Volume 3 Supplement 7
Genetic Analysis Workshop 16
Incorporating multiplemarker information to detect risk loci for rheumatoid arthritis
 Xuexia Wang^{1, 2},
 Huaizhen Qin^{1} and
 Qiuying Sha^{1}Email author
DOI: 10.1186/175365613S7S28
© Wang et al; licensee BioMed Central Ltd. 2009
Published: 15 December 2009
Abstract
In genomewide association studies, new schemes are needed to incorporate multiplelocus information. In this article, we proposed a twostage slidingwindow approach to detect associations between a disease and multiple genetic polymorphisms. In the proposed approach, we measured the genetic association between a disease and a singlenucleotide polymorphism window by the newly developed likelihood ratio testprincipal components statistic, and performed a slidingwindow technique to detect disease susceptibility windows. We split the whole sample into two subsamples, each of which contained a portion of cases and controls. In the first stage, we selected the top R windows by the statistics based on the first subsample, and in the second stage, we claimed significant windows by falsediscovery rate correction on the pvalues of the statistics based on the second subsample. By applying the new approach to the Genetic Analysis Workshop 16 Problem 1 data set, we detected 212 out of 531,601 windows to be responsible for rheumatoid arthritis. Except for chromosomes 4 and 18, each of the other 20 autosomes was found to harbor risk windows. Our results supported the findings of some rheumatoid arthritis susceptibility genes identified in the literature. In addition, we identified several new singlenucleotide polymorphism windows for followup studies.
Background
Rheumatoid arthritis (RA) is a common chronic destructive disease of an unknown complex etiology. Both genetic and environmental bases are thought to contribute to this disease. The human leukocyte antigen (HLA) region major histocompatibility complex (MHC) on chromosome 6 (6p21.3) is known to be associated with RA. This region is the only one that has been consistently shown to be both linked and associated with RA across all populations. It extends over 3.6 Mb and is divided into three subregions (classes I, II, and III). It is a highly dense area containing about 220 genes, many of which are thought to have immunoregulatory functions [1]. Recently, matured genotyping technology and availability of large casecontrol collections have made it possible to detect mild risk loci. The Genetic Analysis Workshop 16 (GAW16) Problem 1 data set is such a large scale casecontrol study which contains genotypes at 531,689 singlenucleotide polymorphisms (SNPs) on chromosomes 122 for 868 cases and 1194 controls.
Recently, many approaches, e.g., Hotelling's T^{2} test [2, 3] and the linkage disequilibrium (LD) contrast tests [4, 5], have been proposed to detect multiplemarker association. The Hotelling's T^{2} test and the LD contrast test compare the means and the variancecovariance matrices of genotype scores between cases and controls, respectively. More recently, we proposed a likelihood ratio testprincipal component (LRT_PC) to compare the means and the variancecovariance matrices of genotype scores simultaneously [6]. However, all of these approaches only allow a SNP region of several to tens of markers.
In this article, we report a novel genomewide slidingwindow approach to detect genetic association between a trait and SNP regions. This approach integrated the LRT_PC with the concept of sliding window [7] and the basic idea of twostage approaches [8]. Applied to the GAW16 Problem 1 data set, our approach yielded results that support the findings in the literature of some RA susceptibility genes on chromosomes 1, 2, and 6 and detected more SNP windows for followup studies.
Methods
LRT_PC statistic
