEUR\u003c/sub\u003e, LD\u003csub\u003eEAS\u003c/sub\u003e, and LD\u003csub\u003eAFR\u003c/sub\u003e references. We successfully identified 941 95% credible sets for 1,144 locus-trait associations from EUR-GWASs (\u003cb\u003etable S27\u003c/b\u003e), 1,660 for 1,273 locus-trait associations from EUR-EAS-GWASs (\u003cb\u003etable S28\u003c/b\u003e), and 2,073 for 1,266 locus-trait associations from EUR-EAS-AFR-GWASs (\u003cb\u003etable S29\u003c/b\u003e). We also found 53 causal variants (PP\u0026thinsp;\u0026gt;\u0026thinsp;0.8) from EUR-GWASs (\u003cb\u003etable S30\u003c/b\u003e), 142 from EUR-EAS-GWASs (\u003cb\u003etable S31\u003c/b\u003e), and 356 from EUR-EAS-AFR-GWASs (\u003cb\u003etable S32\u003c/b\u003e) (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eB), generating 453 unique causal variants across these GWASs.\u003c/p\u003e \u003cp\u003eWe aligned the 95% credible sets with one causal variant assumption across these fine-mapping analyses, generating 467 aligned 95% credible sets (\u003cb\u003etable S33\u003c/b\u003e). We used Wilcoxon rank-sum test to compare the size and maximal PP of these 95% credible sets. The median size of these 95% credible sets was 33, 14, and 10 for one, two, and three-ancestry fine-mapping, demonstrating significant differences (Wilcoxon rank-sum test: \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;2.25 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;47\u003c/sup\u003e for EUR vs EUR-EAS; \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;8.83 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;28\u003c/sup\u003e for EUR vs EUR-EAS-AFR; \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;5.97 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;6\u003c/sup\u003e for EUR-EAS vs EUR-EAS-AFR; Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eC). The median maximal PP values in these 95% credible sets was 0.09, 0.19, and 0.28 for one, two, and three-ancestry fine-mapping, also demonstrating significant differences (Wilcoxon rank-sum test: \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;1.62 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;35\u003c/sup\u003e for EUR vs EUR-EAS; \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;3.80 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;19\u003c/sup\u003e for EUR vs EUR-EAS-AFR; \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;2.33 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;5\u003c/sup\u003e for EUR-EAS vs EUR-EAS-AFR; Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eC).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eFunctional annotations\u003c/h2\u003e \u003cp\u003eWe used three approaches to perform functional annotations for the 453 unique causal variants (PP\u0026thinsp;\u0026gt;\u0026thinsp;0.8) from fine-mapping. We used ANNOVAR\u003csup\u003e\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e\u003c/sup\u003e to annotate functional consequences, and found 226 (49.9%) variants in the intergenic region, 164 (36.2%) in the intron region, 6 (1.3%) in the UTR region, and 6 (1.3%) missense variants. We then assessed deleteriousness using the combined annotation-dependent depletion (CADD) score\u003csup\u003e\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e\u003c/sup\u003e, identifying 13 (2.9%) pathogenic variants (CADD score\u0026thinsp;\u0026gt;\u0026thinsp;20). Finally, we evaluated the regulatory function of causal variants by RegulomeDB (RDB)\u003csup\u003e\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e\u003c/sup\u003e. We found 200 (44.2%) variants with regulatory potential (RDB\u0026thinsp;\u0026lt;\u0026thinsp;4). These results (\u003cb\u003etables S30-32\u003c/b\u003e) improve the understanding of genetic mechanisms of cerebellar structure. For example, as a causal variant of cerebellar vermis X volume (PP\u0026thinsp;=\u0026thinsp;0.87), rs145919520 (17p13.1) is a missense variant of \u003cem\u003eCLUH\u003c/em\u003e. \u003cem\u003eCLUH\u003c/em\u003e plays a crucial role in maintaining functional mitochondria in axons\u003csup\u003e\u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e\u003c/sup\u003e. As a fine-mapped causal variant of cerebellar vermis VIII volume (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eD), rs9956387 (18q21.1) is a missense variant of \u003cem\u003eSKOR2\u003c/em\u003e, serving as a transcriptional regulator in Purkinje cells during cerebellum development\u003csup\u003e\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e48\u003c/span\u003e\u003c/sup\u003e. The lead variant rs7540842 (1q25.2) mapped to \u003cem\u003eASTN1\u003c/em\u003e was identified as a fine-mapped causal variant of right cerebellar lobule IX volume (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eE). \u003cem\u003eASTN1\u003c/em\u003e affects the cerebellar volume, neuronal migration, and development of Purkinje cells\u003csup\u003e\u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eColocalization with gene expression\u003c/h2\u003e \u003cp\u003eFor loci of 1,144 locus-trait associations from EUR-GWASs for cerebellar volumetric traits, we used Coloc\u003csup\u003e\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e\u003c/sup\u003e to perform the Bayesian colocalization to identify the shared causal variants with eQTLs (\u003cem\u003eP\u0026thinsp;\u0026lt;\u003c/em\u003e\u0026thinsp;5 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;8\u003c/sup\u003e) of human cerebellar tissue from MetaBrain (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://metabrain.nl/\u003c/span\u003e\u003cspan address=\"https://metabrain.nl/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e)\u003csup\u003e51\u003c/sup\u003e. Colocalization was defined as the posterior probability of shared causal variant (PP.H4) over 0.8. We found 160 colocalizations between 71 genes and 30 cerebellar volumetric traits (\u003cb\u003eFig. S7A\u003c/b\u003e; \u003cb\u003etable S34\u003c/b\u003e). Among them, 36 genes had colocalization with at least two cerebellar volumetric traits, such as \u003cem\u003eSLC44A5\u003c/em\u003e, \u003cem\u003ePTK2\u003c/em\u003e, and \u003cem\u003eZFHX4\u003c/em\u003e colocalized with 14, 13, and 6 traits, respectively. \u003cem\u003eSLC44A5\u003c/em\u003e is involved in lipid metabolism, and its SNPs have been associated with education attainment\u003csup\u003e\u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e52\u003c/span\u003e\u003c/sup\u003e and major depressive disorder\u003csup\u003e\u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e53\u003c/span\u003e\u003c/sup\u003e. \u003cem\u003ePTK2\u003c/em\u003e encodes FAK, a cell adhesion tyrosine kinase, implicating in synaptic branching\u003csup\u003e\u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e54\u003c/span\u003e\u003c/sup\u003e and linking to congenital cerebellar hypoplasia\u003csup\u003e\u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e55\u003c/span\u003e\u003c/sup\u003e. \u003cem\u003eZFHX4\u003c/em\u003e acts as a marker of GABAergic interneurons in the forming deep nuclei of the cerebellum\u003csup\u003e\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003ePathway enrichment\u003c/h2\u003e \u003cp\u003eBased on 71 prioritized genes whose cerebellar expression colocalized with locus-trait associations from EUR-GWASs for cerebellar volumetric traits, we input these genes to g:Profiler (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://biit.cs.ut.ee/gprofiler/gost\u003c/span\u003e\u003cspan address=\"https://biit.cs.ut.ee/gprofiler/gost\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e), a web server for functional enrichment analysis\u003csup\u003e\u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e56\u003c/span\u003e\u003c/sup\u003e, to perform the pathway enrichment analysis based on the pre-specified pathways in GO (15,472 biological processes) database\u003csup\u003e\u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e57\u003c/span\u003e\u003c/sup\u003e. We found 12 enrichment pathways (\u003cem\u003eP\u003c/em\u003e\u003csub\u003e\u003cem\u003ec\u003c/em\u003e\u003c/sub\u003e \u0026lt; 0.05, Benjamini-Hochberg FDR corrected), demonstrating the nervous system development, neurogenesis, and system development (\u003cb\u003eFig. S7B\u003c/b\u003e; \u003cb\u003etable S35\u003c/b\u003e) as the top three biological processes.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eGenetically informed cerebellar subregions\u003c/h2\u003e \u003cp\u003eThe cerebellum consists of two hemispheres (including both cortex and white matter) connected by a central structure called the vermis. Based on its fissures, the cerebellum is categorized into three lobes: the anterior (lobules I-V), posterior (lobules VI-IX), and flocculonodular (lobule X) lobes. Nevertheless, the genetically informed subregions of the cerebellum remain unclear. Here, we utilized the high-definition likelihood (HDL) method\u003csup\u003e\u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e58\u003c/span\u003e\u003c/sup\u003e to estimate bivariate genetic correlations across the volumes of 27 distinct cerebellar substructures. Based on the genetic correlation matrix (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e, \u003cb\u003etable S36\u003c/b\u003e), we used genomic structural equation modelling (gSEM)\u003csup\u003e\u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e59\u003c/span\u003e\u003c/sup\u003e to perform genetic clustering analyses, and defined an acceptable fit of the model as the comparative fit index (CFI)\u0026thinsp;\u0026gt;\u0026thinsp;0.90 and standardized root-mean-squared residual (SRMR)\u0026thinsp;\u0026lt;\u0026thinsp;0.10\u003csup\u003e59\u003c/sup\u003e. We assessed a common factor model, yielding a poor model fit (CFI\u0026thinsp;=\u0026thinsp;0.51, SRMR\u0026thinsp;=\u0026thinsp;0.15). We then evaluated the anatomical scheme (anterior lobe, posterior lobe, flocculonodular lobe, vermis, and white matter), which also resulted a poor fit (CFI\u0026thinsp;=\u0026thinsp;0.79, SRMR\u0026thinsp;=\u0026thinsp;0.12). To determine the best-fit model, we conducted exploratory factor analysis (EFA) based on the genetic correlation matrix of cerebellar substructure volumes. We identified a nine-factor model that could explain 78.1% of the total genetic variance (\u003cb\u003etable S37\u003c/b\u003e). The subsequent confirmatory factor analysis (CFA) confirmed the nine-factor model by exhibiting an almost perfect fit (CFI\u0026thinsp;=\u0026thinsp;0.99, SRMR\u0026thinsp;=\u0026thinsp;0.058; Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e). These results indicate that the cerebellum may be divided into nine subregions, each with a unique genetic architecture. The vermis substructures were clustered into a single subregion. In cerebellar hemispheres, identical substructures are grouped into the same subregions, such as the white matter subregion. Consistent with anatomical divisions, the bilateral anterior lobes (lobules I-V) and the bilateral flocculonodular lobes (lobule X) were also clustered into two distinct subregions, respectively. However, the cerebellar posterior lobe had complex genetic architecture and was divided into five subregions. Although lobules VI and IX were two distinct subregions, lobules CrusI, CrusII, VIIb, VIIIa, and VIIIb were clustered into three subregions that differ from the anatomical subdivisions. The lobules CrusII, VIIb, and VIIIa were included in one subregion, while lobules CrusI and VIIIb were clustered into two distinct subregions.