The tumor suppressor adenomatous polyposis coli and caudal related homeodomain protein regulate expression of retinol dehydrogenase L.

Development of normal colon epithelial cells proceeds through a systematic differentiation of cells that emerge from stem cells within the base of colon crypts. Genetic mutations in the adenomatous polyposis coli (APC) gene are thought to cause colon adenoma and carcinoma formation by enhancing colonocyte proliferation and impairing differentiation. We currently have a limited understanding of the cellular mechanisms that promote colonocyte differentiation. Herein, we present evidence supporting a lack of retinoic acid biosynthesis as a mechanism contributing to the development of colon adenomas and carcinomas. Microarray and reverse transcriptase-PCR analyses revealed reduced expression of two retinoid biosynthesis genes: retinol dehydrogenase 5 (RDH5) and retinol dehydrogenase L (RDHL) in colon adenomas and carcinomas as compared with normal colon. Consistent with the adenoma and carcinomas samples, seven colon carcinoma cell lines also lacked expression of RDH5 and RDHL. Assessment of RDH enzymatic activity within these seven cell lines showed poor conversion of retinol into retinoic acid when compared with normal cells such as normal human mammary epithelial cells. Reintroduction of wild type APC into an APC-deficient colon carcinoma cell line (HT29) resulted in increased expression of RDHL without affecting RDH5. APC-mediated induction of RDHL was paralleled by increased production of retinoic acid. Investigations into the mechanism responsible for APC induction of RDHL indicated that beta-catenin fails to repress RDHL. The colon-specific transcription factor CDX2, however, activated an RDHL promoter construct and induced endogenous RDHL. Finally, the induction of RDHL by APC appears dependent on the presence of CDX2. We propose a novel role for APC and CDX2 in controlling retinoic acid biosynthesis and in promoting a retinoid-induced program of colonocyte differentiation.

Colon cancer arises from distinct genetic events that initiate and promote tumor formation. An inherited colon cancer predisposition, familial adenomatous polyposis, results from mutations in a single gene known as adenomatous polyposis coli (APC). 1 This syndrome is characterized by the appearance of hundreds to thousands of colon adenomas in affected individuals. Recent investigations have generated a model describing downstream events controlled by APC. In the current model, APC regulates the activity of a transcriptional pathway that may control colonocyte proliferation (for a review, see Refs. 1 and 2). It does so by regulating the levels of ␤-catenin, a protein found initially to function as a link between extracellular adhesion molecules and the cytoskeleton. It appears, however, that ␤-catenin also regulates transcription through a partnership with TCF/LEF transcription factors. In cells expressing functional APC, APC acts to repress ␤-catenin levels through regulation of ubiquitin-mediated proteolysis. Low levels of ␤-catenin prevent activation of TCF/LEF. In cells harboring mutated APC, ␤-catenin accumulates. This accumulation allows assembly of ␤-catenin⅐TCF/LEF complexes and activation of the transcriptional capabilities of TCF/LEF (1,2). ␤-catenin⅐TCF/LEF-dependent transcriptional activation of specific cell cycle regulatory genes, like c-myc (3) and cyclin D1 (4), may underlie the development of colon adenomas and colon carcinomas (1)(2)(3)(5)(6)(7)(8). In addition to sustained cell proliferation, colonocytes within adenomas and carcinomas display differentiation defects (9 -13). For example, crypts from colon adenomas are deficient in mucin-producing goblet cells (14), one of the three predominant, terminally differentiated cell types seen within normal colon crypts. Although APC/␤-catenin pathway target genes such as c-myc and cyclin D1 offer mechanistic insights into disregulation of colonocyte proliferation, few of the current APC pathway target genes have easily identifiable roles in cellular differentiation.
In addition to the nuclear hormone receptors, retinoid responsiveness within cells is governed by retinoid availability (35,36). For the most part, cells acquire retinoids in the form of retinol, an inactive precursor. Tissues must, therefore, convert retinol into RA in order to activate the network of nuclear receptors required to evoke retinoid transcriptional responses. The enzymes that catalyze these conversions fall into three distinct classes that include the alcohol dehydrogenases, the short-chain dehydrogenases/reductases, and the aldehyde dehydrogenases. Alcohol dehydrogenases and short-chain dehydrogenase/reductase enzymes convert retinol into the aldehyde, retinal. Further conversion of retinal into RA is carried out by the aldehyde dehydrogenase enzyme family. Enzymes in each class have broad substrate specificities and can oxidize or reduce many physiologically important alcohols or aldehydes including ethanol, steroids, and retinoids. The actions of RA, in turn, can be limited by catabolism via cytochrome P450 enzymes (35,36). Although the biochemistry of these retinoid biosynthetic and metabolic enzymes is emerging, little is known about the regulation of these enzymes within tissues or specific cell types.
