Genetic polymorphisms in vitamin D receptor VDR/RXRA influence the likelihood of colon adenoma recurrence

Abstract

Low circulating levels of vitamin D have been implicated in colorectal cancer risk. The biological actions of the hormonal form of vitamin D, 1,25(OH)₂D₃, are mediated by the vitamin D receptor (VDR), which heterodimerizes with retinoid X receptors (RXRs). We applied a single nucleotide polymorphism (SNP) tagging approach to assess the association between genetic variation in RXRA and VDR and odds of metachronous colorectal neoplasia after polypectomy in a pooled population of two studies. A total of 32 tagSNPs in RXRA and 42 in VDR were analyzed in 1,439 participants. A gene-level association was observed for RXRA and any (P = 0.04), and proximal (P = 0.03) metachronous neoplasia, but no gene-level associations were observed for VDR. Although no single SNP in VDR was related to any metachronous adenoma after correction for multiple comparisons, the association between RXRA SNP rs7861779 and proximal metachronous neoplasia was of borderline statistical significance (OR=0.68; 95% CI=0.53–0.86; unadjusted p=0.001; adjusted p=0.06). This finding was observed in both of the individual studies when examined independently. Haplotypes within linkage blocks of RXRA support an approximately 30% significant reduction in odds of metachronous neoplasia arising in the proximal colon among carriers of specific haplotypes, which was strongest (OR_proximal = 0.67; 95% CI, 0.52 – 0.86) for carriers of a CGGGCA haplotype (rs1805352, rs3132297, rs3132296, rs3118529, rs3118536, rs7861779). These results represent a novel finding indicating that allelic variation in RXRA is related to metachronous colorectal neoplasia, and may be of particular importance in the development of proximal lesions.

Introduction

A protective effect of vitamin D on colorectal cancer (CRC) was first proposed in 1980 (1). Subsequent epidemiological studies have reported significant inverse associations between measured blood levels of 25(OH)D, a biomarker used to approximate systemic vitamin D levels, and either colorectal cancer (2–6) or its precursor, colorectal adenoma (7). Several mechanisms of action for the putative anti-tumor properties of vitamin D have been proposed from extensive cell culture studies. The hormone metabolite of vitamin D, 1,25(OH)₂D₃, has been shown to play a key role in the maintenance of cellular proliferation (8) and differentiation (9, 10); modulation of the cell cycle (11), and regulation of members of the Bcl-2 family, which are involved in apoptosis (10, 12). 1,25(OH)₂D₃ has also been demonstrated to upregulate E-cadherin, an important component of cellular adhesion (13, 14) resulting in the translocation of β-catenin from the nucleus to the plasma membrane (13, 15).

1,25(OH)₂D₃ mediates its action as a ligand by binding to the vitamin D receptor (VDR), which commonly forms a heterodimer with retinoid X receptors (RXRs) such as RXRα, an isoform of RXR which has been implicated in colorectal carcinogenesis (16). Formation of VDR-RXR heterodimers releases co-repressor proteins and recruits coactivators, resulting in increased VDR transcriptional activity at vitamin D-responsive elements (VDREs) in VDR target genes (17), including those involved in the regulation of vitamin D metabolites (18). Both VDR and RXRα are members of the steroid nuclear receptor superfamily (19), whose members and their ligands have been identified as potential targets for the prevention and treatment of several different cancers (17, 20). To date, the biological role of RXRα in relation to cancer has not been investigated as thoroughly as that of VDR, although RXR binding is absolutely required for transcriptional activation by VDR (21), and RXRα dysregulation has been identified to have carcinogenic effects in the colon (16). Thus, RXRα likely has a major regulatory role in pathways related to VDR including effects on 1,25(OH)₂D₃-mediated anti-neoplastic activities in the colon.

Although a number of studies have evaluated limited candidate SNPs in the VDR gene for their association with the development of colorectal adenoma (22–26), none have used a tag SNP approach (27) to evaluate allelic variation in VDR and odds of metachronous colorectal adenoma. In addition, despite the clear biologic coupling of RXRα and VDR in mediating 1,25(OH)₂D₃ activity at the cellular level, association of genetic variation in RXRA with cancer outcomes is limited to studies of prostate and biliary cancers (28, 29). Therefore, in the current work, a tag SNP approach was employed to probe common genetic variation in the RXRA and VDR genes as well as to construct haplotype blocks where appropriate to determine the role of these genes in the development of metachronous colorectal neoplasia.

