Introduction

Essential hypertension is a complex disease characterised by a sustained elevation in blood pressure with no known underlying medical or biological cause. The heritability of hypertension is thought to range from 30% to 60%, with variable clinical presentation and drug response due to multiple contributory genes, genetic/ethnic heterogeneity and environmental effects [1], [2]. Genome-wide linkage, association, and candidate gene studies in patients and animal hypertension models have identified an array of associated variants for hypertension that include sodium channels, the sympathetic nervous system and adrenergic pathways, and the renin-angiotensin-aldosterone system (reviewed in [1], [3], [4]).

Although high dietary salt intake may be associated with hypertension, blood pressure response to salt in individuals is heterogeneous and most likely due to the genetic background of individuals. Thus, in some cases, hypertension appears to be salt-sensitive and reduction of dietary salt results in a significant decrease in blood pressure, and conversely salt resistance if there is little or no change. The underlying genetics of this type of hypertension, however, have not been fully elucidated [5]. Evidence to date suggests the involvement of variants of sodium channels, renal ion channels, the mineralocorticoid receptor, and related proteins affecting synthesis and levels of mineralocorticoids (reviewed in [2]). Additional proposed disease variants include those of the α-adducin gene (increasing sodium/potassium pump activity), the glucagon gene, (reducing cAMP production and impairing natriuresis), the serum- and glucocorticoid-regulated kinase (SGK1) gene (enhancing aldosterone–induced expression of sodium channels) [6], and genes involved in arachidonic acid metabolism [7] (reviewed in [8]). Additionally, variation in the Cyp4A10 gene, a member of the cytochrome P450 family involved in the functional regulation of sodium channels, may cause salt-sensitive hypertension [9].

Salt-sensitive hypertension appears to be more prevalent in people of indigenous African origin, for example it was found in 73% of African-American hypertensive patients when compared with 51% of the hypertensive population and 26% of normotensive individuals [10], [11]. Of interest, it has also been shown that Caucasians secrete a water load more rapidly than indigenous Africans, perhaps due to a difference in renal water handling. This may also have an impact on occurrence of salt-sensitive hypertension in the two populations [12]. In a study from South Africa, suppressed plasma renin activity (an index of salt sensitivity) was significantly lower in both normotensive and hypertensive indigenous African patients compared to Caucasians despite comparable sodium intake. This was particularly evident in normotensives where 14.9% of Caucasians and 70% of indigenous Africans have a plasma renin activity <1.1 ng/ml/hr [13]. We have previously shown a relationship of the R563Q mutation of the SCNN1b gene with hypertension, but the prevalence of this SNP in the hypertensive population is 5% and <1% in normotensives and cannot account for the suppressed plasma renin activity in 70% of indigenous Africans [14].

Although it can be shown that certain risk factors and diseases vary as a function of ethnicity, it is difficult to translate this effectively into therapeutics until the underlying genotypes are better understood. The ultimate aim would be to optimise treatment based solely on genotype and regardless of ethnicity. However in the first instance, population diversity and ethnic differences in disease risk present an avenue of research towards understanding the underlying disease genetics [15].

Computational identification of most likely etiological genes can facilitate more efficient identification of genes of diagnostic, prognostic and therapeutic value by presenting strong candidates for future empirical research. The aim of this study, therefore, is to use computational approaches to prioritize and present most likely disease gene candidates for salt-sensitive hypertension, for further empirical analysis by translational researchers. In this study, we have used gene functional annotation associated with hypertension and salt-sensitivity to predict and rank novel candidates for salt-sensitive hypertension. Our analysis uses a computational approach that combines text-mining of PubMed abstracts for potentially relevant genes, and extensive mining of gene annotation data. We have then analysed Affymetrix single nucleotide polymorphism (SNP) data in a total of 126 indigenous South African individuals, spanning five distinct indigenous South African population groups, for top-scoring candidate genes from the computational analysis. We have compared allele frequencies at all SNPs assayed in indigenous Southern Africans, with those in the Caucasian population assayed in the HapMap project [16]. We have also looked for variation in the copy number of candidate genes between the different population groups. Our intention is to prioritise by computational approaches most likely candidate genes for salt-sensitive hypertension; and to prioritise further those candidates that are sufficiently different between the South African and Caucasians to potentially underlie susceptibility to salt-sensitive hypertension in indigenous Southern African populations.

