Defining regulatory polymorphisms

The dissection of genetic factors defining our individual susceptibility to complex disease traits is being tackled across the field of clinical medicine, offering the promise of novel insights into disease pathogenesis and therapeutic targets together with the tailoring of management to the individual patient. However, resolving the genetic factors underlying complex diseases has proved problematic, with typically multiple genes involved of individually modest magnitude, which together only form part of a multifactorial disease process [1, 2]. For these reasons linkage analysis has met with limited success, and while a genomic region may be localised, fine mapping typically proves problematic. Association studies have proved useful, often based on a candidate gene approach, with whole-genome association studies now becoming feasible [3]. The premise of such genetic analysis is that functionally important genetic variation results in differing clinical phenotypes. As discussed below, the challenges of mapping genetic susceptibility loci are compounded by difficulties in defining specific functional polymorphism(s) at an experimental level. At present it is typically left unresolved as to whether a disease-associated polymorphism is itself functionally important or acting only as a marker for a coinherited, perhaps as yet unidentified, genetic polymorphism.

Classification of variation

DNA sequence polymorphisms are usually defined as variation present at greater than 1% frequency in the population. The most common are single nucleotide polymorphisms (SNPs) in which one of the four possible nucleotides in the DNA sequence is substituted by another, occurring on average every 800 nucleotides across the genome. Other sequence polymorphisms include deletions and insertions of one or more nucleotides, rearrangements and repeating sequences which may be short tandemly repeated motifs of one to six nucleotides (microsatellites) or longer repeating ‘minisatellites’. The majority of DNA sequence polymorphisms are of no functional importance, although some alter the structure of the resulting peptide through, for example, amino acid changes [4]. Such coding polymorphisms have classically been implicated in monogenic Mendelian disorders with their consequences typically predictable and amenable to testing. In contrast, regulatory polymorphisms occurring outside exonic regions have long been postulated to be important modulators of gene expression and evolutionary change [5], but there is only now growing evidence that this is the case [6].

Regulatory polymorphisms can be classified into two groups. The first are cis-acting, in other words acting on the copy of the gene present on that allele and typically present in or near the locus of the gene that it regulates. This may arise, for example, through the sequence change occurring in a regulatory DNA binding site and altering the affinity with which a regulatory protein is recruited and hence how the gene is expressed. Alternatively, the regulatory polymorphism may be trans-acting, a polymorphism in one gene affecting the expression of another gene. There is evidence from differences in strains of mice and yeast that trans-acting loci were responsible for much of the observed differences in expression [7, 8, 9], but in the context of complex disease susceptibility work has focused on potential cis-acting variation and forms the basis of this review.

Evidence for regulatory polymorphisms

If such regulatory polymorphisms exist, differences in expression of genes within and between populations would be expected. Variation in genome-wide gene expression has been associated with phenotypic variation in a number of different organisms, notably budding yeast. A genetic linkage analysis of global expression patterns in a cross between two strains of budding yeast demonstrated heritable differences in expression [9]. Among naturally occurring isolates of Saccharomyes cerevisiae, population genetic variation is correlated with phenotypic variation [10, 11]. Differences in gene expression have been associated with phenotypic variation in Drosophila development [12], within and between populations of teleost fish [13] and primate species [14], and in human lymphoblastoid cell lines [15]. Here variation in gene expression was observed between lines established from unrelated and related individuals, with evidence of familial aggregation [15]. Moreover, this variation in expression has recently been used as a quantitative trait to carry out linkage mapping at a genomic level in defined pedigrees of lymphoblastoid lines [16]. These data suggest that only a minority of effects are likely to be operating in cis, and that a number of regions show strong linkage to expression of many genes, possibly representing hotspots of transcriptional regulation. For some putative cis-regulators, typing additional SNP markers confirmed the association at a population and familial level [16]

There is growing evidence that allele-specific differences in gene expression (Fig. 1) occur in autosomal genes which do not show genomic imprinting (imprinted genes being the small number of genes in which allelic expression is dependent on parent of origin) [17]. The finding of allele-specific differences in transcript abundance among non-imprinted genes provides important direct supporting evidence for the existence of regulatory polymorphisms. If such differences are arising from cis-acting regulatory polymorphisms, we would expect allele-specific effects to be heritable, and there is evidence that this is the case [18, 19]. Allelic differences in expression appear context specific, for example, with regard to tissue-type, and to be of modest magnitude (typically 1.5- to 2.0-fold) [18, 19, 20, 21]. At present it is unclear how common allele-specific gene expression is across the genome, ranging from 6% of genes in a murine study [20] to approx. 20% in human lymphoblastoid cell lines [18, 19].

