Novel genetic associations for blood pressure identified via gene-alcohol interaction in up to 570K individuals across multiple ancestries

Heavy alcohol consumption is an established risk factor for hypertension; the mechanism by which alcohol consumption impact blood pressure (BP) regulation remains unknown. We hypothesized that a genome-wide association study accounting for gene-alcohol consumption interaction for BP might identify additional BP loci and contribute to the understanding of alcohol-related BP regulation. We conducted a large two-stage investigation incorporating joint testing of main genetic effects and single nucleotide variant (SNV)-alcohol consumption interactions. In Stage 1, genome-wide discovery meta-analyses in ≈131K individuals across several ancestry groups yielded 3,514 SNVs (245 loci) with suggestive evidence of association (P < 1.0 x 10−5). In Stage 2, these SNVs were tested for independent external replication in ≈440K individuals across multiple ancestries. We identified and replicated (at Bonferroni correction threshold) five novel BP loci (380 SNVs in 21 genes) and 49 previously reported BP loci (2,159 SNVs in 109 genes) in European ancestry, and in multi-ancestry meta-analyses (P < 5.0 x 10−8). For African ancestry samples, we detected 18 potentially novel BP loci (P < 5.0 x 10−8) in Stage 1 that warrant further replication. Additionally, correlated meta-analysis identified eight novel BP loci (11 genes). Several genes in these loci (e.g., PINX1, GATA4, BLK, FTO and GABBR2) have been previously reported to be associated with alcohol consumption. These findings provide insights into the role of alcohol consumption in the genetic architecture of hypertension.

National Heart, Lung, and Blood Institute (NHLBI). Brenda W.J.H. Penninx (BWJHP) has received research funding (non-related to the work reported here) from Jansen Research and Boehringer Ingelheim. Mike A. Nalls (MAN) is supported by a consulting contract between Data Tecnica International LLC and the National Institute on Aging (NIA), National Institutes of Health (NIH), Bethesda, MD, USA. MAN also consults for Illumina Inc., the Michael J. Fox Foundation, and the University of California Healthcare. MAN also has commercial affiliation with Data Tecnica International, Glen Echo, MD, USA. Mark J. Caulfield (MJC) has commercial affiliation and is Chief Scientist for Genomics England, a UK government company. Oscar H Franco (OHF) is supported by grants from Metagenics (on women's health and epigenetics) and from Nestlé (on child health). Peter S. Sever (PSS) is financial supported from several pharmaceutical companies which manufacture either blood pressure lowering or lipid lowering agents, or both, and consultancy fees. Paul W. Franks (PWF) has been a paid consultant in the design of a personalized nutrition trial (PREDICT) as part of a private-public partnership at Kings College London, UK, and has received research support from several pharmaceutical companies as part of European Union Innovative Medicines Initiative (IMI) projects. Fimlab LTD provided support in the form of salaries for author Terho Lehtimäki (TL) but did not have any additional role in the study design to publish, or preparation of the manuscript. Gen-info Ltd provided support in the form of salaries for author Ozren Polašek (OP) but did not have any additional role in the study design to publish, or preparation of the manuscript. The specific roles of these authors are articulated in the 'author contributions' section. There are no patents, products in development, or marked products to declare. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. and the National Institute on Aging (NIA), National Institutes of Health (NIH), Bethesda, MD, USA. MAN also consults for Illumina Inc., the Michael J. Fox Foundation, and the University of California Healthcare. MAN also has commercial affiliation with Data Tecnica International, Glen Echo, MD, USA. Mark J. Caulfield (MJC) has commercial affiliation and is Chief Scientist for Genomics England, a UK government company. OHF is supported by grants from Metagenics (on women's health and epigenetics) and from Nestlé (on child health). Peter S. Sever (PSS) is financial supported from several pharmaceutical companies which manufacture either blood pressure lowering or lipid lowering agents, or both, and consultancy fees. Paul W. Franks (PWF) has been a paid consultant in the design of a personalized nutrition trial (PREDICT) as part of a private-public partnership at Kings College London, UK, and has received research support from several pharmaceutical companies as part of European Union Innovative Medicines Initiative (IMI) projects. Terho Lehtimäki (TL) is employed by Fimlab Ltd. Ozren Polašek (OP) is employed by Gen-info Ltd. There are no patents, products in development, or marked products to declare. All the other authors have declared no competing interests exist. This does not alter the authors' adherence to PLOS ONE policies on sharing data and materials.

