The Recruitment of SOX/OCT Complexes and the Differential Activity of HOXA1 and HOXB1 Modulate the Hoxb1Auto-regulatory Enhancer Function*

Regionally restricted expression patterns of Hox genes in developing embryos rely on auto-, cross-, and para-regulatory transcriptional elements. One example is the Hoxb1 auto-regulatory element (b1-ARE), which drives expression of Hoxb1 in the fourth rhombomere of the hindbrain. We previously showed that HOXB1 and PBX1 activate transcription from the b1-ARE by binding to sequences required for the expression of a reporter gene in rhombomere 4 in vivo. We now report that in embryonal carcinoma cells, which retain characteristics of primitive neuroectodermal cells, the b1-ARE displays higher basal and HOX/PBX-induced activities than in other cell backgrounds. We have identified a bipartite-binding site for SOX/OCT heterodimers within the b1-ARE that accounts for its cell context-specific activity and is required for maximal transcriptional activity of HOX/PBX complexes in embryonal carcinoma cells. Furthermore, we found that in an embryonal carcinoma cell background, HOXB1 has a significantly higher transcriptional activity than its paralog HOXA1. We map the determinants for this differential activity within the HOXB1 N-terminal transcriptional activation domain. By using analysis in transgenic and HOXA1 mutant mice, we extended these findings on the differential activities of HOXA1 and HOXB1 in vivo, and we demonstrated that they are important for regulating aspects of HOXB1 expression in the hindbrain. We found that mutation of the SOX/OCT site and targeted inactivation of Hoxa1 both impair the response of the b1-ARE to retinoic acid in transgenic mice. Our results show that Hoxa1 is the primary mediator of the response of b1-ARE to retinoic acid in vivoand that this function is dependent on the binding of SOX/OCT heterodimers to the b1-ARE. These results uncover novel functional differences between Hox paralogs and their modulators.

The HOX homeodomain-containing transcription factors control cell fate and developmental patterns in all metazoans, leading to the generation of morphological differences along body axes (reviewed in Ref. 1). In most vertebrates the four Hox clusters encode 39 distinct proteins in which the homeodomain (HD) 1 and flanking amino acids dictate the DNA binding specificity by recognizing a restricted set of sites containing the core consensus sequence TNAT(G/T)(G/A) (2)(3)(4). Despite the apparent similarity in consensus DNA recognition sites, HOX proteins can modulate their binding properties and specificity through the concomitant activity of an emerging array of cofactors. Interactions with cofactors such as the HD-containing proteins of the EXD/PBX (PBC) (3)(4)(5)(6) and MEIS/PREP (MEINOX) (7)(8)(9)(10)(11) families can modulate the affinity and stability of DNA binding and regulate transcriptional activity of HOX proteins.
Transgenic analysis in mice has led to the identification of sequences that function as HOX target sites in vivo for the auto-, cross-, and para-regulatory interactions among Hox genes (6,(12)(13)(14)(15). Several of these in vivo target sites are composed of bipartite and overlapping HOX-and PBX-binding motifs (HOX/ PBC sites) and represent useful models for analyzing the function of HOX-containing transcriptional complexes. Proteinprotein contacts involved in functional interactions between the HOX and the EXD/PBX proteins were found to be mediated by both the N-terminal region of the HD (16) and the short conserved hexapeptide or YPWM motif (17)(18)(19), located upstream of the HD in a subset of HOX proteins (paralogy groups [1][2][3][4][5][6][7][8]. Recent crystal structure analysis of HOX/PBX homeodomain complexes bound to DNA target sites with relevance in vivo have provided a useful model for understanding how these interactions can influence specificity of binding (20,21).
The identification of bona fide Hox-responsive enhancers and promoters allowed analysis of the transcriptional properties of these proteins. These studies have shown that HOX proteins share the modular type structure of most eukaryotic transcription factors, featuring separate DNA binding and activator or repressor domains (5,(22)(23)(24)(25)(26). One of the HOX-binding elements examined in most detail in vivo is contained in a highly conserved auto-regulatory enhancer of the Hoxb1 gene (b1-ARE). This is a key cis-regulatory element for the normal rhombomere 4 (r4)-restricted expression of Hoxb1 in the devel-oping hindbrain. Genetic analysis has revealed that in an early phase Hoxa1 and Hoxb1 function synergistically to establish initial r4 identity by triggering the auto-regulatory loop (6,27,28). As a part of this process, the early activation of Hoxb1 and Hoxa1 themselves is directly mediated by RA signaling through the presence of RA response elements (RAREs) located at the 3Ј ends of both genes (28 -31). In later stages Hoxb1 is required to maintain r4 identity (27,28). Even though the b1-ARE does not contain a canonical RARE, it indirectly participates in mediating the RA-dependent ectopic activation of Hoxb1 in anterior regions (r2), through its auto-regulatory ability (6).
Auto-and para-regulation of Hoxb1 are dependent on the cooperative binding of HOXB1 or HOXA1, and a member of the PBX family on three conserved sequence motifs (R1, R2, and R3) in the b1-ARE related to a consensus bipartite HOX/PBCbinding site (5,6). Each of these HOX/PBX repeats contributes to the r4-restricted expression of a reporter gene in transgenic mice and to the indirect RA response. The R3 motif, however, makes the largest contribution to the b1-ARE regulatory activity (6). Interestingly, the b1-ARE is not active in other regions of the embryo where Hoxb1 and Pbx1 are coexpressed (spinal cord, paraxial mesoderm, endoderm, and limb buds), suggesting that region-specific expression in the hindbrain might be determined by differential activity of additional factors.