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eGenetic associations with cognitive and mental health phenotypes\u003c/h2\u003e \u003cp\u003eThe extensive cognitive and mental health assessments in the UKBB dataset provided us a unique opportunity to investigate the associations of PGSs of cerebellar volumetric traits with these cognitive and mental health phenotypes. We utilized summary statistics from EUR-GWASs for the 31 cerebellar volumetric traits to construct the PGS models, which were then applied to estimate the PGS scores of 371,558 unrelated EUR-UKBB participants not included in EUR-GWASs. In these participants, we used PHESANT\u003csup\u003e\u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e60\u003c/span\u003e\u003c/sup\u003e to conduct phenome-wide association studies (PheWASs) to identify associations between 31 PGSs and 198 cognitive and mental health phenotypes (\u003cb\u003etable S38\u003c/b\u003e). We found 141 significant associations (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05/31/198\u0026thinsp;=\u0026thinsp;8.15 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;6\u003c/sup\u003e, Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eA, \u003cb\u003etable S39\u003c/b\u003e) between 31 cerebellar volumetric traits and 20 behavioral phenotypes (two cognitive and 18 mental health phenotypes). For example, the reaction time in pairs matching was associated with PGSs of six cerebellar volumetric traits (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eB), consistent with the longer reaction time in patients with cerebellar lesions\u003csup\u003e\u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e61\u003c/span\u003e\u003c/sup\u003e. The associations of PGS of cerebellar vermis volume with tenseness/restlessness (\u003cem\u003eβ\u003c/em\u003e = -0.007, \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;2.78 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;16\u003c/sup\u003e), happiness (\u003cem\u003eβ\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.015, \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;6.38 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;10\u003c/sup\u003e), and morning drink of alcohol (\u003cem\u003eβ\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.036, \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;3.26 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;8\u003c/sup\u003e, Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eC) were also in line with previous observations\u003csup\u003e\u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e62\u003c/span\u003e,\u003cspan citationid=\"CR63\" class=\"CitationRef\"\u003e63\u003c/span\u003e\u003c/sup\u003e. These results offered further support for the cerebellum's involvement in the regulation of cognition and emotion.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003eGenetic associations with neuropsychiatric disorders (NPDs)\u003c/h2\u003e \u003cp\u003e \u003cb\u003eGenetic correlation.\u003c/b\u003e Based on the EUR-GWAS summary statistics for both cerebellar volumetric traits from this study and NPDs from prior studies (\u003cb\u003etable S40\u003c/b\u003e), we used HDL\u003csup\u003e\u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e58\u003c/span\u003e\u003c/sup\u003e to explore genetic correlations between 31 cerebellar volumetric traits and ten NPDs. We found 11 significant genetic correlations (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;1.61 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;4\u003c/sup\u003e, Bonferroni corrected) between nine cerebellar volumetric traits and four NPDs (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eD, \u003cb\u003etable S41\u003c/b\u003e), including genetic correlations between eight cerebellar volumetric traits and PTSD (all \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;2.96 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;5\u003c/sup\u003e), total cerebellar volume and ALS (r\u0026thinsp;=\u0026thinsp;0.076, \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;1.00 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;4\u003c/sup\u003e), left lobules I-IV volume and BD (r = -0.075, \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;8.52 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;5\u003c/sup\u003e), and left lobule VIIIa volume and depression (r = -0.070, \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;4.42 \u0026times; 10\u003csup\u003e\u0026minus;\u0026thinsp;5\u003c/sup\u003e).\u003c/p\u003e \u003cp\u003e \u003cb\u003eGenetic colocalization.\u003c/b\u003e Based on the EUR-GWAS summary statistics for 31 cerebellar volumetric traits and ten NPDs, we employed Coloc\u003csup\u003e\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e\u003c/sup\u003e to explore genetic loci that are shared between cerebellar volumetric traits and NPDs. Among the 1,144 locus-trait associations from EUR-GWASs, we identified colocalization (PP.H4\u0026thinsp;\u0026gt;\u0026thinsp;0.8; Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eE and \u003cb\u003etable S42\u003c/b\u003e) between 17 cerebellar volumetric traits and six NPDs (AD, BD, depression, PD, PTSD, and SCZ) at 11 loci. We further performed multi-trait colocalization using HyPrColoc\u003csup\u003e\u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e64\u003c/span\u003e\u003c/sup\u003e, to identify the loci shared by cerebellar gene expression, cerebellar volumetric traits, and NPDs. We found nine multi-trait colocalizations (PP\u0026thinsp;\u0026gt;\u0026thinsp;0.8, Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eE and \u003cb\u003etable S43\u003c/b\u003e) at two loci between cerebellar expression of two genes (\u003cem\u003eSNX31\u003c/em\u003e, \u003cem\u003eARHGAP27\u003c/em\u003e), five cerebellar volumetric traits, and three NPDs (AD, PD, and PTSD). One locus (17q21.31) was shared by cerebellar \u003cem\u003eARHGAP27\u003c/em\u003e expression, left and right cerebellar white matter volumes, and three NPDs (AD, PD, and PTSD) (PP\u0026thinsp;=\u0026thinsp;0.80\u0026ndash;0.93, \u003cb\u003etable S43\u003c/b\u003e). Another locus (8q22.3) was shared by cerebellar \u003cem\u003eSNX31\u003c/em\u003e expression, three cerebellar volumetric traits (left cerebellar cortex volume, total cerebellar volume, and cerebellar vermis IX volume), and AD (PP\u0026thinsp;=\u0026thinsp;0.95\u0026ndash;0.96, \u003cb\u003etable S43\u003c/b\u003e). \u003cem\u003eSNX31\u003c/em\u003e is a member of the sorting nexin family, which is essential for synaptic function, and dysregulation of sorting nexin is observed in patients with AD\u003csup\u003e\u003cspan citationid=\"CR65\" class=\"CitationRef\"\u003e65\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eIn this study, we performed GWASs on 31 cerebellar volumetric traits obtained from deep-learning segmentation in 57,071 participants across three ancestries (EUR, EAS and AFR). We discovered that deep-learning outperformed atlas-based segmentation in heritability estimation, genetic discovery, and polygenic prediction. We identified 353 independent loci in sex-combined univariate GWASs, including 241 new loci. We then performed sex-stratified univariate GWASs and additionally discovered four new loci. We finally conducted sex-combined and sex-stratified multivariate GWASs and further found 162 new loci. Although most associations were shared by ancestries and sexes, we still found 19 ancestry-specific and eight sex-specific associations. We prioritized 453 causal variants and 71 genes and clustered 27 cerebellar anatomical substructures into nine subregions based on their genetic architecture. PGSs of cerebellar volumetric traits were correlated with 20 cognitive and mental health phenotypes. These cerebellar volumetric traits showed genetic correlations with four and genetic colocalizations with six neuropsychiatric disorders.\u003c/p\u003e \u003cp\u003eA distinctive contribution of our study to the field of neuroimaging genetics is the demonstration that deep-learning-based cerebellar segmentation surpassed atlas-based cerebellar segmentation in heritability estimation, genetic discovery, and genetic risk prediction. The SUIT atlas is the most commonly used probabilistic atlas of the human cerebellum\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e. The cerebellar substructure volumes of each participant can be obtained by non-linearly registering structural magnetic resonance images to the atlas known as SUIT-based cerebellar segmentation. As the most reliable atlas-based segmentation for the human cerebellum, it has been applied to calculate cerebellar substructure volumes in all previous GWASs on cerebellar volumetric traits\u003csup\u003e\u003cspan additionalcitationids=\"CR19 CR20\" citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e–\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e. As a probabilistic atlas, SUIT cannot precisely estimate the boundary of each cerebellar substructure for each individual. The imprecise phenotyping may undermine heritability estimation, GWAS discovery, and PGS prediction\u003csup\u003e\u003cspan citationid=\"CR66\" class=\"CitationRef\"\u003e66\u003c/span\u003e\u003c/sup\u003e. Recently, deep learning methods have been applied to cerebellar segmentation, achieving superior performance compared to atlas-based segmentation. For instance, CerebNet is proposed as a fast and reliable deep-learning pipeline for detailed cerebellar segmentation and showed higher precision, reliability, and sensitivity than other methods including SUIT segmentation\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e. The improved heritability estimation, GWAS discovery, and PGS prediction for cerebellar volumetric traits obtained from CerebNet indicate that more precise phenotyping is essential for genetic analyses of human phenotypes.\u003c/p\u003e \u003cp\u003eA significant contribution of our study to the field of neuroimaging genetics is the discovery of new genetic associations with cerebellar volumetric traits through precise phenotyping and comprehensive multi-ancestry GWASs. For instance, we found 1,484 locus-trait associations (\u003cem\u003eP\u003c/em\u003e \u0026lt; 1.61 × 10\u003csup\u003e− 9\u003c/sup\u003e) in sex-combined univariate GWASs for 31 cerebellar volumetric traits, comprising 1,102 new associations, which is 3.3 times the 447 associations reported in previous GWASs for these traits\u003csup\u003e\u003cspan additionalcitationids=\"CR19 CR20\" citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e–\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e. A new association was observed between rs17379472 (16q21) and bilateral lobule VI volumes. The lead variant is a missense variant of \u003cem\u003eADGRG1\u003c/em\u003e (also called as \u003cem\u003eGPR56\u003c/em\u003e), leading to cerebellar hypoplasia\u003csup\u003e\u003cspan citationid=\"CR67\" class=\"CitationRef\"\u003e67\u003c/span\u003e\u003c/sup\u003e. We found that increased ancestry diversity in statistical fine-mapping could enhance the identification of causal variants and decrease the average size of 95% credible sets, emphasizing the importance of including ancestrally diverse populations in GWASs\u003csup\u003e\u003cspan citationid=\"CR68\" class=\"CitationRef\"\u003e68\u003c/span\u003e\u003c/sup\u003e. Although most genetic associations were shared by ancestries and sexes, we also found 19 ancestry-specific and eight sex-specific associations, providing the potential genetic substrates for the observed differences in cerebellar volumetric traits between ancestries and sexes\u003csup\u003e\u003cspan citationid=\"CR69\" class=\"CitationRef\"\u003e69\u003c/span\u003e,\u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e70\u003c/span\u003e\u003c/sup\u003e. By leveraging the distributed genetic influences across cerebellar substructure volumes, we performed multivariate GWASs for cerebellar volumetric traits and additionally discovered 162 new loci. For example, we found a locus (lead variant rs111511908) that was only significant in multivariate GWASs. The variant is an eQTL of \u003cem\u003eCACNB4\u003c/em\u003e, linked to neurodevelopmental disorder including cerebellar atrophy\u003csup\u003e\u003cspan citationid=\"CR71\" class=\"CitationRef\"\u003e71\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eAnother valuable aspect of this study is the identification of genetically informed subregions of the human cerebellum. Although correspondence was observed between the majority of anatomical and genetic subdivisions, notable differences were identified in the affiliation of the Crus and VIII sub-lobules within the posterior lobe. Based on the genetic architecture, CrusII, VIIb, and VIIIa were grouped into a single subregion, while CrusI and VIIIb were categorized into two separate subregions. The genetically informed classification of these sub-lobules is partially aligned with the functional and pathological distinctions between Crus I and Crus II, as well as between VIIIa and VIIIb, along with the observed coactivation and joint damage between VIIb and VIIIa. For instance, in language processing, CrusI was activated during the syntactic task, while CrusII was activated during the semantic task\u003csup\u003e\u003cspan citationid=\"CR72\" class=\"CitationRef\"\u003e72\u003c/span\u003e\u003c/sup\u003e. In patients with bipolar disorder, CrusI exhibited increased functional connectivity with other brain regions, while CrusII showed decreased connectivity\u003csup\u003e\u003cspan citationid=\"CR73\" class=\"CitationRef\"\u003e73\u003c/span\u003e\u003c/sup\u003e. In long-term plasticity, drum training resulted in increased volume in VIIIa but decreased volume in VIIIb\u003csup\u003e\u003cspan citationid=\"CR74\" class=\"CitationRef\"\u003e74\u003c/span\u003e\u003c/sup\u003e. Additionally, synergy among the CrusII, VIIb, and VIIIa has been frequently documented in earlier studies. As examples, VIIb and VIIIa were coactivated in both working memory and attention tasks\u003csup\u003e\u003cspan citationid=\"CR75\" class=\"CitationRef\"\u003e75\u003c/span\u003e\u003c/sup\u003e and jointly damaged in schizophrenia patients with persistent auditory verbal hallucinations\u003csup\u003e\u003cspan citationid=\"CR76\" class=\"CitationRef\"\u003e76\u003c/span\u003e\u003c/sup\u003e; CrusII and VIIb exhibited volume reduction in patients with ASD\u003csup\u003e\u003cspan citationid=\"CR77\" class=\"CitationRef\"\u003e77\u003c/span\u003e\u003c/sup\u003e, PTSD\u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e, and schizophrenia\u003csup\u003e\u003cspan citationid=\"CR78\" class=\"CitationRef\"\u003e78\u003c/span\u003e\u003c/sup\u003e; and CrusII and VIIIa encoded anticipatory mechanisms for dexterous object manipulation\u003csup\u003e\u003cspan citationid=\"CR79\" class=\"CitationRef\"\u003e79\u003c/span\u003e\u003c/sup\u003e. Therefore, the genetic parcellation of the posterior cerebellar lobe may provide a new framework for understanding the function and damage of its sub-lobules.