Abstract
Heavy alcohol consumption is an established risk factor for hypertension; the mechanism by which alcohol consumption impact blood pressure (BP) regulation remains unknown. We hypothesized that a genome-wide association study accounting for gene-alcohol consumption interaction for BP might identify additional BP loci and contribute to the understanding of alcohol-related BP regulation. We conducted a large two-stage investigation incorporating joint testing of main genetic effects and single nucleotide variant (SNV)-alcohol consumption interactions. In Stage 1, genome-wide discovery meta-analyses in %131K individuals across several ancestry groups yielded 3,514 SNVs (245 loci) with suggestive evidence of association (P < 1.0 x 10 −5 ). In Stage 2, these SNVs were tested for independent external replication in %440K individuals across multiple ancestries. We identified and replicated (at Bonferroni correction threshold) five novel BP loci (380 SNVs in 21 genes) and 49 previously reported BP loci (2,159 SNVs in 109 genes) in European ancestry, and in multi-ancestry meta-analyses (P < 5.0 x 10 −8 ). For African ancestry samples, we detected 18 potentially novel BP loci

Introduction
Hypertension is a major risk factor for cardiovascular disease (CVD) [1], which in 2015 alone was estimated to cause about 10.7 million deaths worldwide [2]. The prevalence of hypertension in the US is~46% for those of African ancestry compared to~33% for European ancestry and~30% for Hispanic ancestry [3] based on previous blood pressure (BP) guidelines (The Seventh Report of the Joint National Committee on Prevention) [4]. Recently, based on the 2017 American College of Cardiology/ American Heart Association high BP guideline, the overall prevalence of hypertension among US adults is estimated at 45.6% [5]. Blood pressure levels are influenced by alcohol consumption independently of adiposity, sodium intake, smoking and socio-economic status [6]. Alcohol shows a dose-dependent effect on systolic BP (SBP) after adjusting for environmental confounders [7].
Genome-wide association studies (GWAS) have identified more than 400 single nucleotide variants (SNVs) for BP [8][9][10][11][12][13][14] and about 30 SNVs for alcohol consumption [15][16][17]. However, few studies have explored SNV-alcohol interactions in relation to BP [18,19], in part due to the large sample sizes required to obtain adequate power [18,20]. SNVs, which effect differ by level of alcohol consumption, can harbor modest marginal effects and might therefore be missed by standard marginal effects association screening. As previously demonstrated, a joint test of main genetic effect and gene-environmental interaction can have higher power [21] to identify such variants.
Within the CHARGE Gene-Lifestyle Interactions Working Group [22,23], we studied a total of 571,652 adults across multiple ancestries to identify variants associated with SBP, diastolic BP (DBP), mean arterial pressure (MAP), and pulse pressure (PP). We tested a model that included a joint model of SNV main effect on BP and SNV-alcohol consumption interaction, in each ancestry and across ancestries. Alcohol consumption was defined by two categories: (I) as current drinking (yes/no), and (II) in the subset of drinkers, as light/ heavy drinking (1-7 drinks/week or !8 drinks/week). Individual cohort results were metaanalyzed using a modified version of METAL applicable to the statistics summary results accounting for interactions [24]. We also performed multi-trait correlated meta-analyses [25,26] in participants of European ancestry using the joint model P-values from each meta-analysis of all four BP traits.