Methods

Subjects were drawn from participants of the Wheat Bran Fiber (WBF) and Ursodeoxycholic Acid (UDCA) trials conducted at the University of Arizona as described previously (30, 31). The WBF study was conducted to determine the effect of a high-fiber versus a low-fiber cereal supplement on metachronous colorectal neoplasia among participants in a randomized, double-blind clinical trial (30). Participants in the study had at least one histologically confirmed colorectal adenoma removed no more than 3 months before entry into the trial and were between 40 and 80 years of age. A follow-up colonoscopy was completed in 1310 participants (32) and no effect of the fiber supplement on metachronous neoplasia was observed. The UDCA trial was a phase III, double- blind, placebo controlled study conducted to determine the effect of UDCA on colorectal metachronous neoplasia (31). A total of 1,192 participants completed the UDCA trial (31) with no effect of UDCA on metachronous neoplasia. The University of Arizona Human Subjects Committee and Institutional Review Board approved both the WBF and UDCA trials.

Endpoint ascertainment

Metachronous colorectal neoplasia was defined as adenomas or cancers (n=7) detected by colonoscopy at least six months after randomization to the parent trials. In the past, these lesions have been defined as ‘recurrences’; however, due to the possibility of some of the ‘recurrent’ lesions having been those missed at baseline colonoscopy, it was recommended that the terminology be changed from ‘recurrent’ to ‘metachronous’. Personnel at each study site reviewed endoscopy and pathology reports and extracted data regarding size, histology, number, and location, followed by central pathology review at each site. Lesions were classified as proximal if they were located at or proximal to the splenic flexure; those distal to the splenic flexure, including the rectum, were categorized as distal.

SNP Selection and Genotyping

SNP selection for this platform utilized Haploview Tagger to identify bin tags from a European Caucasian (CEU) population. Initial tag SNPs and LD blocks were identified from HapMap data release #16c.1, June 2005, on NCBI B34 assembly, dbSNP b124. Tag SNPs from these data were identified utilizing the following criteria: Minor allele frequency > 5%; pairwise r² > 0.95; and at least 60 base pairs between neighboring SNPs (33, 34). SNPs located at the 5’ and 3’ ends of an LD block were also included. SNPs with little or no linkage disequilibrium were selected from HapMap or dbSNP at a density of 1 per kb. In addition to this tag SNP selection strategy, high-interest SNPs that have been reported in the literature were also chosen, including BsmI (rs1544410), TaqI (rs731236), and FokI (rs2228570) restriction endonuclease sites, as well as the Cdx-2 binding site in the promoter region (rs11568820).

Genotyping of the samples was performed on the Illumina Golden Gate platform (Illumina^®, San Diego, CA). Briefly, DNA was activated with streptavidin/biotin and added to a hybridization mixture. After hybridization, the samples were washed followed by extension, ligation and cleanup. Universal primers labeled with Cy3 or Cy5 were utilized for PCR labeling of the DNA. Next, the labeled DNA was allowed to hybridize with the Sentrix Array Matrix (SAM) and finally, placed in the BeadArray Reader for quantitation of the fluorescence signal. Analysis of the output from the fluorescence reading was managed using Bead Studio software (Illumina^®, San Diego, CA). SNPs were considered to have failed genotyping if they met at least one of the following criteria: Illumina GenTrain Score < 0.4; 10% GC Score < 0.25; AB T Dev > 0.1239, call frequency < 0.95; intra-plate replicate errors > 2; parent-parent-child errors > 2; or discordance with HapMap > 3. Participants were genotyped for a total of 49 SNPs in VDR and 41 in RXRA.

The VDR FokI polymorphism (rs2228570, 27823C>T) failed both the Illumina GoldenGate™ and the Sequenom Iplex™ technologies. We genotyped the sample set for high-interest SNPs that failed the Illumina GoldenGate™ platform using the GenomeLab™ SNPStream 12-plex technology from Beckman Coulter at the University of Arizona Genetics Core; because FokI is a commonly studied SNP and is not in high LD with any other SNP, it was included in the SNPStream platform. This technology couples base extension chemistry with a subsequent hybridization step to a glass slide. Genotype data passed quality control measures if they met the following criteria: distributed into three statistically significant clusters, failed to generate genotypes from all 181 blank wells, showed Mendelian consistency for Coriell CEPH trios, and showed >97% concordance among lab blinded replicates. Any samples that exhibited low fluorescence intensity or generated an ambiguous genotype call were classified as failed and not included in the data. Genotype frequencies were found to be in Hardy-Weinberg equilibrium.