\u003c/p\u003e \u003cp\u003eIn PheWAS, we identified significant associations between PGSs of all cerebellar volumetric traits and mental health phenotypes, indicating that all cerebellar subregions may play a role in emotion processing. These findings confirmed and extended previous observations of the involvement of several cerebellar subregions in emotion processing\u003csup\u003e\u003cspan citationid=\"CR80\" class=\"CitationRef\"\u003e80\u003c/span\u003e,\u003cspan citationid=\"CR81\" class=\"CitationRef\"\u003e81\u003c/span\u003e\u003c/sup\u003e. Consistent with earlier studies\u003csup\u003e\u003cspan citationid=\"CR79\" class=\"CitationRef\"\u003e79\u003c/span\u003e,\u003cspan citationid=\"CR82\" class=\"CitationRef\"\u003e82\u003c/span\u003e\u003c/sup\u003e, we found that left CrusII and bilateral VIIIa were associated with cognitive performance. Therefore, these results verified the role of the cerebellum in cognitive and emotional processing. We also identified genetic correlations or colocalization of cerebellar volumetric traits with AD, PD, ALS, PTSD, BD, SCZ, and depression, offering potential genetic substrates underlying the observed cerebellar impairments in these disorders\u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e,\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e,\u003cspan additionalcitationids=\"CR15\" citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e–\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e,\u003cspan citationid=\"CR73\" class=\"CitationRef\"\u003e73\u003c/span\u003e,\u003cspan citationid=\"CR76\" class=\"CitationRef\"\u003e76\u003c/span\u003e\u003c/sup\u003e. In the multi-trait colocalization analyses, we found two gene (\u003cem\u003eARHGAP27\u003c/em\u003e and \u003cem\u003eSNX31\u003c/em\u003e) whose cerebellar expression showed colocalization with both cerebellar volumetric traits and NPDs. For instance, the expression of \u003cem\u003eARHGAP27\u003c/em\u003e was colocalized with cerebellar white matter volumes and three NPDs (AD, PD, and PTSD). \u003cem\u003eARHGAP27\u003c/em\u003e is a Rho GTPase-activating protein, acting as a key player in neurodegeneration\u003csup\u003e\u003cspan citationid=\"CR83\" class=\"CitationRef\"\u003e83\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eSeveral limitations should be noted when interpreting our findings. First, the study population primarily comprised EUR individuals, resulting in unstable evaluation of heritability in individuals of EAS and AFR. Additionally, a lenient threshold was used to select AFR participants, which does not accurately represent the whole population of African ancestry. Second, although we have harmonized the cerebellar volumetric traits to minimize the influence of imaging scanners, confounding factors may still exist and limit the interpretation of our results. Finally, even if we have used the sample-weighted method to obtain the cross-ancestry LD reference, we cannot precisely estimate the LD structures in cross-ancestry population, which may influence analyses that require LD information.\u003c/p\u003e \u003cp\u003eIn conclusion, we confirmed that precise phenotyping is essential in neuroimaging genetic analyses. Through precise phenotyping and extensive multi-ancestry analyses, we identified 407 new loci associated with cerebellar volumetric traits, including sex-specific and ancestry-specific associations. We developed the first genetically informed atlas of the cerebellum and offered genetic insights into the associations between the cerebellum and cognition, emotion, and neuropsychiatric disorders. These findings may enhance our understanding of the genetic architecture of the cerebellum and its links to brain functions and disorders.\u003c/p\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003cdiv id=\"Sec20\" class=\"Section3\"\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec22\" class=\"Section2\"\u003e \u003cdiv id=\"Sec23\" class=\"Section3\"\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec24\" class=\"Section2\"\u003e \u003cdiv id=\"Sec25\" class=\"Section3\"\u003e \u003c/div\u003e \u003cdiv id=\"Sec26\" class=\"Section3\"\u003e \u003c/div\u003e \u003cdiv id=\"Sec27\" class=\"Section3\"\u003e \u003c/div\u003e \u003c/div\u003e "},{"header":"Methods","content":"\u003ch2\u003eParticipants\u003c/h2\u003e\u003cp\u003eThe participants included in this study were sourced from three distinct datasets: the UK Biobank (UKBB; \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.ukbiobank.ac.uk/\u003c/span\u003e\u003cspan address=\"https://www.ukbiobank.ac.uk/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) study\u003csup\u003e\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u003c/sup\u003e, the Chinese Imaging Genetics (CHIMGEN; \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://chimgen.tmu.edu.cn/\u003c/span\u003e\u003cspan address=\"http://chimgen.tmu.edu.cn/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) study\u003csup\u003e\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u003c/sup\u003e, and the Adolescent Brain Cognitive Development (ABCD; \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://abcdstudy.org/\u003c/span\u003e\u003cspan address=\"https://abcdstudy.org/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) study\u003csup\u003e\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e\u003c/sup\u003e. The UKBB study enrolled approximately 500,000 participants aged 40 to 69 years from 22 research centers throughout the United Kingdom\u003csup\u003e\u003cspan citationid=\"CR84\" class=\"CitationRef\"\u003e84\u003c/span\u003e\u003c/sup\u003e. This study was approved by the National Health Service Research Ethics Service (21/NW/0157), and written informed consent was obtained from each participant. We accessed to the data under application number 75556. The CHIMGEN study recruited 7,306 healthy Chinese Han participants aged 18 to 30 years from 32 research centers, which was approved by the Medical Research Ethics Committees of all institutions, and all participants provided written informed consent\u003csup\u003e\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u003c/sup\u003e. The ABCD study is a longitudinal cohort comprising over 10,000 children aged 9 to 10 years at their baseline assessment from 21 research centers. This study received approval from a central Institutional Review Board (IRB) at the University of California, San Diego for the majority of research centers, or from a local IRB for a few research centers. All parents or caregivers provided the written informed consent and all children provided the written assent\u003csup\u003e\u003cspan citationid=\"CR85\" class=\"CitationRef\"\u003e85\u003c/span\u003e\u003c/sup\u003e. We accessed to the data under application ID 17607. For each dataset, participants with qualified structural magnetic resonance imaging (MRI) and genomic data were included in this study.\u003c/p\u003e\u003ch2\u003eData preparation\u003c/h2\u003e\u003cp\u003e\u003cb\u003eUKBB.\u003c/b\u003e Of 487,207 participants who passed the initial quality control (QC) of genetic data\u003csup\u003e\u003cspan citationid=\"CR84\" class=\"CitationRef\"\u003e84\u003c/span\u003e\u003c/sup\u003e, we excluded 651 participants with sex chromosome aneuploidy and 186 with sex mismatch, remaining 486,370 participants with qualified genomic data, including 44,179 with structural MRI data. We employed CerebNet to segment the cerebellum into 30 substructures\u003csup\u003e\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e. After excluding 919 participants with brain tumors, imaging artifacts, incomplete cerebellum coverage, or segmentation error, the volume of each cerebellar substructure was obtained for the 43,260 remaining participants. Following the removal of 125 participants whose substructure volume was outside five times the median absolute deviation (MAD) from the median, 43,135 participants were retained. As most participants reported European ancestry (EUR), we utilized SNPweights\u003csup\u003e\u003cspan citationid=\"CR86\" class=\"CitationRef\"\u003e86\u003c/span\u003e\u003c/sup\u003e to verify the ancestry of each participant against the EUR reference panel from the 1000 Genomes Project phase 3 (1KGP), ensuring that the EUR proportion was greater than 80%\u003csup\u003e87\u003c/sup\u003e. Ultimately, we included 40,826 EUR participants from the UKBB dataset. Using the criteria of minor allele frequency (MAF) ≥ 0.005, imputation information score (info) ≥ 0.6, and \u003cem\u003eP\u003c/em\u003e ≥ 1 × 10\u003csup\u003e− 7\u003c/sup\u003e in Hardy-Weinberg equilibrium (HWE), we finally included 9,683,756 autosomal and 317,880 X-chromosomal bi-allelic variants.\u003c/p\u003e\u003cp\u003e \u003cb\u003eCHIMGEN.\u003c/b\u003e Among the 7,306 CHIMGEN participants, 7,195 with DNA samples were genotyped by Illumina ASA-750K (Asian Screening Array) that was specially designed for Asians. Details for the sample- and variant-level QC, genetic principal component analysis (PCA), and genetic data imputation are provided in our previous study\u003csup\u003e\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e. All 7,163 participants with qualified genetic data also had structural MRI data. After excluding 61 participants without qualified structural MRI data, we used CerebNet\u003csup\u003e\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e to calculate the cerebellar substructure volumes for the remaining 7,102 participants. Following the removal of 19 participants whose subregion volume exceeded five times the MAD from the median, we included 7,083 East Asian (EAS) participants from the CHIMGEN dataset in the genetic analyses. The analyses were performed for 8,790,144 imputed autosomal and 227,164 X-chromosomal bi-allelic variants (MAF ≥ 0.005, info ≥ 0.6, and \u003cem\u003eP\u003c/em\u003e\u003csub\u003e\u003cem\u003eHWE\u003c/em\u003e\u003c/sub\u003e ≥ 1 × 10\u003csup\u003e− 7\u003c/sup\u003e).\u003c/p\u003e\u003cp\u003e \u003cb\u003eABCD.\u003c/b\u003e Among the 11,760 participants from the ABCD dataset, 11,101 remained after genetic QC, including 10,660 with qualified structural MRI data. We utilized CerebNet\u003csup\u003e\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e to calculate cerebellar substructure volumes and excluded 264 participants whose substructure volume was more than five times the MAD from the median, resulting in 10,396 participants. We also utilized SNPweights\u003csup\u003e\u003cspan citationid=\"CR86\" class=\"CitationRef\"\u003e86\u003c/span\u003e\u003c/sup\u003e to identify EUR participants with an EUR proportion over 80%\u003csup\u003e87\u003c/sup\u003e. As African descent (AFR) was also commonly seen in ABCD participants, we employed the criteria from a prior study\u003csup\u003e\u003cspan citationid=\"CR88\" class=\"CitationRef\"\u003e88\u003c/span\u003e\u003c/sup\u003e to identify the participants with African descent using the reference panels for African, European, and East Asian from 1KGP. A participant was classified as having African descent if AFR proportion exceeded 5%, along with EUR proportion \u0026lt; 80% and EAS proportion \u0026lt; 5%. We finally included 5,808 EUR and 3,354 AFR participants from the ABCD dataset. To maintain consistency with the genomic build of the UKBB and CHIMGEN datasets, we converted variants of the ABCD dataset from GRCh38/hg38 to GRCh37/hg19. With the same criteria (MAF ≥ 0.005, info ≥ 0.6, and \u003cem\u003eP\u003c/em\u003e\u003csub\u003e\u003cem\u003eHWE\u003c/em\u003e\u003c/sub\u003e ≥ 1 × 10\u003csup\u003e− 7\u003c/sup\u003e), we finally included 9,064,819 autosomal and 278,985 X-chromosomal bi-allelic variants for the 5,808 EUR participants, and 7,162,933 autosomal and 232,441 X-chromosomal bi-allelic variants for the 3,354 AFR participants.