Regulatory Interactions between the Human HOXB1, HOXB2, and HOXB3 Proteins and the Upstream Sequence of the Otx2 Gene in Embryonal Carcinoma Cells*

Vertebrate Hox and Otxgenes encode homeodomain-containing transcription factors thought to transduce positional information along the body axis in the segmental portion of the trunk and in the rostral brain, respectively. Moreover,Hox and Otx2 genes show a complementary spatial regulation during embryogenesis. In this report, we show that a 1821-base pair (bp) upstream DNA fragment of the Otx2 gene is positively regulated by co-transfection with expression vectors for the human HOXB1, HOXB2, and HOXB3 proteins in an embryonal carcinoma cell line (NT2/D1) and that a shorter fragment of only 534 bp is able to drive this regulation. We also identified the HOXB1, HOXB2, and HOXB3 DNA-binding region on the 534-bp Otx2 genomic fragment using nuclear extracts from Hox-transfected COS cells and 12.5 days postcoitum mouse embryos or HOXB3 homeodomain-containing bacterial extracts. HOXB1, HOXB3, and nuclear extracts from 12.5 days postcoitum mouse embryos bind to a sequence containing two palindromic TAATTA sites, which bear four copies of the ATTA core sequence, a common feature of most HOM-C/HOX binding sites. HOXB2 protected an adjacent site containing a direct repeat of an ACTT sequence, quite divergent from the ATTA consensus. The region bound by the three homeoproteins is strikingly conserved through evolution and necessary (at least for HOXB1 and HOXB3) to mediate the up-regulation of the Otx2 transcription. Taken together, our data support the hypothesis that anteriorly expressed Hox genes might play a role in the refinement of the Otx2 early expression boundaries in vivo.

Hox gene transcription factors are key players in the establishment of the anteroposterior positional values along the body axis, as shown by the analysis of their mutant phenotypes in animals as evolutionary diverged as nematodes, insects, and mammals (1). Hox genes are expressed in segmental structures along the anteroposterior axis in developing vertebrate embryos according to spatially and temporally restricted patterns, which correlate with their physical arrangement in the cluster, following the 3Ј-anterior/early 5Ј-posterior/late colinearity rule (1)(2)(3)(4)(5). The homeodomain is the DNA-binding motif necessary for the biological function of HOX proteins. However, despite markedly divergent biological effects occurring in vivo, the homeodomains of many different HOX proteins share a very high sequence similarity and can bind in vitro with similar affinities to many DNA sites containing an ATTA core sequence (6 -8).
Hox genes are expressed in many parts of the vertebrate central nervous system but are not expressed in the rostral brain, i.e. in forebrain and in midbrain. The early organization of the developing rostral brain in higher vertebrates and the existence of a segmental patterning in this region is still the object of intense debate. The expression pattern of putative regulatory genes has revealed that the rostral brain can be divided into discrete longitudinal and transverse domains (9). One of the genes showing such a regionally restricted expression in the forebrain is Otx2, a homeobox-containing gene isolated in mouse and humans as one of the two counterparts of the Drosophila orthodenticle gene, which is involved in the control of the head development in flies (10 -12). Orthologues of Otx2 have subsequently been cloned in several other species (13)(14)(15)(16). Recently, three groups have shown that targeted Otx2 inactivation in mice leads to embryonal lethality, with embryos showing severe gastrulation defects and deletions of rostral brain (17)(18)(19).
Otx2 mRNA and protein are present at 5.5-5.7 days postcoitum (d.p.c.) 1 in the embryonic portion of the primary ectoderm (or epiblast). Subsequently, between 7.0 and 7.5 d.p.c., Otx2 expression becomes progressively restricted to the anterior portion of the embryo, mainly the neuroectoderm of the headfold, fated to give rise to fore-and midbrain (11,20). As Otx2 expression turns off in the posterior regions, Hox genes are progressively activated, in such a way that Otx2 is eventually restricted to the most anterior third of the embryo, and Hox genes are restricted to the most posterior two-thirds (21,22).
