Hypertension is a leading risk factor for cardiovascular disease and stroke and it is estimated that over 41% of adults in the U.S. will have hypertension by 2030 with a disproportionate burden among ethnic minorities [1
]. Multiple lines of evidence indicate that inter-individual variation in blood pressure (BP) is influenced by both genetic [2
] and non-genetic factors including socioeconomic status (SES) [3
] and psychosocial factors [4
]. Given that complex traits such as BP are likely shaped by multiple risk factors as well as their interactions with one another, a deeper understanding of the gene-by-socioeconomic/psychosocial factor interactions on BP may help to identify individuals that are genetically susceptible to high BP in specific social contexts.
Genome-wide association studies (GWAS) have identified significant, replicated predictors of systolic (SBP) and diastolic blood pressure (DBP) in populations of primarily European, African and Asian ancestries [5
]. Many of the identified loci harbor genes with plausible biological roles in BP regulation [5
]. Several physiological pathways have been proposed for these genes in the pathogenesis of hypertension, many of which are likely to be modulated by non-genetic factors. For example, the ATP2B1
gene, which was strongly associated with both SBP and DBP [5
], encodes PMCA1, a plasma membrane calcium/calmodulin-dependent ATPase that is expressed in vascular endothelium and is involved in calcium pumping from the cytosol to the extracellular compartment [11
]. This signal transduction pathway is influenced by oxidative stress and inflammation [12
], which are known to be associated with lifestyle (e.g., physical activity, diet, smoking), sociodemographic, and psychosocial factors (e.g., stress) [14
In parallel, epidemiological studies have shown that socioeconomic and psychosocial factors are also related to BP levels. Several studies have linked low SES at the individual level [3
], neighborhood level [18
] and over the life course [19
] to high BP. Possible pathways mediating these effects include influences of low SES on health behaviors such as salt intake or lack of physical activity as well as possible links between SES and stress, which has been hypothesized to be associated with BP through neuroendocrine mechanisms [20
]. Psychosocial factors have also been linked to high BP, possibly through effects on behavior or direct impact on sympathetic and hypothalamic-pituitary-adrenal (HPA) axis activity [3
]. For example, high levels of anger have been associated with progression from prehypertension to hypertension [20
], depressive symptoms have been associated with increased SBP and DBP [22
] as well as hypertension incidence [24
] and stress has been associated with BP progression [25
]. Nevertheless, findings for some psychosocial factors, such as depressive symptoms, have been inconsistent across studies and important questions remain regarding the relative importance of socioeconomic and psychosocial factors in explaining variability in BP.
The proper understanding and quantification of the etiologic roles of genes and environments in the causation of complex diseases will require consideration of gene-by-environment interactions [26
]. Specifically, the influence of genetic predictors may be enhanced or suppressed by the presence of a certain environmental context. Likewise, the influence of environmental exposures may be modified by genetic predisposition. The investigation of gene-by-environment interactions in the genomic era remains in its infancy in part because of the lack of large population studies with rich environmental and genomic measurements as well as analytic methods that can effectively utilize the data. In addition, the majority of gene-by-environment interaction studies for complex diseases and traits such as BP have focused on demographic factors (e.g., age and body mass index (BMI)) or health behaviors (e.g., smoking, salt intake) [27
] and have had cross-sectional designs. Very few have examined the socioeconomic or psychosocial environments in which genes operate and even fewer have used repeated measures data. However, the pathways through which socioeconomic and psychosocial factors are hypothesized to influence BP (which may include health behaviors but also physiologic processes such as stress response, oxidative stress, inflammation, or changes in the immune system) are likely to interact with genetic predispositions and these relationships may change over time.
In this study, we used longitudinal data from four large epidemiologic cohorts of European and/or African ancestry to investigate interactions between known socioeconomic/psychosocial and genetic risk factors on variation in BP. We used a novel genomic region-based method for repeated measures analysis, Longitudinal Gene-Environment-Wide Interaction Studies (LGEWIS) [28
], to evaluate interactions rather than testing each single nucleotide polymorphism (SNP) individually. Region-based approaches such as LGEWIS may be advantageous for trans-ethnic analysis because they are able to detect interactions even in the presence of genetic heterogeneity in ancestrally diverse populations. Testing all SNPs in a region simultaneously substantially reduces multiple testing burden compared to traditional one-at-a-time SNP analysis. Furthermore, region-based analyses are able to detect the cumulative small effects of multiple SNPs within the region that may be missed by single SNP approaches that rely on the presence of high impact SNP effects. Longitudinal approaches to genetic association and gene-by-environment interaction studies that allow for varying outcome and exposure trajectories also have improved power compared to cross-sectional analysis [28
]. This work represents an important first step toward comprehensively integrating social and biological factors to better understand the determinants of BP in multiple race/ethnic groups.