We reported previously (5) that the human HOXB1 and PBX1 proteins cooperatively activate transcription from a basal promoter under the control of the b1-ARE in transfected mammalian cells. Comparing the functional selectivity of proteins from a range of Hox paralogy groups, we showed (5) that only a restricted subset of HOX proteins (HOXA1, HOXB1, and HOXB2) are able to activate transcription from the b1-ARE in cooperation with PBX1. Selective recognition of the R3 motif by the HOXB1/PBX1 complex is mediated by the N terminus of the HOX homeodomain, whereas the major transcriptional activator domain is provided by the HOXB1 N-terminal region (5,26).
In this study we show that the basal and HOX/PBX-induced transcriptional activities of b1-ARE, in comparison to other cell lines, are significantly higher in murine or human embryonal carcinoma (EC) cells, which retain phenotypic and molecular characteristics of primitive neuroectodermal cells (32,33). In this context, HOXB1 displays a stronger transcriptional activation than its paralog HOXA1, in cooperation with PBX1. The differential activity between HOXB1 and HOXA1 is not based on the DNA-binding properties of the HOXB1/PBX1 complex but rather on specific determinants within the HOXB1 transcriptional activation domain that selectively interact with the transcriptional machinery. Furthermore, we show that the full transcriptional activity of HOX/PBX heterodimers on the b1-ARE is dependent on the binding of a SOX/OCT heterodimeric complex to a bipartite site located immediately upstream of the R3 sequence. Mutation of this site, as well as the targeted inactivation of the Hoxa1 gene, impairs the response of the b1-ARE to RA treatment in transgenic mice and the establishment of a full ectopic auto-regulatory circuit of Hoxb1 in vivo. Our results show that the product of the Hoxa1 gene is required for the response of Hoxb1 to RA, and this activity requires a functional SOX/OCT-binding site within the b1-ARE. Together, these results have uncovered novel differences among HOX paralogs and their modifiers.

EXPERIMENTAL PROCEDURES
Expression Vectors and Reporter Plasmids-All expression constructs are derivatives of the SV40 promoter-based expression vector pSG5 (34). HOXB1, HOXA1, HOXB2, and PBX1 expression vectors were described previously (5). pSGB1/A1HD was generated by swapping the region spanning the FDWM motif and the homeodomain of HOXB1 (aa 175-257) with the corresponding region of HOXA1 (aa 206 -289), whereas pSGA1/B1CT was generated by substituting the C-terminal region of HOXA1 (aa 290 -336) with that of HOXB1 (aa 258 -296). The mutated cDNAs were cloned into the BamHI site of pSG5. pGAL1-147 (35) contains the DNA binding domain of yeast GAL4 (amino acids 1-147). pGALVP16 was generated by cloning an PCR-amplified region encompassing the C terminus (last 80 aa) of the VP16 protein in frame with the GAL1-147 protein at the BamHI site of the pGAL1-147 vector. PCR was carried out with Pfu polymerase (Stratagene). All PCR-generated fragments were sequenced on both strands, and expression of all proteins was preliminarily tested using a T7 polymerase-based transcription and reticulocyte lysate-based translation system (Promega).
For transgenic mouse analysis, the wild type Hoxb1 5Ј-flanking StuI-HindIII fragment containing the b1-ARE was cloned into the lacZ basal reporter vector BGZ40 (38) as described previously (6). Point mutations in the SOX/OCT-binding sequence (TAAT to CCGG and TTTGTC to CGCTGT) were generated by site-directed mutagenesis in M13 or inverse PCR, followed by double-strand sequencing.
Cell Culture and Transfection-COS-7 cells were maintained in Dulbecco's modified Eagle's medium, supplemented with 10% fetal calf serum (Life Technologies, Inc.), 100 IU/ml penicillin, and 100 g/ml streptomycin. P19 and NT2/D1 cells were maintained in ␣-minimum Eagle's medium. Transfections were carried out by CaPO 4 precipitation (39). In a typical transfection experiment, 8 g of reporter plasmid, 4 -8 g of expression construct, and 0.2 g of pCMV-␤-gal (CLONTECH) as an internal control were used per 10-cm dish. 48 -60 h after transfection, cells were washed and lysed directly on the plate with a solution containing 1% Triton X-100, 25 mM glycyl-glycine, pH 7.8, 15 mM MgSO 4 , 4 mM EGTA, 1 mM DTT. Extracts were collected, centrifuged to clear the supernatant, and assayed for luciferase and ␤-galactosidase expression as described (40). To rule out the existence of intrinsic differences between COS-7 and P19 cells in sustaining transcriptional activation, we tested for the capability of a GAL4-VP16 chimera to activate transcription from a GAL4-responding reported in both cell lines. The GAL4-VP16 chimera gave identical results in terms of transactivation in both cell lines (data not shown).
