Chemoprevention of Colorectal Neoplasia by Estrogen: Potential Role of Vitamin D Activity

Postmenopausal hormone replacement therapy lowers colon cancer incidence. In humans, the mechanism is unknown, but animal models suggest that it may involve activation of the vitamin D receptor (VDR) pathway.

The aims of our study were to determine whether estrogen intervention affects global gene expression in rectal mucosal biopsies and whether vitamin D–related genes are affected.

Estradiol was given to raise serum estradiol to premenopausal levels in 10 postmenopausal women under close nutritional control. Primary end points were expression of VDR, CYP24A1, CYP27B1, and E-cadherin in rectal mucosa by reverse transcription-PCR and examining response to estradiol by genome-wide arrays. Responses in gene expression in rectal biopsies to estrogen were determined in each subject individually and compared with a human estrogen response gene array database and a custom array in vitro–generated database.

Cluster analysis showed that subjects maintained their overall gene expression profile and that interindividual differences were greater than intraindividual differences after intervention. Eight of 10 subjects showed significant enrichment in estrogen-responsive genes. Gene array group analysis showed activation of the VDR pathway and down-regulation of inflammatory and immune signaling pathways. Reverse transcription-PCR analysis showed significant up-regulation of VDR and E-cadherin, a downstream target of vitamin D action.

These data suggest that the chemopreventive action of hormone replacement therapy on colon neoplasia results, at least in part, from changes in vitamin D activity. Evaluation of gene arrays is useful in chemopreventive intervention studies in small groups of subjects.

Selection of chemopreventive agents to lower the risk of colorectal cancer in humans requires extensive clinical studies. It has become increasingly accepted that chemopreventive regimens, just like chemotherapeutic regimens for cancer treatment, should include more than a single agent (1). Selection of multiple different compounds is complex, but they probably should alter different pathways of carcinogenesis simultaneously. Because the mechanism of action of many chemopreventive agents is unknown, it is important to simplify preliminary clinical study design to determine the composition, tissue, and cellular targets of such colorectal chemoprevention regimens. An approach that we have developed is to measure rectal gene expression in small numbers of volunteers before and after administration of a putative chemopreventive regimen to gain insight into pathways that are altered by such intervention.

Postmenopausal hormone replacement therapy is associated with decreased incidence and death rate of colon cancer in epidemiologic studies (2–6) and intervention trials (3, 7). A recent population-based case-control study found that it was also associated with strong protection against rectal cancer (8). In the colon, estrogen effects are mediated predominantly by estrogen receptor (ER) β (9) and loss of expression of this receptor is associated with advanced Duke's staging in colon cancer (10). However, ERα also may play a role in the anticancer effects of estrogen signaling. Recent observations suggest that loss of signaling via ERα may be a contributor to colorectal cancer development in the APC/Min mouse model of colorectal carcinogenesis (11). Furthermore, evidence suggests that the chemopreventive effects of estrogen may, at least in part, be mediated through vitamin D receptor (VDR) signaling. VDR is a known estrogen-responsive gene (12) and colonocytes express VDR at relatively high levels; however, the expression of VDR decreases in the adenoma-carcinoma sequence (13). Low vitamin D intake is associated with colon cancer in humans (14) and rodents (15) and vitamin D supplementation decreases tumor load in Apc^Min−/+ mice (16). In dimethylhydrazine-induced murine colon carcinogenesis, estrogen prevented tumor formation and increased expression of colonic VDR (17). Estradiol replacement also prevented intestinal tumorigenesis and ameliorated enterocyte migration and intercellular adhesion in the Apc^Min−/+ colorectal carcinoma mouse model (18). Furthermore, estrogen increases expression of VDR in rat intestine (19, 20) and estrogen administration restores the responsiveness of the intestine to 1,25-dihydroxyvitamin D [1,25(OH₂)D] in ovariectomized premenopausal (21) and postmenopausal women (22). Other tissues and cell systems, such as uterus (23), liver (24), and breast cancer cells (25), also respond to estradiol by increased VDR expression.

In the present study, we investigated whether administration of estradiol to postmenopausal volunteers with low estradiol levels results in changes in the VDR pathway in rectal biopsy specimens. The primary end points were changes in vitamin D–related gene expression in the rectal mucosa after confirming that the estrogen intervention had an effect by assessing estrogen-responsive genes in rectal biopsies.