The influence of socioeconomic and psychosocial factors on BP and hypertension is widely recognized [4
]. Advances in genomic technologies and multi-cohort collaborations are now providing evidence for the additional influence of hundreds of genes [5
]. However, most of the heritability of BP remains unexplained, likely in part due to the presence of gene-by-environment interactions [27
]. A deeper understanding of the context-dependent genetic effects on BP may yield insight into the biological mechanisms of hypertension etiology and facilitate personalized medicine approaches to BP control that may include lifestyle modifications, more aggressive prevention or treatment approaches, and/or pharmacogenomics. Further, a more complete understanding of the similarities and differences in context-dependent effects across multiple race/ethnic groups may help in the development of effective strategies for reducing health disparities in hypertension.
To our knowledge, this study is one of the first to meta-analyze the effects of socioeconomic and psychosocial factors on BP in multiple cohorts of European and African ancestry. Consistent with prior literature [3
], we found that lower adult SES, as measured by having less than a high school degree, was associated with increased BP. Lower childhood SES was most strongly associated with SBP in EA but was not associated with BP in AA. Of the three psychosocial stressors evaluated (anger, depressive symptoms, and chronic burden), all were associated with both SBP and DBP in AA participants, except for depressive symptoms with SBP. However, they were not significantly associated with BP in EA. All observed associations were in the expected direction (lower SES and childhood SES and greater anger, depressive symptoms, and chronic burden were associated with higher BPs).
A previous study in EA and AA ARIC participants found that high levels of trait anger were associated with progression from prehypertension to hypertension [20
]. A recent study in AA from JHS [25
] also found outwardly-expressed anger is associated with BP stage progression. Hostility, a trait closely related to anger, was associated with hypertension risk in the Coronary Artery Risk in Young Adults study (CARDIA) as well [56
]. In this study, outward/trait anger associated with continuous measures of SBP in AA, with each increase of 1 unit on the outward/trait anger score (range 0–4) conferring an increase of 0.85 mmHg. For AA, a 1 unit increase in anger was also associated with a 0.68 mmHg increase in DBP. The strong relationship between anger and BP in this study sample is perhaps expected, since ARIC and JHS comprise two of the four cohorts meta-analyzed here. In EA, although the relationship between anger and BP was in the expected direction, the effect was not significant. Biological mechanisms for this relationship may include sympathetic nervous system hyperactivity and arousal due to anger and psychological stress [20
Although some studies have demonstrated increased risk of hypertension [24
], BP progression (an increase in BP stage) [25
], or increases in BP [22
] with depression/depressive symptoms, other studies show inverse [59
] or no relationships [4
]. Differences in findings may be due to study design (duration of follow-up, measurement of depression, and age or race/ethnic composition of the study sample). In this study, repeated measures analysis showed no association between depressive symptoms and either SBP or DBP in EA and only a weak relationship with SBP in AA. However, the relationship between depressive symptom score and DBP was significant in AA (p
= 0.002), with a 10% increase in depressive symptom score associated with a 0.18 mmHg increase in DBP. Potential mechanisms for the relationship between depression and hypertension include both behavioral (e.g., smoking, physical activity, adherence to antihypertensive medication) and biological (e.g., inflammation, altered night-time BP dipping as a result of insomnia) mechanisms [60
We used a novel genomic region-based, repeated measures approach to assess the relationship between 33 genomic regions known to be associated with SBP, DBP, or both. These regions were identified in a GWAS meta-analysis of EA (29 regions) [5
] and a GWAS meta-analysis of AA followed by trans-ethnic meta-analysis (an additional 4 regions) [6
]. Of these 33 regions, 25 (75%) had a marginal association p
-value of < 0.2 for SBP and/or DBP in our meta-analysis of EAs, with the majority of these (17 regions) achieving p <
0.2 in both. In AA, a smaller number of regions (16, 48%) had a marginal association p
-value < 0.2, including only 5 regions with p <
0.2 for both SBP and DBP. These regions with marginal p <
0.2 were carried forward for genomic region-by-psychosocial factor interaction analysis. However, only 12 and 10 regions in EA were significant with SBP and DBP, respectively, at a False Discovery Rate of 5%. No regions were significant after multiple testing correction in AA, likely because most of the regions were identified in meta-analysis of EA only, as well as the significantly smaller sample size of AA compared to EA participants in this study.
