Integration of anteroposterior and dorsoventral regulation of Phox2b transcription in cranial motoneuron progenitors by homeodomain proteins

How specific neuronal types are generated at defined locations within the developing neural tube is still poorly understood. During early development of the central nervous system (CNS), neural progenitors acquire regionally restricted positional addresses by responding to graded inductive signals intersecting along the AP and DV axes(Jessell, 2000; Lumsden and Krumlauf, 1996). A grid-like set of positional cues is established within an initially equivalent group of cells that, together with the activation of a neurogenesis program(Bertrand et al., 2002), is translated into specific neuronal fates. Moreover, distinct neuronal types are generated in specific temporal order, often from common pools of neural progenitors (Jessell, 2000; Pattyn et al., 2003a).

In vertebrates, distinct sets of spatially restricted homeodomain (HD)transcription factors provide transcriptional readouts of AP and DV positional addresses in neural progenitors. Along the AP axis, distinct progenitor domains are generated by the nested expression patterns of the Hox HD-containing genes (Lumsden and Krumlauf,1996). In the ventral neural tube, DV positional addresses are instead conferred by Nkx-, Dbx-, Pax- and Irx-class HD proteins(Briscoe et al., 2000). Along both axes, auto- and crossregulatory activities among HD factors are required to refine and/or maintain progenitor domains(Briscoe et al., 2000; Dasen et al., 2003; Maconochie et al., 1997; Popperl et al., 1995). The combined activity of HD genes is thought to activate other sets of transcription factors that, in turn, regulate the expression of unique neuronal phenotypes (Lee and Pfaff,2001). Neuronal identity is therefore the result of a complex regulatory network of transcription factors acting sequentially. Despite the increasing knowledge about genetic cascades and epistatic relationships among HD factors, little is known about their direct downstream targets and how AP and DV molecular inputs are integrated for precise spatiotemporal transcriptional regulation.

Hox genes are involved in the specification of motoneuron (MN) subtype identities along the AP axis, both at spinal cord and hindbrain levels(Barrow and Capecchi, 1996; Cooper et al., 2003; Dasen et al., 2003; Davenne et al., 1999; Gaufo et al., 2000; Gaufo et al., 2003; Gavalas et al., 1997; Goddard et al., 1996; Guidato et al., 2003; Jungbluth et al., 1999; Pattyn et al., 2003a; Studer et al., 1996; Tiret et al., 1998). In ventral r4, for example, the development of facial branchiomotor (BM) neurons depends on Hoxb1 and Hoxb2 functions, raising the question of their direct molecular target(s). Transcriptional specificity of Hox factors is achieved upon heterodimerization with Pbx HD factors, murine homologs of Drosophila extradenticle (exd), and binding of bipartite PH sites (Chan et al.,1994; Maconochie et al.,1997; Mann and Affolter,1998; Mann and Chan,1996; Popperl et al.,1995). Pbx-Hox binding and transcriptional activity are further enhanced by Prep or Meis proteins, murine homologs of Drosophilahomothorax (hth), and additional members of the TALE (three-amino acid-loop-extension) class of HD factors(Burglin, 1997). By binding of distinct P/M sites in the vicinity of PH sites and direct interaction with Pbx, Prep/Meis/Hth proteins facilitate the formation of transcriptionally active ternary complexes (Berthelsen et al., 1998a; Berthelsen et al.,1998b; Ferretti et al.,2000; Gebelein et al.,2002; Jacobs et al.,1999; Ryoo et al.,1999). For example, Hox-Pbx-Prep complexes are involved in the maintenance of Hoxb1 and Hoxb2 transcription in r4(Ferretti et al., 2000; Jacobs et al., 1999; Maconochie et al., 1997; Popperl et al., 1995). Despite these insights into the regulation of Hox-mediated transcription, and Hox gene involvement in MN development, so far no direct Hox target gene has been identified that is required for MN specification and/or differentiation.

In the ventral neural tube, specification of neuronal progenitors requires HD factors that function as repressors of other repressors, that is by transcriptional derepression of downstream targets(Briscoe et al., 2000; Muhr et al., 2001). Similar derepression strategies have been conserved in various tissues or animal systems, indicating that they are efficient ways of keeping control of target gene expression (Barolo and Posakony,2002). Nonetheless, a derepression strategy involves the existence of transcriptional activators driving neuronal specification in suitably derepressed environments. Recently, retinoic acid signaling was identified as one of such activator pathways, involved in somatic MN (sMN) specification in the spinal cord (Novitch et al.,2003). While providing a rationale for how activators and repressors may interact to drive cell fate in a specific progenitor domain,the molecular mechanisms that allow switching from transcriptional repression to activation of target genes remain poorly understood. It is also important to investigate whether neuronal specification in other regions of the vertebrate CNS may rely on similar mechanisms. In the hindbrain, for example,it is unknown how the repressor activities of Nkx HD proteins may integrate with Hox factors to achieve spatially restricted regulation of MN programs and downstream targets.

