Linkage Disequilibrium and Haplotype Analysis of COX‐2 and Risk of Colorectal Adenoma Development

Abstract

Single nucleotide polymorphisms (SNPs) in the promoter and untranslated region of cyclooxygenase (COX)‐2, an inducible enzyme responsible for the synthesis of prostaglandins, have been reported to modulate the risk for many human cancers. We performed comprehensive linkage disequilibrium (LD) and haplotype analyses of 13 single nucleotide polymorphisms of the COX‐2 gene and examined its susceptibility to adenoma development in 72 African American cases and 142 controls. Results revealed significant variation in LD patterns with consequence for adenoma development. Two distinct haplotype blocks were identified; one block covered the coding regions of exon 1, introns and a section of the 3′‐unstranslated region (3′‐UTR), whereas the second block resided solely in the 3′‐UTR region. A haplotype in block 1 increased the risk of adenoma development by threefold (odds ratio [OR]= 2.9, confidence interval [CI]= 1.8–3.7, P= 0.002). Regression analysis showed, increase in copies of minor alleles of 6,064(T>C) polymorphism associated with increased odds of adenoma development by 80% (OR = 1.80, CI = 1.09–3.21, P= 0.034), 10,848(G>A) by 84% (OR = 1.84, CI = 1.05–3.23, P= 0.034) and 10,935(A>G) by 32% (OR = 1.32, CI = 1.12–3.69, P= 0.036). These results support the hypothesis that COX‐2 gene might play a role in the etiology of colon cancer and warrant further investigation in other cancers. Besides, these variations should be taken into account for disease‐based association studies in which the COX‐2 polymorphism is considered as a candidate gene. Clin Trans Sci 2012; Volume 5: 60–64

Introduction

Single nucleotide polymorphisms (SNPs), in the promoter and 3′‐untranslated region (3′‐UTR) of cyclooxygenase (COX)‐2 have been reported to modulate the risk for many human cancers, including esophageal, ¹ breast, ² prostate, ³ colorectal, ⁴ , ⁵ , ⁶ gastric, ⁷ bladder, biliary tract, ⁸ and lung cancer. ⁹ , ¹⁰ , ¹¹ Inflammation is emerging as a major area of interest in the development and progression of many of these cancers, including colorectal cancer. COX‐2, an inducible enzyme, responsible for the synthesis of prostaglandis, plays a critical role in inflammation activity. ¹² , ¹³ This overproducing genetic variant of the genes seems worthy of future studies in several human pathologies that are mediated by the inducible activity of the isoenzyme.

Colorectal cancer is a multifactorial disease with complex etiology and studies have shown the role genetics play in its susceptibility. ⁵ , ¹⁴ , ¹⁵ SNPs are linked within a gene locus and are inherited as several gene variants of different functional properties. Numerous SNPs, have been identified in COX‐2 gene and some recent studies has reported possible association between inflammation, COX‐2 gene and adenoma development. ¹⁵ , ¹⁶ , ¹⁷ Yet these studies focus on just a small number of SNPs. Moreover, little is known about the role that sequence variation and mutation of the COX‐2 gene may contribute to the risk of colorectal adenoma development. Due to the high degree of linkage disequilibrium (LD) observed between SNPs within genomic blocks, ancestral disease variants may be uncovered through evaluation of the underlying haplotypes. Haplotype‐based association studies have, therefore, been proposed as a comprehensive approach to identify causal genetic variation underlying complex diseases and have become an important tool in clinical studies. ¹⁸ , ¹⁹ Recently, we reported a possible influence of allelic variant of the COX‐2 gene on the risk of colorectal adenoma development. ¹⁵ To expound and facilitate future genetics studies of this important gene to development of colorectal adenoma and other cancers, we performed comprehensive LD and haplotype analyses of 13 different SNPs of the COX‐2 promoter gene.

Materials and Methods

Study population

African American patients referred for colonoscopy to the Division of Gastroenterology, Howard University Hospital, between September 2000 and October 2003 were studied. Details of the study have been previously described. ¹⁵ In brief, the purpose of the study was explained to the patients and, those committed to participate signed an informed consent approved by Howard University Institutional Review Board before the colonoscopy. Patients were eligible as cases if colonoscopy resulted in a first diagnosis of colorectal adenomatous or hyperplastic polyp as confirmed by histology. Patients were eligible as controls if they were free from all polyps and had no history of colorectal adenomas/cancer. Patients with a history of inflammatory bowel disease, malabsorption, any cancer, history of chemotherapy, and interferon treatment were excluded. Seventy‐two patients qualified as cases and 146 as controls.

