Genome-wide association study of nocturnal blood pressure dipping in hypertensive patients

Rationale and results of the GENRES study (clinicaltrials.​gov NCT03276598, registered retrospectively) have been previously published elsewhere [13‐15]. In brief, GENRES is a double-blind, randomized, placebo-controlled cross-over study that was designed to search for genetic variation predicting response to four classes of antihypertensive drugs, conducted at Helsinki University Hospital between years 1999 and 2004 (Additional file 1: Figure S1). Initially, a total of 313 Finnish men (aged 35 to 60 years) with moderate hypertension (DBP ≥ 95 mmHg or previous use of antihypertensive medication) were screened for the study. Secondary hypertension, drug-treated diabetes mellitus, congestive heart failure and CAD were among the exclusion criteria. The study participants underwent 4 different one-month single-drug treatment periods (losartan 50 mg, amlodipine 5 mg, hydrochlorothiazide 25 mg, and bisoprolol 5 mg daily) separated by one-month placebo periods. Twenty-four-hour ABPM was carried out at the end of each drug and placebo period using a device equipped with a position sensor (Diasys Integra; Novacor, Rueil-Malmaison, France). Each subject gave written informed consent and the study was approved by the Ethics Committee of Helsinki University Central Hospital and National Agency of Medicines of Finland.

For the discovery GWAS, we used the ABPMs recorded at the end of each placebo period. BP readings were taken every 15 min when standing and every 30 min when supine. Intense physical activity was not permitted. Daytime was defined as hours between 7 am and 10 pm. Single observations were excluded from the analysis due to low pulse pressure (< 15 mmHg if SBP < 120 mmHg and < 20 mmHg if SBP > 120 mmHg), high heart rates (≥ 110 bpm), lying down during daytime, standing up or being awake at nighttime or high physical activity. For a recording to be accepted, > 15 daytime and > 7 nighttime measurements were required. The mean SBP and DBP dipping of all placebo periods were used as the study variables. A total of 235 unrelated subjects were successfully genotyped and had ABPM data from at least one placebo period. To reduce the effect of day-to-day variation in the dipping phenotype, only the 204 subjects that had ≥3 accepted placebo ABPMs available (173 subjects with four and 31 subjects with three ABPMs) were included in the discovery GWAS.

DYNAMIC (clinicaltrials.​gov NCT01742702) is an on-going study to investigate hemodynamic changes in primary and secondary hypertension with non-invasive hemodynamic measurement [16, 17]. A total 188 subjects with ABPM dipping available were genotyped. One subject was excluded due to low success rates of genotyping, one subject was excluded due to morbid obesity (body mass index (BMI) > 40) and 3 subjects were excluded due to first degree relativeness; ultimately, 183 subjects were included in this replication cohort. BP was measured with Microlife WatchBP O3 monitor (Microlife AG, Widnau, Switzerland) every 20 min during the day (7 am to 10 pm) and every 30 min during the night (10 pm to 7 am). Single observations were excluded using the same criteria as described for GENRES. Each subject gave written informed consent and the study was approved by the Ethics Committee of Tampere University Hospital.

DILGOM was a population survey conducted in 2007 to assess the effects of environment and genetics on obesity and metabolic syndrome. Four hundred and ninety-four subjects of DILGOM participated in a cardiovascular substudy. In 2014, 290 still living participants of the cardiovascular substudy underwent re-examination including a 24-h ABPM to compare novel and traditional BP measurement methods [18]. Both BP dipping and genetic data were available for 207 unrelated subjects. Four subjects that were morbidly obese (BMI > 40) and 23 subjects that were over the age of 75 were excluded, resulting in 180 subjects to be included in the replication cohort. BP was measured with a Microlife WatchBP O3 monitor (Microlife AG, Widnau, Switzerland) every 20 min during the day (7 am to 10 pm) and every 30 min during the night (10 pm to 7 am). Night-time BP was defined as the mean of all BP values of the actual sleeping period and daytime BP as the mean of all other BP values. Each subject gave written informed consent and the study was approved by the Ethics Committee of the Hospital District of Southwest Finland.

