Antibody-stratified association analysis

The meta-analysis includes cohorts restricted to AChR-Ab-positive MG cases (n=2798) and ICD-based cohorts potentially incorporating patients from other antibody-mediated MG or seronegative cases (n=2910). Notably, for the 12 reported index SNPs we found no significant differences in effect sizes between these two data sources except for the MHC locus (Supplementary Fig. ⁴).

Genome-wide association study of early-onset MG

We additionally conducted a GWAS on a subset of 1391 EOMG cases utilizing the available phenotypic information. The control group included 22,407 individuals (effective n_half=2056). Our discovery GWAS of EOMG was performed on 7,542,347 SNPs (λ_GC=1.024) (Supplementary Fig. ⁵). We found genome-wide significant associations with four loci and seven additional index SNPs with P<1e⁻⁶. Summary statistics for SNPs with P<1e⁻⁴ are included in Supplementary Data ².

The results were replicated in 871 23andMe cases that reported a diagnosis of EOMG and 109,843 control subjects, all of whom were under 50 years old. Overall, we report two loci that reached genome-wide significance in the meta-analysis of discovery and replication GWAS (Table 2). SOS1, which is the only unique locus not found in the combined MG GWAS, and TNIP1 were not replicated (Table 1B, Fig. 1b, Supplementary Fig. ⁶).

Genome-wide association study of late-onset MG

We further conducted a GWAS of 2404 LOMG cases and 64,103 controls not included in the EOMG GWAS (effective n_half=3701). The LOMG discovery GWAS included 6,310,403 SNPs (λ_GC=1.015) and identified three genome-wide significant loci (Supplementary Fig. ⁷) and four additional index SNPs below a P-value of 1e⁻⁶. Results for all SNPs with P<1e⁻⁴ are included in Supplementary Data ³. A replication study was conducted in the 23andMe dataset of 3989 cases and 226,643 controls because information on the age at diagnosis was unavailable. We found four genome-wide significant loci in the meta-analysis of discovery and replication, all of which overlapped with the combined MG GWAS meta-analysis (Table 3, Fig. 1c, Supplementary Fig. ⁸).

Human leukocyte antigen results

We performed association analyses of 135 imputed HLA class I and II alleles in a subsample of 1927 cases and 5549 controls with available genotypes. Subsequently, we conducted separate analyses for EOMG (1080 cases and 3321 controls) and LOMG (846 cases and 2179 controls). Effects sizes for the strongest associations across the three subgroups are visualized in Fig. 2.

Open in a separate window

Fig. 2

HLA association analysis results.

The forest plot displays the top risk-conferring and protective HLA allele across all main analyses. The log₁₀ of the odds ratio (OR) from inverse-variance-weighted fixed-effects meta-analysis is indicated as diamonds for each HLA allele and dataset used along with the 95% confidence intervals (error bars). Different P-value significance levels are indicated by asterisks for nominally significant (P<0.05*), Bonferroni-corrected (P<3.70e⁻⁴**), and genome-wide (P<5e⁻⁸***). MG myasthenia gravis; EOMG early-onset myasthenia gravis, LOMG late-onset myasthenia gravis.

Our analysis revealed HLA-B*08:01 as the top risk-conferring HLA allele for MG (OR=2.349, P=1.15e⁻⁵², SE=0.056; Supplementary Data ⁴, Supplementary Fig. ⁹). The top protective allele was HLA-C*05:01, (OR=0.599, P=3.21e⁻¹⁰, SE=0.082). After conditioning on the top variant rs4143332 (LD-r² of 0.988 to HLA-B*08:01), or complement component 4 (C4) expression levels HLA-C*05:01 remained nominally significant (rs4143332: OR=0.6869, P=5.662e⁻⁶, SE=0.0828; C4 expression: OR = 0.68991SE=0.0865, P=1.774e⁻⁵), indicating partial independence from HLA-B*08:01 and C4. After conditioning on C4A and C6B expression levels HLA-B*08:01 remained genome-wide significant (OR=2.34527, SE=0.0777, P=5.69e⁻²⁸).