Fig. 1
figure 1

Investigation of regulatory polymorphism: allele-specific analysis of mRNA. Strategy: This approach uses the presence of an exonic transcribed SNP (shown here as G/A) to resolve the allelic origin of transcribed mRNA [18]. By analysis of mRNA from cells derived from an individual heterozygous for the marker SNP, an internally controlled system is established in which relative allele-specific gene expression can be estimated. Uses: Allele-specific quantification of mRNA is a useful in vivo approach to resolving functionally important haplotypes using transcribed marker SNPs. It provides a direct assessment of the relative abundance of allele-specific transcript in a natural chromosomal context in which the normal regulatory machinery and chromatin environment are operating. Limitations: The assay requires accurate and sensitive quantification of relative transcript abundance, typically based on primer extension methods. A major limitation is that for many genes and for the majority of haplotypes no exonic marker is present to allow resolution of transcript origin. Some information may be achieved by using intronic SNPs to study relative expression of unspliced RNA; a further approach is to use a different indirect measure of gene expression, namely phosphorylated Pol II loading by haploChIP in living cells [80]. The allele-specific density of Pol II loading can be used in the same way as transcript abundance except that as the Pol II is being measured in situ by crosslinking it to DNA, any SNP marker can be used within 2 kb 5′ or 3′ to the gene including promoter and 3′UTR SNPs which considerably expands the number of haplotypes that can be interrogated

Full size image

The elucidation of allele-specific gene expression using SNP markers to discriminate transcribed RNA should prove a powerful approach in screening for regulatory polymorphisms. It is important to note, however, that it serves to identify functionally important haplotypes (the coinherited polymorphisms on an allele) rather than specific polymorphisms, unless sufficient complexity of the underlying haplotypic structure can be resolved so as to fine map observed differences down to a specific region or individual polymorphism. The consequences of regulatory polymorphism are highly context dependent, depending, for example, on the cell type, prevailing environmental conditions and exogenous stimuli. This is compounded by the effects of covariation at other genomic sites, both within and without the locus. Such effects mean that the logical approach of investigating the effects of putative regulatory polymorphisms by correlating the levels of protein product derived from the gene between cells or individuals of differing genotype (Fig. 2) is often confounded.

Fig. 2
figure 2

Investigation of regulatory polymorphism: assays of secreted protein. Strategy: An intuitive approach is to compare levels of protein produced from the gene of interest in individuals of differing genotype, for example, homozygous AA or BB or heterozygous AB. Uses: This approach can be highly informative where sufficient numbers of individuals are assayed on multiple occasions using appropriate controls to minimise confounding by environmental or experimental factors. Limitations: The approach is potentially confounded at many levels including environmental factors and other variables affecting the levels of expression of a gene between individuals such as differences in receptor-ligand interaction and signal transduction as well as translation and posttranslational effects. Genetic variation on the compared haplotypes may confound interpretation of the differences seen with the chosen marker SNP

Full size image

Reporter gene assays

The most widely used experimental tool to interrogate the significance of specific putative regulatory polymorphisms are reporter gene assays [22] in which cells are transiently transfected with allele- or SNP-specific promoter constructs (Fig. 3). The results of reporter gene experiments are highly context dependent, in terms of both the biological system used and reporter gene construct design. A recent analysis of 144 functionally significant human genetic polymorphisms analysed by reporter gene assays in physiologically relevant cell lines revealed most were in the proximal regulatory promoter regions of genes, but over one quarter were located more than 1 kb upstream or 3′ to the transcriptional start site [23]. Overall an equal number of examples were found of effects of polymorphic activator or repressor binding and analysis of ancestral gain or loss of transcription factor binding was consistent with this. Notable among classes of transcription factor implicated as showing polymorphic modulation of allelic binding were USF, Sp1, nuclear factor (NF) κB, GATA and Oct-1, most often demonstrated in vitro by gel shift assays (Fig. 4). Overall Rockman and Wray [23] estimate that humans are heterozygous at more functional cis-regulatory sites than at amino acid positions, with 10,700 functional biallelic cis-regulatory polymorphisms in a typical human. A recent study found that 34% of promoter polymorphisms significantly modulated reporter gene expression more than 1.5-fold in at least one of three transfected human cell lines in a screening exercise of the proximal promoters of 170 opportunistically selected genes [24]. Systematic surveys of promoter haplotypes by reporter gene analysis on chromosomes 21 [25] and 22q11 [26] show functional variation among 18% and 20% of polymorphisms tested, respectively, a proportion similar to that reported on analysis of allele-specific expression for other genomic regions [18, 19].