Electrophoretic Mobility Shift Assay (EMSA)-To obtain total cell protein extracts, cells were collected from confluent plates, washed with phosphate-buffered saline, pelleted, frozen with liquid nitrogen, and lysed by resuspension in 5 volumes of Extraction Buffer (10 mM Hepes, pH 7.9, 0.4 M NaCl, 0.1 mM EDTA, 0.5 mM DTT, 5% glycerol, 0.5 mM phenylmethylsulfonyl fluoride, 1% Trasylol). The lysate was then centrifuged for 30 min at 34,000 rpm in a Beckman Ti-50 rotor, and the supernatant was stored in aliquots at Ϫ80°C. To obtain embryonic cell extracts, about 50 mouse embryos were collected at 9.5 dpc, washed with phosphate-buffered saline, and lysed as described above. Gel retardation analysis was performed by preincubating the cell extracts (8 g) for 30 min on ice in 20 l of binding buffer (75 mM NaCl, 20% Ficoll, 10 mM Tris-HCl, pH 7.5, 0.5 mM EDTA, 10 mM DTT, 3 g of poly(dG-dC) or poly(dI-dC)), together with 2 l (0.5 ng, 5 ϫ 10 4 cpm) of 32 P-endlabeled oligonucleotide probe. Competition experiments were carried out by adding 2 l of ␣-Oct1 (Santa Cruz Biotechnology), ␣-Oct3/4 (kindly provided by Hans Shöler), or ␣-Sox-2 (a gift from Marco E. Bianchi) antisera before adding the probe. The incubation mixture was resolved by electrophoresis on a 5% polyacrylamide gel in 0.25ϫ TBE at 10 V/cm. Gels were dried and exposed to a Kodak X-AR film at Ϫ70°C. The b1-ARE oligonucleotide probe sequence was 5Ј-AGCTTGTGTCTT-TGTCATGCTAATGATTGGGGGG-3Ј.
Generation and Analysis of Transgenic Mice-Transgenic mice were produced by microinjection of DNA into fertilized eggs from crosses between F1 (CBA ϫ C57) females and males. Whole mount ␤-galactosidase reporter activity in founder embryos was performed as described previously (41). RA exposure was achieved by treating pregnant females with embryos at 7.5 dpc by oral gavage with 200 ml of sesame seed oil containing all-trans-retinoic acid (Sigma), diluted from a 25 mg/ml stock solution in dimethyl sulfoxide, for a final dose of about 20 mg/kg of maternal body weight (42). The inductive response was assayed in embryos at 9.5 dpc.

RESULTS
The Transcriptional Activity of the HOXB1/PBX1 Complex Is Maximal in a Neuroectodermal Cell Background-We have shown previously that the HOXB1/PBX1 heterodimer is able to activate transcription of a reporter gene under the control of the b1-ARE in mammalian cell lines (5,26). To investigate the influence of the cell context on this system, we compared the transcriptional activity of the HOXB1/PBX1 complex in different cell lines, including COS-7, HeLa, NIH3T3, P19, and NT2/D1 cells. The murine P19 and the human NT2/D1 EC cells are originally derived from germ cell tumors and retain many characteristics of primitive neuroectodermal cells, including the ability to differentiate into neurons (32,33) and activate the four Hox gene clusters upon induction with retinoic acid (43)(44)(45). To measure transcriptional activity, reporter constructs were made by placing a luciferase gene under the control of the adenovirus major late basal promoter either alone (pAdMLluc) or in combination with the 150-bp b1-ARE Hoxresponsive region (pAdMLARE). These were cotransfected with expression constructs for HOXB1 and PBX1 into the various cell lines and assayed for reporter activity (Fig. 1). In COS-7, HeLa, or NIH3T3 cells the basal activity of the reporters with and without the b1-ARE enhancer were identical ( Fig. 1 and data not shown). In contrast, in both P19 and NT2/D1 EC cells the construct containing the b1-ARE enhancer (pAdMLARE) displayed a 7-fold higher basal level, compared with the control (pAdMLluc). In non-EC cells (e.g. COS-7 in Fig. 1), coexpression of HOXB1 and PBX1 led to a 8-fold transactivation of the pAdMLARE reporter. In P19 and NT2 cells, however, coexpression of HOXB1 and PBX1 led to significantly higher levels of transactivation, ϳ20and ϳ60-fold, respectively, over the pAdMLARE basal activity of the reporter, corresponding to a ϳ140and ϳ360-fold increase in activity, respectively, of the enhancer-less pAdMLluc reporter (Fig. 1).
Transfection of the PBX1 expression vector alone had no effect on reporter activity in any cell context, whereas transfection of HOXB1 alone led to a 2-3-fold activation only in EC cells presumably through interaction with the endogenous levels of PBX1 expressed in these cells. The higher activity of the HOXB1/PBX1 complex on the b1-ARE target was found not to be due to a higher transfection efficiency in EC cells, as the percentage of transfected cells was found to be comparable to that of the other cell lines (data not shown). Furthermore, we observed no broad differences in the capability of EC versus COS-7 cells to sustain transcriptional activation (see "Experimental Procedures"). Our findings that the basal activity and the ability of the HOXB1/PBX1 complex to act on the b1-ARE are much higher in EC cells, compared with other cell lines, suggest either the presence of EC-specific enhancing factors or COS-7 inhibiting components that modify the b1-ARE regulatory potential.
HOXB1 and HOXA1 Differentially Activate Transcription through the b1-ARE in Neuroectodermal Cells-We previously reported that only HOX proteins belonging to paralogous groups 1 and 2 can bind and activate the full b1-ARE or multimerized versions of R3 in cultured cells in a PBX-dependent manner (5). To compare directly the activity of different HOX/ PBX complexes in an EC cell background, expression constructs for HOXA1, HOXB1, and HOXB2 were cotransfected in P19 cells, together with the reporters containing the entire b1-ARE (pAdMLARE) or a trimer of the R3 motif (5). The basal activity of pAdMLARE was higher than that of pAdMLR3 and was only weakly induced by cotransfection with HOX proteins alone (Fig. 2). On the pAdMLR3 reporter, consistent with our previous results using non-EC cells, the activities of the HOXB1/PBX1 and the HOXA1/PBX1 complexes were comparable, resulting in an 80 -100-fold transactivation. The HOXB2/PBX1 complex was also active although it remained lower (15-fold transactivation) (Fig. 2). In contrast, on the full b1-ARE reporter (pAdMLARE) while HOXB1 and PBX1 induced a 20-fold transactivation (see also Fig. 1), surprisingly HOXA1 only weakly activated (6-fold) the reporter expression in combination with PBX1 (Fig. 2). HOXB2 displayed an even lower (4-fold) activity with PBX1 (Fig. 2). These results indicate that in the EC cell background HOXB1 activates transcription more efficiently than its paralog HOXA1 in cooperation with PBX1 through the entire b1-ARE element. This differential activity was never observed in non-neuroectodermal cell lines.
