Angiotensinogen gene polymorphism, again?

The recent technological advances of molecular biology led to an explosive growth of genetic research in hypertension. Most of the studies have sought association between polymorphism in ‘candidate’ genes and hypertension with a case–control design, or tried to quantify the effect of one allele of the polymorphic site on blood pressure in a sample of the general population. In its simplest form, polymorphism, which is the result of a mutation, is a single base mutation which substitutes one nucleotide for another. The polymorphism at the site harboring such change is called single nucleotide polymorphism (SNP, read snip) [1]. SNPs may happen in any portion of any gene as exons (which, in this case, may or may not change the coded aminoacid: nonsynonymous versus synonymous SNP), non-coding regulatory regions (5′ UTR or 3′ UTR), introns and intergenic regions. They are defined by a frequency > 1% in at least one population and undergo stable inheritance. Their density is very high, being as many as 0.5–3 per 1000 bp [2,3]. In rare cases, SNPs may affect the phenotype. Although it is intuitively apparent that the amino acid substitutions can change the function of the protein, even a synonymous polymorphism in a coding region or a SNP in an intronic region can affect gene function by altering the stability or mRNA splicing, whereas a SNP in a regulatory region can alter the level of mRNA expression [4]. Most studies in hypertension have produced plausible evidence for the involvement of SNPs in several genes [5], even if practically all positive findings have been challenged by negative ones. Furthermore, because many explanations may be given for a negative finding, such as (i) limited statistical power; (ii) heterogeneity of the populations studied; (iii) poor distinction between cases and controls; (iv) low genotype relative risk for any of the described molecular variants; and (v) the practical difficulty in taking into account context dependency and epistasis, the picture remains quite discouraging for the non-expert in the field [6]. The large number of contradictory results, together with the continuous evolution of genetic and statistical and sampling strategies, has prompted many investigators to adopt a negative attitude for the entire approach. The very first gene that was found to be associated with human hypertension is angiotensinogen (AGT) [7]. In their seminal paper, Jeunemaitre et al. [7] not only demonstrated that AGT locus is genetically linked to essential hypertension, but also that its threonine allele of the M235T polymorphism (methionine← threonine) is associated with higher blood pressure values and higher serum angiotensinogen concentrations. To date, the article by Jeunemaitre et al. [7] has received almost 1000 citations (ISI, Web of Science), and the results of similar studies have been reviewed in three large meta-analyses concluding that the 235T allele of angiotensinogen is indeed, albeit weakly, associated with an increased risk for hypertension in Caucasians, possibly in Japanese and probably not in Africans [8–10]. Two of the reviews also confirmed an association of the 235T allele with serum AGT concentration [9,10]. One study warned of the risk that the positive associations were possibly inflated by publication bias but, when publication bias was taken into account, the role of 235T allele remained confirmed [9]. It might be expected that the story is ended and the reader may legitimately ask ‘why angiotensinogen gene again?'. In this issue of the journal, Wu et al. [11] report an association of a particular haplotype of AGT gene with serum AGT concentration in a sample of African-Americans from the Maywood Study. We will take this opportunity to provide a non-technical update regarding the new strategies and perspectives involved the study of the genetic mechanisms of hypertension. The concept of linkage disequilibrium and haplotype block Simple statistical association cannot prove that 235T allele is the mutation that increases blood pressure and AGT secretion. It may well only be in linkage disequilibrium (LD) with the causal mutation(s). LD simply indicates the non-random association of alleles in short segments of a chromosome. When LD is very high, alleles of linked markers can sometimes be used as surrogates for the state of nearby loci. The 235T allele has now definitely been excluded as the causal one and a another SNP at position −6 in the promoter, in apparent complete LD with M235T, has been proposed [12]. As several other SNPs have been proposed [13–19] (mostly in the promoter region), continued hunting makes the story become more complicated. In spite of the complexity of genetic structure for any gene, association studies on AGT, as well as on other genes, generally focus on one SNP in one gene at a time. The use of haplotypes (as a combination of several biallelic polymorphisms) is in its starting phase because two major limitations have been removed. Because no specific information is available concerning the phase for each haplotype, the most probable haplotype configuration for each subject must be reconstructed statistically. Two general types of reconstruction algorithms have been proposed. The first determines the most likely set of haplotypes based on the genotype of each individual and the distribution of genotypes in the population. Among these reconstruction algorithms are the parsimony [20], maximum likelihood [21] and Bayesian methods [22]. The second approach uses genotypes of family members to determine the most probable set of haplotypes, based on the alleles shared between specific pairs of parents and offspring [23]. This approach is generally more precise, because it can consider information about both the frequencies in the general population and transmission within specific families. Automation and cost reduction in genotyping allows far more productivity whereas the discovery that LD is not uniformly distributed along the genome allows much simpler genetic analyses. Although allele frequencies are simple to compare for a biallelic polymorphism, in the case of haplotypes, the number of possible groups of individuals sharing the same haplotype may be extremely high because it is theoretically equal to 2n where n = number of biallelic polymorphisms present in the DNA segment under investigation (Fig. 1). For example, in the study by Jeunemaitre et al. [19], 10 SNPs were identified, which means 210 = 1024 possible haplotypes. Fortunately, relatively few actual