The complex spatial and temporal regulation of the Otx2 and Hox genes can be mimicked experimentally both in vivo and in vitro. In fact, during early mouse development, high doses of retinoic acid (RA) (a molecule involved in embryonic pattern formation and specification of regional identities in central nervous system, axial skeleton, and limbs) severely repress Otx2 expression, leading to a reduction in the extent of foreand midbrain at later stages and, conversely, to an expansion of the hindbrain together with a strong induction of the expression of Hoxb-1 and Hoxb-2 anteriorly (23,24). In a human embryonal carcinoma cell line (NT2/D1), RA treatment abolishes constitutive Otx2 expression and strongly activates Hox gene expression (11,25,26). In addition, in NT2/D1 cells the activity of a 5Ј genomic fragment of the mouse Otx2 gene driving a reporter gene showed a 6 -9-fold decrease after RA treatment (23), while the activity of Hox gene regulatory sequences is strongly enhanced (27)(28)(29). Taken together, these data suggest that the complementary spatial and temporal regulation of the Otx2 and Hox genes might be a consequence of the RA gradient present in the embryo and raise the possibility that Otx2 and Hox gene expression might mutually exclude each other by means of regulatory interactions.
The restriction of the Otx2 expression to the anterior neuroectoderm has been shown to require both positive and negative signals coming from the underlying mesendoderm (21). In this report, we explore the possibility that anteriorly expressed Hox genes play a role in the early regulation of Otx2 expression through activation or repression of the transcriptional activity of the 5Ј Otx2 genomic sequences. We show that a 1821-bp 5Ј genomic fragment of the mouse Otx2 gene, whose activity has already been shown to be down-regulated by RA (23), can also directly interact with Hox genes. The transcriptional activity of the 5Ј-flanking Otx2 genomic sequences driving a reporter gene in NT2/D1 cells was increased by co-transfection with HOXB1, HOXB2, and HOXB3-encoding expression vectors, suggesting that Hox genes are not involved in the down-regulation of Otx2 expression in the posterior part of the embryo, but rather they might be involved in the refinement of Otx2 expression boundaries in the anterior compartment. Hox genes expressed more posteriorly during later stages of development were unable to transactivate Otx2 expression. The binding sites of the three HOX proteins have been identified and shown to be necessary for the transactivation in transfected cells by HOXB1 and HOXB3 but not by HOXB2. The same sites were also recognized by nuclear extracts obtained from 12.5-day-old mouse embryos. Our data suggest that HOXB1, HOXB2, and HOXB3 might directly activate the Otx2 promoter and, possibly, play a role in the early regulation of the Otx2 gene expression.

EXPERIMENTAL PROCEDURES
Reporter Plasmid and cDNA Expression Vectors-To obtain the pOTX2Luc⌬Ϫ1219, the pOTX2Luc⌬Ϫ710, and pOTX2Luc⌬ϩ163 constructs, a 1821-bp EcoRI-StyI, a 1310-bp ClaI-StyI, and a 439-bp HaeIII-StyI DNA fragment, respectively, of the mouse OTX2 genomic upstream sequence, extending to position ϩ600 in their 3Ј-ends, was cloned into the SmaI site of the pXP2Luc reporter plasmid (30). The pOTX2LucATTAbox-TK construct was obtained cloning a HinfI-HaeIII 534-bp fragment into the SmaI site of the pT81Luc reporter plasmid. The HBE1 binding site was mutagenized by polymerase chain reaction with the mismatched primers HBE1m (5Ј-ATCCAGACTACTGCAGAG-GTGAAAATGAT-3Ј), containing a PstI site in the mutated sequence, to obtain the pOTX2LucATTAbox-TKm construct.
The pSG-HOXB1 cDNA construct was reconstructed by a 5Ј genomic 370-bp TaqI-PstI sequence and a 700-bp PstI-StuI partial cDNA fragment into the pBluescript-KS Ϫ vector and then cloned into the EcoRI (filled in) site of the pSG5 mammalian expression vector (31). The pSG-HOXB2 construct was obtained by an EcoRI partial digestion of a 1700-bp genomic fragment cloned into the EcoRI site of the pSG5 vector. We have already described the pSG-HOXB3, pCT-HOXC6, and pSG-HOXD3 cDNA expression vectors (27,32).
Cell Culture and Transfection Assays-NT2/D1 cells were cultured in high glucose Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum (all from Life Technologies, Inc.) in 5% CO 2 in air-humidified atmosphere. NT2/C1 cells were transfected by calcium phosphate co-precipitation as described previously (32) with 5-8 g of expression vector, 10 -12 g of reporter plasmid, and 1-2 g of either pRSV-␤-gal or pPGK-␤-gal as internal control. Luciferase and ␤-galactosidase assays were performed as described previously (32). All transfections were carried out in duplicate batches, using at least two different DNA preparations of each expression construct in at least three separate experiments.