In order to define the molecular pathways that may govern APC-dependent differentiation of colonocytes, we have analyzed gene expression profiles in colon adenomas and carcinomas compared with normal colon. Our analyses revealed that colon adenomas and carcinomas show consistent down-regulation of the RA biosynthetic enzymes retinol dehydrogenase 5 (RDH5) and retinol dehydrogenase-like (RDHL). Given that loss of RA biosynthetic genes may contribute to the lack of differentiation observed in colon adenomas and carcinomas, we investigated the regulatory mechanisms that control the expression of RDH5 and RDHL. We found in a survey of normal human tissues that, whereas both RDHL and RDH5 were expressed in the colon, RDHL expression appears relatively restricted to the colon. Reintroduction of APC into APC-deficient colon cancer cells induced RDHL expression. Furthermore, the intestinal specific transcription factor, CDX2, targets RDHL and acts synergistically with APC in the induction of RDHL. Since RA holds known differentiation properties, our data create a model wherein APC controls RA biosynthesis as a potential mechanism for regulating colonocyte differentiation.

EXPERIMENTAL PROCEDURES
Microarray Analysis-Microarray analyses were performed as described previously (37). Briefly, total RNA was extracted from surgically excised and microdissected colon tissue samples using Trizol reagent (Invitrogen). Poly(A) RNA was selected using an Oligotex Kit (Qiagen). First-strand cDNA probes were generated by reverse transcription of 1 g of purified mRNA with SuperScript II (Invitrogen) after the addition of Cy3-dCTP or Cy5-dCTP (Amersham Biosciences). Probes were purified and reconstituted in 30 l of 5ϫ SSC, 0.1% SDS, 0.1 g/ml salmon sperm DNA, and 50% formamide. After denaturation at 94°C, the hybridization mix was deposited onto the slide under a coverslip.
Hybridizations were performed overnight at 42°C in a humidified chamber on glass slides displaying 4608 human genes in duplicate. In every case, RNA from adenoma or carcinoma tissue was compared with a pool of normal tissue RNA from six separate donors. Since RNA availability of most samples was limiting, most samples were only surveyed on our first slide (containing 4608 genes), which contained the RDH5 gene but not the RDHL gene. Since some of the samples (numbers 3, 6, 18, 22, and 25) contained plentiful RNA, we were able to survey their gene expression on our entire gene set (totaling 38,000 genes and including the RDHL gene). Following hybridization, slides were washed for 10 min in 1ϫ SSC, 0.2% SDS and then for 20 min in 0.1ϫ SSC, 0.2% SDS. Slides were dipped in distilled water and dried with compressed air, and the fluorescent hybridization signatures were captured using the "Avalanche" dual laser confocal scanner (Amersham Biosciences). Fluorescent intensities were quantified using ArrayVision 4.0 (Imaging Research). Differentially expressed genes were selected when the Log 2 (Cy5 signal/Cy3 signal) exceeded 1.85-fold. This cut-off was selected based on the control comparisons examining normal colon versus normal colon samples. This value exceeds the variability seen in these control comparisons. Differential expression was determined using Student's t test for each gene in each comparison. The p value must not have exceeded 0.05.
PCR was performed in duplicate (or triplicate for 18 S rRNA) with a master mix consisting of cDNA template, buffer (500 mM Tris, pH 8.3, 2.5 mg/ml bovine serum albumin, 30 mM MgCl 2 ), dNTPs (2 mM), TaqStart antibody (Clontech), Biolase DNA polymerase (Bioline), gene-specific forward and reverse primers (10 M), and SYBR Green I (Molecular Probes, Inc., Eugene, OR). The PCR conditions are as follows: 35 cycles of amplification with 1-s denaturation at 95°C and 5-s annealing at 54°C for RDHL, 58°C for RDH5, 60°C for cyclin D, and 53°C for 18 S rRNA. A template-free negative control was included in each experiment.
The copy number was measured by comparing gene amplification with the amplification of standard samples that contained 10 3 to 10 7 copies of the gene or 10 5 to 10 9 for 18 S rRNA. The relative expression level of each gene was calculated by averaging the replicates and then dividing the average copy number of gene X by the average copy number of 18 S rRNA. The S.E. value of the ratios was calculated using a confidence interval.
Cell Culture and Drug Treatments-HT29, HCT116, and RKO colon adenocarcinoma cells were cultured as recommended by the American Type Culture Collection. HT29 APC-inducible and LacZ-inducible cells were kindly provided by Dr. Bert Vogelstein (Johns Hopkins Oncology Center, Baltimore, MD).
