Retinoic Acid Regulation of Cdx1: an Indirect Mechanism for Retinoids and Vertebral Specification

Abstract

Retinoic acid (RA) is required for diverse developmental programs, including vertebral specification. Both RA receptor disruption and excess RA result in homeotic transformations of the axial skeleton. These effects are believed to occur through altered expression of Hox genes, several of which have been demonstrated to be direct RA targets. Members of the cdx (caudal) homeobox gene family are also implicated in regulating Hox expression. Disruption of cdx1 results in vertebral homeotic transformations and alteration of Hox expression boundaries; similar homeosis is also observed in cdx2 heterozygotes. In Xenopus, gain or loss of Cdx function affects vertebral morphogenesis through a mechanism that also correlates with altered Hox expression. Taken together with the finding of putative Cdx binding motifs in several Hox promoters, these data strongly support a role for Cdx members in direct regulation of expression of at least some Hox genes. Most retinoid-responsive Hox genes have not been demonstrated to be direct RA targets, suggesting that intermediaries are involved. Based on these findings, we hypothesized that one or more cdx members may transduce the effects of RA on Hox transcription. Consistent with this, we present evidence that cdx1 is a direct RA target gene, suggesting an additional pathway for retinoid-dependent vertebral specification.

Retinoids such as retinoic acid (RA) play key roles in vertebrate development (43). The retinoid signal is transduced by two families of nuclear receptors: the RA receptors (RARs) and their isoforms (RARα1 and -α2; RARβ1, -β2, -β3, and -β4; and RARγ1 and -γ2) and the retinoid x receptors (RXRα, -β, and -γ). These receptors mediate ligand-dependent target gene transcription typically by binding as heterodimers to cis-acting RA response elements (RAREs) (6, 14, 29, 47).

Vertebral specification is believed to be governed by a Hox “code.” This model is supported by a multitude of studies which demonstrate that anterior or posterior shifts in Hox expression or the ablation of specific Hox genes often leads to alterations in somite identity, as inferred by vertebral homeosis (10, 13, 31). Transplantation experiments suggest that vertebral specification, and hence Hox expression boundaries, is established in the unsegmented paraxial mesoderm at or shortly following gastrulation (36).

Both RAR knockout studies and studies on the effects of excess RA demonstrate roles for retinoids in vertebral morphogenesis (10, 31). RARγ null mice display axial malformations, including vertebral homeotic transformations (25). Although disruption of either RARα or RARβ2 does not affect skeletal development (27, 33), both receptors collaborate with RARγ in vertebral development, as judged by the marked increase in frequency and severity of axial skeletal defects in the corresponding double null mutants (26). A role for Hox genes in this program is suggested by the finding of altered expression of some Hox members following RA treatment in vivo. Moreover, certain Hox mutants are phenocopies of the axial transformations observed in RAR mutants (25, 26). However, despite these correlations, few RA-responsive Hox genes have been shown to be direct RAR targets (11, 24, 35, 38, 39, 45).

Several lines of evidence suggest that vertebrate caudal homologues are key regulators of Hox expression. The murine caudal homologues cdx1, cdx2, and cdx4, are expressed in overlapping domains in the primitive streak region, with expression maintained in the posterior embryo through embryonic day 12.5 (E12.5) (5, 12, 34). These expression patterns suggest that a gradient of cdx function exists in the posterior embryo, which may reflect a means of regulating expression of different cohorts of Hox genes during somite specification (30). Consistent with this, cdx1 null mutants as well as cdx2 heterozygotes exhibit vertebral homeotic transformations (8, 46), which, in the former case, correlate with altered expression of certain Hox genes. The finding of consensus Cdx response elements in the promoter regions of several Hox loci (5, 46) further supports a role for Cdx members in direct regulation of Hox expression. Similar observations in Xenopus and Caenorhabditis elegans suggest that this pathway may be conserved (17, 20).

As most RA-responsive Hox genes are not known to be direct RAR targets, we hypothesized that a cdx member(s) may function as an intermediate. In support of this, we present evidence that cdx1 responds to RA and RAR ablation in vivo in a manner consistent with it being a direct retinoid target. These results suggest an indirect pathway by which RA regulates Hox expression via direct control of cdx1.

MATERIALS AND METHODS

Animals.

