Arterial stifness measures have been reported as independent markers of vascular agein1g,2, hypertension3,4, cardiovascular disease (CVD5,)6 and mortality6,7. Carotid-femoral pulse wave velocity (PWV) is rthefeerence standard method for measuring arterial stifness. Howe,vaenr alternative and more convenient non-invasive emthod is to record the digital blood volume waveformnsguisnifra-red fnger sensors (photoplethysmograp8,hwy)here measurements can be recorded in a seated positionatrher than supine position required for carotid-foermal PWV recordings. Vis automatic technique is able to detect the waveform formed by the digital volumee, pwuhlisch is created by two components. First, pressure is transmitted from the leg ventricle to the fnger (direct conmenpto) whilst the second component is due to the transmiiossn of pressure from the heart to the lower boedfyec(tred component) via the aorta. Ve digital volume pulsaenctherefore be visualised as a dicrotic waveforamn,d the interval between the peaks of the direct and refecetd components can be recorded to derive the artelrisatifness index (ASI), when divided into the individuals9height. Higher ASI values refect arterial walls with greater stifness, due to the earlier arrival of wave refeicot n. ASI has been shown to have close agreement whiotther techniques measuring arterial stifness, including PWV (r= 0.58, P < 0.01) and augmentation index (r= 0.80, P < 0.01)9.

Previous studies have evaluated the utility of ASaIs a potential clinical marker of CVD, having shogwonod sensitivity (87%) and specifcity (87%) when diferentiating between older men with coronary artery diassee (CAD) and younger men without CAD9. In a study of asymptomatic middle-aged patienttsh, e mean ASI value was signifcantly higher (P = 0.002) in individuals with at least one signifcant(>50%) coronary stenotic plaque than those without a>50% stenotic lesio1n0. More recently, ASI has been shown to be an independent predictor Locus C1orf21 TEX41 MRVI1 FOXO1

SNV of CVD, myocardial infarction and mortality inUtKheBiobank coho1r1t. During a median follow-up period of 2.8 years, the risk of CVD and myocardial infarcntifoor individuals with higher ASI were 27% and 38%higher respectively. For all-cause mortality, each standard deviation chnage in ASI has a hazard ratio of 1.08 (95% co-nf dence interval [CI], 1.0531.12) in a multi-variant adjusted model11.

Our current knowledge of the biological factorspaanthdways contributing to arterial stifness is limteid. Heritability studies of arterial stifness using PWV measurements have suggested moderate genetic croibnution with estimates up to 0.53 in twin studies12,13 and ranging between 0.26 and 0.40 in population ustdies14,15. Ve identifcation of single nucleotide variants (SNVmsa)y improve the current understanding of the mechaisnms controlling arterial stifness and may be a usefudladition to the development of disease risk modPerlse.vious genome-wide association studies (GWASs) for artelrsitaifness were mostly performed using PWV as thheepnotype. One of the frst arterial stifness GWASs waos n 644 individuals from the Framingham Heart Stu,duysing carotid-brachial PWV as the phenotype, and no vanrtias reached genome-wide signifcanc1e6. Subsequent genetic studies have reported inconsistent fndings but nonheave studied ASI as the phenotypic measure17. Ve wealth of data available within the UK Biobank study ofetrhse opportunity to produce more robust fndingaslianrge population cohort. We therefore performed a GWAS to identify SNVs associated with ASI aneddetxhpelobirological mechanisms underlying arterial stifness.

Four loci identie昀؀d for arterial stin昀؀ess index. In total, 143,590 UK Biobank participants met theniclusion criteria to derive ASI at the baseline visit (Supplementary Fig. S1). Ager genetic quality control (QC)cleuxsions, three GWASs were performed on 127,121 inddivuials of European ancestry (age 5±6 8.1, 48.1% males) with a mean ASI of 9.0 m/s. Our primary GWAS was preformed on rank based inverse normal transformeNdT(I) residuals and included mean arterial pressure (MAP) as a covariate. We also performed analyses eluxcding MAP and on untransformed ASI. Ve baseline visit charaecrtistics of the individuals included in the GWAS aer summarised in Supplementary Table S1. Following the GWAS, we reviewed the results and quantile-quantile (QQ) plots. Vere was minimal genomic infation in testasttistics (lambda = 1.097 for all models) under polygenic inheritance (Supplementary Fig. S2). Ve genome-wideSNV heritability of ASI in our cohort was estimd aatte 6.1% (standard error 0.4%).