Gene-by-environment interaction studies on BP and related traits are now being reported and catalogued [27
], including large-scale genome-wide analyses of genetic interactions with lifestyle factors (alcohol consumption [62
], smoking [63
], sodium intake [66
], age [67
], and BMI [68
]). To our knowledge, only one genome-wide study has examined interactions between socioeconomic or psychosocial factors and BP. In this study of 3836 participants from the Framingham Heart Study (FHS), two genome-wide significant and three suggestive gene-by-education interactions on SBP (1 interaction), DBP (3 interactions) and pulse pressure (1 interaction) were identified [69
]. Although the 5 genes identified were all biologically related to BP, none were identified in GWAS studies used to select genomic regions for evaluation in this study, so they were not included in our analysis. We also identified one study that examined interactions between polymorphisms in biological candidate genes and socioeconomic/psychosocial factors on BP. In this study of 208 AAs, neighborhood SES was found to interact with glucocorticoid receptor polymorphisms to influence cortisol levels but not BP [70
]. Finally, in the paper describing the LGEWIS method, He et al. investigated interactions between genes known to influence BP and perceived or geographic information system (GIS)-based measures of healthy food and physical activity environment in MESA participants [28
]. Interactions were observed for SBP between the genomic region indexed by rs10850411 and perceived healthy food availability in EA, and between the CACNB2
genomic region and density of recreational facilities in Hispanic Americans. Although we evaluated both of these genomic regions in our study, we did not observe interactions between these regions and socioeconomic/psychosocial factors on BP.
In this multi-cohort study examining socioeconomic/psychosocial factor interactions with genomic regions identified through large, replicated BP GWAS, we observed two significant interactions after correction for multiple testing. Outward/trait anger was found to have an interaction with the C10orf107
gene region on DBP in EA. C10orf107
, or CABCOCO1
, is a relatively uncharacterized gene but the mouse homolog encodes a protein with a predicted coiled-coil domain and a CLAMP motif with a leucine zipper domain that has calcium-binding activity [71
]. Results of tissue-specific gene expression patterns from the GTEx Analysis Release V6p (dbGaP Accession phs000424.v6.p1) [72
] showed that this protein is most highly expressed in the testis, spinal cord and brain tissues, pituitary gland, prostate and lung. Further research is necessary to characterize the biological mechanisms that underlie the interaction between C10orf107
and anger on DBP. To ensure that the interaction was not detected due to gene-environment correlation, we used LGEWIS to assess the association between the C10orf107
genomic region and anger (p
> 0.1 in all cohorts, indicating no gene-environment correlation). Single SNP analysis in this region revealed no SNPs that appeared to be driving this association in multiple cohorts. This finding in EA illustrates the ability of genomic region-based analysis techniques to detect interaction signals that reflect a shift in the entire distribution of interaction effects across the genomic region, due to the presence of multiple SNPs that have small influences, even when no SNPs with high-impact interaction effects are present.
We also identified an interaction between the HFE
gene region and depression symptom score on DBP in AA. HFE
encodes a membrane protein that regulates iron binding, which influences absorption of iron in the small intestine and recycling of iron in macrophages [73
] and has also been shown to demonstrate immune-related activities that bridge adaptive and innate immunity [74
]. Mutations in HFE
can cause hereditary hemochromatosis (HH) [75
], characterized by iron accumulation that can lead to tissue damage. Hemochromatosis may be associated with BP through increased serum ferritin (iron) levels that alter heart morphology or cause metabolic abnormalities such as insulin resistance [76
]. Early signs of hemochromatosis include fatigue, malaise, joint pain and swelling and enlarged liver [78
]. In a recent study of 395 subjects with HFE
-related HH, 41% of the HH patients meeting the criteria for fibromyalgia syndrome, characterized by chronic joint pain and fatigue, also met the criteria for depression. To ensure that the interaction observed in this study was not due to gene-by-environment correlation, we tested the HFE
genomic region for association with depressive symptom score in each of the AA cohorts using LGEWIS. None of the cohorts demonstrated significant association (all p
> 0.1, data not shown), so we concluded that the interaction was not due to correlation between HFE
and depressive symptoms.
