Biological Pathways of Long-Term Visit-to-Visit Blood Pressure Variability in the American Population: Cardiovascular Health Study and Women’s Health Initiatives

Department of Epidemiology, College of Public Health, University of Arkansas for Medical Sciences, Little Rock, AR, USA Department of Biomedical Informatics, College of Medicine, University of Arkansas for Medical Sciences, Little Rock, AR, USA Department of Human Genetics-Graduate School of Public Health, University of Pittsburgh, Pittsburgh, PA, USA Department of Epidemiology and Biostatistics, School of Community and Rural Health, University of Texas Health Science Center, UT Health Northeast, Houston, TX, USA


Introduction
Epidemiological studies show that blood pressure varies; it oscillates over days (short term), as well as months and years (long term) [1]. e variation reflects the degree of blood pressure control and its long-term burden on the cardiovascular system [2]. Visit-to-visit blood pressure variability (VVBPV) is associated with organ damage [2]. Several observational studies documented a relationship between blood pressure variability and cardiovascular morbidity and mortality independently of the absolute mean arterial blood pressure across clinical visits.
In the Women's Health Initiative (WHI) study, the researchers compared the highest quintile of VVBPV versus the lowest quintile. e hazard ratio (HR) for stroke and its 95% confidence interval (CI) were 1.46 (1. 15-1.85). In the Cardiovascular Health Study (CHS), a one-unit increase in the VVBPV increased the risk for myocardial infarction by 11%: HR and its 95% CI were 1.1 (1.0-1.3).
In previous studies, investigators located many genetic variants that are associated with hypertension [3,4]. Epidemiological studies have shown that hypertension and blood pressure have been considered polygenic genetic traits, which are regulated by many biological pathways. However, the literature has little information on the biological mechanisms that may regulate long-term blood pressure variability. Additional work is warranted to identify these biological mechanisms. In this study, we analyzed genetic data from the WHI and CHS studies to locate SNPs associated with VVBPV and discern the involved biological pathways to provide insight into the architecture of this phenotype.

Methods
e Institutional Review Board at the University of Arkansas for Medical Sciences approved this study. We used the genetic data of two cohorts, WHI and CHS. e GWAS of the CHS included 1264 hypertensive patients who had blood pressure. e CHS GWAS study had two subcohorts: the White and African American cohorts. For the GWAS of WHI, the total number of the hypertensive participants was 8889.
e WHI cohort has three subcohorts: Garnet (Whites), IMS (Whites), and Share (African Americans). e blood pressure readings of the participants of these two CHS and WHI cohorts were collected over 11 years. Systolic and diastolic blood pressure (SBP and DBP) were measured at each annual study visit by certified staff and according to standardized procedures and instruments. Blood pressure was measured with a mercury sphygmomanometer after the participant was seated and had rested for five minutes. We included participants who had at least four visit-to-visit blood pressure readings. We used the average of at least two within-visit readings to determine the SBP and DBP of each visit. To account for the effect of the antihypertensive medication, we added 10 mmHg and 5 mmHg to the systolic and diastolic blood pressure, respectively, as it was done in previous studies [5,6]. We defined the two VVBPV of SBP and DBP phenotypes as the standard deviation about the participant's regression line (SDreg); a regression line was fitted for SBP and for DBP, separately, across visits. e estimation of SDreg assumes a linear temporal trend.
For quality control, we excluded SNPs with genotype missingness >5%, samples with low call rates (<95%), and SNPs with genotype frequencies that deviated from Hardy-Weinberg equilibrium (HWE), i.e., P < 10E − 5. Because our study was population-based, we only included unrelated individuals. To eliminate the possibility of spurious associations due to population structure, we computed principal components to detect population stratification and used ten components in the statistical analysis to adjust for it. To further assure the quality of samples, we excluded monomorphic SNPs: samples with very low (<4 SDs) heterozygosity would indicate poor DNA quality, and samples with very high (>4 SDs) heterozygosity would suggest sample contamination. We used the LiftOver algorithm to map our data to genetic assembly GRCH37 [7], and SHAPEIT software package [8] was used to phase our genotype data and detect flipped SNPs. Plink [9] was used to map the forward DNA strand. e Michigan Imputation Server was used to impute untyped and missing. For SNP imputation, the 1000 Genome Project was used as the reference panel for imputation [10]. Following a common variant, common disease hypothesis (CV-CD), we filtered out SNPs that had a minor allele frequency (MAF) < 0.05 (rare and low-frequency SNPs). After imputation, we excluded SNPs that had r 2 < 0.30 or MAF < 0.05.