\u003c/p\u003e\u003ch2\u003eCerebellar segmentation\u003c/h2\u003e\u003cp\u003eIn this study, we employed CerebNet\u003csup\u003e\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e, a fast and reliable deep-learning pipeline, to automatically segment the cerebellum into 30 substructures using brain structural MRI data without any preprocessing. CerebNet uses a FastSurferCNN deep-learning model tailored for cerebellar segmentation, showing superior accuracy, test-retest reliability, extensibility, and sensitivity compared to other segmentation methods\u003csup\u003e\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e. These 30 cerebellar substructures comprised five for the vermis (VI, VII, VIII, IX, and X), 20 for the hemisphere (left and right I-IV, V, VI, CrusI, CrusII, VIIb, VIIIa, VIIIb, IX, and X), along with five global ones (left white matter, right white matter, left cortex, right cortex, and vermis). CerebNet can automatically calculate the volumes of these 30 cerebellar substructures. To test whether CerebNet outperforms traditional segmentation methods in genetic analyses of cerebellar volumetric traits, we also computed the volumes of these cerebellar substructures by integrating the FreeSurfer automatic segmentation\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e and the spatially unbiased infra-tentorial template (SUIT)\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e, referred to as SUIT + FS. SUIT is the most detailed and widely used cerebellar atlas, subdividing the cerebellum into eight vermis and 20 hemisphere substructures. By combining three substructures of vermis VII into a single substructure and two substructures of vermis VIII into one substructure, we derived 25 cerebellar substructures similar to those from CerebNet. The vermis volume can be calculated by summing all the vermis substructures, whereas FreeSurfer-ASEG can estimate the volumes of bilateral white matter and cortex. Thus, we established the correspondence in 30 cerebellar volumetric traits between CerebNet and SUIT + FS.\u003c/p\u003e\u003cp\u003eFor UKBB participants, we downloaded the volume data (SUIT + FS) for cerebellar substructures from the database\u003csup\u003e\u003cspan citationid=\"CR89\" class=\"CitationRef\"\u003e89\u003c/span\u003e\u003c/sup\u003e. For CHIMGEN participants, we used the volume data for cerebellar substructures from SUIT + FS that have been utilized in a prior study\u003csup\u003e\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e. For ABCD participants, we used the CHIMGEN pipeline (SUIT + FS) to obtain the volume data for cerebellar substructures. Specifically, we used computational anatomy toolbox (CAT 12, version r1364, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://dbm.neuro.uni-jena.de/cat\u003c/span\u003e\u003cspan address=\"http://dbm.neuro.uni-jena.de/cat\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) to preprocess brain structural MRI data. After correcting for imaging inhomogeneity due to B1-field bias, brain structural images were segmented into gray matter (GM), white matter, and cerebrospinal fluid using an adaptive Maximum A Posterior (MAP) technique\u003csup\u003e\u003cspan citationid=\"CR90\" class=\"CitationRef\"\u003e90\u003c/span\u003e\u003c/sup\u003e. Based on the 10,294 ABCD participants, the Diffeomorphic Anatomical Registration Through Exponentiated Lie Algebra (DARTEL) algorithm\u003csup\u003e\u003cspan citationid=\"CR91\" class=\"CitationRef\"\u003e91\u003c/span\u003e\u003c/sup\u003e was used to create a population-specific GM template in the Montreal Neurological Institute (MNI) space. The segmented GM images were normalized to the population-specific GM template using the DARTEL algorithm and resampled into cubic voxels of 1.5 mm. Modulation was applied to the GM images to preserve absolute GM volume (GMV). Based on the voxel-wise GMV map, the SUIT approach in combination with FreeSurfer-ASEG was used to calculate the volumes of cerebellar substructures.\u003c/p\u003e\u003cp\u003eTo evaluate the reliability of CerebNet segmentation, we utilized this method to compute the volumes of cerebellar substructures based on the brain structural MRI data obtained at two separate time points from the same participants (CHIMGEN: n = 26; UKBB: n = 2,944; and ABCD: n = 4,076). For each test-retest dataset, we calculated the intraclass correlation coefficient (ICC) of the volume data of each substructure obtained at the two time points across all participants. We found high reliability in the three datasets (CHIMGEN: ICC = 0.62–0.97; UKBB: ICC = 0.88–0.96; ABCD: ICC = 0.87–0.98; \u003cb\u003etable S2\u003c/b\u003e).\u003c/p\u003e\u003cp\u003eBrain structural MRI data from the UKBB, CHIMGEN, and ABCD datasets were acquired using various scanners, which inevitably introduces bias into genetic analyses. Compared to including imaging sites as a covariate in the regression model, ComBat harmonization is a more effective method for eliminating the MRI scanner effects while preserving biological variability\u003csup\u003e\u003cspan citationid=\"CR92\" class=\"CitationRef\"\u003e92\u003c/span\u003e\u003c/sup\u003e. We tested the impact of ComBat harmonization in two participants who visited different centers and were scanned at 28 MRI scanners utilized for the CHIMGEN data acquisition. In each participant, we utilized CerebNet to calculate the volumes of 30 cerebellar substructures based on the MRI data obtained from each scanner, and used the coefficient of variation (CV) to evaluate the between-scanner variations of these traits. In both participants, CVs of these volumetric traits prior to harmonization were significantly reduced (Wilcoxon rank-sum test; Participant 1: \u003cem\u003eP\u003c/em\u003e = 8.32 × 10\u003csup\u003e− 10\u003c/sup\u003e; Participant 2: \u003cem\u003eP\u003c/em\u003e = 1.40 × 10\u003csup\u003e− 5\u003c/sup\u003e) after harmonization (\u003cb\u003eFig. S8\u003c/b\u003e). Additionally, as the skewed data distribution would violate the assumption of normality when using a linear regression model for genetic analyses, quantile normalization was applied to these cerebellar volumetric traits.\u003c/p\u003e\u003ch2\u003eComparing genetic findings between cerebellar segmentation methods\u003c/h2\u003e\u003cp\u003eAlthough CerebNet demonstrates greater accuracy, test-retest reliability, and sensitivity compared to other segmentation methods, including SUIT + FS\u003csup\u003e22\u003c/sup\u003e, its advantages in genetic analyses are still unclear. Here, we investigated whether cerebellar substructure volumes obtained from CerebNet exhibit higher heritability, more genetic associations, and improved predictive performance of polygenic scores (PGS) compared to those derived from SUIT + FS.\u003c/p\u003e\u003cp\u003e \u003cb\u003eHeritability.\u003c/b\u003e We used GCTA-GREML\u003csup\u003e\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u003c/sup\u003e to estimate single nucleotide polymorphism (SNP)-based heritability of each cerebellar substructure volume obtained from the two segmentation approaches in 40,826 UKBB\u003csub\u003eEUR\u003c/sub\u003e, 7,083 CHIMGEN\u003csub\u003eEAS\u003c/sub\u003e, 5,808 ABCD\u003csub\u003eEUR\u003c/sub\u003e, and 3,354 ABCD\u003csub\u003eAFR\u003c/sub\u003e participants, respectively. In these analyses, we controlled for age at scanning, genetically determined sex, age × sex, total intracranial volume (TIV), genetic principal components (PCs; top 40 for UKBB, 10 for CHIMGEN, and 32 for ABCD), and genetic batch (only for UKBB and ABCD). After calculating the genetic relationship matrix (GRM) for autosomal SNPs included in each GWAS, we estimated the variance of the trait explained by these SNPs, referred to as SNP-based heritability. The Wilcoxon rank-sum test was used to compare the difference in heritability of these traits obtained from CerebNet and SUIT + FS.\u003c/p\u003e\u003cp\u003e \u003cb\u003eGenetic discovery.\u003c/b\u003e To test whether CerebNet improves genetic discovery, we performed GWASs for 30 cerebellar substructure volumes obtained from CerebNet and SUIT + FS in 40,826 UKBB\u003csub\u003eEUR\u003c/sub\u003e participants. We generated a sparse GRM based on the full-dense GRM from the heritability estimation using a cutoff of 0.05. Based on the sparse GRM (\u003cb\u003etable S3\u003c/b\u003e), we employed the mixed linear model (MLM) implemented in fastGWA\u003csup\u003e\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u003c/sup\u003e to conduct GWAS (additive effect) with the same covariates as heritability estimate. We used the Wilcoxon rank-sum test to compare the number, \u003cem\u003eP\u003c/em\u003e-value, and power of variant-trait associations (\u003cem\u003eP\u003c/em\u003e \u0026lt; 5 × 10\u003csup\u003e− 8\u003c/sup\u003e) derived from the two segmentation methods.\u003c/p\u003e\u003cp\u003e \u003cb\u003ePGS prediction.\u003c/b\u003e To test whether CerebNet can improve PGS prediction compared to SUIT + FS, we employed the UKBB\u003csub\u003eEUR\u003c/sub\u003e data as the base dataset and the CHIMGEN\u003csub\u003eEAS\u003c/sub\u003e, ABCD\u003csub\u003eEUR\u003c/sub\u003e, and ABCD\u003csub\u003eAFR\u003c/sub\u003e data as three distinct target datasets. PRSice-2\u003csup\u003e28\u003c/sup\u003e was used to obtain the best-fit PGS model at the optimal \u003cem\u003eP\u003c/em\u003e-value threshold for each cerebellar volumetric trait obtained from each segmentation method based on the UKBB-GWAS summary statistics for the trait. The best-fit model was defined as the PGS model with the largest explained variance (R\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e) for the volumetric trait in the base dataset. The best-fit PGS model was then applied to each target dataset to calculate the R\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e of PGS for the trait in the target dataset. Then we used Wilcoxon rank-sum test to compare R\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e values of the 30 cerebellar volumetric traits obtained from CerebNet and SUIT + FS in each target dataset.\u003c/p\u003e\u003ch2\u003eComprehensive GWASs\u003c/h2\u003e\u003cp\u003eAfter showing that CerebNet improves the genetic discovery of cerebellar substructure volumes, we performed comprehensive GWASs on the autosomal and X-chromosomal variants to investigate the genetic architecture of 31 cerebellar volumetric traits derived from CerebNet by further including the total cerebellar volume. GWASs included sex-combined and sex-stratified analyses, further dividing into univariate and multivariate analyses. In each subcategory, we conducted four single-dataset GWASs (UKBB\u003csub\u003eEUR\u003c/sub\u003e-GWAS, ABCD\u003csub\u003eEUR\u003c/sub\u003e-GWAS, CHIMGEN\u003csub\u003eEAS\u003c/sub\u003e-GWAS, and ABCD\u003csub\u003eAFR\u003c/sub\u003e-GWAS), one EUR-GWAS meta-analysis, and two cross-ancestry (EUR-EAS and EUR-EAS-AFR) GWAS meta-analyses using the same covariates as heritability estimate. We considered sex-combined univariate GWASs as the main results, while conducting other GWASs for supplementary insights. We reported the study-wide significant associations (\u003cem\u003eP\u003c/em\u003e \u0026lt; 1.61 × 10\u003csup\u003e− 9\u003c/sup\u003e) in univariate GWASs and the genome-wide significant associations (\u003cem\u003eP\u003c/em\u003e \u0026lt; 5 × 10\u003csup\u003e− 8\u003c/sup\u003e) in multivariate GWASs. The participants, genetic variants, and GWAS parameters are presented in \u003cb\u003etable S3\u003c/b\u003e.\u003c/p\u003e\u003cp\u003e \u003cb\u003eSex-combined single-dataset univariate GWAS.\u003c/b\u003e In each of the sex-combined single-ancestry dataset (UKBB\u003csub\u003eEUR\u003c/sub\u003e, CHIMGEN\u003csub\u003eEAS\u003c/sub\u003e, ABCD\u003csub\u003eEUR\u003c/sub\u003e, and ABCD\u003csub\u003eAFR\u003c/sub\u003e), we used MLM in fastGWA\u003csup\u003e\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u003c/sup\u003e to conduct univariate GWASs for the 31 cerebellar volumetric traits while controlling for the predefined covariates.\u003c/p\u003e\u003cp\u003e \u003cb\u003eSex-combined univariate EUR-GWASs.\u003c/b\u003e Based on the summary data of sex-combined UKBB\u003csub\u003eEUR\u003c/sub\u003e-GWASs and ABCD\u003csub\u003eEUR\u003c/sub\u003e-GWASs for cerebellar volumetric traits, we utilized the inverse-variance-weighted (IVW) fixed effect model implemented in METAL\u003csup\u003e\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e\u003c/sup\u003e to conduct EUR-GWAS meta-analyses.\u003c/p\u003e\u003cp\u003e \u003cb\u003eSex-combined univariate cross-ancestry GWASs.\u003c/b\u003e For each cerebellar volumetric trait, we used the IVW fixed effect model in METAL\u003csup\u003e\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e\u003c/sup\u003e to conduct EUR-EAS-GWAS meta-analysis based on the summary data of EUR-GWAS and CHIMGEN\u003csub\u003eEAS\u003c/sub\u003e-GWAS. We also used the same model to conduct EUR-EAS-AFR-GWAS meta-analysis based on the summary data of EUR-GWAS, CHIMGEN\u003csub\u003eEAS\u003c/sub\u003e-GWAS and ABCD\u003csub\u003eAFR\u003c/sub\u003e-GWAS.\u003c/p\u003e\u003cp\u003e \u003cb\u003eSex-stratified GWASs.