The RARγ, RARα1, and RARα1/γ null mice used in the present study have been described previously (26, 27). RARγ heterozygous and null embryos were generated from RARγ^+/− intercrosses, whereas RARα1 and RARα1/γ null embryos were derived from RARα1^−/−γ^+/− intercrosses. Wild-type embryos were obtained either from RARγ^+/− matings, from intercrosses of wild-type stock from the RARγ colony (C57BL/6-129Sv hybrid), or from CD-1 intercrosses. No overt differences in gene expression or RA response were noted between any of these backgrounds. Females were dosed by oral gavage with all-trans RA dissolved in corn oil to a final delivery of 10 or 100 mg/kg of body weight at E7.5, E8.5, or E9.5 (noon of the day of plug appearance was considered E0.5). Animals were sacrificed 1 to 8 h posttreatment, and embryos were dissected in phosphate-buffered saline (PBS), fixed overnight in 4% paraformaldehyde, dehydrated through a methanol series, and stored at −20°C in 100% methanol. Yolk sacs were used to establish genotype by PCR as described previously (19). In some experiments, embryos were treated as described above and the presomitic caudal embryonic region was dissected out, snap frozen, and stored at −80°C prior to RNA isolation.

In situ hybridization analysis and embryo culture.

Embryos were pooled by stage, genotype, and RA treatment and rehydrated. Whole-mount in situ hybridization was performed as described previously (50), using a riboprobe generated from the cdx1 cDNA (34). After hybridization, embryos were cleared and photographed. Some specimens were then postfixed in 4% paraformaldehyde–0.2% glutaraldehyde at 4°C for 30 min, rinsed in several changes of PBS, embedded in Paraplast (Fisher), and sectioned.

Embryo culture was performed essentially as described previously (15). Embryos were dissected out in PBS containing 10% fetal bovine serum and stored briefly in Dulbecco's modified Eagle's medium (DMEM) (Life Technologies) buffered with HEPES. Embryos were cultured in DMEM-rat serum (50:50) preequilibrated with 5% O₂–5% CO₂ in N₂ at 37°C. Cultures were maintained for 4 h in the presence of RA (10⁻⁹ to 10⁻⁷ M in dimethyl sulfoxide [DMSO]) or vehicle (0.1%) prior to in situ hybridization analysis. In some experiments, cycloheximide (30 mg/ml) or the vehicle (EtOH, 0.1%) was also included in the culture medium for 30 min prior to addition of RA or DMSO. To monitor de novo protein synthesis in the latter experiments, 100 μCi of [³⁵S]methionine was added per ml, and incorporation of the label was assessed by filter binding as described previously (3).

Cell culture and transfection analysis.

F9 embryocarcinoma cells were maintained in DMEM (Life Technologies) supplemented with glucose (4.5 g/liter), 10% fetal bovine serum, and gentamicin (10 μg/ml). For routine culture, cells were passaged every third day into gelatinized 100-mm tissue culture plates and cultured at 37°C in 5% CO₂. For Northern blot experiments, cells were seeded in 100-mm plates (approximately 10⁶ cells/plate) and treated the following day with all-trans RA (1 μM) dissolved in DMSO (final concentration, 0.1%). Control cultures were treated with DMSO only. For Northern blot analysis, cells were harvested 2 to 48 h posttreatment, snap frozen, and stored at −80°C prior to RNA extraction.

To assess the requirement for de novo protein synthesis, cells were treated for 30 min with 15 or 30 μg of cycloheximide/ml or with the vehicle alone prior to RA treatment. Cells were then harvested, snap frozen, and stored as described above prior to Northern blot analysis. Parallel cultures were incubated with 40 μCi of [³⁵S]methionine/ml, and incorporation of the label was assessed by filter binding.

For transfection analysis, cells were passaged into gelatin-treated six-well cluster plates (approximately 10⁵ cells/well) and transfected 24 h later using the calcium phosphate method. DNA mixtures were comprised of 1.5 μg of luciferase reporter construct, 0.75 μg of a lacZ expression vector as an internal control, and pBluescript KS(+), to a final concentration of 5 μg DNA per transfection. The following day, the medium was replenished and cells were treated with RA or DMSO, and culture was continued for 24 h. Monolayers were then rinsed twice in ice-cold PBS, and cells were disrupted by addition of 250 μl of lysis buffer (0.1 M Tris-Cl [pH 8], 1% NP-40, 1 μM dithiothreitol) for 5 min at room temperature. Cell lysates were collected and assessed for luciferase and β-galactosidase as described previously (3), and β-galactosidase activity was used to correct for transfection efficiency. Results were corrected for background (empty expression vector) and expressed as the means of three independent transfections. Unless otherwise stated, each experiment was repeated a minimum of three times.