Genetic associations for BP identified via gene-alcohol interaction
The overall description of the CHARGE Gene-Lifestyle Interactions Working Group was previously reported [22,23]. We studied the joint model of SNV main effect and SNV-alcohol consumption interaction for BP in a two-stage study design, as depicted in S1 Fig. GWAS discovery (Stage 1), was conducted in each of 47 multi-ancestry cohorts including a total of 130,828 individuals of African ancestry (N = 21,417), Asian ancestry (N = 9,838), Brazilian (4,415), European ancestry (N = 91,102), and Hispanic ancestry (N = 4,056) (S1-S4 Tables and S1 Note). A total of 3,514 SNVs (245 loci) attained P < 1.0 x 10 −5 in Stage 1 meta-analyses (for at least one combination of BP trait and alcohol consumption status in one ancestry or multi-ancestries). We considered a locus to be independent, if our lead variant (i.e., most significant) was in low linkage disequilibrium (LD, r 2 0.2) and at least 500 kb away from any variant associated with BP in previous GWAS (P 5.0 x 10 −8 ). The meta-analysis distributions of-log 10 P-values of observed versus-log 10 P-values expected (QQ plots) are shown in S2 and S3 Figs.
The 3,514 SNVs were taken forward to replication, Stage 2, which included 440,824 individuals from 68 cohorts of African ancestry (N = 5,041), Asian ancestry (N = 141,026), European ancestry (N = 281,380), and Hispanic ancestry (N = 13,377, S5-S8 Tables and S1 Note). We identified and replicated (Stage 2, at Bonferroni correction P < 0.0002) five novel BP loci in European ancestry, four loci on 8p23.1 and one locus (FTO) on 16q12.2, which included 380 SNVs in 21 genes. These findings achieved genome-wide statistical significance (P < 5.0 x 10 −8 ) in Stage 1 and Stage 2 combined meta-analyses. Tables 1 and 2 show the most significant SNVs per BP trait, per alcohol consumption and gene for European ancestry participants. The loci containing novel BP associations at 8p23.1 were detected for all four BP traits in current drinkers and in light/heavy drinkers. The regional association plots on chromosomes 8p23 and 16q12 in European ancestry are shown in S4 and S5 Figs. For African ancestry, 18 potentially novel BP loci were found in discovery (P 5.0 x 10 −8 ), but without replication (Table 3). Further, we performed combined meta-analyses of Stage 1 and Stage 2 across all ancestries, which reproduced our European ancestry findings (P 5.0 x 10 −8 , Table 4 and S9 Table). We also identified and replicated 49 previously reported BP loci (2,159 SNVs in 109 genes) for European ancestry participants (S10 Table). For African Ancestry, and multi-ancestry analyses, additional reported BP loci were significant (P < 5.0 x 10 −8 ) in Stage 1 and Stage 2 combined meta-analyses (S11 and S12 Tables). Manhattan plots for BP trait and alcohol consumption status are shown in S6-S15 Figs, for Stage 1 and Stage 2 combined meta-analyses of European, African and Asian ancestries.
Finally, we leveraged the added power of correlated meta-analysis [25,26] for BP traits to detect additional variants. We performed correlated meta-analysis on P-values from METALmeta-analysis [24] of DBP, SBP, MAP and PP traits separately for current drinkers and light/ heavy drinkers in Stage 1 European ancestry cohorts. A variant was considered pleiotropic if the P-METAL-meta reached P 0.0001 in two or more BP traits and the correlated metaanalysis P-value was P 5.0 x 10 −8 [27]. We identified eight novel BP loci (11 genes, Table 5),  the above five novel loci (14 genes, Tables 1 and 2), and the 22 previously reported BP loci (49 genes).

Protein-protein interactions and pathways enriched for BP genes
The protein-protein interactions (PPI) analyses showed that several novel gene proteins are important hubs in interaction with many other proteins. For example, MAPKAPK2 (1q32.1, Table 5) interacts among others with BAG2, LISP1 and ELAVL1. ELAVL1 interacts also with

Discussion
This is the first large-scale study to systematically evaluate the role of joint effect of main gene and gene-alcohol interaction on BP in a very large meta-analysis across multiple ancestries.  Associations for blood pressure identified via gene-alcohol interaction Table 5. Novel SNVs/Genes associated with BP traits from correlated meta-analysis in European ancestry in Stage 1. Associations for blood pressure identified via gene-alcohol interaction  (Tables 1 and 2). Chromosome 8p23.1 is a complex region of deletions and replications, with repeated inverse structures [39,40]. We identified four LD blocks in 8p23.1 (Fig 1). The significant GWAS results on 8p23.1 are from European ancestry participants in Stage 1, Stage 2 follow up, and combined Stage 1 and Stage 2 metaanalyses. For this region, the evidence of genetic associations was identified from all four BP traits at both current drinking and light/heavy drinking status (Tables 1 and 2). The association on 8p23.1 found in the large European ancestry sample may also occur in other ancestries. The genome-wide significance levels in meta-analysis of European ancestry combined with African (5 genes), Asian (2 genes), and/or Hispanic (9 genes) ancestries have shown small improvements in their P-values compared to European ancestry meta-analysis alone (Tables 4  and S9). For some of these associated SNVs on 8p23.1, the allele frequencies in European ancestry are higher than in African ancestry (e.g., rs4841294: 0.44 versus 0.25, respectively), and Hispanic Ancestry (e.g., rs34919878: 0.42 versus 0.25, respectively). These findings suggest the presence of cross-population association patterns between European, African, and Hispanic ancestries, although they are not genome-wide significant in African and Hispanic ancestries presumably because of small sample sizes. Several of the genes residing on 8p23.1 have been reported for alcohol metabolism and/or dependence. Overexpression of PINX1 was reported to be associated with alcohol-related Associations for blood pressure identified via gene-alcohol interaction cirrhosis and fibrosis [41]. The transcription factor GATA4 has been reported to be associated with alcohol dependence in several studies [42][43][44][45]. GATA4 was suggested to regulate atrial natriuretic peptide (ANP, officially known as NPPA) modulating the amygdala's response to alcohol dependence [39] and is associated with BP [46]. In addition, a suggestive GWAS finding was observed between a variant near BLK-LINC00208 with alcohol dependence [47]. The S2 Note provides a comprehensive summary of novel and neighboring genes and their potential biological relevance.