\u003c/b\u003e We repeated the above-mentioned GWASs in females and males, while controlling for the predefined covariates except for genetically determined sex and age × sex.\u003c/p\u003e\u003cp\u003e \u003cb\u003eM\u003c/b\u003e \u003cb\u003eultivariate GWASs.\u003c/b\u003e Considering the correlation structure of cerebellar volumetric traits, we employed C-GWAS\u003csup\u003e\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e\u003c/sup\u003e (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://github.com/Fun-Gene/CGWAS\u003c/span\u003e\u003cspan address=\"https://github.com/Fun-Gene/CGWAS\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) to combine GWAS summary data of correlated traits to enhance the discovery of loci associated with cerebellar volumetric traits. C-GWAS was conducted on the volumes of 27 non-overlapping cerebellar substructures (\u003cem\u003eP\u003c/em\u003e \u0026lt; 5 × 10\u003csup\u003e− 8\u003c/sup\u003e).\u003c/p\u003e\u003cp\u003e \u003cb\u003ePopulation stratification.\u003c/b\u003e We estimated the population stratification for each univariate GWAS using genomic control inflation factor (λ\u003csub\u003eGC\u003c/sub\u003e) and linkage disequilibrium score regression (LDSC) intercept\u003csup\u003e\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u003c/sup\u003e. λ\u003csub\u003eGC\u003c/sub\u003e was calculated as the median of the resulting chi-squared (χ\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e) test statistics (z scores) divided by 0.4549, which is the expected median of the χ\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e distribution with one degree of freedom. As high λ\u003csub\u003eGC\u003c/sub\u003e indicates either genomic inflation or polygenicity, we used LDSC intercepts to identify genomic inflation based on LD scores. We considered no population stratification if the LDSC intercept is close to one.\u003c/p\u003e\u003cp\u003e \u003cb\u003eLinkage disequilibrium (LD) references.\u003c/b\u003e We used imputed genotype data from 3,354 ABCD\u003csub\u003eAFR\u003c/sub\u003e, 7,083 CHIMGEN, and 40,826 UKBB participants with qualified genetic and imaging data to construct LD references for African ancestry (LD\u003csub\u003eAFR\u003c/sub\u003e), East Asian ancestry (LD\u003csub\u003eEAS\u003c/sub\u003e) and European ancestry (LD\u003csub\u003eEUR\u003c/sub\u003e), respectively. We further constructed two merged LD references (LD\u003csub\u003eEUR−EAS\u003c/sub\u003e and LD\u003csub\u003eEUR−EAS−AFR\u003c/sub\u003e) using the sample-weighted method based on the above-mentioned data. We also constructed female-specific and male-specific LD references using the same strategies. The LD references used in each analysis are presented in \u003cb\u003etable S3\u003c/b\u003e.\u003c/p\u003e\u003cp\u003e \u003cb\u003eIndependent associations and loci.\u003c/b\u003e For each GWAS, the matched LD reference was used to identify independent variant-trait associations by PLINK clumping\u003csup\u003e\u003cspan citationid=\"CR93\" class=\"CitationRef\"\u003e93\u003c/span\u003e\u003c/sup\u003e with the following steps: (1) all significant variants were included in a list of prioritized variants; (2) the most significant one was defined as the first lead variant (independent variant), and variants within 1 Mb from or in LD with (r\u003csup\u003e2\u003c/sup\u003e \u0026gt; 0.1) the lead variant were clumped; (3) the remaining variants formed a new list, and then step (2) was repeated; and (4) the iterative process stopped until the list was empty. We identified independent locus-trait associations by: (1) creating loci for all independent variants by adding 1 Mb to both sides; (2) merging overlapping loci; (3) merging loci if an independent variant of one locus was in LD (r\u003csup\u003e2\u003c/sup\u003e \u0026gt; 0.1) with independent variant of another locus; and (4) merging loci overlapped with the major histocompatibility complex (MHC) or 8p23.1 region into one locus. The associations of the remaining loci with this trait were defined as independent locus-trait associations. For all variant-trait associations, we merged the lead variants with an LD r\u003csup\u003e2\u003c/sup\u003e \u0026gt; 0.1 or within 1 Mb. The remaining variants were defined as independent signals. We created loci for all locus-trait associations by merging 1 Mb to both sides, and then merged the overlapping loci to identify independent loci.\u003c/p\u003e\u003cp\u003e \u003cb\u003ePooling significant associations.\u003c/b\u003e As the sample size of EUR participants was at least four times larger than those of other ancestries in this study, we used the LD\u003csub\u003eEUR\u003c/sub\u003e to pool the significant associations from all GWASs. We used the above-mentioned strategies to identify independent variant-trait associations, locus-trait associations, signals, and loci.\u003c/p\u003e\u003cp\u003e \u003cb\u003eIdentifying new associations and loci.\u003c/b\u003e As prior GWASs for cerebellar volumetric traits\u003csup\u003e\u003cspan additionalcitationids=\"CR19 CR20\" citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e–\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e were conducted mainly in EUR participants, we utilized the same strategies and LD\u003csub\u003eEUR\u003c/sub\u003e reference to pool the GWAS results (\u003cem\u003eP\u003c/em\u003e \u0026lt; 1.61 × 10\u003csup\u003e− 9\u003c/sup\u003e). We identified 614 known variant-trait associations, 447 locus-trait associations, 217 signals, and 135 loci. Based on these results, we defined a new variant-trait association if the variant was 1 Mb away from and not in LD (r\u003csup\u003e2\u003c/sup\u003e \u0026lt; 0.1) with any variants of the trait in the list of known variant-trait associations; a new locus-trait association when the locus was not overlapped with loci of known locus-trait associations and all lead variants in the locus were not in LD (r\u003csup\u003e2\u003c/sup\u003e \u0026lt; 0.1) with any lead variants in the loci of all known locus-trait associations; a new independent signals when the signals was 1 Mb away from and not in LD (r\u003csup\u003e2\u003c/sup\u003e \u0026lt; 0.1) with any known signals; and a new locus when the locus was not overlapped with any known loci and all lead variants in the locus were not in LD (r\u003csup\u003e2\u003c/sup\u003e \u0026lt; 0.1) with any lead variants in known loci.\u003c/p\u003e\u003ch2\u003eAllelic effect heterogeneity\u003c/h2\u003e\u003cp\u003eThe Cochran's Q test (CQ-test) was employed to assess the allelic-effect heterogeneity of pooled variant-trait associations (\u003cem\u003eP\u003c/em\u003e \u0026lt; 1.61 × 10\u003csup\u003e− 9\u003c/sup\u003e) among the sex-combined univariate EUR-GWAS, CHIIMGEN\u003csub\u003eEAS\u003c/sub\u003e-GWAS and ABCD\u003csub\u003eAFR\u003c/sub\u003e-GWAS. CQ-test was conducted for variants included in all three GWASs. We defined ancestry-shared associations as variant-trait associations with \u003cem\u003eP\u003c/em\u003e ≥ 0.05 in CQ-test and ancestry-specific associations as those with \u003cem\u003eP\u003c/em\u003e\u003csub\u003e\u003cem\u003ec\u003c/em\u003e\u003c/sub\u003e \u0026lt; 0.05 (Bonferroni correction for the number of tested associations) in CQ-test. CQ-test was also used to identify sex-shared and sex-specific associations in EAS, AFR, and EUR, respectively. With LD\u003csub\u003eEAS\u003c/sub\u003e, we pooled the variant-trait associations from sex-stratified univariate CHIIMGEN\u003csub\u003eEAS\u003c/sub\u003e-GWASs. The CQ-test was used to test the allelic-effect heterogeneity for the pooled associations between females and males. The same procedures were then applied to sex-stratified univariate EUR-GWASs, and ABCD\u003csub\u003eAFR\u003c/sub\u003e-GWAS. We defined variant-trait associations with \u003cem\u003eP\u003c/em\u003e ≥ 0.05 in CQ-test as sex-shared associations and those with \u003cem\u003eP\u003c/em\u003e\u003csub\u003e\u003cem\u003ec\u003c/em\u003e\u003c/sub\u003e \u0026lt; 0.05 (Bonferroni correction for the number of tested associations) in CQ-test as sex-specific associations.\u003c/p\u003e\u003ch2\u003eFunctional annotations\u003c/h2\u003e\u003cp\u003e \u003cb\u003eGenomic location and functional consequence.\u003c/b\u003e For each GWAS, we used ANNOVAR\u003csup\u003e\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e\u003c/sup\u003e to categorize the variants from significant variant-trait associations based on their genic position, such as exon, intron, untranslated region, and intergenic region. We used the combined annotation-dependent depletion (CADD) score to prioritize deleterious and pathogenic variants with scores above 20\u003csup\u003e45\u003c/sup\u003e. We also employed the RegulomeDB score to prioritize the variants in regulatory elements\u003csup\u003e\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003e \u003cb\u003eStatistical fine-mapping.\u003c/b\u003e We employed the matched LD reference to perform statistical fine-mapping to identify the 95% credible set and the causal variants with posterior probability (PP) above 0.8 for each locus of significant locus-trait associations (\u003cem\u003eP\u003c/em\u003e \u0026lt; 1.61 × 10\u003csup\u003e− 9\u003c/sup\u003e). We employed SuSiE-R\u003csup\u003e\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e,\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e\u003c/sup\u003e to conduct single-ancestry fine-mapping based on EUR-GWASs, while we utilized SuSiEx\u003csup\u003e\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e\u003c/sup\u003e to performed two-ancestry fine-mapping based on EUR-EAS cross-ancestry GWASs and three-ancestry fine-mapping based on EUR-EAS-AFR cross-ancestry GWASs. To compare the performance of the three fine-mapping strategies, we applied these methods to all locus-trait associations identified by any of the three GWASs. We aligned the 95% credible sets with one causal variant assumption across the three fine-mapping strategies, and used the Wilcoxon rank-sum test to compare the size and max PP of 95% credible sets.\u003c/p\u003e\u003cp\u003e \u003cb\u003eColocalization with gene expression.\u003c/b\u003e Coloc\u003csup\u003e\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e\u003c/sup\u003e (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://chr1swallace.github.io/coloc/\u003c/span\u003e\u003cspan address=\"https://chr1swallace.github.io/coloc/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) was used to perform Bayesian colocalization to identify the shared loci between those associated with cerebellar volumetric traits in EUR-GWAS and expression quantitative trait loci (eQTLs; \u003cem\u003eP \u0026lt;\u003c/em\u003e 5 × 10\u003csup\u003e− 8\u003c/sup\u003e) of human cerebellar tissue. With the default priors (\u003cem\u003eP\u003c/em\u003e\u003csub\u003e\u003cem\u003e1\u003c/em\u003e\u003c/sub\u003e = 1 × 10\u003csup\u003e− 4\u003c/sup\u003e, \u003cem\u003eP\u003c/em\u003e\u003csub\u003e\u003cem\u003e2\u003c/em\u003e\u003c/sub\u003e = 1 × 10\u003csup\u003e− 4\u003c/sup\u003e, and \u003cem\u003eP\u003c/em\u003e\u003csub\u003e\u003cem\u003e12\u003c/em\u003e\u003c/sub\u003e = 1 × 10\u003csup\u003e− 5\u003c/sup\u003e), we considered genetic colocalization when PP.H4 (the posterior probability of shared causal variant) was greater than 0.8. The cis-eQTL data of cerebellar tissue (715 tissue samples from 492 EUR individuals) were downloaded from MetaBrain\u003csup\u003e\u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e51\u003c/span\u003e\u003c/sup\u003e (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://metabrain.nl/\u003c/span\u003e\u003cspan address=\"https://metabrain.nl/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). We defined prioritized genes as those colocalized with cerebellar volumetric traits (PP.H4 \u0026gt; 0.8).\u003c/p\u003e\u003cp\u003e \u003cb\u003ePathway enrichment.\u003c/b\u003e We performed the pathway enrichment analyses by inputting the prioritized genes to g:Profiler website (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://biit.cs.ut.ee/gprofiler/gost\u003c/span\u003e\u003cspan address=\"https://biit.cs.ut.ee/gprofiler/gost\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e)\u003csup\u003e56\u003c/sup\u003e. From the pre-specified GO pathways (15,472 biological processes)\u003csup\u003e\u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e57\u003c/span\u003e\u003c/sup\u003e, we identified the significant pathways at \u003cem\u003eP\u003c/em\u003e\u003csub\u003e\u003cem\u003ec\u003c/em\u003e\u003c/sub\u003e \u0026lt; 0.05 (Benjamini-Hochberg FDR corrected).