Statistical Analysis

Participants from the WBF and UDCA trials with both genetic data and trial endpoint data available were pooled, yielding 1,530 participants. Of these, a total of 91 participants reported a race/ethnicity other than white. Since the trials did not have adequate numbers of the other racial/ethnic groups to appropriately address the issue of population stratification, all genetic analyses were limited to the white participants, yielding a final sample size of 1,439 individuals. Prior to the statistical analysis, all SNPs which failed the criteria listed above, were monomorphic, or had an excessive heterogeneity score were excluded from the data set (VDR = 7; RXRA = 9), leaving at total of 42 VDR SNPs and 32 RXRA SNPs for inclusion in the final statistical analyses.

We used a multi-stage approach to test overall association of a gene via principal component analysis, finer resolution with a single SNP analysis, and characterization of effects via haplotype analysis. Principal components (PCs) were generated to capture the set of available SNPs within each gene (32 on RXRA, 42 on VDR) where the PCs are linear transformations of the original SNP data. The PCs were modeled using logistic regression, utilizing an 80% explained-variance threshold in determining how many PCs to include in the models (35). Three different outcomes were modeled for each gene: any metachronous colorectal neoplasia, proximal metachronous neoplasia, and distal metachronous neoplasia. Proximal and distal lesions were evaluated as distinct outcomes based on a priori evidence that these may represent biologically disparate pathologies with potentially distinct genetic etiologies (36). Exclusion of rectal lesions from the distal endpoint did not result in any material changes to the point estimates; hence, they were included as distal colorectal lesions. A p-value for the overall gene-outcome association was obtained from a likelihood ratio test comparing the model with PCs versus an intercept-only model, with degrees of freedom equal to the number of PCs.

Individual SNPs were evaluated using logistic regression models, utilizing additive, dominant, and recessive modes of inheritance due to the lack of existing evidence regarding the true mode of inheritance for these SNPs. Preliminary statistical models were adjusted for age and sex; however, these covariates had no meaningful effect on the estimates, and crude models are presented in the Results. After obtaining uncorrected p-values for these associations, a multiple comparisons adjustment was utilized that is specifically designed for correlated tests due to linkage disequilibrium and the exploration of different modes of inheritance (37).

Linkage disequilibrium plots were generated using the mapLD package available as part of the R project (http://www.r-project.org) to identify haplotype blocks. Haplotypes within each block were estimated with fastPHASE version 1.2 using default iteration settings (38). Haplotypes were evaluated for associations with any metachronous colorectal neoplasia as well as by colorectal sub-site-using logistic regression to obtain odds ratios and 95% confidence intervals. An overall p-value for each haplotype block was obtained with a likelihood ratio test. All statistical analyses were conducted with R version 2.7.2 or SAS version 9.1. All statistical tests were two-sided and considered significant at a value of p<0.05. D’ values for SNPs were determined with Haploview 4.1 (http://www.broad.mit.edu/mpg/haploview/). Significant SNPs and haplotypes identified in the uncorrected analysis were evaluated separately in WBF and UDCA to assess heterogeneity between the two populations.

Results

Characteristics of the genotyped participants included in the pooled analysis of metachronous neoplasia are presented in Table 1. The mean age of the pooled study population was 66.2 ± 8.2, with 66.7% of the participants being male, 24.9% reporting a family history of colorectal cancer, and 43.7% having had a polyp prior to the baseline examination. At follow-up, 45.9% of the participants had a metachronous adenoma; of these, 33.1% had proximal lesions and 24.0% had distal neoplasia, with the remainder having had adenomas at both colon sub-sites. In order to evaluate the overall gene-level association of VDR and RXRA, PC analysis was conducted with any, proximal, and distal metachronous neoplasia as the primary endpoints (Table 2). At the gene level, RXRA was significantly associated with any (P = 0.04) and proximal (P = 0.03), but not with distal metachronous neoplasia. No significant association was observed at the gene level for VDR and metachronous neoplasia.

Table 1.

Characteristics of the pooled population, WBF and UDCA.