We identifed four genome-wide signifcant loTciE(X41, FOXO1, C1orf21 and MRVI1) for ASI in our primary analysis (Table 1 and Fig. 1). Ve regional association plots are indicated in F2ig.V. ree of the locTi(EX41, FOXO1 and C1orf21) were genome-wide signifcant in the secondary GWAS that excluded the MAP adjustment (Supplementary Table S2a and Supplementary Fig. S3) and in the GWAS of untransformed ASI (Supplemeanryt Table S2b). Ve magnitude of efect sizes and assoctiaons of the signifcant loci were similar acroses pthrimary and secondary models. We observed no independent signals at any of the ASI loci.

Ve most signifcant associated variants in our primary GWAS were: rs1006923, an intronic variantTaEtX41 (β = 0.0293, P = 5.3 × 10−12), rs7331212 an intronic variant atFOXO1 (β = 0.0301, P = 2.2 × 10−11) and rs1930290 (β = −0.0230, P = 1.1 × 10−8), located in the gene region of the open readirnagmfe C1orf21. Ve intronic variant rs10840457 within the MRVI1 gene region on chromosome β11=(−0.0236, P = 3.4 × 10−8) was only genome-wide signifcant in our primary analysis, hoewver there was some support in the secondary analeys (P = 3.3 × 10−7) and untransformed ASIP( = 5.4 × 10−8).

Forty-three variants at non-overlapping loci hadgsguestive genome-wide signifcanceP(< 1 × 10−5) with ASI in our primary analysis (Supplementary Table S3). A few variants are located close to potentianlcdaidate genes of interest: rs9521719 in theCOL4A2 gene (β = 0.0215, P = 1.1 × 10−8), rs8107744 in the RSPH6A gene (β = −0.0276, P = 1.2 × 10−7). Other variants of note were rs371147897 foundtihne OR4A47 gene region (P = 1.6 × 10−7), rs9501489 located inDDR1 (P = 4.0 × 10−6) and rs149320025 located in theFUCA1 gene region on chromosomeP1 =( 4.0 × 10−7).

FunctionalannotationoffourASIloci. In our primary analysis, 219 candidate variants were identifed by SNP2GENE function in the Functional Mapping and Annotation of Genome-wide Association Studies (FUMA) platform18. Ve majority of SNVs and their proxie2s>(r0.8) were located in introns (63%), 31% were loceadt in intergenic regions, there was one acceptor splicaerviant at MRVI1, rs11042902 and the remainder were located in exons and 3′-untranslated regions (3′-UTRs). Of the 219 variants, 151 variants mapped to the four genome-wide signifcant loci.

Gene-based analysis, as computed by multi-marker analysis of genomic annotation (MAGM19A,w)hich mapped the output SNVs from BOLT-LMM to 18,666 periontcoding genes, identifed four genes that reached the gene-wide significance thresholdP(= 2.67 × 10−6). COL4A2 was the most significant gene for ASI (P = 1.41 × 10−8) and MRVI1 (P = 3.08 × 10−7), FOXO1 (P = 7.54 × 10−7) and FOXO3 (P = 2.34 × 10−6) were also signifcant (Supplementary Fig. S4). Two furtrhgeenes, TCF20 located 2.9 Mb downstream to rs1006923 (P = 5.37 × 10−6) and FBXO46 located on chromosome 1P9=( 6.72 × 10−6) indicated suggestive signifcance. Ve FBXO46 gene is located less than 67 kb away from rs8107744at the RSPH6A locus on chromosome 19 that was signifcant from single variant analyses usingnutransformed ASI valuesP( = 1.4 × 10−8, Supplementary Table S2b).

To identify further candidate genes at each locuse wreviewed results from expression quantitative tirtaloci (eQTL) analyses across 53 tissue types from Genotey-pTissue Expression (GTEx) database from FUM1A8. Signifcant eQTLs were observed foSrLC25A15 at the FOXO1 locus across several tissues including oesophagus, transformed fbroblasts and sun exposed skin in thloe wer leg (lead SNV rs12865518P, = 2.04 × 10−8), ZEB2 at the TEX41 locus in the aorta (lead SNV rs225238P3,= 3.85 × 10−5) and C1orf21 and APOBEC4 at the C1orf21 locus in nerve and brain tissue (Supplementary Table S4).

