Familial Analysis of Epistatic and Sex-Dependent Association of Genes of the Renin–Angiotensin–Aldosterone System and Blood PressureCLINICAL PERSPECTIVE
Background—Renin–angiotensin–aldosterone system genes have been inconsistently associated with blood pressure, possibly because of unrecognized influences of sex-dependent genetic effects or gene–gene interactions (epistasis).
Methods and Results—We tested association of systolic blood pressure with single-nucleotide polymorphisms (SNPs) at renin (REN), angiotensinogen (AGT), angiotensin-converting enzyme (ACE), angiotensin II type 1 receptor (AGTR1), and aldosterone synthase (CYP11B2), including sex–SNP or SNP–SNP interactions. Eighty-eight tagSNPs were tested in 2872 white individuals in 809 pedigrees from the Victorian Family Heart Study using variance components models. Three SNPs (rs8075924 and rs4277404 at ACE and rs12721297 at AGTR1) were individually associated with lower systolic blood pressure with significant (P<0.00076) effect sizes ≈1.7 to 2.5 mm Hg. Sex-specific associations were seen for 3 SNPs in men (rs2468523 and rs2478544 at AGT and rs11658531 at ACE) and 1 SNP in women (rs12451328 at ACE). SNP–SNP interaction was suggested (P<0.005) for 14 SNP pairs, none of which had shown individual association with systolic blood pressure. Four SNP pairs were at the same gene (2 for REN, 1 for AGT, and 1 for AGTR1). The SNP rs3097 at CYP11B2 was represented in 5 separate pairs.
Conclusions—SNPs at key renin–angiotensin–aldosterone system genes associate with systolic blood pressure individually in both sexes, individually in one sex only and only when combined with another SNP. Analyses that incorporate sex-dependent and epistatic effects could reconcile past inconsistencies and account for some of the missing heritability of blood pressure and are generally relevant to SNP association studies for any phenotype.
- blood pressure
- cardiovascular disease
- molecular epidemiology
- population genetics
- renin–angiotensin system
- twin study
The renin–angiotensin–aldosterone system (RAAS) has a central role in the physiology of blood pressure and is implicated in the pathogenesis of hypertension and cardiovascular disease. The RAAS cascade begins with the rate-limiting cleavage of angiotensinogen (AGT) by renin to produce angiotensin I (Ang I).1 The subsequent conversion by the angiotensin-converting enzyme (ACE) of Ang I forms angiotensin II (Ang II) that is responsible for a broad range of physiological actions, many of which are mediated through the angiotensin receptor type 1 (AT1), including vasoconstriction and the release of aldosterone from the adrenal glands. Ang II and aldosterone are the classical effectors of the RAAS with complementary actions to raise blood pressure. The enzymes, hormones, and receptors of the RAAS provide targets for effective therapies for hypertension and cardiovascular disease. They also have been popular candidates for genetic studies.
See Editorial by Brian J. Morris
Several reviews and meta-analyses summarize the vast number of studies of the RAAS genes in cardiovascular physiology, disease, and treatment.2–4 Genes encoding renin (REN), angiotensinogen (AGT), angiotensin-converting enzyme (ACE), angiotensin type 1 receptor (AGTR1), and aldosterone synthase (CYP11B2) feature prominently. Functional polymorphisms in these genes have been associated with significant variation in the levels of RAAS components including angiotensinogen, ACE, and renin.2 They have also been associated with high blood pressure and other cardiovascular diseases including coronary artery disease, preeclampsia and diabetic nephropathy.2
The results of gene studies of individual RAAS candidates have been heterogeneous in terms of the nature of the associations and the magnitude of the estimated phenotypic effects. Some of these inconsistencies have been attributed to inadequate statistical power and racial differences. Other factors, including sex-dependent effects5 and gene–gene interaction (epistasis) are also potential contributors.