The genotyping methods and quality control steps (including exclusions for identity-by-state clustering, gender check, Hardy-Weinberg equilibrium P value < 1 × 10^− 5 and minor allele frequency < 0.01) for GENRES have been described in detail before [13]. The DNA samples of the GENRES study subjects were genotyped at the Institute for Molecular Medicine Finland (Helsinki, Finland) using the Illumina HumanOmniExpress BeadChip (Illumina, San Diego, CA, USA). After quality control steps a total of 631,844 autosomal SNPs were available for the analysis. Subsequent imputation was performed using IMPUTE2 [19] and the Hapmap2 CEU release 22 reference panel after pre-phasing with ShapeITv2. However, only the genotyped SNPs were used for the GWAS analysis.

The DNA samples of the DYNAMIC study subjects were genotyped at the Institute for Molecular Medicine Finland (Helsinki, Finland) using the Illumina HumanOmniExpress BeadChip (Illumina, San Diego, CA, USA). One subject was excluded due to low success rate (< 95%). The six SNPs used for replication passed the same quality control steps as described for GENRES above.

The genotype data of the DILGOM study subjects was derived from previously conducted GWASs. The DILGOM DNA samples were genotyped with Illumina Cardio-MetaboChip and Illumina 610 K arrays (Illumina, San Diego, CA, USA) at the Wellcome Trust Sanger Institute (Cambridge, UK) and the Institute for Molecular Medicine Finland (Helsinki, Finland). Unobserved SNPs were imputed using 1000 Genomes haplotypes Phase I integrated variant set release (ShapeITv2) in NCBI build 37 version June 2014. If the selected SNP was not genotyped, then the imputed value was used. Info scores for the imputed SNPs used were 0.989 for rs16984571 and 0.998 for rs1230361.

Left ventricular hypertrophy (LVH) was assessed in GENRES from the transthoracic echocardiograms recorded during the first placebo period and from electrocardiograms (ECGs) recorded at the end of each placebo period [20]. Echocardiographically determined left ventricular mass (LVM, in grams) was calculated as 0.8 × [1.04 × ((interventricular septal thickness + left ventricular end-diastolic diameter + posterior wall thickness)³ – left ventricular end diastolic diameter³)] + 0.6 and left ventricle mass index (LVMI) by dividing LVM by body surface area (square meters). LVMI was available for 227 GENRES subjects that were successfully genotyped for rs4905794. The standard resting 12-lead ECG was recorded with Marquette MAC 5000 electrocardiograph (GE Marquette Medical Systems, Milwaukee, Wisconsin, USA) at the end of each placebo period and all measurements were performed blinded to all other data. Sokolow-Lyon voltage (SV1 + RV5 or SV1 + RV6, whichever was greater) and Cornell voltage-duration product [(SV3 + RaVL) × QRS duration] were used as markers of LVH. The indices were calculated from digital ECGs from automatically made ECG measurements with a visual confirmation for all subjects who had completed the whole study. Mean values of all placebo periods were used. A total of 185 subjects, that were successfully genotyped for rs4905794, had these ECG measurements available.

Baseline demographics and BP dipping data of all study subjects were analyzed using IBM SPSS Statistics 22.0 (IBM Corp., Armonk, NY, USA). SBP and DBP dipping were analyzed separately as continuous variables, and were calculated as night-to-day BP ratio [3] and expressed as percentage [(mean nighttime BP / mean daytime BP) × 100]. Both variables were adjusted for corresponding daytime BP using linear regression. Lifestyle factors (age, BMI, waist-hip ratio, antihypertensive use before entry to the study, duration of hypertension, smoking, serum creatinine level and daily urinary excretion of sodium after the first placebo period were tested for association with stepwise linear regression and included as covariates if P value for association was < 0.10 for either systolic or diastolic night-to-day BP ratio. The final regression models are presented in Additional file 2: Table S1. These regression variables were approximately normally distributed. The effect of population stratification on night-to-day BP ratio was tested with principal components (PCs) generated with program Smartpca from Eigensoft package [21]. A reduced dataset of 114,487 SNPs with r² < 0.5 and the long-range linkage disequilibrium (LD) regions excluded [22] was used. Three significant PCs for population stratification with P values < 0.10 were identified with program Twstats from Eigensoft package, but they were not associated with either systolic or diastolic night-to-day BP ratio residuals (all P values > 0.30) and were thus not included in GWAS analysis. The discovery GWAS was done using covariate-adjusted night-to-day BP ratio residuals and linear regression under additive genetic model in PLINK v1.07 [23]. P values < 5 × 10^− 8 were considered as genome-wide-level significant, while P values < 1 × 10^− 5 were considered as suggestively significant.

