Single nucleotide polymorphisms (SNPs) are the most common form of genomic variations occurring on average every 1000 nucleotides. The vast majority of SNPs are neutral allelic variants, however the few that do influence a phenotype in a measurable way, are important for understanding the underlying genetics of human health. SNPs are the focus of a large number of human genetics studies attempting to understand their impact on complex diseases like Alzheimers, Parkinsons, diabetes, etc. Most SNPs, by the virtue of their location within genes (introns, 3'-UTRs, etc) or between genes, are considered most likely to be benign and not to contribute to a phenotype, whether it may be the manifestation of a disease or quicker metabolism of a drug. Among the group of SNPs located within coding regions of genes and causing a change in the peptide sequence (non-synonymous SNPs or 'nsSNPs') or among SNPs located within promoters (regulatory SNPs or rSNPs), a majority may not influence the overall activity of the protein or the gene expression. With the per-SNP validation and genotyping cost relatively high, it is increasingly important to develop strategies to predict functionally relevant SNPs in silico. The SNP databases in public domain, like NCBI/dbSNP and HGVbase, have facilitated this by highlighting all nsSNPs and also further classifying the location of the amino acid within the encoded proteins 1 to more accurately predict the detrimental effects of a change in peptide sequence. Several recent studies have attempted to focus on the subset of nsSNPs that most likely influence phenotype 23456. Of the approximately 4.5 Million SNPs in dbSNP 7, an estimated 10,000 nsSNP exist and approximately 10–15% of those are projected to be damaging 6. Comparatively fewer attempts have been made to predict and validate functional promoter SNPs 8.

Transcriptional regulatory regions in the 5'-FR of human genes encode short (often < 25 bp) 910 sequences which serve as targets for binding of transcription factors (TFs). Understanding the conditions of binding, specificity and identity of the factors would help us understand the mechanism of regulation of human genes. Eukaryotic TFs tolerate considerable sequence variation in their target sites and recent bioinformatics works 111213 have developed methods to model the DNA binding specificity of individual TFs 10. Such matrices, although highly accurate 914, are less specific in the identification of sites with in vivo function 11, mainly due to our limited understanding of additional factors involved in TF specificity such as factor cooperative binding, protein-protein interactions, chromatin superstructures and TF concentrations. Currently the most successful approach to overcome this information gap is based on the assumption that sequences conserved between species (here human and mouse) would most likely mediate biological function 1516171819.

The 7TM (7 trans-membrane domain proteins), also known as the hetero-trimeric GTP-binding protein (G protein)-coupled receptors (GPCRs) are members of a large family with an estimated 700 genes in the human genome 20. By some estimates, nearly 60% of drugs marketed today target directly or indirectly the GPCR family members 21. Several studies have collectively analyzed the occurrence, and importance, of coding SNPs to pharmaceutical efforts, in this family of genes 222324. Characterizing polymorphisms that are located in the 5'-FR of these genes and that influence expression has been reported earlier. We therefore selected a subset of clinically and pharmacologically important GPCR genes and their 5'-FR sequences to test our bioinformatics and laboratory experimentation approach for prediction of functionally important SNPs in regulatory regions. Our selection system evaluates the influence of SNPs on TF-DNA complex stability, and further investigates the influence of such SNPs on promoter activity. We present a proof-of-concept for such a strategy and identify issues and problem-areas for future developments.

The lists of the full names and Ensembl ENSG numbers 25 of the 176 GPCR genes are shown in additional files [See Additional file 1]. From a total of approximately 800 SNPs in proximal 5 kb regions, less than 200 were mapped to regions of mouse-human genome conservation. Of these approximately 200 SNPs, 36 were predicted to influence TF binding, in regions of sequence conservation of over 70% in human-mouse; the alignments for two such regions are indicated in additional files [See Additional file 2]. Table 1 lists the 21 genes, along with the SNPs, TFs and TFBS sequences and positions relative to transcription start site. These 36 candidate SNPs in Table 1 were qualified by our selection criteria, as described in the methods section, and were predicted to influence the binding of TFs in a qualitative manner. The absolute binding score of the TF differed by at least 2 units between the two alleles. Ten SNPs within the 5'-FR of 7 genes were selected for EMSA tests and are shown in bold letters in Table 1. The choice of these genes was based on our understanding of their significance in human physiology and relevance to research interests within the GPCR research community. Table 2 shows the results from the EMSA experiments, where values in each column for Allele 1 and Allele 2 are the ratios of measurements from each of the 5 different concentrations of the competitor oligomers (labeled × 5 through × 25 in Table 2 and in Figure 1) divided by the measurements without competitor (labeled 'C' in Figure 1). The decrease in level of the labeled product as a consequence of increasing non-labeled oligomer concentration is an indication of the efficiency of displacement, thereby reflecting the relative stability of the DNA-protein complex. A marginal increase in level of radio-labeled complex, instead of a decrease (Table 2 rs1800508) is generally considered to be due to additional factor involvements. Table 2, column 'Ratio' shows the difference, calculated for the highest concentration of the non-labeled competitor (25-fold), between the efficiency of competition between a perfect-match competitor and the allelic mismatch competitor. Values close to 1.00 indicate no difference in rate of competition between the two alleles, and therefore no relative difference in stability of the DNA-protein complex. While 4 SNPs exhibit mobility shift with a difference of 2-fold or more (rs267412, rs509813, rs945032 and rs2528521), three SNPs exhibit a moderate, nevertheless reproducible shift (rs1799722, rs2882225 and rs1538251). Finally, three SNPs fail to show any significant shift (rs968554, rs267413 and rs1800508). In all, seven out of ten polymorphic markers indicated reproducible binding differences between the alleles. Figure 1 shows pictures of EMSA gels for two of the SNPs, rs1799722 and rs2528521.