As the use of a multimerized version of the HOX/PBC repeat 3 target sequence apparently relaxes the paralog selectivity, allowing HOXA1 and HOXB2 to function as effectively as HOXB1, we considered the possibility that other enhancer sequences within the b1-ARE might influence the activity of the different HOX/PBX heterodimers. In order to analyze the contribution of different regions of the enhancer in restricting transcriptional activation, we cotransfected reporter constructs containing sequential 5Ј deletions of the b1-ARE ( Fig. 3B) with HOXB1, HOXA1, and PBX1 in P19 cells. In deletions with increasing size there was a progressive reduction in the overall level of transactivation that progressively reduced expression to 70, 50, and 30% of the entire b1-ARE (Fig. 3A). However, the differential ability of HOXB1/PBX1 versus HOXA1/PBX1 to activate these sites was maintained even on a reporter containing only repeat 3 and its surrounding sequences (ARE⌬R1 ϩ R2, Fig. 3A). Considering that a reporter containing a multimer of the R3 motif alone responds equally well to HOXB1 or HOXA1 (Fig. 2), this result suggests that sequences flanking R3 are important in mediating the selective preference of the b1-ARE for HOXB1.
A SOX/POU-OCT-binding Site Is Necessary for Full Transcriptional Activity of the b1-ARE in EC Cells-In the b-1ARE, located between the R2 and R3 HOX/PBC sites, there is a bipartite sequence motif that is highly conserved in the human, mouse, chicken and pufferfish Hoxb1 locus (6). In this motif (designated SOct in Fig. 4), the 5Ј-most part (TCTTTGTC) closely resembles the target sequence for the HMG box protein SOX-2 (46 -48), whereas the 3Ј part (ATGCTAAT) shows a high degree of similarity with the consensus recognition sequence for POU/Octamer-binding proteins (49). To test the role of this putative SOX/OCT heterodimer-binding site in the function of the b1-ARE in EC cells, we generated a 11-bp mutation (ARESOctm) encompassing both the SOX and the POU/OCT sites in the context of the whole b1-ARE sequence (Fig. 4B). In cotransfection experiments this SOX/OCT site mutation reduced transactivation of the HOXB1/PBX1 complex by 50% and that of HOXA1-PBX1 by 77% (Fig. 4A). Interestingly, this SOX/OCT mutation completely abolished the EC cell-specific basal activity of the b1-ARE, as well as the weak but detectable transactivation by HOXB1 in the absence of PBX1 (Fig. 4A). To help distinguish between the contribution of the SOX-specific versus the POU/OCT-specific sequence on the b-1ARE activity, an additional mutation (ARE⌬R1ϩR2ϩSox, Fig. 4B) was made in the context of the smallest fragment (ARE⌬R1ϩR2) showing differential response to transactivation by HOXB1/PBX1 and HOXA1/PBX1 (see Fig. 3). This mutation, like the ARE SOctm mutation, completely abolished the basal activity of the reporter in P19 cells, reduced transactivation by the HOXB1/ PBX1 complex by 30%, and almost completely suppressed HOXA1/PBX activity (Fig. 4A).
These results indicate that the increased basal activity of the b1-ARE in EC cells compared with COS cells is due to ECspecific or EC-enriched factors interacting with the SOX/OCT site and that the SOX-specific hemisite is necessary for this activity. The same factors appear to affect the overall transcriptional activity of HOX/PBX heterodimers on the b1-ARE through this site. Furthermore, our data show that transactivation by HOXA1/PBX1 is more dependent upon the SOct motif than that by HOXB1/PBX1.
The SOX/OCT Site Is Cooperatively Bound by SOX-2/OCT1 and SOX-2/OCT3/4 Heterodimers-Since the conserved SOX/ OCT site has a functional role in the b1-ARE enhancer activity in EC cells, we looked for EC cell-specific binding activities on this site by EMSA. The octamer-binding proteins OCT1 and OCT3/4 and the HMG box protein SOX-2 are expressed at high levels in EC cells and were previously shown to bind and regulate cooperatively the activity of developmentally regulated enhancers such as that of FGF-4 (50, 51). Hence, we used a double-stranded oligonucleotide spanning the SOX/OCT site of the FGF-4 enhancer as a control. The b-1ARE EMSAs were carried out using as probe a double-stranded oligonucleotide containing the SOX/OCT site (SO-ARE) in combination with total cell extracts obtained from either P19 or NT2 EC cells (Fig. 5). Except where indicated, poly(dG-dC) was used instead of poly(dI-dC) as nonspecific DNA competitor to allow binding by HMG box proteins (50). As shown in Fig. 5 (lanes 2 and 6), in both P19 and NT2/D1 cell extracts a major SOX-2/OCT3/4 complex and a minor SOX-2/OCT1 complex bind the b1-ARE probe. These were specifically competed and/or supershifted by the addition of ant-OCT1 and anti-OCT3/4 antibodies to the binding reaction (Fig. 5, lanes 3 and 4 and 7 and 8). Addition of poly(dI-dC) competed out the binding of the heterodimeric complexes and allowed the formation of monomeric OCT1 and OCT3/4 complexes (Fig. 5, lanes 5 and 9).