Preparation of Nuclear Extracts, DNase I Footprinting, and Gel Retardation Assays-Crude nuclear extracts from COS-7 transfected cells were prepared 48 h after transfection as described by Dignam et al. (33).
Embryonic nuclear extracts were obtained from a pool of 12.5 d.p.c. mouse embryos from which the posterior (abdominal) part of the body had been surgically removed. Then the remaining portion of the body was separated in two: the head (12.5 d.p.c. brain nuclear extracts) and the trunk (12.5 d.p.c. trunk nuclear extracts). Crude nuclear extracts were then prepared from embryonic tissues as described previously (32).
In order to perform DNase I footprinting experiments, the same HinfI-HaeIII 534-bp fragment used for the construction of pOTX2-LucATTAbox-TK plasmid was cloned into a HincII-SmaI-digested pGEM3 vector, and the resulting construct was digested with SspI, end-labeled with a standard T4 polynucleotide kinase reaction, and cut at a SacI site present in the polylinker. The resulting 455-bp DNA fragment was purified by polyacrylamide gel electrophoresis and then used for the DNase I footprinting assays as described previously (34), in the presence of 2-8 l of COS nuclear extracts or 0.5-2 l of HOXB3 homeodomain (and immediately flanking residues, 4 amino acids at the N terminus and 10 amino acids at the 3Ј terminus) produced in E. coli by a T7 promoter-based expression system, prepared as already described (32). 32 P-Labeled 35-mer double-stranded oligonucleotide HBE1 (5Ј-TATCCAGACTACTAATTAGGTGAAAATGATTACTG-3Ј) was used as probe in gel retardation assays as described previously (32) with 2-4 l of nuclear extracts. A 500-fold molar excess of the double-stranded HBE1 or HBE1m (5Ј-ATCCAGACTACTGCAGAGGTGAAAATGAT-3Ј) oligonucleotides was used for the competition assays. A polyclonal antiserum raised against the bacterially produced HOXB3 homeodomain (32) was also used in gel retardation assays.
In order to assess if these 5Ј-flanking sequences are able to mediate a regulation by HOX proteins, we co-transfected in NT2/D1 cells the pOTX2Luc deletion constructs together with expression vectors encoding HOXB1, HOXB2, and HOXB3. These Hox genes are known to be already expressed at 7.5 d.p.c., when the concomitant transition in Otx2 and Hox gene expression takes place (4). Full-length cDNAs coding for human HOXB1 and HOXB2 (35) were cloned into the SV40 FIG. 1. Reporter and expression constructs used in co-transfection assays. A, the reporter constructs pOTX2Luc⌬Ϫ1219, pOTX2Luc⌬Ϫ710, and pOTX2Luc⌬ϩ163 were generated by cloning three overlapping upstream genomic sequences of the mouse Otx2 gene, extending from position Ϫ1219, Ϫ710, and ϩ163, respectively, to position ϩ602 from the transcription start site, upstream of the luciferase (LUC) coding region in the pXP2Luc vector. The reporter construct pOTX2LucATTAbox-TK was constructed by cloning in front of the minimal TK promoter (black box) in the pT81Luc vector the Otx2 genomic sequence from position Ϫ371 to ϩ163. Panel B, full-length human or mouse HOX cDNA sequences (HOX cDNAs) were cloned in the mammalian expression vector pSG5, under the control of the SV40 early promoter (pSG-HOX). The arrows indicate transcription start sites in all constructs. early promoter-based mammalian expression vector pSG5 to generate the pSG-HOXB1 and pSG-HOXB2 constructs (Fig.  1B). The pSG-HOXB3 plasmid has been previously described (32). In NT2/D1 cells, the reporter construct pOTX2Luc⌬Ϫ1219 showed a 12-, 25-, and 9-fold increase in transcriptional activity when co-transfected with pSG-HOXB1, pSG-HOXB2, and pSG-HOXB3, respectively. Likewise, the construct pOTX2Luc-⌬Ϫ710 showed an 8-, 12-, and 6-fold increase in transcriptional activity if co-transfected with pSG-HOXB1, -HOXB2, and -HOXB3, respectively, while a higher level of basal transcriptional activity was also noted ( Fig. 2A). The pOTX2Luc⌬ϩ163 construct, used as negative control, failed to show any increase in transcriptional activity in co-transfection assays ( Fig. 2A). The extent of the observed transactivation of the reporter constructs was directly dependent on the concentration of the HOX proteins used in the assay (data not shown).
