Genetic variation at RAB3GAP2 and its role in exercise-related adaptation and recovery

Skeletal muscle ber composition and capillary density inuence physical performance and whole-body metabolic properties. ~45% of the variance in ber type is heritable, which motivated us to perform a genome-wide association study of skeletal muscle histology from 656 Swedish men. Four independent variants were associated (p < 5x10 − 8 ) with proportion of type IIx bers or capillary-to-ber ratio (C:F). The strongest signal localized to the rs115660502 variant, where the G-allele corresponded with increased C:F and reduced skeletal muscle expression of the proximal gene, RAB3 GTPase Activating NonCatalytic Protein Subunit 2 (RAB3GAP2). The G-allele was less frequent in elite short-track sprinters and more frequent in endurance athletes than in matched non-athlete (population) controls; RAB3GAP2 expression was reduced by high-intensity intermittent training. RAB3GAP2 protein was not uniformly expressed in muscle tissue but localized to the endothelium and capillaries. Experimental reduction of RAB3GAP2 in human endothelial cells led to increased tube formation in vitro, to regulation of secreted factors promoting angiogenesis and T-cell activation, to reduced intracellular levels of von Willebrand factor (VWF) and, post-implantation, to increased endothelial cell density in vivo in mice. The amount of RAB3GAP2 in skeletal muscle was positively associated with exercise-induced release of VWF in vivo in humans. By regulating the release of protein factors (VWF, CD70, TNC, TNXB, MCP1, IGFBP3, COL1A1, TFPI2 and tPA), RAB3GAP2 inuences tness adaptation after exercise by improving muscle healing and promotion of capillary formation. CD70 mediating capillary 16,17 . The RAB3 protein family controls exocytosis of neurotransmitters and 16 , and mutations in RAB3GAP2 cause Martsolf syndrome and Warburg Micro Syndrome 2, diseases characterized by delayed neurodevelopment and cognitive impairments 25 . To our knowledge, no previous studies have linked RAB3GAP2 to exercise-response phenotypes. genome-wide signicant SNPs were directly typed in the MM and MEI cohorts. The imputation error rates vary from variant to variant (i.e., from 0-4% for common homozygotes, 0-86% for heterozygotes), and these error rates were inversely related to the imputation info score. To assure that ndings were not biased by factors related to genotype imputation, genetic association analyses were re-run using directly genotyped data, which did not indicate any major inuence of imputation on the results. Results from the directly genotyped, rather than imputed, data are generally presented in the manuscript. Analyses conditioning on variants associated with the index traits at a level of genome-wide signicance (P<5x10 -8 ) were undertaken in in corresponding regions. In case more than one variant located on the same haplotype was found at genome-wide signicance level, and the associated variants were not in a strong LD with each other, we tested for secondary / independent association. For this purpose, an analysis was performed, where association analysis of all genetic variants from the haplotype region was re-run adjusting for the genome-wide signicant variants sequentially one-by-one. If the association with P<5x10 -8 remained after adjusting for another neighboring genome-wide signicant variant, we concluded that the two genome-wide signicant variants represented independent genetic signals. Otherwise, their association was capturing the same genetic signal. The ‘Metafor’ command in R software 43 as used to make Forest plots, and the LocusZoom browser was used for creating regional association plots 44 . Bio-Rad. tissue lysate blots, a pre-blotted membrane with different human tissue lysates was purchased (INSTA-Blot NBP2-31378 Novusbio). For analyses of CRISPR/Cas9 knockdown eciency on RAB3GAP2 the cells were lysed in 2% SDS in PBS with protease inhibitor cocktail (Roche), passed through QIAshredder minispin columns (Qiagen) to reduce the viscosity, and the protein content was quantied using BCA Protein Assay Kit (ThermoFisher). 20 mg of protein was run on a 4-20% Mini-Protean TGX SDS-PAGE gel (Bio-Rad) and electroblotted onto nitrocellulose membrane (Bio-Rad). From this point, both human tissue lysate blots and the CRISPR knockdown analysis, the same blotting protocol was used. The membrane was rinsed in TBS, blocked in 5% bovine serum albumin in TBS, for 1 h at room temperature, then transferred to antibody incubation buffer (1% bovine serum albumin in TBS with 0.05% Tween-20) with anti-RAB3GAP2 diluted 1:2000, and incubated overnight at 4 °C. Next, the membrane was washed 4 x 5 min in wash buffer (TBS with 0.05% Tween-20) and incubated with HRP-conjugated anti-rabbit diluted 1:10000 in antibody incubation buffer, for 1 h, at room temperature. The membrane was washed 4 x 5 min in wash buffer and developed using SuperSignal West Pico PLUS Chemiluminescent Substrate (ThermoFIsher) and a CCD camera (Bio-Rad). For the second blotting step the membrane was stripped in Restore Western Blot Stripping Buffer (ThermoFisher), then the above-described blotting protocol was used, but using mouse anti-tubulin (diluted 1:5000) or rabbit anti-GAPDH (diluted 1:10000) antibodies, and species-specic secondary HRP-conjugated antibodies.