The statistically most significant SNPs from each locus with P value < 1 × 10^− 5 were selected for replication in two independent Finnish cohorts. For the replication cohorts similar statistical approach was used. Bonferroni corrected P value < 0.004 was regarded as a successful replication [0.05 / (2 studies × 6 SNPs)], considering that systolic and diastolic dipping are highly correlated. P value < 0.05 with the same direction of effect was considered as suggestive replication. We also conducted meta-analyses of the top BP dipping (night-to-day BP ratio) SNPs in all available studies using inverse-variance model with fixed effects in METAL [24]. We defined significant results as P values < 5 × 10^− 8. P values < 1 × 10^− 5 were considered to represent a suggestive association.

In addition, we tested association of the genome-wide-significant SNP (rs4905794) with LVMI, Sokolow-Lyon voltage and Cornell voltage-duration product measured in GENRES subjects to further explore the nature of the SNP in relation to target organ damage. We tested for significant covariates and adjusted LVMI with age (standardized β = 3.3, P = 0.007) and Sokolow-Lyon voltage with BMI (standardized β = − 0.6, P = 0.0003) using stepwise linear regression (other covariates tested were smoking, waist-hip ratio and previous use of antihypertensive medication). BP levels were not included as covariates (although highly significant) as we presumed the effect of the genotypes to be principally mediated through BP levels. No covariates were significant for Cornell voltage-duration product.

A SNP set-based approach was used to test for association of circadian genes with BP dipping (night-to-day BP ratio). The 24 genes listed in the circadian rhythm pathway in KEGG (Kyoto Encyclopedia of Genes and Genomes) PATHWAY database [25, 26] were selected for testing. All successfully genotyped SNPs located inside the transcript boundaries of each circadian gene were included. Transcript boundaries of the circadian genes were derived from Ensembl Genome Browser (release 87, in NCBI37/hg19) [27, 28]. The set-based tests were conducted in PLINK v1.07 [23]. Using the SBP and DBP dipping (night-to-day BP ratio) residuals calculated as described above, a mean statistic was calculated for each gene based on the single-SNP association results under a linear model. Single SNPs were included in the set if association P value was < 0.05 and r² was < 0.5 with other SNPs selected with maximum of five SNPs included for each gene. The dataset was then permuted 10,000 times calculating empirical P value for each gene describing the number of times the permuted set-statistic exceeded the original one for the gene. Bonferroni corrected P value of 0.002 (0.05/24) was used to correct for multiple testing (24 genes).

Finally, we sought to replicate the previously published genetic associations, based on candidate gene studies, with BP dipping. This included five SNPs from Leu et al. [10], two SNPs from Sheng et al. [11] (one of which is shared with Leu et al.) and five SNPs from Wirtwein et al. [12]. Two of these SNPs were not genotyped in GENRES and thus imputed values were used. We considered Bonferroni corrected P value < 0.0045 (0.05/11 SNPs tested) and the same direction of effect as successful replication and P value < 0.05 and the same direction of effect as suggestive replication.

We assessed the consequence of the leading SNPs of the loci that achieved P value < 1 × 10^− 5, using VEP (variant effect predictor) database [29, 30]. To look for putative functional consequences, we annotated each SNP to the closest gene using Ensembl database (release 87) [27, 28] and conducted an expression quantitative trait locus (eQTL) analysis in publicly available Genotype Tissue Expression (GTEx) database [31, 32]. We then used PhenoScanner [33, 34] to find if any of the SNPs had previously reported cardiovascular trait associations at a nominal significance of P < 0.05.

The Neale lab has made available basic GWAS results for over 2000 phenotypes from the UK Biobank data of ~ 337,000 unrelated individuals of British ancestry [35]. We looked up our leading SNPs for BP dipping through MR-Base PheWAS database [36] to see if they were associated with SBP or DBP, or with seven sleep traits (sleep duration, getting up in morning, chronotype, nap during day, sleeplessness, snoring, daytime dozing) in the Neale lab data, at a nominal significance of P < 0.05.