A close inspection of the DNA sequence surrounding the transcription start site revealed the presence of several copies of the ATTA sequence ( Fig. 4), which is contained in most of the HOX/HOM-C target sites (8). We then decided to clone the HinfI-HaeIII 534-bp ATTA-rich sequence from position Ϫ371 to ϩ163 in the pT81Luc expression vector (30) to generate the pOTX2LucATTAbox-TK construct, which contains also the thymidine kinase (TK) minimal promoter in front of the luciferase reporter gene in order to increase its basal activity (Fig. 1A). Co-transfection of the pOTX2LucATTAbox-TK with the pSG-HOXB1, -HOXB2, and -HOXB3 showed a 15-, 17.5-, and 7-fold increase in its basal transcriptional activity, respectively (Fig. 2B).
To assess the specificity of the transactivation on the Otx2 promoter of the anteriorly expressed Hox genes, expression vectors for other Hox cDNAs (namely, HOXC6 and Hoxd-8) were co-transfected in NT2/D1 cells together with pOTX2-LucATTAbox-TK, but they did not cause any increase in the Otx2 promoter basal activity. We have already reported that an efficient production of the proteins showing no effect on the Otx2 promoter was achieved with the constructs used in the present study (32). However, co-transfection with an expression vector encoding for HOXD3, which belongs (together with HOXB3) to paralogy group 3, caused an increase in the activity of the reporter construct similar to that caused by HOXB3 (Fig.  2B). These results indicate that the ability to transactivate the Otx2 promoter is not a feature shared by all members of the Hox gene family but is apparently common only to the anteriorly expressed Hox genes, namely to those belonging to the paralogy groups 1, 2, and 3. Similar transfection experiments were also conducted in HeLa and NIH3T3 cells, but no effect was observed on the Otx2 promoter, suggesting that the cellular environment plays a crucial role in the transactivating activity of HOXB1, -B2, and -B3 (data not shown).
HOXB1, HOXB2, and HOXB3 Bind to the Otx2 Promoter, but HOXB2 Differs in DNA Binding Specificity-The binding of the HOXB1, HOXB2, and HOXB3 proteins to the Otx2 promoter sequence was tested by DNase I footprinting assay. Nuclear extracts were prepared from COS-7 cells transfected with the pSG-HOXB1, pSG-HOXB2, and pSG-HOXB3 constructs and tested on the Otx2 455-bp SspI-HaeIII DNA fragment containing most of the pOTX2ATTAbox-TK construct (Fig. 4). The only protection signal clearly detected with HOXB1 and HOXB3 on these two probes was from position Ϫ177 and Ϫ186, respectively, to position Ϫ133 from the major transcription start site (Fig. 3A). A weaker and shorter protection signal centered near position Ϫ180 was present with HOXB2-containing nuclear extracts. The protection was also confirmed on the other DNA strand (data not shown). No other clear protections were observed at the other ATTA sites present in the DNA fragment, indicating that (at least in the DNase I footprinting assay) a certain degree of DNA binding specificity was achieved by HOXB1, HOXB2, and HOXB3-containing nuclear extracts, although homeoproteins are known to bind to virtually any DNA containing an ATTA core sequence in vitro (8), suggesting that other determinants (i.e. PBX-or PBX-like factors) present in the COS nuclear extracts were able to raise the in vitro DNA binding specificity of HOX genes (41). Finally, an additional weaker protection containing an ATTA core sequence and a hypersensitive site was found near position Ϫ213 with HOXB1 only (Figs. 3A and 4). DNase I footprinting experiments performed on the remaining 5Ј part of the pOTX2LucATTAbox-TK construct did not reveal any other protection (data not shown).