HLA-B*08:01 was also the top risk-conferring allele associated with EOMG (OR=4.677, P=2.18e⁻⁹⁴, SE=0.075; Supplementary Data ⁵, Supplementary Fig. ¹⁰) while DRB1*07:01 was the top protective (OR=0.3996, P=1.11e⁻¹⁶, SE=0.111). After conditioning on the top variant rs2596565 (LD-r² of 1 to HLA-B*08:01), the protective effect of HLA-DRB1*07:01 remained genome-wide significant (OR=0.529, P=2.63e⁻⁸, SE=0.115), indicating HLA-DRB1 as an independent HLA-haplotype. When additionally conditioning on C4 expression both effects remained genome-wide significant (HLA-B*08:01: OR=5.00531 SE=0.0.1082P=4.453e⁻⁵⁰; DRB1*07:01: OR=0.34311, SE=0.1334, P=1.048e⁻¹⁵)

The top HLA allele associated with LOMG was HLA-DRB1*03:01 which implied a protective effect (OR=0.479, P=2.37e⁻⁹, SE=0.123; Supplementary Data ⁶, Supplementary Fig. ¹¹). After conditioning on C4 expression the effect remained genome-wide significant (OR=0.36400, SE=0.1583P=1.724e⁻¹⁰). We observed a significant opposite direction effect for HLA-DRB1*03:01 in MG (OR=1.916, P=2.72e⁻³⁰, SE=0.057) and much more pronounced in EOMG (OR=3.689, P=6.12e⁻⁷¹, SE=0.073). The top risk-conferring allele DRB1*07:01 in LOMG did not reach genome-wide significance (OR=1.414, P=8.86e⁻⁵, SE=0.088).

Complement component 4 results

We performed association tests and meta-analyses of imputed C4 haplotypes in 1927 MG cases and 5549 controls, 1080 EOMG cases and 3321 controls, and 846 LOMG cases and 2179 controls. Two isoforms of C4 are encoded by the genes C4A and C4B located at the MHC class III region. These vary in size and copy number combinations, resulting in long forms (L; ~21 kilobases) and short forms (S; ~14 kilobases). Specifically, we examined four common structural haplogroups of C4A (A) and C4B (B) (BS, AL-BS, AL-BL, and AL-AL). We calculated multiple logistic regression models with BS as the reference haplogroup, which has the fewest gene copy numbers. The resulting odds ratios ranged from 0.46–0.54 in MG, 0.21–0.27 in EOMG, and 1.83–2.27 in LOMG (Supplementary Data ⁷).

Gene prioritization results

To link loci implicated by the GWAS to protein-coding genes, we applied positional mapping, expression quantitative trait loci (eQTL), and chromatin interaction gene mapping. This resulted in 52 mapped genes across 11 loci, excluding the extended MHC region. Four of these loci (rs6433501, rs231779, rs6861227, rs215918) only contained a single protein-coding gene (CHRNA1, CTLA4, TNIP1, TBX18). Furthermore, our analyses highlighted the index SNP rs2476601 in PTPN22 on chromosome 1 (PIP=66%) and rs231775 in CTLA4 on chromosome 2 (PIP=14%; max. PIP=19%; LD-r²=0.97 with lead variant rs231779) as non-synonymous variants. On chromosome 11 there were no protein-coding genes within the defined locus boundaries. For the remaining loci, positional, eQTL, and chromatin interaction mapping nominated multiple genes in the loci, none of which can be prioritized with high confidence. Gene-level results are shown in Supplementary Data ⁸, and annotated results for SNPs in the credible set with PIP>0.01 are included in Supplementary Data ⁹.

We performed drug target enrichment analysis on the set of 52 prioritized genes as input to the Genome for REPositioning drugs (GREP) pipeline²⁰. Significant enrichment was observed only for the ATC category “ectoparasiticides, incl. scabicides, insecticides and repellents” (OR=28.73 and Fisher’s exact P=4.3e⁻²). However, this result does not withstand multiple testing corrections, and the target gene ALDH2 is situated within the complex chromosome 12 locus.

