Hypertension is a major risk factor for vascular and renal disease and thereby contributes to substantial worldwide morbidity and mortality [1]. Although blood pressure (BP) has been shown to have significant heritability (30-60%) [2], the genes conferring susceptibility to hypertension remain largely unknown. A recent genome-wide association study (GWAS) analysed 79,447 single-nucleotide polymorphisms (SNPs) in the American Old Order Amish, a closed population descended from a small number of common founders who emigrated from Switzerland in the early 1700s who have a relatively homogeneous lifestyle [3]. Their initial genome-wide screen of 542 subjects from the Amish Family Diabetes Study [4], in which families were ascertained through a proband with type 2 diabetes, identified a cluster of SNPs in the STK39 (serine threonine kinase 39) locus on chromosome 2q24.3 that were associated with BP (P = 8.9 × 10^-6 to 9.1 × 10^-5) [3]. Although the P-value did not reach the conventionally accepted threshold for genome-wide significance [5], the inclusion of additional data from an Amish and four non-Amish Caucasian cohorts led to a significant result (P = 1.6 × 10^-7) in the combined dataset of 7,125 individuals. The replication cohorts included a further population of 2,842 individuals from a case-control study of diabetes and three smaller groups not selected for diabetes.

The STK39 gene encodes the SPAK (Ste20-related proline-alanine-rich kinase) protein which interacts with ion cotransporters involved in salt transport and osmotic cell volume regulation, including the thiazide-sensitive and loop diuretic-sensitive cotransporters involved in renal salt excretion [6, 7]. The STK39 gene contains 18 exons spanning approximately 300 kb on chromosome 2q24.3; SNPs associated with BP by GWAS were located within introns 1-8, with no coding or splice variants identified by sequencing [3]. Using transfection experiments with luciferase reporter constructs Wang and colleagues demonstrated that alleles of one SNP (rs35929607), located in a conserved region of intron 2, altered expression of the reporter constructs in vitro. This SNP was in linkage disequilibrium (LD) with the SNPs identified by GWAS, suggesting a possible functional mechanism for the observed association.

Despite these results, other large GWAS and meta-analyses involving up to 71,225 Caucasian individuals have failed to identify association between STK39 SNPs and BP that satisfies genome-wide significance thresholds [8–10]. However, such data do not exclude an effect of this locus on BP, since the statistical threshold for replicating an association differs from the threshold for genome-wide 'discovery'. Differences in the actual SNPs genotyped, patterns of LD, and phenotypic variation between populations might contribute to the lack of genome-wide significance in these studies, compared to the GWAS by Wang and colleagues. One recent GWAS looked specifically for evidence of correlation between STK39 SNPs and BP in 1017 African American subjects and provided some supportive evidence for an association [11]. SNPs in the STK39 region did not reach genome-wide significance, but the number of significantly associated SNPs at a nominal p threshold of p < 0.05 was somewhat higher than expected by chance alone (9/136 for systolic BP and 33/136 for diastolic BP, compared to 7/136 expected by chance).

We aimed to test whether the STK39 SNPs identified in the report by Wang and colleagues [3] were associated with BP in a large family-based Caucasian cohort ascertained through probands with essential hypertension [12, 13]. This cohort has a higher average BP (134 v 128 mmHg systolic) and lower proportion of diabetics (2.5 v 22.2%) compared to the Amish population used for the initial genome-wide study by Wang and colleagues, but has 24-hour ambulatory BP recordings which provide a more reproducible and precise measure of true BP, with better prediction of cardiovascular events, compared to clinic BP measures [14, 15]. Selecting the same SNPs typed by Wang and colleagues and using a cohort enriched for hypertension with ambulatory BP data should increase the power to detect effects associated with these variants in our population.