We also checked if the genetic variants identifiefdor ASI were associated with other traits using PhenoScanner20. Genome-wide associations were observed for vatrsiaant three of the four ASI loci. Variants at TEX41 (rs1006923) and MRVI1 (rs10840457) were associated with blood pressureratits and ASI from analyses in UK Biobank21. Signifcant associations were also observed wituhlspe wave peak to peak time (PPT) for rs1006923 (TEX41) and rs7331212 at the FOXO1 locus by the Neale lab21 that has publicly released GWAS results for over 4,200 phenotypes found within UK Biobank. Ve varitaant the TEX41 locus was additionally associated with CAD22 (Supplementary Table S5).

Discussion Our main fnding was the identifcation of four geneo-wmide signifcant loci for ASI in a large Europeanc-estry based population cohort, despite the relatively lo(0w.06) heritability observed for ASI. We furtheor lflowed-up with several candidate genes using bioinformatics analyses at each of the identifed loci.

Previous genetic association studies for arteritaiflsness, as summarised by Loganet al.17, have mainly used PWV as the phenotypic measure of stifness and theeprorted fndings were limited. A meta-analysis including GWAS results of 20,634 individuals from 9 disvceory cohorts and of 5,306 individuals from twoicraetpilon cohorts, all from European ancestry, identifed olonceus on chromosome 14 in t3h′-eBCL11B gene desert. Vis gene desert was shown to be associated with the coartid-femoral PWV (rs7152623, discovePr=y 2.8 × 10−10, replicationP = 1.4 × 10−6)23. However, variants at this locus were not statisctailly signifcantly associated with ASI (P = 0.08) in our dataset.

More recently, a GWAS for brachial-ankle PWV (baPW), Vinvolving 402 Korean patients (mean age 59 years, 59% male) with diagnosed CVD has been reported by Parektal.24. Two SNVs were found to be associated with baPWV 3 rs7271920 (P = 7.15 × 10−9) and rs10125157 (P = 8.25 × 10−7). Neither variant was signifcant (rs7271920, P = 3.51 × 10−1; rs10125157, P = 3.10 × 10−1) in the replication cohort in their study that i nlucded 1,206 individuals. We also observed non-signifcanretsults for both variants (rs7271920P, = 0.17; rs10125157, P = 0.81) in our study. It is not too surprising toseorvbe a lack of signifcant fndings for the twoavnatrsi reported by Park et al.24 in the UK Biobank cohort. A contributing factory mbeathe diference in the size of the popu-la tions, and the diferent ethnicities, and that the reported variants by Parket al.24 may be false positive fndings as there was no replication in their study.

Ve lack of replication of results on PWV in UKaBnikoabnd across PWV GWASs may also be due to theklac of methodological standardisation to derive PWV, as it can be measured at diferent sites such asidc-afroemt oral or brachial-ankle. Our fndings suggest that ASI andPWV may have diferent aetiologies and thus may pvriode independent data on the underlying biological mechanisms and potential cardiovascular riskofarsc.tA previous expert consensus statement on arterial stifness decsribed PWV as a measure of regional stifness, whAileSI is seen as a surrogate marker of stifness through warveefection assessments25. In other words, whilst PWV is determined mainly by the speed at which waveform tarvels, for ASI, other factors such as the refectpivoe int would also impact on its measurements leading to potential diferences in their aetiologies.

Our most signifcant variant for ASI, rs1006923 at thTeEX41 locus has previously been reported to be sig-nif icantly associated with CAD in a mixed population GWAS that included UK Biobank participan22t.sIt is located 129 kb upstream and in low linkage disequilibrium L(D; r2 = 0.19) to rs1438896 a variant reported by Warrenet al.26 with blood pressure traits. Another variant at thTeEX41 locus rs183032 is 50 kb downstream to our lead variant with a moderate LD (r2 = 0.38) is associated with aortic stenosis in an Iclaendic cohor2t7. Helgadottir (2018)27 suggested the tumour growth factoβr(T-GF-β) throughZEB227 as a candidate gene. ZEB2 has a role as a DNA-binding transcriptional repressor that inetracts with the main signal transducers for TGFβ-receptors (SMADs). Considering that TGFβ- is involved in the regulation of vascular smootuhscmle diferentiation, as well as in the collagen up-regulation in the vascaurl wall, changes in the expression of TGβF-have the potential to alter arterial stifness. We note additional supoprt foZrEB2 from eQTL data in aortic tissue in our analyses (Supplementary Table S4).