DNA extraction

A brief sample of blood was collected by venipuncture into EDTA glass tubes for RBC and buffy coat was kept for DNA extraction for analysis of COX‐2 polymorphism. Blood samples were centrifuged at 2,000 g for 10 minutes and blood components (serum, plasma, and buffy coat) were separated manually. DNA was isolated by using a simple proteinase K treatment, according to manufacturer recommendation (Qiagen, Alameda, CA, USA).

Genotyping of polymorphisms

Genotyping was as described previously. ¹⁵ Thirteen haplotype‐tagging SNPs in the COX‐2 gene were selected and genotyped. All SNPs, except for one, were from the regulatory regions distributed over the entire gene (three in the promoter region at positions −466, −663, and −861; one in exon 1 at 2,331; two in intron 5 at 5,072 and 5,229; one in intron 6 at 5,625; one in intron 7 at 6,064; five in the 3′‐UTR of exon 10 at 8,334, 8,494, 10,494, 10,848, and 10,935; Figure 1 ). All 13 SNPs were genotyped in all study participants. All SNPs were tested using Taqman assay with MGB chemistry (Applied Biosystems, Inc., CA, USA).

Figure 1.

Structure and location of the 13 polymorphisms at the PTGS2 locus. Positions in base pairs as per GenBank sequence D28, 235. Exons are shown as dark boxes, untranslated regions as light boxes. Vertical lines represent SNPs, with their positions in the sequence indicated.

Statistical analysis

Genotype and allele frequencies were estimated by gene counting. Hardy–Weinberg equilibrium (HWE) test were performed for each SNP to determine deviations of observed genotypic distribution from expected distribution. We used the software Haploview [Version 2.1.1] to determine the LD structure of the SNPs within the COX‐2 gene and to test for HWE.

Haplotype construction and LD analysis

Haplotype of all possible SNP pairs were constructed using Haploview and the haplotype frequencies estimated by Permutation methods. ²⁰ , ²¹ The nonrandom distributions of the haplotype were assessed separately for cases and controls by calculating pairwise LD coefficient, D′. In large samples, D′= 1 indicates complete LD, that is, no evidence for recombination between the SNP pairs—such SNPs are good surrogates for each other; D′= 0 indicates no LD. “Strong” LD was defined as having pairwise D′ > 0.85. Haplotype block structure was examined using the criteria of Gabriel et al., ¹⁸ which utilizes the 90% confidence bounds of D′ to define sites of historical recombination between SNPs. Patterns of LD are visualized using Haploview.

Logistic regression analyses

We used conditional logistic regression models to examine the effects of COX‐2 gene on risk of adenoma development adjusting for age and gender. To avoid ramification of multicollinearity, because SNPs that are in LD are highly correlated and serve as good surrogates for each other, only SNPs that exhibit strong evidence of recombination with each other were used in the regression models. For each identified SNP, we modeled a participant's genotype under an additive model that is equivalent to coding of the number of copies of the minor allele (reference allele) that the participant possesses. Several models, each corresponding to a “set of SNPs” that mutually exhibit strong evidence of recombination with one another within the set were fitted and the final (most parsimonious) model was selected based on the Akaike's information criterion (AIC), defined as AIC =−2 × log(likelihood) + 2 × (number of parameters). The model with the minimum AIC fits the data best.

Results

In summary, LD patterns vary substantially between subjects susceptible to adenoma development and controls.

Allelic associations

Data consisted of 218 African American individuals, 56% males, with mean † SD age 54.9 † 10.6 years. All SNPs used in the study were identified previously ¹⁵ to be tag SNPs and captured most common variations in the African American population. The minor alíele frequency (MAF) at each marker ranged between 8% and 49%. Table 1 shows alíele frequencies and Hardy–Weinberg Test P values. All 13 SNPs, except SNP466 (rs689,462) were in Hardy‐Weinberg equilibrium (P= 0.02) for the combined data as well as for the cases and control subgroups. All 13 SNPs genotyped exhibited heterozygosity.

Table 1.

Characteristics of the SNP in the promoter and coding region of the cyclooxygenase (COX)‐2 gene and their minor alíele frequencies (MAF).

Db SNP rs no. cluster ID	Region	Position	MAF (All)	MAF Cases	MAF Controls	Hardy‐ Weinberg test, P value
rs689,462	Promoter	466(A>C)	0.24	0.21	0.26	0.015
rs689,464	Promoter	663(GT>del)	0.26	0.22	0.27	0.563
rs20,415	Promoter	861 (G>A)	0.08	0.07	0.08	0.575
rs2,745,557	Exon 1	2,331 (C>T)	0.16	0.13	0.18	0.202
rs4,648,274	Intron 5	5,072(A>C)	0.14	0.12	0.15	0.368
rs20,432	Intron 5	5,229(G>T)	0.44	0.44	0.45	0.669
rs2,066,826	Intron 6	5,625 (G>A)	0.35	0.36	0.34	0.370
rs4,648,276	Intron 7	6,064(T>C)	0.14	0.11	0.15	0.656 0.095
rs4,648,291	Exon 10	8,344(TTATA>del)	0.40	0.42	0.38	0.381
rs5,275	Exon 10	8,494 (T>C)	0.36	0.36	0.37	0.381
rs689,470	Exon 10	10,494(T>C)	0.37	0.36	0.37	0.422
rs4,648,306	Exon 10	10,848(G>A)	0.14	0.11	0.16	0.362
rs4,648,308	Exon 10	10,935(A>G)	0.47	0.49	0.46	0.122

MAF = minor alíele fre quency; A>B implies B is the minor alíele.