Here, we have investigated the transcriptional regulation of the paired HD transcription factor Phox2b. In the mouse hindbrain, Phox2bis expressed in longitudinal columns, spanning several rhombomeres, that identify distinct populations of neural progenitors and postmitotic neurons(Pattyn et al., 1997). Phox2b is an obligatory determinant of cranial BM and visceral motor(VM) neuron specification (Dubreuil et al., 2002; Pattyn et al.,2000). BM and VM neurons – collectively referred to as vMN– innervate the muscles of the branchial arches and the parasympathetic ganglia, respectively. All vMNs are generated from a common ventral progenitor domain, pMNv, which is equivalent to the spinal p3 domain generating V3 interneurons (Briscoe et al.,2000; Pattyn et al.,2003a; Pattyn et al.,2003b). Throughout the hindbrain, the pMNv domain expresses Nkx2.2 and Nkx2.9, as well as Nkx6.1 and Nkx6.2 (Briscoe et al.,2000; Pattyn et al.,2003b). However, the distribution of vMN subtypes is rhombomere specific. Rhombomere 1 does not generate vMNs, r2-4 generate only BM neurons,whereas r5-7 generate both BM and VM neurons. In Phox2b knockout mice, vMN progenitors either do not exit the cell cycle or switch to a serotonergic fate (Dubreuil et al.,2000; Pattyn et al.,2000; Pattyn et al.,2003a). Thus, Phox2b acts as a binary switch in the selection of vMN or serotonergic fate. In Hoxb1 and Hoxb2knockout mice, maintenance of Phox2b expression is impaired and this results in facial BM to serotonergic fate switch in ventral r4(Pattyn et al., 2003a).

We show that Phox2b is a direct target of Hoxb1 and Hoxb2. We identify a conserved Phox2b enhancer containing separate PH and P/M sites, both of which are essential for ventral r4 regulation. We further show that transcriptional cooperation among Hox, Pbx, Prep factors and Nkx2.2, via a derepression mechanism, is an important component of Phox2benhancer activity. In addition, cooperation between Hox paralogs 1 or 2 and Nkx2 factors is required in vivo to generate ectopic Phox2b-expressing vMNs. These findings provide a molecular rationale to explain how AP and DV inputs are integrated on the Phox2b promoter to drive restricted expression in the r4 pMNv domain and facial MN fate.

Expression vectors were as described: mouse Hoxa2(Pasqualetti et al., 2000),mouse Hoxb1, human HOXB2, and Pbx1a (Di Rocco et al., 1997), Prep1(Berthelsen et al., 1998a), and chick Nkx2.2, Nkx2.2HD-VP16, Nkx2.2HD-EnR and Nkx2.2ΔTN(Muhr et al., 2001). To yield P2b_10/lacZ, a SalI-NdeI 10 kb fragment, including the Phox2b endogenous promoter and 5′ sequences, was cloned in front of a NLS-lacZ reporter cassette. P2b_2.8/lacZ was obtained by cloning the XbaI-XbaI 2.8 kb proximal fragment(Fig. 1) into the BGZ40 reporter (Studer et al.,1996). The P2b_0.38 fragment was PCR-amplified using the following primers: 5′GTTGACTAGTGGACGAAGAAGGGGGGAAACA3′ (sense) and 5′ACTTACTAGTAGTATATAGTCCTCATAATAAACTTG3′ (antisense). P2b_0.38 was cloned into the SpeI site of BGZ40 (P2b_0.38/lacZ) or into p-βglob-Luc (P2b_0.38/Luc), containing a humanβ-globin minimal promoter driving Luciferase expression. P2b_0.38mPH/Luc, P2b_0.38mPH/lacZ,P2b_0.38ΔPM/lacZ and P2b_0.38mPM/lacZ were generated with the Stratagene QuickChange™ Site-directed Mutagenesis Kit. To introduce nucleotide changes in the PH site the following primers were used:5′TTCCAAGTAGCGTCGTTCGATTAAAGGGCAGGAGCTGGTTAGAAG3′ (sense) and 5′CTTCTAACCAGCTCCTGCCCTTTAATCGAACGACGCTACTTGGAA3′ (antisense). To introduce mutations in the P/M site the following primers were used:5′GCCCAATAGACGGATGAGTTAGTAAAAAGCGCCAGCAATAAG3′ (sense) and 5′CTTATTGCTGGCGCTTTTTACTAACTCATCCGTCTATTGGGC3′ (antisense) for P2b_0.38ΔPM/lacZ; and 5′GCCCAATAGACGGATGAGTTATCTGTATGTAAAAAGGCGCAGC3′ (sense) and 5′GCTGGCGCTTTTTACATACAGATAACTCATCCGTCTATTGGGC3′ (antisense) for P2b_0.38mPM/lacZ. To construct P2b_10ΔPH/lacZ, a 1.3 kb NheI-XmaI fragment containing the P2b_0.38 enhancer was subcloned from the P2b_10/lacZ into pBluescript KS (Stratagene). Deletion of 11 bp, including the PH site was obtained by using the following primers: 5′TCATTATTTCCAAGTAGCGTAGGGCAGGAGCTGGTTAGAA3′ (sense) and 5′TTCTAACCAGCTCCTGCCCTACGCTACTTGGAAATAATG3′ (antisense). The wild-type 1.3 kb fragment in P2b_10/lacZ was replaced by the fragment including the PH deletion to obtain P2b_10ΔPH/lacZ. The p3xPH/Luc and p3xPH/lacZ reporter plasmids were obtained by cloning into the NheI site of p-βglob-Luc or the XbaI site of BGZ40, respectively, the following double stranded fragment containing three copies of the PH site and adjacent sequences (17 bp):5′ACTTACTAGTCGTGATTGAATTAAAGGCGTGATTGAATTAAAGGCGTGATTGAATTAAAGGACTAGTACTT3′with its complementary antiparallel oligonucleotide. The reporter plasmid p3xmPH/Luc was constructed as p3xPH/Luc, except that the PH site was mutated as in P2b_0.38mPH/Luc and P2b_0.38mPH/lacZ.