To assess RA regulation in stable transfectants, 50 μg of the parental 2-kb cdx1 reporter vector was linearized and cotransfected with 5 μg of a neomycin selection vector. Cells were selected by culture in the presence of 300 μg of G418 (Life Technologies)/ml for 2 weeks. Clones (approximately 100) were pooled and used to assess RA response by luciferase assay as described above.

Northern blotting and representative cDNA analysis.

Total RNA was extracted from frozen embryos or cell pellets by using Trizol (Life Technologies) according to the manufacturer's directions. Fifteen micrograms of total RNA was resolved by electrophoresis through a formaldehyde gel and subjected to Northern blotting using Hybond N (Amersham) as described by the manufacturer. To quantify differences in embryonic gene expression, caudal tissue (posterior to the closed neural tube) was used for the generation of representative cDNA by PCR as previously described (18), followed by analysis by Southern blotting. Hybridizations were performed overnight at 42°C in a formamide-based buffer (40% formamide, 0.9 M sodium chloride, 50 mM sodium phosphate, 2 mM EDTA, 4× Denhardt's solution, 0.1% sodium dodecyl sulfate [SDS]) supplemented with 0.1 mg of denatured salmon sperm DNA/ml; denatured probe (approximately 10⁶ cpm/ml) prepared by random priming was used. Blots were washed in 2× SSC–0.1% SDS (1× SSC is 0.15 M NaCl plus 0.015 M sodium citrate) three times at 65°C, followed by three washes in 0.2× SSC–0.1% SDS at the same temperature, and signal was revealed by autoradiography using X-Omat film (Kodak). For representative cDNA analysis, following autoradiography, densitometry was performed using Alpha Imager IS-1000 software (Alpha Innotech Corporation, San Leandro, Calif.). Values were normalized with respect to β-actin and expressed as fold change relative to untreated controls.

EMSA.

The region of the cdx1 promoter region conferring RA response, as defined by transfection analysis, was scanned by initially using fragments amplified by PCR. Each fragment was purified, end labeled with T4 polynucleotide kinase, and tested for RAR and RXR binding by electrophoretic mobility shift assay (EMSA). The putative RARE identified by this approach was evaluated for binding by EMSA with a double-stranded end-labeled oligonucleotide harboring either the sequence 5′-AAGGGTCGTGACCCT or the mutated sequence 5′-AAGGGCAAGTTCCCT (altered nucleotides are underlined). The end-labeled double-stranded oligonucleotide 5′-GGGTAGGGTTCACCGAAAGTTCACTCGCA, harboring a consensus RARE (DR5), was used as a positive control in all binding assays. Nuclear extracts from Cos cells which had been either mock transfected or transfected with expression vectors encoding RARα and RXRγ were used as a source of protein. Binding reaction mixtures containing approximately 2 ng of probe (50,000 cpm) and 2 μl of nuclear extract (3 μg of protein) were equilibrated for 30 min at room temperature and separated by electrophoresis through a 5% polyacrylamide gel containing 0.25× Tris-borate-EDTA. For antibody supershifts, 0.5 μl of anti-RARα antibody (Santa Cruz) was added to protein extracts and equilibrated on ice for 30 min prior to addition of probe and further incubation as described above. Specificity of binding was assessed by competition with a 100-fold excess of unlabeled RARE nucleotides (sequences noted above) or with nucleotides harboring an SP-1 binding motif (5′-TCGATCGGGGCGGGGCGA). In other EMSA experiments, electrophoresis was initiated at various times after probe addition or with various amounts of transfected Cos cell extracts.

Isolation of genomic sequences and derivation of plasmids.