BP genes interacting with alcohol show association with alcohol metabolism or dependence
FTO (16q12.2) variants in interaction with alcohol consumption were significant for BP in European ancestry (Table 2) and in combined meta-analysis of European and Asian ancestries ( Table 4). FTO is involved in the regulation of thermogenesis and the control of adipocyte differentiation into brown or white fat cells [48]. FTO variants have been associated in diverse ancestries with obesity-related traits [49,50], as well as alcohol consumption and alcohol dependency [51,52]. Frequency of alcohol consumption was suggested to modify the effect of FTO variants on body mass index [53].
IL10 (interleukin 10,~49 Kb upstream of rs3813963, Table 5) has been associated with hypertension [54] and with alcoholic cirrhosis [55]. MALAT1 (ncRNA,~390 Kb upstream of rs201407003) is upregulated in the cerebellum, hippocampus and brain stem of alcoholics [56], which may represent an important mechanism for alcohol actions in the central nervous system.
It is worth to note that the allele frequencies for several potential SNVs in African ancestry (Table 3) are low (<0.10) but they are monomorphic in Europeans, which may suggest African-specific associations. Even though we did not have true replications for African ancestry associations (some of them due to missing SNVs or very low sample size in Stage 2), the identified candidate loci include genes previously related to alcohol consumption and dependence (Table 3). GABRB1 [57] (4p12) and GABBR2 [58] (9q22.33, 143 kb upstream of rs73655199) are major neurotransmitters in the vertebrate brain, representing ligand-gated ion channels and have been shown to associate with alcohol dependence. EYS (6q12) displayed association with alcohol dependence in multi-ancestry population studies for rare [59] and common [60] variants. LINGO2 (9p21.1) was reported to be associated with age at onset of alcohol dependence in the Collaborative Study on the Genetics of Alcoholism [16]. ERCC6 (10q11.23) participates in DNA repair in response to oxidative stress [61]. Carriers of Arg1230Pro at ERCC6 had a decreased risk for laryngeal cancer, strongest in heavy smokers and high alcohol consumers [62]. CHAT (10q11.23, 136 kb downstream of rs4253197) encodes an enzyme that catalyzes the biosynthesis of the neurotransmitter acetylcholine, and binge ethanol in adolescents was reported to decrease CHAT expression [63]. BAG3 (10q26.11, 183 Kb downstream of rs201383951) was also suggested to contribute to alcohol-induced neurodegenerations [64]. A mouse study suggested that BAG3 exerts a vaso-relaxing effect through the activation of the PI3K/Akt/eNOS signaling pathway, and may influence BP regulation [64]. A GWAS identified association of BAG3 with dilated cardiomyopathy [65], and suggestive association with alcohol dependence [44]. SGK1 (409 kb upstream of rs76987554) is associated with increased BP [66] and may contribute to the mechanisms underlying behavioral response to chronic ethanol exposure [67]. In addition, our two potential genes by alcohol interaction, TARID (rs76987554) and CDH17 (rs115888294), have been recently reported association with BP in African ancestry, which supports our findings [68].

