Populations

This study includes self-reported Hispanic and African American adults with genotyped or imputed Metabochip genetic variants from the Population Architecture using Genomics and Epidemiology (PAGE) Study [19] cohorts and the Family Blood Pressure Program (FBPP) [20]. Samples from the Multi-Ethnic Study of Atherosclerosis (MESA) were used for replication. Study descriptions and methods for BP measurements are included in the Methods A in S1 File. All participants have given consent for genetic studies and data sharing and all studies have Institutional Research Board approval.

To generalize novel findings to individuals of European ancestry, we used summary data from the HapMap II meta-analysis of the International Consortium of Blood Pressure (ICBP), which included a discovery sample of 69,395 European-descent individuals [9]. Summary statistics from meta-analyses are publicly available and were downloaded from http://www.ncbi.nlm.nih.gov/projects/gap/cgi-bin/study.cgi?study_id=phs000585.v1.p1

Genotyping and Outcomes

The Metabochip array design includes approximately 200,000 single nucleotide polymorphisms (SNPs) selected from GWAS loci as well as dense genotyping for fine-mapping of loci, as previously described [21]. Genotyping on PAGE samples was performed within PAGE [22]. Descriptions of the FBPP and replication study genotyping are summarized in the Methods A in S1 File. Standard QC filters were applied for samples and SNPs in PAGE: SNPs with call rates <95%, >2 replication errors, Hardy-Weinberg Equilibrium p-value<1 x 10⁻⁶, Illumina GenTrain Scores <0.6, Illumina Cluster Separation Scores <0.4 or discordant calls as specified in [22] were excluded. A subset of WHI African Americans had genome-wide data available from the SHARe GWAS; Metabochip genotypes were imputed in this sample, as described in Liu et al. [23]. Imputed genotypes with rsq<0.30 were excluded from analyses. Ancestry principal components were determined using Eigensoft software separately for each study [24]. Ancestry outliers and first-degree relatives were excluded for PAGE studies except for SOL, which adjusted for first degree relatives. Mt. Sinai BioMe used genotypes from the Illumina HumanOmniExpress chip, which were imputed to 1000 Genome (1000G) Phase I integrated haplotype panels (March 2012) using IMPUTE2. Only SNPs with imputation quality> 0.4 and available on the Metabochip were examined in Mt. Sinai BioMe.

All studies, except Mt. Sinai BioMe, have standardized measures of BP, including sitting systolic and diastolic BP measured by trained personnel, according to a standardized protocol that specified a rest period of 5 minutes prior to measurement, bias-free instrumentation and timed, repeat measurements (Methods A in S1 File). Mt. Sinai BioMe BP measures were extracted from electronic medical records (EMR). Hypertension was defined by a systolic BP > 140 and diastolic BP > 90 mm Hg or use of anti-hypertensive medication [25]. To correct for treatment effects, we added constants of 10 and 5 mm Hg to the systolic BP and diastolic BP, respectively, for individuals taking a BP lowering medication [7].

Statistical analysis

Each study prepared race/ethnicity-specific summary models. Genotypes were coded using additive genetic models, or allelic dosage for imputed SNPs. For studies of unrelated individuals, we used linear regression models to test associations of genotypes with quantitative BP traits. For the studies that included related family members, we used methods to account for the correlation structure due to genetic relatedness. Specifically, in the FBPP studies, we used variance component models for quantitative traits and liability models for hypertension to account for the family-based study design. The HCHS/SOL has approximately 2000 related individuals, and a complex sampling design was used for recruitment. Therefore, we used the W-PS method proposed by Lin et al, which is a weighted version of generalized estimation equations, to account for unequal inclusion probabilities and family relationships[26]. All of these models were adjusted for age, sex (except WHI), body mass index (BMI) and principal components of ancestry.

Summary results for Hispanics from each study were combined for an inverse variance weighted fixed-effects meta-analysis of Hispanics only. Similarly, all African American results were combined for an inverse variance weighted fixed-effects meta-analysis of African Americans. Results from these race-specific meta-analyses are presented. We then conducted trans-ethnic meta-analyses of African Americans and Hispanics by combining all study-specific African American and Hispanic summary results from PAGE and the FBPP using (1) a fixed effects and (2) a random-effects approach described by Han and Eskin [27] as implemented in their software, METASOFT (http://genetics.cs.ucla.edu/meta). Between study/race-ethnicity effect sizes consistency was assessed using the Q test (Chi-squared p-value) and the I² metric, where low I² suggests little between-study/race-ethnicity variability. For the meta-analyses, we filtered out SNPs with a minor allele frequency less than 0.01. Results for all analyses were reported as Betas and standard errors (SE).

We sought to replicate our main findings from the discovery meta-analyses in MESA Hispanics and African Americans with the same protocols used in the discovery analysis (see Methods A in S1 File), and validated our findings in European ancestry samples using publically available ICBP data. A conservative SNP discovery significance threshold for the Metabochip array, P < 2.8 x 10⁻⁷ was used to account for the numbers of tests performed in the discovery analyses.