haplotypes usually account for the vast majority of alleles (five in the case of the study by Yanai et al. [15]). Why there are so few haplotypes within each block compared to the expected is still matter of debate, and this is summarized very well elsewhere [24,25]. Briefly, it appears that our genome consists of a long sequence of segments (30–50 kb long) of very low recombination, called haplotype blocks, alternate with shorter sequences of high recombination (Fig. 2). As a consequence of the block structure, every haplotype that is block based behaves as a single variation. Accordingly, each individual has two complementary forms of the haplotype-block, that are equivalent to ‘alleles’ and that jointly are equivalent to ‘genotype', in this case defined as diplotype. If, for a single SNP the possible genotype are only 3 (there can be individuals homozygous for either allele or heterozygous, for a haplotype, the possible diplotypes are equal to n×[(n+1)/2]). This simple finding adds tremendous power to genetic analysis. The haplotype block may be considered as a modern synonym of locus. Any haplotype block can contain a large number of SNPs, with many possible haplotypes, but only a few are needed to describe almost all of the population. Zhu et al. [26] recently demonstrated that the AGT gene is contained in a single haplotype block, both in a Caucasian and in an African-American population (the latter recruited in Maywood as the present one), and, in either case, only five haplotypes describe 99% of the population. In the present issue of the journal, Wu et al. [11] extend the observation by testing the association of one of the described haplotypes to hypertenson and serum angiotensinogen concentration [11]. They first completely sequenced a subsample of their population, starting 2 kb before the beginning and ending 1 kb after the end of the transcribed region of the gene, together with all the introns, for a total of approximately 15.5 kb, identifying 91 SNPs. Subsequently, they defined a binning algorithm to select the minimal number of SNPs that were indeed useful for the analysis, based on the observation that if two SNPs are in complete LD, it is enough to genotype one to infer the genotype of the other. Without a doubt, the technique is more complex but, in the end, it allows several SNPs to be aggregated in a bin. All the SNPs in a bin (called tagged SNPs) are in some way synonymous, and only one tagged SNP needs to be genotyped to gain information for the entire bin. The five major bins out of the 12 identified were studied and haplotypes constructed since statistical significance was never reached for an association of any of the bins tested individually with serum AGT concentration (note that testing a bin means testing the SNP that identifies the bin). Haplotype analysis demonstrated that one haplotype was associated with higher AGT concentration. We calculated that such a haplotype is responsible for approximately 7% of the variance of the serum AGT concentration, which is in agreement with the recently obtained estimate of 5% by Brand et al. [27] on a sample of different ethnicity, with different genetic strategies.Fig. 1: Upper panel: schematic representation of a biallelic SNP. T←C substitution in position 6, giving rise to alleles a and b. Lower panel: Schematic representation of possible haplotypes in the same DNA segment. Two more substitutions appear: A←G in position 4 and A←C in position 9. This can give rise to 23 = 8 possible haplotypes (a to h). Each haplotype may have different frequency within a population so that, most often, two or three haplotypes contain the greatest proportion of the population (95% or more), whereas some haplotypes never occur.Fig. 2: An idealized representation of four blocks of the block-like structure of linkage disequilibrium. Four blocks with regions of low haplotype diversity are separated by recombinational hot spots. For simplicity, the dimension (physical length of the DNA segment) for both blocks and hot spots have been kept constant. Lines below the blocks represent examples of the number of common haplotypes that might be present for each block. To allow comparison, seven SNPs have been hypothesized for each block. Considering that the number of possible haplotypes is 27 = 128, the four hypothetical blocks present two, three, two and four common haplotypes, respectively. This means that it is enough to genotype any of the seven SNPs of block 1 to gain complete information for use in an association study for block 1. The lower part of the plot indicates that, within each block, LD decays only gradually with distance, or not at all. Conversely, LD decays much more steeply within hot spots. Modified with permission [24].What does all this mean and where are we going? Wu et al. [11] not only confirm that AGT gene polymorphysms affect serum AGT concentration also in African-Americans, but also present a very powerful technique of genetic analysis. This result could never have been reached with pairwise analysis of all the 91 detected SNPs due to the multiple comparisons. However, the use of haplotypes is not just a statistical trick. In fact we should remember that each haplotype is made up by several base-pairs, arranged in linear sequences: some of them can be polymorphic (=SNPs) and in a haplotype-block there may be more than one SNP, say for example 2 or 3 or 4. When this is the case, all these SNPs may be contributing to increase the susceptibility to the trait: if a haplotype-block include a whole gene, these polymorphic variants (SNPs) could be present in different sub-regions of the gene itself, say the promoter, intron 1 and 3'UTR, and only their joint presence constitutes a risk factor for the trait. The results indicate that, within a chromosomal segment of the haplotype block containing the AGT gene, there is one or more SNP in LD with the haplotype associated with higher AGT concentration, and that (or those) is the causal mutation. At present, this kind of analysis is still cumbersome but should add precision to association studies in complex diseases. As usual, technology is developing quickly and methods for the complete automation of molecular haplotyping are already being proposed [28]. In spite of ‘ome’ inflation [29], the awareness that technology cannot do everything and that we still lack a better and deeper understanding of intermediate phenotypes and their relationships in the development of complex diseases, claims have already been made for the starting of the ‘human phenome project’ [30].