Regulatory features of BP genes
Analysis of our significant BP variants for cis-transcription regulation via HaploReg [29] (S13 Table) showed that in total about 11% of variants were localized in promoter histone marks, 55% in enhancer histone marks, 34% at DNAse hypersensitive sites, 10% located at protein regulatory binding sites, and 88% were predicted to change regulatory protein binding motifs. These feature findings are inflated, because several variants are in LD blocks. Several of our variants had P-values 5.0 x 10 −8 for being eQTLs for one or more target genes. The rs2921053 is the best eSNV regulating the transcription of SGK223 in thyroid tissue (Pvalue = 1.04 x 10 −67 ). Thyroid hormones are known to affect BP, heart and cardiovascular system [69].

Pathways enriched for BP genes
Our findings, TNKS (Table 1), FSTL5 and MAPKAPK2 (Table 5) and many other genes from PPI networks (S17 Fig), are part of Wnt/beta-catenin [70] signaling pathway. The TNKS forms a complex for degrading β-catenin (CTNNB1) [70] in interaction with AXIN1, AXIN2, and glycogen synthase kinase 3β (GSK-3β) (S17 and S18 Figs). The Wnt/beta-catenin pathway is known to be involved in renal injury and fibrosis induced by hypertension [71]. In addition, TNKS is involved in the regulation of GLUT4 trafficking in adipocytes [72]. Other findings from correlated meta-analysis also contributed to pathways. For example, rs206648224 is intronic to DYRK3, 37 Kb upstream of MAPKAPK2, and 119 Kb downstream of IL10. MAP-KAPK2 is a stress-activated serine/threonine-protein kinase involved in cytokine production especially for TNF and IL6, and phosphorylates among others LSP1, already identified in association with BP [9]. MAPKAPK2 [73] augments and FSTL5 [74] diminishes the expression of Wnt/β-catenin signaling pathway.

Limitations
Despite large sample sizes in Stages 1 and 2 (%131K individuals and %440K individuals, respectively), our novel variants (8p23 and 16q12) are common in their allele frequencies. For an analysis of gene by alcohol interactions in BP, even larger sample sizes are required to have sufficient power for detecting (and replicating) variants with lower allele frequency in the genome.
Our findings were based on a joint test of the main and interaction effects, which limits our ability to statistically differentiate the effect of interaction from the main effect. However, there is evidence that several of our novel and previously reported findings suggest association with alcohol consumption and dependency.
For African ancestry, the findings were not replicated, due to low sample size in Stage 2 (%3K individuals) versus Stage 1 (%21K individuals) and because seven potential variants for African ancestry were not available in Stage 2.
There are fewer associations of SNVs interacting with light/heavy drinkers compared to current drinkers, which is probably due to the reduced sample size in light/heavy drinkers. We also found an association in light/heavy drinkers which is not present in current drinkers. The LOC105374235 gene interacts with light/heavy drinkers for SBP but does not interact with current drinkers for SBP in African ancestry (Table 3 and S10 Fig). These findings suggest that novel loci for BP can be expected to be discovered when increasing the sample size for light/ heavy drinkers.
The two Brazilian cohorts (from discovery only) were included in the multi-ancestry metaanalyses. However, their association results did not contribute to SNV-alcohol interactions for BP traits, which could be in part to the relative small sample size (4,415 subjects) affecting the power of associations in the joint gene-environmental interaction model.

Conclusion
We identified and replicated five novel loci (380 SNVs in 21 genes) via joint test of main genetic effect and gene-alcohol interaction, and eight novel loci (11 genes) using correlated meta-analysis in European ancestry. We also found 18 potentially novel BP loci in discovery (P 5.0 x 10 −8 ) in gene-alcohol interaction model in African ancestry participants, but without replication. In addition, we identified 49 loci previously reported for BP (2,159 SNVs in 109 genes) using the joint test for interaction in European and multi-ancestries meta-analyses. Several of these SNVs/genes are related to alcohol metabolism and dependence, have evidence for regulatory features, and are enriched in pathways for cardiovascular disease, hypertension and blood pressure homeostasis. Our findings provide novel insights into mechanisms of BP regulation and may highlight new therapeutic targets.

Methods
Individuals between the ages of 18-80, who participated in the studies, provided written informed consent and approval by their research ethics committees and/or institutional review boards. The description of each participating study cohort is shown in S1 Note.