\u003c/p\u003e\u003ch2\u003eGenetic clustering of cerebellar substructures\u003c/h2\u003e\u003cp\u003eTo cluster the 27 non-overlapping cerebellar substructures, we used the high-definition likelihood (HDL) method\u003csup\u003e\u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e58\u003c/span\u003e\u003c/sup\u003e to compute genetic correlations of these substructures based on EUR-GWASs. HDL is an extension of LDSC that can effectively reduce the variance of estimates, thereby increasing precision. We used the imputed HapMap3 reference panel from UKBB after excluding the MHC region. Based on the genetic correlation matrix, we performed genomic structural equation modelling (gSEM)\u003csup\u003e\u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e59\u003c/span\u003e\u003c/sup\u003e. First, we used the common factor model to test whether a single factor was sufficient for genetic clustering. Second, we tried an anatomical parcellation scheme. If these two models cannot achieve an acceptable fit, we used the nScree function of the nFactor R package to conduct exploratory factor analyses (EFA) to identify the optimal number of factors. The resulting factor model from EFA was validated using confirmatory factor analysis (CFA). The model fit was assessed using comparative fit index (CFI) and standardized root-mean-squared residual (SRMR). An acceptable model fit is defined as CFI \u0026gt; 0.9 and SRMR \u0026lt; 0.1, while a good model fit is defined as CFI \u0026gt; 0.95 and SRMR \u0026lt; 0.05\u003csup\u003e59\u003c/sup\u003e.\u003c/p\u003e\u003ch2\u003eGenetic relationships with other phenotypes\u003c/h2\u003e\u003cp\u003eTo investigate the genetic relationships between cerebellar volumetric traits and other cerebellum-related phenotypes, we performed three types of analyses: (1) we conducted the phenome-wide association study (PheWAS) to identify the associations of the PGS of each cerebellar volumetric trait with 198 cognitive and mental health phenotypes; (2) we investigated the genetic correlations between 31 cerebellar volumetric traits and ten neuropsychiatric disorders (NPDs); and (3) we calculated the genetic colocalizations between 31 cerebellar volumetric traits and ten NPDs.\u003c/p\u003e\u003cp\u003e \u003cb\u003ePheWAS.\u003c/b\u003e We used the PHESANT package\u003csup\u003e\u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e60\u003c/span\u003e\u003c/sup\u003e in R to conduct PheWAS to identify associations (\u003cem\u003eP\u003c/em\u003e \u0026lt; 0.05/31/198 = 8.15 × 10\u003csup\u003e− 6\u003c/sup\u003e, Bonferroni corrected) between PGSs of 31 cerebellar volumetric traits and 198 cognitive and mental health phenotypes (\u003cb\u003etable S38\u003c/b\u003e)\u003csup\u003e\u003cspan citationid=\"CR94\" class=\"CitationRef\"\u003e94\u003c/span\u003e\u003c/sup\u003e, while controlling for age, sex, the first 40 genetic PCs, genotype batch, and assessment centers. The base dataset comprised EUR participants included in EUR-GWASs for cerebellar volumetric traits, while the target dataset consisted of 371,558 unrelated UKBB Caucasians not included in EUR-GWASs. Based on the EUR-GWAS summary statistics of cerebellar volumetric traits and UKBB\u003csub\u003eEUR\u003c/sub\u003e-LD reference, we used PRS-CS\u003csup\u003e\u003cspan citationid=\"CR95\" class=\"CitationRef\"\u003e95\u003c/span\u003e\u003c/sup\u003e to calculate the PGSs of cerebellar volumetric traits for each participant in target datasets. PRS-CS used a Bayesian regression framework and used a continuous shrinkage (CS) prior on effect size of genetic variant, which is robust to varying genetic architectures.\u003c/p\u003e\u003cp\u003e \u003cb\u003eGenetic correlation.\u003c/b\u003e HDL was used to calculate genetic correlations (\u003cem\u003eP\u003c/em\u003e \u0026lt; 0.05/31/10 = 1.61 × 10\u003csup\u003e− 4\u003c/sup\u003e, Bonferroni corrected) between 31 cerebellar volumetric traits and 10 NPDs based on EUR-GWAS summary statistics (\u003cb\u003etable S40\u003c/b\u003e) and the UKBB reference panel (excluding the MHC region) provided by HDL\u003csup\u003e\u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e58\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003e \u003cb\u003eGenetic colocalization.\u003c/b\u003e Coloc\u003csup\u003e\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e\u003c/sup\u003e (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://chr1swallace.github.io/coloc/\u003c/span\u003e\u003cspan address=\"https://chr1swallace.github.io/coloc/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) was used to identify the loci shared by locus-trait associations (\u003cem\u003eP\u003c/em\u003e \u0026lt; 1.61 × 10\u003csup\u003e− 9\u003c/sup\u003e) of each cerebellar volumetric trait and locus-disease associations (\u003cem\u003eP\u003c/em\u003e \u0026lt; 5 × 10\u003csup\u003e− 8\u003c/sup\u003e) of each NPD based on the EUR-GWAS summary statistics. With the default priors (\u003cem\u003eP\u003c/em\u003e\u003csub\u003e\u003cem\u003e1\u003c/em\u003e\u003c/sub\u003e = 1 × 10\u003csup\u003e− 4\u003c/sup\u003e, \u003cem\u003eP\u003c/em\u003e\u003csub\u003e\u003cem\u003e2\u003c/em\u003e\u003c/sub\u003e = 1 × 10\u003csup\u003e− 4\u003c/sup\u003e, and \u003cem\u003eP\u003c/em\u003e\u003csub\u003e\u003cem\u003e12\u003c/em\u003e\u003c/sub\u003e = 1 × 10\u003csup\u003e− 5\u003c/sup\u003e), we considered colocalization if PP.H4 was greater than 0.8. For the locus-trait associations of cerebellar volumetric traits that were colocalized with both cerebellar gene expression and NPDs (PP.H4 \u0026gt; 0.8), we further performed multi-trait colocalization using hypothesis prioritization in multi-trait colocalization (HyPrColoc)\u003csup\u003e\u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e64\u003c/span\u003e\u003c/sup\u003e to identify the colocalization (PP \u0026gt; 0.8) of the three traits.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003eData availability:\u003c/p\u003e\n\u003cp\u003eAll GWAS summary statistics in this study are available. GWAS summary statistics will be deposited in the GWAS catalog.\u003c/p\u003e\n\u003cp\u003eCode availability:\u003c/p\u003e\n\u003cp\u003eThis paper does not report original code. We made use of publicly available software and tools in this study. All relevant software and code are described in the text and can be found at references cited. All codes used to generate results are publicly available (https://github.com/xuehui2014/Deep-learning-segmentation-and-multi-ancestry-GWAS-enhance-genetic-discovery-of-the-cerebellum).\u003c/p\u003e\n\u003cp\u003eAcknowledgements:\u003c/p\u003e\n\u003cp\u003eWe are grateful to participants and researchers of CHIMGEN, UKBB and ABCD, who generously donated their time to make this resource available. We acknowledge funding from the National Natural Science Foundation of China (82430063, 82030053 to C.Y., 82402253 to J. F., 8247052 to W. Q.).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAuthor contributions\u003c/p\u003e\n\u003cp\u003eC. Y., H. X., and J. F. designed the study and wrote the manuscript. H. X., J. F., H. D., S. W. analyzed the data. C. Y., W. Q., M. L. supervised this work. All authors critically reviewed the manuscript.\u003c/p\u003e\n\u003cp\u003eCompeting interests\u003c/p\u003e\n\u003cp\u003eAuthors declare that they have no competing interests.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eWang, V. Y. \u0026amp; Zoghbi, H. Y. Genetic regulation of cerebellar development. \u003cem\u003eNature Reviews Neuroscience\u003c/em\u003e \u003cstrong\u003e2\u003c/strong\u003e, 484-491, doi:Doi 10.1038/35081558 (2001).\u003c/li\u003e\n\u003cli\u003eSathyanesan, A.\u003cem\u003e et al.\u003c/em\u003e Emerging connections between cerebellar development, behaviour and complex brain disorders. \u003cem\u003eNature Reviews Neuroscience\u003c/em\u003e \u003cstrong\u003e20\u003c/strong\u003e, 298-313, doi:10.1038/s41583-019-0152-2 (2019).\u003c/li\u003e\n\u003cli\u003eGlickstein, M. What does the cerebellum really do? \u003cem\u003eCurr Biol\u003c/em\u003e \u003cstrong\u003e17\u003c/strong\u003e, R824-R827, doi:DOI 10.1016/j.cub.2007.08.009 (2007).\u003c/li\u003e\n\u003cli\u003eStrick, P. L., Dum, R. P. \u0026amp; Fiez, J. A. Cerebellum and Nonmotor Function. \u003cem\u003eAnnual Review of Neuroscience\u003c/em\u003e \u003cstrong\u003e32\u003c/strong\u003e, 413-434, doi:10.1146/annurev.neuro.31.060407.125606 (2009).\u003c/li\u003e\n\u003cli\u003eBuckner, R. L. The Cerebellum and Cognitive Function: 25 Years of Insight from Anatomy and Neuroimaging. \u003cem\u003eNeuron\u003c/em\u003e \u003cstrong\u003e80\u003c/strong\u003e, 807-815, doi:10.1016/j.neuron.2013.10.044 (2013).\u003c/li\u003e\n\u003cli\u003eStoodley, C. J.\u003cem\u003e et al.\u003c/em\u003e Altered cerebellar connectivity in autism and cerebellar-mediated rescue of autism-related behaviors in mice. \u003cem\u003eNat Neurosci\u003c/em\u003e \u003cstrong\u003e20\u003c/strong\u003e, 1744-1751, doi:10.1038/s41593-017-0004-1 (2017).\u003c/li\u003e\n\u003cli\u003eSeidman, L. J., Valera, E. M. \u0026amp; Makris, N. Structural brain imaging of attention-deficit/hyperactivity disorder. \u003cem\u003eBiol Psychiatry\u003c/em\u003e \u003cstrong\u003e57\u003c/strong\u003e, 1263-1272, doi:10.1016/j.biopsych.2004.11.019 (2005).\u003c/li\u003e\n\u003cli\u003ePicard, H., Amado, I., Mouchet-Mages, S., Oli\u0026eacute;, J. P. \u0026amp; Krebs, M. O. The role of the cerebellum in schizophrenia: an update of clinical, cognitive, and functional evidences. \u003cem\u003eSchizophrenia Bull\u003c/em\u003e \u003cstrong\u003e34\u003c/strong\u003e, 155-172, doi:10.1093/schbul/sbm049 (2008).\u003c/li\u003e\n\u003cli\u003eDiedrichsen, J., Balsters, J. H., Flavell, J., Cussans, E. \u0026amp; Ramnani, N. A probabilistic MR atlas of the human cerebellum. \u003cem\u003eNeuroimage\u003c/em\u003e \u003cstrong\u003e46\u003c/strong\u003e, 39-46, doi:10.1016/j.neuroimage.2009.01.045 (2009).\u003c/li\u003e\n\u003cli\u003eFischl, B.\u003cem\u003e et al.\u003c/em\u003e Whole brain segmentation: Automated labeling of neuroanatomical structures in the human brain. \u003cem\u003eNeuron\u003c/em\u003e \u003cstrong\u003e33\u003c/strong\u003e, 341-355, doi:Doi 10.1016/S0896-6273(02)00569-X (2002).\u003c/li\u003e\n\u003cli\u003eHuggins, A. A.\u003cem\u003e et al.\u003c/em\u003e Smaller total and subregional cerebellar volumes in posttraumatic stress disorder: a mega-analysis by the ENIGMA-PGC PTSD workgroup. \u003cem\u003eMolecular Psychiatry\u003c/em\u003e \u003cstrong\u003e29\u003c/strong\u003e, 611-623, doi:10.1038/s41380-023-02352-0 (2024).\u003c/li\u003e\n\u003cli\u003eFu, X., Chen, Y., Luo, X., Ide, J. S. \u0026amp; Li, C. R. Gray matter volumetric correlates of the polygenic risk of depression: A study of the Human Connectome Project data. \u003cem\u003eEur Neuropsychopharmacol\u003c/em\u003e \u003cstrong\u003e87\u003c/strong\u003e, 2-12, doi:10.1016/j.euroneuro.2024.06.004 (2024).\u003c/li\u003e\n\u003cli\u003eMooney, M. A.\u003cem\u003e et al.\u003c/em\u003e Smaller total brain volume but not subcortical structure volume related to common genetic risk for ADHD. \u003cem\u003ePsychol Med\u003c/em\u003e \u003cstrong\u003e51\u003c/strong\u003e, 1279-1288, doi:10.1017/s0033291719004148 (2021).\u003c/li\u003e\n\u003cli\u003eKerestes, R.\u003cem\u003e et al.\u003c/em\u003e Cerebellar Volume and Disease Staging in Parkinson\u0026apos;s Disease: An ENIGMA-PD Study. \u003cem\u003eMovement Disord\u003c/em\u003e \u003cstrong\u003e38\u003c/strong\u003e, 2269-2281, doi:10.1002/mds.29611 (2023).\u003c/li\u003e\n\u003cli\u003eGellersen, H. M., Guell, X. \u0026amp; Sami, S. Differential vulnerability of the cerebellum in healthy ageing and Alzheimer\u0026apos;s disease. \u003cem\u003eNeuroimage Clin\u003c/em\u003e \u003cstrong\u003e30\u003c/strong\u003e, 102605, doi:10.1016/j.nicl.2021.102605 (2021).\u003c/li\u003e\n\u003cli\u003eQiu, T.\u003cem\u003e et al.\u003c/em\u003e Precentral degeneration and cerebellar compensation in amyotrophic lateral sclerosis: A multimodal MRI analysis. \u003cem\u003eHuman Brain Mapping\u003c/em\u003e \u003cstrong\u003e40\u003c/strong\u003e, 3464-3474, doi:10.1002/hbm.24609 (2019).\u003c/li\u003e\n\u003cli\u003eStrike, L. T.\u003cem\u003e et al.\u003c/em\u003e Heritability of cerebellar subregion volumes in adolescent and young adult twins. \u003cem\u003eHuman Brain Mapping\u003c/em\u003e \u003cstrong\u003e45\u003c/strong\u003e, doi:10.1002/hbm.26717 (2024).\u003c/li\u003e\n\u003cli\u003eChambers, T.\u003cem\u003e et al.\u003c/em\u003e Genetic common variants associated with cerebellar volume and their overlap with mental disorders: a study on 33,265 individuals from the UK-Biobank. \u003cem\u003eMolecular Psychiatry\u003c/em\u003e, doi:10.1038/s41380-022-01443-8 (2022).\u003c/li\u003e\n\u003cli\u003eTissink, E.\u003cem\u003e et al.\u003c/em\u003e Genome-wide association study of cerebellar volume provides insights into heritable mechanisms underlying brain development and mental health. \u003cem\u003eCommunications Biology\u003c/em\u003e \u003cstrong\u003e5\u003c/strong\u003e, 710, doi:10.1038/s42003-022-03672-7 (2022).\u003c/li\u003e\n\u003cli\u003eSmith, S. M.\u003cem\u003e et al.