To confirm the presence of the protected DNA sequence on the Otx2 genomic upstream sequence, we performed a DNase I footprinting assay with a bacterially produced HOXB3 homeodomain and immediately flanking residues (4 amino acids at the N terminus and 10 amino acids at the 3Ј terminus) on the 455-bp SspI-HaeIII DNA fragment; a larger and strong protection centered around the same site was detected from position Ϫ176 to position Ϫ92 (Figs. 3B and 4). Since it has been reported that the protection signals found with homeodomains are shorter than that obtained with entire proteins (32), we think that the longer protection observed in our assay is due to the simultaneous binding of several homeodomain molecules and that co-factors present in COS nuclear extracts might  HOXB1, HOXB2, and HOXB3 proteins. Panel A, 10 g of the pOTX2-Luc⌬Ϫ1219, pOTX2Luc⌬Ϫ710, or pOTX-2Luc⌬ϩ163 reporter constructs were cotransfected in NT2/D1 cells together with 5 g of expression vectors encoding HOXB1, HOXB2, or HOXB3. The open bars indicate the basal activity of the three reporter constructs, while data from co-transfection assays with HOX constructs are represented by filled bars. Luciferase activity is expressed in -fold induction, and the basal transcriptional activity is arbitrarily considered equal to 1. pOTX2Luc⌬Ϫ710 showed a 4-fold higher level of basal transcriptional activity than pOTX2Luc⌬Ϫ1219. Panel B, the pOTX2LucATTAbox-TK reporter construct was co-transfected as in panel A, but additional more posteriorly expressed HOX-encoding constructs were used in the assay. Data are expressed as in panel B.
inhibit HOX proteins from binding to flanking DNA sequences. In fact, several DNA-hypersensitive sites were also noticed, which presumably mark the borders of the DNA-bound HOXB3-homeodomain (Fig. 3B). The HOXB3 homeodomaincontaining bacterial extracts bind also to other sites containing an ATTA core sequences, again showing that bacterial extracts display a less stringent DNA binding specificity than COS nuclear extracts (data not shown). The sequence of the entire AT-rich region is reported in Fig. 4; the protected region is boxed and the presence of unprotected ATTA sites is highlighted in boldface type. The region protected by HOXB1 and HOXB3 shared a core sequence containing two palindromic TAATTA binding sites, whereas the site protected by HOXB2 contained a direct repeat of an ACTT core sequence, quite divergent from the ATTA core consensus sequence (Fig. 4) but almost identical to the sequence (GCTTACTT versus ACTTACTT) found protected by HOXB2 on the A ␥-globin enhancer (36).
On the basis of footprinting information, we designed a 35mer double-stranded oligonucleotide (HBE1, HOX-binding element 1; underlined in Fig. 4) representing the sequence most strongly protected by the nuclear extracts containing HOXB1 and HOXB3 but not HOXB2. As expected, nuclear extracts containing HOXB1 formed a retarded DNA-protein complex with the HBE1 double-stranded oligonucleotide (Fig. 5A, lane  4, which was specifically competed by a 500-fold molar excess of unlabeled double-stranded oligonucleotide HBE1 but not by a comparable excess of unlabeled HBE1m double-stranded oligonucleotide, in which four bases inside the palindromic TAATTA site have been mutagenized (Fig. 5A, lanes 5 and 6, and Fig. 6). Control nuclear extracts from mock-transfected COS cells (Fig.  5A, lanes 1-3) and HOXB2-containing nuclear extracts failed to show any specifically retarded complex (Fig. 5A, lanes 7-9). Similarly, HOXB3-containing COS nuclear extracts formed multiple DNA-protein complexes, and at least three of them were specifically competed by a 500-fold molar excess of HBE1 but not by HBE1m (Fig. 5B, lanes 5-7), while the very faint bands detected with the control extracts were not specific (Fig.  5B, lanes 1-3). In addition, an antiserum against the HOXB3 homeodomain, which had already been shown to interfere with the DNA binding capability of HOXB3 (32), was also able to abolish the DNA binding activity of HOXB3-containing extracts on the HBE1 site (Fig. 5B, lanes 5 and 8), while the preimmune serum did not show any effect (data not shown).  8 -9) and mock-transfected (lanes 10 -11) COS nuclear extracts. Lane 1, no protein added to the reaction (Ϫ). Lanes 4 and 7, DNase I digestions were carried out at a higher DNase I concentration. On the left, bars represent the protected sequences; their position from the major transcription start site and the proteins responsible for the protection are also indicated. A hypersensitive site found only with the HOXB1-containing extracts is also shown on the right. Panel B, the same DNA fragment was incubated with 0.5, 1, and 2 l of control (lanes 3-5) or HOXB3 homeodomain-containing (lanes 6 -8) crude bacterial extracts. Lane 1, G ϩ A. Lanes 2 and 9, no protein added to the reaction (Ϫ). A bar on the left represents the protected sequence, and the region not protected by the COS HOXB3-containing extracts is indicated by a dashed line. Four DNase I-hypersensitive sites (black arrows) are indicated on the right.