In comparison, at least five complexes with similar intensity were detected on the control FGF-4 probe, and they correspond to the individual monomers and the SOX-2/OCT1 and SOX-2/ OCT3/4 heterodimers (Fig. 5, lanes 11 and 15). The identity of all complexes was confirmed by anti-OCT1 and anti-OCT3/4 antibodies (Fig. 5, lanes 12 and 13 and 16 and 17), and again the addition of poly(dI-dC) allowed only binding of the monomeric OCT complexes (Fig. 5, lanes 14 and 18). These results indicate that OCT1 and OCT3/4 bind exclusively as heterodimers with SOX-2 to the SOX/OCT sequence in the b1-ARE.
To check whether binding of the SOX-2/OCT complexes on the b1-ARE SOX/OCT site also occurs in embryonic tissues, we used total cell extracts obtained from the central nervous system of 9.5-dpc mouse embryos. As shown in Fig. 6 (left panel), strong binding of both SOX-2/OCT3/4 and SOX-2/OCT1 complexes to the b1-ARE sequence was observed in embryo extracts, with a pattern similar to that observed with EC cell extracts. The identity of the complexes was also confirmed with anti-OCT and anti-SOX-2 antibodies and by the differential use of poly(dG-dC) or poly(dI-dC) as nonspecific competitors (Fig.  6). A much weaker binding was observed on the FGF-4 probe, particularly for the SOX-2/OCT3/4 complex (Fig. 6, right panel). These results show that the b1-ARE sequence binds protein complexes containing SOX and OCT proteins present in embryonic tissues at a stage corresponding to the maximal activity of the b1-ARE in the developing hindbrain (6).
SOX-2 and OCT1 Cooperatively Enhance the Activity of the b1-ARE in COS Cells-Taken together, the experiments above suggest that the different activity of the b1-ARE in EC cells compared with COS cells could be mediated by differences in the availability of SOX and OCT proteins. To assess this issue directly, we cotransfected expression plasmids for the SOX-2, OCT1, and OCT3/4 into COS-7 cells together with the b1-ARE reporter in the presence or absence of HOXB1 and PBX1 (Fig.  7). No activation of reporter expression was observed by transfection of any of the individual plasmids in the absence of HOXB1 and PBX. Cotransfection of SOX-2 and OCT1 caused a 4-fold induction of the reporter basal activity bringing it closer to that observed in EC cells. More important, the activity of SOX-2/OCT1 had an additive effect on that of the HOXB1/ PBX1 complex on the b1-ARE and caused a further 2-3-fold increase in its transcriptional activity (Fig. 7). OCT1 alone, but not SOX-2, had some effect on the activity on the HOXB1/PBX1 complex, whereas OCT3/4 had no effect, neither alone nor in combination with SOX-2. These results confirm that the SOX/ OCT site within the b1-ARE significantly contributes to its enhancer activity by recruiting SOX/OCT heterodimers in the appropriate cell context that are able to increase the transcriptional activity of HOX/PBX complexes.
The SOX/OCT-binding Site Is Necessary for Ectopic Activation of b1-ARE by Retinoic Acid in Vivo-In view of our data indicating that the SOX/OCT-binding sequence flanking R3 is important both for the basal activity of the b1-ARE and for the activation by HOX proteins in cell culture, we investigated the role of these sequence using transgenic analysis. Deletions 3Ј of R3 had no effect on the in vivo regulatory activity of the enhancer (data not shown). Two additional mutants were generated in the SOct sites in the context of a highly conserved 331-bp StuI/HindIII fragment containing the b1-ARE (Fig. 8).
In the first case, where we introduced a mutation (TAAT to CCGG) into the more 3Ј OCT-specific hemisite, lacZ reporter expression appeared normal (Fig. 8). At 9.5-10.0 dpc reporter staining was correctly restricted to r4 and second arch neural crest in embryos not exposed to RA and ectopically induced in r2 in all embryos (n ϭ 10) that had been exposed to RA at an earlier time (Fig. 8, A-D). This is identical to the properties of the wild type b1-ARE and shows that mutation of the POU homeodomain-binding sequence is still compatible with full enhancer activity. In contrast, a second variant that alters both the OCT (TAAT to CCGG) and the SOX (TTTGTC to CGCTGT)binding hemisites completely abolished (4/5 embryos), or dramatically reduced (1/5 embryo), the RA-induced ectopic stripe of reporter expression in r2 of embryos exposed to RA (Fig. 8, E-G). The same mutation had no obvious effect (7/7 embryos) on normal r4 expression (Fig. 8, E-G, and data not shown). However, there may be a quantitatively reduced level of activity in this mutated b1-ARE element that is hard to detect, since ␤-galactosidase staining allows only a qualitative estimate of transgene expression. These data show that a complete SOX/ OCT site, and presumably efficient binding of SOX/OCT heterodimer(s), in combination with the HOX/PBC repeats are necessary in vivo for mediating the RA response controlled by the conserved auto-regulatory elements in the b1-ARE.