Fig. 3
figure 3

Investigation of regulatory polymorphism: the reporter gene assay. Strategy: Reporter gene assays are a powerful approach to resolve the effects of DNA sequences on gene expression. The DNA sequence of interest, for example a promoter region spanning an SNP, is placed upstream of a reporter gene whose expression can be measured. The reporter gene construct is then inserted (transfected) into a cell and expression assayed. Uses: This robust approach allows the effects of polymorphisms to be assayed, for example, comparing the expression of reporter gene constructs differing only by the nucleotide(s) of interest. The assay is highly sensitive and with appropriate controls is reproducible and specific. Limitations: The assay is an in vitro approach as the transfected DNA sequences lack the native chromatin configuration which may be essential to accurate interpretation of the consequences of genetic polymorphism. The design of reporter constructs is critical, notably the choice of which portions of the naturally occurring regulatory regions of the gene locus within which the polymorphism is found to include in the reporter gene. Any functional effect of an SNP may also be dependent on its naturally occurring haplotype: many early studies assayed SNPs in isolation rather dissecting their naturally occurring coinherited combinations. Results of reporter gene assays are highly context specific with respect to choice of cell type to transfect; stimulus used for induction of gene expression; the mode and efficiency of transfection; and the DNA plasmid design and preparation

Full size image
Fig. 4
figure 4

Investigation of regulatory polymorphism: the gel shift assay of protein-DNA interactions. Strategy: The ‘gel shift’ or electrophoreitic mobility shift assay investigates protein-DNA binding. Short DNA probes corresponding to a genomic DNA sequence of interest are synthesised, annealed as a duplex and radiolabelled. These are then usually incubated with a crude nuclear extract or a recombinant protein of interest. On binding by protein to DNA the mobility of the probe on electrophoresis is retarded (‘shifted’). The nature and specificity of these complexes can then be resolved. Uses: A highly specific and sensitive assay to investigate relative binding affinities of proteins to the two allelic forms of an SNP. Specificity can be resolved using unlabelled competitor probes. The nature of retarded complexes can be investigated by UV crosslinking experiments and using antibodies to abolish or ‘supershift’ bound complexes. Limitations: This is an in vitro assay, typically used as a screening tool using crude nuclear lysates for hypothesis generation which does not define whether binding actually occurs in vivo. The DNA probes used are short and lack a native chromatin structure. The absence of flanking sequences and conformational effects may lead to discrepancies in observed binding

Full size image

Other strategies to resolve regulatory SNPs include using computational based approaches to predict regulatory elements, such as promoters and sites of regulatory protein binding within them, combined with interspecies comparisons to identify evolutionary conservation [27]. A number of software programmes are available which can facilitate prediction of the effect of nucleotide substitutions on likely regulatory protein binding [28]. For specific regulatory transcription factor proteins the sophistication of such predictions is rapidly increasing based on experimentally derived protein-DNA binding data, for example, for NF-κB/Rel [29].

To illustrate the approaches to defining regulatory polymorphisms that I have outlined, I will discuss the evidence for regulatory polymorphisms in the context of a number of complex disease traits. This discussion demonstrates the limitations of experimental strategies used to resolve regulatory polymorphisms and also the successes that have been achieved. Given the breadth of the field this review does not attempt to be comprehensive or to discuss in detail the evidence for disease associations at a population genetic level but rather serves to introduce the subject to a clinically based readership and to discuss the challenges that lie ahead.

Regulatory polymorphisms are important in complex disease traits: lessons from analysis of a VNTR at the IDDM 2 locus

Regulatory polymorphisms are increasingly recognised to contribute to complex disease traits although the number of examples with both a clear-cut genetic association and defined molecular mechanism remain small. A number of examples involving SNPs are discussed below, including the role of polymorphism of CTLA4, encoding cytotoxic T lymphocyte antigen 4, in determining susceptibility to autoimmune disease [30] and polymorphism of DARC, encoding Duffy antigen receptor for chemokines, in malaria [31]. Variable number tandem repeat (VNTR) sequences (minisatellites) have been associated with a number of complex disease traits, but only in a minority of cases has there been evidence of direct functional consequences arising from polymorphism within the VNTR itself. One notable example is the insulin/IGF-2 (INS-IGF2) VNTR found approx. 600 bp from the transcriptional start site of INS, encoding insulin.

The VNTR is important to INS transcriptional regulation [32] and is found only in primates, suggesting that it represents a recent evolutionary change [32]. It comprises 14–15 bp tandem repeating sequences with the consensus ACAGGGGTSYGGGG [33] from which three classes can be defined based on length. Individuals homozygous for the shorter class I polymorphism were associated with higher rates of type I diabetes in case-control studies [34, 35]. This association was mapped to a 4.1 kb region spanning INS [36] within which the strongest disease association lies at the VNTR, but two additional SNPs in the locus may be involved [37, 38]. Overall, homozygosity for class I alleles was seen to confer increased risk while class III alleles were dominantly protective [37, 39, 40]. Genetic susceptibility to type I diabetes is a good example of a polygenic, multifactorial disease in which a number of loci other than the IDDM 2 locus at INS have been implicated through association studies, linkage analysis and genome scanning [41].