Phenotypes, alcohol consumption, and study cohorts
SBP (in mmHg) and diastolic BP (DBP in mmHg) were measured at resting or sitting positions by averaging up to three BP readings at the same clinical visit. To account for the reduction in BP levels due to anti-hypertensive medication use, the BP levels were adjusted by adding 15 mm Hg to SBP and 10 mm Hg to DBP values. After adjustment, mean arterial pressure (MAP) was defined as the sum of two-thirds of DBP and one-third of SBP, and pulse pressure (PP) was estimated as the difference between SBP and DBP. Hypertension was defined whether participants presented: (i) SBP ! 140 mm Hg, (ii) DBP ! 90 mm Hg, and/or (iii) taking anti-hypertensive medication. For quality control (QC), SE-N (i.e., inverse of the median standard error versus the square root of the sample size) plots were produced [75]. If cohortspecific analytical problems existed, they were corrected.
Definition of "a dose or a drink" is about 17.7 grams of ethanol, which is the amount of a typical beverage of 12 oz. (354.882 ml) bottle or can of beer, a 5 oz. (147.868 ml) glass of wine, or a standard 1.5 oz. (44.3603 ml) shot of 80-proof spirits, such as gin, vodka, or whiskey [76]. Alcohol consumption was defined by two categories: (I) as current drinking (yes/no), and (II) in the subset of drinkers, as light/heavy drinking (1-7 drinks/week or !8 drinks/week).

Genotyping
Genotyping was performed using Illumina (San Diego, CA, USA) or Affymetrix (Santa Clara, CA, USA) arrays. 1000 Genomes Imputation was implemented using MACH and Minimac, IMPUTE2, and/or BEAGLE software, based on the cosmopolitan panel from Phase I Integrated Release Version 3 Haplotypes (2010-11 data freeze, 2012-03-14 haplotypes). Dosages from 1000 Genomes were used in 106 cohorts out of 115 Stage 1 and Stage 2 cohorts. If 1000 Genomes were not available in a cohort, dosages based on HapMap Phase II / III reference panel (2 Stage 1 cohorts and 4 Stage 2 cohorts) or genotyped data (3 Stage 2 cohorts) were used in the analyses. Information of study characteristics, genotyping, imputation, covariates, and analyses are summarized for Stage 1 in S1-S4 Tables, and for Stage 2 in S5-S8 Tables.

Interaction association analysis
Each Stage 1 and Stage 2 cohort conducted a joint statistical model analysis [24]: where SNV is the dosage of the genetic (G) variant, E is the alcohol consumption (current drinker or light/heavy drinker) effect, SNV Ã E is SNV-alcohol interaction effect, b values are the respective beta coefficients from regression analysis and C represents covariates (age, sex, principal components (PCs), and other study-specific covariates). The joint model provides estimates of b G and b GE , robust estimates of the corresponding standard errors (SEs) and covariance, and P-values from the joint 2 degree-of-freedom Wald test. The SNV effect (b G ) is context-dependent and thus should not be interpreted as the "main effect" [23]. Principal components were derived from genotyped SNVs and used for controlling population stratification and genomic confounding effects. Each cohort decided the number of PCs to be included in the joint statistical model analysis, as shown in S4 Table (Discovery, in Stage 1) and S8 Table  (Replication, in Stage 2). Particularly for African ancestry, it was required to include at the least the first PC and additional PCs as appropriate.
The association analyses were implemented by programming in R or using ProbABEL [77] for studies of unrelated individuals, or by GenABEL/MixABEL [78] or MMAP (O'Connell, unpublished; personal communication), which account for family relatedness.

Meta-analysis and quality control
We employed a modified METAL software [24] to perform 2 degrees of freedom joint metaanalysis, using the inverse-variance weighted fixed-effects approach. We applied multiple steps of QC, both at cohort association analysis and at meta-analysis level, implemented with EasyQC, an R package [75]. They included filtering of markers with imputation quality < 0.5; with minor allele frequency < 1%; minor allele count 10; if alleles were mismatched when comparing the cohort's alleles with the 1000 Genomes cosmopolitan panel; and/or if the allele frequencies were different from those of the 1000 Genomes. In addition, a cohort participated in the meta-analysis if it had more than 50 individuals consuming alcohol. The meta-analysis results were reported if they had more than 5,000 individuals and if at least two studies for each SNV contributed to the analysis. Markers with meta-heterogeneity P < 1.0 x 10 −6 were dropped. We used (double) study-and meta-level genomic control corrections to account for population stratification accumulated across studies or due to unaccounted relatedness. Distributions of-log 10 P-values of observed versus-log 10 P-values expected (QQ plots) are shown in S2 and S3 Figs.