Transfections and Luciferase Assays-Fugene 6 (Roche Applied Science) and LipofectAMINE Plus (Invitrogen) were used to transfect HCT116 cells and RKO cells, respectively, according to the manufacturers' protocols. Cells were seeded at a density of 100,000 cells/well in 12-well plates and transfected the next day. Transfections were performed using 0.6 g of DNA (including 0.06 g of normalization vector, 0.12 g of reporter vector, and 0.42 g of expression vector), and, in the absence of further treatment, cells were harvested 24 h after the start of transfection. In certain experiments, medium containing charcoalstripped serum as well as either RA mixture (see above) or ethanol vehicle was added to cells 24 h after transfection, and cells were harvested 24 h after medium change. Luciferase values were analyzed using a Dual Luciferase Assay System (Promega). Transfection efficiencies were normalized by dividing the firefly luciferase activity by the Renilla luciferase activity for each sample. Data in each experiment are presented as the mean Ϯ S.D. of duplicates from a representative experiment. All experiments were performed at least three times.
Electrophoretic Mobility Shift Assays-22-Mer oligonucleotides representing Ϫ338 to Ϫ359 in the RDHL promoter containing putative CDX2 sites were annealed and 5Ј-end-labeled with [␣-32 P]ATP using T4 kinase (MBI Fermentas). The labeled oligonucleotides were then purified over a micro Bio-spin 6 column (Bio-Rad) to remove unincorporated nucleotides. For binding reactions, 6 g of extract from 293 cells overexpressing CDX2 were incubated with 1 l (50,000 -200,000 cpm) of 32 P-labeled probe in a 10-l final volume of 1ϫ binding buffer (20% glycerol, 5 mM MgCl 2 , 2.5 mM EDTA, 2.5 mM dithiothreitol, 250 mM NaCl, 50 mM Tris-HCl (pH 7.5), and 0.25 mg/ml poly(dI-dC)-poly(dI-dC)). For competition and supershift assays, a 10-or 50-fold molar excess of unlabeled oligonucleotide or 2 g of anti-His tag antibody or anti-CDX2 antibody were added 15 min prior to the addition of labeled oligonucleotide. Complexes were separated on a 6% acrylamide gel in 1ϫ TBE at 250 V for 3 h. The gel was then dried and exposed to a phosphor screen and visualized on a PhosphorImager (Amersham Biosciences).
Northern Blotting-Total RNA was isolated using Trizol (Invitrogen) followed by poly(A) RNA selection using a Poly(A) Tract mRNA Isolation kit (Promega). Poly(A) RNA was fractionated through formaldehyde-containing agarose gels and transferred onto nylon membranes (Amersham Biosciences). Probes were generated using the Rediprime II random prime labeling system (Amersham Biosciences) supplemented with [␣-32 P]dCTP. Hybridizations with 32 P-labeled probes were carried out using ULTRAhyb buffer (Ambion) as recommended by the manufacturer.
RA Extraction and HPLC Analysis-Cells were treated with 100 nmol ATROL at 80 -90% confluence for 12 h. Medium was removed, and cells were scraped into PBS for protein quantification. After the addition of 100 nmol of internal standard TTNPB, the medium was acidified with 6 N HCl (0.03ϫ volume) and extracted with an equal volume of hexane containing 0.1 mg/ml butylated hydroxytoluene. The resulting solution was mixed vigorously and spun down at 11,500 rpm for 20 min. Following centrifugation, the organic phase was transferred to a glass vial, dried under nitrogen, and reconstituted in 100 l of 1:1 Me 2 SO/ MeOH for HPLC analysis. Extracted RA is expressed as pmol of RA/mg of protein/mmol of TTNPB.
Retinoid analysis and quantification was performed using a reversed phase Phenomenex Luna C18, 4.6 ϫ 250 mm, 5 particle size analytical column. Retinoids were eluted with a gradient starting at 80% acetonitrile, 20% ammonium acetate, pH 5.0, to 100% acetonitrile in 40 min, with a flow rate of 1.5 ml/min at 350 nm. Identification of retinoid peaks from the extracts was done by comparing elution positions with matching retinoid standards. Quantification of extracted retinoids was performed by relating the area of the peak to areas obtained by the analysis of known quantities of retinoid standards.

Retinol Dehydrogenases Are Down-regulated in Neoplastic
Colon-In order to identify signaling pathway alterations in neoplastic colon, we performed microarray expression analyses on colon adenomas and carcinomas in comparison with a pool of normal colon tissue. A striking feature of our colon tumor progression data was that ϳ80% of the differentially expressed genes were down-regulated in adenoma and carcinoma tissues as compared with normal (data not shown). Two genes in particular showing down-regulation in colon adenomas and carcinomas caught our attention. The first gene encoded RDH5, an enzyme that catalyzes the conversion of retinol into retinal. The second gene encoded RDHL, a recently described, novel retinol dehydrogenase (described by Soref et al. (38), but referred to as hRDH-TBE). Each of these genes was downregulated at least 2-fold in ϳ70% of both adenoma and carcinoma tissues relative to normal (Fig. 1A).