There is some functional evidence that the INS-IGF2 VNTR alters gene expression in the pancreas where class I alleles have been associated with modest levels of increased expression relative to class III alleles. Autopsy studies of human pancreas from fetal [42] and adult [37, 40] tissues show mRNA to be relatively higher for class I than class III alleles but observed differences were modest. Insulin secretion in vivo was lower in class III/II than class I/I homozygous individuals in some but not all studies [43, 44, 45, 46]. This is consistent with results obtained using reporter constructs transfected into a rodent pancreatic β cell line where class I alleles show 1.5- to 3-fold more expression than class III alleles [47]. One exception is the class I allele 698 (λHI) which showed lower expression than class III alleles when compared in fetal rat islet cells and in insulinoma cell lines using native insulin and heterologous promoter constructs [32]. Interestingly this particular allele showed no disease association with type I diabetes susceptibility, which may account for the apparent dichotomy with other class I alleles showing increased expression or alternatively be dependent on sequence rather than length differences between alleles [37]. Such sequence heterogeneity may account for differences in binding by regulatory proteins. For example, while the transcription factor Pur-1 was shown to bind the VNTR and activate transcription from a linked downstream promoter, different repeats bound Pur-1 with differing affinities [32]. These differences in affinity may relate to structural differences between repeats, in particular the presence of G-quartets. These unusual four-stranded DNA structures arise through non-Watson-Crick base-pairing as the INS-IGF2 VNTR is guanin rich [48, 49]. Pur-1 binding was found to depend on both intra- and intermolecular G-quartet formation of the INS-IGF2 VNTR [49].

Further clues as to how VNTR alleles may modulate disease susceptibility arose from studies of INS expression in the thymus [50, 51]. Given that insulin is a target autoantigen in type I diabetes it was postulated that insulin is normally expressed in the thymus to induce tolerance. Differential expression of INS associated with the VNTR may in turn then affect selection of insulin-specific T-lymphocytes and predispose to β-cell autoimmunity. INS was found to be transcribed in fetal thymi with proinsulin and insulin proteins present [50, 51]. Relative INS expression was significantly higher (two- to three-fold) with class III alleles than class I when mRNA was assayed in individuals heterozygous for INS-IGF2 VNTR I/III using a linked transcribed marker polymorphism in the INS 3′UTR [51]. Using the same approach, the allelic ratio was found to be converse in the pancreas, suggesting tissue-specific transcriptional effects. The same differences in relative allelic expression were observed in an independent study showing protective class III VNTRs were associated with higher steady-state levels of INS mRNA expression [50]. The higher levels of proinsulin and insulin in the thymus associated with class III VNTRs are postulated to promote self-tolerance through negative selection of insulin-specific T-lymphocytes in development, cells which play a critical role in the pathogenesis of type I diabetes.

Whether the INS-IGF2 VNTR also modulates expression of the adjacent imprinted gene IGF2 is controversial and may be tissue specific. Higher levels of IGF2 mRNA were associated with class I than class III alleles in human placenta using competitive reverse transcription PCR, consistent with reporter gene experiments in a hepatoma cell line using INS-IGF2 constructs [52]. However, other workers have shown no difference in IGF2 expression in thymus or pancreas associated with VNTR alleles [53]. Whether the VNTR acts as a long-range control element affecting the expression of both INS and IGF2 remains to be resolved.

Defining functionality among coinherited polymorphisms: CCR5 and AIDS progression

Complex multifactorial disease traits such as susceptibility to infectious disease have been found to typically involve several host susceptibility genes of individually modest magnitude within which a number of functionally important variants may be present [54]. Linkage disequilibrium between genetic markers may confound not only localisation on linkage mapping or association study but also the results of functional experiments to localise specific regulatory variation. Detailed definition of underlying haplotypic structure is needed in order to help resolve specific functional polymorphism(s), which may include both coding and regulatory variation as demonstrated by the work on variation at CCR5 and AIDS progression.

A number of host genetic factors have been shown to influence transmission and disease progression of the human immunodeficiency virus (HIV) 1 [55]. Notable among these is polymorphism of CCR5, encoding CC chemokine receptor 5, a major chemokine coreceptor of HIV-1 necessary for viral entry into cells. A number of coding polymorphisms have been identified, including a rare 32-bp deletion of the CCR5 open reading frame (Δ32) which results in a failure of the protein to be transported to the cell surface. Individuals homozygous for CCR5 Δ32 allele do not express CCR5 and are highly resistant to HIV-1 infection [56, 57]. Individuals heterozygous for CCR5 Δ32 have a delayed progression to AIDS and have a lower HIV-1 viral load early in disease [58].