Methods

Selection of candidate genes by Gene Ontology (GO) annotation

Data was accessed from the Ensembl database (Ensembl_mart_47) [17]. GO annotations were selected using AmiGO ([18], www.geneontology.org) to reflect a range of pathways and functions. An overview of the functions and pathways included is shown in Table 1 (1–9). The full descriptions of all GO terms used are shown in Supplementary Data File S1.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Criteria for terms used to identify candidate genes.

https://doi.org/10.1371/journal.pone.0012989.t001

Results

Identification of candidate genes

In total, 2057 unique genes were included across all gene lists, and were used as the primary set of candidate genes. Each of these was then assayed for the various characteristics (see Table 1), by both text-mining and GO annotation, and assigned a cumulative score, shown in Table 2. The genes were ranked by this score, and the top scoring candidates prioritised as most likely candidate disease genes (scoring for all candidates shown in Supplementary Data File S3). The top scoring candidates were curated to exclude any spurious results. The top ranking genes were PTH - Parathyroid hormone precursor and AGTR1 - Type-1 angiotensin II receptor. A selection of additional likely candidates was made from the genes ranked in the top twenty positions, as shown in Table 3 (full set of top candidates shown in Supplementary Data File S4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Scoring of candidates selected for further analysis (full data shown in file S2).

https://doi.org/10.1371/journal.pone.0012989.t002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Candidate genes selected for further analysis.

https://doi.org/10.1371/journal.pone.0012989.t003

Comparative analysis of SNP frequencies in candidate genes, between populations

A total of 1079 SNPs were selected from the Affymetrix data as associated with the prioritised candidate genes, falling either within the genes or within 100 000 bp either upstream or downstream of the gene (Supplementary Data File S2) (the SCNN1b gene was not included in this analysis as it was not one of the prioritised genes). Of these, 303 had allele frequencies that were significantly different between Caucasian and all indigenous South African populations at a threshold of p<0.05. There were 159 SNPs that had allele frequencies that were significantly different at a threshold of p<0.001. The categories and numbers of SNPs selected are shown in Table 4. Additionally, the numbers of SNPs falling within the coding region of the gene and not the flanking sequence are shown. These were distributed across the candidate genes as shown in Table 5, and results for a selection of genes are shown in Figure 1 (SNP distribution in all genes and flanking sequences are shown in supplementary file S5). The distribution of SNPs within the genes and excluded from the flanking sequence are also shown.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Examples of distribution of SNPs with significantly different allele frequencies when comparing indigenous Southern African and Caucasian populations (all genes shown in supplementary data file S5).

Gene exon structure (bold vertical bars), chromosome and base pair position (horizontal axis) are shown. Each black cross represents a SNP with allele frequencies that differ between Caucasian and South African populations at the level of significance indicated, and the grey crosses represent all SNPs analysed. A. EDNRA (many SNPs within the gene) B. AGT (a few SNPs within the gene, most SNPs in flanking regions); C. ANG (no SNPs within the gene); D. PTH (few SNPs assayed within the gene).

https://doi.org/10.1371/journal.pone.0012989.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. The categories and numbers of all SNPs selected.

https://doi.org/10.1371/journal.pone.0012989.t004

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. The numbers of SNPs selected for each candidate gene.

https://doi.org/10.1371/journal.pone.0012989.t005

Discussion

The high ranking genes in the candidate gene list included several previously identified causative genes for hypertension. The top scoring AGTR1 gene has been extensively implicated in controlling blood pressure and volume in the cardiovascular system and pharmacologic agents that interrupt the action of angiotensin II action by antagonizing the AGTR1 receptor are highly successful in the treatment of angiotensin II-dependent hypertension [28]. AGT, AGTR1, ACE and REN are all members of the renin-angiotensin-aldosterone pathway (as shown in www.pharmgkb.org, the ACE inhibitor pathway). ACE inhibitors and AGTR1-blocking drugs are already extensively exploited as therapy for hypertension. Given that these genes are well annotated and form the foundation of the characteristics used to select new candidates, it is unsurprising that they are highly scored in the analysis and act as an internal control showing that the selective process is accurate. The same is true of NPPA- Atrial natriuretic peptide [29], although interestingly this gene was not selected by atrial natriuretic peptide-specific GO annotation, but rather by association in PubMed abstract with terms including sodium reabsorption/excretion, low renin and aldosterone levels.

Despite our previous description of the association of hypertension with the R563Q mutation, the SCNN1b gene was not identified as a good candidate gene. It only fulfilled one of the characteristics that we used to prioritise candidates (“sodium transport”), and therefore was not assigned a high priority as a candidate gene. The focus of this computational analysis is to synthesise various sources of information to highlight less obvious good candidates, such as PTH, rather than selecting more obvious candidates such as SCNN1b based only on a one-dimensional analysis of the direct link between gene and disease. The limitations of detecting genetic associations with hypertension must also be noted, whereby detection of common variants with small effects is more likely than detection of a rare variant dominant effect like the R563Q mutation. [30].