Biologically, sex is associated with important differences in the regulation of arterial pressure and renal function by the RAAS.6 Also, the impact of genetic variation on regulation of many autosomal genes is known to be dependent on sex.7 Of note, some RAAS genetic studies that stratified by sex have reported sex differences in the associations between cardiovascular disease and AGT,8–10 ACE,11 and AGTR1.12
Given the biological cascade and interdependence of the individual components of the RAAS, there is potential for interaction between polymorphisms of RAAS genes. As proposed by Moore and Williams,13 the failure to replicate some single locus associations might be because the impact of single alleles on cardiovascular risk is dependent on genetic variation at other loci (ie, epistasis). Indeed, it has been argued that simply modeling genetic polymorphisms as independent and additive effects on the phenotype is not a realistic biological model and that epistasis should be included,14 although others have questioned the relative importance of epistasis in complex traits.15
Many studies have tested for epistasis in relation to polymorphisms of RAAS genes and have suggested interactions between AGT and ACE,9,16–18 between multiple genes,19 and within haplotypes comprised of certain combinations of variants within individual genes.20–24 Other studies have been unable to identify evidence that might indicate epistasis.25,26
Except for a small number of studies (such as Zhu et al23), most genetic association studies of the RAAS have examined only a single marker at each gene. Unfortunately, such approaches do not fully capture the genetic variation in and around each gene, some of which may code changes that are involved in sex-dependent and epistatic interactions.
In the present study, we sought to better understand the role of RAAS gene variants in blood pressure variation in the general population by improving on previous approaches. We undertook a comprehensive gene-wide mapping of REN, AGT, ACE, AGTR1, and CYP11B2 against blood pressure in 2-generation adult families recruited to the VFHS (Victorian Family Heart Study)27 and used variance components modeling to test for sex-dependent and epistatic effects. Family-based designs have unique advantages over population-based designs as they are robust against population substructure, permit more sophisticated modeling and multiple hypothesis testing28 including interactions,29 and allow better detection of genotyping errors.30
The details of the recruitment of subjects for the VFHS have been published previously27 and provided in the Methods section in the Data Supplement. The Ethics Review Committee of the Alfred Hospital, Melbourne, approved the study, and informed consent was obtained from all participants. In brief, a volunteer sample of 767 white adult families enriched with families containing twins (70 monozygotic [MZ] pairs and 84 dizygotic [DZ] pairs) and comprising both parents and at least one natural offspring was recruited. Data were also available for 89 individuals (41 incomplete families).
In this study, we used the average of 2 lying and 2 standing measurements of systolic blood pressure (SBP) and diastolic blood pressure (DBP). For subjects receiving antihypertensive treatments, we adjusted the recorded pressures by adding 10 and 5 mm Hg to SBP and DBP, respectively, as justified previously31 and shown to provide an appropriate adjustment.32
The selection and fate of SNPs in this study is summarized in Figure I in the Data Supplement. TagSNPs were selected from the Hapmap database (data release #24, November 08)33 to provide coverage of each of the 5 RAAS genes including 10 kb up and downstream of each gene. These 104 markers captured 262 validated SNPs with minor allele frequencies of >1%. They comprised 17 SNPs for REN (with rs11240688 tagging the commonly genotyped -5312C/T polymorphism at r2=1 according to the SNAP database, www.broadinstitute.org/mpg/snap/ldsearch.php), 16 for ACE (with rs4353 tagging the I/D polymorphism34), 29 for AGT (with rs6687360 tagging the M235T polymorphism, r2=0.83), 33 for AGTR1 (with rs5186 as the A1166C polymorphism), and 9 for CYP11B2 (with rs1799998 tagging the −344C/T polymorphism, r2=0.90).