Introduction
One of the most pervasive threats to contemporary human health is the sedentary nature of our lifestyles. The bene cial impact of exercise on health is determined in part by the damage caused to skeletal muscle and the rate at which damaged muscle tissue repairs. Indeed, this cyclical process of damage and repair dictates tness adaptation. These adaptations include skeletal muscle hypertrophy, improvements in substrate utilization, ber type switching (from Type IIx to Type IIa) and changes leading to increased blood ow and perfusion such as increased capillary density.
Sustained aerobic exercise has many established health bene ts including reductions in risk of cardiovascular disease. However, people vary in their response to exercise 1,2 , especially those who are untrained, acute bouts of exhaustive exercise occasionally results in life-threatening events such as myocardial infarction, hemorrhagic stroke and provoked venous thromboembolism [3][4][5][6] .
Genetics has proven powerful for the discovery of pathogenic pathways and processes across many diseases. Such knowledge has been exploited in drug development, where information about genetic perturbation can help narrow the search for drugable targets 7 . Exercise is increasingly viewed as a therapeutic adjunct or alternative to drugs. As with drug development, it is plausible that identifying genetic perturbations in uencing pathways through which exercise impacts human health might lead to discoveries that broaden therapeutic options, either by facilitating the development of exercise mimetics or by optimizing individualized exercise interventions.
The purpose of this study was to identify genetic variants associated with exercise-related muscle tissue phenotypes and elucidate the underlying molecular mechanisms that link exercise with health. n = 11) and ~-4%, (p Meta-FDR = 0.007, n = 69 vs 8) respectively (Figure 2a-b, Supplement table S4), which in turn was inversely correlated with C:F ratio (rho= -0.38, p=0.03, n=32) ( Figure 2c). The reduction of RAB3GAP2 expression following HIIT was replicated in a meta-analysis of 8 datasets with acute HIIT (~-7%, P meta = 2.9 x 10 -3 , n = 54) and in 3 datasets with prolonged HIIT (~-6%, P meta = 1.5 x 10 -3 , n = 34) in MetaMEx 14 . The RAB3GAP2 protein is widely expressed in human tissues, including skeletal muscle (Supplement Figure S6). Using immunohistochemistry, we localized the RAB3GAP2 protein to skeletal muscle capillaries and endothelium, but it was absent in muscle bers and larger vessels (Figure 2d-i). This distinction in vascular localization was supported by the presence of RAB3GAP2 in human microvascular endothelial cells (HMEC) and absence in human umbilical vein endothelial cells (HUVEC) of macrovascular origin (Supplement Figure S7). rs115660502 is located close to the transcriptional start site of RAB3GAP2 (~20 kb upstream) and, in two ENCODE cell lines (Huh7 and Medullo) 15 , within a DNase hypersensitive region. Several transcription factors implicated in angiogenesis (e.g. STAT3, FOSL1, SIRT6, TAL1, and GATA2) bind in close proximity (500-1000 bp) to rs115660502 in the ENCODE database 15 . In summary, the G-allele and HIIT are both associated with lower RAB3GAP2 expression, and the protein is localized to the endothelium and capillaries.