Hoxa1 Is Required for RA-induced Ectopic Activation of b1-ARE in Vivo-Our results obtained by EC cell transfection indicate that the transactivation of the b1-ARE by a HOXA1/ PBX1 complex is weaker and more dependent on the presence of a SOct site as that of the HOXB1/PBX1 complex. This differential activity, given that the response of the b1 ARE to retinoids in embryos requires the SOct site, might reveal HOXA1 as being the primary component involved in mediating the response of Hoxb1 to RA in vivo. To test this idea, we crossed a transgenic reporter line including the b1-ARE (31) into a Hoxa1 mutant genetic background (52) and assayed its response to RA in mutant and wild type embryos. This Hoxb1/ lacZ transgene does not contain the 3Ј-RAREs responsible for Hoxb1 activation in the ectoderm (28,31) and in the endoderm (53), and therefore any response to RA would be indirect. Untreated wild type and heterozygous Hoxa1 ϩ/Ϫ embryos show a robust r4 lacZ expression (Fig. 9A), whereas homozygous Hoxa1 Ϫ/Ϫ untreated embryos show a significantly reduced expression in r4 corresponding to the reduced r4 territory in these mutants (Fig. 9C). Upon treatment with RA, 6/6 Hoxa1 wild type and 9/10 Hoxa1 ϩ/Ϫ heterozygous embryos responded by ectopically expressing the transgene in the anterior hindbrain (Fig. 9B), whereas 5/5 homozygous Hoxa1 mutant embryos failed to induce ectopically transgene expression (Fig. 9D).
Our results in vivo indicate that the ectopic activation of this element by RA requires both the presence of a functional SOX/ OCT site and the presence of a functional Hoxa1 gene. This implies that for HOXA1 to function efficiently on the b1-ARE, it is important that SOX and OCT proteins are recruited to the SOct site. It is interesting that even though HOXA1 and HOXB1 both participate in regulating normal r4 expression (27,28), we find that, unlike the RA response, transgene expression is maintained in r4 despite mutation of the SOct site. This suggests that in vivo as in the EC cells, HOXB1 is not as dependent as HOXA1 upon SOX and OCT proteins and is able to activate the b1-ARE in r4.
Determinants for the Differential Activity of HOXB1 and HOXA1 Reside within the HOXB1 N-terminal Domain-To examine whether the differential activation of HOXA1 and HOXB1 on the b1-ARE is related to intrinsic structural differences between these proteins, we used a domain swap approach. The HOXA1 and HOXB1 proteins have almost identical homeodomains (54), and they share a conserved FDWM motif just N-terminal to the homeodomain that serves as the HOX/PBX interaction surface, necessary for cooperative DNA binding and transcriptional activation (reviewed in Ref. 3). HOXA1 and HOXB1 bind the b1-ARE R3 site with the same affinity in vitro and have similar transcriptional activity on the full b1-ARE element in non-EC cells and in transgenic mice (5,6,55). However, the two proteins differ completely in their C-terminal and N-terminal regions (56), and for HOXB1 the N-terminal region contains the major transcriptional activation domain used in the context of heterodimers with PBX1 (5,26). Therefore, we designed two chimeric mutants by replacing either the N terminus (aa 1-205) and the C terminus, or only the C terminus (aa 290 -336) of HOXA1, with the corresponding regions of HOXB1 (aa 1-174 and 258 -296) (Fig. 10B), leaving the HOXA1 homeodomain and FDWM regions intact (Fig. 10B).
Expression vectors encoding the HOXA1/B1CT and the HOXB1/A1HD chimeric proteins were cotransfected with PBX1 in P19 cells together with reporters under the control of either the entire b1-ARE (pAdMLARE) or R3 (pAdMLR3). In controls, both chimeras bind the b1-ARE R3 element in vitro with comparable affinity, and both activate transcription from the pAdMLR3 reporter in EC cells and COS-7 cells (Fig. 10A and data not shown). Testing the response using the entire b1-ARE control region, the transcriptional activity of the chimera carrying only the C-terminal replacement remained low and comparable to that of unmodified HOXA1. In contrast, the activity of the chimera containing the combined C-and N-terminal replacement was enhanced and comparable to that of wild type HOXB1 (Fig. 10A).  (6). The embryo in A and B was not exposed to ectopic RA, whereas that in C and D was isolated from a female given RA at 7.75 dpc by oral gavage. This mutation does not impair the ability of the r4 enhancer to direct expression to r4 (A and B) or to mediate a response to ectopic doses of RA as indicated by the induction of a second stripe of expression in r2 (C and D). E-G, lateral (E and F) and dorsal (G) views of reporter expression in 9.0 -10.5-dpc embryos treated with RA and carrying the combined mutations in the SOX-(TTTGTC to CGCTGT) and OCT (TAAT to CCGG)-specific hemisites. In both untreated and RA-treated cases r4 expression is not affected (E-G and data not shown). However in the majority of cases (4/5) the double mutant construct fails to mediate a response to ectopic RA (E), and in the single example of an RA response the induction in r2 is very patchy and incomplete (F and G), compared with wild type or the single TAAT mutation (C and D). A schematic diagram of the transgenic constructs and mutations is indicated below the panels. The three bipartite HOX/PBX repeats are indicated as R1-R3.
