Hypermethylation-associated Inactivation of the Cellular Retinol-Binding-Protein 1 Gene in Human Cancer¹

Retinoids, analogues of vitamin A, are known to control cellular signals involved in cell growth, differentiation, and carcinogenesis (1, 2). Retinoids suppress preneoplastic lesions and prevent the development of second primary cancers (3, 4). 9-cis-Retinoic acid inhibits mammary tumors induced by N-nitroso-N-methylurea (5). However, some tumors show resistance to retinoid action by mechanisms still largely unknown. Retinoid activity is mediated by nuclear hormone receptors. The RARs³ α, β, and γ and RXRs α, β, and γ act as ligand-activated transcription factors. Disruption of RARs and RXRs results in developmental defects and neoplastic transformation (1, 2). Aberrant retinoid signaling in cancer has been found in the leukemogenic role of the dominant-negative PML-RARα fusion protein in acute promyelocytic leukemia (6) and the demonstration of down-regulation of RARβ2 expression in solid human carcinomas (7, 8, 9) caused by CpG island promoter hypermethylation and loss of heterozygosity (10, 11, 12, 13).

Other key components in retinoid activity are the CRBPs. The CRBPs belong to the family of fatty acid-binding proteins, and three members have been described: CRBP2 and CRBP3, which show tissue-specific expression (14, 15); and CRBP1, which is expressed widely (16). Retinoic acid (the main active metabolite of retinol) is present in the circulation, but most tissues rely on the uptake and cytosolic metabolism of retinol to activate RARs and RXRs. CRBPs possess a high-affinity binding for retinol, possibly functioning as a chaperone-like molecule to regulate this prenuclear phase of retinol signaling (14). Furthermore, the critical role of CRBP1 is demonstrated in the CRBP1 knock-out mice, which are very susceptible to hypovitaminosis A syndrome (17). Recently, CRBP1 down-regulation in breast cancer cell lines (8) and tumors (18) has been observed. The mechanism behind this inactivation has not been established. We investigated the possible role of CpG island aberrant methylation in CRBP1 silencing and its relationship to RARβ2 methylation and dietary retinol intake.

The cancer cell lines used in this study were MCF-7, MDA-MB-231, MDA-MB-468, T47D, Hs578, MDA-435 (breast); SW48, DLD1, RKO, SW837, HCT-116, SW480, Colo-320, HT29 (colon); H358, H1618, H249, H209, H157, H460, H69, DMS53, A549, H1299, H1752 (lung); DuPro, LnCAP, PC3, Dn145 (prostate); UMSCC1, HN12 (head and neck); D283, DaOY (glioma); Raji (lymphoma), KG1a (leukemia); HeLa (cervix); and T24 (bladder). Cell lines were maintained in appropriate medium and treated with demethylating agent 5-aza-2′-deoxycytidine as described (19). Primary malignancies were obtained at the Johns Hopkins Hospital in Baltimore, MD and the Hospitals Sant Pau and Duran i Reynals in Barcelona, Spain and have been examined for other epigenetic changes (19, 20, 21). For 131 colorectal cancer patients, dietary history was obtained according to the following cohort design. These were among the patients consecutively diagnosed and operated at the Hospital Duran i Reynals during 1996 that provided informed consent, biological specimens to extract DNA, and completed the questionnaires. They were interviewed after diagnosis, usually within 15 days before or after surgical removal of the tumor. The interview was carried out by trained personnel using a structured questionnaire to avoid potential recall biases on epidemiological factors and a dietary history based on the average food consumption 1 year previous to the disease (22). Nutrient intakes (retinol and carotene) were estimated using food composition tables. To adjust for total energy intake, nutrient densities were calculated dividing nutrient intake by total kilocalories estimated from macronutrients plus alcohol (23).