Characteristics	WBF	UDCA	Pooled
	n=611	n=828	n=1,439
Mean age, y ± SD	66.0 ± 8.2	66.4 ± 8.2	66.2 ± 8.2
Sex, Male, n (%)	403 (66.0%)	560 (67.6%)	963 (66.9%)
Family history of colorectal cancer1, n (%)	113 (19.9%)	234 (28.3%)	347 (24.9%)
Previous polyps2, n (%)	228 (41.2%)	355 (45.5%)	583 (43.7%)
Any metachronous neoplasia3, n (%)	313 (51.2%)	347 (41.9%)	660 (45.9%)
Proximal metachronous neoplasia, n (%)	225 (37.3%)	248 (30.1%)	473 (33.1%)
Distal metachronous neoplasia, n (%)	160 (26.5%)	183 (22.2%)	343 (24.0%)
Rectal metachronous neoplasia4, n (%)	15 (2.5%)	20 (2.4%)	35 (2.4%)

¹History of colorectal cancer in parent or sibling.

²History of colorectal polyps before qualifying colonoscopy.

³Colorectal adenoma or cancer detected during trial follow-up.

⁴Rectal adenoma only

Table 2.

Association between RXRA and VDR genes and metachronous colorectal neoplasia.

	Any metachronous			Proximal metachronous			Distal metachronous
	neoplasia			neoplasia			neoplasia
	OR	95%CI		OR	95%CI		OR	95%CI
RXRA
PC1	1.01	(0.96–1.05)		1.01	(0.97–1.06)		0.97	(0.93–1.03)
PC2	0.90	(0.83–0.99)		0.88	(0.80–0.97)		1.01	(0.91–1.12)
PC3	1.12	(1.00–1.25)		1.10	(0.98–1.24)		1.03	(0.91–1.17)
PC41	1.09	(0.94–1.28)		1.10	(0.94–1.30)		1.15	(0.96–1.38)
LRT p-value2			0.04			0.03			0.49
VDR
PC1	0.98	(0.93–1.03)		1.00	(0.95–1.06)		1.00	(0.94–1.06)
PC2	0.97	(0.92–1.02)		0.96	(0.91–1.02)		1.00	(0.93–1.06)
PC3	1.03	(0.95–1.10)		0.99	(0.92–1.07)		1.06	(0.98–1.16)
PC4	0.96	(0.88–1.04)		0.94	(0.86–1.03)		0.97	(0.98–1.16)
PC5	1.06	(0.93–1.19)		1.02	(0.89–1.16)		1.02	(0.88–1.18)
PC6	1.13	(0.99–1.30)		1.14	(0.99–1.31)		1.01	(0.86–1.17)
PC7	1.03	(0.89–1.20)		1.00	(0.85–1.16)		0.97	(0.82–1.15)
PC81	1.19	(1.02–1.40)		1.21	(1.02–1.43)		0.98	(0.81–1.17)
LRT p-value2			0.11			0.18			0.96

¹An 80% explained-variance threshold is used for including principal components (PC) in the model.

²P-value for each model is from a likelihood ratio test (LRT) with degrees of freedom equal to the number of principal components.

Next, the relationships between individual SNPs within each gene for any, proximal and distal metachronous neoplasia were evaluated using additive, dominant, and recessive inheritance models for each SNP. After evaluating 32 SNPs in RXRA and 42 in VDR, a total of five SNPs in RXRA and seven in VDR were significantly associated with metachronous neoplasia. Adjustment for multiple comparisons revealed one association of borderline statistical significance for RXRA (rs7861779) and proximal adenoma (OR=0.68; 95% CI=0.53–0.86; unadjusted p=0.001; adjusted p=0.06; Supplemental Table 1). Further, when this SNP was investigated in relation to proximal metachronous adenoma in the WBF and UDCA trials separately with unadjusted logistic regression models, this result was observed in each study separately, with an OR (95% CI) of 0.64 (0.45–0.91) in the WBF trial and 0.70 (0.50–0.98) in the UDCA trial. This SNP was also significantly associated with the presence of proximal lesions at baseline in cross-sectional analyses (p=0.05; data not shown). Considering the potential implications of the differing proportions of participants with a family history of colorectal cancer or who reported previous polyps between the WBF and UDCA trials, the analyses for RXRA rs7861779 were repeated in separate models stratified by these variables. No material differences in the point estimate were observed between those with a family history and those without, nor for those who reported having previous polyps versus those who did not. No significant relationships were observed for any SNPs in VDR and metachronous adenoma after adjustment for multiple comparisons (Supplemental Table 2), including the following high-interest SNPs: TaqI (OR=1.00; 95% CI=0.86–1.17), BsmI (OR=1.01; 95% CI=0.87–1.18), FokI (OR=0.93; 95% CI=0.80–1.09), or the Cdx-2 binding site (OR=0.95; 95% CI=0.79–1.13).