The addition to the binding reaction of the ␣-HOXB3 serum markedly improved the formation of unspecific DNA-protein complexes present in the control extracts at very low amounts (Fig. 5B, lanes 4 and 8).
Interestingly, sequence comparison of Otx2 5Ј-flanking regions in mouse and Xenopus showed stretches of high similarity in the 5Ј upstream region from position Ϫ371 to ϩ163 and nearby sequences, 2 with 75% identity in the footprinted sequence (Fig. 7). This level of identity is quite remarkable for noncoding sequences from very divergent species, suggesting that the function mediated by this region might be conserved through evolution too and, thereby, be relevant for Otx2 gene regulation.
HOXB1, HOXB2, HOXB3, and Nuclear Factors from Mouse Embryos Bind to the Same Region in the Otx2 Upstream Regulatory Sequence-To assess whether HOXB1, HOXB2, and HOXB3 expressed in embryonic tissues were able to recognize the same sequences bound by the nuclear extracts prepared from COS HOXB1-, HOXB2-, and HOXB3-transfected cells, nuclear extracts from 12.5 d.p.c. mouse embryos were prepared and used in a footprinting assay. Both OTX2 and the three above mentioned HOX proteins are expressed in the mouse at this developmental stage (11,21,37). The embryonic nuclear extracts from regions expressing paralogy group 1-3 HOX proteins were prepared by surgical removal of anterior limbs and all of the posterior (i.e. post-thoracic) body. Subsequently, separation of the head from the rest of the trunk was surgically achieved, and two separate preparations of nuclear extracts were then performed. The resulting extracts were allowed to react with the 455-bp Otx2 5Ј genomic fragment, where they protected from DNase I digestion the same region recognized by HOXB1, HOXB2, and HOXB3 (Fig. 8). No differences were observed between the DNA binding properties of the brain and trunk extracts, suggesting that only anteriorly expressed Hox genes were responsible for the binding activity found on this region (Fig. 8) 4. Sequence of the upstream genomic region of the mouse Otx2 gene from position ؊371 to ؉163 from the major transcription start site. The entire Otx2 5Ј genomic sequence cloned in the pOTX2LucATTAbox-TK construct is here reported. The region bound by HOXB1, HOXB2, and HOXB3, as determined by DNase I footprinting (Fig. 3), is boxed, and the HOXB2 binding site is stippled. The underlined sequence is the HBE1 oligonucleotide used in gel retardation assays (see below). The more upstream binding site found protected with HOXB1 only is also boxed. The several unprotected ATTA consensus sites are highlighted in boldface type. The SspI restriction site used for the footprinting assay is indicated above the sequence. An arrow indicates the major transcription start site. Gray shading indicates the sequence protected by the HOXB3 homeodomain. (lanes 1-3), HOXB1-(lanes 4 -6), and HOXB2-transfected COS cells (lanes 7-9) were used in a gel retardation assay with the 32 P-labeled HBE1 double-stranded oligonucleotide. A 500-fold molar excess of unlabeled HBE1 (lanes 2, 5, and 8) or HBE1m (lanes 3, 6, and 9) doublestranded oligonucleotides was also used. The specifically retarded HBE1-protein complex is indicated on the left. Lane 10, no proteins were added in the reaction. Panel B, the same assay as in panel A was performed with nuclear extracts from mock- (lanes 1-4) and HOXB3transfected COS cells (lanes 5-8). 1 l of a polyclonal anti-HOXB3 homeodomain antibody was added to the reaction (lanes 4 -8). The specifically retarded HBE1-protein complexes (comp.) are indicated on the right. C, no protein added to the reaction. F-, unbound probe. A 3-day exposure and an overnight exposure were used for the autoradiographs shown in panels A and B, respectively. struct (bearing the same mutation present in the HBE1m oligonucleotide used for the competition experiments; see above), which was then tested in a co-transfection assay together with pSG-HOXB1, -HOXB2, and -HOXB3 in NT2/D1 cells. This mutation caused an almost complete abrogation of the transactivation exerted by the HOXB1 and HOXB3 expression vectors. However, the extent of the transactivation on the Otx2 promoter by HOXB2 was not affected by the mutation at the HBE1 site (Fig. 6). These data show that the HOXB1 and HOXB3 activity on the Otx2 promoter is directly mediated by the HBE1 site. In contrast, the transactivation