Association Analyses and Meta-analyses

Nearly 90 percent of the markers on the Exome Chip are low-frequency (MAF 0.01–0.05) or rare (MAF <0.01) variants. Power for association was evaluated for MAP assuming a mean of 100 mm Hg with standard deviation of 10 mm Hg using QUANTO⁷⁷ for a sample size n=150,000 at the significance level of 3.4 × 10⁻⁷ for a variant with MAF of 0.0005, 0.001, 0.005, or 0.01. To reach 80% power, an effect size of 5, 3.5, 1.6, or 1.1 mm Hg, is needed, respectively, for a variant with MAF=0.0005, 0.001, 0.005, or 0.01.

We downloaded the phase 3 genotype data for the European ancestry from HapMap project. The phase 3 file “hapmap3_r2_b36_fwd.CEU.qc.poly” includes 1,416,121 variants (1,352,770 with MAF>0.01 and 1,223,919 with MAF> 0.05). We used the PLINK command “show-tags” to estimate the number of common variants (MAF>0.05) that can be tagged by Exome Chip variants. We estimated that 172,220 (linkage disequilibrium r²≥0.5) and 88,186 (linkage disequilibrium r²≥0.8) common SNPs (MAF >0.05) can be tagged by the Exome Chip variants. Compared to the number of variants tagged by a GWAS chip (e.g. Affymetrix 500K), the Exome Chip tags much fewer common variants.

Gene-based (or region-based) testing was performed using the seqMeta package⁷⁸. Covariates included age, age-squared, sex, body mass index (BMI), and principle components (if applicable) to account for population structure. All variants were recoded to conform to the alleles specified in a “Recode” file distributed to each study. In all analyses, variant effects were modeled additively. Conditional analysis was performed to identify independent BP signals at previously reported BP loci^5–15 using the seqMeta package⁷⁸ by adjusting at the cohort level for the previously reported GWAS SNP with the smallest p-value in association analysis. Similarly, for any newly identified locus with multiple variants, conditional analysis was performed by adjusting for the most significant variant in the region to identify non-redundant signals.

Meta-analysis of single variant associations from discovery and follow-up stage results was performed using the inverse variance weighted fixed-effects method⁷⁹ implemented in the seqMeta package⁷⁸. In the discovery stage, the primary meta-analysis was performed in all samples to identify variants showing consistent effects with BP traits across multiple ancestry groups. Secondary analysis was performed in each of the three ancestries separately to identify novel variants with different ancestral origin. Meta-analysis was also performed on results from conditional analysis and compared with the original meta-analysis to identify non-redundant signals. Although we performed association and meta-analysis on all genotyped variants that passed quality control, we only reported results from about 147,000 variants that had minor allele counts (MACs) ≥30 in meta-analyses of all samples. Since the BP traits are highly correlated, we used an array-wide Bonferroni-corrected significance threshold of 3.4 ×10⁻⁷ (=0.05/147,000). The Exome Chip array contains numerous previously published variants or their LD proxies, mostly from GWAS using imputed genotype information for a variety of human traits. Using exome chip experimental genotypes, associations from previous BP GWAS^5–15 were considered significant with P values ≤ 0.05/n, where n is the number of previously identified SNPs or SNPs that showed at least moderate LD (r²≥0.3) on the Exome Chip.

Meta-analysis was also conducted at the gene level to evaluate aggregate effects from multiple non-synonymous and splicing variants with MAFs ≤0.01 (T1) and ≤ 0.05 (T5) in a gene using both the sequence kernel association test (SKAT)⁴³ and the standard burden test^41,42 implemented in the seqMeta package⁷⁸. The standard burden test collapses the rare variants and has optimal properties when these variants all have the same directionality and magnitude of effect on phenotype. In contrast, SKAT aggregates individual variant score test statistics and offers better power compared to the burden test when there are a variety of effect sizes and directions, e.g. both protective and deleterious effects in a gene⁴³. Approximately 17,000 genes were included two or more non-synonymous variants in the primary meta-analysis of all study samples. An association was deemed to be signficant at P<1 ×10⁻⁶ for gene-based tests. Among up to 154,543 individuals of European ancestry from CHD Exome+ Consortium, ExomeBP Consortium, GoT2DGenes Consortium, T2D–GENES consortium (Supplementary Note), gene-based SKAT was applied to HTN and inverse normal transformed DBP, SBP, PP using the RAREMETAL software package⁸⁰. We performed lookup in their SKAT results for the genes that displayed P<1 ×10⁻⁶ in Stage 1 analysis of this study.