Chick eggs were incubated in a humidified chamber, and embryos were staged according to (Hamburger and Hamilton,1951). DNA constructs were injected into neural tubes of stage HH10-12 chick embryos. Electroporation was performed as described previously(Itasaki et al., 1999) using a square wave electroporator. Construct concentrations were: 1.5 mg/ml reporter construct, 1 mg/ml expression vector and 0.5 mg/ml co-injected pCMV/EGFP as a tracer of electroporated cells. Embryos were harvested 24-48 hours after electroporation and processed for immunohistochemistry, in situ hybridization or β-galactosidase staining.

P19 embryonic carcinoma cells were cultured in Dulbecco minimal essential media supplemented by 5% fetal calf serum and 5% delipidated fetal calf serum. Cells were seeded, incubated for 36 hours and transient transfections were performed using Lipofectamine 2000 (Invitrogen) according to the manufacturer's protocol. In a typical experiment, 500 ng of reporter plasmid,250-500 ng of each expression construct and 100 ng of pCMV-β-gal (as a control of transfection efficiency) were used per well in six-well plates. Cells were lysed 40-45 hours after transfection then assayed for luciferase activity. Values were normalized by β-gal activity. Data represent means of duplicate values from representative experiments. All transfections were independently repeated at least three times.

Generation of mouse transgenic embryos was performed as described(Popperl et al., 1995). Embryos were harvested at E10.5. β-Gal was detected either by whole-mount in situ hybridization with a lacZ probe (a kind gift of M. Kmita) or by X-Gal staining.

Proteins were in vitro translated using the coupled TNT transcription/translation system (Promega) in the presence of ³⁵S methionine. Proteins were visualized by SDS-PAGE, followed by autoradiography. EMSA was performed according to (Ferretti et al., 2000). Antibodies used are polyclonal anti-Hoxb1 (Babco),polyclonal anti-Pbx1 and anti-Prep1 (Santa Cruz Biotechnology).

Monoclonal anti-Isl1/2 (Developmental Studies Hybridoma Bank) was used for immunohistochemistry, as described(Tsuchida et al., 1994). Chick Phox2b (Ernsberger et al.,1995), Isl2 (Tsuchida et al., 1994) and Hb9(Lee and Pfaff, 2001) probes were used for in situ hybridization as described previously(Gavalas et al., 1997).

In transient mouse transgenic assay, a 10 kb genomic construct driving the lacZ reporter gene, including the Phox2b promoter and its upstream sequences (P2b_10/lacZ, Fig. 1A) recapitulated most of the endogenous Phox2b expression pattern at E10.5(Fig. 1B,D; data not shown). In the hindbrain, the r2-r6 dorsal columns of Phox2b expression were readily identified, as well as the two ventral columns that included the developing cranial vMNs (Fig. 1B) (Pattyn et al.,1997). To further identify fragments driving expression in Phox2b subset domains, we deleted the 5′-most 7 kb from the 10 kb genomic fragment. The resulting 2.8 kb region was placed upstream of an heterologous promoter driving the lacZ reporter(P2b_2.8/lacZ). Three out of eight P2b_2.8/lacZ transgenic embryos displayed detectable reporter levels at E10.5. Interestingly, lacZ expression was restricted to the ventral neural tube(Fig. 1C; data not shown). In particular, two out of three embryos displayed expression in two ventral columns running throughout the hindbrain and spinal cord (data not shown). In the third embryo, reporter expression was restricted to the ventral region of r4 (Fig. 1C), matching the Phox2b endogenous domain. Thus, the 2.8 kb fragment contains cis-regulatory sequences integrating both AP and DV positional information to drive spatially restricted Phox2b expression in ventral r4.

To identify conserved cis-regulatory elements within the 2.8 kb fragment potentially involved in r4 restricted regulation, we compared its sequence with Phox2b genomic sequences from other organisms. A stretch of 376 base pairs (bp) in the mouse proximal promoter region (GenBank Accession Number AY 640178) was highly conserved in human and rat (97% at the nucleotide level), as well as pufferfish (Fugu) and zebrafish (Danio) Phox2bgenomic regions (Fig. 2A,B). To test for regulatory potential of this conserved region, we used in ovo electroporation in the chick hindbrain. This is a suitable system to study conservation of transcriptional regulatory mechanisms (e.g. Itasaki et al., 1999; Manzanares et al., 2001). We electroporated a lacZ reporter construct carrying the mouse 376 bp Phox2b conserved enhancer (P2b_0.38/lacZ) into neuroepithelial cells along one side of the neural tube of stage HH10-12 chick embryos. High reporter expression levels were restricted to r4 and, to a lesser extent, to r2, whereas only weak activation was found in other hindbrain regions or rostral spinal cord(Fig. 3C). In r4, expression always extended more ventrally than in r2 (arrow, Fig. 3C). Electroporation of a control construct without the enhancer did not result in any activation (data not shown).