These 104 tagSNPs, plus an additional 3 duplicate SNPs (included to test the reliability of the genotyping platform and check for concordance in genotype call), were processed by Sequenom MassARRAY Assay Design software v3.1 (Sequenom). Four of these 107 SNPs were rejected by the software program and excluded entirely, and one well designed by the software to contain a single SNP was not genotyped. The remaining 102 SNPs (including the 3 duplicates) were automatically multiplexed into 4 separate assays containing 33, 29, 27, and 13 SNPs. SNP genotyping was performed using MassARRAY MALDI-TOF technology and Sequenom iPLEX Gold chemistries according to manufacturer’s instructions. Primers for each SNP in each multiplex are shown in Table I in the Data Supplement. Acquired raw genotyping data were analyzed using Typer 4.0 software package (Sequenom).
Genotype and Pedigree Checking
In addition to the usual genotyping quality control checks (SNP Hardy–Weinberg equilibrium [estimated via a likelihood-based method which takes into account relationships between individuals,35 excluding SNPs with P<0.01 and treating with caution any SNPs with 0.01<P<0.05], genotype call rates per SNP and per individual [threshold 90% for each well]), our family-based design permitted error checking between relatives of both genotypes and family relationships. The programs PLINK,36 a user-written R script,37 and RelativeFinder (a component of Merlin)38 were used as described in the Methods section in the Data Supplement. We also compared genotypes for each of the 3 duplicated SNPs. Apparent genotypic errors were resolved by manually re-examining raw genotype data and correcting final genotype calls, setting genotypes to missing, or correcting or modifying pedigrees and relationships.
To assess association of individual SNPs with SBP within families, variance components models were fitted using the software package Solar.39 Age, sex, and age–sex interactions were included as fixed covariates in all models, with separate effects allowed for each generation (parents and offspring).
Each SNP was included first as a continuous exposure (0, 1, or 2 copies of the minor allele) in an additive model and then using the quantitative trait linkage disequilibrium approach and parameterization as implemented in Solar, which tests for and is robust to population stratification.40
Within each family, a multivariate t distribution was assumed for the vector of outcomes conditional on covariates, as this distribution provided a better fit than the multivariate normal distribution. The variance for each individual was σ2, and the covariance between a pair of relatives j and k within a nuclear family was ρjkσ2, where ρjk is one of ρSP, ρPO, ρSIB, ρDZ, and ρMZ, depending on whether j and k are from a spouse, parent–offspring, sibling, DZ, or MZ twin pair. This general model has been used for previous analyses of data from this study.27 The parameters of the model are estimated using maximum-likelihood estimation.
Sex-by-SNP (modeled as additive genotype) interaction effects were then included in the models described above. Finally, all possible models that included a single pair of SNPs and their interaction were fitted. All possible pairwise combinations of SNPs were considered, and in each model, a main effect for each of the 2 SNPs in the model (modeled as additive genotype) and an interaction term for the product of the number of minor alleles of each SNP in the pair were included as fixed effects. The interaction term allows the effect of one SNP to differ according to the genotype at the second SNP. Sex-by-SNP-by-SNP interaction terms were not modeled.
Although multiple testing is clearly a significant issue in genetic association analysis, there is not an ideal way to address this issue, particularly as our main aims were essentially hypothesis generating. As an initial conservative indicator, we calculated a Bonferroni correction to the nominally significant α of 0.05, where the number of independent tests was estimated by adding the effective number of uncorrelated SNPs in each of the 5 genes, estimated by Solar using the methods of Moskvina and Schmidt.41 The effective number of independent SNPs was estimated to be 66 (see below), resulting in a Bonferroni-corrected P value of 0.05/66=0.00076 for individual SNP associations and sex-dependent associations, and P=0.05/(66C2)=0.05/2145=2.3×10–5 for the epistatic association analyses. In the context of hypothesis generation, we report individual SNP associations where the Wald P value for the regression coefficient of the SNP was <0.005. In the sex-by-SNP interaction analyses, we focused on SNPs for which the P value for the interaction term was <0.05, and in the SNP-by-SNP interaction analyses, we focused on SNPs for which the P value for the interaction term was <0.005. For each such model, we investigated the associations more thoroughly and performed model checking.