Transcriptome-wide association study results

TWAS using the individual tissue-based prediction matrix of gene expression identified 138 significant genes whose transcript expression was significantly associated with MG. Permutation and co-localization tests identified 56 and 45 gene hits below the significance threshold. Combining TWAS, permutation, and co-localization tests, we identified 24 significant unique genes with high confidence. The TWAS highlighted individual genes in the highly complex loci on chromosome 12 and 17, as well as genes in additional loci that did not reach genome-wide significance in the GWAS, including PAPPA and EPS15L1 (Supplementary Data ¹⁰, Supplementary Figs. ¹²–¹⁴).

Polygenic risk scoring results

After excluding the test sample, our training GWAS comprised 5318 cases and 431,304 controls. Conservatively adjusting for 50 tests resulted in a Bonferroni-corrected P-value of <0.001. In the test sample of 390 cases and 724 controls the polygenic risk score (PRS) for MG explained 4.21% of variation in disease status (R²_observed) at a P-value thresholds (PT) of <0.001 (P=5.12e⁻⁹, AUC=0.607, R²_liability=0.96%; Fig. 3, Supplementary Data ¹¹). In a sample of 176 EOMG cases and 613 age-matched controls, the MG-PRS performed best at a PT<0.1. It explained 2.6% of the variation between early-onset cases and controls (P=2.58e⁻⁴, AUC=0.579, R²_liability=0.64%; Supplementary Fig. ^15A, Supplementary Data ¹²). In a sample of 213 LOMG cases and 62 matched controls, the MG-PRS explained 7.6% of the variation between cases and controls at PT<0.001 (P=3.36e⁻⁴, AUC=0.643, R²_liability=1.75%; Supplementary Figs. ^15B, Supplementary Data ¹³). We further assessed the performance of the MG-PRS when stratifying cases by antibody profile. We identified 73 cases that tested negative for AChR-Ab, including MuSK- and LRP4-Ab-positive patients. These cases were merged with 343 randomly selected controls. The MG-PRS explained 2.9% of the phenotypic variation at a PT of <0.5. However, the P-value remained above the Bonferroni-corrected threshold (P=7.4e⁻³, AUC=0.556, R²_liability=0.71%; Supplementary Fig. ^16A, Supplementary Data ¹⁴). Finally, we calculated the MG-PRS in 371 AChR-Ab-positive cases and 333 independent controls which explained 4.8% of phenotypic variation at a PT<0.0001 (P=1.18e⁻⁶, AUC=0.620, R²_liability=1.01%; Supplementary Fig. ^16B, Supplementary Data ¹⁵).

Open in a separate window

Fig. 3

Performance of MG polygenic risk scores.

Results of polygenic risk scoring of the combined target sample of CCM and LUMC cases and controls across all 10 PTs. Panel a shows the proportion of variance explained through the logistic regression models for PTs 1-10. Panel b shows the density distribution of the z-transformed best-performing PT (P<0.001) for cases (orange) and controls (blue). Panel c shows the odds ratios (OR) from logistic regression models for PT P<0.001 across ten deciles of the PRS along with the corresponding 95% confidence intervals (error bars). The target sample was scored using the combined MG leave-one-out training dataset of 5318 cases and 431,304 controls.PRS polygenic risk score.

Genetic correlation and variant lookup results

We assessed the genome-wide genetic correlation (r_g) of MG, EOMG, and LOMG with 15 autoimmune and neurological traits via Linkage Disequilibrium Score Regression²¹ (LDSC). We identified five significant genetic correlations with MG. The strongest correlations were observed with type-1 diabetes (r_g=0.523, P=1.42e⁻⁶, SE=0.109) and rheumatoid arthritis (r_g=0.5082, P=1.11e⁻⁶, SE=0.104). For EOMG, the strongest correlation was with systemic scleroderma (r_g=0.686, P=5.18e⁻⁶, SE=0.151), and for LOMG with vitiligo (r_g=0.442, P=9.69e⁻⁵, SE=0.113). The full results are presented in Supplementary Fig. ¹⁷ and Supplementary Data ¹⁶-¹⁸.