Correlated meta-analysis
The genome (millions of SNPs) are under the null hypothesis of no genotype-phenotype association, which is only mildly contaminated with a relatively smaller set of SNVs that are under the alternative. The correlated meta-analysis [25,26] performs a large sampling of genome and produces the polychoric correlation estimator (using SAS PROC FREQ). The estimator measures the relation degree of any non-independence between scans. The correlated meta-analysis corrects the inference for it, retaining the proper type I error structure. The correlated meta-analysis [25,26] uses the Fisher's 1925 method by combining P-values at each location of the genome. This technique uses the fact that for number of scans, sum of −2 ln (p i ), approximately chi-square (X 2 ) with two degrees of freedom. In the case of correlated GWAS, this sum is no longer distributed as a simple X 2 . Instead, the correlated meta-analysis method [25,26] uses an inverse-normal transform, Z i = θ −1 (p i ) forming the N dimensional vector Z of all Z i s. Then, the method applies the basic theorem of multidimensional statistics for the matrix D, if  Z~N(O, E) then DZ~N(O, E∑D'). In particular, when D is a 1×N vector of all 1's, SUM(Z) = D Z~N(0, SUM(∑)), whose tail probability gives the Z meta-analysis P-value. In this case, for estimating ∑, the SNV P-values are dichotomized across the genome as (P 0.5; P > 0.5). The software was developed in SAS.