After full quality control (see Methods section and Figure I in the Data Supplement), genotypes were available for 88 SNPs (13 from REN, 23 from AGT, 31 from AGTR1, 8 from CYP11B2, and 13 from ACE) on ≤2872 individuals in 809 distinct corrected pedigrees. The final number of pedigrees included was larger than the original 808 as some families were split into 2 separate pedigrees because of relationship errors, as described in detail in the Data Supplement. Initial concordance between duplicated SNPs was 99.4% for rs2933250, 100% for rs12695894 (which was almost nonpolymorphic), and 99.2% for rs2675511. Additional details about these 88 SNPs, including minor allele frequency and percentage genotyped and Hardy–Weinberg P value, are included in Table II in the Data Supplement, and linkage disequilibrium (LD) between SNPs in our sample is shown in Figure II in the Data Supplement.
Individual SNP Association
We found that 9 SNPs showed evidence of association with SBP, tested using either the quantitative trait linkage disequilibrium or additive model at P<0.05 (Table 1; Table III in the Data Supplement). Of these, the P values for 2 SNPs (rs8075924 at ACE and rs4277404 at ACE) were less than the Bonferroni-adjusted P value for both tests, and one additional SNP (rs12721297 in AGTR1) had a P value <0.005 but >0.00076 for both tests. All 3 SNPs had 0.005<FDR P value<0.04. Minor alleles were associated with lower SBP for all 3 SNPs, and the estimated effect sizes from the additive model ranged from −1.7 to −2.8 mm Hg for these SNPs (Table 1). These 3 associations appeared robust and remained when analyses included adjustments only for age and sex, when outlying individuals and pedigrees were excluded (92 individuals and 7 pedigrees), when individuals with >10% of genotypes missing (125 individuals) were excluded, and under a variety of different genetic models (eg, dominant, 2df genotype; data not shown). If there were actually no real associations and assuming independence of tests, we would expect a minimum of 3 to 4 significant results at P=0.05 by chance alone. Because 9 of our SNPs were associated at P=0.05, we can be moderately confident that at least some of these associations are not solely due to chance.
No evidence of significant SBP association was seen with the familiar RAAS polymorphisms REN -5312C/T, ACE I/D, AGT M235T, AGTR1 A1166C, or CYP11B2 -344C/T using our SNP proxies. However, the SNP rs6687360 that tags the M235T polymorphism at AGT showed nominal association by quantitative trait linkage disequilibrium (P=0.034) and using an additive model (P=0.046; Table 2).
We found evidence for differential effects according to sex (SNP×sex interaction P value <0.05) for 4 SNPs: rs2478523 at AGT, rs2478544 at AGT, rs11658531 at ACE, and rs12451328 at ACE (Table 3). Interaction analysis results for all SNPs are shown in Table IV in the Data Supplement. No SNP achieved the threshold for significance according to Bonferroni-adjusted P values, and the lowest FDR-adjusted P value was 0.29. We would again expect a minimum of 3 to 4 significant results at P=0.05 by chance. The SNPs rs2478523, rs2478544, and rs11658531 showed effects in males but not in females, where minor alleles were associated with increased SBP with estimated effects ranging from 1.3 to 1.6 mm Hg (Table 3 and Figure 1). The SNP rs12451328 showed effects only in females but not in males with an effect size of ≈1.3 mm Hg (Table 3; Figure 1). Again, these associations were robust in a variety of model checks (data not shown).
