Bioinformatics Advance Access originally published online on October 12, 2006
Bioinformatics 2006 22(23):2960-2961; doi:10.1093/bioinformatics/btl518
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
ssSNPer: identifying statistically similar SNPs to aid interpretation of genetic association studies
Genetic Epidemiology Laboratory, QIMR 300 Herston Road, Brisbane, Queensland, 4006, Australia
 |
ABSTRACT |
|---|
 TOP ABSTRACT REFERENCES |
|---|
Summary: ssSNPer is a novel user-friendly web interface that provides easy determination of the number and location of untested HapMap SNPs, in the region surrounding a tested HapMap SNP, which are statistically similar and would thus produce comparable and perhaps more significant association results. Identification of ssSNPs can have crucial implications for the interpretation of the initial association results and the design of follow-up studies.
Availability: http://fraser.qimr.edu.au/general/daleN/ssSNPer/
Contact: daleN{at}qimr.edu.au
Most papers on genetic association conclude with the obligatory genuflection that the identified variant may not be causal, but in tight linkage disequilibrium (LD) with another that is. Here, I try to quantify this problem and highlight the utility of identifying ‘statistically similar’ single nucleotide polymorphisms (ssSNPs) in the interpretation of initial association results and in the design of follow-up studies.
Given practical limitations on genotyping, investigators are forced to select and prioritize test SNPs from the 
12 million SNPs currently present in the dbSNP (build 126) database (http://www.ncbi.nlm.nih.gov/projects/SNP/). It has therefore become routine to utilize correlations (LD) among nearby variants to guide the selection of informative ‘tag’ SNPs; the most extensive genome-wide resource being the International HapMap Project, currently containing 6.3 million SNPs (http:/www.hapmap.org/).
The degree of LD between alleles at two SNPs can be described in terms of the common measure r. The LD measure r, is the correlation coefficient for a 2 x 2 table and is usually squared to remove the arbitrary sign introduced when the SNP alleles are labelled. The resulting metric (r2) is useful in association analyses as it expresses the proportion of variance in SNP 1 ‘explained’ or predicted by SNP 2 and vice versa. As a consequence, r2 provides a measure of statistical similarity between SNPs. Put another way, r2 measures the statistical power SNP 1 has to predict the genotypes of SNP 2 and vice versa (Hill and Robertson, 1968).
In candidate gene-wide association studies, SNPs may be selected based on plausible biological causality [e.g. nonsynonymous coding SNPs (nsSNPs), predicted transfactor binding sites, predicted miRNA target sites, etc.] and/or tagSNPs may be selected until an r2 
0.8 (for example) is exceeded for all sites (Carlson et al., 2004). The latter case also pertains to positional genome-wide association studies where 100 000s of tagSNPs are required to satisfactorily screen the entire human genome.
Although it is becoming common practice for investigators to utilize LD among nearby variants in the design of association studies, I wish to emphasize that once a significant association is obtained it is important to carefully examine inter-marker LD, as it can have crucial implications for the interpretation of the initial association results and the design of follow-up studies.
To explain, first consider a pair of SNPs with r2 = 1.0; these SNPs are ‘statistically indistinguishable’ (siSNPs) [also previously termed ‘genetically indistinguishable’, giSNPs (Lawrence et al., 2005)] and barring genotype error will produce statistically identical association results. In addition, untyped SNPs with r2 0.8 can be considered to be of high enough similarity (i.e. SNP 1 has 80% power to predict the genotypes of SNP 2 and vice versa) to produce similar and importantly, perhaps more significant association results compared to the original associated test SNP. Therefore, it is recommended that researchers investigate ssSNPs (especially where r2 0.8) and their surrounding plausible causative region(s) in addition to the region(s) initially implicated by the test SNP.
To demonstrate the utility of identifying HapMap ssSNPs, I examined the 1 Mb region surrounding the test SNPs associated in the four studies (Amundadottir et al., 2006; Graham et al., 2006; Grant et al., 2006; Smyth et al., 2006) recently appearing and discussed (Todd, 2006) in Nature Genetics. Briefly, HapMap SNPs (release #21) genotyped in the CEPH trios within 500 kb either side of the most significantly associated (HapMap) test SNP in each study were examined via my novel ssSNPer web interface (described below).