We recently proposed a LRT_PC approach to test the association between a given SNP window and a disease status [6]. To calculate the test statistic, we first perform principal component (PC) analysis to the genotype scores of the sampled individuals. Then, the LRT_PC test statistic is given by , where n and m are the numbers of cases and controls, respectively, and the values are the sample variancecovariance matrices of the first K PCs in cases, controls, and the pooled sample, respectively. Wang et al. [6] showed that the LRT_PC test is more powerful than the Hotelling's T^{2} test and the LD contrast test [2–5] in most cases. The power of the LRT_PC test is perhaps due to its ability to capture the differences of the means and the variancecovariance matrices of genotype scores in cases and controls simultaneously.
Twostage slidingwindow approach
Because the LRT_PC test may be more powerful than other multimarker tests, we wanted to use it to analyze the data set of GAW16 Problem 1. However, the LRT_PC can only be applied to a small chromosome region. To apply the LRT_PC to genomewide association studies, we propose a slidingwindow approach [7]. To use sliding windows, we divide all SNPs into contiguous overlapping windows and apply the LRT_PC in each window. Suppose that we use windows with a window size of S, then, all the SNPs can be divided into windows 1 to S, 2 to S + 1, 3 to S + 2, and so on.
Because we do not know the distribution or asymptotic distribution of the test statistic LRT_PC, we need to use a permutation approach to estimate the pvalue of the test. For a genomewide association study, the number of windows usually is more than 500,000 and the number of permutations usually is no less than 1000 (100,000 permutations were used in this study). The computation is not feasible for the slidingwindow approach discussed above. Thus, we propose a twostage approach. In the twostage approach, we split all individuals into two subsamples. In the first stage, by assuming that all individuals are genotyped at all SNPs, we use the first subsample to select R most promising SNP windows with the largest values of the LRT_PC statistic calculated via the first subsample. In the second stage, only the genotypes at SNPs within the R most promising windows are used. In this stage, we use the second subsample to assess P values for the R selected windows by permutations and claim significance by the falsediscovery rate (FDR) correction in Benjamini and Hochberg [9]. For the twostage approach, we only need to do permutations in the second stage. Thus, the twostage approach is computationally much more efficient than onestage approach.
To analyze the data set of GAW16 Problem 1 using LRT_PC based twostage slidingwindow approach, we use the following settings: window size is 5; the number of windows selected in the firststage, R, is 1000; the sample size of the first subsample is 15% of the total sample (15% cases and 15% controls). In the first stage, the number of PCs used in the LRT_PC test in each window is 5, i.e., we do not perform PC analysis. In the secondstage, the number of PCs used in the LRT_PC test in each window, K, is decided by the fact that the first K PCs can explain 85% of the total variability.
To choose the sample size of the first subsample, we did a power analysis based on a singlemarker test similar to that of Wang et al. [8]. Our results showed that the optimal value of the sample size of the first subsample is between 10% and 30% of the total sample. We use the results based on a singlemarker test as a reference to choose the sample size of the first subsample in this study (15% of the total sample).
As pointed by Skol et al. [10], our proposed twostage approach may be not as powerful as joint analysis. However, the results of Skol et al. also showed that when the sample size of the first subsample is small (15% of the total sample), the power difference between the twostage approach and joint analysis is also small. To compare the power of the twostage and onestage approaches, we have done a small scale simulation study (10,000 SNPs and 1000 permutations). The simulation results showed that when the first subsample is 15% of the total sample, the power difference between the two approaches is also small. In summary, compared with the joint analysis and onestage approach, our proposed twostage approach has a small power loss in exchange for a big increase in computational efficiency.
Results
The details of nonoverlapped significant windows
Window ID  Chr  Physical Location  Genes  References 

