As the first step of the present study, systolic (ASBP) and diastolic (ADBP) 24‐hour ambulatory blood pressure responses to each of the 4 monotherapies were analyzed separately in the GENRES Study (discovery sample). The BP response was calculated as BP after 4 weeks' drug treatment minus mean BP after placebo periods. The mean BP level of all (up to 4) placebo periods, as opposed to the 1 preceding placebo period, was used as the baseline level to reduce BP variation. The approach was supported by several analyses. First, the BP responses to study drugs showed clearly lower variation when mean of all placebo periods was used as the baseline level (Table S1). Second, compared with the other drugs, amlodipine seemed to have a small carry‐over effect (≈−1.5/−0.5 mm Hg) based on placebo BP levels 4 weeks after amlodipine treatment (Table S2). The randomized cross‐over design and the use of all placebo periods as the baseline level eliminates any systematic effect of this finding on the results. In addition, the effect was probably even smaller after an additional 4 weeks when BP response to the next study drug was assessed. Third, the higher variation of placebo BP levels when only 1 placebo period was used can be seen in Table S2 as higher SDs. Finally, the preceding study treatment and the order of the drug treatment periods had no effect on BP response to any of the study drug when they were tested with regression analysis (GLM Univariate procedure of IBM SPSS Statistics program, version 19).
For the genome‐wide analyses, ASBP and ADBP response residuals were generated using IBM SPSS Statistics program and stepwise linear regression. The following covariates were tested with P<0.10 as an inclusion condition: corresponding placebo ABP (mean of all periods), age, earlier use of antihypertensive medication (coded as 0/1), current smoking (coded as 0/1), body mass index, daily urinary sodium excretion after the first placebo period, and serum creatinine level after the first placebo period. For non‐normally distributed covariates, normalized values were used. The covariates selected for calculation of BP response residuals (in mm Hg) for each study drug are listed in Table S3.
The genome‐wide association analysis was done using covariate‐adjusted BP responses and linear regression under an additive genetic model with program PLINK.20 (link) For each of the 4 study drugs, we report here the 20 genetic loci with the lowest P values based on either ASBP or ADBP responses. P values <5×10−8 were considered as significant at the genome‐wide level.
In the second step, replication analyses of the 20 best loci associated with losartan, bisoprolol, and hydrochlorothiazide responses were carried out using the data from PEAR, GERA I, GERA II, and SOPHIA studies. In each case, the analyses were confined to study participants of European ancestry only, and age, gender, baseline BP, and principal components to account for ancestry were used as covariates. Accordingly, we replicated losartan/GENRES data using losartan/SOPHIA and candesartan/GERA II data, bisoprolol/GENRES data using atenolol/PEAR data, and hydrochlorothiazide/GENRES data using both hydrochlorothiazide/PEAR and hydrochlorothiazide/GERA I data. Successful replication was defined as P values below the Bonferroni‐corrected level of 2.5×10−4 (number of individual tests: 99 SNPs in 5 replication study analyses×SBP+DBP responses=198) and the same direction of effect. Suggestive replication was defined as P values <0.05 and the same direction of effect.
As the third step, meta‐analyses of the top 20 GENRES 24‐hour ambulatory blood pressure response SNPs in all available studies were performed using inverse‐variance model with fixed effects in METAL.21 (link) We defined significant results as P values <5×10−8. In addition, P values <1×10−5 were considered to represent a suggestive association.
For the genome‐wide analyses, ASBP and ADBP response residuals were generated using IBM SPSS Statistics program and stepwise linear regression. The following covariates were tested with P<0.10 as an inclusion condition: corresponding placebo ABP (mean of all periods), age, earlier use of antihypertensive medication (coded as 0/1), current smoking (coded as 0/1), body mass index, daily urinary sodium excretion after the first placebo period, and serum creatinine level after the first placebo period. For non‐normally distributed covariates, normalized values were used. The covariates selected for calculation of BP response residuals (in mm Hg) for each study drug are listed in Table S3.
The genome‐wide association analysis was done using covariate‐adjusted BP responses and linear regression under an additive genetic model with program PLINK.20 (link) For each of the 4 study drugs, we report here the 20 genetic loci with the lowest P values based on either ASBP or ADBP responses. P values <5×10−8 were considered as significant at the genome‐wide level.
In the second step, replication analyses of the 20 best loci associated with losartan, bisoprolol, and hydrochlorothiazide responses were carried out using the data from PEAR, GERA I, GERA II, and SOPHIA studies. In each case, the analyses were confined to study participants of European ancestry only, and age, gender, baseline BP, and principal components to account for ancestry were used as covariates. Accordingly, we replicated losartan/GENRES data using losartan/SOPHIA and candesartan/GERA II data, bisoprolol/GENRES data using atenolol/PEAR data, and hydrochlorothiazide/GENRES data using both hydrochlorothiazide/PEAR and hydrochlorothiazide/GERA I data. Successful replication was defined as P values below the Bonferroni‐corrected level of 2.5×10−4 (number of individual tests: 99 SNPs in 5 replication study analyses×SBP+DBP responses=198) and the same direction of effect. Suggestive replication was defined as P values <0.05 and the same direction of effect.
As the third step, meta‐analyses of the top 20 GENRES 24‐hour ambulatory blood pressure response SNPs in all available studies were performed using inverse‐variance model with fixed effects in METAL.21 (link) We defined significant results as P values <5×10−8. In addition, P values <1×10−5 were considered to represent a suggestive association.