Of the almost 4000 (88C2=3828) pairwise SNP-by-SNP combinations, interaction models identified 14 pairs with Wald P values (comparing models with and without interaction terms) below the nominal level of 0.005 (Table 4). No interactions were significant using the Bonferroni-adjusted P value 2.3×10−5, and the lowest FDR-adjusted P value was 0.10. Table 4 shows the genes at which the SNPs are located. The top 14 interacting pairs involved 19 individual SNPs, with the SNP rs3097 (at CYP11B2) represented in 5 pairs and 5 other SNPs seen in 2 pairs each (rs7079, rs4295, rs10494849, rs11240688, and rs12639531). Four of the 14 pairs suggested interaction between SNPs at the same gene with 1 pair for AGT, 2 pairs for REN, and 1 pair for AGTR1. For the other 10 pairs, only one pair suggested interaction between genes that could be considered sequential steps in the RAAS cascade (AGTR1 and CYP11B2). None of the top 14 SNP pairs was among those showing individual associations with SBP (Table 2), and none of the well-known RAAS polymorphisms showed evidence of interaction. Interestingly, one individual SNP from the top 14 interacting pairs, rs2478544, also showed evidence of sex-dependent effects (Table 3). We would expect a minimum of 19 significant results at P=0.05 by chance.
The interaction parameter (interaction β) provides an estimate of the effect on SBP of the different SNP genotype combinations. As the main effect size for SNP1 (in a model without interactions) reflects the mean SBP change with each additional minor allele of SNP1 (ie from 0 to 1 minor alleles and from 1 to 2 minor alleles), the interaction effect can be thought of as the difference of differences, as it reflects the difference between the SNP1 effect size for SNP2 major allele homozygotes and the SNP1 effect size for SNP2 heterozygotes. The direction and magnitude of interaction effects are shown in Table 4 and in absolute terms range from 1.4 to 5.3 mm Hg. These associations were also robust (data not shown).
Figure 2 shows a Circos plot for all individual SNP associations and the top 100 SNP–SNP interactions. A large number of interactions were between REN on chromosome 1, and SNPs more distal to AGTR1, and a moderate number were between AGT and SNPs proximal to AGTR1. The figure also shows that 2 of the 3 strongest SNP–SNP interactions were between SNPs in the same gene. Most SNPs typed were in AGTR1, but none of these were strongly associated with SBP either individually or via interactions. In contrast, ACE was covered by fewer SNPs but included the strongest single-SNP associations and one of the 3 top SNP–SNP interactions. Full results for the top 100 interactions are included in Table V in the Data Supplement.
There was a high correlation between SBP and DBP (r=0.71), and results for DBP (see Tables VI through VIII in the Data Supplement) were similar to those for SBP.
The current study examined association between population variation in blood pressure and 5 candidate genes (REN, AGT, ACE, AGTR1, and CYP11B2) that encode important elements of the RAAS. The role of the RAAS in blood pressure homeostasis and in the pathophysiology of hypertension is undisputed, but there has been considerable disagreement regarding the association of RAAS gene variants with blood pressure. Part of this inconsistency might be the result of unrecognized effects of sex5 or other interacting genes in determining individual gene variant effects.14. There is both physiological and some genetic evidence for sex-dependent effects.6,8–12 As a cascade of reactions with integrated feedback mechanisms,1 the a priori likelihood of interaction between variants in the RAAS genes also seems relatively high.
Our aim was first to identify individual genetic associations with blood pressure and then explore associations that are dependent on sex and those that depend on interactions with other variants of the RAAS genes. We relied on carefully standardized phenotypic measurements (with adjustment for potential confounders such as drug treatments) and comprehensive gene-wide mapping with extensive quality checking. Importantly, this was a study in families enriched with DZ and MZ twins. This afforded not only high-level checking for data quality, but also sophisticated variance component modeling of the blood pressure and genetic data.