Sequences were isolated from a murine phage genomic library. A BamHI-NotI fragment containing the endogenous transcription initiation site (16) and extending approximately 2 kb 5′ was ligated into the promoterless luciferase expression vector pXP2 (37), and subsequent deletion constructs were prepared by using convenient restriction sites. A reporter bearing the putative cdx1 RARE was obtained by ligating either the double-stranded oligonucleotide 5′-AAGGGTCGTGACCCCT, harboring the wild-type sequences, or the mutated sequence 5′-AAGGGCAAGTTCCCCT into pTK109-Luc. A positive RA-responsive control, RARE-Luc, was derived by ligating the double-stranded oligonucleotide GGGTAGGGTTCACCGAAAGTTCACTCGCA, bearing the RARβ2 consensus RARE, into pTK109-Luc. Site-directed mutagenesis was performed using a Transformer kit (Clontech). All constructs were confirmed by sequencing.

RESULTS

RA induces cdx1 in vivo.

In untreated wild-type embryos at E8.5, cdx1 transcripts were abundant in the ectoderm and mesoderm in the primitive streak region, with weaker expression in caudal neuroepithelium, in agreement with previous studies (Fig. 1A) (34). Eight hours following gavage with RA (100 mg/kg), expression was markedly increased (Fig. 1B) in all germ layers of the caudal embryo (Fig. 1D, compare to the control in 1C), with induction detectable as early as 1 h posttreatment (data not shown). Note that in this and other experiments, embryos to be compared were processed in parallel using the same probe to control for variables in signal intensity. Note also that experiments were terminated when strong staining was observed in any of the pooled samples. Therefore, untreated embryos were sometimes understained to clearly demonstrate the effect of RA.

FIG. 1.

Induction of cdx1 by RA in vivo. (A and B) cdx1 expression in E8.5 embryos treated with a vehicle (A) or 8 h following gavage with 100 mg of RA/kg (B). (C and D) Sagittal sections of embryos shown in panels A and B, respectively. Note the marked induction in signal throughout the caudal embryo. (E, F, and G) Semiquantitative analysis of cdx1 expression. Representative cDNA Southern blots were generated from E8.5 caudal tissue, as described in Materials and Methods, and hybridized to either cdx1 (E) or β-actin gene (F) probes. Compare basal expression (panel E, lanes 2 through 4) to expression following an RA treatment (lanes 6 through 8). β-Actin gene expression from the same material (F) was not affected. Densitometry from the blots shown in panels E and F was performed, and cdx1 expression levels were normalized to β-actin gene expression levels. Results (G) are expressed relative to those for untreated controls and indicate a ninefold induction. Due to saturation of the X-ray film in the lanes from the treated samples, this is likely an underrepresentation of regulation. (H to J) Ex vivo response of cdx1 to RA. E8.5 embryos were excised and cultured for 4 h in the presence of a vehicle (H), 10⁻⁹ M RA (I), or 10⁻⁷ M RA (J), following which cdx1 expression was assessed by whole-mount in situ hybridization. Bar, 500 μm (A, B, H, I, and J) and 250 μm (C and D).

In order to calculate induction, semiquantitative PCR was employed to compare cdx1 transcript abundance in control and treated caudal embryo tissue. Treated embryos clearly exhibited a strong induction of message relative to untreated controls (Fig. 1E), whereas actin transcript abundance was not affected (Fig. 1F). Densitometric assessment of this regulation revealed a ninefold induction of cdx1 signal when normalized for actin transcript abundance (Fig. 1G).

Embryo culture was employed to allow administration of known concentrations of RA and to more precisely regulate the time of exposure. In these experiments, both doses of RA tested (10⁻⁹ and 10⁻⁷ M) rapidly induced cdx1 expression (Fig. 1I and J, compare to control in 1H). Notably, the lower concentration is in close agreement with the K_d of the RARs for RA (2), further supporting a physiological role for retinoids in regulating cdx1 expression. Similar results were obtained by using a low dose of RA (10 mg/kg) in vivo at E8.5 (data not shown) and by using tissue culture models (see below).

RA induces cdx1 expression at several developmental stages.