of the Otx2 promoter by HOXB2 is not mediated by the HBE1 site, in accordance with the lack of binding to the HBE1 element shown by HOXB2 in vitro (Figs. 3A and 5A). DISCUSSION A huge amount of data during the last decade, including the analysis of mouse knockout mutants, has demonstrated that positional information along the anteroposterior axis of the vertebrate trunk is transduced by Hox genes, which encode a large family of proteins evolutionary related to Drosophila HOM-C gene products (1, 4). Recently, some of the molecules involved in the early organization of the vertebrate rostral brain have also been identified, and it has been suggested that they might play a role in the early patterning of the developing head in longitudinal and transverse domains (9). However, during development Hox genes are not expressed in the vertebrate rostral brain, whereas forebrain-specific genes are not expressed in the segmental portion of the trunk. Otx2 is present at 5.5-5.7 d.p.c. in the whole embryonic portion of the epiblast, while at later stages of development (7.0 -7.5 d.p.c.) it becomes progressively restricted to the anterior portion of the embryo, mainly the neuroectoderm of the headfold. In contrast, the expression of Hox genes begins to appear in the posterior part of the epiblast and progressively moves to more anterior regions. Thus, Otx2 and Hox genes show a complementary spatial regulation during embryogenesis. Moreover, Otx2 and Hox genes have been shown to be down-and up-regulated, respectively, by retinoic acid both in vivo and in vitro (11,21,23). How Hox and Otx2 patterning information can complement each other during embryogenesis and how their mutually exclusive expression patterns are established still remain un- FIG. 6. A mutation in the ATTA core sequence of the HBE1 site abolishes the transactivation by HOXB1 and HOXB3 but not by HOXB2. The reporter construct pOTX2LucATTAbox-TK was mutagenized in the TAATTA core sequence (bottom part) of the HBE1 site to generate the pOTX2LucATTAbox-TKm construct, and the extent of its transactivation by the HOXB1, HOXB2, and HOXB3 proteins was compared with the one of the wild type construct. Data are expressed in -fold activation over the basal activity of the reporter construct pOTX2LucATTAbox-TK (wt) or pOTX2LucATTAbox-TKm (mut). solved questions. In principle, it was possible that HOX genes mediated the Otx2 restriction to the anterior third of the embryo through down-regulation of Otx2 expression in the posterior two-thirds. If this was the case, it was also possible that in vertebrates RA acts by activating Hox gene expression, which in turn would repress Otx2 expression by a direct interaction with its cis-regulatory sequences. On the contrary, we provide here evidence that Otx2 is, indeed, a candidate target for regulation by HOX proteins, but anterior HOX proteins activate, rather than repress, Otx2 transcription.

FIG. 5. Gel retardation assays of the HBE1 double-stranded oligonucleotide by nuclear extracts from HOXB1-, HOXB2-, and HOXB3-transfected COS cells. Panel A, nuclear extracts from mock-
In fact, a 1821-bp 5Ј genomic fragment of the Otx2 promoter is positively regulated by co-transfection with the expression vectors for HOXB1, HOXB2, and HOXB3 in an embryonal carcinoma cell line (NT2/D1), and a shorter fragment of only 534 bp is able to drive this regulation. HOXB1 and HOXB3 bind to a sequence containing two palindromic TAATTA sites, which bear four copies of an ATTA core sequence, a common feature of most HOM-C/HOX binding sites (8), while HOXB2 binds to an adjacent direct ACTT repeat. Mutagenesis of the TAATTA motif abolished HOXB1-and HOXB3-mediated, but not HOXB2-mediated, transactivation. These findings indicate that Hox genes in NT2/D1 cells up-regulate the expression of the Otx2 gene and that a direct role for Hox genes in the down-regulation of the early Otx2 expression in the posterior mouse embryo is unlikely. Moreover, a repressive role for HOXB1, HOXB2, and HOXB3 proteins is unlikely, because these homeoproteins have already been shown to act as transcriptional activators on other cis-regulatory sequences (32,36,38,39). However, we cannot rule out the possibility that other determinants or co-factors necessary for the repression of the Otx2 promoter might be absent (or present in a limiting amount) in the NT2/D1 cells used in the transfection assay.