Bioinformatics analyses
The annotation of variants was sourced from NCBI dbSNP build 138 (hg19) during the analyses and updated to dbSNP build 150 (hg38) for reporting results. Our candidate SNVs for BP were questioned if they resided in any of regulatory marks, analyzing information from the NCBI Entrez gene, dbSNP, Encyclopedia of DNA Elements Consortium (ENCODE) project and the Roadmap Epigenomics Mapping Consortium (ROADMAP), as summarized by Hap-loReg [28,29], and RegulomeDB [30,31].
HaploReg (v.4.1) queries were used to identify functional annotations including the chromatin state segmentation on the Roadmap reference epigenomes, conserved regions by GERP and SiPhy, the experiments of DNAse hypersensitivity and ChIP-seq experiments from ENCODE. UCSC Genome Browser and GENCODE were used for gene annotations. We calculated the proximity of each variant to a gene.
GWAS3D [33] (accessed on 03.15.2017) was used to analyze genetic variants that may affect regulatory elements, by integrating annotations from cell type-specific chromatin states, epigenetic modifications, sequence motifs and cross-species conservation. The regulatory elements are inferred from the genome-wide chromosome interaction data, chromatin marks in different cell types measured by high-throughput chromosome conformation capture technologies (5C, ChIA-PET and Hi-C) from ENCODE, Gene Expression Omnibus (GEO) database, published resources and regulatory factor motifs. We gathered also evidence for eQTLs based on GTEx (v. 7), GRASP software and special gene expression reported results [83,84].
The importance of our novel and potential novel BP genes (Tables 1-5) were mined by means of four methods: enrichment analysis, protein-protein interactions (PPI), analytical gene expression cis-regulation, and analytical gene expression trans-regulation.
The GeneGO and Literature Lab of ACUMENTA software (accessed on 03.15. 2017) were used for enrichment analysis. We tested if novel genes were significantly enriched among prespecified gene sets defined in pathways, or by shared roles in particular diseases or biological processes from Gene Ontology. The GeneGO enrichment analysis consists of matching unique gene symbols of possible targets for the "common", "similar" and "unique" sets with gene symbols in functional ontologies. The probability of a random intersection between a set of gene symbols, the size of target list with ontology entities, is estimated by P-value of a hypergeometric intersection. The lower P-value means higher relevance of the entity to the dataset, which shows in higher rating for the entity.
Literature Lab is an interface between experimentally-derived gene lists and scientific literature in a curated vocabulary of 24,000 biological and biochemical terms. It employs statistical and clustering analysis on over 17.5 million PubMed abstracts (from 01.01.1990 to the present) to identify pathways (809 pathways), diseases, compounds, cell biology and other areas of biology and biochemistry. The analysis engine compares statistically the submitted gene set to 1,000 random gene sets generated in the analysis to identify term relationships that are associated with the gene set more than by chance alone.
The BP candidate genes were assessed via PPI of databases from Biological General Repository for Interaction Datasets (BioGrid), Escherichia coli K-12 (EcoCyc), and Human Protein Database (HPRD) as summarized by the National Center for Biotechnology Information (NCBI, accessed on 02.28.2017). The gene list from PPI was evaluated using igraph package [85]. The network was built using our programs in SAS, to a Pajek format and imported into igraph in R language. "Google" PageRank algorithm provided the importance of genes (website pages) in a network, which was implemented by igraph.
Information of data analysis tools and databases, including their website links (when available) and the corresponding literature citations, are provided in S15 Table. Supporting information S1 Note. Description of participating studies. Study descriptions of discovery cohorts (Stage 1) and replication cohorts (Stage 2).  Table. Descriptive analyses for discovery data (Stage 1) in current drinkers. Characteristics of blood pressure (BP) in current drinkers (yes or no), within sub-sample of individuals with or without anti-hypertensive (BP Lowering) medications, and in combined samples; SBP, systolic BP; DBP, diastolic BP; MAP, mean arterial pressure; PP, pulse pressure; N, number of individuals; mean, mean levels; SD, standard deviation of mean; Min, minimum value; Max, maximum value; For each BP trait (SBP, DBP, MAP, and PP), the extreme BP values were winsorised if a BP value was greater than 6 SD, above or below the mean, setting the BP value exactly at 6 SDs from the mean. (XLSX) S2 Table. Descriptive analyses for discovery data (Stage 1) in light/heavy drinkers. Characteristics of blood pressure (BP) in light/heavy drinkers (1-7 drinks/week or !8 drinks/week), within sub-sample of individuals with or without anti-hypertensive (BP Lowering) medications, and in combined samples; SBP, systolic BP; DBP, diastolic BP; MAP, mean arterial pressure; PP, pulse pressure; N, number of individuals; mean, mean levels; SD, standard deviation of mean; Min, minimum value; Max, maximum value; For each BP trait (SBP, DBP, MAP, and PP), the extreme BP values were winsorised if a BP value was greater than 6 SD, above or below the mean, setting the BP value exactly at 6 SDs from the mean. (XLSX) S3 Table. Descriptive analyses for blood pressure (BP) stratified by alcohol consumption for discovery data (Stage 1). Characteristics of systolic BP and diastolic BP, after correcting for BP lowering medication and winsorizing observations. (XLSX) S4  RegulomeDB scoring has classes defined as 1b, 1d and 1f: likely to affect binding and linked to expression of a gene target, with details: 1b (eQTL + TF binding + any motif + DNase footprint + DNase peak); 1d (eQTL + TF binding + any motif + DNase peak); 1f (eQTL + TF binding/DNase peak), 2a and 2b: likely to affect binding, 3a: less likely to affect binding, 4, 5, and 6: minimal binding evidence, and 7: no data. This software was accessed on 06.15.2017. Regulatory function measured by Promoter histone marks, Enhancer histone marks, DNase (DNAse hypersensitivity), Proteins bound, Motifs changed. (XLSX) S14 Table. Novel SNVs/genes associated with BP traits for eSNV/eQTL using GTEx. Target genes (Tissues and P-Values). Association findings from European Ancestry (novel) and correlated meta-analysis (novel variants). The annotation of variants was sourced from NCBI dbSNP build 138 (hg19) during the analyses and updated to dbSNP build 150 (hg38) for reporting results. scoring has classes defined as 1b, 1d and 1f: likely to affect binding and linked to expression of a gene target, with details: 1b (eQTL + TF binding + any motif + DNase footprint + DNase peak); 1d (eQTL + TF binding + any motif + DNase peak); 1f (eQTL + TF binding/DNase peak), 2a and 2b: likely to affect binding, 3a: less likely to affect binding, 4, 5, and 6: minimal binding evidence, and 7: no data. This software was accessed on 06. 15 the Erasmus University Medical Center Rotterdam, the University of Utrecht, the Radboud Medical Center Nijmegen, and the Parnassia Bavo group, all in the Netherlands. We are grateful to all adolescents who participated in this research and to everyone who worked on this project and made it possible. UKB (United Kingdom Biobank, www.ukbiobank.ac.uk): This research has been conducted using the UK Biobank Resource. The UK Biobank data were analyzed from the data set corresponding to UK Biobank access application no. 236, application title "Genome-wide association study of blood pressure", with Paul Elliott as the PI/applicant. This work was supported by the UK-CMC and the BP working group.