In the mouse, exogenous RA can induce vertebral homeosis from E7.5 to E9.5 in a manner that is coincident with altered Hox expression (22). We therefore determined if cdx1 responded to RA throughout this window. In untreated embryos at E7.5, cdx1 expression was observed in the ectoderm and mesoderm of the primitive streak region (Fig. 2A) as described previously (34), whereas RA elicited a strong induction of expression in the entire streak region at this stage (Fig. 2B). Interestingly, treatment appeared to induce cdx1 precociously in embryos where expression either had not yet commenced or was only weakly detected (Fig. 2E, compare with F). At E9.5, cdx1 transcripts in the caudal embryo were present at low levels (Fig. 2I). At this stage, hybridization was also observed in the forelimb bud mesenchyme, with weaker expression sometimes observed in the presumptive dermamyotome (Fig. 2I) (34). Four hours following treatment at E9.5, message was markedly induced in all of these domains, with a stronger signal consistently observed in the dermamyotome (Fig. 2J).

FIG. 2.

cdx1 expression in wild-type and RARα1/γ mutant embryos. Shown are results for whole-mount analysis of cdx1 expression in E7.5 wild-type embryos without (A and E) or 4 h following (B and F) RA treatment. Note induction throughout the primitive streak region in the treated samples. (C and D) Expression in untreated (C) and RA-treated (D) E7.5 RARα1/γ mutant embryos. Note the reduced expression in the untreated mutant relative to the wild type (C versus A) and the decreased effects of RA on expression in the mutant background relative to controls (compare mutants in panels C and D to wild types in panels A and B). (I and J) cdx1 expression in wild-type embryos at E9.5 without (I) and following (J) RA treatment. In untreated embryos, weak expression of cdx1 was observed in the tail bud, with transcripts also evident in the forelimb bud and dermamyotome (asterisk and arrowhead, respectively, in panel I); RA strongly induced expression in all of these domains (J, compare to panel I). In RARα1/γ null embryos at E8.5 (G and H) and E9.5 (K and L), RA induction was reduced relative to that seen in the wild-type samples (compare effects of treatment in mutants in panels G and H and panels K and L to the effect on wild-type controls in Fig. 1A and B and Fig. 2I and J, respectively). Bars, 50 μm (A through F), 100 μm (G and H), and 75 μm (I through L).

cdx1 expression is altered in RAR null mutants.

cdx1 expression was not overtly different in wild-type controls and RARα1 or RARγ single mutants at E7.5 to E9.5 (data not shown). In contrast, E7.5 RARα1/γ double mutants always exhibited reduced cdx1 expression in the primitive streak region (Fig. 2C, compare with A). In marked contrast, transcript levels in the primitive streak region of these mutants at E8.5 were often comparable to levels in wild-type embryos (Fig. 2G, compare with 1A; also data not shown). At E9.5, expression in the tail bud was either comparable or too weak to compare between RARα1/γ mutants and controls. Similar variability was also seen with regard to expression in the limb buds and dermamyotome of these mutants, with expression sometimes weaker in the mutants than in stage-matched wild-type controls (Fig. 2K, compare with I). However, differences in expression were not consistently observed between RARα1/γ mutants and controls at E9.5. Similar variability in signal intensity was often seen in control E9.5 samples, suggesting highly variable and dynamic expression of cdx1 at this stage. Such variance precluded an accurate determination of the effects of RAR loss on cdx1 levels in these embryos.

We also determined whether RARα1 and/or RARγ was required for induction of cdx1 by exogenous RA. Following treatment at E8.5, caudal cdx1 expression in RARα1 null embryos was comparable to expression in the wild type, whereas induction in RARγ mutants was only modestly reduced relative to that in the wild-type controls (data not shown). In contrast, induction in RARα1/γ mutants was markedly compromised in all normally responsive domains (primitive streak, dermamyotome, and forelimb bud) at all stages examined (compare untreated mutants in Fig. 2C, G, and K with the treated stage-matched samples in Fig. 2D, H, and L; also compare the relative induction in these double mutants to that seen in wild-type specimens at comparable stages (Fig. 2; see Fig. 1A and B for E8.5 wild-type embryos).

cdx1 is a direct RA target.

To further investigate the effects of RA on cdx1 expression, we employed F9 embryocarcinoma cells. In these cultures, RA up-regulated cdx1 transcript levels as early as 2 h after treatment, with a maximum level attained after 24 to 48 h (Fig. 3A). In both F9 cells (Fig. 3B) and embryo cultures (Fig. 3C through F), this response was independent of de novo protein synthesis, as induction was evident in the absence or presence of cycloheximide (cycloheximide treatment resulted in ≥95% inhibition of de novo protein synthesis in either system). Notably, in embryo cultures, cdx1 message was increased by cycloheximide treatment alone (Fig. 3E, compare to C), whereas a further increase in message abundance was seen upon subsequent treatment with RA (Fig. 3F). This superinduction effect suggests both an increase in transcription and a stabilization of message, a common phenomenon for immediate-early target genes.