Data Analysis.
We conducted GWAS analysis, metaanalysis, gene-based association analysis, and pathway analysis. For GWAS analysis, we converted the genetic data files of the study cohorts into file formats that can be analyzed by the MaCH2qtl software package. We used the MaCH2qtl to fit linear regression models and account for the uncertainty of the imputed data. We analyzed the association of the log (SDReg) of SBP and DBP (separately) with each SNP after adjusting for age, sex (only for CHS), mean arterial blood pressure across visits, number of clinical visits, average time interval between visits, smoking, dyslipidemia, and ten principal components. We assumed an additive effect of allele dosage on blood pressure readings. We set level of significance at 5.0E − 8 to minimize false-positive results.
For the transethnic meta-analysis, we used random-effect-Han's and Eskin's model (RE-HE) to pool the results of the five cohorts.
e RE-HE model has more statistical power than the traditional random-effect or fixed-effect methods to detect a significant association in the presence of interstudy heterogeneity due to differences in the ethnic background of pooled populations [11]. To increase statistical power, we conducted Gene-Based Association Analysis using GCTA64. To compute the gene P values, we used the 1000 Genomes Project panel as a reference population to account for linkage disequilibrium (LD) between SNPs. Figure 1 summarizes the pipeline, which we used to conduct the pathway analysis. We first selected the significant genes based on P values ranging from 10 − 7 to 10 − 6 and analyzed these genes for biological plausibility using DAVID Bioinformatics Resources 6.8 [12]. e detailed process for both the DBP and SBP variability is shown in Figure 1. e pathways were considered significant if the Fisher exact P values were equal to or smaller than 0.05 and the minimum gene count belonging to a pathway was at least two. Table 1 shows the characteristics and number of the WHI and CHS participants. We conducted 5 GWAS analyses (see the supplementary materials: GWAS Analysis, Supplementary Table 1, and Supplementary Figure 1).

Gene-Based Analysis and Functional Annotation.
We used summary-level data from the meta-analysis results of the genome-wide association studies and linkage disequilibrium data from a reference American population to do gene-based association analysis. We identified gene loci that could be associated with VVBPV: (1)

SBP Variability.
For SBP variability, 36 of the top 100 genes were significant ( Table 2). Further genetic analyses located the significant genes that play role in cardiovascular disease pathways (Table 3). ese pathways included the GABAergic neurologic pathway, which regulates the sympathetic activity and vascular tone. In the GABAergic pathway ( Figure 2), three significant genes were identified ADCY4, PRKCB, and GABRA4. In addition to the neurological pathway, two pathways related to the vasculature were detected (Figure 3). e significant genes PRKCB and ADCY4 can influence the RAP1 protein-signaling pathway that regulates integrin production; integrin production controls normal blood flow. Other important pathways included trafficking of GluR2-containing AMPA receptors and G alpha (z) signaling events, which regulate cardiac function.

DBP Variability.
For DBP variability, 40 of the top 100 genes were significant (Table 2) and the biological pathway analysis generated three significant pathways: the hippo signaling pathway, the CDO myogenesis pathway, and O-glycosylation of TSR domain-containing protein pathway (Table 4 and Figures 4 and 5). e last two pathways have a relation with the cardiac muscle and play a role in DBP variability [13,14]. ree genes-CTNNA2, TEAD4, and BIRC5-were significant in the gene-based analysis. ey were identified in the hippo signaling pathway (Figure 4).