Although each of the 5 genes has been associated previously with blood pressure or hypertension, we were only able to confidently identify SBP association with individual variants at loci encoding the angiotensin-converting enzyme (rs8075924 in ACE and rs4277404 at ACE) and the angiotensin type 1 receptor (rs12721297 at AGTR1). The estimated blood pressure effects per allele of these variants were relatively small (1.5–3 mm Hg) but greater than those estimated for SNPs identified by genome-wide association studies (GWAS) that are typically <1 mm Hg.43 Interestingly, for all 3 SNPs, the minor alleles were associated with lower, rather than higher, SBP. Of the RAAS variants that have been reported previously for the 5 genes (including the ACE I/D and the AGTR1 A1166C polymorphisms), only the AGT M235T polymorphism (tagged by rs6687360) showed any evidence suggestive of similar association with SBP.
However, 2 SNPs at AGT (rs2478523 and rs2478544) did show evidence of sex-dependent association, with effects on SBP only evident in men. One of these AGT SNPs, rs2478523, was also nominally individually associated with SBP overall, suggesting that the association signal in males was strong enough to overcome the lack of association in females when both sexes were combined. We also identified 2 SNPs at the ACE locus (rs11658531 and rs12451328) that showed sex-dependent association, the former evident in men only and the latter evident in women only. Neither of these ACE SNPs had shown individual association with SBP.
Previous analyses stratified by sex have focused on the M235T polymorphism of AGT, which shows a degree of LD (r2=0.6) with our sex-dependent SNP rs2478523. Consistent with our findings, Jeunemaitre8 had shown linkage of blood pressure with M235T in male but not female sibling pairs. However, other studies9,10 have suggested the M235T association to be more evident in women than in men. It is difficult to fully reconcile these data, but overall, it does appear that sex plays an important role in the association between blood pressure and AGT gene variation and should be considered in further studies.
The sex-specific findings for ACE reveal potential complexities, whereby 2 ACE variants (rs11658531 and rs12451328) in moderate LD (r2=0.54) show sex dependency but in opposite directions. Framingham data analysis has revealed that the ACE I/D polymorphism (with which neither our ACE rs11658531 nor rs12451328 exhibit significant LD) is associated with hypertension and DBP in men, but not in women.11 No other ACE variants were tested in the Framingham sample, precluding a direct comparison with our data.
The SNP–SNP interactions suggested in these analyses comprised SNPs that individually showed no significant association with SBP. The most significant evidence for interaction was between 2 SNPs (rs7079 and rs2004776) both at AGT. Interestingly, one of these SNPs (rs2004776) was in moderate LD (r2=0.6) with the well-known M2345T polymorphism and itself has been associated independently with blood pressure.44 The estimated pressure effects of previous studies of rs2004776 were modest (<0.6 mm Hg) and showed some internal inconsistency. Given that the effects of AGT rs2004776 might depend on the AGT rs7079 genotype, the past discrepancies might be resolved if interacting genotypes such as rs7079 were taken into account. Three other studies have suggested that haplotypes comprising certain variants in AGT (including M235T) were preferentially associated with blood pressure in different racial groups20 or with hypertension.24 However, these previous studies did not test for possible epistatic interactions that might have been relevant here also.
Evidence from our data for interaction between SNPs at the same gene was also suggested for REN and AGTR1. Two other studies have found certain haplotypes of variants at REN to be associated with hypertension,22,23 suggesting the possibility of epistatic relationships. However, the REN polymorphisms in the earlier studies are not in LD with the epistatic SNPs in our analyses.
The remainder of interactions implicated pairs of SNPs derived from separate genes. Only one pair involved genes (AGTR1 and CYP11B2) that might be considered adjacent steps in the RAAS cascade. Others, in physiological terms, appeared disparate, such as AGT with CYP11B2, at almost opposite ends of the RAAS cascade. Interestingly, CYP11B2 that encodes aldosterone synthase was implicated in half of the top 14 SNP–SNP interactions, most often through the SNP rs3097 that paired with 3 separate SNPs at AGT and 2 SNPs at ACE. Such a finding raises the possibility that rs3097 might play a central role in epistatic networks.