Enrolled in this unblinded study were 10 postmenopausal women (mean age, 58.3 ± 4.6 y) at risk for colorectal neoplasia with a history of colorectal adenoma resection or a first-degree relative with colorectal cancer or adenoma and with circulating serum estradiol concentrations in the postmenopausal range (<45 pg/mL). Six of the volunteers were white (non-Hispanic), three were black (non-Hispanic), and one other ethnicity (Table 1A). All subjects were healthy and received a comprehensive history and physical examination before enrollment. Subjects were excluded if they had a history of cancer other than nonmelanoma skin cancer, subtle mammary tumors revealed by breast examination, previous major intestinal surgery, malabsorption, or those on estrogen/progesterone replacement or if they had supplemental vitamin D intake.

Volunteers maintained their prestudy diet for at least 2 wk before the experimental period as well as during the 4-wk study. Their diet supplied sufficient calories and nutrients to maintain their energy balance and contained calcium/vitamin D in amounts that approximate current intakes (Table 1B). Dietary intake was monitored by 3-d food records twice before study start, and 24-h recalls were obtained during weekly telephone calls by the trained research nutritionists of The Rockefeller University Hospital. During these telephone interviews, the diet of each volunteer was reviewed and they were instructed how to maintain their prestudy diet and their weight ± 1.5%.

Complete blood counts, comprehensive AMA chemistry panel, prothrombin time and partial thromboplastin, serum 25-hydroxyvitamin D [25(OH)D] and 1,25(OH₂)D, estradiol and protein-bound sex hormone levels, 24-h urine creatinine, and calcium excretion were measured. From flexible proctosigmoidoscopy after a 60 mL tap water enema, ∼12 rectal biopsies were taken from 4 quadrants 10 to 15 cm from the anal verge. Ten biopsies were immediately placed in Nunc cryotubes and frozen in liquid nitrogen for extraction of RNA, and 2 biopsies were fixed in 10% buffered formalin and processed for histology. Histologic examination of the biopsies showed that ∼90% of cells were from the epithelium layer.

Studies were approved by the Institutional Review Board of The Rockefeller University Hospital and informed consent was obtained from all subjects. At study start, subjects were provided with daily doses of estradiol (0.5-1 mg; Barr Laboratories) to raise their serum estradiol levels to premenopausal range (>50 pg/mL). Estradiol levels were also measured at study end. Compliance was maximized by selection of motivated volunteers, estradiol ingestion was evaluated by pill counts at study end, and diets were monitored weekly by nutritionists. In addition, biopsies were taken from two individuals not enrolled in this trial, which were used as in vivo controls. In these subjects, four separate biopsy samples were obtained from different quadrants of the rectum from the same sigmoidoscopy procedure and total RNA was extracted for microarray analyses. Gene expression analysis was done on RNA samples from each of these four biopsy samples separately to serve as an in vivo control for differences between biopsy samples.

Frozen biopsies were maintained in liquid nitrogen until used for RNA extraction for gene arrays and reverse transcription-PCR (RT-PCR). Total RNA was extracted from rectal biopsies using the Trizol method (Invitrogen). Trizol-extracted RNA was further purified using Qiagen RNeasy kits (Qiagen, Inc.), yielding high-quality RNA suitable for RT-PCR and microarray analyses. RNA quality was verified by analysis on Agilent 2100 Bioanalyzer (Agilent Technologies), and RNA was quantified by NanoDrop (NanoDrop Technologies). All the samples were run concurrently. Total RNA (500 ng) was used for in vitro transcription and cRNA amplification and labeling using Ambion Illumina kit according to the manufacturers' instructions. Biotin-labeled cRNA was labeled with fluorescent dye in The Rockefeller University Gene Array Facility and hybridized onto Sentrix HumanRef-8 24K Expression Array Bead Chip (Illumina). Arrays then were scanned by the Illumina Bead Station laser scanning imaging system.