Reduced expression of RAB3GAP2 in human endothelial cells stimulates tube formation
To test if RAB3GAP2 expression is causally related to capillary formation, RAB3GAP2 was knocked-down using CRISPR-Cas9 (mRNA by ~70 -80% in HMECs), with corresponding protein reduction (Supplement gure S8). The level of knockdown was stable and sustained over 12 passages (Supplement gure S8).
We then undertook an in vitro tube formation assay using HMECs, where RAB3GAP2 knockdown led to a 2.8-fold increase in number of loops (i.e., ring structures) at 6 hours of differentiation and a 2.9-fold increase at 24 hours (p Mann-Whitney = 8.0 x 10 -7 and p Mann-Whitney = 7.1 x 10 -5 respectively, n = 11 and 13). The number of branching points increased by ~30% in RAB3GAP2 knocked-down cells compared with wildtype at 6 hours of differentiation (p Mann-Whitney = 8.8 x 10 -6 , n = 11) and increased by ~32% at 24 hours (p Mann-Whitney = 7.6 x 10 -4 , n = 13). The total tube length formed was ~22% longer in RAB3GAP2 knocked-down cells compared to wildtype at 6 hours of differentiation (p Mann-Whitney = 8.0 x 10 -7 , n = 11) and ~27% longer at 24 hours (p Mann-Whitney = 2.4 x 10 -5 , n = 13). Quanti cation and example images of tube formation in HMECs from the different conditions are shown in gure 3a-d.
In an in vivo angiogenesis plug assay, Matrigel containing HMECs with or without RAB3GAP2 knockdown were injected in the groin region of NGS mice. The resulting gel plugs were harvested after 7-days, sectioned, and stained using Masson's Trichrome. The total HMEC cell area was ~50% larger in RAB3GAP2 knocked-down cells compared to wildtype (p Mann-Whitney = 0.036, n = 3-5). Quanti cation and example images of cell density of HMECs are shown in Figure   3e-f. In summary, experiments in human endothelial cells replicate the GWAS association with C:F, whereby knockdown of RAB3GAP2 leads to a gain-offunction with increased tube formation in vitro and increased cell density in vivo in mice.
RAB3GAP2 regulates the secretion of proteins promoting angiogenesis and T-cell activation RAB3GAP2 is known to regulate vesicle tra cking 16 . To identify secreted proteins regulated by RAB3GAP2 and driving the effects described above, we performed a RNA sequencing screen in HMECs with or without CRISPR-Cas9-mediated knockdown of RAB3GAP2 expression. 496 genes with lower and 993 genes with higher expression in RAB3GAP2 knocked down cells were identi ed at FDR 5% (Supplement table S5). In addition, 154 secreted proteins were screened in the medium of HMECs using 4 Olink panels with or without knockdown of RAB3GAP2 (Supplement table S6). Combining these results, RAB3GAP2 knockdown in uenced both the mRNA expression and protein secretion of eight factors (Figure 4a-h). Four of these factors are known to stimulate angiogenesis, i.e., Tenascin C (TNC), Insulin like growth factor binding protein 3 (IGFBP3), Plasminogen activator tissue type or tPA (PLAT), C-C motif chemokine ligand 2 / MCP-1 (CCL2), all had higher concentration in RAB3GAP2 knocked down cells. Three of the eight factors are known to inhibit angiogenesis, i.e., Tenascin XB (TNXB), Collagen, type I, alpha 1 (COL1A1) and Tissue factor pathway inhibitor 2 (TFPI2), all had lower concentration in RAB3GAP2 knocked down cells. However, the most strongly regulated factor was the T-cell activating protein cluster of differentiation 70 (CD70), with a ~3fold higher RNA expression in RAB3GAP2 knocked down cells (p adj = 1.4 x 10 -40 , n = 8) and a ~2.5-fold higher protein concentration in the medium from RAB3GAP2 knocked down cells compared to wildtype (p Wilcoxon = 0.03, n = 6). In summary, lower RAB3GAP2 levels lead to pro-angiogenic levels of proteins like TNC, IGFBP3, tPA and MCP-1 and possibly to T-cell activation by increased secretion of CD70.
RAB3GAP2 regulates von Willebrand factor secretion RAB3GAP2 heterodimerizes with the catalytic subunit RAB3GAP1 to inhibit RAB3 activity 17 . In endothelial cells, RAB3 (A and D isoforms) stimulate the release of WPB (Weibel-Palade bodies) 18,19 , vesicles that are highly enriched for von Willebrand factor (VWF) but also contain angiogenic factors 20 . Characterized as uorescence intensity per cell, CRISPR-Cas9-mediated knockdown of RAB3GAP2 expression in HMECs suggested reduced intracellular VWF levels (n = 3, 20-24 cells per experiment, Figure 4i-j). This decrease was quanti ed in independent experiments using an ELISA assay to measure the intracellular VWF concentration. Intracellular VWF concentration was ~11% lower in RAB3GAP2 knocked-down cells compared with wildtype (p Wilcoxon = 7.8 x 10 -3 , n = 8, Figure 4i), indicating increased VWF secretion.
Exercise acutely stimulates VWF secretion 21 , with VWF secretion being inversely related to physical tness 22 Figure 4k). Moreover, exercise-induced VWF secretion was inversely associated with VO 2 max (expressed in ml·kg -1 ·min -1 ) (-216 ± 77 ng/ml [b ± SE], p = 9.0 x 10 -3 , n = 38, adjusted for VWF before the Wingate test and peak power per kg body weight). Finally, we demonstrated that basal RAB3GAP2 mRNA level was positively associated with change in circulating VWF concentration following the Wingate test (2128 ± 728 ng/ml [b ± SE], p = 6.0 x 10 -3 , n = 38, adjusted for VWF before the Wingate test and peak power per kg body weight). Blood groups are known to affect circulating VWF concentration 23 , but adjusting for blood group did not materially in uence the association between RAB3GAP2 muscle expression and VWF levels after exercise (data not shown).
Given that the G allele at rs115660502 is associated with lower RAB3GAP2 expression and experimental lowering of RAB3GAP2 expression reduces intracellular VWF levels, we hypothesized that this allele is associated with higher circulating VWF levels in the non-exercised state, which we con rmed in a GWAS meta-analysis dataset (n = 43,775) 24