Genetic correlation analyses performed with 835 medical endpoints in FinnGen R8 and MG highlighted 17 traits that passed the Bonferroni adjusted P-value of 5.88e⁻⁵ (correcting for overall 850 tests). We found the strongest correlations between MG and autoimmune hyperthyroidism (r_g=0.586, P=6.27e⁻⁷, SE=0.118), and seropositive rheumatoid arthritis, strict definition (r_g=0.50, P=1.09e⁻⁵, SE=0.114). Results for all traits below nominal significance (P<0.05) are presented in Supplementary Data ¹⁹.

We have further assessed the genetic correlation between datasets ascertained through clinical studies without antibody-based inclusion criteria (effective n_half=507), clinical studies limited to AChR-ab cases (effective n_half=5,067) and electronic healthcare records (EHR) (effective n_half=4293). We observe the highest correlation between AChR-ab filtered and clinical datasets (r_g=0.6831, SE=0.2137, P=0.0014). However, correlations between EHR based and clinical samples (r_g=0.4774, SE=0.2430, P=0.0494) and between EHR based and AChR-ab filtered (r_g=0.4021, SE=0.1323, P=0.0024) were nominally significant.

We conducted a lookup of index SNPs included in the discovery GWAS and those in LD (LD-r²>0.1) in GWAS Catalog to assess pleiotropy. Furthermore, GWAS Catalog, UK Biobank, and FinnGen R6 datasets were searched for associations with individual index SNPs. The lookups revealed large pleiotropy with other autoimmune diseases and immune-related molecular phenotypes, including white blood cell count. GWAS Catalog results for all SNPs with P<1e⁻⁴ are included in Supplementary Data ¹. Genome-wide significant associations with other traits for each index SNP are shown in Supplementary Fig. ¹⁸.

Inclusion criteria

Cases met the international classification of diseases (ICD) diagnostic criteria for MG (ICD-10 code G70.0 or ICD-9 code 358.0). Newly obtained control samples were filtered for common autoimmune diseases due to the overall high genetic correlation with MG²⁷ (Supplementary Data ²³). All individuals included in the analyses were of European ancestry and provided written informed consent. All included cohorts received approval by their respective local institutional review boards. More detailed sample descriptions are included in the Supplementary Information.

Genomic quality control and imputation

Quality control and imputation were performed using software implemented in the Rapid Imputation for Consortias Pipeline (RICOPILI)⁵³ for all cohorts that provided genotypes. We applied standard quality filters to retain individuals and SNPs. These filters include SNP missingness <0.05, subject missingness <0.02, autosomal heterozygosity deviation F_het<0.2, SNP missingness after sample exclusion <0.02, minor allele frequency (MAF)>0.01, a difference in SNP missingness between cases and controls <0.02, and Hardy-Weinberg equilibrium of P>10e⁻⁶ in controls and P>10e⁻¹⁰ in cases. Related individuals were filtered based on identity by descent segments (PI-HAT>0.2). One member of the related pairs was retained while cases were preferred to control subjects. Externally generated GWAS summary statistics were aligned to Genome Reference Consortium Human Build 37 and filtered to include only SNPs with a MAF>0.01 if necessary. A principal component analysis of the genotyped SNPs was performed via EIGENSTRAT⁵⁴ and plotted to identify and exclude ancestral outliers by visual inspection. After quality control, all cohorts showed a genomic inflation factor (λ_GC) of 1.034–1.073. Imputation was carried out using the Haplotype Reference Consortium reference panel release 1.1⁵⁵. Haplotype pre-phasing and imputation were performed using EAGLE version 2.4.1⁵⁶ and MINIMAC3⁵⁷.

Genome-wide association meta-analyses

We ran GWAS on imputed dosage files of SNPs with a MAF>0.01 using additive logistic regression models for MG versus controls, LOMG versus controls, and EOMG versus controls. By default, we included the first six principal components (PCs) in the models as covariates. Results were meta-analyzed using METAL⁵⁸ with the effect size estimates weighted by the inverse of the corresponding standard errors (SE). We defined an independent signal as the area around an index SNP with a P-value<5e⁻⁸ and a linkage disequilibrium (LD) of r²<0.1 within a 3-megabase window. We conducted leave-one-out GWAS to identify additional significant candidate loci for replication that were potentially lost due to heterogeneity in our overall sample. All newly identified or previously reported GWAS loci were subsequently confirmed or rejected via replication analyses, including meta-analysis and binomial sign-tests.