To test whether estrogen-responsive genes were significantly enriched after estradiol intervention, we evaluated a publicly available Estrogen Responsive Genes Database (ERGDB; version 2.0 data for humans; ref. 12), from which 970 unique estrogen-responsive genes were imported into GeneSpring software for analyses. These data are well established but contain entries from many human tissues. In addition, a colon-specific translational experimental in vitro system was developed (in vitro control). Briefly, HCT116 colon cancer cells, purchased from the American Type Culture Collection, were maintained in DMEM (Invitrogen) with 100 units/mL penicillin and 100 μg/mL streptomycin in a normal atmosphere with 5% CO₂ at 37°C. Cells were plated in six-well plates at 125,000 per well, allowed to attach for 24 h, and treated with 1,000 pg/mL β-estradiol for 24 h, and total RNA was collected for microarray analyses.

Expression data were imported and analyzed by GeneSpring software after normalization steps. The hierarchical clustering algorithm used to generate the dendrogram is based on the complete linkage (26). Distance between two individual samples was calculated by Pearson distance with normalized expression values, and differences in gene expression in individual subjects were determined by filtering fold change for each subject separately. The expression cutoff difference was 1.3-fold. (The array suggested minimum cutoff is 1.2.)

Genes differentially expressed following estrogen administration were tested for similarity with genes from the ERGDB and genes modulated by estradiol in the HCT116 colon cancer cells in vitro system. Gene lists also were subjected to gene ontology (GO) analysis using the GO function resident in the GeneSpring software package version 7.3. Statistical significance of overlap between gene groups was calculated conservatively using standard Fisher's exact test and P values were adjusted with Bonferroni multiple testing correction. P values of ≤0.05 were considered significant. Additionally, an Ingenuity Pathway Analysis was done for group analysis of estradiol intervention and to test whether the VDR pathway is significantly enriched.⁷

For RT-PCR, 2 to 5 μg of total RNA, in triplicate, were used as template for cDNA synthesis using SuperScript III First-Strand kit (Invitrogen). cDNA was diluted in water and amounts corresponding to 10 ng of original RNA were used for quantitative gene expression by RT-PCR. RT-PCR used Taqman Gene Expression Assay probes and primers (Applied Biosystems) and the ABI Prism 7900 RT-PCR system at The Rockefeller University Gene Array Core Facility. The following genes were assayed: VDR, E-cadherin, CYP27B1, CYP24A1, ERα, and ERβ. 18S rRNA endogenous controls were adjusted for sample loading and variation in reverse transcription efficiency. Three independent RT-PCRs, with respective endogenous controls, were used to calculate results.

Statistical analysis of gene arrays is described above. RT-PCR samples were assessed in triplicates and mRNA levels were expressed as log ratio of relative expression levels before or after treatment. The null hypothesis, that the log ratio is equal to zero, was tested by two-tailed paired t test. Benjamini and Hochberg multiple testing correction was applied adjusting P value for multiple comparisons.

Paired t tests were used to compare serum concentrations of estradiol, 25(OH)D, and 1,25(OH₂)D before and after estradiol administration. Pearson correlation coefficients measuring the strength of linear relationships between two variables were calculated for changes in mRNA after estradiol administration. Significance was P < 0.05.

All subjects except one completed the 4-week experimental study period. This subject (#2) dropped out for personal reasons and was replaced. Two subjects described mild breast tenderness and one noticed vaginal blood spotting during estrogen administration. Mean subject weight was 81.1 kg and body mass index was 29.5 (Table 1A). Mean serum estradiol concentration at study start was 17 ± 10 pg/mL (mean ± SD) and at study end was 46.6 ± 19 pg/mL (P < 0.001), indicating that the subjects consumed estrogen as prescribed by the protocol. Baseline serum 25(OH)D and 1,25(OH₂)D concentrations were in the reference range and did not change significantly during the study (Table 1C). Subjects consumed ∼1,600 kcal/d with 63 ± 15 g of fat, similar to values reported in the 1999 to 2000 National Health and Nutrition Examination Survey diet for 40- to 60-year-old women (Table 1B) and food intake and weight did not change significantly during the study.

A primary aim of this study was to determine whether estrogen intervention could cause detectable changes in gene expression profiles in human rectal mucosal biopsy specimens. A hierarchical clustering analysis was done on the microarray results using distance similarity measures and complete linkage (Fig. 1). These 20 samples were divided into 10 main groups each corresponding to a single subject with the preintervention and postintervention samples always clustering nearest to each other. These results indicate that an individual's overall gene profile (for the 24,000 interrogated genes) was preserved during the 4-week study. The average Pearson's distance, a measure of the degree of similarities or differences between samples, for the preintervention and postintervention biopsy specimens was 0.45 ± 0.06, and the average distance between the four samples from the in vivo controls was 0.26 ± 0.06, which is significantly less than for the estrogen trial subjects (P < 0.01). As expected, the six samples from the in vitro control cells produced the smallest average Pearson's distance of 0.02 ± 0.002 and served as a measure of the least achievable distance from this microarray system.