In summary, the conserved P2b_0.38 enhancer responded in a rhombomere-restricted fashion to the activity of endogenous factors in the chick hindbrain. Such response reminded that of Hox-regulated enhancers(Maconochie et al., 1997; Jacobs et al., 1999; Ferretti et al., 2000; Popperl et al., 1995),suggesting that Hox-responsive sequences are present in the P2b_0.38 enhancer.

To investigate transactivation by Hox transcription factors, we transfected murine embryonal carcinoma P19 cells, a suitable system for the analysis of the transcriptional activity of Hox proteins(Di Rocco et al., 2001; Saleh et al., 2000). A Luciferase reporter construct driven by the conserved Phox2benhancer in front of a minimal promoter (P2b_0.38/Luc) was co-transfected along with Hoxb1, HOXB2 or Hoxa2 expression vectors. A three-to fourfold increase of P2b_0.38/Luc basal transcriptional activity was observed with either one of the Hox vectors(Fig. 3A; not shown). To extend these findings in vivo, we co-electroporated Hox vectors with P2b_0.38/lacZ (Fig. 3E,G,I) in the chick neural tube. We observed a marked upregulation of P2b_0.38/lacZ activity in the hindbrain and rostral spinal cord of embryos co-electroporated with either Hoxb1 (n=21/32),HOXB2 (n=10/12) or Hoxa2 (n=12/18), when compared with P2b_0.38/lacZ alone (Fig. 3C). Unlike the modest transactivation observed in P19 cells(Fig. 3A), this robust effect indicated that the in vivo activity of electroporated Hox proteins may be enhanced by the presence of endogenous co-factors. We therefore tested whether Hox transcriptional activity on P2b_0.38/Luc was improved by co-transfections with Pbx1a and Prep1 expression vectors(Fig. 3A). Pbx1a or Prep1 alone were unable to stimulate the reporter activity more than two- to threefold. By contrast, co-transfection of either Hoxb1, HOXB2 or Hoxa2 with Pbx1a and Prep1 co-factors resulted in a significant 13-14-fold enhancement of transcription(Fig. 3A; and not shown).

In summary, the Phox2b conserved enhancer can be transactivated by Hoxb1, HOXB2 or Hoxa2 both in cultured cells and chick neural tube. Furthermore, the observed Hox-mediated transcriptional activity is enhanced by the co-factors Pbx1a and Prep1. These results strongly suggest that DNA binding site(s) for Hox proteins and their co-factors are present in the P2b_0.38 enhancer.

Indeed, sequencing of the Phox2b enhancer revealed the presence of a putative bipartite PH-binding site (TGATTGAA)(Fig. 2B). Notably, its nucleotide sequence was identical to that of the low-affinity PH binding site of repeat 2 (R2) of the Hoxb1 autoregulatory (b1-ARE) r4 enhancer(Popperl et al., 1995)(Fig. 2C). Moreover, it shared fairly high conservation with the PH site present in the Hoxb2 r4 enhancer, also regulated by Hoxb1(Ferretti et al., 2000; Jacobs et al., 1999; Maconochie et al., 1997)(Fig. 2C). Similar to the Hoxb1 and Hoxb2 r4 enhancers, we also found a conserved P/M site (TTGTCATG), downstream of the PH site(Fig. 2B,C). The Phox2b P/M site and its flanking nucleotides exactly matched the sequence found in the Hoxb1 r4 enhancer and shared six out of eight nucleotides with that in the Hoxb2 r4 enhancer(Ferretti et al., 2000; Jacobs et al., 1999). Interestingly, unlike the previously identified PH and P/M sites lying in relative proximity to each other, the Phox2b P/M site was 147 nucleotides distant from the PH site (Fig. 2B).

To test for direct binding, we run EMSA assays using Hox, Pbx1a and/or Prep1 in vitro translated proteins on a 30 bp oligonucleotide probe containing the Phox2b PH site and flanking sequences, or on a PCR-amplified radiolabeled fragment (233 bp) including both PH and P/M sites in their native context (Fig. 4A,B). As for the Hoxb1 and Hoxb2 r4 enhancers(Berthelsen et al., 1998a; Ferretti et al., 2000; Jacobs et al., 1999; Popperl et al., 1995), the PH-containing probe (Fig. 4A)was readily bound by Pbx-Prep heterodimers, whereas no binding was observed with Pbx or Prep on their own (Fig. 4A, lane 4 arrow; not shown). Moreover, nucleotide changes of the PH site (same mutation as in Fig. 3B,D and 6B,D; see below) abrogated binding of the Pbx-Prep heterodimer (not shown). By contrast, we did not detect Hoxb1-Pbx1a heterodimers nor Hoxb1-Pbx1a-Prep1 heterotrimers,indicating low in vitro binding affinity of the PH site for Hox-containing multimeric complexes (Fig. 4A,lanes 3,5). Such binding behavior was consistent with that of the PH site in the R2 of the b1-ARE r4 enhancer (Popperl et al., 1995), which did not show cooperative binding of Pbx-Hox complexes in vitro, while contributing to Hoxb1 r4 expression in vivo(Popperl et al., 1995) (see Discussion).