Only one HapMap SNP (rs12243326, r2 = 0.969) 20 087 bp upstream would likely produce an association result similar (i.e. with r2 0.8) to the original association of rs12255372 with type 2 diabetes (Grant et al., 2006). Given that both rs12243326 and rs12255372 are within the TCF7L2 gene, it is reasonable to conclude that further studies aimed at identifying the causal variant(s) should concentrate solely on TCF7L2.
Similarly, only one HapMap SNP (rs2111485, r2 = 0.882) would have high power to produce an association result comparable to the type 1 diabetes (T1D) associated nsSNP rs1990760 (exon 15) (Smyth et al., 2006). The ssSNP rs2111485 is 13 515 bp downstream from rs1990760, lies 13 053 bp downstream of IFIH1 and 23.5 kb upstream from the next nearest gene (FAP). Hence the ssSNP information supports the biological evidence indicating IFIH1 to be the strongest candidate. However, two ssSNPs {rs984971 (intergenic), rs2068330 (intron 14)} with r2 = 0.72 were identified within the other functional candidate gene in the region mentioned by the authors (KCNH7). Importantly, these two SNPs were also reported by Smyth et al., (2006) to be highly associated with T1D. Hence, although IFIH1 remains the more likely candidate due to association of the nsSNP rs1990760 and further studies should (initially) concentrate on this gene, the possibility remains that there are other regulatory variants within the IFIH1 region, including KCNH7 and FAP.
In contrast, there are 17 ssSNPs (Table 1) ranging from 15 305 bp upstream to 89 846 bp downstream of the rs2280714 SNP most strongly associated with elevated IRF5 expression levels, which is associated with increased risk to systemic lupus erythematosus (Graham et al., 2006), making the association interpretation more complex. In particular, all but one of the 17 ssSNPs (rs752637) is 3' to IRF5 and lie within the TNPO3 gene. The authors provide a strong case for a role of IRF5 based upon expression data and biological plausibility; however they note that the SNPs most strongly associated with IRF5 expression are well downstream of IRF5 and do not lie in a recognizable regulatory region and hypothesize that an additional (unknown) genetic variant in tight LD with rs2280714 may drive the expression phenotype. Consequently, to thoroughly investigate the association and identify the causal variant(s), further analyses should involve both the IRF5 and neighbouring TNPO3 gene.
|
Correspondingly, 19 ssSNPs (Table 1) ranging from 815 bp 5' to 54 322 bp 3' were identified for the rs1447295 SNP associated with prostate cancer (Amundadottir et al., 2006). Intriguingly, there are currently no genes identified within the 55 137 bp region spanned by the 19 ssSNPs. However, there are numerous conserved regions and two pseudogenes (DQ515896 and DQ515897) within the ssSNP-identified region which provide likely targets for further study. Of course, as in all HapMap-based analyses, other (ss)SNPs, not currently genotyped in the HapMap may exist which have the potential to assist investigators identify the causal variant(s), thus highlighting the need for continued SNP discovery.
The ssSNPer web interface is a simple yet powerful tool, which allows users to merely upload to our web server a file specifying a list of test HapMap refSNP (rs) IDs and a file containing HapMap SNP genotype data (step-by-step instructions for obtaining HapMap genotype data for regions up to 5 Mb are provided) to determine the number and location of ssSNPs in the surrounding region. Single test SNP ssSNPer output first includes a graph providing an overview of the statistical similarity (r2), based on expectation-maximization (EM) estimated haplotype frequencies (Excoffier and Slatkin 1995), between the test SNP and all surrounding HapMap SNPs. Following the graph is a list of the surrounding SNPs which have at least 50% similarity (i.e. r2 0.5) first sorted from highest to lowest r2 and then according to map order (i.e. from pter to qter). Map distances are given as base pair from the test SNP. The final section reports all r2 values (i.e. 0 
r2 1) between the test SNP and all HapMap SNPs across the submitted region in map order; these data are suitable for plotting along with other map landmarks (e.g. genes, conserved regions, etc). Multiple test SNP ssSNPer output is restricted to the section listing ssSNPs with r2 0.5 and is also useful for choosing replacement SNPs in the late stages of association study design, where ssSNPs can be preferentially chosen with respect to their r2 values to replace previously selected SNPs which do not fulfil assay design requirements.