For EMSA-positive markers with no validation information at dbSNP, HGVbase or Celera Discovery Systems™ we did a validation analysis using RFLP (restriction fragment length polymorphism) on DNA samples from 25 healthy Nordic individuals. The three SNPs which failed to show positive gel-shift results (serotonin receptor 5HT-1A: rs968554; DRD1: rs267413; and BK-2: rs1800508) were not investigated any further. For dopamine receptor D1 (DRD1) polymorphism (rs267412) the genotype distribution was found to be TT = 30%, AA = 20% and AT = 50%, and for calcitonin receptor promoter (CT-R) polymorphism rs2528521, it was GG = 40%, AA = 30% and GA = 30%. While rs267412 was found to be in Hardy Weinberg Equilibrium (HWE), rs2528521 was not. A larger population sample should be genotyped to accurately measure HWE for both these loci. Bradykinin B2 (BK-2) promoter polymorphism rs1799722 is noted to be polymorphic (major allele C: 56%) at NCBI/dbSNP. The polymorphic nature of this locus (rs1799722) was also confirmed by crosschecking at Celera Discovery Systems™. Also for the second BK-2 SNP rs945032, the allele frequency information was found at dbSNP (major allele = 80%). NCBI's dbSNP provided no allele frequency information about rs1538251 adenosine-A3 receptor (ADORA3), and on sequencing 20 DNA samples this marker proved to not be polymorphic in Nordic sample population, and was eliminated from further analysis (data not shown). The genotype frequency of muscarinic acetylcholine receptor M1 (CHRM1) SNP rs509813 was documented at Celera Discovery Systems™. The contig-position of rs2882225 (follicle simulating hormone receptor; FSHR) was not in agreement between the three major public genome databases, i.e. NCBI, Ensembl and Santa Cruz Genome Assembly (UCSC). It was mapped within the transcript for FSHR by NCBI/dbSNP, and completely absent from Ensembl and UCSC 26. Therefore of the seven SNPs exhibiting positive EMSA, five SNPs (in four genes) qualified for analysis of their influence on promoter activity.

For expression analysis in living cells, the published promoter regions, or putative regulatory 5'FR of up to 2 Kb, were cloned and basal levels of luciferase were monitored. Repeated attempts to clone the promoter region of DRD1 failed (data not shown). Furthermore the position of rs267413 is mapped at -4446 nucleotides with respect to the transcription start site, whereas the minimum length of the genomic fragment known to drive DRD1 expression is limited to 2571 nucleotides. Considering the distal position of the marker, we decided not to examine rs267413 further in this study. The promoter regions of BK-2, CHRM1 and CT-R were cloned successfully. A total of four dissimilar human cell lines (HeLa, Hep2G and SK-N-MC, HEK293) were used to monitor the influence of the four SNPs (rs945032, rs1799722, rs2528521 and rs509813) to investigate differences in expression that are possibly due to differences in TF expression in different cell types. BK-2 promoter SNPs rs945032 (genotype = GG) and rs1799722 (genotype = AA) showed approximately 40%-60% higher activity in HeLa cells as compared to their other homozygote alleles AA and GG, respectively (Figure 2). The BK-2 SNP rs945032 behaves in a reciprocal manner in two (HeLa, and HEK293) cell types. The BK-2 SNP rs1799722 allele 'C' increases expression only in HeLa while decreasing expression in the other three cell types, similar to CHRM1 marker rs509813. CT-R marker rs2528521 and CHRM1 marker rs509813 failed to show any influence on luciferase expression levels in HeLa and SK-N-MC cells. Finally, the CT-R SNP rs2528521, allele 'C', influences expression in a significant manner, but only in one cell line (Hep2G). Hence, different alleles behave differently in different cell environments.

SNPs that are located within coding regions and result in a change in the peptide sequence may be classified as 'damaging' or 'altering' if predicted to be in structurally or functionally important sites of the three dimensional structure of the protein. It is less straightforward to predict the functional importance of SNPs within regulatory regions. TFs tolerate variation in their binding sites and all positions in a site do not contribute equally to the binding energy. Therefore, the quantitative effect a given rSNP has on gene expression depends on its position and the bases involved. Complex human diseases like Parkinson's, diabetes and obesity are polygenic diseases, where many predisposing genetic and environmental factors together, over a period of time, cause a disease state. Differences in expression of genes and cellular concentrations of proteins due to common polymorphisms in 5' regulatory regions could equally elucidate gene function as the examination of non-synonymous SNPs in coding regions. We decided to test the 5'-FR of a group of physiologically and clinically important genes, for SNPs within TFBS, which could potentially influence the kinetics of binding affinity.