FIG. 3.

cdx1 induction is independent of de novo protein synthesis. (A) cdx1 expression in F9 cells treated with RA (lanes 2, 4, 6, 8, and 10) or DMSO (lanes 3, 5, 7, 9, and 11) for the indicated times. (B) Northern blot of RNA from F9 cells treated for 4 h with RA (lanes 2, 4, and 6) or DMSO (lanes 1, 3, and 5) with the indicated amounts of cycloheximide. Both blots were reprobed for β-actin as a loading control. (C and D) cdx1 expression in cultured embryos. Embryos were cultured for 4 h in the presence of a vehicle (C), 20 μg of cycloheximide (CHX)/ml (E), or 10⁻⁶ M RA (D) or RA plus cycloheximide (F) prior to in situ hybridization. Results shown are typical of several experiments. Bar, 400 μm (C to F).

We further investigated the mechanism of transcriptional regulation by using transfection approaches. The genomic region of cdx1, comprising approximately 2 kb of 5′ sequences, including the endogenous transcriptional start site and a portion of the 5′ untranslated region (5′-UTR) (16), was as potent as a synthetic RARE at eliciting an RA response in transient-transfection assays in F9 cells (Fig. 4A). Removal of sequences comprising the endogenous transcriptional start site and 5′-UTR revealed that the remaining, nontranscribed, sequences elicited an induction in transfection assays when coupled to a heterologous promoter (data not shown; see also below), suggesting that the RA response was not mediated through posttranscriptional mechanisms operating via the 5′-UTR. Finally, the 2-kb reporter was used to establish stably integrated reporter cell lines. These cell lines responded to low (10⁻⁹ M) levels of RA in dose-response experiments (data not shown), indicating that, as observed in vivo, this region conferred a response to physiological levels of RA.

FIG. 4.

Identification of an RARE in the cdx1 promoter. (A) F9 cells were transfected with the indicated reporter constructs, and luciferase activity was determined 24 h following treatment with 10⁻⁶ M RA. Results are expressed as fold induction by RA relative to that in untreated cultures. Bm, BamHI; Bs, BstXI; P, PstI; X, XmnI. Numbering to the left indicates the 5′-most position relative to the transcriptional start site. DR5 is the positive control reporter construct. (B) The cdx1 RARE motif was tested for receptor binding by EMSA using the radiolabeled oligonucleotide probes indicated at the bottom. Lane 1, no extract; lanes 2 and 9, mock-transfected cells; lanes 3 through 8 and 10 through 15, cdx1 RARE and DR5 probe, respectively, incubated with extracts from cells transfected with RARα plus RXRγ. Specific binding complexes (“Specific”) were not affected by nonspecific competitor (lanes 8 and 15) or by mutated cdx1 RARE (lanes 6 and 14) but were competed by excess DR5 (lanes 7 and 12) or Cdx1 RARE (lanes 5 and 13). “Super Shift” indicates complexes formed by incubation with anti-RARα (lanes 4 and 11). (C) Transfection analysis, as in panel A, indicates that point mutation of the RARE sequences (∗) in the context of the 2-kb parental cdx1 promoter results in complete loss of RA induction in F9 cells. A single copy of the cdx1 RARE motif confers an RA response with a heterologous promoter, and this effect is lost upon mutation of these sequences (Cdx1 RARE*). (D and E) Stability of the receptor association with cdx1 RARE compared to the canonical DR5 RARE motif. (D) Comparable amounts of the probes were incubated with different amounts of extracts from receptor-transfected Cos cells for 30 min before electrophoresis. The protein (Prot.) amounts used were 3.00, 1.00, 0.30, 0.10, 0.03, 0.01, and 0.00 μg/incubation. (E) cdx1 RARE and DR5 probes were incubated with 3 μg of Cos extract for the indicated amount of time, and binding complexes were resolved by electrophoresis.