The SNPs we identified are not known to have functional effects themselves or be in LD with known functional variants. However, if our observations are verified, the apparent pairing of disparate genes of the RAAS might reflect underlying molecular links. For example, the pairs might have emerged because the SNP variants (or others in strong LD) might interfere with the binding or actions of factors such as a noncoding RNA that might normally coordinate expression of these genes toward a physiological goal.45 In this context, we tested variants in and around only the selected RAAS genes. It is possible that other epistatic interactions with other genetic components of RAAS or genes outside the RAAS are important for blood pressure control. Such possibilities invite the computationally daunting prospect of genome-wide analyses of epistasis.46
None of our genotyped SNPs altered amino acid coding. However, there is evidence to suggest that the minor allele of AGT rs7079, a 3′ untranslated region variant, creates or enhances a binding site for several microRNAs (miRNAs).47 miRNAs generally bind to gene 3′ untranslated regions and serve to destabilize the transcribed mRNA, silencing translation to protein.48 AGT rs7079 might, therefore, function to regulate AGT through reduction in translation of angiotensinogen protein. In addition, there is evidence that several of our genotyped SNPs are expression-quantitative trait loci (associated with altered expression of genes) for RAAS genes in various tissues. For example, the minor allele of AGT rs2004776, shown to interact with AGT rs7079 in our study, is associated with reduced AGT mRNA in brain tissue (Broad Institute GTEx portal: www.gtexportal.org). Therefore, our data might be indicative of an interaction between the effect of rs2004776 on AGT transcription and of rs7079 on AGT translation via 3′ untranslated region microRNA binding. The exact mechanism of interaction is unclear but warrants further study.
The use of family designs in genetic studies, as recommended by previous authors,28 provides opportunities for additional quality control checks taking advantage of Mendelian relationships. A brief analytic comparison of fully cleaned and partially cleaned data (described in the Data Supplement) suggests that, as previously observed in linkage analyses of blood pressure,49 results and power may be affected by data quality in some circumstances. A more detailed and thorough investigation of this topic may confirm this observation and elucidate the situations where this may be most important.
In summary, variants in genes of the RAAS have shown evidence of 3 different forms of association with SBP. First, they can be associated individually, independent of sex. Second, they can be associated, but only in one sex. Third, they can be associated, but only in the presence of another polymorphism. In light of our findings, it seems likely that GWAS data might reveal more of the missing heritability for blood pressure if both sex-dependent and epistatic effects were included. The estimated magnitude of the effect sizes for SBP were not vastly different between the 3 patterns, being between 1 and 5 mm Hg. Yet for sex-dependent analyses and particularly for epistatic analyses, the number of tests required to identify such effects is increased and significance thresholds fall accordingly. The identification of sex-dependent epistatic interaction would be even more challenging. Even in our exploratory analyses with a limited number of genotypes, in a relatively large family study, none of our suggested SNP–SNP interactions achieved the Bonferroni P value adjusted for multiple testing. Finally, if the patterns of epistatic interaction are to be understood in the context of underlying molecular links between variants across the genome, then sophisticated bioinformatic modeling predicated on a thorough functional knowledge of both coding and noncoding DNA will be required.
Our findings have likely implications for genetic epistatic studies of other biological systems in which precursors, enzymes, active molecules, receptors, and second messengers form a cohesive operational unit with potential for interaction and feedback. Relevant cardiovascular-related phenotypes to which our approaches might be applied include the sympathetic nervous system, cholesterol metabolism, and coagulation pathways. Additionally, we have highlighted the importance of sex-stratified genetic analyses for sex-dependent phenotypes. Our data will help inform the design and analysis of such future work.
We thank Margaret Stebbing, the general practitioners and research nurses, Professor John Hopper, and Twins Research Australia, Professor Graham G. Giles, and the Health 2000 Study for their contributions to subject recruitment.