A common strategy for modeling the binding preferences of a transcriptions factor is to construct a position weight matrix (PWM) from known binding sites. PWMs are probabilistic models that capture the nucleotide preference at each position of a TFBS as well as the differential contribution of positions to the overall binding energy. When a putative binding site sequence is assessed using a PWM, a score is obtained that theoretically should be proportional to the binding energy between the TF and that sequence 10. It has been convincingly shown, and generally accepted, that by considering DNA sequence conservation between mouse and human, the over-predictive nature of TFBS modeling can be significantly remedied. Therefore we chose for this work, not to include 'negative controls', that is, SNPs from outside of mouse-human conserved regions. We do indeed think that larger studies in the future should perhaps incorporate certain number of such negative controls to validate the theoretical predictions. Thus, using PWMs it should be possible to overcome the difficulties with rSNP detection stated above. We reasoned that if the score difference between alleles is large, it should correspond to a difference in gene expression that is reproducible in living cells. Using the JASPAR database of PWMs 27 and a phylogenetic footprinting strategy previously shown to be successful 17, we developed a method to detect putative TFBS and identify rSNPs likely to affect TF binding significantly. By incorporating phylogenetic footprinting, the method reported in this study emphasizes SNPs present in genomic regions that are highly conserved between human and mouse, thereby increasing the probability of a downstream functional influence of variations within theses sequences.

Electromobility Shift Assays (EMSA) produce DNA-protein binding interactions in artificial conditions. Therefore in silico prediction methods based on other in vitro or in vivo selection technologies, like 'systematic evolution of ligands by exponential enrichment' (SELEX), may not agree with the experimental outcome of EMSA procedures. Since the construction of PWMs is often concluded from published records based on SELEX enrichment approaches, it is informative to experimentally validate the predicted binding site using methods like EMSA. Therefore, we validated a subset of our predictions with in vitro electro-mobility shift. We used a stringent selection criterion, that is, qualified only alleles demonstrating an absolute binding score difference of 2.0 or more. A stringent selection criterion would no doubt decrease excessive hits and false positives at the expense of certain loss of true positives. Our results showed that approximately 60%-70% (i.e. 7 out of the selected 10 SNPs) of predicted sites (Table 2) bind proteins from HeLa nuclear extracts.

We finally attempted to correlate the EMSA findings from the 10 SNPs with influences on actual promoter activity within living cells. Due to mapping discrepancies of one SNP and failure to clone one promoter, we tested only four out of the seven EMSA-positive SNPs identified thus far. Of the four SNPs tested in a promoter-less expression vector, two (rs945032 and rs1799722) indicated significant influence on promoter activity, while two showed convincing and reproducible, yet comparatively limited influence (rs2528521 and rs509813) on promoter activity. The influence of these polymorphisms (rs945032 and rs1799722) indicate that any given functional variation within a regulatory region might exert a measurable influence within the context of a cell type determined by the TF expression profile of the cells and perhaps competitive binding of the TF to overlapping multiple binding sites.

There are several factors in the current approach which indicate that there are far more rSNPs than currently detected using available technologies. The EMSA assays employ HeLa nuclear extract, thereby limiting our findings to the TF expression repertoire of only HeLa nuclei. The TFBS package used a limited collection of high quality PWMs, which unfortunately represent only a small part of the approximately 2,000 known human and mouse TFs. The theoretical thresholds set for selections of alleles which are predicted to differentially bind TF require further rigorous testing to ensure that the present selection is optimal.

From a total of approximately 200 SNPs in evolutionally conserved 5'-FR of 176 human GPCR genes, our prediction algorithm selected 36 SNPs with possible influence on TFBS. When ten of these 36 SNPs were tested for mobility shifts, seven exhibited a positive result, and four of these were further tested for influence on promoter activity using an in situ reporter system. Finally, two of the four showed significant and reproducible influences which were dependent on the cell environment. Thus starting from a large pool of potential regulatory SNPs, we successfully identified a small fraction that actually influenced promoter activity. We therefore propose a method for effective selection of functional, regulatory SNPs, in evolutionary conserved 5'-FR regions of human genes, as a means for identification of candidate SNPs for genetic association analysis studies.

TF: Transcription Factor; TFBS: transcription factor binding site; SNP: Single Nucleotide Polymorphism; GPCR: G-protein coupled receptors; 5'-FR: 5' flanking regulatory region.

Original Concepts and supervision: CW, WWW, BL and SM-T; Bioinformatics: PGE, BL and SMT; Running costs: SMT and CW; Manus preparation: SMT, PGE, CW; Electromobility Shift Assays and RFLP assays: YM; Luciferase expression: MAF and YM.