Discussion
We undertook a sequence of integrated studies to discover, validate and mechanistically elucidate novel molecular features underlying skeletal musclemediated effects of exercise in health. We began by undertaking a GWAS of exercise-related skeletal muscle phenotypes and leveraged these ndings to discover novel mechanisms linking exercise to health traits. Speci cally, we show that a RAB3GAP2 variant (rs115660502) is associated with C:F ratio and that the frequency of the effect allele (G) varies between elite athletes and population controls. We uncovered that both the G allele and HIIT exercise training decrease RAB3GAP2 muscle expression, which, through either mechanism, regulates the secretion of VWF, T-cell activating and angiogenesis promoting proteins, resulting in enhanced muscle repair and capillary formation. Experimental reduction of RAB3GAP2 expression leads to increased angiogenesis both in vitro and in vivo. Finally, we show that the level of RAB3GAP2 in muscle modulates the exercise-induced secretion of VWF in vivo. Taken together, these results show that RAB3GAP2 plays a key role in muscle adaptation to exercise, both acutely by regulating VWF and CD70 release and chronically by mediating capillary formation.
RAB3GAP2 encodes a regulatory subunit that together with RAB2GAP1 (encoding the catalytic subunit) form the GTPase-activating complex with speci city for the RAB3 Rab-GTPase-activating proteins (RAB3A, RAB3B, RAB3C and RAB3D) 16 Exercise acutely stimulates VWF secretion 21 with the quantity of VWF secreted being inversely related to physical tness 22 , consistent with the broader known effects of exercise in cardiovascular risk. Our data help explain in part why an acute bout of strenuous exercise might transiently raise the propencity for thrombotic events (at least in high risk individuals), whereas regular exercise (i.e., repeated bouts of acute exercise that enhance aerobic tness) lowers overall cardiovascular risk in the long-term in virtually all population groups.
Our ndings suggest that in those carrying the rs115660502 G allele, skeletal muscle recovery may be accelerated; yet these same individuals may experience higher basal levels of VWF, an established risk factor for venous thrombosis 26,27 . A reduction of RAB3GAP2 activity, either genetically (via presence of the G allele) or by exercise, results in reduced intracellular VWF concentrations, presumably owing to increased basal secretion of WPBs. These conclusions are supported by the in vitro and in vivo experiments in HMECs, the higher circulating levels of resting VWF in carriers of the G allele and by the in vivo response to Wingate testing.
The secretion of VWF, CD70 and proteins promoting angiogenesis in response to exercise likely re ects the elevated risk of injury when intensive exercise ensues. This mechanism might have evolved to reduce risk of bleeding, minimize muscle microruptures and help repair muscle tissue after exercise, as well as to promote tness adaptation through de novo capillary formation. This implies that the regulation of these factors is a rapid process that clearly outpaces de novo production of VWF and formation of new WPBs. Subsequently, the endothelium secretes all VWF present in its cells. However, the lower VWF concentration in G allele carriers and/or well-trained individuals suppresses this acute response (as observed with the Wingate test, both with respect to RAB3GAP2 in muscle and VO 2max ), which may reduce risk of cardiovascular events associated with acute strenuous exercise. These observations may partially explain why exercise training reduces cardiovascular risk, whereas acute and intense exertion in untrained (or unhealthy/older) individuals can sometimes cause harm.
In summary, naturally occurring genetic perturbation of RAB3GAP2 promotes capillary formation and the release of VWF in a manner consistent with exercise. The higher frequency of the rs115660502 variant allele in highly trained athletes may be ergogenic, by enhancing exercise-induced capillary formation and muscle repair. However, this variant may also predispose risk of thrombotic events in sedentary individuals.  ml/kg/min, respectively) were recruited from two specialist Swedish ski schools. In addition, one senior male cross-country skier was recruited (age and VO 2max were 38 years and 69.5 ml/kg/min). All participants competed at a national level, six were members of national junior development teams and one was a senior World championship medalist. All subjects provided written informed consent and the study protocol was approved by the Regional Ethical