1  1  792429, 1071463  FAM87B, C1orf159  
2  1  149769454, 149786537  FCRL3  [13] 
3  2  139347733, 139370993  NXPH2  
4  2  158103376, 158124755  CYTIP  
5  2  192062916, 192117323  MYO1B  
6  2  193087360, 193117091  STAT4  [12] 
7  2  204487030, 204527849  
8  2  217410540, 217413523  
9  3  61980695, 61998337  PTPRG  
10  3  112000554, 112031324  
11  5  25934777, 25967191  
12  5  111055309, 111062116  
13  5  137614229, 137669929  GFRA3  
14  5  175278829, 175508504  
15  7  35309745, 35321578  
16  8  20380853, 20411350  
17  9  34654488, 34675940  CCL27  
18  9  84179401, 84220779  SLC28A3  
19  9  91845119, 91920813  
20  10  5016096, 5053944  
21  10  49685217, 49698112  WDFY4  
22  10  87994785, 88004329  GRID1  
23  11  67868248, 67886182  LRP5  
24  11  68807443, 68825321  
25  12  130500249, 130524477  hypothetical LOC116437  
26  13  73411181, 73419864  KLF12  
27  13  113652806, 113781019  FAM70B  
28  14  31921536, 31942503  AKAP6  
29  14  80898925, 80930993  STON2  
30  15  72678184, 72721493  
31  16  2452524, 2582219  C16orf59  
32  16  9214168, 9232588  
33  16  12651732, 12676977  
34  16  64715655, 64730155  
34  16  64715655, 64730155  
35  16  64735494, 64764278  
36  16  67451454, 67514126  TMCO7  
37  17  34194598, 34251205  PIP4K2B  
38  17  68919134, 68930918  SDK2  
39  19  41254897, 41282169  WDR62  
40  22  24746347, 24757608  MYO18B  
41  22  24760841, 24782942  
42  22  43076969, 43084289 
For validation purposes, we used the SNP Search Engine to find genes which contain or are near to SNPs that were discovered. We found rs2357135 nesting at "2q32", which is near to gene "STAT 4" and thus supported the finding in Remmers et al. [12]. In addition, our discoveries supported the findings of FCRL3 on chromosome 1 [13] and HLA region (MHC) on chromosome 6 [1]. Additionally, we detected many novel risk windows for followup studies. Except for chromosomes 4 and 18, each of the other autosomes was found to carry risk windows. For example, we detected SNP rs2047465 which nests in gene SDK2 on chromosome 17.
Discussion
In this article, we proposed a twostage slidingwindow approach and, in each window, our recently proposed LRT_PC test was applied to test the association between a window and a disease. Then, we applied the method to GAW16 Problem 1 to detect risk windows for RA. Different existing RA association studies discovered diverse susceptibility genes on all chromosomes except for MHC candidate genes on chromosome 6. Our analysis supported the findings of some RA susceptibility genes that have been identified to be associated with RA in the literature.
It is intractable to formulate the null distribution of the LRT_PC statistic and thus the permutation approach must be applied to evaluate the pvalues of top R promising windows (R = 1000 was used in this study). For a given R, the number of permutations must be large enough to obtain accurate pvalues. Thus, the choice of R partially depends on computational capacity. Further efforts are needed to determine the optimal value of R.
In this study, we used 5 PCs, i.e., we did not perform PC analysis in LRT_PC test for each window of size 5 in the first stage. We did not perform PC analysis (or use 5 PCs) in the first stage for two reasons. One is that we have to use the same number of PCs in different windows so that the values of the test statistic in different windows are comparable. The other is that in the first stage, we only use the test statistic to rank the SNP windows and the results are similar when using different numbers of PCs (results not shown). Another remaining question regarding the proposed method is how to choose the window size. Most often, researchers use windows of size 3. In the LRT_PC test, we use PC analysis to reduce the dimension, and thus we can use a larger window size (i.e., window size of 5).
Conclusion
In this article, we proposed a twostage slidingwindow approach to detect the association between SNP windows and a disease status. Application to GAW16 Problem 1 data set illustrated its practical advantages. Our results supported the findings of several genes which were identified to be responsible for RA in the literature. We also discovered several additional SNP windows for followup studies.
List of abbreviations used
 FDR:

Falsediscovery rate
 GAW16:

Genetic Workshop Analysis 16
 HLA:

Human leukocyte antigen
 LD:

Linkage disequilibrium
 LRTPC:

Likelihood ratio testprincipal component
 MHC:

Major histocompatibility complex
 PC:

Principal component
 RA:

Rheumatoid arthritis
 SNP:

Singlenucleotide polymorphism
Declarations
Acknowledgements
The Genetic Analysis Workshops are supported by NIH grant R01 GM031575 from the National Institute of General Medical Sciences. This work was supported by NIH grants R01 GM069940 and the OverseasReturned Scholars Foundation of Department of Education of Heilongjiang Province (1152HZ01).
This article has been published as part of BMC Proceedings Volume 3 Supplement 7, 2009: Genetic Analysis Workshop 16. The full contents of the supplement are available online at http://www.biomedcentral.com/17536561/3?issue=S7.
Authors’ Affiliations
References
 Newton JL, Harney SM, Wordsworth BP, Brown MA: A review of the MHC genetics of rheumatoid arthritis. Genes Immun. 2004, 5: 151157. 10.1038/sj.gene.6364045.View ArticlePubMed
 Xiong M, Zhao J, Berwinkle E: Generalized T ^{2} test for genome association studies. Am J Hum Genet. 2002, 70: 12571268. 10.1086/340392.PubMed CentralView ArticlePubMed
 Fan R, Knapp M: Genome association studies of complex diseases by casecontrol designs. Am J Hum Genet. 2003, 72: 850868. 10.1086/373966.PubMed CentralView ArticlePubMed
 Zaykin DV, Westfall PH, Young SS, Karnoub MA, Wagner MJ, Ehm MG: Testing association of statistically inferred haplotypes with discrete and continuous traits in samples of unrelated individuals. Hum Hered. 2002, 53: 7991. 10.1159/000057986.View ArticlePubMed
 Wang T, Zhu X, Elston RC: Improving power in contrasting linkagedisequilibrium patterns between cases and controls. Am J Hum Genet. 2007, 80: 911920. 10.1086/516794.PubMed CentralView ArticlePubMed
 Wang X, Zhang S, Sha Q: A new association test to test multiplemarker association. Genet Epidemiol. 2009, 33: 164171. 10.1002/gepi.20369.PubMed CentralView ArticlePubMed
 Yang H, Lin C, Fann C: A slidingwindow weighted linkage disequilibrium test. Genet Epidemiol. 2006, 30: 531545. 10.1002/gepi.20165.View ArticlePubMed
 Wang H, Thomas DC, Peer I, Stram DO: Optimal twostage genotyping designs for genomewide association scan. Genet Epidemiol. 2006, 30: 356368. 10.1002/gepi.20150.View ArticlePubMed
 Benjamini Y, Hochberg Y: Controlling the false discovery rate: a practical and powerful approach to multiple testing. J Roy Stat Soc Ser B. 1995, 57: 289300.
 Skol AD, Scott LJ, Abecasis GR, Boehnke M: Joint analysis is more efficient than replicationbased analysis for twostage genomewide association studies. Nat Genet. 2006, 38: 209213. 10.1038/ng1706.View ArticlePubMed
 Browning SR, Browning BL: Rapid and accurate haplotype phasing and missingdata inference for wholegenome association studies by use of localized haplotype clustering. Am J Hum Genet. 2007, 81: 10841097. 10.1086/521987.PubMed CentralView ArticlePubMed
 Remmers EF, Plenge RM, Lee AT, Graham RR, Hom G, Behrens TW, de Bakker PI, Le JM, Lee HS, Batliwalla F, Li W, Masters SL, Booty MG, Carulli JP, Padyukov L, Alfredsson L, Klareskog L, Chen WV, Amos CI, Criswell LA, Seldin MF, Kastner DL, Gregersen PK: STAT4 and the risk of rheumatoid arthritis and systemic lupus erythematosus. N Engl J Med. 2007, 357: 977986. 10.1056/NEJMoa073003.PubMed CentralView ArticlePubMed
 Kochi Y, Yamada R, Suzuki A, Harley JB, Shirasawa S, Sawada T, Bae SC, Tokuhiro S, Chang X, Sekine A, Takahashi A, Tsunoda T, Ohnishi Y, Kaufman KM, Kang CP, Kang C, Otsubo S, Yumura W, Mimori A, Koike T, Nakamura Y, Sasazuki T, Yamamoto K: A functional variant in FCRL3, encoding Fc receptorlike 3, is associated with rheumatoid arthritis and several autoimmunities. Nat Genet. 2005, 37: 478485. 10.1038/ng1540.PubMed CentralView ArticlePubMed
Copyright
This article is published under license to BioMed Central Ltd. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.