At the second ASI locusF,OXO1 represents a good candidate gene, rs7331212 is located within the O class of the forkhead family of transcription factors. Vis gereenome-wide signifcant association of the samareiavnt with pulse wave PPT21, which is the time interval between the peak values of the direct and refected components of the pulse waveform using to calculate A2S8I. Although the specifc function oFOfXO1 gene is not well described, it may play a role in blood pressure regulation.eScpifcally, lack ofFOXO1 has been shown to reduce expression of angiotensinogen, which is a precursor of anngsiion tIeI that mediates vasoconstriction, in knockomuotuse models29. Furthermore,FOXO1 is involved in the signalling axis that regulates mindi3n0, which has a role in neointima formation where there is vascular smooth mcluescell proliferation. Importantly, the signifcaenocf the association between our lead variant aFtOXO1 and ASI persisted ager adjustment for MAP in our study.

At our third ASI locus, there are few candidate geneCs.1orf21 is an uncharacterised protein-coding gene that at present has not been functionally annotated.

At the 4th locus rs10840457 is located near theMRVI1 (Murine Retrovirus Integration Site 1 Homologo,als known as IP3R-associated cGMP kinase substrateI(RAG)). Ve MRVI1 gene is responsible for encoding the MRVI1/IRAG protein, which is present in a number of tissuesnciluding aorta and trachea31, and is involved in smooth muscle contractility. Specifcally, there inshibition of calcium release from endoplasmiccrueltuim fo-l lowing co-expression oIRfAG and cGMP-dependent protein kinase type Iβ (cGKIβ) in the presence of cGMP31. In a study using IRAG-knockout mice, the authors concluded that signnalgli of cGKβIvia IRAG is a vital functional component in the regulation of smooth mtuoscnle and intracellular calcium by nitric oxide andatrial natriuretic peptide32. At this locusMRVI1 represents an interesting candidate, but we note this is a locus that was not genome-wide signifcant in the secondary analyses, thus further validation will be required.

Gene-based testing by MAGMA revealed COL4A2 as the most signifcant gene association with ASnI oiur cohort, withFOXO1, MRVI1 and FOXO3 also signifcant. Ve lead SNV from the GWAS at tChOeL4A2 locus is rs9521719 (P = 1.1 × 10−7). COL4A2, along with the adjacent geneCOL4A1, encodes the protein subunits of type IV collagen forming heterotrimers. Type IV colilsagaevnital structural component of basement membnreas and mutations in these genes are seen in disorders suchas myopathy, intracerebral haemorrhage and glaucoam33. A previous GWAS reported rs3742207 located near thCeOL4A1 to have strong associatioPn=( 7.08 × 10−7) with PWV in a cohort comprising of 4,221 Sardinian iindduiavls34. Vis variant was successfully replicated internalyl in 1,828 individuals and also in 813 Amish individuals. Ve PWV variant rs3742207 did not show any ascsoiation with ASI in our cohorPt =( 0.95). Vis result may add additional support on ethre being a diference in the genetic mechanism of ASI compared to PWV, both markers of arterial stifness.

We found the genome-wide SNV heritability of ASbIeto6%. Vis estimate is much lower than the reportde heritability for carotid-femoral PWV, which ranged between 0.363104,.3450.Vis dissimilarity might be explained by the diferences in stifness measure, covariates included and populations. In additnio,due to the methodol o-g ical inconsistencies mentioned above, it should be noted that the heritability of PWV would diferddienpgenon the site of measurement. Mitchelelt al.14 reported moderate heritability for carotid-fem oPWralV (h2 = 0.40) in their study of 1,480 participants in the FraminghamStudy ofspring cohort. However, when using theslceosmmon approach of measuring PWV between the carontidd barachial artery, the heritability estimate was lower (h2 = 0.09). In twin studies, the reported heritability estimates range between 0.38 and 0.5312,13,36 where the PWV measurements were either located at the wrist (aorto-radial) or foot (aorto-dorsalis-pedis).