We next constructed linkage blocks within each gene using the available SNP data to derive haplotype groups within each blockable region in order to evaluate the possibility of epistatic effects (See Supplemental Figure 1 for the block structure using D’ for both VDR and RXRA). Three major blocks on RXRA were identified, and VDR was also reduced to three blocks within which the SNPs were highly correlated. We next evaluated haplotype constructed within each of the primary blocks using the most common haplotype in that block as the referent group.

As shown in Table 3, for the RXRA gene, a GAGA haplotype in block 1 (rs11102986, rs11103473, rs10776909, and rs12004589) was associated with reduced odds for proximal metachronous neoplasia (OR, 0.70; 95% CI, 0.54 – 0.92; block global p=0.02). In addition, a haplotype within block 2 of RXRA designated CGGGCA (rs1805352, rs3132297, rs3132296, rs3118529, rs3118536, and rs7861779) was similarly associated with a reduction in odds for any (OR, 0.74; 95% CI, 0.59 – 0.92; block global p=0.03) and proximal (OR, 0.67; 95% CI, 0.52 – 0.86; block global p=0.0097) metachronous neoplasia. A single haplotype in block 1 of VDR, designated GAA (rs11574143, rs731236, rs1544410), was related to reduced odds of proximal adenoma (OR=0.42; 95% CI=0.19–0.96) and increased odds of distal adenoma (OR=2.05; 95% CI=1.08–3.88), though the overall block was not statistically significant for either sub-site (p=0.18 and p=0.15 for proximal and distal adenoma, respectively).

Table 3.

Haplotypes on RXRA and VDR genes and association with metachronous colorectal neoplasia.