DNA methylation patterns in the CpG island of CRBP1 were determined by methylation-specific PCR (24). Methylation-specific PCR distinguishes unmethylated from methylated alleles in a given gene on the basis of sequence changes produced after bisulfite treatment of DNA, which converts unmethylated, but not methylated, cytosines to uracil and subsequent PCR by use of primers designed for either methylated or unmethylated DNA (24). Primer sequences of CRBP1 for the unmethylated reaction were 5′-GTGTTGGGAATTTAGTTGTTGTTGTTTT-3′ (sense) and 5′-ACTACCAAAACAACAACTACCAATACTACA-3′ (antisense); and for the methylated reaction, 5′-TTGGGAATTTAGTTGTCGTCGTTTC-3′ (sense) and 5′-AAACAACGACTACCGATACTACGCG-3′ (antisense). Primers for RARβ2 were: unmethylated reaction, 5′-TTGGGATGTTGAGAATGTGAGTGATTT-3′ (sense) and 5′-CTTACTCAACCAATCCAACCAAAACAA-3′ (antisense); and for the methylated reaction, 5′-TGTCGAGAACGCGAGCGATTC-3′ (sense) and 5′-CGACCA ATCCAACCGAAACGA-3′ (antisense).

RT-PCR was performed as described previously (18, 19). The PCR primers used were 5′-TTGTGGCCAAACTGGCTCCA-3′ (sense) for exon 1 and 5′-ACACTGGAGCTTGTCTCCGT-3′ (antisense) for exon 3 of CRBP1, which amplify a 320-bp product. Glyceraldehyde-3-phosphate dehydrogenase served as a positive control (19).

All comparisons for statistical significance were performed by use of χ² or Fisher’s exact test, as appropriate, with all Ps representing two-tailed tests and statistically significant at 0.05. For the cases where dietary information was collected, binary logistic regression models adjusted for age and sex were developed. Retinol and carotene intakes were categorized into quartiles to avoid the effect of extreme values. OR and 95% confidence intervals were calculated for quartiles 2 to 4 compared with the first one. Tests for linear trend on the ORs were calculated using the categorized variable as quantitative.

CRBP1 is a candidate for hypermethylation-associated inactivation because a 5′-CpG island is located around the transcription start site. All normal tissues analyzed, including lymphocytes, breast, colon, bone marrow, endometrium, kidney, liver, and lung, were completely unmethylated at the CRBP1 promoter (Fig. 1,A). The CRBP1 CpG island was fully methylated in 17 of 37 cancer cell lines studied: MCF-7, MDA-MB-231, T47D, Hs578, MDA-MB-435, SW48, DLD1, RKO, SW837, DuPro, LnCAP, PC3, Dn145, H358, D283, RAJI, and KG1a (examples in Fig. 1,B). Two lung cancer cell lines were partially methylated (H1299 and H1752). The remaining 17 cell lines were unmethylated at the CRBP1 promoter. Expression of the CRBP1 transcript was assessed by RT-PCR in several cell lines. The cell lines MDA-MB-231, MCF-7, SW837, SW48, and Dn145, hypermethylated at the CRBP1 promoter, did not express CRBP1, whereas MDA-MB-468, normal lymphocytes, and normal colon, unmethylated at the CRBP1 promoter, had strong CRBP1 expression. The treatment of the methylated cancer cell lines with the demethylating agent 5-aza-2′-deoxycytidine restored the expression of the CRBP1 transcript (Fig. 1 C).

To assess the prevalence of promoter hypermethylation of CRBP1 in vivo, we examined 553 human primary tumors. The highest incidence of CRBP1 promoter hypermethylation was found in non-Hodgkin’s lymphomas, where 39 of 65 (60%) had CRBP1 epigenetic inactivation (examples in Fig. 2). Among different histological subtypes of lymphoma, CRBP1 aberrant methylation was present in 24 of 39 (62%) diffuse, 5 of 9 (56%) follicular, 2 of 2 (100%) Burkitt’s, and 8 of 15 (53%) lymphomas of other histologies.

Aberrant CpG island methylation of the CRBP1 was also common in colorectal (57%; 101 of 177), gastric (42%; 8 of 19), and liver (30%; 10 of 33) tumors, as well as leukemia samples (28%; 15 of 53). Additionally, CRBP1 hypermethylation was also observed in breast (19%; 9 of 48), bladder (18%; 5 of 27), non-small cell lung (15%; 6 of 41), glioma (14%; 3 of 21), and head and neck (7%; 1 of 15) tumors. No CRBP1 promoter methylation was observed in endometrial (0 of 33) or ovarian (0 of 21) tumors. The data are summarized in Table 1.