Replication sample

We replicated our results in a sample ascertained through 23andMe, Inc. The diagnosis of MG was self-reported by the participants. Control individuals were selected at random, constituting 5% of the 23andMe control group, to enhance computational efficiency while minimizing power reduction. We defined EOMG cases as those under 50 years old at participation. 871 individuals met this criterion and were merged with 109,843 randomly drawn controls younger than 50. As no age of onset information was available, we replicated the LOMG discovery GWAS loci using the full 23andMe MG dataset.

Human leukocyte antigen imputation

For all available genotypes, we imputed HLA alleles via a European 1000 Genomes⁵⁹ phase 3 reference panel of 503 individuals with HLA types inferred from sequencing data⁶⁰. The reference was downloaded from the CookHLA⁶¹ GitHub repository and includes 151 HLA alleles with a frequency > 0.01. Pre-phasing and imputation were carried out via SHAPEIT2 and IMPUTE4⁶². We conducted association analyses with the same three dichotomous outcomes as in the GWAS on imputed dosage files. Conditional analyses were conducted via the stepwise inclusion of variants with the lowest P-value as covariates in logistic regression models until no signal below P<1e⁻⁶ was left. To further account for the complex LD structure of the MHC we conducted additional conditional analysis by calculating the C4A and C4B expression levels for each individual based on imputed C4 alleles. We have used the formula proposed by Sekar and colleagues⁶³: C4A expression = (0.47 * AL)+(0.47 * AS)+(0.20 * BL); C4B expression = (1.03 * BL)+(0.88 * BS).

Complement component 4 imputation

We used a reference panel based on 1265 sequenced individuals from the Genomic Psychiatry Cohort⁶⁴ to infer C4 haplogroups, which were imputed using Beagle 5.4⁶⁵. We initially filtered the imputed C4 variants based on the gene composition levels (e.g., BS, AL-BS, AL-BL, AL-AL) by allele frequency (>0.01) in controls and imputation quality (INFO>0.80). Subsequently, we removed individuals with dosage sums ≤1.9 to avoid these being attributed to the reference group in the logistic regression model. Dosages for all remaining C4 variants were then modeled together in a logistic regression for each outcome, incorporating the first six principal components and excluding the BS reference group, denoting the shortest C4 variant. Fixed-effects meta-analyses for MG, EOMG, and LOMG were conducted in R version 4.3.2⁶⁶ using meta version 7.0–0⁶⁷.

Gene prioritization

In order to map SNPs within the identified loci to protein-coding genes we applied the SNP2GENE module implemented in FUMA version 1.5.2⁶⁸ via positional (10 kilobase window), eQTL (GTEx version 8 tissues, database of immune cell expression eQTLs), and chromatin interaction mapping (HiC of adult and fetal cortex, and GSE87112 tissues). We used the discovery GWAS summary statistics along with the set of 11 pre-defined index SNPs as input files for FUMA. Due to its complex LD structure we excluded the extended MHC region (Chromosome 6: 25–35Mb) from all analyses. We additionally filtered the mapped genes with the LD-based locus boundaries defined by RICOPILI (LD-r²>0.1). Furthermore, we intersected FUMA functional annotations with SNPs in the 95% credible set with a PIP>0.01 (calculated via the R package coloc 5.2.3 finemap.abf module⁶⁹).

To evaluate the enrichment of clinical indication categories (ICD-10 and ATC) of druggable target genes in our GWAS, we utilized the set of genes prioritized by FUMA as input to the GREP pipeline²⁰. The pipeline employs Fisher’s exact tests to determine whether the gene set demonstrates enrichment in genes targeted by medications.