Due to small tissue sizes and correspondingly low yields of mRNA samples, we obtained limited data on RDHL by microarray (see "Experimental Procedures"). We thus used quantitative RT-PCR to assess the expression levels of RDH5 and RDHL in an additional 10 patient-matched normal and carcinoma colon tissues. Fig. 1B shows that expression of RDHL was decreased at least 2-fold in 9 of 10 carcinoma samples in comparison with the matched, normal appearing colon tissue, whereas RDH5 was decreased in 6 of 10 samples examined. As a positive marker for distinguishing colon carcinoma from normal, cyclin D1 expression levels were increased in each of the carcinoma samples relative to normal (data not shown) (39,40). We also noted the loss of tags corresponding to RDHL in two Only samples 3, 6, 18, 22, and 25 were in sufficient quantity to analyze for both RDH5 and RDHL expression. B, quantitative RT-PCR was performed on total RNA from matched patient normal colon and adenoma (sample 267) or carcinoma (all samples except 267) for the indicated genes. Each sample was first normalized to 18 S rRNA. The data were then plotted as -fold decrease in gene expression for each neoplastic colon sample relative to its matched normal colon sample. Values shown represent the mean Ϯ S.D. for three independent determinations. colon carcinoma sample data sets deposited into the SAGE (serial analysis of gene expression) data base, an observation recently reported by Buckhaults et al. (41).
We examined the tissue distribution of RDH5 and RDHL by performing Northern analyses on mRNAs from multiple human tissues. Hybridization with full-length RDHL identified a 1.9-kb mRNA species that was primarily expressed in the colon ( Fig. 2A). Although there are three potential splice variants of the RDHL gene (as deposited in GenBank TM , accession clones AF067174, AF240698, and AF240697), we confirmed by RT-PCR that normal colon expresses the isoform corresponding to clone AF067174 (data not shown), the same splice variant characterized as a retinol dehydrogenase (38). Expression of RDHL was low but also detectable in heart, spleen, placenta, and lung ( Fig. 2A). In contrast, a probe specific for RDH5 hybridized to a 1.4-kb mRNA species with the highest levels appearing in liver and kidney ( Fig. 2A). RDH5 was also present in heart, skeletal muscle, colon, and small intestine. Analysis of the blot with a probe specific for ␤-actin indicated similar mRNA loading in each lane (Fig. 2A).
Colon Tumor Cells Lack RDHL and RDH5 and Fail to Convert Retinol to Retinoic Acid-Given the absences of RDH5 and RDHL in colon carcinomas, we examined their expression levels in a panel of commonly studied colon carcinoma cell lines. Among the cell lines examined were cells known to harbor mutations in APC or ␤-catenin (Caco-2, Colo 205, DLD-1, HT29, SW480, and HCT116). Only RKO cells contain both wild type APC and ␤-catenin. In agreement with the data from human tumors, each of seven colon cancer cell lines expressed low or undetectable levels of RDHL and RDH5 in comparison with normal colon (Fig. 2B). To assess the capacity of these cell lines to produce retinoic acid, we quantified retinoic acid production following incubation of each cell type with 100 nmol of all-trans-retinol for 24 h. Retinoic acid was extracted from tissue culture medium using acidified hexane and retinoic acid levels determined by HPLC in comparison with the internal extraction standard TTPNB. Due to the lack of a method for culturing primary colon epithelial cells, we chose to compare the results from the colon cancer cell lines to normal human mammary epithelial cells, which express RDH5 at levels similar to that of normal colon (data not shown). Fig. 2C shows that each of the colon cancer cell lines converted retinol into retinoic acid poorly.