Ve main strength of our study is the very large sapmle size that, despite the low estimated heritabiltiy for ASI, has enabled the identifcation of genetic varnitas not found in previous studies. Mean arterialepsrsure has been shown to have a strong infuence on arterial stifn1e5sso it was included as a covariate in the fully adjusted model, rather than systolic blood pressure (SBP)dandiastolic blood pressure (DBP). Despite using thlaergest cohort to date in identifying genetic variants faroterrial stifness, the main limitation of our stu disythat loci discovered therein require formal validation in aindependent dataset. However, an external study osfimilar sample size with ASI measurements is currently lacking. We note if we use a more stringePn-tvalue for reporting, P ≤ 1 × 10−8, then two loci would remain signifcanTt E(X41 and FOXO1). In addition, our study cohort consisted of middle-aged individuals of European-ancestryo,suor fndings may not be generalised to other agoegurps and ethnic populations.

In conclusion, we identifed four loci signifcanatslsyociated with ASI, an independent predictor of CDV and mortality. Ve two most signifcant loTciE, X41 and FOXO1, have SNV associations that may alter arterial stifness through blood pressure regulation and vascular smooth muscle diferentiation. Ourltsreasluso suggest an important role of calcium in the regulation moof osth muscle tone contributing to arterial stifn. eFsusrther research will be necessary to validate our discovedr loci in a separate cohort and their confrmati ocann potentially lead to the development of risk models that can be used in clinical practice.

Study population. Ve UK Biobank is a large population-based prospective study>o5f00,000 participants aged 40369 years at baseline recruited between 2006 and 2010 in England, Scotland and Wales. During the initial assessment visit, a broad range of biochemi,ccalinical and genotype data were collected and patricipants had a number of physical measurements. Ve UK Biobaknstudy was approved by the North West Multi-Centre Research Ethics Committee and all enrolled individauls have provided written informed consent foreccotiolln, storage, make availability of their data for hea-rltehlated research. All methods were carried out inaccordance with the relevant guidelines and regulations.

Arterial sti昀؀ness index measures. Pulse wave ASI (UK Biobank Field 21021), measured ni m/s, was derived using the pulse waveform obtained at the fgner (preferably index fnger of the non-dependentnhda though can be placed on any fnger or thumb) with an infra-red sensor (PulseTrace PCTAM,2CareFusion, USA). Ve shape of the volume waveform in the fnger isedctilry related to the time it takes for the pulse wvaeforms to travel through the arterial tree in the lower body and to be refected back to the fnger. Measuremenetrsewmade by clipping the device to a fnger and the readingsimade over 10315 seconds. As the participants9 hgeihts were unknown until ager the recording of the wavefor mta,dtahe actual ASI values were calculated (ASI = height/ PPT) by UK Biobank outside the assessment centre visit.

Data were taken at the UK Biobank Assessment Centres during the baseline recruitment between 29th April 2009 and 1st October 2010 inclusive (Nmax= 169,822). UK Biobank participants are free to witdhraw at any time and so 15 individuals were removed based on ethapplication specifc list of anonymised IDs. In didivuals with absence of notch position in the pulse waveform (Field 4204) were exclude=d 2(n5,288) as the notch in the digital volume pulse waveform indicates the refecdtecomponent of the pressure transmitted and is theefrore required to calculate PPT. Outlier ASI values, defned as three inter-quartile ranges below the frst quartile or above the third quartile (n= 81), were also removed from analyses. Finally, ptaicripants with missing height, weight or blood pressure measurements were removed from the fnal analyse=s (8n48).

Covariates. Blood pressure was measured using the Omron 705 IeTlectronic blood pressure mo-ni tor (OMRON Healthcare Europe B.V. Kruisweg 577 213N2A Hoofddorp). SBP and DBP were derived as the mean of the two recorded automated measurements (UBKiobank Fields 4079, 4080), except for 1,141 in-di viduals who only had one recorded reading. MAP emstaition was calculated using the traditional formu:la MAP = DBP + 1/3(SBP-DBP). Height (Field 50) was measured usinga Seca 202 device (Seca, Birmingham, UK). For individuals who reported use of anti-hypertevnesmiedications through questionnaires in UK bioba(nFkields 6177, 6153), their SBP and DBP were adjusting by adding 15 mmHg and 10 mmHg respectively to the mean recorded reading2s6,37. Ve ID numbers of the arterial stifness devices uesd were obtained from Field 4206 in UK Biobank.