Next, we analyzed estrogen-responsive genes that were modulated by estradiol intervention. Nine hundred seventy unique, estrogen-responsive genes from all human tissues from a publicly available ERGDB⁸

were present on the Illumina array, and 1,072 genes changed in HCT116 colon cancer cells after exposure to estradiol. There was a significant overlap between these two estrogen-responsive gene lists (P < 0.001). The number of genes that changed their expression significantly after 4-week estradiol intervention varied from 587 to 1,937 among individual subjects. However, statistically significant enrichment for estrogen-responsive genes in ERGDB occurred in 8 of 10 subjects by the highly conservative Fisher's exact test with Bonferroni correction for multiple observations (Table 2). Overall, 6.8% to 10.4% of genes that changed their expression were also present in the estrogen-responsive databases. Therefore, these data clearly confirm the primary aim of our study showing that the rectal mucosa was responsive to the estrogen intervention.

GO analysis⁹

showed that significantly enriched GO categories also significantly overlapped between two estrogen-responsive databases (P < 0.001); a complete summary of significantly enriched GO categories is shown in Supplementary Table. Group analysis of gene expression following estradiol intervention using Ingenuity Pathway Analysis showed significant down-regulation of genes mediating inflammatory signaling and immune responses (Table 3). The antigen presentation pathway showed down-regulation in multiple components of MHC I and II and the complement system pathway showed down-regulation in C1q, C2, and C2a. Interleukin (IL)-4 signaling pathway was also significantly down-regulated and there was a down-regulation in chemokine signaling pathway and natural killer cell signaling. There were significant changes in genes involved in VDR activation, suggesting that estradiol may act, at least in part, through activation of the VDR pathway (Fig. 2). Multiple downstream target RNAs of VDR, such as CYP24A1, TGFβ2, CDKN1A, cyclin C, OPN, RANKL, LRP5, IGFBP-6, IGFBP-5, CST6, SEMA3B, SERPINB1, MADD, GADD45A, PPARD, SULT2A1, HSD17B2, and PDGFα, were found to be up-regulated in the whole-genome array experiment (Fig. 2A). Additionally, there was a down-regulation in genes such as CYP27B1, RANTES, GM-CSF, IL-2, IL-12, and IFNγ. These genes are downstream of VDR-mediated transrepression and would be expected to decrease their expression on activation of VDR pathway (Fig. 2B).

To test the second aim of our study, expression of vitamin D–related genes VDR, CYP24A1 (24,25α-hydroxylase), and CYP27B1 (1α-hydroxylase), ERα, ERβ, and E-cadherin (a downstream target of vitamin D action) was assessed by RT-PCR in 10 paired preintervention/postintervention biopsies. Estrogen treatment increased significantly VDR (P = 0.006) and E-cadherin (P = 0.02; Fig. 3) expression without affecting CYP27B1, ERα, or ERβ mRNA levels. The mucosal levels of CYP24A1 were, at best, close to limit of detection of our assay and several samples were below this detection limit. Given these caveats, no significant differences in CYP24A1 mRNA were seen in samples by RT-PCR but Ingenuity Pathway Analysis revealed increased CYP24A1 expression levels via group analysis. This result is expected if VDR is activated because CYP24A1 is one of its downstream targets (Fig. 2).

The present studies show a method of analyzing gene changes induced in the rectal mucosa by a preventive agent intervention in a small group of subjects under controlled nutritional conditions. We studied the effects of estradiol, administered to postmenopausal women to raise circulating levels to premenopausal levels, on gene expression in rectal biopsies. The tissue response for the estradiol intervention in these postmenopausal women was determined by comparing the changes in gene expression measured in the rectal epithelium with a published database on general effects of estrogen in vivo and with changes found in colon cancer cells exposed to estradiol in vitro. We have shown previously the utility of comparing changes in gene expression of rectal biopsies induced by in vivo sulindac treatment with those that occurred during incubation of colon cancer cells in vitro (27). The present data show that genes that were modulated by estrogen intervention in rectal biopsies showed significant overlap with ERGDBs showing rectal mucosal response to the estradiol intervention.