The requirement for an inductive signal from anterior mesendoderm to stabilize Otx2 expression in the ectodermal layer has already been reported (14,21). In particular, it has been shown that isolated explants from mouse ectoderm become committed to express Otx2 in a cell-autonomous fashion only by the midstreak stage of development, corresponding to the time of onset of Hox gene expression (21). These findings have been implemented by data from homozygous mutant Otx2 Ϫ/Ϫ mice showing that the inductive signal from the mesendoderm is dependent upon a functional Otx2 allele. In fact, lacZ expression in Otx2 Ϫ/Ϫ mouse embryos, sustained by the vector used in the disruption experiment, was strongly maintained in the external layer, the endodermal component of the mesendoderm, but was almost undetectable in the embryonal ectoderm (17). It is conceivable that Hox genes might increase the Otx2 expression in the mesendoderm up to a threshold value necessary to achieve a proper inductive signal on the overlying ectoderm. This hypothesis is further supported by the observation that the Otx2 expression becomes progressively stronger as its restriction to the anterior neuroectoderm takes place.
Although in the NT2/D1 embryonal carcinoma cell line it is possible to reproduce the complementary transcriptional regulation of Otx2 and Hox gene expression by retinoids, we still do not know if these cells indeed provide an environment truly equivalent to the one found in the developing embryo. However, very recently Kimura et al. (40) identified a 49-bp Otx2 cis-regulatory element that is necessary and sufficient to drive a transgene expression in the cephalic mesenchyme of the developing mouse. In addition, they identified in pufferfish Fugu rubripes an Otx2 cis-regulatory sequence functionally equivalent to the murine 49-bp element and performed a mutational analysis of the murine genomic sequences that were most conserved between mouse and pufferfish. In this manner, the authors found that only mutagenesis of the same TAATTA core sequences identified in this report (Fig. 9) leads to a complete loss of transgene expression in the developing mouse embryo; these data demonstrate that the TAATTA motifs identified in our DNA binding and mutational analysis play a relevant functional role not only in vitro, but also in vivo.
The observation that Hox genes (namely, HOXC6 and Hoxd-8) that, at later stages of development, will be expressed in more posterior domains are not able to transactivate the Otx2 promoter shows that such interaction is not a common feature of all homeodomain-containing genes but rather is restricted only to the most anteriorly expressed. HOXB1, HOXB2, and HOXB3 are all able to bind to the same region in the Otx2 promoter but significantly differ in DNA binding specificity. HOXB1 and HOXB3 bind to the same site, with minor differences in the length of the protected sequence, whereas the adjacent site protected by HOXB2 contained a direct repeat of an ACTT core repeat, quite divergent from the ATTA consensus sequence. An almost identical sequence (GCT-TACTT versus ACTTACTT) was found protected in the A ␥globin enhancer by an activity specifically found in the K562 cell line, afterward identified as HOXB2 (36).
Finally, the data here presented indicate that, despite a well documented lack in DNA binding specificity shown by HOM/ Hox genes in vitro (4), the HOXB1, HOXB2, and HOXB3containing COS nuclear extracts (but not the HOXB3-containing bacterial ones) show some in vitro specificity, because in the genomic fragment used in the footprinting assay many other ATTA sites are present that are not recognized by any of the three homeoproteins. The DNA binding specificity shown by COS nuclear extracts might be due to determinants present in the unpurified nuclear extracts such as, for example, PBX or PBX-like proteins (41). The same DNA binding specificity is shared by nuclear extracts prepared from 12.5 d.p.c. mouse embryos, showing that the COS nuclear extracts can be considered quite representative of the in vivo situation. Moreover, the relevance of the protected sequence within the Otx2 Ϫ371 to ϩ163 region is also supported by the striking homology (75% identity) that we found between species as divergent as mouse and Xenopus, even higher than the reported homology between the pufferfish and mouse genomic sequences (40). It is here worth noting that we cannot establish if quantitative differences in DNA binding shown by the three homeoproteins used in this study are due to minor differences in the rate of production in COS cells or in their DNA binding affinity.
In conclusion, we have shown that HOXB1, HOXB2, and HOXB3 are able to positively interact with the Otx2 upstream regulatory sequence in an embryonal carcinoma cell line, and we have characterized their DNA-binding sites on this promoter; HOXB1 and HOXB3 bind to similar sequences, while HOXB2 displays a completely different DNA binding specificity. The region bound by the three homeoproteins is strikingly conserved through evolution and necessary (at least for HOXB1 and HOXB3) to mediate the regulation of the Otx2 promoter. Taken together, our data support the hypothesis that Hox genes might play a role in the refinement of the Otx2 early expression boundaries in vivo.