Sources of Funding
This work was supported by the Victorian Health Promotion Foundation, the National Health and Medical Research Council of Australia, and the Heart Foundation of Australia. This research was facilitated through the Twins Research Australia receives support from the National Health and Medical Research Council through a Centre of Research Excellence Grant, from which Dr Scurrah is also partially funded. Dr Ellis was supported by an Australian Research Council Future Fellowship.
The Data Supplement is available at http://circgenetics.ahajournals.org/lookup/suppl/doi:10.1161/CIRCGENETICS.116.001595/-/DC1.
- Received August 11, 2016.
- Accepted March 2, 2017.
- © 2017 American Heart Association, Inc.
- Konoshita T
- Braliou GG,
- Grigoriadou AM,
- Kontou PI,
- Bagos PG
- Dimas AS,
- Nica AC,
- Montgomery SB,
- Stranger BE,
- Raj T,
- Buil A,
- et al
- Dhanachandra Singh Kh,
- Jajodia A,
- Kaur H,
- Kukreti R,
- Karthikeyan M
- O’Donnell CJ,
- Lindpaintner K,
- Larson MG,
- Rao VS,
- Ordovas JM,
- Schaefer EJ,
- et al
- Charita B,
- Padma G,
- Sushma P,
- Deepak P,
- Padma T
- Castellano M,
- Glorioso N,
- Cusi D,
- Sarzani R,
- Fabris B,
- Opocher G,
- et al
- Marciante KD,
- Bis JC,
- Rieder MJ,
- Reiner AP,
- Lumley T,
- Monks SA,
- et al
- Harrap SB,
- Stebbing M,
- Hopper JL,
- Hoang HN,
- Giles GG
- Hoffmann TJ,
- Vansteelandt S,
- Lange C,
- Silverman EK,
- DeMeo DL,
- Laird NM
- Ancelin ML,
- Carrière I,
- Scali J,
- Ritchie K,
- Chaudieu I,
- Ryan J
- Blangero J,
- Goring HH,
- Kent JW Jr.,
- Williams JT,
- Peterson CP,
- Almasy L,
- et al
- 37.↵R Core Team. R: A Language and Environment for Statistical Computing. Vienna, Austria: R Core Team; 2015.
- Abecasis GR,
- Cherney SS,
- Cookson WOC,
- Cardon LR
- Havill LM,
- Dyer TD,
- Richardson DK,
- Mahaney MC,
- Blangero J
- Storey JD,
- Tibshirani R
- Johnson AD,
- Newton-Cheh C,
- Chasman DI,
- Ehret GB,
- Johnson T,
- Rose L,
- et al
- Obama T,
- Eguchi S
- Cannell IG,
- Kong YW,
- Bushell M
The renin–angiotensin–aldosterone system (RAAS) is a key physiological control system for blood pressure and a proven therapeutic target for cardiovascular disease prevention. Variants in and around the genes encoding the individual components of the RAAS have been inconsistently associated with blood pressure and cardiovascular disease. In this article, we use family-based studies to explore the potential that interactions between these variants (epistasis) and different associations between men and women might account for some of the discrepancies in genetic analyses of the RAAS. We combined comprehensive single-nucleotide polymorphism selection and informative statistical methodology to explore sex-dependent and epistatic effects relevant to RAAS gene associations with systolic blood pressure. Three single-nucleotide polymorphisms (rs8075924 and rs4277404 at ACE and rs12721297 at AGTR1) were strongly associated with systolic blood pressure after adjustment for multiple testing. We found evidence of both sex-dependent and epistatic effects for other RAAS single-nucleotide polymorphisms, although these associations were not significant under stringent multiple testing corrections. Nevertheless, we recommend that such epistatic and sex-dependent effects should be routinely modeled and assessed in larger studies. In particular, sex-dependent and epistatic effects for the RAAS gene variants in relation to cardiovascular disease and for responses to and side effects of therapy targeting RAAS blockade merit investigation. The RAAS genes remain important candidates for cardiovascular disease and treatment effects, but their precise genetic mechanisms remain unresolved.