Our hypothesis was that the vitamin D system in the colon was the functional target for estradiol. It has been established that postmenopausal hormone replacement therapy is associated with decreased incidence and death rate of colon cancer in both epidemiologic (2–6) and intervention (3, 7) trials. Experimental evidence from animal studies has pointed to several potential mechanisms of action of estradiol, including changes in ER signaling (10, 11, 18) or in bile salt composition (28, 29). However, additional data suggest that this action may, at least in part, be mediated through VDR signaling. It has been shown previously (12) that VDR is an estrogen target gene and that expression of VDR is decreasing in high-grade (poorly differentiated or undifferentiated) colon tumors (13). Low vitamin D intake is associated with colon cancer both in humans (14) and in rodents (15). We found that estrogen treatment increased highly significantly expression of VDR as well as that of E-cadherin, a downstream target of vitamin D (Fig. 3). There was a trend for increase in ERα mRNA but no change was observed for ERβ (Fig. 3). Others have observed that administration of estrogen or estrogenic compound to APC/min mice resulted in increased expression of ERβ (18, 30). We have no explanation for this discrepancy except for interspecies differences in the response to estradiol. The cellular action of vitamin D is mediated through the production of calcitriol, 1,25(OH₂)D, and its interaction with VDR present in many cells of the body. Because it is not possible to quantify 1,25(OH₂)D in biopsies, we analyzed surrogate markers of activation of the VDR pathway focusing on downstream targets. Our RT-PCR data confirmed that estrogen administration increased mRNA expression of VDR as well as of a downstream target of vitamin D action, E-cadherin. In addition, the whole-genome array analysis showed that several other downstream targets of VDR were also elevated and target genes of VDR-mediated transrepression were down-regulated (Fig. 2). These observations support our hypothesis that estrogen may regulate vitamin D activity through autocrine/paracrine modulation of the colorectal VDR pathway. Unfortunately, biopsy tissues were insufficient to measure the actual VDR protein expression. However, because there are over 30 corepressors and coactivators along the VDR signaling pathway, protein expression would not necessarily reflect functional status (31). Furthermore, regulation of gene expression by 1,25(OH₂)D may be highly tissue specific (32). Up-regulation of the VDR might have led to increased uptake of vitamin D by colonocytes but we have no direct evidence of this.

The hypothesis that estrogen acts on the colon through vitamin D action is supported also by observations in other systems. Estradiol increases the production of 1,25(OH₂)D in Caco2 colon cancer cells, CYP27B1 mRNA levels (33), and the expression of rat intestinal VDR (19, 20). It has been hypothesized that colonic 1,25(OH₂)D derives from conversion of circulating 25(OH)D by CYP27B1 present in the colonocytes. The balance between the activities of CYP27B1 and the degradative enzyme CYP24A1 determines epithelial cell content of 1,25(OH₂)D (33, 34). Our present RT-PCR data showed no significant effects of estradiol on CYP27B1 and CYP24A1 expression, although the latter was up-regulated in gene array experiments.

Estrogen also reduced components of inflammatory and immune response pathways such as those linked to antigen presentation, IL-4 signaling, and chemokine signaling. The biological significance of these observations in the colorectum is presently unknown. Estrogen attenuates chemokine signaling in human monocytes (35) and inflammatory responses following injury (36, 37). Postmenopausal hormone replacement therapy is protective of disease activity in women with inflammatory bowel disease (38). Because of the link between inflammation and colon carcinogenesis, this role of estrogen warrants further studies.

In summary, data from this small but carefully conducted study suggest that estradiol administration results in significant changes of estrogen-responsive genes and seems to increase VDR signaling in the colorectal epithelium of postmenopausal women via up-regulation of VDR, a known estrogen-responsive gene. Furthermore, estrogen decreased multiple components of inflammatory signaling and immune response pathways in the rectal mucosa. Our data suggest that evaluation of gene arrays may be helpful in chemopreventive intervention studies in small groups of subjects if a set of specific hypotheses is tested. We conclude that chemopreventive activity of hormone replacement therapy on colon neoplasia results, at least in part, from changes in vitamin D activity.

No potential conflicts of interest were disclosed.