To further address the relevance of CRBP1 hypermethylation, we examined CRBP1 methylation in preinvasive lesions of the gastrointestinal tract. CRBP1 promoter hypermethylation was observed in 21 of 34 (62%) colorectal adenomas (Fig. 3,A) and 13 of 21 (62%) gastric adenomas. This rate is similar to the percentage found in the invasive colorectal and gastric tumors and suggests that methylation-associated inactivation of CRBP1 is an early event in human tumorigenesis. To confirm that the methylation of this 5′ region of the CRBP1 gene was functionally relevant, we examined the expression of CRBP1 using RT-PCR in 8 colorectal adenomas. Five colorectal adenomas without CRBP1 methylation expressed high levels of CRBP1 mRNA, whereas all 3 adenomas with CRBP1 aberrant methylation had undetectable CRBP1 transcript (Fig. 3 B).

Because alteration of the RARβ2 gene is common in human malignancies (10, 11, 12, 13), we wondered whether any relationship between methylation of RARβ2 and CRBP1 existed. Similar to CRBP1, we found that hypermethylation of the RARβ2 promoter was also a common event in colon cancer (54%; 93 of 172) and non-Hodgkin’s lymphoma (26%; 11 of 42). The most common group observed was those tumors where both genes were concomitantly methylated, 39% (84 of 214), but closely followed for the category where both were unmethylated, 30% (65 of 214). In 20% (43 of 214) of cases, CRBP1 was hypermethylated alone, and only in a minority of cases, 10% (22/214), RARβ2 was hypermethylated alone. No significant differences between colon tumors and lymphomas were observed. Thus, both genes were simultaneously hypermethylated more often than expected by chance (Kappa statistic P < 0.0001), although RARβ2 was methylated more often than CRBP1 (Exact McNemar test, P = 0.013).

For 131 colorectal cancer patients, a dietary history was available as described in “Materials and Methods.” Median daily retinol intake in these patients was 155 μg (interquartile range, 102 to 239). These values were used as cutpoints to define quartiles. We found that cases in the highest quartile of retinol intake were more likely to have CRBP1 and RARβ2 unmethylated. The ORs were 1.9 (P for linear trend = 0.069) for CRBP1 and 2.5 (P for linear trend = 0.039) for RARβ2. Other nutrients analyzed and found unrelated to the methylation status of the genes were vitamin C, E, lycopene, lutein, α-carotene, and β-carotene. The methylation status of CRBP1 was also unrelated to age, sex, or stage at diagnosis in this sample of subjects.

Our data demonstrate that epigenetic disruption of CRPBI is a common event in human cancer. The first described alteration in the retinoid pathway was the leukemogenic role of the PML-RARα fusion protein (6). Later, evidence supported the role of RARβ2 as a tumor suppressor gene: the induction of RARβ2 related to the chemopreventive effects of retinoids (25), the loss of RARβ2 expression in human neoplasms (7, 9), frequent chromosomal losses at 3p21–3p24 where RARb2 is located (26, 27), and the methylation-mediated silencing of RARβ2 (10, 13). We now provide another piece of this puzzle; the epigenetic silencing of CRBP1 is also a common alteration in human cancer.

What are the biological consequences of the methylation-mediated silencing of CRBP1? The loss of CRBP1 may compromise retinoic acid metabolism by diminishing retinol transport and blocking the formation of retinyl esters (14, 28). Interestingly, it has also been demonstrated that the lack of retinoic acid function may increase the activity of the β-catenin-LEF/T-cell factor signaling pathway (29), a central element in malignant cell transformation. Our data also show the relatively common simultaneous inactivation of CRBP1 and RARβ2. One explanation is that CRBP1 may have a function independent of its retinol-binding ability. In this regard, CRBP1 also functions in mammary epithelial cell inhibition of the phosphatidylinositol 3-kinase/Akt survival pathway and suppresses anchorage-independent growth (30).

In addition, aberrant methylation of CRBP1 may have predictive value. Until now, the use of retinoids to prevent or treat human cancer has achieved only modest success (31). The disruption of CRBP1 and RARβ2 by promoter hypermethylation in many of the human neoplasms may in part explain this resistance. However, in colorectal tumors, ∼30% of malignancies did not present any apparent lesion in the retinoid pathway. This subset of tumors may be extremely sensitive to treatment with retinoids.