Haplotype	Count	Any Metachronous				Proximal Neoplasia				Distal Neoplasia
		N	(%)	OR	95%CI	N	(%)	OR	95%CI	N	(%)	OR	95%CI
RXRA, block 11
GTGC	1857	857	(46.1)	1.00	ref	617	(33.5)	1.00	ref	428	(23.2)	1.00	ref
AAAC	523	252	(48.2)	1.09	(0.89–1.32)	185	(35.6)	1.10	(0.89–1.34)	132	(25.4)	1.12	(0.90–1.41)
GAGA	319	126	(39.5)	0.76	(0.60–0.97)	83	(26.1)	0.70	(0.54–0.92)	82	(25.8)	1.15	(0.87–1.51)
GAAC	86	45	(52.3)	1.28	(0.83–1.97)	34	(40.5)	1.35	(0.86–2.11)	22	(26.2)	1.17	(0.71–1.93)
GAGC	86	38	(44.2)	0.92	(0.60–1.43)	26	(30.2)	0.86	(0.54–1.38)	21	(24.4)	1.07	(0.64–1.77)
rare	7	2	(28.6)	0.47	(0.09–2.41)	1	(14.3)	0.33	(0.04–2.75)	1	(14.3)	0.55	(0.07–4.59)
				p=0.1145				p=0.0284				p=0.8075
RXRA, block 22
AGAACG	1989	927	(46.6)	1.00	ref	674	(34.2)	1.00	ref	467	(23.7)	1.00	ref
CAGGAG	482	230	(47.7)	1.05	(0.86–1.28)	165	(34.5)	1.02	(0.82–1.25)	121	(25.3)	1.09	(0.87–1.38)
CGGGCA	383	150	(39.2)	0.74	(0.59–0.92)	98	(25.7)	0.67	(0.52–0.86)	91	(23.9)	1.01	(0.78–1.31)
rare	24	13	(54.2)	1.35	(0.60–3.04)	9	(37.5)	1.16	(0.50–2.66)	7	(29.2)	1.33	(0.55–3.22)
				p=0.0325				p=0.0097				p=0.8246
RXRA, block 33
AAGGG	1853	861	(46.5)	1.00	ref	622	(33.8)	1.00	ref	429	(23.3)	1.00	ref
GCAAA	501	234	(46.7)	1.01	(0.83–1.23)	169	(34.0)	1.01	(0.82–1.24)	122	(24.5)	1.07	(0.85–1.35)
GAGGA	249	112	(45.0)	0.94	(0.72–1.23)	77	(31.2)	0.89	(0.67–1.18)	66	(26.7)	1.20	(0.89–1.62)
GAGAA	186	73	(39.2)	0.74	(0.55–1.01)	49	(26.5)	0.70	(0.50–0.99)	45	(24.3)	1.06	(0.74–1.50)
GAGGG	35	18	(51.4)	1.22	(0.62–2.38)	15	(42.9)	1.47	(0.75–2.88)	7	(20.0)	0.82	(0.36–1.89)
AAAGG	23	9	(39.1)	0.74	(0.32–1.72)	4	(17.4)	0.41	(0.14–1.22)	8	(34.8)	1.75	(0.74–4.16)
AAGGA	12	4	(33.3)	0.58	(0.17–1.92)	3	(25.0)	0.65	(0.18–2.42)	3	(25.0)	1.09	(0.30–4.06)
rare	19	9	(47.4)	1.04	(0.42–2.56)	7	(36.8)	1.14	(0.45–2.91)	6	(31.6)	1.52	(0.57–4.01)
				p=0.6040				p=0.2091				p=0.8086
VDR, block 14
GAG	1407	655	(46.6)	1.00	ref	457	(32.8)	1.00	ref	332	(23.8)	1.00	ref
GGA	1106	508	(45.9)	0.98	(0.83–1.14)	376	(34.2)	1.07	(0.90–1.26)	258	(23.5)	0.98	(0.82–1.18)
AAG	306	130	(42.5)	0.85	(0.66–1.09)	101	(33.1)	1.02	(0.78–1.32)	73	(23.9)	1.01	(0.75–1.35)
GAA	41	18	(43.9)	0.90	(0.48–1.68)	7	(17.1)	0.42	(0.19–0.96)	16	(39.0)	2.05	(1.08–3.88)
rare	18	9	(50.0)	1.15	(0.45–2.91)	5	(27.8)	0.79	(0.28–2.23)	7	(38.9)	2.04	(0.78–5.29)
				p=0.7591				p=0.1792				p=0.1537
VDR, block 25
GCCGG	1018	482	(47.3)	1.00	ref	337	(33.4)	1.00	ref	253	(25.1)	1.00	ref
GCAAG	956	426	(44.6)	0.89	(0.75–1.07)	300	(31.5)	0.92	(0.76–1.11)	236	(24.8)	0.98	(0.80–1.21)
AACAA	381	169	(44.4)	0.89	(0.70–1.12)	120	(31.7)	0.92	(0.72–1.19)	93	(24.5)	0.97	(0.74–1.28)
GACAA	368	178	(48.4)	1.04	(0.82–1.32)	134	(36.9)	1.17	(0.91–1.50)	76	(20.9)	0.79	(0.59–1.06)
GCCAA	128	57	(44.5)	0.89	(0.62–1.29)	48	(37.5)	1.19	(0.82–1.75)	26	(20.3)	0.76	(0.48–1.20)
GACAG	18	6	(33.3)	0.56	(0.21–1.49)	5	(27.8)	0.77	(0.27–2.17)	1	(5.6)	0.18	(0.02–1.33)
rare	9	2	(22.2)	0.32	(0.07–1.54)	2	(22.2)	0.57	(0.12–2.75)	1	(11.1)	0.37	(0.05–3.00)
				p=0.3939				p=0.4642				p=0.1518
VDR, block 36
AAGAA	1011	477	(47.2)	1.00	ref	334	(33.4)	1.00	ref	250	(25.0)	1.00	ref
GGGGA	977	433	(44.3)	0.89	(0.75–1.06)	307	(31.6)	0.92	(0.76–1.11)	237	(24.4)	0.97	(0.79–1.19)
AAAAG	649	291	(44.8)	0.91	(0.75–1.11)	224	(34.7)	1.06	(0.86–1.31)	148	(22.9)	0.89	(0.71–1.13)
AGGGA	184	93	(50.5)	1.14	(0.84–1.57)	61	(33.7)	1.02	(0.73–1.42)	40	(22.1)	0.85	(0.58–1.25)
rare	57	26	(45.6)	0.94	(0.55–1.60)	20	(35.1)	1.08	(0.62–1.89)	11	(19.3)	0.72	(0.37–1.41)
				p=0.4719				p=0.7460				p=0.7207

¹RXRA, block 1 includes: rs11102986, rs11103473, rs10776909, rs12004589

²RXRA, block 2 includes: rs1805352, rs3132297, rs3132296, rs3118529, rs3118536, rs7861779