Materials And Methods
Review Board, Umeå University. 2) Spanish athletes from different disciplines (n = 141), including triathletes (n = 16) 35 . All athletes were Spanish (Caucasian) men with solid expertise in international competitions. Blood/saliva samples were collected over 10 years in different places of Spain from athletes who were runners (mostly 5,000m and above), professional road cyclists, rowers, canoeists, and triathletes. Except for professional cyclists,  Athletes could be subdivided further into categories de ned by their level of performance: national-level athletes (n = 28), who were competitive at national-level competition in Jamaica and the Caribbean, and international-level athletes (n = 86), who had competed at major international competition for Jamaica. Forty-six of these international athletes had won medals at major international competitions or held sprint world records. All subjects provided written informed consent to participate in the study, which was approved by the Ethics Committee of University of West Indies. Genotyping of the variants achieving genome-wide signi cance was done and the allele frequencies were compared to the corresponding non-athlete controls of Swedish, Spanish, Polish and Jamaican population allele frequencies.

Phenotypes
In the ULSAM, MM, MEI and MSAT cohorts, muscle biopsies were taken with a 6 mm Bergström needle and frozen in liquid nitrogen, and serial sections (10 or 16 μm) were cut using a cryostat at -20 °C. Capillaries were stained using the double staining method 36 Tables S1-S3. Within each cohort, all phenotypes were pre-adjusted for age and BMI, and inverse rank-normal transformed using the Rankit equation (Bliss 1967) to provide normal data distributions.

Genotyping
Genotyping of the ULSAM was done using the Illumina Omni 2.5M array and Illumina CardioMetabochip (Illumina, California, USA). Genotyping of the MM and MEI cohorts were done using the HumanOmniExpress 12v1 C chips (Illumina, California, USA); genotypes were called using the Illumina Genome studio software, as described in detail elsewhere 38 . Individuals were excluded on the basis of: call rate <95%; gender mismatch; relatedness; and non-European ancestry. Variants were excluded on the basis of: call rate <95% (<99% for MAF <5%); exact p-value for Hardy-Weinberg equilibrium <10 -6 ; and MAF <1%. Genotyping of the different cohorts of imputed genome-wide signi cant variants from the meta-analysis were done using either TaqMan PCR or Sequenom platforms according to the manufacturer's instructions. For TaqMan an ABI Prism Sequence Detection System ABI 7900HT (Applied Biosystems) was used for post-PCR allelic discrimination by measuring allele-speci c uorescence. The results were in Hardy-Weinberg equilibrium (p<1x10 -4 ).