Non-coding SNPs of the CCR5 cis-regulatory region have also been shown to modulate disease transmission and progression [59, 60, 61, 62, 63] although localising effects to specific polymorphisms has proved problematic due to linkage disequilibrium across the locus including neighbouring genes such as CCR2, the product of which can also act as an HIV-1 coreceptor. In order to dissect such effects the haplotypic structure of the highly diverse CCR5 cis-regulatory region was defined, with seven human ‘haplogroups’ and an ancestral haplotype resolved through comparisons with non-human primates [64]. Both CCR5 haplotypes and specific SNPs have been shown to modulate gene expression using reporter gene assays with evidence of differential protein-DNA binding in gel shift assays. For example, the ancestral haplotype HHA led to the lowest level of transcriptional activity in reporter gene assays using haplotype-specific constructs [64].

A G to A SNP of CCR5 at −2459 nt has been associated with disease susceptibility [59, 60, 62, 63], with the G allele forming part of the HHA haplotype and showing lower expression than the A allele on reporter gene analysis [60]. When unstimulated CD14+ monocytes from healthy donors were analysed for CCR5 density, a clear relationship was found with CCR5-2459 genotype: the lowest levels were observed in cells from individuals homozygous GG, intermediate with GA and highest with those homozygous −2459AA [65]. The same relationship was found when macrophage tropic HIV-1 was propagated in vitro in activated peripheral blood mononuclear cells with the G allele showing the lowest levels [65]. A study of Langerhans cells from healthy individuals heterozygous for Δ32 has also shown HIV infection levels to be associated with the CCR5-2459 SNP, with higher levels seen with the AA genotype [66]. Interestingly, while no difference in protein-DNA binding was observed for this SNP on gel shift assays using crude nuclear lysates, differential protein-DNA binding has been reported for a number of other CCR5 alleles, notably at −2554 with recruitment of NF-κB/Rel proteins postulated [64, 67]. The complexity of the task ahead to fully resolve the cis-regulatory SNPs of CCR5 is compounded by the multigenic host factors now shown to influence HIV-1 disease progression, including, for example, variation at chemokine/chemokine receptor loci, receptor ligands such as stromal cell derived factor 1 and RANTES (‘regulated on activation normal T cell expressed and secreted’), and HLA genotype [55].

Interpreting functional analysis of putative regulatory variation: TNF polymorphisms in malaria and rheumatoid arthritis

The functional characterisation of regulatory polymorphisms has been made more difficult by the many potential confounders of commonly used functional assays. Apparently conflicting results often relate to the context specificity within which data should be seen. As noted previously, understanding haplotypic structure often reveals that while a given SNP marker may have been used to distinguish between cells from different individuals, there has been confounding by coinherited functional variants within a subset of the population. Context specificity also extends to the precise experimental conditions used, with results, for example, specific to particular stimuli or cell type. Moreover, the design of experiments can be critical. Underpowered studies may be highly misleading while the particular genetic construct used can be highly specific in the results obtained for a given SNP in a reporter gene assay. These points are well illustrated by work on the TNF locus, which has been the subject of several studies dissecting the functional basis of observed genetic associations.

There has been considerable interest in genetic variation in TNF, encoding tumour necrosis factor (TNF) and susceptibility to a number of infectious and autoimmune conditions, often based on strong evidence of a role for TNF dysregulation in disease pathogenesis. Stable ‘high’ and ‘low’ TNF producers can be identified in a population with a significant genetic component to circulating TNF levels. There is evidence from linkage analysis implicating the TNF locus in susceptibility to mild malaria in African populations [68, 69] while possession of a G to A promoter SNP at −308 nt has been associated with susceptibility to severe malaria in Gambia [70] and Sri Lanka [71]. This SNP does not appear to modulate susceptibility to rheumatoid arthritis, but there is some evidence that possession of TNF-308 may be a useful marker of a patient’s response to anti-TNF treatment [72, 73].

The functional significance of TNF-308 remains controversial (reviewed in [74, 75]). There is some evidence to support an increased level of expression with the TNF-308A allele from reporter gene studies using a range of cell lines and range of stimuli, approx. one-half show a modest increase in levels of transcription with the A allele (predominantly in human cell types) while the remainder show no effect. Effects are highly context specific depending on cell type and stimulus [76] and features of reporter gene construct design. For example, detecting a difference between alleles may be dependent on inclusion of the TNF 3′ UTR in reporter constructs [77]. A number of studies correlating circulating TNF levels in individuals, or induced levels of TNF from isolated cells using a range of stimuli, have shown either a small increase with the TNF-308A allele or no effect. The small reported study sizes have limited power to detect a modest difference and may well be confounded by haplotypic heterogeneity when comparing groups using a single SNP marker. There is evidence in human B cell lines but not monocyte cell lines that this SNP lies in a site of protein-DNA occupancy using in vitro DNA footprinting [78, 79], and contrasting evidence of whether differential allelic recruitment of protein binding is present [77, 78]; the nature of any binding complex remains unresolved. Allele-specific analysis of gene expression in vivo using human lymphoblastoid cell lines [80] suggested no difference in allelic expression in the specific context analysed.