Haplotype and LD analysis

LD across the 13 functional polymorphisms was examined for the combined data and separately for cases and controls and the corresponding estimated D′ values depicted in a color scheme are shown in Figure 2 . For the combined data, we observed that 12 SNP pairs showed strong evidence of recombination ( Figure 2A ‐white shades). Of these, we identified four distinct sets of “family of SNPs” that mutually exhibit strong evidence of recombination with one another within the family. These are, I ={rs689,462, rs689,464, rs2,066,827, rs4,648,291}, II ={rs689,462, rs689,464, rs4,648,276}, III ={rs689,462, rs689,464, rs4,648,306}, and IV ={rs2,066,827, rs4,648,291, rs4,648,308}.

Figure 2.

Linkage disequilibrium (LD) patterns in (A) Combined data (B) Adenoma cases and (C) Controls. LD is estimated by pairwise LD coefficient, D′. The color scheme is dark gray for “strong” LD, light gray for “uninformative” and white for “evidence of historical recombination”. Patterns of LD are visualized using Haploview.

As portrayed, the pattern of LD in the cases differed substantially from that of the controls. In particular, “strong” LD was observed in 47% (37 of 78) SNP pairs ( Figure 2B ; dark gray shades) in the cases compared with 82% (50 of 78; Figure 2C ) in the controls. The lower than expected degree of LD in the cases even between closely spaced SNPs might be caused by gene conversion. ²² Six SNP pairs in the cases showed strong evidence of historical recombination ( Figure 2 ; white shades) compared with 11 in the controls. Interestingly, of the 11 SNP pairs in the controls that showed evidence of recombination only one pair (rs689,462 and rs689,464), both markers in the promoter region were neighboring SNPs. Furthermore, the markers (rs689,462) and (rs689,464) also showed evidence of recombination with five other markers: (rs2,066,826), (rs4,648,276), (rs4,648,291), (rs4,648,306), and (rs4,648,308). In contrast, among the cases, marker (rs689,462) and (rs689,464) showed evidence of recombination with only marker (rs2,066,826). Thus, the results of LD support the high‐risk haplotype for adenoma development.

Haplotype blocks

Two distinct haplotype blocks were identified; one block covered the coding regions of exon 1, introns and a section of the 3′‐UTR, whereas the second block resided solely in the 3′‐UTR region ( Table 2 ). A haplotype in block 1 associated with a threefold increased risk of adenoma development (OR = 2.9, CI = 1.8–3.7, P= 0.002). There was no association of any haplotype in block 2 with adenoma development. We previously reported ¹⁵ significant inverse association between the heterozygous genotype at the rs20,432 SNP in intron 5 and rs4,648,308 SNP in exon 10. The human genome is comprised of genomic blocks that display little evidence of historical recombination and low haplotype diversity. ¹⁸ , ¹⁹ Analysis showed evidence of historical recombination in 7.7% of SNP pairs in cases compared with 14.1% in controls. Although this data provide little support that there is a strong adenoma susceptibility allele at COX‐2, they do suggest the intronic region may harbor a variant that modulate disease risk. Fedorova and Fedorov ²² had reported that intronic variants may modulate disease risk by regulating gene expression, gene splicing, or transcript stability.

Table 2.

Association between haplotypes in linkage disequilibrium (LD) blocks of COX‐2 and adenoma development.

Haplotype frequency
Haplotype+	All data	Cases, Control	Chi‐square	P value
Block 1
CATGT(T)TT	0.204	0.222,0.194	0.459	0.4983
CAGGT(A)CC TATGT(T)TT	0.182 0.158	0.171,0.188 0.135,0.171	0.168 0.892	0.6815 0.3451
CCGAT(A)CC CAGAC(T)CC	0.144 0.138	0.126,0.153 0.114,0.150	0.531 1.005	0.4664 0.3162
CATGT(T)CC CAGAT(A)CC	0.081 0.071	0.079, 0.082 0.124,0.043	0.014 9.216	0.9053 0.0024
CAGGT(T)CC	0.020	0.029,0.016	0.757	0.3842
Block 2
GG	0.473	0.494, 0.462	0.358	0.5498
GA	0.386	0.392, 0.382	0.036	0.8490
AA	0.141	0.114,0.155	1.264	0.2609

+(T) = TTATT, (A) = del; block 1 covered the coding regions of exon 1, introns, and a section of the 3’ untranslated region (3′‐UTR), whereas block 2 resided solely in the 3′‐UTR region.