PTH is also top-scoring alongside AGTR1. It has a less defined role in essential hypertension, which makes it an interesting novel candidate for salt-sensitive hypertension. Several lines of research support a potential role for PTH in the pathogenesis of salt-sensitive hypertension. It has been proposed that the level of PTH may be raised as a secondary response to hypertension [31], [32], as a response to dysregulated sodium excretion from the kidney. It has, however, also been shown that administration of PTH to healthy volunteers under normal glycaemic conditions, in the presence of controlled circulating insulin levels, results in a significant and consistent increase in blood pressure and concurrently in ionised calcium concentration. The increase in blood pressure was strongly correlated to changes in calcium ion concentration [33]. Similarly, hypercalcaemia and hypertension were observed in response to administration of supraphysiological levels of PTH [[34]. Further observations have been made previously that in uraemic patients with secondary hyperparathyroidism, in patients with secondary hyperparathyroidism, and in genetically pre-hypertensive patients a positive correlation is found between blood pressure and PTH/Calcium ion concentrations [35], [36], [37]. In humans salt-induced increases in blood pressure have been linked to increases in intracellular free calcium [38]. The calcium binding to the Ca-sensing receptor expressed in the thick ascending limb inhibits the Na,K, 2CL cotransporter, which decreases both sodium reabsorption and thus influences blood pressure [39]. Finally, the prevalence of hypertension is known to be high in patients with primary hyperparathyroidism [40].

Additionally, PTH levels are found to be elevated in African American individuals when compared to those of Caucasian descent [41], which further supports a role for PTH as an underlying causative gene for salt-sensitive hypertension in indigenous South African patients. However it is possible that this is an indirect association as these elevated levels may be the result of vitamin D insufficiency, due to less sun-stimulated vitamin D synthesis in dark-skinned people [41], [42].

Finally, in a prospective study, Taylor et al. examined the association between plasma intact PTH levels and the risk of incident hypertension in 481 men without baseline hypertension, and suggested that plasma levels of intact PTH (even within ranges considered normal) are positively and independently associated with a higher risk of incident hypertension. These authors proposed that PTH could potentially serve as a novel target for the prevention of hypertension [43], supporting its ranking as a primary candidate gene in our study. This may also have implications for therapy as these patients could potentially respond better to calcium channel blockers.

Of the additional high-scoring candidates, some have been previously directly associated with hypertension and salt-sensitivity. Candidates EDNRA, EDN1 and EDNRB function together in a regulatory pathway with endothelin-converting enzyme ECE1, which has been implicated in essential hypertension. Additionally, Rothermund et al. proposed a role for EDNRA and EDNRB in salt-sensitive hypertension [44]. Our study suggests that these genes are good candidates and should be revisited for a potential role in salt-sensitive hypertension. The association between the insulin gene, INS, and hypertension has been well documented. There is extensive cross-talk between insulin and angiotensin II regulating the metabolic and circulatory systems, with effects at both the extracellular level whereby ACE controls angiotensin II synthesis and interferes with insulin signalling through angiotensin II regulation and accumulation of bradykinin; and at the intracellular level whereby signal transduction pathways are affected (reviewed in [45], [46]). Clinically, the association between insulin resistance and pre-hypertension has been demonstrated [47], and it has also been proposed that insulin promotes renal sodium retention ([48] and references therein). Therefore insulin is a good candidate for salt-sensitive hypertension, as it has been implicated in both hypertension and sodium retention.

Other high-scoring candidates have not been shown previously to have a direct role in hypertension, although they do have roles in vascular and heart physiology, and have been associated with other diseases. HCN2 and HCN4 are hyperpolarisation-activated cation channels of the HCN gene family, and contribute to spontaneous rhythmic activity in both heart and brain. Expression of HCN4 is predominantly in the heart, as well as the thalamus and the developing central nervous system. Mutations in HCN4 have been associated with the cardiac disease, sinus bradycardia (OMIM #163800). The gene product of ANG, the angiogenin gene, is an inducer of neovascularization, and the RNase activity of ANG is important for its angiogenic activity. It is induced by hypoxia to elicit angiogenesis and is expressed in motor neurons; mutations in ANG are associated with the neurodegenerative disease amyotrophic lateral sclerosis in some populations (OMIM # 611895).