\u003c/em\u003e An expanded set of genome-wide association studies of brain imaging phenotypes in UK Biobank. \u003cem\u003eNature Neuroscience\u003c/em\u003e \u003cstrong\u003e24\u003c/strong\u003e, 737-745, doi:10.1038/s41593-021-00826-4 (2021).\u003c/li\u003e\n\u003cli\u003eFu, J. L.\u003cem\u003e et al.\u003c/em\u003e Cross-ancestry genome-wide association studies of brain imaging phenotypes. \u003cem\u003eNature Genetics\u003c/em\u003e, doi:10.1038/s41588-024-01766-y (2024).\u003c/li\u003e\n\u003cli\u003eFaber, J.\u003cem\u003e et al.\u003c/em\u003e CerebNet: A fast and reliable deep-learning pipeline for detailed cerebellum sub-segmentation. \u003cem\u003eNeuroimage\u003c/em\u003e \u003cstrong\u003e264\u003c/strong\u003e, doi:10.1016/j.neuroimage.2022.119703 (2022).\u003c/li\u003e\n\u003cli\u003eXu, Q.\u003cem\u003e et al.\u003c/em\u003e CHIMGEN: a Chinese imaging genetics cohort to enhance cross-ethnic and cross-geographic brain research. \u003cem\u003eMol Psychiatry\u003c/em\u003e \u003cstrong\u003e25\u003c/strong\u003e, 517-529, doi:10.1038/s41380-019-0627-6 (2020).\u003c/li\u003e\n\u003cli\u003eMiller, K. L.\u003cem\u003e et al.\u003c/em\u003e Multimodal population brain imaging in the UK Biobank prospective epidemiological study. \u003cem\u003eNature Neuroscience\u003c/em\u003e \u003cstrong\u003e19\u003c/strong\u003e, 1523-1536, doi:10.1038/nn.4393 (2016).\u003c/li\u003e\n\u003cli\u003eCasey, B. J.\u003cem\u003e et al.\u003c/em\u003e The Adolescent Brain Cognitive Development (ABCD) study: Imaging acquisition across 21 sites. \u003cem\u003eDev Cogn Neuros-Neth\u003c/em\u003e \u003cstrong\u003e32\u003c/strong\u003e, 43-54, doi:10.1016/j.dcn.2018.03.001 (2018).\u003c/li\u003e\n\u003cli\u003eYang, J.\u003cem\u003e et al.\u003c/em\u003e Genetic variance estimation with imputed variants finds negligible missing heritability for human height and body mass index. \u003cem\u003eNat Genet\u003c/em\u003e \u003cstrong\u003e47\u003c/strong\u003e, 1114-1120, doi:10.1038/ng.3390 (2015).\u003c/li\u003e\n\u003cli\u003eJiang, L. D.\u003cem\u003e et al.\u003c/em\u003e A resource-efficient tool for mixed model association analysis of large-scale data. \u003cem\u003eNature Genetics\u003c/em\u003e \u003cstrong\u003e51\u003c/strong\u003e, 1749-+, doi:10.1038/s41588-019-0530-8 (2019).\u003c/li\u003e\n\u003cli\u003eChoi, S. W. \u0026amp; O\u0026apos;Reilly, P. F. PRSice-2: Polygenic Risk Score software for biobank-scale data. \u003cem\u003eGigascience\u003c/em\u003e \u003cstrong\u003e8\u003c/strong\u003e, doi:10.1093/gigascience/giz082 (2019).\u003c/li\u003e\n\u003cli\u003eBulik-Sullivan, B. K.\u003cem\u003e et al.\u003c/em\u003e LD Score regression distinguishes confounding from polygenicity in genome-wide association studies. \u003cem\u003eNature Genetics\u003c/em\u003e \u003cstrong\u003e47\u003c/strong\u003e, 291-+, doi:10.1038/ng.3211 (2015).\u003c/li\u003e\n\u003cli\u003eChen, X.\u003cem\u003e et al.\u003c/em\u003e Transcriptomic mapping uncovers Purkinje neuron plasticity driving learning. \u003cem\u003eNature\u003c/em\u003e \u003cstrong\u003e605\u003c/strong\u003e, 722-727, doi:10.1038/s41586-022-04711-3 (2022).\u003c/li\u003e\n\u003cli\u003eWiller, C. J., Li, Y. \u0026amp; Abecasis, G. R. METAL: fast and efficient meta-analysis of genomewide association scans. \u003cem\u003eBioinformatics\u003c/em\u003e \u003cstrong\u003e26\u003c/strong\u003e, 2190-2191, doi:10.1093/bioinformatics/btq340 (2010).\u003c/li\u003e\n\u003cli\u003eVan Battum, E.\u003cem\u003e et al.\u003c/em\u003e Plexin-B2 controls the timing of differentiation and the motility of cerebellar granule neurons. \u003cem\u003eElife\u003c/em\u003e \u003cstrong\u003e10\u003c/strong\u003e, doi:10.7554/eLife.60554 (2021).\u003c/li\u003e\n\u003cli\u003eSepp, M.\u003cem\u003e et al.\u003c/em\u003e Cellular development and evolution of the mammalian cerebellum. \u003cem\u003eNature\u003c/em\u003e \u003cstrong\u003e625\u003c/strong\u003e, doi:10.1038/s41586-023-06884-x (2024).\u003c/li\u003e\n\u003cli\u003eXiong, Z.\u003cem\u003e et al.\u003c/em\u003e Combining genome-wide association studies highlight novel loci involved in human facial variation. \u003cem\u003eNat Commun\u003c/em\u003e \u003cstrong\u003e13\u003c/strong\u003e, 7832, doi:10.1038/s41467-022-35328-9 (2022).\u003c/li\u003e\n\u003cli\u003eDas, G.\u003cem\u003e et al.\u003c/em\u003e EphA5 and EphA6: regulation of neuronal and spine morphology. \u003cem\u003eCell Biosci\u003c/em\u003e \u003cstrong\u003e6\u003c/strong\u003e, 48, doi:10.1186/s13578-016-0115-5 (2016).\u003c/li\u003e\n\u003cli\u003eCaballero Oteyza, A.\u003cem\u003e et al.\u003c/em\u003e Motor protein mutations cause a new form of hereditary spastic paraplegia. \u003cem\u003eNeurology\u003c/em\u003e \u003cstrong\u003e82\u003c/strong\u003e, 2007-2016, doi:10.1212/wnl.0000000000000479 (2014).\u003c/li\u003e\n\u003cli\u003eSiddiqui, N.\u003cem\u003e et al.\u003c/em\u003e Force generation of KIF1C is impaired by pathogenic mutations. \u003cem\u003eCurr Biol\u003c/em\u003e \u003cstrong\u003e32\u003c/strong\u003e, 3862-3870.e3866, doi:10.1016/j.cub.2022.07.029 (2022).\u003c/li\u003e\n\u003cli\u003eRamirez, M.\u003cem\u003e et al.\u003c/em\u003e Temporal analysis of enhancers during mouse cerebellar development reveals dynamic and novel regulatory functions. \u003cem\u003eElife\u003c/em\u003e \u003cstrong\u003e11\u003c/strong\u003e, doi:10.7554/eLife.74207 (2022).\u003c/li\u003e\n\u003cli\u003eDe Luca, A.\u003cem\u003e et al.\u003c/em\u003e Sonic hedgehog patterning during cerebellar development. \u003cem\u003eCell Mol Life Sci\u003c/em\u003e \u003cstrong\u003e73\u003c/strong\u003e, 291-303, doi:10.1007/s00018-015-2065-1 (2016).\u003c/li\u003e\n\u003cli\u003eKy\u0026ouml;stil\u0026auml;, K.\u003cem\u003e et al.\u003c/em\u003e A SEL1L mutation links a canine progressive early-onset cerebellar ataxia to the endoplasmic reticulum-associated protein degradation (ERAD) machinery. \u003cem\u003ePLoS Genet\u003c/em\u003e \u003cstrong\u003e8\u003c/strong\u003e, e1002759, doi:10.1371/journal.pgen.1002759 (2012).\u003c/li\u003e\n\u003cli\u003eWang, G., Sarkar, A., Carbonetto, P. \u0026amp; Stephens, M. A simple new approach to variable selection in regression, with application to genetic fine mapping. \u003cem\u003eJ R Stat Soc Series B Stat Methodol\u003c/em\u003e \u003cstrong\u003e82\u003c/strong\u003e, 1273-1300, doi:10.1111/rssb.12388 (2020).\u003c/li\u003e\n\u003cli\u003eZou, Y., Carbonetto, P., Wang, G. \u0026amp; Stephens, M. Fine-mapping from summary data with the \u0026quot;Sum of Single Effects\u0026quot; model. \u003cem\u003ePlos Genetics\u003c/em\u003e \u003cstrong\u003e18\u003c/strong\u003e, doi:10.1371/journal.pgen.1010299 (2022).\u003c/li\u003e\n\u003cli\u003eYuan, K.\u003cem\u003e et al.\u003c/em\u003e Fine-mapping across diverse ancestries drives the discovery of putative causal variants underlying human complex traits and diseases. \u003cem\u003eNature Genetics\u003c/em\u003e, doi:10.1038/s41588-024-01870-z (2024).\u003c/li\u003e\n\u003cli\u003eWang, K., Li, M. \u0026amp; Hakonarson, H. ANNOVAR: functional annotation of genetic variants from high-throughput sequencing data. \u003cem\u003eNucleic Acids Res\u003c/em\u003e \u003cstrong\u003e38\u003c/strong\u003e, e164, doi:10.1093/nar/gkq603 (2010).\u003c/li\u003e\n\u003cli\u003eRentzsch, P., Witten, D., Cooper, G. M., Shendure, J. \u0026amp; Kircher, M. CADD: predicting the deleteriousness of variants throughout the human genome. \u003cem\u003eNucleic Acids Research\u003c/em\u003e \u003cstrong\u003e47\u003c/strong\u003e, D886-D894, doi:10.1093/nar/gky1016 (2019).\u003c/li\u003e\n\u003cli\u003eDong, S. C.\u003cem\u003e et al.\u003c/em\u003e Annotating and prioritizing human non-coding variants with RegulomeDB v.2. \u003cem\u003eNature Genetics\u003c/em\u003e \u003cstrong\u003e55\u003c/strong\u003e, 724-726, doi:10.1038/s41588-023-01365-3 (2023).\u003c/li\u003e\n\u003cli\u003eZaninello, M.\u003cem\u003e et al.\u003c/em\u003e CLUH maintains functional mitochondria and translation in motoneuronal axons and prevents peripheral neuropathy. \u003cem\u003eScience Advances\u003c/em\u003e \u003cstrong\u003e10\u003c/strong\u003e, doi:10.1126/sciadv.adn2050 (2024).\u003c/li\u003e\n\u003cli\u003eWang, B. P., Harrison, W., Overbeek, P. A. \u0026amp; Zheng, H. Transposon mutagenesis with coat color genotyping identifies an essential role for Skor2 in sonic hedgehog signaling and cerebellum development. \u003cem\u003eDevelopment\u003c/em\u003e \u003cstrong\u003e138\u003c/strong\u003e, 4487-4497, doi:10.1242/dev.067264 (2011).\u003c/li\u003e\n\u003cli\u003eAdams, N. C., Tomoda, T., Cooper, M., Dietz, G. \u0026amp; Hatten, M. E. Mice that lack astrotactin have slowed neuronal migration. \u003cem\u003eDevelopment\u003c/em\u003e \u003cstrong\u003e129\u003c/strong\u003e, 965-972, doi:10.1242/dev.129.4.965 (2002).\u003c/li\u003e\n\u003cli\u003eWallace, C. Eliciting priors and relaxing the single causal variant assumption in colocalisation analyses. \u003cem\u003ePlos Genetics\u003c/em\u003e \u003cstrong\u003e16\u003c/strong\u003e, doi:10.1371/journal.pgen.1008720 (2020).\u003c/li\u003e\n\u003cli\u003ede Klein, N.\u003cem\u003e et al.\u003c/em\u003e Brain expression quantitative trait locus and network analyses reveal downstream effects and putative drivers for brain-related diseases. \u003cem\u003eNature Genetics\u003c/em\u003e \u003cstrong\u003e55\u003c/strong\u003e, 377-+, doi:10.1038/s41588-023-01300-6 (2023).\u003c/li\u003e\n\u003cli\u003eOkbay, A.\u003cem\u003e et al.\u003c/em\u003e Polygenic prediction of educational attainment within and between families from genome-wide association analyses in 3 million individuals. \u003cem\u003eNature Genetics\u003c/em\u003e \u003cstrong\u003e54\u003c/strong\u003e, 437-+, doi:10.1038/s41588-022-01016-z (2022).\u003c/li\u003e\n\u003cli\u003eDahl, A.\u003cem\u003e et al.\u003c/em\u003e Phenotype integration improves power and preserves specificity in biobank-based genetic studies of major depressive disorder. \u003cem\u003eNature Genetics\u003c/em\u003e \u003cstrong\u003e55\u003c/strong\u003e, 2082-2093, doi:10.1038/s41588-023-01559-9 (2023).\u003c/li\u003e\n\u003cli\u003eRico, B.\u003cem\u003e et al.\u003c/em\u003e Control of axonal branching and synapse formation by focal adhesion kinase. \u003cem\u003eNature Neuroscience\u003c/em\u003e \u003cstrong\u003e7\u003c/strong\u003e, 1059-1069, doi:10.1038/nn1317 (2004).\u003c/li\u003e\n\u003cli\u003eDi Gregorio, E.\u003cem\u003e et al.\u003c/em\u003e A de novo X;8 translocation creates a PTK2-THOC2 gene fusion with THOC2 expression knockdown in a patient with psychomotor retardation and congenital cerebellar hypoplasia. \u003cem\u003eJournal of Medical Genetics\u003c/em\u003e \u003cstrong\u003e50\u003c/strong\u003e, 543-551, doi:10.1136/jmedgenet-2013-101542 (2013).\u003c/li\u003e\n\u003cli\u003eKolberg, L.\u003cem\u003e et al.\u003c/em\u003e g:Profiler-interoperable web service for functional enrichment analysis and gene identifier mapping (2023 update). \u003cem\u003eNucleic Acids Research\u003c/em\u003e \u003cstrong\u003e51\u003c/strong\u003e, W207-W212, doi:10.1093/nar/gkad347 (2023).\u003c/li\u003e\n\u003cli\u003eAshburner, M.\u003cem\u003e et al.\u003c/em\u003e Gene Ontology: tool for the unification of biology. \u003cem\u003eNature Genetics\u003c/em\u003e \u003cstrong\u003e25\u003c/strong\u003e, 25-29, doi:Doi 10.1038/75556 (2000).\u003c/li\u003e\n\u003cli\u003eNing, Z., Pawitan, Y. \u0026amp; Shen, X. High-definition likelihood inference of genetic correlations across human complex traits. \u003cem\u003eNat Genet\u003c/em\u003e \u003cstrong\u003e52\u003c/strong\u003e, 859-864, doi:10.1038/s41588-020-0653-y (2020).\u003c/li\u003e\n\u003cli\u003eGrotzinger, A. D.\u003cem\u003e et al.\u003c/em\u003e Genomic structural equation modelling provides insights into the multivariate genetic architecture of complex traits. \u003cem\u003eNat Hum Behav\u003c/em\u003e \u003cstrong\u003e3\u003c/strong\u003e, 513-525, doi:10.1038/s41562-019-0566-x (2019).\u003c/li\u003e\n\u003cli\u003eMillard, L. A. C., Davies, N. M., Gaunt, T. R., Smith, G. D. \u0026amp; Tilling, K. Software Application Profile: PHESANT: a tool for performing automated phenome scans in UK Biobank. \u003cem\u003eInt J Epidemiol\u003c/em\u003e \u003cstrong\u003e47\u003c/strong\u003e, 29-35, doi:10.1093/ije/dyx204 (2018).\u003c/li\u003e\n\u003cli\u003eMolinari, M.\u003cem\u003e et al.\u003c/em\u003e Cerebellum and procedural learning: evidence from focal cerebellar lesions. \u003cem\u003eBrain\u003c/em\u003e \u003cstrong\u003e120 ( Pt 10)\u003c/strong\u003e, 1753-1762, doi:10.1093/brain/120.10.1753 (1997).\u003c/li\u003e\n\u003cli\u003eLevisohn, L., Cronin-Golomb, A. \u0026amp; Schmahmann, J. D. Neuropsychological consequences of cerebellar tumour resection in children - Cerebellar cognitive affective syndrome in a paediatric population. \u003cem\u003eBrain\u003c/em\u003e \u003cstrong\u003e123\u003c/strong\u003e, 1041-1050, doi:DOI 10.1093/brain/123.5.1041 (2000).