APC Regulates RDHL Expression-Since RDH5 and RDHL expression was reduced in most colon adenomas and carcinomas as well as colon carcinoma cell lines with APC pathway defects, we began considering mechanisms that would account for their lack of expression. We gave particular attention to the possibility that the regulation of RDH5 and RDHL was connected to the APC pathway. To address this, we utilized the HT29 cell line containing a ZnCl 2 -inducible APC gene constructed and described by Morin et al. (42). In the absence of ZnCl 2 , the APC-inducible cells only express mutant forms of APC. Upon the addition of ZnCl 2 , wild-type APC expression is induced. An HT29 cell line containing a ZnCl 2 -inducible LacZ gene served as a negative control. We treated each cell line with 100 M ZnCl 2 for 24 h and analyzed endogenous gene expression by Northern analysis. Fig. 3A demonstrates that induction of APC led to a 3.2-fold induction of RDHL without any detectable increase in RDH5 expression. LacZ-inducible cells showed no induction of RDHL or RDH5. The induction of RDHL following restoration of wild-type APC was also time-dependent. Quantitative RT-PCR analysis detected induction of RDHL as early as 8 h following the addition of 100 M ZnCl 2 (Fig. 3B). Induction of RDHL persisted for 12 and 24 h but declined by 48 h when cell viability became limiting. The temporal induction of RDHL was consistent with the temporal induction of APC.
Given that APC induced the expression of RDHL (Fig. 3, A  and B), we expected to see an increase in retinol dehydrogenase activity following induction of APC. To accomplish this, cells were again induced to express APC and RDHL. Cells were then provided 50 nM retinol for 4 h. Following incubation, RA was extracted and quantified as above. Consistent with the induction of RDHL, we observed a statistically significant increase in the ability of HT29 cells to convert retinol into RA after induction of APC. We saw no increase in RA production in the LacZ-inducible cells.
APC Regulation of RDHL Is Independent of ␤-Catenin-APC induction of RDHL (Fig. 3, A and B) focused our attention on understanding the transcriptional mechanisms that control RDHL expression downstream of APC. The most obvious possibility was that elevated levels of ␤-catenin served to repress RDHL expression. We thus examined the first 1000 base pairs of the RDHL promoter as well as the RDH5 promoter for putative TCF/LEF consensus sequences ((A/T)(A/T)CAAAG).

FIG. 2. RDHL and RDH5 expression is low in colon cancer cell lines and is accompanied by poor conversion of retinol into RA.
A, a human multiple tissue Northern blot containing a minimum of 1 g of polyadenylated RNA was probed with the full-length coding sequences for RDHL, RDH5, and ␤-actin (for loading control). Exposure times for the three hybridizations ranged from 1 to 3 days in order to emphasize the relative tissue distributions of RDH5 and RDHL. B, poly(A) RNA was harvested from the indicated colon cancer cell lines and analyzed by Northern blot with probes for RDHL, RDH5, and glyceraldehyde-3-phosphate dehydrogenase (GAPDH) (for loading control). C, indicated cell lines were incubated for 8 h with 50 mM retinol. RA levels in cell culture medium were determined following extraction with hexane-HCl and separation by reversed phase HPLC. Retinoic acid values are shown as the mean Ϯ S.D. for three replicate experiments. Fig. 4A shows a schematic illustrating that the RDHL promoter contains one putative TCF/LEF binding site, whereas the RDH5 promoter contains two. To test whether ␤-catenin could regulate the RDHL promoter through TCF/LEF, we cloned the RDHL and RDH5 promoters and fused each to a luciferase reporter gene. We then examined the ability of a constitutively active form of ␤-catenin (␤-catenin S37A) to repress luciferase expression in 293 cells, a cell line that is often used to identify ␤-catenin target genes. As a positive control, we used TOP-FLASH, which contains multimerized TCF/LEF sites fused to a luciferase reporter gene. Fig. 4B demonstrates that whereas ␤-catenin S37A is functional, its expression did not induce the RDHL promoter or repress the promoter basal activity.
Since it was possible that endogenous levels of ␤-catenin served to maximally repress RDHL:LUC, thus masking any effect of ␤-catenin S37A, we decided to inhibit endogenous ␤-catenin using a dominant-negative LEF1 construct (DN-LEF). This construct contains the DNA binding domain of LEF1 but lacks the ␤-catenin interaction domain, thereby preventing ␤-catenin from regulating transcription from TCF/LEF sites. We expected that if ␤-catenin normally represses the RDHL promoter, then inhibiting endogenous ␤-catenin with DN-LEF should lead to induction of RDHL:LUC. RDH5:LUC and FOP-FLASH (the equivalent reporter vector to TOP-FLASH, but with mutated TCF/LEF sites) served as negative controls. Fig. 4C shows that whereas DN-LEF can inhibit the activity of ␤-catenin, RDHL:LUC is not induced. Based on this result, we discarded the model wherein ␤-catenin directly regulates RDHL.
