Meta-analysis of blood pressure and the CYP11B2 polymorphism highlights the need for better designed studies

The distribution of blood pressure in populations is almost certainly dependent on a mosaic of many genetic loci each with a minute contribution and is under the influence of multiple gene–gene [1] and gene–environment [2] interactions. Unravelling the genetic determinants of hypertension mainly rests on three approaches. First, whole-genome scans search for linkage peaks with traits of interest, and in the case of positive linkage, lead to further experimental and clinical studies to characterize potential loci of interest. Second, high-throughput genotyping using modern chip technology is like casting a net in a pond containing plenty of fish with the hope that subsequent sorting will identify the goldfish (i.e. the rare polymorphic markers carrying the breakthrough information). Finally, the candidate gene approach builds on existing pathophysiological knowledge and common polymorphisms in genes known to be involved in some aspects of blood pressure regulation. Usually, a few seminal studies raise the interest in a particular, preferably functional, polymorphism in a candidate gene. Next, large-scale case–control designs or association studies in patient cohorts or in the population at large attempt to reproduce the initial observations, but rarely generate consistent findings [3]. This is not surprising because, as outlined elsewhere [4,5], methodological issues limit the interpretation of many published studies. Key obstacles are the lack of standardization, the arbitrary dichotomization of continuous phenotypes, the inappropriate selection of cases and controls, population admixture and stratification, an insufficient sample size, and a failure to account for confounders or environmental factors. Quantitative systematic reviews, commonly known as meta-analyses, offer the possibility to provide pooled estimates across a large number of studies on a given cardiovascular trait in relation to a genetic polymorphism and to identify sources of heterogeneity in the published results. The –344C/T polymorphism in the aldosterone synthase gene (CYP11B2) ranks high among the genetic variants most frequently studied in relation to cardiovascular and renal phenotypes. In the present issue of the journal, Sookoian et al. [6] reviewed 42 studies, of which 24 were population-based and 18 hospital-based. The number of pooled reports was 19 for hypertension as dichotomous endpoint (11 225 subjects), 13 for systolic and diastolic blood pressures as continuous variables (n = 1775), 14 for the plasma aldosterone concentration (n = 2872), and eight for the plasma renin activity (n = 1428). The authors excluded heterozygotes from analysis. The bold conclusion, that –344CC homozygotes had a 17% lower risk of hypertension than their –344TT counterparts, hinged on the pooled odds ratio computed from a fixed-effect model [0.83; 95% confidence interval (CI) = 0.76–0.91; P < 0.001] across widely heterogeneous studies [6]. The odds ratio calculated from a random-effects model did not reach statistical significance (0.89; CI = 0.76–1.04; P = 0.13). In the review by Sookoian et al. [6], hypertension was a blood pressure higher than 140 mmHg systolic or 90 mmHg diastolic or treatment with antihypertensive drugs. Unlike recommendations in recent guidelines [7], untreated subjects with a systolic blood pressure of 140 mmHg or diastolic blood pressure of 90 mmHg were therefore classified as normotensive. The continuous analyses across 13 studies and 15 groups of subjects did not confirm the main conclusion because neither fixed-effects nor random-effects models demonstrated a significant difference between CC and TT homozygotes in systolic (P = 0.41 and 0.54, respectively) or diastolic (P = 0.61 and 0.85, respectively) blood pressures. Treated patients were excluded from the meta-analysis of blood pressure as continuous phenotype. In most quantitative overviews, heterogeneity is a major issue. Ioannidis et al. [3] noted between-study heterogeneity in 26 of 55 meta-analyses of genetic association across 579 studies [3]. The magnitude of the genetic effect differed significantly in larger compared to smaller studies in ten (18%), 20 (36%), and 21 (38%) meta-analyses, as assessed by tests of rank correlation, regression on the standard error, or regression on the inverse of the variance, respectively [3]. The largest studies usually yielded more conservative pooled estimates than complete meta-analyses, which included all studies. In 14 (26%) meta-analyses, the association under study was stronger in the first studies than in subsequent research [3]. Heterogeneity in studies relating blood pressure to genetic variants is to be expected. For example, blood pressure behaves as an age-related quantitative trait. Advancing age increases salt sensitivity [8,9], decreases the gain of the baroreceptor reflex [10], reduces renal perfusion and renin activity [11], and attenuates the dampening effects of the large arteries on both systolic and diastolic blood pressures [12]. In addition to age, race and sex are also important determinants of the blood pressure level and the activity of the renin–angiotensin–aldosterone axis. In the meta-analysis by Sookoian et al. [6], the