Logistic regression analysis

For the combined data, we identified four distinct sets of “family of SNPs” that mutually exhibit strong evidence of recombination with one another within the family (See subsection, Haplotype and LD Analysis). Results of separate conditional logistic regression models corresponding to the four distinct set of “family of SNPs” are shown in Table 3 . Analysis showed Model III with the minimum AIC, though not much different from that of II and IV. We, therefore, discuss estimates from these models. Results showed allelic association with adenoma development. In particular, an increase in copies of minor alleles of 6,064(T>C) [rs4,648,276] polymorphism associated with increased odds of adenoma development by 80% (OR = 1.80, CI = 1.09–3.21, P= 0.034), 10,848(G>A) [rs4,648,306] by 84% (OR = 1.84, CI = 1.05–3.23, P= 0.034) and 10,935(A>G) [rs4,648,308] allele by 32% (OR =, CI =,1.12–1.69, P= 0.036). All three SNPS, reside in the 3′‐UTR region. There was no significant detectable interaction with age or sex and the SNPs.

Table 3.

Conditional logistic regression analysis for SNPs that exhibit strong evidence of recombination with one another. Additive genetic models, equivalent to coding of the number of copies of the minor alíele (reference alíele) that the participant possesses were fitted.

	Model I*	Model II	Model III	Model IV
SNP(ma)	Odds ratio [CI]	Odds ratio [CI]	Odds ratio [CI]	Odds ratio [CI]
rs6,894,62 466(A>C)	0.97 [0.46–2.06]	1.10 [0.55–2.22]	1.11 [0.54–2.23]
rs689,464 663(GT>del)	1.59 [0.75–3.39]	1.54 [0.74–3.24]	1.54 [0.74–3.12]
rs2,066,826 5,625(G>A)	1.01 [0.71–1.68]			1.24 [0.83‐l.84]
rs4,648,276 6,064(T>C)		1.8 [1.09–3.21]a
rs4,648,291 8,344(TTATA>del)	1.13 [0.74–1.72]			1.19 [0.81–1.72]
rs4,648,306 10,848(G>A)			1.84 [1.05–3.23]a
rs4,648,308 10,935 (A>G)				1.32 [1.12–3.69]a
−2 × loge(likelihood)	261.52	258.18	257.29	259.25
AIC	269.52	264.18	263.29	265.25

AIC = Akaike's information criterion =−2 × log_e(likelihood) + 2 × (no. of parameters).

^a P < 0.05.

*Each model contains SNPs that mutually exhibit strong evidence of recombination with one another.

Discussion and Conclusion

The COX‐2 pathway has been recognized to be important in the development and progression of human cancer. ²³ Numerous studies have suggested a major role for COX‐2 in the initiation, promotion, and progression of cancers in different organs. ²⁴ , ²⁵ , ²⁶ However, little is known about the role of sequence variation within COX‐2 in colorectal cancer. Recently, we reported a possible influence of allelic variant of the COX‐2 gene on the risk of colorectal adenoma development. LD makes tightly linked variants strongly correlated, producing cost savings for association studies. In the present study, we performed a comprehensive LD and haplotype analyses of 13 SNPs in the regulatory regions over the entire COX‐2 gene and examine possible interaction of the SNPs to the risk of adenoma development. Results revealed significant LD distributional patterns with probable consequences for colorectal adenoma development. These findings may indicate potential epistasis of the COX‐2 gene may play a role in mediating susceptibility to colorectal adenoma development. Fedorova and Fedorov ²² had reported that intronic variants may modulate disease risk by regulating gene expression, gene splicing, or transcript stability. Because COX‐2 activity plays a vital role in inflammation, this overproducing genetic variant of the genes seems worthy of future studies in several human pathologies that, apart from colorectal cancer, are mediated by the inducible activity of the isoenzyme.

Although there are few studies that have investigated haplotype patterns and its association to adenoma development, these studies examined only a small number of functional SNPs. To the best of our knowledge, this is one of limited studies that have thoroughly examined LD and haplotype patterns in a moderately large candidate SNPs and its association to colorectal adenoma development and the only one in an African American population. Probable limitations of the study include the inclusion of multiple SNPs for a relatively small sample size, and the fact that this study was performed in a single population from a single center and, therefore, warrants confirmation in a larger and independent group of subjects.

In conclusion, our findings indicate that LD patterns vary substantially between subjects susceptible to adenoma development and controls. Although additional studies in independent populations replicating these findings are desirable, these variations should be taken into account for disease‐based association studies in which the COX‐2 gene is considered as a candidate.