These results show that specific DNA recognition and cooperative binding with PBX1, provided by the region encompassing the homeo-and FDWM domains, are not responsible for the differential activity of HOXA1 and HOXB1 on the b1-ARE. Conversely, the HOXB1 N-terminal region, comprising the transcriptional activation domain, appears to contain specific determinants that are able to respond to cues from the cell background leading to a higher activity with respect to HOXA1. DISCUSSION Mammalian HOX proteins are a large family of transcription factors that control cell identity, differentiation, and patterning in animal embryonic development. Due to the similarities in their structure, in vitro binding abilities, and in vivo function, it is important to understand the factors and components that serve to regulate the selectivity and specificity of HOX transcription complexes during activation of downstream target genes. In this study we have addressed some of these issues by examining the regulatory properties and functional requirements of an in vivo Hox-responsive target sequence represented by an auto-regulatory enhancer (b1-ARE) from the Hoxb1 gene. We have used human or murine EC cells, which retain most of the characteristics of primitive neuroectodermal cells (32,33), as a model system to test the overall transcriptional activity of the b1-ARE. We found that the b1-ARE has both higher basal and higher HOX/PBX-induced activities in neuro-ectodermal EC cells as compared with other cell backgrounds. The enhanced EC activity correlates with the presence of a bipartite motif in the b1-ARE that binds SOX and OCT proteins and is necessary for the optimal response of b1-ARE to transcriptional activation by HOX/PBX heterodimers both in vitro and in vivo. Surprisingly, despite recent evidence suggesting that Hox paralogs are functionally equivalent (57), our analysis revealed that there are differences in the way that the b1-ARE responds to HOXB1 and HOXA1, which correlate with differences in their N-terminal domains. Together, our results have uncovered some novel aspects of how HOX proteins interact with their in vivo target sequences, identified cis-elements that can modulate HOX/PBX complex activities, and raised a number of important issues on the functional differences between paralogous HOX proteins.
Cis-determinants in the b1-ARE-Our analysis has revealed that a highly conserved sequence motif located immediately upstream of a HOX/PBX sequence (R3) is an important determinant of the cell type-specific restriction of the b1-ARE activity. This bipartite element mediates high affinity binding of a SOX/OCT heterodimer that contributes to enhancer activity in addition to, and in combination with, HOXA1/PBX1 and  (6). A and B, Hoxa1 wild type background. The embryo in A was not exposed to ectopic RA, whereas that in B was isolated from a female given RA at 7.75 dpc by oral gavage. The Hoxb1-ARE lacZ reporter is ectopically activated by RA treatment and displays a stripe of expression in r2 in 6/6 embryos (B). C and D, Hoxa1Ϫ/Ϫ background. The Hoxb1/lacZ transgene fails to be expressed ectopically in r2 in 5/5 embryos isolated from females treated with RA at 7.75 dpc by oral gavage (D). In both untreated (C) and RA treated (D) cases r4 expression is not affected. ov, otic vesicle. HOXB1/PBX1 heterodimers. This motif (SOct) allows cooperative binding of SOX-2/OCT1 and SOX-2/OCT3/4 heterodimers present in extracts of murine or human EC cells and mouse embryos. The binding of the SOX-2/OCT1 complex to the b1-ARE site actually appears to be stronger than that observed with the SOX/OCT-binding sequence located in the developmentally regulated FGF-4 enhancer (50,51). This difference could be explained by the different spacing between the SOXand OCT-specific hemisites on the two bipartite sequences. In the b1-ARE sequence, the SOX and Octamer sites actually overlap by 1 base pair, whereas in the FGF-4 enhancer the two sites are spaced apart by 2 base pairs. Since it was previously shown that the distance between the two sites is critical for the assembly of the SOX/OCT heterodimer in the context of the FGF-4 enhancer (46), it seems likely that the configuration of the b1-ARE site allows a different interaction between SOX-2 and OCT1 or OCT3/4, leading to a stronger binding of the SOX-2/OCT complexes. The slightly different mobility of the SOX-2/OCT3/4 complex on the b1-ARE versus the FGF-4 sequence (see Fig. 5) might be caused by a differential bending of the DNA induced by SOX-2 that might also affect binding strength. High affinity binding of a SOX-2/OCT3/4 complex was recently reported on a closely spaced bipartite sequence found in the upstream regulatory element of the murine UTF1 transcription factor (58).
The Hoxb1 SOX/OCT (SOct) site contains at its 5Ј end a potential binding site (TGACAA) for the TALE homeodomain proteins MEIS and PREP. This is adjacent to R2 and separated from R3 by 17 nucleotides. A similar combination of a TALE site and a HOX/PBX site is also found in an r4-restricted cross-regulatory enhancer from the Hoxb2 gene, except that the sites are separated by 8 base pairs (11,15). In the case of Hoxb2 there is no overlapping SOX/OCT site, but the TALE and HOX/PBX sites synergize in vitro and in vivo to allow the formation of a trimeric and transcriptionally active HOXB1-PBX1-MEIS/PREP complex required for enhancer activity in transgenic assays (10,11). In the context of the b1-ARE, however, MEIS1 or PREP1 is not necessary for DNA binding of HOX/PBX heterodimers and, accordingly, is not required to bind DNA to enhance the transcriptional activity of the HOXB1/PBX1 complex in transfected cells (8). Indeed, in EMSAs none of the retarded complexes formed on the complete SOct sequence by either EC cell or mouse embryo extracts appeared to contain MEIS or PREP proteins (data not shown).