We also aimed to confirm whether the SNPs have cis-acting effects on STK39 expression in vivo using allelic expression imbalance (AEI). This technique measures the relative amount of transcript arising from each allele in individuals heterozygous for a transcribed polymorphism. Unequal amounts of transcript from each allele indicate the presence of cis-acting effects on gene expression [16]. The allelic expression ratio can be used to quantify the degree of AEI and test whether particular SNPs are associated with cis-acting effects, even in the absence of LD between the transcribed and cis-acting SNPs [17]. The degree to which data from transfection studies in cell lines represents the in vivo situation in complex human tissues is questionable [18]. AEI assesses alleles expressed in their native environment, and because comparison of alleles is made within samples, trans-acting influences and experimental insults are the same for each allele which maximises the sensitivity for detecting cis-acting effects [16].

The STK39 SNPs reported to be associated with BP and hypertension by Wang and colleagues [3] did not show a significant association with BP in our British Caucasian families. Other published GWAS and two large recent meta-analyses did not identify associations achieving genome-wide significance levels for STK39 polymorphisms and BP, despite powerful analyses of combined cohorts involving 34,433 and 71,225 Caucasian individuals which successfully identified multiple other susceptibility loci [9, 10]. Although failure to achieve genome-wide significance levels does not exclude an effect of this locus on BP, any effect of sequence variation at STK39 on BP is too small to be detected in our data and is perhaps something particular to the Amish population. We also found no association with BP for the chromosome 9p21.3 SNP rs4977950 which was the top signal in the GWAS by Wang and colleagues (P = 9.1 × 10^-8), and the only SNP to achieve significance at the genome-wide threshold in their study.

Inadequate power is an unlikely explanation for the lack of association in this study as the maximum plausible genetic effect for the typed SNPs on BP phenotypes, which we calculated based on the observed mean effect and 95% confidence intervals, was low in all cases and our data exclude an effect as large as the 3 mmHg increase in systolic BP per copy of the risk allele reported by Wang and colleagues in the Amish population. Families in the present study were selected through a hypertensive proband and had a higher proportion of hypertensive individuals and wider distribution of BP values; this should increase the power to detect an association with BP phenotypes compared to the cohorts in the study by Wang and colleagues, which were not selected for BP. The use of 24-hour ambulatory BP recordings in our study provides a more reproducible assessment of "usual" BP and reduces misclassification due to "white-coat" or masked hypertension that may occur with isolated clinic measurements [14, 15]; this should also reduce the noise and increase the power of our study. The heritability of night-time BP measures was significantly higher than clinic BP measures, which should increase the power for detecting genetic effects. It is a frequently observed phenomenon in genetic epidemiology that studies first reporting a novel association often show a more extreme odds ratio than subsequent replication studies [22]; if the effect size estimated by Wang et al is an upwardly biased estimate this might explain the negative association in our study and the other reported GWAS. The effect due to STK39 alleles observed by Wang et al was strongest in the Amish population and substantially weaker in the other Caucasian cohorts tested in their study. Our data suggest that the effect of the typed STK39 SNPs on BP is at most modest in a UK Caucasian population. If the reported association was caused by LD between the typed SNPs and another functional variant, then weaker LD in other populations, compared to the Amish population, could account for the results; this is particularly pertinent since closed founder populations are expected to exhibit extensive LD. Heterogeneity between the populations, including differences in population history, recruitment strategy, and phenotypes (such as the proportion of diabetics and hypertensives) might also contribute to the differences in genetic associations we have observed. Our analysis was performed using an additive model because this was the model used to report the most significant overall association by Wang and colleagues.