Genotypic data. Central QC and imputation of genotypic data perfoedrmby UK Biobank has been previously described38. Briefy, genotypic data was obtained through eitrhUeK Biobank Axiom or UK BiLEVE Axiom arrays (Afymetrix Research Service Laboratory, Santa Clara, California, USA). Ve Haplotype Reenfceer Consortium (HRC) and the merged UK10K sequencing+1000 Genomes were used as a reference panels for imputation with preference for the HRC panel wheSreNPs were present in both panels. Vis study utilised the refreshed genetic dataset made available by UK Biobank in July 201378.

Genetic QC was performed in 164,835 individuals whounderwent assessment of arterial stifness using fgner photoplethysmography and had available genotypictad.aVis process excluded participants with either high missingness or high heterozygosity defned by UK Bbiaonk in their genotypic data (n= 345), as well as those with mis-match between self-reported and inferred sex from the genotype=s(1n16).

A 4-way k-means clustering analysis was performed accordintgo data from the frst and second principal components (PC1 and PC2) using the 8pvclust9 R paacgke (version 2.0303)9 to objectively identify the main ethnic groups (White, Asian, African and Chinese) within UK Biobank. <White= participants were defned for those present in the <White= cluster for both PC1 and PC2 analyses. Vey also needed to match their self-reproted ancestry though m<ixed, o<ther= and m<issing= were treated as being broad ethnicity. After restrictiotno European-ancestry only, 127,121 individuals remainde. Lastly, we performed SNV-level QC to exclude 2,73571 SNVs from genotyped SNVs in these individuals with the following thresholds: minor allele frequency (MAF) of 1%, Hardy-Weinberg equilibrium (P-value of 1× 10−6) and missingness of 0.015 using PLINK 14.90. Ager exclusions, a fnal total of 546,505 model SNVs was generated for our GWAS analyses.

from the regression of ASI against the phenotyopivcacriates (i.e., all except genotyping array and PsC) before performing genetic analyses as the distribution dindot approximate a normal distribution (Supplemenrtya Fig. S5). Ve heritability of ASI explained by addivte genetic variation was estimated using a varianec components method (BOLT-REML)51.

We performed three GWASs using BOLT-LMM (v2.3.2) sogwa4r1eacross our 546,505 model SNVs and ~9.9 million imputed SNVs with MA≥F 1% and imputation information (INFO) sco≥re0.3. Our primary ASI GWAS was on the INT residuals including the followingvcaoriates: age, age2, sex, weight, genotyping array (UK Biobank vs UK BiLEVE), device used to obtain pulse wavefo,rsmmoking status (current vs non-current smokersM), AP and frst 10 principal components (PCs). As a sensivtity analyses, we also performed a GWAS using unatnrsformed ASI values and same covariates as those included in our primary analysis.

A secondary GWAS where MAP was excluded as a covaraite was also performed. All GWASs assumed a linear mixed infnitesimal model method under an additiveegnetic model implemented in BOLT-LMM. Vis method accounts for cryptic population structure and allsothwe inclusion of related individuals permittingregater power compared to principal component analy4s1i.s

Genome-wide signifcance was defned asP-value ≤ 5 × 10−8 and R (version 3.5.1) statistical Sogwar4e2 was used to generate Manhattan plots and QQ plots. Regionasaslociation plots were made for genome-wide sigannitflcoci using the LocusZoom web-based platform (locuszorogm)w. oith the hg19/1000 Genomes (Nov 2014) EUR Bui.ld