The TNF-308 SNP forms an extended haplotype with polymorphisms in neighbouring gene LTA, encoding lymphotoxin α, including LTA+252. Intriguingly, allele-specific differences in gene expression were observed at LTA with haplotypes bearing this SNP [80]. There is some evidence that LTA+252 modulates reporter gene expression and possibly protein-DNA binding [81] but, strikingly, it shows a very strong association with susceptibility to myocardial infarction on a genome-wide association study [81]. There is also evidence of association between the TNF-308/LTA+252 haplotype and rheumatoid arthritis in a study of individuals possessing HLA DRB1*04, the major determinant of disease susceptibility [82]. The picture is potentially further complicated by other LTA haplotypes showing functional effects. For example, the LTA+80 SNP shows evidence of allele-specific binding by a transcriptional repressor, ABF-1 [83]. This highlights the importance of a full understanding of haplotypic structure if both the functional effects of regulatory SNPs are to be resolved and disease associations finely mapped.

A number of other TNF SNPs found on different haplotypes have also been implicated in susceptibility to disease. For example, a TNF SNP at −238 nt has been associated with severe malarial anaemia [84] and protection from severe malaria [79, 85] while a linked SNP at

−376 nt is associated with susceptibility to severe malaria [79]. There is evidence that TNF-376 lies in a complex area of protein-DNA binding and acts to modulate binding of the transcription factor Oct-1 in human monocytes in vitro, increasing constitutive gene expression in a reporter gene model [79]. There is also some evidence that TNF-238 lies in the binding site of an as yet unidentified transcriptional repressor [86] and is associated with disease severity in rheumatoid arthritis [87, 88].

Regulatory polymorphisms in monogenic disease and complex traits: lessons from analysis of variation at F7

There has been considerable interest in genetic variation of F7, the gene encoding coagulation factor VII, both to determine the molecular basis of the autosomal recessive condition factor VII deficiency and in complex disease susceptibility, notably ischaemic heart disease. Factor VII is a vitamin K dependent factor essential for haemostasis which initiates the extrinsic pathway of coagulation and leads to localised generation of thrombin. It is striking, but not perhaps surprising, that the effects of the regulatory polymorphisms incriminated in factor VII deficiency and heart disease should be found to differ so markedly in their magnitude of effect on gene expression. Here, as in other examples of regulatory polymorphisms associated with complex multifactorial traits, the effects on gene expression appear modest. This highlights the need for experimental designs to be sufficiently sensitive to detect such differences.

The clinical phenotype of factor VII deficiency is highly variable and has been associated almost exclusively with coding changes in the gene, predominantly missense mutations but also deletions, splice site abnormalities and nonsense mutations [89]. Analysis of the promoter regions of individuals with severe factor VII deficiency has identified rare examples of patients possessing promoter SNPs with drastic effects on factor VII expression through modulation of transcription factor binding.

A naturally occurring T to G transversion at −61 nt was found to reduce reporter gene expression to 6.7% vs. wild type on transfection into a human hepatocyte cell line [90]. This SNP was found to prevent binding of the transcription factor hepatic nuclear factor (HNF) 4 on gel shift assay using human liver nuclear extracts. HNF4 had been previously shown to bind to this region of the promoter by DNA footprinting analysis and gel shift assay, and to be functionally important to promoter activity [91]. On cotransfection of an HNF4 expression construct in a non-hepatocyte cell line that does not constitutively support factor VII activity, transactivation was seen with the wild-type factor VII promoter construct but not with the rarer allele bearing the polymorphism [90]. A further C to T SNP at −55 nt also modulated binding by HNF4 in vitro on electrophoretic mobility shift assay, reducing reporter gene expression to 9.7% vs. wild type on transient transfection into HepG2 cells [92]. A C to G SNP at −94 nt prevents recruitment of Sp1 on electrophoretic mobility shift assay, reducing expression to 5.8% of wild type on reporter gene expression in HepG2 cells [93]. This area had been shown to be important to factor VII gene expression on promoter analysis [94].

These findings contrast with functional analysis of putative regulatory polymorphisms associated with susceptibility to ischaemic heart disease. Atherosclerotic plaque disruption which results in binding of tissue factor to circulating factor VII is a major cause of thrombosis in myocardial infarction. High levels of plasma factor VII are a significant predictor of death due to ischaemic heart disease [95], and genetic factors are reported to account for approx. one-third of plasma factor VII levels between individuals [96]. Resolving the functionally important variant(s) has, however, proved problematic. A coding region G to A SNP which results in a change in amino acid 353 from arginine to glutamine was associated with a significant reduction in plasma factor VII coagulant levels [97], and with susceptibility to ischaemic heart disease [98, 99] although this remains controversial [100, 101, 102]. Strong linkage disequilibrium was found between this SNP and a 10-bp insertion polymorphism in the 5′ untranslated region (position −323) with evidence that the disease association [99, 103] and functional effect may lie rather at this site. Low levels of plasma factor VII [104] and a reduction in reporter gene expression by one-third in a human hepatic cell line [91] were reported. However, evidence from a population in which the two polymorphisms were found in some individuals on different haplotypes suggest that the amino acid change does affect hepatic secretion independently of the insertion polymorphism [105].