frequency of the –344C allele ranged from 0.28 in African Americans to 0.31 in Japanese, and to 0.47 in Caucasians. Sookoian et al. [6] addressed heterogeneity by applying a random-effects model, which assumes that the genetic effects might be genuinely different across studies and which allows for within-study and between-study variance components. By contrast, fixed-effect models ascribe differences between study results to chance alone. Random-effects models produce wider confidence intervals when between-study heterogeneity exists, but otherwise estimates from fixed and random effects models are quite similar. In line with common practice in the literature, Sookoian et al. [6] also addressed heterogeneity by limiting their analyses to large studies with a sample size of at least 500 [13], or by the computation of pooled estimates across race and age groups [14,15]. The strategy of additionally removing studies from ethnically homogeneous groups raises concern. In any scientific analysis, deleting outliers is arbitrary and therefore controversial. The dividing line between removing true distortion and embellishing the data to fit the hypothesis is extremely thin. Whether the impact factor of a journal in which a report is published truly reflects a study's quality and whether the computation of a summary statistic for a single study dealing with a particular ethnic group is meaningful remains open for discussion. Phenotype–genotype associations without supporting evidence of underlying mechanistic pathways have little clinical relevance. The product of CYP11B2 is required for the final steps in the biosynthesis of aldosterone [16,17]. The –344C/T biallelic polymorphism affects binding of human steroidogenic factor 1 (SF1). Angiotensin II and K+, the major physiological regulators of the aldosterone production, utilize the same SF1 cis-elements to regulate the expression of aldosterone synthase [18]. In vitro, SF1 binds almost five-fold stronger to the –344C than the –344T allele [19]. However, H295R human adrenal cells, transfected with either the C or T allele in the presence of increasing concentrations of an SF1 expression plasmid, did not reveal differences in the reporter construct activity under basal conditions or after forskolin treatment [19]. Only after incubation with angiotensin II did C-allele transfected cells show a minute preferential increase (10%) in reporter activity [19]. These findings obtained in vitro cast doubt on the functionality of the –344C/T polymorphism in vivo. It is in linkage disequilibrium with other polymorphisms near or within CYP11B2[20,21] or it might only become functional through epigenetic interaction with other genes [1,22,23]. Whatever the underlying mechanism, Sookoian noticed significantly lower plasma renin activity in –344CC homozygotes compared to their TT counterparts [6], but did not find significant genetic differences in the plasma aldosterone concentration. Unlike the analysis of blood pressure as a continuous phenotype, the latter computations included patients on antihypertensive drug treatment [6]. In keeping with other studies in humans [24,25], we observed the highest aldosterone excretion rate and the lowest plasma renin activity in –344TT homozygotes [1]. Other studies demonstrated an increased frequency of the –344TT genotype or the 173Lys allele, which is in linkage disequilibrium with the –344T allele in patients with low-renin hypertension [26]. In conclusion, although not everyone might subscribe to the conclusion of the meta-analysis by Sookoian et al. [6], the Argentine team have to be congratulated for their comprehensive review of the literature. Their meta-analysis highlights the need for more efficient studies of genetic associations. Blood pressure and the activity of the renin system change with age. Cross-sectional studies, in which outcome and exposure are simultaneously recorded, have a limited capability to dissect associations of complex traits, such as blood pressure, with genetic risk factors. Notably, in our prospective population study [1], the –344CC genotype enhanced the risk of hypertension associated with the DD genotype of the angiotensin-converting enzyme gene and mutated adducin, whereas in cross-sectional analyses of the same population, high blood pressure was associated with the –344T allele. Complex phenotypes are not only under the influence of multiple genes, but also genetic determinants interact with anthropometric characteristics, environmental factors and lifestyle. For genetic variants potentially influencing blood pressure or acting through the renin–angiotensin–aldosterone system, dietary salt intake [2] and adiposity [27] should be accounted for. A detailed analysis of the influence of certain characteristics on phenotype–genotype associations requires access to the individual data of each subject [28]. Finally, in an era of robotic high-throughput genotyping, the true challenge for future studies undoubtedly lies in the labour-intensive collection of high-fidelity phenotypes. The reproducibility of these phenotypes is of key importance because, to attain sufficient statistical power in prospective studies, a multicentre approach, as used in the European [2] or Chinese [22] Projects on Genes in Hypertension, will be indispensable.