The Role of OCT and SOX Proteins in b1-ARE Enhancer Activity in EC Cells-Transient transfection analysis showed that the SOct site is responsible for the high basal transcriptional activity of the b1-ARE sequence in EC cells and is also required for fully activated transcription in the presence of HOXB1 and PBX1. Interestingly, transfection of SOX-2 and OCT1 in non-EC cells (COS-7), which do not express endogenous SOX proteins, significantly increases both the basal activity of the b1-ARE reporter and the HOXB1/PBX1-mediated transactivation. The nature of the OCT partner seems to be important, as transfection of SOX-2 and OCT3/4 had no effect on the b1-ARE. Conversely, a SOX-2/OCT3/4 complex but not a SOX-2/OCT1 complex was reported to have a positive effect on the transcriptional activity of the FGF-4 (51), osteopontin (47), and UTF1 (58) enhancers in HeLa cells. These results indicate that the b1-ARE SOX/OCT site might have unique binding properties in vivo compared with previously reported bipartite sites and facilitates functional interaction specifically for the SOX-2/OCT1 complex. Therefore, both sequence and spacing between the two hemisites appear to be important general factors in dictating specific binding of heterodimeric complexes between SOX-2 and alternative OCT proteins to this class of target elements.
In Vivo Role for the SOX/OCT Site in b1-ARE Regulation-These experiments suggest a model whereby the bipartite SOX/ OCT site contributes to the activity of the b1-ARE enhancer in vivo by recruiting SOX-2/OCT1 heterodimers. This in turn could either increase the stability or the affinity of binding by HOXB1-PBX1 and HOXA1/PBX1 complexes and/or activate transcription in synergy with such complexes. In support of this, combined mutations in both parts of the SOX/OCT site show that it is necessary for full in vivo activity of the b1-ARE in transgenic mouse embryos.
Genetic and regulatory analyses have previously shown that Hoxa1 and Hoxb1 function synergistically to regulate the early r4 expression of Hoxb1 (7.5-8.5 dpc) and that this segmental expression is maintained from 8.5 dpc and onwards by Hoxb1 itself (6,27,28,59). Mutation of both the SOX-and OCTspecific hemisites in the b1-ARE does not abolish reporter expression in r4 at 9.5 dpc (Fig. 8). This expression in later stages implies that endogenous HOXB1/PBX complexes are still able to activate the mutant b1-ARE reporter. This is consistent with our findings in EC cells that showed HOXB1/PBX complexes could stimulate transcription in the absence of the SOct site, although not at maximal levels, whereas the HOXA1/ PBX complexes displayed an absolute requirement for this site in transactivation (Fig. 4). In contrast, mutation of the SOct site impairs the ability of the b1-ARE reporter to generate a stripe of ectopic expression in r2 in response to in utero RA treatment (Fig. 8). Furthermore, this ectopic RA response is also abolished in Hoxa1 mutant embryos (Fig. 9). These results suggest that, as in EC cells, the interaction of HOXA1/PBX complexes with factors binding to the SOct site is essential for generating complexes with sufficient activity to trigger the b1-ARE auto-regulatory loop in r2. This underscores the importance of the SOct site for in vivo activity and differential dependence of HOXA1/PBX versus HOXB1/PBX complexes upon this site for modulating transcriptional activity. It is interesting that the ectopic response of the transgene to RA depends upon HOXA1 and that HOXB1 does not compensate. We have found that the reason for this is that while both the Hoxa1 and Hoxb1 genes have 3Ј-RAREs necessary for early neural expression, the Hoxb1 3Ј-RARE does not respond to in utero RA treatment (data not shown).
A mutation in the OCT-specific hemisite, which is recognized by the POU homeodomain of the Octamer factor, has little or no effect in transgenic mice. This suggests that the SOX/OCT complex can assemble in vivo on a suboptimal target sequence, through interaction of the SOX factor with the POU-specific domain of the Octamer factor. Similar data on the relative importance of the POU-specific sequence for binding of OCT1 and OCT3/4 to a bipartite SOX/OCT site in vitro was also observed in the case of the UTF1 upstream regulatory element (58).
Selectivity and HOX Protein Determinants-The homeodomains of HOXB1 and HOXA1 are almost identical, and in combination with PBX proteins display a virtually indistinguishable binding specificity. Therefore, the differential activity of HOXB1 or HOXA1 complexes with PBX most likely resides in specific interactions with other DNA-binding factors and/or with the transcriptional machinery. Indeed, our analysis demonstrates that a critical determinant of the difference between the HOXB1/PBX1 complex and that of HOXA1/PBX1 resides in the HOXB1 N-terminal transcriptional activation domain. A chimeric protein, where the N-terminal domain of HOXA1 is replaced with that of HOXB1, makes it indistinguishable from wild type HOXB1 in activating the b1-ARE in cooperation with PBX1. This suggests that DNA recognition is not the key variable involved in modulating the activity of group-1 HOX proteins on the b1-ARE target and that factors interacting with the N terminus of these HOX proteins are responsible for mediating their differential activities in a tissue-or cell-specific manner.
Although we have focused on how distinct HOX proteins in the heterodimers differentially interact with cofactors, diversity in PBX partners may also contribute to the selectivity. The existence of both the PBX and other novel HOX cofactors might explain the observed tissue-specific restriction of some Hoxresponsive enhancers in vivo. A recent report (57) has challenged the view that paralogous HOX proteins have gained intrinsic functional diversity in the course of evolution. They propose that paralogous proteins are functionally equivalent, and it is only the relative levels or domains of their expression that govern their unique activities. Our results, conversely, show the existence of intrinsic differences in function between the HOXA1 and HOXB1 proteins. In this regard, the difference between HOXB1 and HOXA1 in their ability to interact functionally with the same Hoxb1 auto-regulatory element in a specific cell context is a clear example of non-redundancy between products of Hox genes belonging to the same paralogous group. We feel that this observation may also extend to other paralogous groups, and it will be important to characterize the nature of the proteins interacting with the N-terminal domains of the HOX proteins.