To add to this complexity, the 10-bp insertion is in linkage with SNPs at −122 (T/C) and −401 (G/T); reporter gene experiments in a human hepatic cell line suggests that the maximal repression of expression was seen only when the naturally occurring haplotype containing all three polymorphisms was used rather than the polymorphisms in isolation [106]. Moreover, a further disease associated haplotype of F7 has been identified (−670C/−402A) [102] and associated with increased reporter gene expression in hepatoma cells [102]. The F7-402 A allele has been independently associated with increased transcriptional activity on transient transfection and with increased factor VII plasma levels [107]. Other non-coding polymorphisms of F7 have also been associated with levels of factor VII, notably of intron 7, where higher numbers of copies of a 37-bp repeat sequence were associated with mRNA expression levels through differential efficiency in mRNA splicing [108]. There is a report that this intronic repeat is associated with risk of myocardial infarction [98].

Resolving a clear molecular mechanism whereby a regulatory polymorphism modulates transcriptional regulation: Duffy binding protein and susceptibility to malaria

An elegant example of how a regulatory polymorphism may act to alter gene expression through a highly context specific effect on transcriptional regulation is shown by work on the Duffy binding protein. The presence of the Duffy blood group antigen on the surface of erythrocytes is essential for invasion by the malarial parasite Plasmodium vivax. Those individuals who lack the Duffy protein are completely protected from malaria due to this parasite [109, 110]. The molecular basis for this was found to be a promoter SNP in the FY gene (alias DARC, ‘Duffy antigen receptor for chemokines’) [31]. In the presence of the T to C SNP at -46 nt, a binding site for the transcription factor GATA-1 was found to be disrupted which in a reporter gene system reduced gene expression 20-fold. It is striking that in Duffy negative individuals, possessing the FY*0 allele bearing this SNP, the Duffy protein is not expressed on erythrocytes but is expressed on other cell types. This arises because GATA-1 is an erythroid-specific transcription factor, and hence the functional modulation of gene expression by the FY-46 SNP is cell-specific. The FY*0 allele found in Duffy negative individuals is at or near fixation in most sub-Saharan African populations but is very rare outside Africa, and this has been postulated as strong evidence of the action of natural selection. Given that vivax malaria is present at significant levels in non-African populations, this suggests that in evolutionary terms the mutation occurred after the proposed major human migrations out of Africa. It has been postulated that the strong subsequent selection pressure resulted in non-African populations showing different and greater numbers of SNPs at the Duffy locus than African populations, the opposite of the situation characteristically seen in other genomic regions where African populations are typically more diverse [111].

Looking beyond transcription: regulatory polymorphism and gene regulation

In order to advance further our understanding of regulatory polymorphisms in complex disease traits, it will be necessary to consider more broadly the potential role of DNA sequence polymorphisms in terms of the different facets of gene regulation which may be modulated (for reviews of gene expression see [112, 113]). In the examples described above, we have been concerned largely with cis-acting regulatory polymorphisms putatively modulating transcriptional initiation. This is the multifactorial process whereby gene expression is initiated through formation of a preinitiation complex including the enzyme RNA polymerase II and associated factors. Transcription factors and coregulator proteins facilitate and define this process, with modulation through DNA binding site sequence polymorphism altering the affinity of protein-DNA interactions and the control of transcriptional initiation. Following phosphorylation of the carboxyterminal domain of Pol II, nascent preRNA is transcribed through elongation with subsequent processing through capping, cleavage and polyadenylation with splicing of the transcript. The stability and localisation of RNA, and its translation into protein, are also all highly regulated and orchestrated. There is evidence that many of these steps occur concurrently, both physically and functionally. For example, splicing may be occurring while transcript is being synthesised. The relevance of this to our discussion of regulatory polymorphisms in complex disease traits is that the DNA, and in turn RNA, sequence remains pivotal to regulation throughout gene expression: sequence specificity for transcription factor binding is equally applicable to the rapidly growing list of other regulatory factors such as elongation and splicing factors. Many of the data to date have derived from allele-specific transcription factor binding and promoter analysis using reporter gene assays, but in many instances no definitive mechanism has been demonstrated beyond relative allelic differences in levels of transcription. A broader appreciation is therefore needed of the process of gene expression and how it may be potentially modulated.