Deletion analysis and transfection assays mapped the RA response region between −694 and −185 relative to the transcription start site (Fig. 4A). As a typical RARE (DR5) was not observed in these sequences, EMSAs were employed to identify the element. These experiments identified the motif AAGGGTCGTGACCCT as a target for RAR and RXR binding and demonstrated that all parameters of the cdx1 RARE investigated by EMSA were identical to those exhibited by the control DR5 element (Fig. 4B, left and right sides, respectively). In particular, receptor binding to cdx1 sequences was efficiently competed with excess unlabeled self or consensus DR5 RARE sequences, but not by an SP-1 binding motif or a mutated cdx1 RARE. Conversely, the putative cdx1 RARE, but not the mutated element, competed efficiently for binding to the DR5 RARE. Specificity was further confirmed by supershift assays, which demonstrated the presence of RARα in the complex. Although this supershift was not quantitative (for unknown reasons), the cdx1 and DR5 elements exhibited identical degrees of antibody binding (Fig. 4B, compare supershift binding between lanes 4 and 11).

The relative affinity between the cdx1 and DR5 motifs was also assessed either by varying the RAR-RXR concentration or by comparing relative binding as a function of reaction time. These data suggest that the cdx1 motif is tightly associated with receptor complexes comparable to the DR5 control element, differing approximately twofold (Fig. 4D and data not shown). Moreover, the cdx1 sequences exhibited only modestly slower kinetics of association relative to the DR5 element (Fig. 4E). These data are consistent with the finding that the cdx1 RARE was absolutely essential for retinoid response in the context of the 2-kb promoter, as mutation of this motif completely abolished the response in F9 cells (Fig. 4C). Moreover, a single copy of this element was also sufficient to confer an RA response to a heterologous basal promoter (Fig. 4C). These cdx1 sequences bear remarkable similarity to the rat growth hormone promoter TRE, the thyroid hormone response element, which has previously been shown to confer an RA response in transfection assays (49). The finding that this motif is perfectly conserved in the human cdx1 promoter (data not shown) further suggests a conserved and important role for this element in directing expression.

DISCUSSION

Many Hox genes respond to RA both in tissue culture and in vivo, and this relationship is believed to be a principal means by which retinoids act in vertebral specification (10). Although a number of Hox genes have been shown to be direct RA targets, the mechanism(s) by which RA affects expression of most of the responsive Hox genes is largely unknown. Our present findings demonstrate that cdx1 is a direct RA target, which, together with the established relationship between Cdx and Hox gene expression, strongly suggests a novel pathway for retinoids in axial specification.

Contribution of RA to cdx1 expression.

Findings from several models illustrate a role for caudal family members in anterior-posterior patterning. cdx1 null mutants exhibit anterior vertebral homeosis which is coincident with posterior shifts in the expression boundaries of certain Hox genes; similar vertebral defects are also observed in cdx2 heterozygotes (8, 46). In Xenopus, gain or loss of the function of Xcad (the frog homologue of cdx4) results in patterning defects which correlate with altered Hox gene expression along the anterior-posterior axis (17). Taken together with the presence of potential Cdx response elements in a number of Hox promoters (7, 46), these data support a role for Cdx members in the direct control of Hox expression.

Our finding that cdx1 is a direct RA target is in agreement with a number of observations. The homeotic transformations and rib fusions observed in cdx1 null offspring (46) are reminiscent of the axial skeletal malformations exhibited by certain RAR null offspring (26). These similarities occur with respect to both the nature of the defects and their location along the vertebral column, being largely restricted to the cervical region in both classes of mutants. Consistent with this finding, a reduction in cdx1 message was apparent in RARα1/γ mutants at E7.5. This is in agreement both with the window during which the cervical vertebrae are presumed to be specified and with the high frequency of vertebral defects observed in RARα1/γ mutants relative to RARα1 null offspring (which appear to be normal). However, reduction of cdx1 expression was not observed in RARγ null embryos. This may relate to the low incidence of homeosis seen in these mutants (25), suggesting that effects on cdx1 may be observed only at a correspondingly low frequency and/or may be too subtle to be readily detected by in situ hybridization techniques.