Discussion
In the present study, we conducted three types of statistical analyses to explore the genetic background of VVBPV. First, we conducted a GWAS to identify variants that could increase VVBPV among Americans. Second, we conducted gene-based analysis to identify potential genes that influence VVBPV. ird, we conducted pathway analyses to interpret the role of top-scoring SNPs located on the identified genes.
Regarding the GWAS analysis, we observed several suggestive, but not statistically significant, associations (P values >5.0E − 8). We used stringent P value thresholds for significance-5.0E − 8-to control for false-positive results  BioMed Research International obtained from the large number of SNPs tested, but we did not observe P values ≤5.0E − 8 because of lack of power. erefore, we conducted gene-based analyses to overcome the lack of power. e gene-based analysis has several advantages: First, the gene is the functional unit of the human genome. When we use the gene as the unit of analysis, we can conduct functional analyses such as biological pathways. Second, such analyses reduce the multipletesting burden substantially; they require a correction for thousands of genes rather than millions of SNPs. Finally, we conducted pathways analyses to interpret the gene-based results into functional information, which might identify the pathogenic mechanisms of VVBPV.

Systolic VVBPV Pathways.
In this study, we identified biological pathways that might control SBP variability. e first pathway is the GABAergic pathway, which regulates important neurotransmitters, e.g., GABA.
GABA is an inhibitory neurotransmitter. It is converted to glutamate by a metabolic pathway called the GABA shunt. In the GABAergic pathway (Figure 2), the ADCY4 gene was significant in our analysis (Table 2). is gene regulates the production of GABA, which can inhibit sympathetic activity [15]. e ADCY4 also encodes for the adenyl cyclase enzyme. e adenyl cyclase is involved in the production of cyclic AMP (c-AMP) and activation of protein kinase A (PKA), which increases GABA release [16]. GABRA4 [15], a significant gene in our analysis, has a role in the production of GABA receptors. ese receptors are regulated by PRKCB [17]. Reduction in GABA will increase the glutamatergic input in the hypothalamic paraventricular nucleus. e increase in glutamate input inhibits the trafficking of the GluR2-containing AMPA receptors [18], which regulates blood pressure [19].
In addition to the neurological pathways, we identified two pathways related to the vasculature (Figure 3). e significant genes related to this pathway were PRKCB and ADCY4. ADCY4 forms c-AMP [12,20] and activates RAP1 (Figure 4). PRKCB can activate RAP1 via the protein kinase C (PKC) and phosphorylation of protein kinase-D (PKD) [21]. e RAP1 signaling pathway controls blood pressure variation by the following mechanisms: (1) it modulates the normal function of endothelial cells and angiogenesis [12]; (2) it affects peripheral vascular resistance (i.e., blood flow and pressure); and (3) it induces myocyte growth and hypertrophy [12]. Activation of RAP1 increases integrin activity. Integrin maintains the myogenic activity of blood vessels and regulates blood flow [12].
is study should be interpreted in the context of a very important limitation. Although VVBPV could have a genetic component regulating it, the effect of the environment cannot be ignored. We had no chance to evaluate the geneenvironment interaction effect because (1) we lacked information on environmental variables, e.g., type and frequency of the antihypertensive medication intake and salt intake, and (2) we had insufficient sample size for evaluating such an effect. is study has many advantages. First, to our knowledge, this is the first study to explore the genetic background of VVBPV in a sample of an American White and Black population. Second, the data of the study cohorts were collected according to comprehensive protocols and thorough quality control measures. ird, the datasets had information on important covariates, which allowed us to adjust for important confounders while conducting the statistical analysis. Fourth, the results of the study may impact the public health field and clinical practice: the discovered pathways could be targeted to control blood pressure variability that leads to adverse cardiovascular outcomes.
In summary, this study shows that VVBPV might have genetic and biological components, which regulate it. Future studies are warranted to complete the following aims: (1) verifying the results of the present study and (2) identifying new pharmacological drugs that target the discovered pathways.
Data Availability e data we have analyzed are owned by the National Heart, Lung, and Blood Institute (NHLBI) and National Center for Biotechnology Information (NCBI). Investigators need to obtain permission from these two institutes before they use the data.

Disclosure
is work was not prepared in collaboration with investigators of the CHS or WHI and does not necessarily reflect the opinions of these investigators or the NHLBI. e CHS datasets used to prepare this manuscript were obtained from dbGaP at http://www.ncbi.nlm.nih.gov/sites/entrez? db�gapthrough dbGaP accession (phs000287.v6.p1).

Conflicts of Interest
e authors declare that they have no conflicts of interest.  Figure 4: Hippo signaling pathway in the DBP variability.   BioMed Research International