Next, we tested whether the binding of a Hoxb1-Pbx1a-Prep1 ternary complex would be stimulated by the simultaneous presence of both PH and P/M motifs, as for the Hoxb1 and Hoxb2 r4-regulated enhancers(Ferretti et al., 2000; Jacobs et al., 1999). Indeed,a slower migrating band was detected upon incubation of Hoxb1, Pbx1a and Prep1 with a probe containing both the PH and P/M sites in their native context,even though spaced by 147 nucleotides (arrow in Fig. 4B, lane 5). This band did not form with only Hoxb1-Pbx1a (lane 3) or Pbx1a-Prep1 (lane 4) pairs, nor with single proteins (data not shown). The presence of all three HD proteins in the complex was further confirmed, as the addition of either anti-Hoxb1,anti-Pbx or anti-Prep antibodies selectively inhibited its formation(Fig. 4B, lanes 6-8). Importantly, deletion of eleven nucleotides encompassing the PH site (same mutation tested in mouse transgenic analysis; Fig. 5) in the context of the entire probe containing the wild-type P/M site prevented all complex formation, demonstrating the requirement of the PH site for efficient multimeric complex association (Fig. 4B, lane 11). Moreover, a point mutant of the P/M site (same mutation tested in chick hindbrain; Fig. 3) in the presence of a wild-type PH site was also unable to bind a ternary complex (Fig. 4B,lane 16), indicating that both the PH and P/M sites are essential for binding of a Hoxb1-Pbx1a-Prep1 complex.

In summary, the combination of low-affinity PH and distant P/M motifs on the Phox2b enhancer can support the assembly of a Hoxb1-Pbx1a-Prep1 ternary complex in vitro. Moreover, both the PH and P/M sites are essential for ternary complex binding. These results support the transcriptional cooperation of the three HD proteins observed in cell culture(Fig. 3A), and raise the question of the specific contribution of the identified sites to the transcriptional regulation of the Phox2b enhancer.

To test the involvement of the PH site in the activity and spatial regulation of the Phox2b enhancer, we introduced nucleotide mutations changing TGATTGAA to TcgTTcgA, while leaving the rest of the enhancer unaltered. Similar nucleotide changes were previously shown to impair PH site activity in the Hoxb1 and Hoxb2 enhancers(Ferretti et al., 2000).

A Luc construct carrying the mutated enhancer(P2b_0.38mPH/Luc) was co-transfected in P19 cells along with Hoxb1,HOXB2, Hoxa2, Pbx1a and/or Prep1 vectors. The transcriptional activity of P2b_0.38mPH/Luc induced by each protein alone remained just above the basal level (compare Fig. 3A with 3B; and not shown). Notably, the enhanced transcriptional response induced by Hox/Pbx or Hox/Pbx/Prep combinations on the wild-type enhancer(Fig. 3A) was abrogated on the P2b_0.38mPH/Luc construct (Fig. 3B).

To investigate the PH site requirement for Hox-mediated spatial regulation,a lacZ construct carrying the mutated Phox2b enhancer(P2b_0.38mPH/lacZ) was electroporated in the neural tube of stage HH10-12 chick embryos. Twenty-four hours later, two main expression differences were observed with the mutant P2b_0.38mPH/lacZ construct,when compared with the wild-type P2b_0.38/lacZ (compare Fig. 3C,D). First, a general decrease of β-gal expression levels and/or number of expressing cells throughout the expression domain. Second, the ventral domain of r4 expression was invariably lost or severely impaired (n=34/34; arrow, Fig. 3D). Mutation of the PH site also severely reduced Hox-induced upregulation of the enhancer, as assessed by co-electroporation of P2b_0.38mPH/lacZ and Hoxb1 or Hox paralog group 2 vectors (compare Fig. 3E,F; data not shown). Thus, the PH site is an essential component of Hox-mediated regulation in ventral r4.

As the PH site contributes to ventral r4 expression in the context of the Phox2b enhancer, we next asked whether it was also sufficient to direct spatially restricted expression in the chick hindbrain. However, unlike the full enhancer (Fig. 3C),the p3xPH/lacZ construct, driven by three copies of a 17 bp oligonucleotide containing the Phox2b PH site and flanking sequences,was weakly active and did not display spatially restricted reporter expression in electroporated embryos (n=32/32, Fig. 6E; see Discussion). Thus,interactions with additional sequences within the Phox2b enhancer are required for the PH site to fully function in vivo. To determine if the P/M motif also contributes to enhancer activity, we generated a six-nucleotide deletion (Fig. 3H) or a four-nucleotide exchange (Fig. 3J; same mutation that abolishes ternary complex binding in vitro in Fig. 4) of the P/M site in the Phox2b enhancer and tested their effects on reporter activity in chick electroporation. Notably, both P/M mutations phenocopied the PH mutation and abrogated ventral reporter expression in r4 (n=30/30, Fig. 3H; n=12/12, Fig. 3J).

Thus, the PH and P/M sites are both essential for Hox-mediated transcriptional regulation of the Phox2b enhancer activity in ventral r4, strongly supporting the involvement of a Hox-Pbx-Prep transcriptionally active ternary complex in vivo.