We have demonstrated that SNPs in the STK39 gene correlate with in vivo allelic expression of this gene in peripheral blood cells. This corroborates and extends the findings from reporter gene constructs in the study by Wang and colleagues. Using a luciferase assay in HELA and HEK293 cell lines they examined the effect on transcription of two SNPs (rs12692877 and rs35929607) that are in conserved elements and which were in complete LD with two of the SNPs associated with BP in their GWAS in the Amish population (rs6749447 and rs3754777 respectively). They demonstrated that the G allele of rs35929607, which was the allele reported to be associated with increased BP, was in isolation associated with a greater than two-fold increase in transcriptional activity compared to the A allele of this SNP, or either allele of rs12692877. Such in vitro studies have limitations since results depend on the constructs that are used and expression is considered outside of the normal chromatin and cellular context, which may not reflect true expression in complex tissues in vivo [16, 18]. Allelic expression analysis allows in vivo assessment of RNA transcripts in their native environment and regulatory context, and by controlling for trans-acting influences has high sensitivity to detect cis-acting effects. We found a stronger association between genotype and allelic expression for rs6749447 than for rs35929607, with the G allele of rs6749447 associated with a 13% increase in expression relative to the T allele. However, if the effect of rs6749447 is not accounted for, then our allelic expression analysis found that the G allele of rs35929607 was significantly associated with increased expression. This is consistent with the in vitro data from Wang and colleagues, although the magnitude of the relative increase in our study was lower at 7%. The difference in the magnitude of effect we observed compared to the in vitro transfection studies is not surprising in view of the fact that in vivo expression is likely to result from the interaction of multiple elements that modulate expression, rather than just a single SNP.

AEI can only be assessed in individuals who are heterozygous for a transcribed polymorphism in the gene of interest - there were no suitable transcribed SNPs in STK39 in Caucasian populations and in the African cohort the number of suitable heterozygotes was relatively small because of a low minor allele frequency (7%) at the transcribed SNP rs1061471. However, measurement of allelic expression ratios within samples is very sensitive for detection of cis-acting influences since trans-acting and experimental factors are identical for each allele (because comparisons are made within samples); this allows significant cis-acting effects to be detected in relatively few samples, as demonstrated in this study. Sensitivity of the technique for the detection of significant effects in equivalent sample sizes has been previously demonstrated [23, 24]. Populations of African-descent show on average smaller regions of high LD and greater heterozygosity compared to Caucasian populations [25, 26], and therefore using a South African cohort increases the sensitivity to separate the effects of different SNPs on expression. This trans-ethnic approach using populations of African descent has been shown to improve the fine mapping of functional polymorphisms associated with quantitative traits when strong LD between markers limits resolution of the functional locus in Caucasian populations [27]. In the case of STK39, using a population of different ethnicity allowed allelic expression to be measured in the absence of transcribed polymorphisms in the Caucasian population. Correlation between STK39 SNPs and expression of this gene does not necessarily mean that these SNPs will influence BP and the finding of an association with STK39 expression in our study is therefore not at odds with the lack of association with BP we observed.

A limitation of the present study is that expression was only tested in white blood cells, rather than in a tissue of potentially greater relevance to BP regulation. However, many cis-acting influences on gene expression are expected to be the same in different cell types [28], although tissue-specific differences have been described [29]. In the case of STK39 this approach is supported by the fact that rs35929607 genotype correlated with expression in immortalised cell lines derived from cervical tumour cells (HELA) and embryonic kidney cells (HEK293) in the study by Wang and colleagues, as well as with expression in blood in the present study. However, the effects may vary in other tissues. We have analysed the SNPs influencing expression in a South African cohort of mixed ethnicity, but the SNPs associated with expression may vary in different populations due to differing allele frequencies and LD patterns. Although the associations of SNPs with STK39 expression that we observe are highly statistically significant, the biological significance of these findings is less certain, since we cannot say what impact such an effect on expression has on disease risk. However, even small differences in gene expression due to genetic factors that are present throughout an individual's lifetime could contribute to differences in common late-onset phenotypes such as hypertension and, as mentioned above, the effects may be greater in tissues related to disease.

Future studies will be necessary to investigate the association of other SNPs at the STK39 locus with BP, or to determine whether rare mutations at this locus contribute significantly to population blood pressure variation, as has been shown for other genes implicated in hypertension causation [30].

STK39 expression is modified by polymorphisms acting in cis, but there is no evidence that these SNPs affecting STK39 transcription are associated with BP in a British Caucasian cohort.