This study is not without limitations. The LGEWIS method only assesses the significance of the interaction term between the genomic region and the socioeconomic/psychosocial factor. However, in some circumstances including gene discovery, it may be optimal to test the joint effects of the genomic region and the interaction simultaneously. This approach has been favored by large consortia now investigating gene-by-environment interactions on a genome-wide scale [79
]. Further work is needed to expand the repeated-measures genomic region-based methods to incorporate the analysis of joint effects. Another limitation is that this study included only 33 of the genomic regions that have been identified as being associated with BP. However, recent GWAS have increased the number of loci to over 100 [7
] and exploring interactions with these newly discovered loci is an important direction for future research. In addition, the 33 genomic regions were selected through GWAS meta-analyses that contained the cohorts analyzed in this study. The ARIC EA participants were included in the discovery sample (N
= 69,395) of the GWAS meta-analysis in EA only (total N
> 200,000 EA) [5
], and all four of the AA cohorts were included in the discovery sample (N
= 29,378) of the trans-ethnic meta-analysis (total N
> 120,000) [6
]. From a genetic perspective, another limitation is that this study only evaluated common SNPs; however, this line of research could be expanded to examine interactions in these gene regions using measures of rare, potentially functional variants such as exome or whole genome sequence data.
From a phenotypic perspective, although we were able to successfully harmonize socioeconomic/psychosocial factors in this study, there is heterogeneity across the cohorts in the measurement of these factors. This is an ongoing challenge for gene-environment interaction studies [80
], emphasizing the need standardized phenotype measures such as those catalogued in the Phoenix Toolkit (www.phenxtoolkit.org
). In addition, we used educational attainment as our sole marker of SES, which is only one facet of individual-level SES and may be confounded with other factors that influence health (such as living in a rural vs. urban area). Educational attainment and the psychosocial factors evaluated in this study may also be confounded by other unmeasured exposures such as smoking, alcohol consumption and physical activity; however, these lifestyle factors may also act as mechanisms by which socioeconomic/psychosocial factors influence BP. Although all cohorts retained over 70% of participants over the course of the study, attrition was more extreme for AA (vs. EA), lower SES (vs. higher SES), and older participants. To help alleviate concerns about bias due to attrition, analyses were controlled for age and sex and were stratified by ethnicity. Sensitivity analyses indicated that further adjustment for adult SES did not attenuate the significant interaction findings (data not shown). We chose to account for antihypertensive use by adding a constant value (+15/10 mmHg) to BP; however, we conducted a sensitivity analysis to evaluate whether an alternative method (using original BP measures and including antihypertensive use as a covariate in all models) influenced they key findings. Effect sizes for significant associations between psychosocial factors and BP, as well p
-values for the two significant genomic region-by-psychosocial factor interactions on DBP, did not change appreciably using this alternative method (data not shown). Finally, we included BMI as an adjustment variable in all models. This approach, used by the discovery GWASs [5
], reduces residual BP variance and boosts power to detect interaction effects; however, it reduces power to detect interactions mediated at least in part by changes in BMI.
We detected two interactions in this study but their effect on BP is small, and although these findings were statistically significant, it is still possible that they were due to chance. Therefore, the clinical relevance of these findings has yet to be determined. Further, although our multiple testing correction accounted for the number of genes we evaluated for interaction, we did not account for testing multiple socioeconomic/psychosocial factors within each ancestry group. Since the socioeconomic/psychosocial factors are correlated, a Bonferroni approach would have been conservative in this instance. However, if we had followed this more stringent multiple testing approach, the interactions we detected would have been attenuated. Although we included four cohorts in this study, we also had limited power to detect effects, which may account for the lack of significance of the associations between some of the genomic regions and BP, especially in AA. Lack of power may also have contributed to inconsistent findings across ancestry groups, both for genomic region associations with BP as well as interactions.
Despite these limitations, this study has several notable strengths including the use of novel genomic region-based methods that allow for repeated measures analysis. Genomic region-based methods to assess interaction are advantageous because they reduce the multiple testing burden compared to testing each SNP individually, they allow for the detection of interactions that are driven by multiple SNPs within a genomic region rather than those being driven by a single genetic factor, and they may also be more suited for trans-ethnic studies because they do not rely on SNPs to have similar linkage disequilibrium patterns across ancestry groups. Meta-analysis across multiple cohorts increases the credibility and generalizability of findings, yet additional replication will be necessary to further confirm these interactions.