Our present data are entirely consistent with retinoid distribution studies. In the mouse, biologically active retinoids are first detected in the primitive streak region at E7.5 (4, 41). This correlates closely with the initial appearance of cdx1 transcripts (34) and the ability of exogenous RA to precociously induce cdx1 at this time. Moreover, cdx1 expression was strongly reduced in RARα1/γ null embryos at E7.5. The finding that expression at E8.5 was not reproducibly altered in the double mutant background is likely related to the fact that retinoid activity is greatly reduced or absent in the primitive streak region commencing at this time (41). These data suggest that RA plays a role in the initial period of cdx1 expression (perhaps to initiate expression) but that an additional factor(s) is involved in maintaining later phases of transcription. In this regard, in Xenopus, fibroblast growth factor (FGF) regulates Hox expression via control of Xcad (17, 40). Taken together with our present findings, this suggests that both retinoid signaling and FGF signaling converge on a common target gene. Indeed, FGF and RA act synergistically in inducing posterior Hox genes in Xenopus (9). However, whether this mechanism can be extrapolated to other vertebrates is unclear, as a relationship between FGF and cdx expression has not been described for the mouse.

RAR-specific regulation of cdx1.

Studies with F9 cells suggest a key role for RARγ in induction of cdx1 (48), an observation that contrasts with our findings in vivo. However, cdx1 induction in F9 cells is only maximal after 24 to 48 h of treatment, in marked contrast to induction in vivo, which is evident 1 h posttreatment. This difference may be due, in part, to the fact that RARγ null F9 cells are refractory to RA-induced differentiation, suggesting that additional RARγ-dependent events impact cdx1 transcription. Interestingly, another RA target gene, CYP26, exhibits similar differences between regulation in vivo and in F9 cells (1, 18). Although the basis for these discrepancies is speculative, these findings may be indicative of common cell-type-specific regulatory mechanisms which govern control of expression of these, and perhaps additional, RA target genes.

RA, Hox expression, and somite specification.

RA has been suggested to be a “posteriorizor” in the activation-transformation model of neurulation (reviewed in reference 42). RA can impart more posterior molecular characteristics on the anterior neuroepithelium, and interference with RAR signaling in Xenopus, or vitamin A deficiency in quail, results in hindbrain patterning defects, presumably due to effects on target genes such as Hoxa-1 and Hoxb-1 (11, 23, 28, 32, 44, 45). However, to date, a somite-specific RARE which is essential for expression of a Hox member with definitive function in paraxial mesoderm has not been described. As defects in cdx1 null mice appear to be related only to somitic Hox misexpression, it is tempting to speculate that RA-dependent vertebral specification may manifest largely through cdx1. In contrast, the nonhomeotic axial patterning defects observed in RARα1/γ double mutant offspring, which are not exhibited by cdx1 mutants, clearly underscore the existence of other retinoid target genes involved in vertebral morphogenesis. The nature of these genes is unknown.

cdx1 and retinoid-induced teratogenesis.

Excess RA has profound effects on vertebrate development and is capable of eliciting, among other malformations, neural tube and limb defects, axial truncation, and homeotic transformation, depending on both the dose and the embryonic stage upon exposure. As overexpression of cdx members can lead to neural tube defects in mouse embryos as well as caudal malformations of Xenopus tadpoles (7, 17), excess RA could conceivably exert some of its teratogenic effects through misexpression of cdx1. In this regard, RARγ is essential for retinoid-induced axial truncation (25). The finding that cdx1 induction in RARγ mutants was only modestly compromised suggests that it is not involved in eliciting this malformation. However, in Xenopus, relatively small changes in Xcad3 gene dosage result in dramatic differences in phenotypic outcome (17). Thus, we cannot exclude the possibility that the relatively small difference in cdx1 induction in RARγ mutants is significant.

Current models suggest that up-regulation of cdx should lead to posterior transformation concomitant with anteriorization of Hox expression domains, whereas loss of Cdx function should lead to the converse situation. RA exposure at E7.5 to E9.5 induces cdx1, yet posteriorization events are seen only following treatment at E7.5. Exposure at later stages results in predominantly anterior transformation with loss of Hox expression (21, 22). At least two possibilities may explain these observations. First, as previously discussed (17), the overall level of Cdx proteins may be of importance, and induction of cdx1 may be offset by reduction of cdx2 (our preliminary observation) and cdx4 (18), thus resulting in a loss of function at E8.5 and E9.5. Alternatively, Cdx members may differentially regulate target genes, as previously suggested (7). Additional studies are needed to address this as well as to further investigate the relationship between RA, cdx, and Hox expression.