To investigate the relevance of the PH motif to Hox-mediated regulation of mouse Phox2b, we generated an 11 bp deletion of the PH site (same mutation that abolishes ternary complex formation in vitro; Fig. 4) in the context of the 10 kb Phox2b genomic construct (P2b_10ΔPH/lacZ), and tested it in mouse transgenic embryos (Fig. 5). Analysis of whole-mount hindbrains of E10.5 transgenic embryos did not reveal overt differences in reporter expression between mutated and wild-type constructs (compare Fig. 5A,B; Fig. 1B; data not shown), indicating that the establishment of the major domains of Phox2b expression in the hindbrain does not depend on an intact PH site. However, a selective, spatially restricted, difference was detected in ventral r4 upon sectioning of embryos carrying the mutant construct(Fig. 5D). B-gal expression in the pMNv progenitor domain was eliminated (six out of 11 embryos) or severely reduced (the five remaining embryos; Fig. 5D), whereas mantle layer (ML) expression was not significantly affected (Fig. 5B,D; data not shown).

The effect of the PH site deletion precisely mirrored the loss or reduction of Phox2b expression in the r4 pMNv progenitor domain of E10.5 Hoxb1^–/– or Hoxb2^–/– mutant embryos, respectively(Davenne et al., 1999; Gaufo et al., 2000) (see Pattyn et al., 2003a). In these mutants, ventral r4 expression of Phox2b at E10.5 is not properly maintained, leading to a facial MN to serotonergic switch of progenitor fate. Our results therefore strongly suggest that maintenance of Phox2b expression by Hoxb1 and/or Hoxb2 in ventral r4 progenitors is directly regulated through the PH site. Moreover, the PH site appears to integrate both AP and DV regulatory inputs as its mutation affects ventral regulation in r4.

Nkx2 proteins are good candidates for providing DV regulatory inputs that restrict Phox2b expression to the pMNv domain. In fact, Nkx2.2/2.9 and Phox2b expression patterns are co-extensive in the pMNv domain (Pattyn et al.,2003a). Moreover, gain-of-function studies have involved Nkx2 proteins in ectopic Phox2b activation(Pattyn et al., 2003b). Thus,Nkx2 patterning factors may interact with Hox and their co-factors to allow high Phox2b expression levels specifically in the ventral r4 progenitor domain. To test whether Hox and Nkx2 factors transcriptionally cooperate to regulate the Phox2b enhancer, we first examined the transcriptional activity of Nkx2 factors on P2b_0.38/Luc in P19 cells. Co-transfection of Nkx2.2 or Nkx2.9 vectors alone did not stimulate reporter activity more than two- to threefold, comparable with the modest activity induced by Hoxb1 alone (Fig. 6A; data not shown). Co-transfection of Nkx2.2 with Pbx and Prep,in the absence of Hoxb1, did not stimulate reporter activity more than fourfold (Fig. 6A). Notably,when Hoxb1 was co-expressed with Nkx2.2, reporter activity was cooperatively stimulated up to tenfold. Further addition of Pbx and Prep co-factors resulted in a synergistic enhancement of transcription up to 20-fold, significantly exceeding the transcriptional enhancement observed with only Hox, Pbx and Prep(Fig. 6A). Importantly,mutation of the PH site within the context of the full enhancer abolished the transcriptional cooperation between Hox, its HD co-factors and Nkx2.2(P2b_0.38mPH/Luc; Fig. 6B).

In principle, the transcriptional enhancement mediated by Nkx2.2 in the presence of Hoxb1 and its co-factors may require DNA binding on the conserved Phox2b enhancer. However, in EMSA assays Nkx2.2 did not bind any potential Nkx consensus binding site or core HD-binding sequence from the Phox2b enhancer (data not shown). Most compellingly, co-transfection of a Nkx2.2HD-VP16 chimeric construct, consisting of the HD of Nkx2.2(Nkx2.2HD) coupled to the VP16 activator domain(Muhr et al., 2001), did not stimulate transcription of P2b_0.38/Luc(Fig. 6A), in keeping with the idea that Nkx2.2 does not bind the enhancer.

In the ventral neural tube, Nkx transcription factors work as transcriptional repressors (Muhr et al.,2001). However, on the Phox2b enhancer Nkx2.2 stimulates transcriptional activation, even though only in the presence of Hoxb1 and its co-factors (Fig. 6A). In principle, this positive effect could also be the result of Nkx2.2 acting as a repressor, by relieving an inhibition on the transcriptional activation stimulated by Hox and co-factors.

We therefore examined the activity of a hybrid construct consisting of Nkx2.2HD coupled to the Engrailed repressor domain (Nkx2.2HD-EnR). This construct functions as a repressor in transfection assays, and mimics the repressive ability of full-length Nkx2.2 in the chick neural tube(Muhr et al., 2001). Strikingly, co-transfection of Nkx2.2HD-EnR in the presence of Hoxb1, Pbx1a and Prep1 led to a 25-fold stimulation of the P2b_0.38/Luc reporter transcription, while co-transfection of Nkx2.2HD-EnR alone had no effect(Fig. 6A). Thus, the observed Nkx2.2-mediated enhancement of Hox, Pbx and Prep-induced transcription is accounted for by its repressor activity, as it can be mimicked by the EnR domain.