³RXRA, block3 includes: rs3118571, rs3118570, rs1536475, rs3132293, rs877954

⁴VDR, block 1 includes: rs11574143, rs731236, rs1544410

⁵VDR, block 2 includes: rs11574026, rs10875695, rs11168293, rs4760655, rs7299460

⁶VDR, block 3 includes: rs4760658, rs4516035, rs11568820, rs7310552, rs7970314

⁷Likelihood ratio test p-value for a model with these haplotypes vs an intercept-only model

When examining associations in the WBF trial and UDCA trial independently, the magnitude of effect observed for each SNP or haplotype was similar for each study, although statistical significance was often demonstrated in only one study or neither study (data not shown). As described above, a notable exception was RXRA SNP rs7861779, which exhibited a statistically significant association with proximal metachronous neoplasia in the WBF (OR=0.64; 95% CI=0.45–0.91) and UDCA (OR=0.70; 95% CI=0.50–0.98) trials individually. In addition, RXRA block 2 haplotype CGGGCA, which includes rs7861779, was significantly associated with any and proximal metachronous neoplasia across the separate and pooled studies.

Because of the biological relationship between VDR and RXRA, interactions between the statistically significant SNP in RXRA, rs7861779, and all of the SNPs in VDR were evaluated. After correction for multiple comparisons, no statistically significant interactions were observed (data not shown).

Discussion

In the current work, a statistically significant association between the RXRA gene and any and proximal metachronous neoplasia was observed, while there was no relationship for VDR at the gene level and neoplasia. After adjustment for multiple comparisons, one SNP in RXRA (rs7861779) was of borderline statistical significance in relation to proximal metachronous adenoma. Haplotype analyses demonstrated that the RXRA block 1 haplotype GAGA, defined by rs11102986, rs11103473, rs10776909, rs12004589, was statistically significantly associated with a 24% reduced odds for proximal neoplasia. In addition, RXRA block 2 haplotype CGGGCA, defined by rs1805352, rs3132297, rs3132296, rs3118529, rs3118536, rs7861779, was significantly related to a similar 26% lower odds of both any and proximal metachronous neoplasia. A single VDR haplotype in block 1 of VDR was related to reduced odds for proximal adenoma and increased odds for distal adenoma; however, the overall block was not significantly associated with lesions at either sub-site.

Using a tagging approach, we found no consistent relationship between polymorphic variation in VDR and odds of metachronous colorectal adenoma, although there is substantial evidence for a role of VDR regulatory events in development of colorectal neoplasia (39). For example, VDR has been reported to interact with β-catenin and inhibit β-catenin signaling events that are commonly disrupted in colonic cancers (13). In the nucleus, β-catenin binds to TCF/LEF enabling the transactivation of target genes as part of the Wnt signaling pathway. Significantly, the activation of this Wnt signaling network, as well as β-catenin-mediated upregulation of proliferative genes, is suppressed in response to 1,25(OH)₂D₃ via the nuclear VDR (13). Furthermore, while it is well-established that VDR heterodimerizes with RXR to exert transcriptional effects and promote cell differentiation (40), β-catenin also acts as a transcriptional co-activator of the RXR-VDR heterodimer resulting in increased transactivation of VDRE-driven genes in the presence of 1,25(OH)₂D₃, and causing enhanced antiproliferation (13, 41). Moreover, RXRα , an isoform expressed in the colonic epithelium, acts via agonists that enhance the interaction between RXR and β-catenin as well as increase the degradation of β-catenin (42, 43). RXR ligands have also been demonstrated to modify the antiproliferative response of 1,25(OH)₂D₃ in two colon cancer cell lines, exhibiting an increased proliferative response in Caco-2 cells, but blocking it in HT-29 cells (44). The reasons for these differential responses are unclear, but the data indicate that RXR acts as a potent modulator of VDR activity. Furthermore, an RXR ligand (9-cis-RA), can act synergistically with VDR-1,25D to induce VDR-mediated transactivation, enable recruitment of coactivators and upregulate E-cadherin expression (45). These actions of the VDR/RXR heterodimer may offer insight into the biological mechanisms by which retinoids and vitamin D metabolites influence the development of colorectal neoplasia, and how genetic variation in these genes may alter this activity (39). We therefore hypothesized that genetic variation in RXR likely affects the biological activity of VDR.