\u003c/li\u003e\n\u003cli\u003eSullivan, E. V.\u003cem\u003e et al.\u003c/em\u003e Disturbed Cerebellar Growth Trajectories in Adolescents Who Initiate Alcohol Drinking. \u003cem\u003eBiol Psychiatry\u003c/em\u003e \u003cstrong\u003e87\u003c/strong\u003e, 632-644, doi:10.1016/j.biopsych.2019.08.026 (2020).\u003c/li\u003e\n\u003cli\u003eFoley, C. N.\u003cem\u003e et al.\u003c/em\u003e A fast and efficient colocalization algorithm for identifying shared genetic risk factors across multiple traits. \u003cem\u003eNature Communications\u003c/em\u003e \u003cstrong\u003e12\u003c/strong\u003e, doi:10.1038/s41467-020-20885-8 (2021).\u003c/li\u003e\n\u003cli\u003eVieira, N., Rito, T., Correia-Neves, M. \u0026amp; Sousa, N. Sorting Out Sorting Nexins Functions in the Nervous System in Health and Disease. \u003cem\u003eMol Neurobiol\u003c/em\u003e \u003cstrong\u003e58\u003c/strong\u003e, 4070-4106, doi:10.1007/s12035-021-02388-9 (2021).\u003c/li\u003e\n\u003cli\u003eManolio, T. A.\u003cem\u003e et al.\u003c/em\u003e Finding the missing heritability of complex diseases. \u003cem\u003eNature\u003c/em\u003e \u003cstrong\u003e461\u003c/strong\u003e, 747-753, doi:10.1038/nature08494 (2009).\u003c/li\u003e\n\u003cli\u003eBahi-Buisson, N.\u003cem\u003e et al.\u003c/em\u003e GPR56-related bilateral frontoparietal polymicrogyria: further evidence for an overlap with the cobblestone complex. \u003cem\u003eBrain\u003c/em\u003e \u003cstrong\u003e133\u003c/strong\u003e, 3194-3209, doi:10.1093/brain/awq259 (2010).\u003c/li\u003e\n\u003cli\u003ePeterson, R. E.\u003cem\u003e et al.\u003c/em\u003e Genome-wide Association Studies in Ancestrally Diverse Populations: Opportunities, Methods, Pitfalls, and Recommendations. \u003cem\u003eCell\u003c/em\u003e \u003cstrong\u003e179\u003c/strong\u003e, 589-603, doi:10.1016/j.cell.2019.08.051 (2019).\u003c/li\u003e\n\u003cli\u003eSch\u0026ouml;ls, L., Bauer, P., Schmidt, T., Schulte, T. \u0026amp; Riess, O. Autosomal dominant cerebellar ataxias: clinical features, genetics, and pathogenesis. \u003cem\u003eLancet Neurol\u003c/em\u003e \u003cstrong\u003e3\u003c/strong\u003e, 291-304, doi:10.1016/s1474-4422(04)00737-9 (2004).\u003c/li\u003e\n\u003cli\u003eSteele, C. J. \u0026amp; Chakravarty, M. M. Gray-matter structural variability in the human cerebellum: Lobule-specific differences across sex and hemisphere. \u003cem\u003eNeuroimage\u003c/em\u003e \u003cstrong\u003e170\u003c/strong\u003e, 164-173, doi:10.1016/j.neuroimage.2017.04.066 (2018).\u003c/li\u003e\n\u003cli\u003eCoste de Bagneaux, P.\u003cem\u003e et al.\u003c/em\u003e A homozygous missense variant in CACNB4 encoding the auxiliary calcium channel beta4 subunit causes a severe neurodevelopmental disorder and impairs channel and non-channel functions. \u003cem\u003ePLoS Genet\u003c/em\u003e \u003cstrong\u003e16\u003c/strong\u003e, e1008625, doi:10.1371/journal.pgen.1008625 (2020).\u003c/li\u003e\n\u003cli\u003eNakatani, H., Nakamura, Y. \u0026amp; Okanoya, K. Respective Involvement of the Right Cerebellar Crus I and II in Syntactic and Semantic Processing for Comprehension of Language. \u003cem\u003eCerebellum\u003c/em\u003e \u003cstrong\u003e22\u003c/strong\u003e, 739-755, doi:10.1007/s12311-022-01451-y (2023).\u003c/li\u003e\n\u003cli\u003eChen, G.\u003cem\u003e et al.\u003c/em\u003e Abnormal cerebellum-DMN regions connectivity in unmedicated bipolar II disorder. \u003cem\u003eJ Affect Disord\u003c/em\u003e \u003cstrong\u003e243\u003c/strong\u003e, 441-447, doi:10.1016/j.jad.2018.09.076 (2019).\u003c/li\u003e\n\u003cli\u003eBruchhage, M. M. K.\u003cem\u003e et al.\u003c/em\u003e Drum training induces long-term plasticity in the cerebellum and connected cortical thickness. \u003cem\u003eSci Rep\u003c/em\u003e \u003cstrong\u003e10\u003c/strong\u003e, 10116, doi:10.1038/s41598-020-65877-2 (2020).\u003c/li\u003e\n\u003cli\u003eBrissenden, J. A., Tobyne, S. M., Halko, M. A. \u0026amp; Somers, D. C. Stimulus-Specific Visual Working Memory Representations in Human Cerebellar Lobule VIIb/VIIIa. \u003cem\u003eJ Neurosci\u003c/em\u003e \u003cstrong\u003e41\u003c/strong\u003e, 1033-1045, doi:10.1523/jneurosci.1253-20.2020 (2021).\u003c/li\u003e\n\u003cli\u003eCierpka, M.\u003cem\u003e et al.\u003c/em\u003e Cerebellar Contributions to Persistent Auditory Verbal Hallucinations in Patients with Schizophrenia. \u003cem\u003eCerebellum\u003c/em\u003e \u003cstrong\u003e16\u003c/strong\u003e, 964-972, doi:10.1007/s12311-017-0874-5 (2017).\u003c/li\u003e\n\u003cli\u003eWang, Y.\u003cem\u003e et al.\u003c/em\u003e Abnormalities in cerebellar subregions\u0026apos; volume and cerebellocerebral structural covariance in autism spectrum disorder. \u003cem\u003eAutism Res\u003c/em\u003e \u003cstrong\u003e18\u003c/strong\u003e, 83-97, doi:10.1002/aur.3287 (2025).\u003c/li\u003e\n\u003cli\u003eLaidi, C.\u003cem\u003e et al.\u003c/em\u003e Cerebellar parcellation in schizophrenia and bipolar disorder. \u003cem\u003eActa Psychiatr Scand\u003c/em\u003e \u003cstrong\u003e140\u003c/strong\u003e, 468-476, doi:10.1111/acps.13087 (2019).\u003c/li\u003e\n\u003cli\u003eMarneweck, M., Barany, D. A., Santello, M. \u0026amp; Grafton, S. T. Neural Representations of Sensorimotor Memory- and Digit Position-Based Load Force Adjustments Before the Onset of Dexterous Object Manipulation. \u003cem\u003eJ Neurosci\u003c/em\u003e \u003cstrong\u003e38\u003c/strong\u003e, 4724-4737, doi:10.1523/jneurosci.2588-17.2018 (2018).\u003c/li\u003e\n\u003cli\u003eWang, X.\u003cem\u003e et al.\u003c/em\u003e Revisiting reward impairments in schizophrenia spectrum disorders: a systematic review and meta-analysis for neuroimaging findings. \u003cem\u003ePsychol Med\u003c/em\u003e \u003cstrong\u003e53\u003c/strong\u003e, 7189-7202, doi:10.1017/s0033291723000703 (2023).\u003c/li\u003e\n\u003cli\u003eErnst, T. M.\u003cem\u003e et al.\u003c/em\u003e The cerebellum is involved in processing of predictions and prediction errors in a fear conditioning paradigm. \u003cem\u003eElife\u003c/em\u003e \u003cstrong\u003e8\u003c/strong\u003e, doi:10.7554/eLife.46831 (2019).\u003c/li\u003e\n\u003cli\u003ePeterburs, J., Blevins, L. C., Sheu, Y. S. \u0026amp; Desmond, J. E. Cerebellar contributions to sequence prediction in verbal working memory. \u003cem\u003eBrain Struct Funct\u003c/em\u003e \u003cstrong\u003e224\u003c/strong\u003e, 485-499, doi:10.1007/s00429-018-1784-0 (2019).\u003c/li\u003e\n\u003cli\u003eStankiewicz, T. R. \u0026amp; Linseman, D. A. Rho family GTPases: key players in neuronal development neuronal survival, and neurodegeneration. \u003cem\u003eFront Cell Neurosci\u003c/em\u003e \u003cstrong\u003e8\u003c/strong\u003e, doi:10.3389/fncel.2014.00314 (2014).\u003c/li\u003e\n\u003cli\u003eBycroft, C.\u003cem\u003e et al.\u003c/em\u003e The UK Biobank resource with deep phenotyping and genomic data. \u003cem\u003eNature\u003c/em\u003e \u003cstrong\u003e562\u003c/strong\u003e, 203-+, doi:10.1038/s41586-018-0579-z (2018).\u003c/li\u003e\n\u003cli\u003eAuchter, A. M.\u003cem\u003e et al.\u003c/em\u003e A description of the ABCD organizational structure and communication framework. \u003cem\u003eDev Cogn Neuros-Neth\u003c/em\u003e \u003cstrong\u003e32\u003c/strong\u003e, 8-15, doi:10.1016/j.dcn.2018.04.003 (2018).\u003c/li\u003e\n\u003cli\u003eChen, C. Y.\u003cem\u003e et al.\u003c/em\u003e Improved ancestry inference using weights from external reference panels. \u003cem\u003eBioinformatics\u003c/em\u003e \u003cstrong\u003e29\u003c/strong\u003e, 1399-1406, doi:10.1093/bioinformatics/btt144 (2013).\u003c/li\u003e\n\u003cli\u003eMukamel, R. E.\u003cem\u003e et al.\u003c/em\u003e Repeat polymorphisms underlie top genetic risk loci for glaucoma and colorectal cancer. \u003cem\u003eCell\u003c/em\u003e \u003cstrong\u003e186\u003c/strong\u003e, doi:10.1016/j.cell.2023.07.002 (2023).\u003c/li\u003e\n\u003cli\u003eNievergelt, C. M.\u003cem\u003e et al.\u003c/em\u003e Genome-wide association analyses identify 95 risk loci and provide insights into the neurobiology of post-traumatic stress disorder. \u003cem\u003eNature Genetics\u003c/em\u003e \u003cstrong\u003e56\u003c/strong\u003e, doi:10.1038/s41588-024-01707-9 (2024).\u003c/li\u003e\n\u003cli\u003eAlfaro-Almagro, F.\u003cem\u003e et al.\u003c/em\u003e Image processing and Quality Control for the first 10,000 brain imaging datasets from UK Biobank. \u003cem\u003eNeuroimage\u003c/em\u003e \u003cstrong\u003e166\u003c/strong\u003e, 400-424, doi:10.1016/j.neuroimage.2017.10.034 (2018).\u003c/li\u003e\n\u003cli\u003eRajapakse, J. C., Giedd, J. N. \u0026amp; Rapoport, J. L. Statistical approach to segmentation of single-channel cerebral MR images. \u003cem\u003eIEEE Trans Med Imaging\u003c/em\u003e \u003cstrong\u003e16\u003c/strong\u003e, 176-186, doi:10.1109/42.563663 (1997).\u003c/li\u003e\n\u003cli\u003eAshburner, J. A fast diffeomorphic image registration algorithm. \u003cem\u003eNeuroimage\u003c/em\u003e \u003cstrong\u003e38\u003c/strong\u003e, 95-113, doi:10.1016/j.neuroimage.2007.07.007 (2007).\u003c/li\u003e\n\u003cli\u003eFortin, J. P.\u003cem\u003e et al.\u003c/em\u003e Harmonization of cortical thickness measurements across scanners and sites. \u003cem\u003eNeuroimage\u003c/em\u003e \u003cstrong\u003e167\u003c/strong\u003e, 104-120, doi:10.1016/j.neuroimage.2017.11.024 (2018).\u003c/li\u003e\n\u003cli\u003ePurcell, S.\u003cem\u003e et al.\u003c/em\u003e PLINK: A tool set for whole-genome association and population-based linkage analyses. \u003cem\u003eAmerican Journal of Human Genetics\u003c/em\u003e \u003cstrong\u003e81\u003c/strong\u003e, 559-575, doi:10.1086/519795 (2007).\u003c/li\u003e\n\u003cli\u003eShen, X.\u003cem\u003e et al.\u003c/em\u003e A phenome-wide association and Mendelian Randomisation study of polygenic risk for depression in UK Biobank. \u003cem\u003eNat Commun\u003c/em\u003e \u003cstrong\u003e11\u003c/strong\u003e, 2301, doi:10.1038/s41467-020-16022-0 (2020).\u003c/li\u003e\n\u003cli\u003eGe, T., Chen, C. Y., Ni, Y., Feng, Y. C. A. \u0026amp; Smoller, J. W. Polygenic prediction via Bayesian regression and continuous shrinkage priors. \u003cem\u003eNature Communications\u003c/em\u003e \u003cstrong\u003e10\u003c/strong\u003e, doi:10.1038/s41467-019-09718-5 (2019).\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"nature-portfolio","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"","title":"Nature Portfolio","twitterHandle":"","acdcEnabled":false,"dfaEnabled":false,"editorialSystem":"ejp","reportingPortfolio":"","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-6694135/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6694135/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThe genetic architecture of cerebellar substructure volumes is crucial for understanding their functions in cognition, emotion, and neuropsychiatric disorders. We utilized deep-learning segmentation and multi-ancestry analysis to identify genetic associations with 31 cerebellar volumetric traits in 57,071 participants. We found deep learning superior to atlas-based segmentation in heritability estimation, genetic discovery, and polygenic prediction. We identified 407 new loci for these traits (241 from univariate, four from sex-stratified, and 162 from multivariate analyses), along with 19 ancestry-specific and eight sex-specific associations. We prioritized 453 causal variants and 71 relevant genes, categorizing these cerebellar substructures into nine genetically informed clusters. We linked cerebellar volumes to 20 cognitive and mental phenotypes and seven brain disorders. These findings provide an overview of the genetic architecture of cerebellar volumes.\u003c/p\u003e","manuscriptTitle":"Deep-learning segmentation and multi-ancestry GWAS enhance genetic discovery of the cerebellum","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-06-09 08:01:26","doi":"10.21203/rs.3.rs-6694135/v1","editorialEvents":[],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"nature-communications","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"NCOMMS","sideBox":"Learn more about [Nature Communications](http://www.nature.com/ncomms/)","snPcode":"","submissionUrl":"https://mts-ncomms.nature.com/","title":"Nature Communications","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"ejp","reportingPortfolio":"Nature Communications","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"11ff3e29-8076-408f-972a-ba16b8a6459d","owner":[],"postedDate":"June 9th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":49134155,"name":"Biological sciences/Genetics/Genetic association study/Genome-wide association studies"},{"id":49134156,"name":"Health sciences/Anatomy"}],"tags":[],"updatedAt":"2025-06-09T08:01:26+00:00","versionOfRecord":[],"versionCreatedAt":"2025-06-09 08:01:26","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6694135","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6694135","identity":"rs-6694135","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}