Imputation & meta-analysis
Genotype data in each cohort were imputed up to 35 million variants from the 1000 Genomes reference panel (all ancestries, March 2012). Prephasing of haplotypes and imputation were performed using ShapeIT 39 and IMPUTE2 40 , respectively. The association within each cohort was performed using SNPTEST frequentist score additive model test 41 . Genetic variants that had poor imputation quality inferred by an info score ≤0.4 and/or had high standard error (SE>10) were excluded from meta-analyses. Fixed effects meta-analyses were undertaken on the summary statistics obtained from the three cohorts using GWAMA 42 , with phenotypes inverse rank-normal transformed and pre-adjusted for age and BMI. Post-meta-analysis QC included removing variants that: a) were detectable only in one cohort and b) had total minor allele count <10. To validate the quality of imputation, all imputed genome-wide signi cant SNPs were directly typed in the MM and MEI cohorts. The imputation error rates vary from variant to variant (i.e., from 0-4% for common homozygotes, 0-86% for heterozygotes), and these error rates were inversely related to the imputation info score. To assure that ndings were not biased by factors related to genotype imputation, genetic association analyses were re-run using directly genotyped data, which did not indicate any major in uence of imputation on the results. Results from the directly genotyped, rather than imputed, data are generally presented in the manuscript. Analyses conditioning on variants associated with the index traits at a level of genome-wide signi cance (P<5x10 -8 ) were undertaken in order to discover possible secondary association signals in corresponding regions. In case more than one variant located on the same haplotype was found to be associated at the genome-wide signi cance level, and the associated variants were not in a strong LD with each other, we tested for secondary / independent association. For this purpose, an analysis was performed, where association analysis of all genetic variants from the haplotype region was re-run adjusting for the genome-wide signi cant variants sequentially one-by-one. If the association with P<5x10 -8 remained after adjusting for another neighboring genome-wide signi cant variant, we concluded that the two genome-wide signi cant variants represented independent genetic signals. Otherwise, their association was capturing the same genetic signal. The 'Metafor' command in R software 43 as used to make Forest plots, and the LocusZoom browser was used for creating regional association plots 44 .

Testing allele frequencies in athletes versus population controls
The ve genome-wide signi cant index variants were genotyped in the four cohorts of elite athletes from different disciplines and populations, i.e. Swedish cross-country skiers 34 , Spanish athletes of different disciplines 35 , Polish skaters and Jamaican sprinters (up to 400 m and in jump and throw events) 45 and in corresponding population-matched controls of non-athlete status. Since the athletes originated from different world populations (Sweden, Spain, Poland, and Jamaica) we normalized the G-allele frequency at rs115660502 in the athletes by G-allele frequency in the control (non-athletes) group from the corresponding population. For example, the ratio of the G-allele frequency in Swedish skiers to G-allele frequency in Swedish controls was ~2.5, while the corresponding ratio for Jamaican sprinters was ~0.5, suggesting that the G-allele of rs115660502 is over-represented in endurance sports (skiing) and under-represented in power sports (sprint). To test for signi cance, resampling with replacement by randomly drawing equal number of controls as athletes and calculating ratios 100.000 times was done to build con dence intervals. The same resampling procedure was repeated for building the empirical distribution of G-allele frequencies in all populations. Next, statistical signi cance was tested by counting the number of times the re-sampled G-allele frequency of the controls was different or equal to the respective athletes. We de ned athletes competing in cross country skiing and triathlon as extreme endurance athletes (i.e., disciplines with > 1.5 h total uninterrupted physically demanding performance time) and athletes competing in sprint running as extreme power athletes (i.e., disciplines with < 60 s total performance time). Other disciplines were de ned as not being extreme in their physical pro le, e.g., skating, basketball, and judo. No difference in frequencies were found for any of the variants associated with Type IIx ber distribution. However, the capillary-increasing G-allele at rs115660502 has a frequency of 4.8% and 3.3% in the Swedish and Spanish non-athlete population controls respectively but had 10% and 6.3% in the Swedish cross-country skiers and Spanish triathletes respectively. By contrast, the G-allele had a 1.1% frequency in the non-athlete Jamaican population controls, which was signi cantly higher compared to the 0.4% frequency of the Jamaican sprinters (p = 0.03). Athletes of non-extreme or mixed sports had a similar G-allele frequency as their corresponding population controls. eQTL analysis eQTL analyses were done using Affymetrix microarray expression data from the MEI (n=39; 23,941 probe sets) and MM (n=38; 22,283 probe sets) cohorts. Genome-wide signi cant variants from the meta-analysis for each phenotype were tested for linear associations with gene expression levels using the 'Matrix_eQTL' command in R 46 . Cis-eQTL (within a 1 Mb window) was performed separately for MEI and MM with the corresponding False Discovery Rate (FDR) correction for multiple testing (FDR<5%). To be consistent with the primary GWAS analyses, we used age and BMI as covariates for eQTL analysis.
The results of the eQTL analysis were meta-analyzed with Stouffer's z-score method. The R commands "hgu133a.db" and "nugohs1a520180.db" were used for probeset-gene annotation.