VDR is a highly polymorphic gene with over 100 identified SNPs, of which only a handful have been studied to determine an association between genetic variation and risk of colorectal adenoma and colorectal cancer. Previous association studies have included BsmI (rs1544410), TaqI (rs731236), ApaI (rs7975232), and FokI (rs2228570) restriction endonuclease sites, as well as the Cdx-2 binding site in the promoter region (rs11568820). None of these SNPs have been consistently independently associated with the risk of adenoma recurrence (23) or colorectal cancer (46); in a recent meta-analysis of FokI and BsmI and colorectal cancer, no significant associations were reported (46), and another review of the literature indicated that the data for VDR SNPs and colorectal cancers may be weaker than for other malignancies (47). In the current work, no relationship between Cdx-2, BsmI, or TaqI and metachronous colorectal adenoma were observed, although a single haplotype in VDR block 1 including TaqI was observed to be statistically significantly associated with distal and proximal lesions, though in opposite directions. Although ApaI was not directly assessed in this population, this SNP has been shown to be in high linkage disequilibrium with TaqI and Bsm I (47), neither of which was significantly related to metachronous neoplasia. Further, these results do not support a relationship between FokI and the development of metachronous colorectal neoplasia, which is in line with recent reviews and meta-analyses of FokI which do not support a strong independent association with colorectal neoplasia (46, 47). Nonetheless, as this SNP may be important in gene by environment interactions, future work will include analyses of interactions between FokI, circulating vitamin D concentrations, and other exposures.

While the association between VDR polymorphic variation and cancer has been studied extensively, only two other studies have investigated the association of polymorphic variation in RXRA and cancer, though neither included colorectal cancer as an endpoint (28, 29). In the work by Ahn et al., no relationship was observed for SNPs in RXRA and risk for prostate cancer (28); however, the study included only 11 SNPs in RXRA. Further, it is likely that the expression and activity of nuclear receptors like RXR vary in a tissue-specific manner (48). Chang et al. reported no significant associations between two RXRA SNPs (rs1536475 and rs1805343) and biliary tract cancers (29). The results of the present study also found no relationship between these two SNPs and risk of metachronous colorectal adenoma (Supplemental Table 1), but did show that RXRA was significantly associated with colorectal neoplasia at the gene level. Further, after corrections for multiple comparisons, one high-interest SNP (rs7861779) was identified, and after characterizing haplotype block structure to assess the association between co-carriage of alleles and metachronous adenoma, RXRA block 1 was observed to be significantly associated with proximal metachronous adenoma. RXRA block 1 contains SNPs located in intron 1; while RXRA block 2, which was found to be significantly associated with any and proximal adenoma, contains SNPs located in introns 2, 4 and 5. SNPs located within introns can potentially affect alternative splicing of RNA (49–51).

Two RXRA haplotype blocks in our population demonstrated significant associations with metachronous adenoma of similar magnitude. The two SNPs of highest interest (rs12004589 and rs7861779) from blocks 1 and 2, respectively, demonstrate strong linkage disequilibrium (D'=0.87) with one another, and as such are not likely to confer independent information. Replication of this work in another population is necessary to determine if the haplotypes and SNPs of interest in our population are generalizable to other populations, and whether they are also related to cancer risk. When examined in the WBF and UDCA trials individually, the RXRA SNP rs7861779 was statistically significantly related to proximal metachronous neoplasia in both studies. Interestingly, rs7861779 was also statistically significantly associated with the presence of proximal adenomas at baseline in cross-sectional analyses (p<0.05; data not shown). Of note, most of the SNPs selected for this investigation are tag SNPs; therefore, the SNPs in these haplotypes are likely not causal, but rather serve as indicators for regions of interest. Therefore, if these data replicate in another population, then further identification of potential causal variants and their functional implications is warranted.

Taken together, laboratory and epidemiological data support a protective role for vitamin D and its receptor, VDR, as well as the VDR heterodimeric partner, RXRα, in colorectal carcinogenesis. However, genetic variation in VDR, as measured here by a tagSNP approach, did not appear to play a major independent role in the formation of metachronous colorectal adenoma. In contrast, RXRA was significantly associated with colorectal neoplasia at the gene level, and these findings suggest a potentially important role of RXRα in the molecular processes leading to colorectal carcinogenesis, particularly of the proximal colon. Because no statistically significant interactions between RXRA SNP rs7861779 and any VDR SNP were observed in the current work, it is possible that the associations observed are mediated via other RXRα heterdimer partners; these pathways should be investigated in future work.