Conditional analysis was performed for genome-wide signifcant locus to detect independent assnosciiga-tio nals using an approximate conditional and joint mtiupl e-SNP (COJO) analysis implemented in genome-wide complex trait analysis (GCTA) to o43l. A secondary signal was declared if all three ohfetfollowing conditions were met: (i) originalP-value of newly identifed variant wa<s1 × 10−6 (ii) ratio between the lead SNV and secondary associatioP-nvalues on a 3lo1g0 scale is 1.5 or less, (i.e., 3lo1g0(P lead)/−log10(P sec) < 1.5) (iii) ratio between the original association and conditional associaPt-iovanlues on a 3lo1g0 scale is 1.5 or less (i.e., 3log10(P)/−log10(P cond)< 1.5) Functional annotation. Summary statistics from our primary model GWAS analyses were uploaded onto the FUMA (v1.3.4c)18 web-based application (http://fuma.ctglab.nl)/ in order to perform functional annotations via its SNP2GENE function. Ve default FUMA settingwsere used whereby signifcant SNVsP( < 5 × 10−8), and those that were in LD (r2 ≥ 0.6), with a MAF ≥ 0.01 were selected for further annotation. In atdiodin, the maximum distance between LD blocks to merge into a locusswsaet at <250 kb and the UKB release 2 European was used as the reference panel population. Ve functional coqnuseences for these SNVs were then obtained with ANNVOAR44. For gene-based analysis, the integrated MAGMA v1.0169 was utilised in which SNVs are mapped to a gene according their genomic location before gene-level asastoiocins with ASI were tested. Here, aP-value with Bonferroni adjustment of 0.05/18697= 2.67 × 10−6 was used to defne genome-wide signifcance. For eQLTmapping, this was implemented by the FUMA platform, GTEx v7 databas4e5 (http://gtexportal.org/hom),e/53 tissue types was used as the gene expression reference data and the signficant threshold was defned as false discovery rat(eFDR) < 0.05. Genome-wide signifcant SNVs and those in LD with2>r 0.8 were examined for previously reported GWAS aos-s ciations with other traits using PhenoScann2e0,ran online search tool containing over 350 millpioubnlicly available association results. Ve signifcance threshold chonsfeor reporting therein waPs-value < 5 × 10−8.

Ve UK Biobank Resource is available, via applicatino, to all bona fde researchers undertaking health-erlated research that is in the public interest. Summary data are available online:www.ukbiobank.ac.uk/data-showcas.e Information on accessing the genetic and phenotdypatea used in this analysis can be found awtww.ukbiobank. ac.uk/using-the-resource./

K.F. is supported by Ve Medical College of Saint rBthaolomew9s Hospital Trust, an independent regisetedr charity that promotes and advances medical and denatl education and research at Barts and Ve London Shcool of Medicine and Dentistry. J.R. acknowledges support from the People Programme (Marie CuriioenAsc)tof the European Union9s Seventh Framework Programme (FP070/7232013) under REA grant agreement n° 608765 and the Medical Research Council grant MR/N025083/ 1. N.A. is supported by a Wellcome Trust Research Training Fellowship (203553/Z/16/Z). We also acknolewdge support from the National Institute for Hhealt Research (NIHR) Cardiovascular Biomedical ResearchCentre at Barts and Queen Mary University of Lond o,n UK. A.M.L. and S.E.P. acknowledge support from thS<meartHeart= EPSRC programme grant ( www.nihr. ac.uk; EP/P001009/1). Vis project was enabled through access to the MRC eMedLab Medical Bioinformatics infrastructure, supported by the Medical Research Council (Grant Number MR/L016311/1). Vis research has been conducted using the UK Biobank Resource (Appcalition 2964). Ve authors wish to thank all UK Bionbka participants and staf.

K.F. contributed in the conceptualisation, formal analysis, methodology, interpretatiohenroesfultts and writing the original drag. J.R. contributed in the formanlaalysis and critically revised the manuscript. H.R.W., A.M.L. and E.T. contributed in the data curation and resources. N.A. contributed in methodology. S.E.P. was involvede in th conceptualisation, interpretation of results, suvpiesrion and critically revised the original manuscprit. P.B.M. was involved in conceptualisation, methodology, foarnmalaylsis, interpretation of results, supervisiondancritically revised the original manuscript. All authors reviewed the manuscript drag.

Supplementary information accompanies this paper athttps://doi.org/10.1038/s41598-019-45703-.0 Competing Interests: Ve authors declare no competing interests.

Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional aoliations.

Open Access This article is licensed under a Creative CommonsAttribution 4.0 International License, which permits use, sharing, adaptation, dsitribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to- the Cre ative Commons license, and indicate if changes werme ade. Ve images or other third party material inhtis article are included in the article9s Creative Commons license, unless indicated otherwise in a crediltine to the material. If material is not included in the article9s Creative Commons license and your intended use is no-t per mitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, vhitstipt://creativecommons.org/licenses/by/4.0/