Quantitative real-time PCR (QPCR)
Veri cation of a sustained CRISPR/Cas9-mediated knockdown of RAB3GAP2 in HMEC cells, and analysis of the expression of RAB3GAP2 mRNA in human skeletal muscle was done using QPCR. RNA was isolated with the RNeasy Plus Mini Kit (Qiagen) and concentration and purity were measured using a NanoDrop ND-1000 spectrophotometer (A260 / A280 > 1. 8

Measurement of von Willebrand factor (VWF)
HMEC-1 cells were seeded on a culture ask (800k cells / 25cm 2 ) and were allowed to grow for 24h at 37 o C and 5% CO 2 . Cells were washed with PBS and lysed in an ice-cold buffer (MCDB131 growth medium (without FBS), Triton-X100 (0.5% vol/vol) (MP Biomedicals (807426)), 1 mM EDTA (Sigma (E6758-100G)) and protease inhibitors (Pierce, Thermo Fisher Scienti c (A32955)) for 20 min at 4 o C on a shaking table. The buffer was collected and centrifuged at 1000g for 5 min, and the supernatant was subsequently analyzed for VWF. Plasma and intracellular levels of VWF were measured using a VWF Human ELISA Kit (Thermo Fisher Scienti c (#EHVWF)).

IF on HUVEC cells ( g. 3 F)
The HMEC cells with or without RAB3GAP2 knockdown were seeded on glass bottomed dishes and cultured until 70-80% con uence. The cells were xed by 3% PFA and permeabilized with Perm buffer (BD). Primary anti-VWF antibody (ab6994, Abcam) was diluted (1:200) in blocking solution (5% donkey serum in PBS) and incubated with the cells for overnight at 4°C. The cells were then washed three times and incubated with cy2 conjugated secondary antibody (Jackson ImmunoResearch, Cat 711-225-152, 1:300) for 2 hours in room temperature. The images were acquired by Meta510 confocal system (Zeiss, Germany) with excitation wavelength 488 nm and emission lter between 500-530 nm. The quanti cation of VWF intensity performed by ZEN2012 software on the cells under same conditions.

Immunoblotting
The following antibodies were used for immunoblotting: rabbit anti-RAB3GAP2 (PA555296, ThermoFisher), mouse anti-a-tubulin (Ab7291, Abcam), rabbit anti-GAPDH (Ab181602, Abcam). Secondary HRP-conjugated antibodies (goat anti-rabbit, goat anti-mouse) were from Bio-Rad. For tissue lysate blots, a preblotted membrane with different human tissue lysates was purchased (INSTA-Blot NBP2-31378 from Novusbio). For analyses of CRISPR/Cas9 knockdown e ciency on RAB3GAP2 the cells were lysed in 2% SDS in PBS with protease inhibitor cocktail (Roche), passed through QIAshredder minispin columns (Qiagen) to reduce the viscosity, and the protein content was quanti ed using BCA Protein Assay Kit (ThermoFisher). 20 mg of protein was run on a 4-20% Mini-Protean TGX SDS-PAGE gel (Bio-Rad) and electroblotted onto nitrocellulose membrane (Bio-Rad). From this point, for both human tissue lysate blots and the CRISPR knockdown analysis, the same blotting protocol was used. The membrane was rinsed in TBS, blocked in 5% bovine serum albumin in TBS, for 1 h at room temperature, then transferred to antibody incubation buffer (1% bovine serum albumin in TBS with 0.05% Tween-20) with anti-RAB3GAP2 diluted 1:2000, and incubated overnight at 4 °C. Next, the membrane was washed 4 x 5 min in wash buffer (TBS with 0.05% Tween-20) and incubated with HRP-conjugated anti-rabbit diluted 1:10000 in antibody incubation buffer, for 1 h, at room temperature. The membrane was washed 4 x 5 min in wash buffer and developed using SuperSignal West Pico PLUS Chemiluminescent Substrate (ThermoFIsher) and a CCD camera (Bio-Rad). For the second blotting step the membrane was stripped in Restore Western Blot Stripping Buffer (ThermoFisher), then the above-described blotting protocol was used, but using mouse anti-tubulin (diluted 1:5000) or rabbit anti-GAPDH (diluted 1:10000) antibodies, and species-speci c secondary HRP-conjugated antibodies.