For our replication of previously reported GWAS SNPs (identified mainly in populations of European ancestry) in our multi-ethnic population, we defined nominally significant as ‘p<0.05 and not corrected for multiple testing.’

Given the different recruitment and BP measurement in the Mt. Sinai BioMe sample, we performed a sensitivity analysis excluding the study from the Hispanic meta-analysis, and reviewed findings for consistency with the total Hispanic results. In a secondary analysis, we measured associations between hypertension and selected BP-associated SNPs using logistic regression models and reported odds ratios (OR) and 95% confidence intervals (95%CI).

We also fine-mapped 22 published BP loci previously identified in European and East Asian ancestry populations that were available on the Metabochip [Tables D-E in S1 File]. For these analyses, we focused on the African ancestry samples as this population, on average, has greater haplotype diversity and shorter LD blocks compared to Hispanics [28, 29]. In other words, a given SNP in a genomic locus of interest will typically have fewer numbers of SNPs highly correlated with it in African Americans, making it somewhat easier to reduce the size of a genomic locus of interest, and thus the number of potential causal SNP(s). To investigate SNPs in each BP-associated locus, we assessed whether the most significant SNP at each locus in African Americans was distinct from, but correlated (r² ≥ 0.50) with the index (i.e. previously reported GWAS) SNP based on LD information from the 1000 Genomes Project (1000G) European ancestry populations (EUR). Using Locus Zoom plots, we also plotted other candidate functional variants in the region. The fine-mapping significance threshold was set at the Bonferroni-corrected threshold of P<0.05/number of Metabochip SNPs at each locus. Other LD estimates in the 1000G populations were obtained using the LDlink tool [30].

Discovery and replication in Hispanics

In the discovery analysis, querying all SNPs on the Metabochip, there was little evidence for inflation due to population stratification in the Hispanic samples (Figure A in S1 File). Meta-analyses of the Hispanic studies identified 14 SNPs in three loci significantly associated with either systolic BP or diastolic BP (Table 2, and Tables A-B in S1 File). These SNPs were all in known BP loci rather than in Metabochip loci linked with other cardiometabolic traits. The most significant SNP (rs2586886/G) located in intron 1 of potassium channel, subfamily K, member 3 (KCNK3) on chromosome 12 was associated with a 0.72 mm Hg increase in diastolic BP (P = 5.2 x 10⁻⁹) (Table 2). This SNP was also associated with increased systolic BP (P = 3.8 x 10⁻⁴), and a 13% increased odds of hypertension (OR = 1.13; 95%CI: 1.06–1.22; P = 4.0 x 10⁻⁴) in 5822 hypertension cases and 13,477 controls. Other SNPs associated with BP in Hispanics were located at two previously established loci, FGF5 (chromosome 4) and ATXN2- SH2B3 (chromosome 12) (Tables A-B, and Figure B in S1 File). At FGF5, conditional analyses of the most significant SNP, rs1458038, which is also reported in individuals of European ancestry [31], supported the evidence for a single association at the locus (Table A in S1 File). At the ATXN2-SH2B3 locus, the most significant SNP rs11065979 was in LD with a missense variant rs3184504 (minor allele frequency = 0.28, P = 1. 4 x 10⁻⁷, r² = 0.91) (Table B and Figure B in S1 File) [9]. A nearby intronic SNP in HECTD4 (rs11066188) also reached array-wide significance (P = 1.3 x 10⁻⁷) and was in LD with the aforementioned ATXN2-SH2B3 variants (r² = 0.74–0.84 in 1000G AMR).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Main findings^a for meta-analysis of diastolic and systolic BP in PAGE Hispanic samples (n~19,706).

https://doi.org/10.1371/journal.pone.0164132.t002

The rs2586886/KCNK3 SNP replicated in independent Hispanic samples from MESA for diastolic BP (P = 0.02, β = 0.78) (Table 3). This SNP is also significantly associated with systolic and diastolic BP in the European ancestry GWAS meta-analysis from the ICBP, and with systolic BP in African Americans, but was borderline significant for diastolic BP in African Americans (Table 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Validation of the KCNK3 rs2586886 associations with BP in independent multi-ethnic populations.

https://doi.org/10.1371/journal.pone.0164132.t003

In sensitivity analyses excluding Mt. Sinai BioMe, rs2586886/KCNK3 remained significantly associated with diastolic BP, P = 3.1 x 10⁻⁹. P-values for other diastolic and systolic SNPs increased to values above the significance threshold, P = 2.8 x 10⁻⁷. However, none of the SNPs showed evidence of heterogeneity by study in the meta-analysis of all Hispanics including Mt. Sinai, all heterogeneity P>0.4 and I²~0. Using PAGE Hispanic results, we performed a look-up of previously identified GWAS SNPs in 22 loci (Table C in S1 File). GWAS SNPs from 15 loci replicated in Hispanics at p<0.05.