Transcriptome-wide association study

Transcriptome-wide association study (TWAS) was performed using FUSION⁷⁰ to predict tissue-specific gene expression based on the MG discovery GWAS summary statistics excluding the extended MHC region (Chromosome 6: 25–35Mb). Gene expression weights from three Genotype-Tissue Expression (GTEx) version 8⁷¹ tissues were used for transcriptomic imputation and association testing: Muscle Skeletal (n=8602), Nerve Tibial (n=11,360), and Whole Blood (n=8059). Gene associations were considered significant at a P-value<0.05/N of genes per tissue. A permutation test, which shuffles the quantitative trait loci (QTL) weights (N_max=1000) was performed to correct inflated associations from by-chance QTL co-localization. Co-localization was computed for genes with a P-value<1e-5. A posterior probability value≥0.75 was considered as evidence for the expression-QTL-GWAS pair influencing both the expression and the GWAS trait in a particular region.

Polygenic risk scoring

PRS analyses were performed to assess the capacity of the discovery results to predict MG, EOMG, and LOMG in an independent sample. Additionally, we evaluated PRS performance in AChR-Ab positive- and negative cases. We performed a leave-one-out GWAS meta-analysis to generate independent training and test datasets. PRS for MG were calculated via the LD-clumping and P-value thresholding method in PLINK 1.9⁷². The training dataset was clumped to account for LD, retaining only the most significant SNP within 500 kilobases and an LD of r²>0.25. PRS was generated across ten P-value thresholds (PT): 5e⁻⁸, 1e⁻⁶, 0.0001, 0.001, 0.01, 0.05, 0.1, 0.2, 0.5, 1. The first six PCs were included in all logistic regression models as covariates. Control subjects were matched by age for onset-specific analyses and split randomly in antibody-specific analyses. The observed phenotypic variation was measured via Nagelkerke’s pseudo-R² and the predictive performance via the area under the receiver operator curve (AUC).

Genetic correlation analyses and variant lookup

The bivariate genetic correlations of MG, EOMG, and LOMG with 15 pre-selected neurological and autoimmune traits were computed via LDSC⁷³. We downloaded publicly available summary statistics^{26,31,74–85} from the GWAS catalog website [https://www.ebi.ac.uk/gwas/]⁸⁶ and formatted summary statistics from Dr. Alkes Price’s research group’s repository [https://alkesgroup.broadinstitute.org/sumstats_formatted]²².

We have additionally performed genetic correlation analyses between MG and 835 medical endpoints, utilizing data from digital health record data of the 342,499 participants included in FinnGen Release 8⁵¹. We limited the medical endpoints to those with at least one genome-wide significant hit in FinnGen, indicating genetic susceptibility to the trait and ensuring an adequate sample size. We have excluded the FinnGen sample from our MG summary statistics for this purpose.

In order to assess pleiotropy, we conducted a lookup of all SNPs in the discovery GWAS and those in LD (LD-r²>0.1) in GWAS Catalog (version from September 2018)⁸⁶ for associations with other traits. Additionally, we performed a lookup for the 11 individual index SNPs highlighted by the combined meta-analysis, excluding the MHC in data sources aggregated by Open Targets Genetics. The data sources include associations identified by the SAIGE study and the Neale lab conducted in the UK Biobank, summary statistics from GWAS Catalog, and FinnGen Release 6⁸⁷.

Heritability estimates

We used LDSC²² and the genome-based restricted maximum likelihood (single-component GREML) module implemented in GCTA²³ to assess h²_SNP of MG in our sample. We used summary statistics including all samples in the discovery GWAS and samples with available genotypes to estimate h²_SNP via LDSC. We further merged the PLINK files of all available genotypes as input for GCTA along with the first six PCs. We assumed a population prevalence of 20 per 100,000 individuals to transform the heritability estimates into the liability scale.