CRISPR/Cas9 genomic editing of HMEC cells
Synthetic sgRNA containing the sequence ATCTCCAACCAATGATCTTA complementary to the protein-encoding sequence in RAB3GAP2 exon 3 (cut site at Leu-88) was obtained from Integrated DNA Technologies (USA). Electroporation of CRISPR/Cas9-sgRNA complex into HMEC cells: 135 pmol Cas9 protein (IDT) was combined with 150 pmol sgRNA in total volume of 5 ul and incubated for 15 min. Next, the ribonucleoprotein complex was combined with 1 million HMEC cells suspended in 100 ul nucleofection buffer kit V (Lonza) and electroporated using T-020 program on Amaxa nucleofector IIb (Lonza), and the cells were seeded in 6-well plates. To evaluate the CRISPR/Cas9 editing e ciency genomic DNA was extracted from the cells ve days postnucleofection using DNeasy Blood & Tissue Kit (Qiagen) and used to PCR amplify the genomic region surrounding the sgRNA-targeted locus. The primers used were: GGAGAAAGAGGGAAATGGAGAG and CCCACAGGAAGAAGAAGGAAATA. PCR was performed using AmpliTaq Gold 360 Master Mix (ThermoFisher) following manufacturer's protocol, but with annealing temperature of 58 °C. PCR amplicons were then puri ed using GeneJET PCR Puri cation Kit (ThermoFisher) and submitted for Sanger sequencing (Euro ns Genomics) using the primer AAACTTCCTGGGCTCCAAGATTG. Additionally, the CRISPR/Cas9 editing e ciency was evaluated using Alt-R Genome Editing Detection Kit (IDT) that employs T7 endonuclease to detect indel-caused heteroduplexes in re-hybridized PCR amplicons. The edited cells were then expanded, and RAB3GAP2 protein content was detected using immunoblotting.

Matrigel Tube-Forming Assay
The Matrigel tube-forming assay is a speci c and extensively used tool for studying angiogenesis in vitro, and a detailed experimental protocol is previously described 50 . Brie y, HMEC-1 cells, cultured to 80-90% con uence in growth medium, were seeded on 12-well plates precoated with Matrigel at a density of 50k cells / well, and were allowed to incubate for 24h at 37 o C in 5% CO 2 . Images of the tubular network formation (between 3-4 images / well) were captured using a brightfield microscope at regular time intervals during the 24 h period. Network formation was analyzed and quanti ed using the online automated image analysis platform myWim (Wimasis Image Analysis, Cordoba, Spain). The identity of the different images was not disclosed during the analysis.
In vivo angiogenesis plug assay All animal experiments were approved by the local ethical committee for animal care in Lund (M167-15). Matrigel plugs were prepared according to Nowak-Sliwinska et al 51 . Harvested HMECs in PBS were mixed with ice cold, overnight thawed Matrigel (GFR, Phenol Red-free, BD Biosciences), at a nal concentration of 5 mg/ml Matrigel, without additional supplements and kept on ice prior to injection. Eight to eighteen-week-old female NSG mice were anaesthetized with iso urane, the groin was shaved and cleaned before injection of 100µl cold Matrigel solution containing 1x10 6 HMECs per plug into the groin area. After 30 seconds gelling time, the needle was removed, and mice were removed from anesthesia. Following 7 days, mice were anesthetized with 2.5% Avertin (12,5 mg/kg body weight; Sigma-Aldrich), plugs were removed and xed overnight before para n embedding. Sections were stained with the connective tissue stain, Trichrome (Abcam ab150686). "-" and "+" indicates direction of effect for the non-reference allele, "x" = either phenotype or genotype not measured in the dataset