RDHL Is Regulated by CDX Transcription Factors-Since ␤-catenin and DN-LEF did not appear to regulate RDHL expression, we considered the possibility that RDHL is controlled by APC independently of ␤-catenin. To identify new candidates for transcriptional regulation of RDHL, we examined the RDHL promoter for additional, canonical transcription factor binding sites. In this analysis, we found that the first 1000 base pairs of the RDHL promoter contained seven TTTAT motifs ( Fig. 4A) that have been shown to bind to the caudal related homeodomain proteins, CDX1 and CDX2 (43). By comparison, the same region of the RDH5 promoter contained only one of these motifs (Fig. 4A). CDX1 and CDX2 have important roles in regulating gastrointestinal development in vertebrates (44 -47), and they, like RA, have antiproliferative, prodifferentiative effects in colon cells (48 -51). Their expression is highly specific to the intestines, and they are down-regulated in certain human colon tumors (52)(53)(54). Finally, CDX2 is mutated in RKO cells, the only cell line used in Fig. 2, B and C, that harbors wild type APC and ␤-catenin. We, therefore, hypothesized that the CDX transcription factors could control RDHL expression. To test this hypothesis, we asked whether CDX1 and CDX2 could activate RDHL:LUC. Fig. 5A shows the induction of RDHL:LUC activity in HCT116 colon cancer cells cotransfected with CDX1 or CDX2. Both CDX1 and CDX2 induced RDHL:LUC but failed to induce RDH5:LUC.
To confirm the binding of CDX2 to putative binding elements within the RDHL promoter, we designed an oligonucleotide corresponding to nucleotides Ϫ359 to Ϫ338 within the RDHL promoter and used this in an electrophoretic mobility gel shift assay. Nuclear protein extracts from 293 cells transfected with His-tagged CDX2 showed two complexes that were effectively competed by the addition of a cold, unlabeled competitor oligonucleotide. Antibodies to either the His tag or CDX2 protein supershifted the complexes, thus confirming binding by CDX2. Finally, we asked whether the expression of CDX2 in a CDX2deficient cell line caused induction of endogenous RDHL. Fig.  5C shows that expression CDX2 in RKO cells causes the induction of the endogenous levels of RDHL as measured by Northern analysis.
RDHL:LUC spans Ϫ2228 to ϩ1071 of the RDHL promoter and contains 18 putative CDX2 response elements. Note that RDHL:LUC contains 10 putative TCF/LEF sites (Fig. 6A) but does not appear to be regulated by ␤-catenin (Fig. 4, B and C).
In an attempt to pinpoint the region of the RDHL promoter that is responsive to CDX2, we made a series of promoter deletions (Fig. 6A), fused each upstream of the luciferase gene, and assessed the ability of CDX2 to induce each reporter construct. Fig. 6B shows a gradual decrease in CDX2 induction as putative CDX2 sites are eliminated from the promoter. Outside of the RDHL TATA-box, no single mutation eliminated CDX2 induction of RDHL (data not shown).
APC and CDX2 Regulate RDHL Synergistically-The above data suggested the possibility that APC and CDX2 regulate RDHL. Two pieces of evidence suggested that APC and CDX2 may work in a dependent manner to regulate RDHL. First, RDHL is down-regulated in tumors and cell lines harboring wild type APC. For example, sequence analysis of the APC mutation cluster region in the nine carcinomas showing reduced RDHL expression in Fig. 1B revealed only six with mutations (data not shown). This suggests intermediates between APC and RDHL. Consistent with this, RKO cells contain wild type APC and ␤-catenin but lack CDX2. To examine the relationship between APC and CDX2 in the regulation of RDHL, we examined whether APC can induce RDHL:LUC in a colon carcinoma cell line, SW480, that contains very low levels of CDX2 (data not shown). Fig. 7 shows that APC alone was incapable of inducing RDHL:LUC in SW480 cells. In contrast, CDX2 alone activated RDHL:LUC ϳ2-fold. Induction of RDHL: LUC was dramatically enhanced by co-transfection of APC with wild-type CDX2, suggesting interaction of the two in regulating RDHL. RDH5:LUC was not induced by any of the treatments. DISCUSSION The existence of biosynthetic and metabolic pathways for retinoids implies that the control of cellular responses to retinoic acid must, at one level, reside in the control of RA biosynthesis and metabolism. In the present study, we provide data that support the loss of RA biosynthetic enzymes as a downstream consequence of APC mutation, but not necessarily a consequence of ␤-catenin disregulation. Our results are consistent with a model wherein APC and the intestine-specific tran-  Fig. 4. B, nuclear protein extracts from 293 cells transfected with His-tagged CDX2 were mixed with a 32 P-labeled oligonucleotide representing putative CDX binding sites (Ϫ359 to Ϫ338) within the RDHL promoter (lane 2). Two complexes were shifted by CDX2 and competed away by unlabeled oligonucleotide (lanes 5 and 6). These complexes were also supershifted by antibodies to the His tag and to CDX2 (lanes 3 and 4). n.s., nonspecific. C, RKO cells were transfected with the expression vectors pCDNA3.1 or pCDNA3.1-CDX2. After 24 h, cells were harvested, and poly(A) RNA was isolated and analyzed by Northern blot with probes for RDHL, CDX2, and ␤-actin (for loading control). scription factor, CDX2, control the RA biosynthetic capacity in colon cells. Evidence for this model includes the following. (i) Expression levels of RDH5 and RDHL were depressed relative to normal in over 70% of the neoplastic tissues that we examined; (ii) wild type APC induced only RDHL in an APC-deficient colon carcinoma cell line; (iii) APC induction of RDHL paralleled increased RA production in HT29 cells; (iv) the intestine-specific transcription factor CDX2 activated an RDHL promoter construct as well as endogenous RDHL expression; and (v) APC induction of RDHL was enhanced by the presence of wild-type CDX2.