We want to acknowledge the participants and investigators of the FinnGen study, the UK Biobank, the Million Veteran Program, BioVU, the Estonian Biobank, deCODE genetics, and the International Multiple Sclerosis Genetics Consortium for sharing data with us. We would like to thank the research participants and employees of 23andMe for making this work possible. Following biobanks are acknowledged for delivering biobank samples to FinnGen: Auria Biobank (www.auria.fi/biopankki), THL Biobank (www.thl.fi/biobank), Helsinki Biobank (www.helsinginbiopankki.fi), Biobank Borealis of Northern Finland (https://www.ppshp.fi/Tutkimus-ja-opetus/Biopankki/Pages/Biobank-Borealis-briefly-in-English.aspx), Finnish Clinical Biobank Tampere (www.tays.fi/en-US/Research_and_development/Finnish_Clinical_Biobank_Tampere), Biobank of Eastern Finland (www.ita-suomenbiopankki.fi/en), Central Finland Biobank (www.ksshp.fi/fi-FI/Potilaalle/Biopankki), Finnish Red Cross Blood Service Biobank (www.veripalvelu.fi/verenluovutus/biopankkitoiminta), Terveystalo Biobank (www.terveystalo.com/fi/Yritystietoa/Terveystalo-Biopankki/Biopankki/) and Arctic Biobank (https://www.oulu.fi/en/university/faculties-and-units/faculty-medicine/northern-finland-birth-cohorts-and-arctic-biobank). All Finnish Biobanks are members of BBMRI.fi infrastructure (www.bbmri.fi). Finnish Biobank Cooperative -FINBB (https://finbb.fi/) is the coordinator of BBMRI-ERIC operations in Finland. The Finnish biobank data can be accessed through the Fingenious® services (https://site.fingenious.fi/en/) managed by FINBB. The FinnGen project is funded by two grants from Business Finland (HUS 4685/31/2016 and UH 4386/31/2016) and the following industry partners: AbbVie Inc., AstraZeneca UK Ltd, Biogen MA Inc., Bristol Myers Squibb (and Celgene Corporation & Celgene International II Sàrl), Genentech Inc., Merck Sharp & Dohme LCC, Pfizer Inc., GlaxoSmithKline Intellectual Property Development Ltd., Sanofi US Services Inc., Maze Therapeutics Inc., Janssen Biotech Inc, Novartis AG, and Boehringer Ingelheim International GmbH. This research is based, in part, on data from the Million Veteran Program, Office of Research and Development, Veterans Health Administration. Funding for D.F.L. was provided by a Career Development Award CDA-2 from the Veterans Affairs Office of Research and Development (1IK2BX005058-01A2). Funding for Murray.S. and J.G. was provided from a Veterans Affairs Office of Research and Development Merit Award (I01CX001849). One dataset used for the analyses described were obtained from Vanderbilt University Medical Center’s BioVU which is supported by numerous sources: institutional funding, private agencies, and federal grants. These include the NIH funded Shared Instrumentation Grant S10RR025141; and CTSA grants UL1TR002243, UL1TR000445, and UL1RR024975. Genomic data are also supported by investigator-led projects that include U01HG004798, R01NS032830, RC2GM092618, P50GM115305, U01HG006378, U19HL065962, R01HD074711; and additional funding sources listed at https://victr.vumc.org/biovu-funding/. P.M. is Einstein Junior Fellow funded by the Einstein Foundation Berlin and acknowledges funding support by the Einstein Foundation Berlin (EJF‐2020–602; EVF‐2021–619, EVF-BUA-2022-694) and the Leducq Foundation for Cardiovascular and Neurovascular Research (Consortium International pour la Recherche Circadienne sur l’AVC). M.G.H. receives financial support from the LUMC (Gisela Thier Fellowship 2021), Top Sector Life Sciences & Health to Samenwerkende Gezondheidsfondsen (LSHM19130), Prinses Beatrix Spierfonds (W.OR-19.13). The LUMC is part of the European Reference Network for Rare Neuromuscular Diseases [ERN EURO-NMD] and the Netherlands Neuromuscular Center. S.R. has received funding from the German Research Foundation (Deutsche Forschungsgemeinschaft - DFG) (grant number 461427996). The Estonian Biobank work was supported by Personal research funding: Team grant PRG1291. S.R. and A.M. received funding from the German Research Foundation (Deutsche Forschungsgemeinschaft) via the Clinical Research Group KFO 5023 BeCAUSE-Y.