The baseline characteristics of GENRES subjects (the discovery sample) are described in Table 1. Ambulatory BP data used for calculation of the mean BP dipping during placebo administration was derived from four (n = 173) or three (n = 31) recordings. The Manhattan plots of the GWAS results for BP dipping from the discovery cohort are shown in Fig. 1 and the quantile-quantile plots are depicted in Additional file 3: Figure S2. As detailed in Table 2, one locus (rs4905794 at chromosome 14q32.2 near BCL11B) reached genome-wide significance threshold for SBP dipping (systolic night-to-day BP ratio β = − 4.8%, P = 9.6 × 10^− 9). The association of rs4905794 was very similar for DBP dipping (diastolic night-to-day BP ratio β = − 4.3%, P = 2.2 × 10^− 6). Further five potential loci were identified with suggestive significance (P < 1 × 10^− 5) (Table 2). Local Manhattan plots of the best loci are presented in Additional file 4: Figure S3A-F. A total of 99 SNPs in 62 loci had P values < 1 × 10^− 4 for either SBP or DBP dipping.

The GENRES subjects received four different antihypertensive monotherapies in a rotational fashion separated by one-month placebo periods (Additional file 1: Figure S1). We also tested if the associations of the statistically most significant SNPs from the six loci with P < 1 × 10^− 5 derived from the placebo periods were similar when dipping data during drug administration was analyzed. The results are shown in Additional file 5: Figure S4A-F. The effects of the SNPs were consistently in the same direction as during placebo periods and the associations remained statistically significant.

In the replication step, the lead SNPs (most significantly associated SNP) from the genome-wide significant locus and the five loci with P values < 1 × 10^− 5 were followed up in two independent Finnish cohorts, DYNAMIC and DILGOM (Table 1). The majority of the DYNAMIC subjects had elevated blood pressure while DILGOM comprised mostly of normotensive individuals (Table 1). The results of the replication analysis are summarized in Table 2. In DYNAMIC, rs4905794 showing genome-wide significance in GENRES had an effect in the same direction, but the association was not statistically significant (β = − 0.8%, P = 0.4 for systolic night-to-day BP ratio and β = − 1.6%, P = 0.13 for diastolic night-to-day BP ratio). The association did not, however, replicate in the population-based DILGOM cohort. Nocturnal BP dipping (night-to-day BP ratio) values according to rs4905794 genotypes in the three study samples are illustrated in Fig. 2. Another SNP, rs16984571 located in KCNS3 intron, showed effect in the same direction as in GENRES in both replication cohorts, but the associations did not reach statistical significance. In a meta-analysis of all three studies (combined n = 567) employing inverse-variance model with fixed effects no SNP showed genome-wide significant P values (Additional file 2: Table S2).

To explore the possible association of the genome-wide-significant SNP (rs4905794) with markers of cardiovascular target organ damage, we tested for the association of this SNP with indices of LVH in GENRES. Using all GENRES subjects with LVMI and rs4905794 genotype data available (n = 227, mean LVMI 98.1 ± 18.5 g/m²), rs4905794 was found to significantly associate with LVMI, with the allele associated with larger dipping also associating with smaller left ventricular mass (G allele β = − 7.6 g/m², P = 0.02) (Fig. 3a). In the 185 GENRES subjects with ECG recordings and rs4905794 genotype available, the mean Sokolow-Lyon voltage was 25.6 ± 6.7 mm and mean Cornell voltage-duration product was 1444 ± 528 mm × ms. The rs4905794 genotype was associated with Sokolow-Lyon voltage (G allele β = − 2.8 mm, P = 0.03) (Fig. 3b). For Cornell voltage-duration product this association had a tendency to the same direction (G allele β = − 112 mm × ms, P = 0.3) (Fig. 3c). If BP dipping was included as a covariate in the analyses, the P values for the associations remained the same.

In GENRES GWAS, no SNPs mapping within the transcript boundaries of the selected circadian genes had P values < 1 × 10^− 4 (Additional file 2: Table S3). Using set-based tests to evaluate the combined effect of independent SNPs in each gene, the retinoic acid-related orphan receptor genes (RORA, RORB, RORC) had the smallest P values (Additional file 2: Table S4). RORB had a suggestively significant P value 0.03 for association with SBP dipping, but this association did not survive multiple testing. Replication analysis for circadian gene SNPs from two earlier candidate gene studies is summarized in Additional file 2: Table S5. These SNPs did not significantly replicate in our study.

In our study, rs9818870 was associated with both systolic and diastolic night-to-day BP ratio (β = 1.4%, P = 0.03 and β = 2.3%, P = 0.002, respectively) in consistent direction (Additional file 2: Table S5), thus further supporting the association of this SNP with BP dipping. The associations of the other four CAD SNPs found by Wirtwein et al. [12] did not significantly replicate in our study.