Next, we asked whether the PH site is sufficient to mediate Hox and Nkx2.2 cooperation. Co-transfection in P19 cells of the p3xPH/Luc construct,which contains three copies of a 17 bp oligonucleotide including the PH site and its flanking sequences, along with Hoxb1, Pbx1a, Nkx2.2 or Nkx2.2HD-EnR vectors alone did not stimulate reporter activity more than twofold(Fig. 6C). Co-expression of Hoxb1 and Pbx1a enhanced p3xPH/Luc activation by about sevenfold. Importantly, co-transfection of Hoxb1 and Pbx1a with Nkx2.2 or Nkx2.2HD-EnR resulted in a robust synergistic stimulation of reporter activity by 14-fold and 20-fold, respectively (Fig. 6C), reproducing the effect observed with the full enhancer. This effect was abolished upon mutation of the PH site (p3xmPH/Luc; Fig. 6D). Notably,co-expression of Hoxb1 and Pbx1a with a truncated version of Nkx2.2 lacking the N-terminal TN domain (Nkx2.2ΔTN), which mediates interaction with co-repressors of the Groucho (Gro)/TLE family(Muhr et al., 2001), almost abolished transcriptional synergy (Fig. 6C).

Finally, the modest activity observed with Nkx2.2 (or Nkx2.2HD-EnR) when co-transfected alone with p3xPH/Luc(Fig. 6C), suggested that Nkx2.2 is not sufficient on its own to stimulate transcription at the PH site in the absence of Hox and Pbx co-factors, the endogenous levels of which are low in P19 cells (Saleh et al.,2000). We therefore tested whether overexpressing Nkx2.2 stimulated transcriptional activity at the PH site in the chick neural tube, a context in which Hox and its co-factors are endogenously available. Interestingly, although weakly active alone(Fig. 6E; see above), the p3xPH/lacZ reporter expression was significantly stimulated by co-electroporation of either Nkx2.2 or Nkx2.2HD-EnR vectors (n=14/16, Fig. 6F; n=10/12, Fig. 6G).

Altogether, these data indicate that Nkx2.2-mediated derepression,partially regulated through interaction with Gro corepressor(s), alleviates a repressive activity at, or in the vicinity of, the PH site, allowing transcriptional activation by Hox and Pbx factors(Fig. 6H,I; see Discussion). These data provide a molecular framework for understanding how AP and DV molecular inputs are integrated on the Phox2b enhancer and have relevance for the mechanism of generation of vMNs at specific hindbrain locations.

The ectopic expression of Hoxb1 or Hoxa2 in r1, an area normally devoid of BM neurons, led to the generation of ectopic facial or trigeminal BM neurons, respectively(Jungbluth et al., 1999). However, ectopic BM neurons were only detected ventrally, despite widespread Hox expression throughout the dorsoventral extent of r1, suggesting the requirement for an additional ventral input for BM neuron specification(Jungbluth et al., 1999). Nkx2 proteins could provide this ventral regulatory input, since electroporation of Nkx2.2 in the chick hindbrain is sufficient to induce ectopic Phox2b expression and generation of BM neurons at dorsal neural tube levels (Pattyn et al.,2003b).

To investigate in vivo cooperation of Hox and Nkx2 factors in ectopic BM neuron generation, we first evaluated the AP distribution of ectopic Phox2b-expressing cells induced by forced Nkx2.2 expression. Stage HH10-12 embryos were electroporated and analyzed 48 hours later. Interestingly, ectopic Phox2b-expressing cells were detected at dorsal levels but only up to r2, i.e. within the Hox⁺domain, and never in r1 (Fig. 7C). Conversely, forced Hoxb1 or Hoxa2 expression throughout the hindbrain resulted in ectopic Phox2b-expressing cells in r1, but only ventrally, i.e. within the Nkx2.2⁺ domain(Fig. 7A,B). No ectopic Phox2b expression was detected at dorsal levels (the dorsoventral extent of electroporation was assessed by GFP co-injection; data not shown). Interestingly, only the combination of either Hoxb1/Nkx2.2 or Hoxa2/Nkx2.2 vectors could stimulate the generation of ectopic Phox2b⁺,Isl1⁺, Isl2^–, Hb9^– vMNs in dorsal r1 (Fig. 7D-I; data not shown).

Thus, the generation of ectopic Phox2b-expressing BM neurons requires in vivo cooperation of Hox paralogs 1 or 2 and Nkx2 HD factors.

Hox transcription factors direct the patterning of a variety of structures in the developing embryo and are thought to regulate numerous genes, but to date only a few direct targets have been identified, mainly in Drosophila (Capovilla et al.,1994; Capovilla et al.,2001; Galant et al.,2002; Lohmann et al.,2002; Vachon et al.,1992). In the vertebrate hindbrain, Hox genes regulate rhombomere-specific neuronal patterning, but it is unclear how they may link early neural patterning to the establishment of neuronal fates, as their direct downstream effectors remain elusive. We provide the first direct link between Hox function in the hindbrain and the expression of a downstream effector, Phox2b, that is an obligatory determinant of cranial vMN specification (Brunet and Pattyn,2002).