The computational approach we have used has selected both hypertension-associated and novel candidate genes for salt-sensitive hypertension. We have used a variety of gene characteristics and functional annotations to select most likely candidates for this disease. Although text- and data-mining do rely on existing information to inform the results, such computational approaches are able to synthesise huge volumes of existing data both through the biomedical literature, and gene annotation in genomic databases. The systematic analysis and categorisation of such volumes of existing data allow novel connections and inferences to be made computationally. The scoring system is based on the premise that a candidate gene selected by multiple sources of evidence is a better candidate than one selected by fewer lines of evidence (as previously demonstrated in [49], [50], [51]), and therefore allows objective weighting of the value and extent of such inferences. In general, the ability of our method to identify a number of genes already implicated in hypertension indicates that the methodology we have used is sound, and that the characteristics by which we have prioritised candidates are appropriate. Our method, however, has also highlighted novel candidates for which disparate existing knowledge recorded in the biomedical literature and in functional gene annotation has not previously been collated to support a role in salt-sensitive hypertension.

The sample set of indigenous South Africans is relatively small, and because of this, the information about allele frequency and copy number variation is presented here as preliminary data. Because of the high incidence of hypertension in indigenous South Africans (for example up to 25% in urban Zulu individuals [52]), of which up to 50% show suppressed plasma renin activity, an indirect measure of salt sensitivity [13], it is however likely that a genetic predisposition to this disease is represented in a fair proportion of the population, and that the evidence for this can be found in our sample set when compared to Caucasians. Such information can be invaluable to the researcher designing studies to identify the underlying genetics of salt-sensitive hypertension: as with all candidate disease gene studies, we present here a prioritisation of alleles for further investigation – only further empirical analysis of appropriately sized sample sets would be able to confirm the association of particular alleles with salt-sensitive hypertension.

It is difficult to make a direct comparison between the numbers, positions and types of SNPs identified for each candidate gene, as the gene sizes and total number of SNPs are so varied. The distribution of SNPs within candidate genes can be described in several approximated categories, and examples are shown in Figure 1. In some genes, there are no selected SNPs falling within the gene itself despite many SNPs being assayed in those regions, and those in flanking sequences are fairly distant from the gene – these are AGT, ANG, HCN4 and NPPA. In some, a few SNPs are selected within and close to the gene, although many more SNPs were assayed than selected in these regions – these are REN, ACE, EDNRB and INS. We propose that these candidates that show fewer differences in allele frequency are less likely to underlie the salt-sensitive hypertension that is seen in indigenous Southern African populations. In a few candidates, many of the assayed SNPs were selected falling within or close to the gene – these are EDNRA, EDN1 and AGTR1. Finally, in a few of the candidates, few SNPs were assayed within or close to the gene, although these were generally selected as having significantly different alleles between the populations – these are HCN2, and the primary candidate PTH. The two latter categories contain candidates that appear to have substantial differences in allele frequencies between Caucasian and the indigenous Southern African populations, and as such are more likely to contain variants that are responsible for the salt-sensitive hypertension that is prevalent in the indigenous Southern African population. Our analysis of copy number variation did not suggest that the candidate genes investigated here have any alteration in copy number at the gene level in either of the population groups. Copy number variation of these genes is therefore unlikely to underlie the salt-sensitive hypertension that is prevalent in the indigenous Southern African populations (although it is interesting to note that the heatmaps generated for the genes AGTR1 and INS do show evidence of population-specific variation at sites within the genes - see supplementary data file S6).

The selection of top-scoring candidates in this study is inevitably affected by some inherent bias: appropriate genes that have more extensive annotations are more likely to be selected in the study than those that have not yet been analysed, so it is more difficult to select entirely novel candidates. However, we have aimed to combine existing knowledge about hypertension and salt sensitivity to draw new conclusions about candidate genes for salt-sensitive hypertension. Computational analysis thus allows the synthesis of existing information to make novel predictions.

Due to the population-specific nature of salt-sensitive hypertension, we propose that the SNPs in the prioritised candidate genes that show significantly different allele frequencies between Caucasian and indigenous African populations provide good targets for further clinical and empirical research. We provide here a list of good candidate genes and alleles related to salt-sensitive hypertension that have been selected and prioritised by computational methods. These prioritized genes may have a significant contribution to the occurrence of salt-sensitive hypertension, and thus warrant further investigation by translational researchers.

Supporting Information

Acknowledgments

The Medical Research Council of South Africa, Affymetrix (USA), and the Centre for Proteomic and Genomic Research (notably Dr Reinhard Hiller) at the University of Cape Town, are acknowledged for their support of the generation of the SNP6 data on the indigenous Southern African populations.