Studies in model organisms highlight the importance of RA biosynthesis and metabolism in development and differentiation. For example, deletion of the retinoid-metabolizing P450 enzyme CYP26 in mice disrupted anteroposterior axis development, normal hind brain patterning, vertebral identity, and development of posterior structures (55)(56)(57). Similarly, mice with a targeted disruption of the retinaldehyde dehydrogenase gene Raldh2 die in midgestation, display shortening along the anteroposterior axis, and fail to form limb buds (58). It is clear from these studies that the biosynthesis and metabolism of RA plays an important role in development and differentiation. Loss of regulatory control of RA biosynthetic genes could, therefore, alter cell growth and differentiation in the colon, thus contributing to the development of colon adenomas and carcinomas. Support for the loss of retinoid biosynthesis in contributing to tumor formation comes from recent studies showing that certain breast cancer cell lines failed to synthesize RA (59). Moreover, reintroduction of ALDH6, a retinaldehyde dehydrogenase, restored the ability of MCF-7 breast cancer cells to synthesize RA (60). Finally, retSDR1 and LRAT, two genes involved in retinol storage, were found to be lost in neuroblastoma (61) and prostate cancer (62), respectively.
Few studies have revealed the regulatory mechanisms that control expression of RA biosynthetic genes in any tissues. The loss of RDH5 and RDHL expression in colon cancer emphasizes the importance of understanding these regulatory mechanisms. We found that reintroduction of the tumor suppressor APC induced RDHL in colon cancer cell lines. Our findings imply that the loss of APC in most colon cancers could account for the similar loss of RDHL that we observed by microarray and RT-PCR (Fig. 1, A and B). Although mutations in APC appear to predict loss of RDHL, loss of RDHL does not appear to predict the presence of an APC mutation. Indeed, a number of tumors and at least two cell lines failed to express RDHL despite harboring wild type APC. This suggests a defective, intermediate step between APC and RDHL.
In explaining this, we focused our attention on the possibility that RDHL was repressed by ␤-catenin. This, however, appears not to be the case for two reasons. First, overexpression of ␤-catenin and a dominant-negative LEF1 construct failed to regulate an RDHL promoter construct either positively or negatively (Fig. 4, B and C). Second, RKO cells lack RDHL expression but express wild type APC and ␤-catenin. This lack of regulation by ␤-catenin implies a function for APC that is independent of its known role in regulating levels of ␤-catenin. A number of studies suggest a ␤-catenin-independent function of APC in cellular differentiation. Mariadason et al. (63) showed that neither inhibition of ␤-catenin transcriptional activity nor translocation of ␤-catenin to the membrane was sufficient to induce two of four differentiation markers examined in a Caco-2 cell model of intestinal differentiation. In addition, Dang et al. (64) have demonstrated ␤-catenin-independent regulation of gut Kruppel-like factor 4 by APC and CDX2. It is therefore likely that ␤-catenin independent pathways can promote differentiation in colonocytes.
In considering ␤-catenin-independent regulation of RDHL, our attention was drawn to the CDX transcription factors. CDX1 and CDX2 were likely candidates for regulation of RDHL for a number of reasons. First, like RDHL, they are highly expressed in the colon (52,53) and lost in certain colon adenocarcinomas (52)(53)(54). Second, they have been shown to cause growth arrest and to promote differentiation upon re-introduction into colon cancer cell lines (48,51). Finally, CDX2 is mutated in RKO cells. Consistent with these findings, the RDHL promoter contains several consensus CDX binding elements (Fig. 6A) that are recognized by CDX2 in vitro (Fig. 5B) and that appear required for CDX2 activation of an RDHL promoter construct (Fig. 6B). In addition, CDX2 induced endogenous RDHL in RKO cells (Fig. 5C). Altogether, these data support a model whereby CDX2 activates endogenous RDHL expression in colonocytes.