Control of splicing

Genetic variation in both coding and non-coding DNA has been shown to affect the complex process of splicing whereby coding RNA sequences are identified and joined together. Polymorphism in exonic or intronic splicing regulatory elements can lead to exon skipping, activate cryptic splice sites or alter the balance of alternatively spliced isoforms [114]. Modulation of alternative splicing has been proposed as a mechanism underlying the strong observed disease association between susceptibility to autoimmune disease and variation at CTLA4, encoding cytotoxic T lymphocyte antigen 4 [30]. Disease susceptibility mapped to a non-coding region 6.1 kb 3′ to CTLA4 and mRNA isoforms were found to be correlated with genotype. In particular, a disease-susceptibility haplotype bearing the marker CT60G produced less of the soluble isoform of CTLA-4 which lacks exon 3 [30]. The molecular mechanism for this remains to be resolved. This contrasts with a C to G nucleotide transition in exon 4 of PTPRC, encoding a transmembrane protein tyrosine phosphatase, receptor-type C (also known as CD45) which was found to modulate splicing [115]. Possession of this SNP, 77 nucleotides downstream of the splice acceptor junction, was associated with susceptibility to multiple sclerosis although this remains controversial [116, 117]. PTPRC is expressed on all nucleated haemopoietic cells and is essential for T cell receptor signal transduction. In the presence of the SNP, an exonic splicing silencer element (designated ESS1) was found to be disrupted which normally functions to repress the weak 5′ splice site of CD45 exon 4 [115]. This results in expression of high levels of high molecular weight isoforms in affected individuals by activated lymphocytes [118, 119] and has been demonstrated on transfection of DNA constructs [120].

Translational efficiency

DNA sequence variation may potentially modulate translational initiation and efficiency. An example of the latter in the context of complex disease susceptibility is a polymorphism of F12, encoding the serine protease factor XII which is the first coagulation factor in the intrinsic pathway of the coagulation cascade. This SNP, denoted 46C/T, has been associated with high risk of coronary heart disease, possibly as a result of reduced fibrinolysis due to low plasma factor XII levels [121], and with risk of venous thromboembolism [122]. The SNP lies in the 5′UTR of F12, 4 bp upstream of ATG translation initiation codon [123]. The T allele, and in particular the TT genotype, is significantly associated with lower levels of factor XII [123, 124, 125] and creates a novel methionine initiating codon that reduces the translation efficiency of factor XII. Both alleles are equally transcribed; however, transcription/translational analysis showed less factor XII was produced from cDNA containing 46T than 46C [123].

Conclusions

Genetic variation plays a significant role in determining susceptibility to complex, multifactorial disease traits ranging from autoimmune to infectious disease. Typically multiple genes have been implicated, each of modest magnitude of effect and fine mapping has proved problematic. There is increasing evidence at a population and experimental level that regulatory polymorphisms are important, and given the strong recent data that genetic variation underlies differences in gene expression this is only likely to increase. Data from in vitro reporter gene assays and allele-specific quantification of gene expression suggest the proportion of non-coding SNPs which may be functional could be much higher than expected. This may also reflect the need for greater rigor in the technical interpretation of data from such approaches. A number of examples where functional regulatory polymorphisms have been found to contribute to complex traits have been discussed, including disease associated genes where both coding and regulatory SNPs have been implicated such as CCR5 and HIV-1 disease. Regulatory polymorphisms have also been found to underlie rare monogenic traits. In the case of F7 the difference in magnitude of effect was striking for SNPs underlying factor VII deficiency and those implicated in ischaemic heart disease. In general for complex traits, a modest magnitude of effect has typically been found which has implications for the sensitivity of experimental approaches which should be applied for their detection. Differences in expression associated with regulatory polymorphisms appear highly context specific, with effects dependent on the specific stimulus, tissue or cell type analysed. This is highlighted by the Duffy binding protein and INS VNTR which showed altered expression in erythrocytes and the thymus, respectively, context specificity which appears critical to the resulting biological phenotypes of protection from malarial parasite invasion and type I diabetes. To date most work has focused on modulation of the process of transcriptional initiation, with data derived largely from reporter gene analysis and gel shift assays of protein-DNA binding. The value and limitations of such approaches have been discussed, with often apparently contradictory results relating to differences, for example, in the biological systems studied, and the design of DNA constructs.

The precise combination of coinherited polymorphisms on a given allele or haplotype are critical to both mapping polygenic disease and in designing and interpreting assays investigating putative regulatory polymorphisms. In many cases it is not possible to demonstrate that a specific SNP is functional, but rather that the haplotype shows allele-specific differences in expression with further resolution needed. Improved approaches for in vivo analysis of regulatory polymorphisms are required which allow the naturally occurring regulatory machinery to operate in a native chromatin environment. There is a need for a broader appreciation of the complexities of gene regulation, and the multiple points at which sequence polymorphism may modulate control of gene expression so that alteration of processes as diverse as transcriptional elongation, splicing, RNA stability and translation can also be considered.