Replication, fine mapping and discovery in African Americans

We first attempted to replicate the 22 BP loci previously identified in GWAS of BP in European ancestries using African American samples. These meta-analyses identified SNPs in FGF5, ATP2B1, TBX3 and LOC339593 nominally associated with systolic BP (Table D in S1 File), and SNPs in FGF5, SLC39A8, ARHGAP42, ADM, CPLX3, LOC339593 and ZNF831 nominally associated with diastolic BP, all P <0.05 (Table E in S1 File). To fully capture the European-ancestry GWAS SNP ‘signal’, we then extended the association analyses in African Americans to include all SNPs in modest LD (r²≥0.5 in 1000G EUR) with each reported GWAS SNP. These analyses identified additional associations in African Americans at MTHFR, EBF1, C10orf107, CYP17A1, NT5C2, CSK, and PLCD3 for either diastolic or systolic BP with P-values ranging from 6.9 x 10⁻⁵ to 0.04, (Tables D-E in S1 File).

Rs56153133/CLCN6 was significantly associated with diastolic BP (P = 6.9 x 10⁻⁵) at the locus-level significance threshold (P<1.4 x 10⁻⁴) in African Americans (Figure C in S1 File). This SNP is in high LD (r² = 0.98 in 1000G EUR) with the previously reported GWAS SNP rs17367504 in the MTHFR locus. However, rs17367504/MTHFR was not associated with diastolic BP in African Americans (P = 0.24), in spite of its relatively high correlation with rs56153133 (r² = 0.65 in PAGE African Americans). In a model conditioning on rs17367504, rs56153133 remained significantly associated with diastolic BP, P = 1.3x10^-6, providing evidence to support its independence from rs17367504. However, the rs56153133-diastolic BP association did not replicate in the independent, though smaller (n = 2550) MESA African American sample, P = 0.88.

The strongest associations in the systolic BP fine mapping were located in the 15q26.1 locus (FURIN), but none of the SNPs reached the fine-mapping significance threshold for the locus (P<5 x 10⁻⁴). The most significant (P = 8.0 x 10⁻⁴) SNP in this locus in African Americans is rs6224, an intronic variant in FURIN. Rs6224/FURIN, is in low LD (r² = 0.08) in African Americans with the prior GWAS SNP in this region, rs2521501/FES, and in modest LD (r² = 0.46) with it in the 1000G EUR (Figure D in S1 File).

In a full Metabochip scan testing for new SNP associations with BP traits in African Americans, no SNPs reached the discovery (multiple testing-adjusted) significance threshold (P<2.8 x 10⁻⁷).

Trans-ethnic meta-analyses

To increase power to identify associations in the discovery scan, we also performed trans-ethnic meta-analyses by combining results from Hispanics and African Americans. We used two approaches for the meta-analyses: a fixed-effects model, and a random-effects model optimized to detect associations under heterogeneity, as implemented in MetaSoft [27]. While no novel loci were identified, there were eight array-wide significant (P<2.8 x 10⁻⁷) SNPs in ULK4, and two SNPs in 5`region of FGF5 associated with diastolic BP (Table F in S1 File).

Several SNPs were also significantly associated with systolic BP including rs2023843, a novel intronic SNP in HOTTIP (in the HOXA-EVX1 locus) (Figure E in S1 File). This SNP replicated in MESA Hispanics, P = 0.038 and was borderline significant in MESA African Americans, P = 0.064. Rs2023843 is in modest LD (r² = 0.27 in 1000G AFR) with rs11564022/ HOXA-EVX1 previously reported in African Americans (and not available on the Metabochip), though LD with it in 1000G AMR Hispanics is very low (r² = 0.03). SNPs upstream of FGF5 were also significantly associated with systolic BP in the trans-ethnic analysis, and most appear to be mainly driven by the Hispanic American results. The most significant trans-ethnic SNP at the FGF5 locus, rs13125101 is in high LD (r²>0.8 in CEU) with the prior GWAS SNPs (rs16998073 and rs11099098) suggesting that it may reflect the same signal (Figure F in S1 File). While this locus is not novel, we identified a significant FGF5 intronic SNP not previously associated with BP, rs36034102. However this SNP was not associated with systolic BP (P = 0.25) in the smaller sample of MESA Hispanics (n = 2106) although the direction and magnitude of the beta coefficient were similar, β = 0.89 in MESA, β = 1.05 in PAGE (data not shown).

All SNPs significant in the fixed-effects trans-ethnic meta-analyses were also significant when using the Han and Eskin (HE) random-effects meta-analysis models with the exception of one systolic BP SNP (rs72656599/5’ of FGF5), which was borderline significant in the random-effects model. As expected, P-values for the two models were generally comparable, with fixed effects models having lower P-values, except in the presence of high heterogeneity (I²≥ 80) between the race/ethnicity groups. In cases of high heterogeneity, the HE random-effects P-values were lower than those from fixed effects models.