|
Acknowledgments |
|---|
The author would like to thank Professors Nick Martin, Peter Visscher and Grant Montgomery and the anonymous reviewers for valuable comments on earlier drafts of this manuscript. The author is supported by an NHMRC Fellowship (#339462). This research was supported in part by NHMRC Program Grant (#389892). Conflict of Interest: none declared.
|
FOOTNOTES |
|---|
Associate Editor: Martin Bishop
Received on September 8, 2006; revised on October 5, 2006; accepted on October 5, 2006
|
REFERENCES |
|---|
|
TOPABSTRACTREFERENCES |
|---|
Amundadottir, L.T., et al. (2006) A common variant associated with prostate cancer in European and African populations. Nat. Genet, . 38, 652–658[CrossRef][Web of Science][Medline].
Carlson, C.S., et al. (2004) Selecting a maximally informative set of single-nucleotide polymorphisms for association analyses using linkage disequilibrium. Am. J. Hum. Genet, . 74, 106–120[CrossRef][Web of Science][Medline].
Excoffier, L. and Slatkin, M. (1995) Maximum-likelihood estimation of molecular haplotype frequencies in a diploid population. Mol. Biol. Evol, . 12, 921–927[Abstract].
Graham, R.R., et al. (2006) A common haplotype of interferon regulatory factor 5 (IRF5) regulates splicing and expression and is associated with increased risk of systemic lupus erythematosus. Nat. Genet, . 38, 550–555[CrossRef][Web of Science][Medline].
Grant, S.F., et al. (2006) Variant of transcription factor 7-like 2 (TCF7L2) gene confers risk of type 2 diabetes. Nat. Genet, . 38, 320–323[CrossRef][Web of Science][Medline].
Hill, W.G. and Robertson, A. (1968) Linkage disequilibrium in finite populations. Theor. Appl. Genet, . 38, 226–231[CrossRef].
Lawrence, R., et al. (2005) Genetically indistinguishable SNPs and their influence on inferring the location of disease-associated variants. Genome Res, . 15, 1503–1510
Smyth, D.J., et al. (2006) A genome-wide association study of nonsynonymous SNPs identifies a type 1 diabetes locus in the interferon-induced helicase (IFIH1) region. Nat. Genet, . 38, 617–619[CrossRef][Web of Science][Medline].
Todd, J.A. (2006) Statistical false positive or true disease pathway? Nat. Genet, . 38, 731–733[CrossRef][Web of Science][Medline].
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  R. M. Spaapen, H. M. Lokhorst, K. van den Oudenalder, B. E. Otterud, H. Dolstra, M. F. Leppert, M. C. Minnema, A. C. Bloem, and T. Mutis Toward targeting B cell cancers with CD4+ CTLs: identification of a CD19-encoded minor histocompatibility antigen using a novel genome-wide analysis J. Exp. Med., November 24, 2008; 205(12): 2863 - 2872. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M. D Tobin, M. Kahonen, P. Braund, T. Nieminen, C. Hajat, M. Tomaszewski, J. Viik, R. Lehtinen, G A. Ng, P. W Macfarlane, et al. Gender and effects of a common genetic variant in the NOS1 regulator NOS1AP on cardiac repolarization in 3761 individuals from two independent populations Int. J. Epidemiol., October 1, 2008; 37(5): 1132 - 1141. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 Z. Z. Zhao, P. M. Pollock, S. Thomas, S. A. Treloar, D. R. Nyholt, and G. W. Montgomery Common variation in the fibroblast growth factor receptor 2 gene is not associated with endometriosis risk Hum. Reprod., July 1, 2008; 23(7): 1661 - 1668. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 G Kristjansdottir, J K Sandling, A Bonetti, I M Roos, L Milani, C Wang, S M Gustafsdottir, S Sigurdsson, A Lundmark, P J Tienari, et al. Interferon regulatory factor 5 (IRF5) gene variants are associated with multiple sclerosis in three distinct populations J. Med. Genet., June 1, 2008; 45(6): 362 - 369. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
Z. Z. Zhao, D. R. Nyholt, L. Le, S. Thomas, C. Engwerda, L. Randall, S. A. Treloar, and G. W. Montgomery Genetic variation in tumour necrosis factor and lymphotoxin is not associated with endometriosis in an Australian sample Hum. Reprod., September 1, 2007; 22(9): 2389 - 2397. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||