We found that APC may require CDX2 to induce RDHL (Fig.  7). It is presently unclear how APC and CDX2 may interact in regulating RDHL. We were unable to detect changes in expression of CDX2 following APC induction in HT29 cells (data not shown), despite previous reports that CDX2 is induced by APC in HT29 cells (65). The idea that APC may not regulate transcription of the CDX2 gene is suggested by the finding that familial adenomatous polyposis tissues express normal levels of CDX2 protein (54). However, a role for APC in post-translational regulation of CDX2 cannot be dismissed, since there does appear to be a relationship between APC and CDX2. First, familial adenomatous polyposis tissues lack expression of the reported CDX2 target gene, Kruppel-like factor 4 (64). Furthermore, Aoki et al. (66) have recently reported that Apcϩ/Ϫ Cdx2ϩ/Ϫ mice showed increased colonic polyp numbers in comparison with Apcϩ/Ϫ Cdx2ϩ/ϩ mice. This suggests a requirement for inactivation of both APC and CDX2 in colon polyp formation and raises the possibility that CDX2 protein is not functional in familial adenomatous polyposis colon polyps despite its normal levels of expression (54).
Our finding that APC and CDX2 control RDHL fits with recent studies suggesting that RA can inhibit the oncogenic effects of ␤-catenin. Easwaran et al. (33) demonstrated that RA-bound RA receptor ␣ can bind to and sequester ␤-catenin, leading to down-regulation of ␤-catenin transcriptional activity. The same group later showed that the differentiationassociated morphological changes induced by RA in a breast cancer cell line are probably mediated through cadherin-de-pendent recruitment of ␤-catenin to the cellular membrane (34). Interestingly, the ability of RA to induce ␤-catenin translocation was independent of its ability to inhibit ␤-cateninmediated transcription (34), suggesting that RA has separable roles in inhibiting proliferation and promoting differentiation. Still, the finding that a number of genes can be synergistically activated by Wnt-1 and RA (67) warrants additional investigations to clarify the relationship between Wnt signaling and retinoids in colon tumor development.
Two separate studies have characterized the enzymatic activity of RDHL. Soref et al. (38) demonstrated that RDHL (referred to as hRDH-TBE in their publication) increased the ability of tracheobronchial epithelial cells to convert retinol to RA, whereas Chetyrkin et al. (68) determined that RDHL (referred to as 3␣-HSD in their publication) prefers different substrates in vitro. Specifically, Chetyrkin et al. (68) found that RDHL was 100 times more efficient as a 3␣-hydroxysteroid dehydrogenase than as a retinol dehydrogenase. Their in vitro studies demonstrated that RDHL can catalyze the conversion of 3␣-tetrahydroprogesterone (allopregnanolone) to dihydroprogesterone and 3␣-androstanediol to the potent androgen, dihydrotestosterone. Presently, the prevalent enzymatic activity of RDHL in colonocytes remains uncertain. We found that induction of APC was paralleled by an increase in RDHL but not RDH5. This was also accompanied by an increased ability of the cells to convert retinol into RA (Fig. 3C) and supports findings that RDHL may convert retinol into retinaldehyde in cells.
Although the roles for retinoids in intestinal epithelial cell function remain unclear, studies in vitamin A-deficient rats predict an important function for retinoids in intestinal mucosa. Glick et al. (24) found that vitamin A-deficient rats have decreased levels of TGF-␤2 in their intestinal mucosa. Administration of RA to these vitamin A-deficient animals restored TGF-␤2 levels. This work emphasizes the importance of retinoids in controlling expression of these ligands in vivo and illustrates the responsiveness of intestinal mucosa to retinoids. In other studies, vitamin A-deficient animals displayed gastrointestinal abnormalities that included decreased mucus production, expansion of proliferation zones within colon crypts, and ion flux alterations (17)(18)(19)(20)(21)(22)(23). In addition, retinoid analogs have proven effective in preventing 5-azoxymethane-induced colon carcinoma formation in rats (17,(25)(26)(27)(28). Our findings suggest that the RA response pathway is a target for inactivation in colon cancer. We provide data that support the silencing of RA biosynthesis as a downstream consequence of APC mutation but not necessarily a consequence of ␤-catenin disregulation. We offer a new model explaining the potential relationship between APC, CDX2, retinoid biosynthesis, and differentiation. Specifically, APC and CDX2 may control intracellular levels of RA and, ultimately, an RA-mediated program of differentiation. Although several target genes for APC/␤-catenin⅐TCF/LEF have been described, there have been no reports of specific, prodifferentiation signaling pathways, like retinoids, that are under the direct control of APC. Ultimately, this work could lead directly to a testable clinical hypothesis aimed at pharmacological restoration of retinoid activity.