We have focused on the regulation of Phox2b expression in the pMNv domain of ventral r4. The r4 pMNv domain gives rise to the facial BM and inner ear efferent neurons of the VIIth cranial nerve(Bruce et al., 1997; Pattyn et al., 2003a; Simon and Lumsden, 1993; Tiveron et al., 2003), and analysis of knockout mice revealed that Hoxb1 and Hoxb2 are required for maintenance of the late phase of Phox2b expression in this progenitor domain (Pattyn et al., 2003a). We provide several lines of evidence supporting a direct regulation by Hoxb1 and/or Hoxb2, involving Pbx and Prep/Meis proteins as co-factors. First,ectopic expression of Hoxb1 or Hox paralog group 2 in the chick neural tube can induce ectopic Phox2b expression(Fig. 7). Second, a conserved 376 bp enhancer, enclosed within a 2.8 kb Phox2b genomic fragment that drives ventrally restricted r4 expression in the mouse(Fig. 1), contains separate PH and P/M sites whose conserved sequences are hallmarks of Hox-mediated transcriptional regulation in r4 (Fig. 2) (Ferretti et al.,2000; Jacobs et al.,1999; Popperl et al.,1995). An intact PH site is required for Hoxb1-, HOXB2- or Hoxa2-mediated transactivation of the Phox2b enhancer, in both P19 cells and chick hindbrain (Fig. 3). Importantly, both the PH and P/M motifs are essential for binding in vitro of a Hox-Pbx-Prep ternary complex and for enhancer activity in ventral r4 of chick embryos. Finally, mutation of the PH site selectively impairs the regulation of a mouse 10 kb Phox2b transgenic construct,recapitulating endogenous Phox2b expression(Fig. 1), in the ventral r4 pMNv domain (Fig. 5). Moreover,the effect of the PH mutation faithfully mimics the endogenous Phox2bdownregulation observed in Hoxb1 and Hoxb2 knockout mice(Pattyn et al., 2003a).

Our results further suggest that Hoxa2 could also directly regulate the Phox2b enhancer. However, analysis in Hoxa2knockout mice did not reveal obvious Phox2b expression defects in ventral r4, indicating a major role for Hoxb1 and Hoxb2 at that level. By contrast, in Hoxa2 mutants Phox2b expression is lost in the r2-r3 dorsal columns(Davenne et al., 1999). Sequences mediating regulation by Hoxa2 in dorsal columns may reside outside the 2.8 kb Phox2b genomic construct, as this fragment drives only ventral expression in transgenic mice (Fig. 1C).

Altogether, our data lead us to conclude that, in the ventral r4 pMNv domain, Phox2b is a direct target of Hoxb1 and Hoxb2.

Similar to the Hoxb1 and Hoxb2 r4 enhancers, we found separate PH and P/M sites embedded within the Phox2b enhancer. Nevertheless, the in vivo output of Hox regulation on these three enhancers is rather different, as the Phox2b PH or P/M sites mediate a transcriptional response restricted to ventral progenitors, despite widespread Hoxb1 and Hoxb2 distribution throughout r4. This is in keeping with the observation that endogenous Phox2b expression is upregulated in sharp columns of selected progenitor domains at distinct DV levels. Comparing the nature and function of bipartite PH and P/M sites in the context of the Hoxb1, Hoxb2 and Phox2b enhancers may therefore provide clues of how Phox2b regulation is spatially constrained.

In the Hoxb2 enhancer, only one PH site is present that shows cooperative binding of Hoxb1 and Pbx/Exd proteins in vitro and is required for r4 expression in vivo (Maconochie et al.,1997). By contrast, the b1-ARE enhancer contains three PH motifs(R1-R3). Mutational analysis in the mouse indicated that all three PH sites are cooperatively required for high levels of r4 expression(Popperl et al., 1995),although with distinct individual contributions. Among the three Hoxb1 PH sites, the R2 sequence precisely matches that of the Phox2b PH octamer core (Fig. 2C). Like the Phox2b PH site, the R2 repeat did not bind Hoxb1/Exd heterodimers in vitro, nor Hoxb1 or Exd alone, although it is necessary for optimal r4 activity (Popperl et al., 1995). Thus, the Hoxb1 R2 repeat requires cooperative interactions with adjacent sequences in the b1-ARE to fully function in vivo. Similarly, a trimerized Phox2b PH site was not sufficient on its own to direct r4 restricted expression in the chick hindbrain (Fig. 6E), unlike the sufficiency for r4 expression of multimerized Hoxb1 R3 or Hoxb2 high-affinity PH sites(Maconochie et al., 1997; Popperl et al., 1995). Nonetheless, the PH motif was necessary, in the context of the Phox2benhancer, for mediating the transcriptional cooperation of Hox, Pbx and Prep/Meis co-factors and for in vivo regulation in ventral r4 both in chick and mouse hindbrain (Figs 3, 5). Thus, the Phox2blow-affinity PH site, while representing a necessary site of integration of r4 activity, operates in vivo mainly through cooperative interactions with its surrounding regulatory environment, even in the presence of high endogenous levels of binding factors.

Cooperative interactions of PH sites with nearby sequences are important for in vivo specificity of Hox-Pbx complexes in both vertebrate and invertebrate Hox-regulated enhancers(Jacobs et al., 1999; Ferretti et al., 2000; Manzanares et al., 2001; Di Rocco et al., 2001; Mann and Affolter, 1998). We show that a distant P/M site makes an essential contribution to the binding specificity of the PH element, allowing formation of a Hox-Pbx-Prep complex in vitro, as ternary complexes were not observed on DNA probes containing mutations of either PH or P/M sites (Fig. 4). Moreover, regulation of the Phox2b enhancer in ventral r4 requires the integrity of both PH and P/M sites(Fig. 3), indicating the formation of transcriptionally active Hox-Pbx-Prep complexes in vivo. Although this functional behavior is reminiscent of that of the P/M element in the Hoxb2 r4 enhancer, it differs from that of the Hoxb1 P/M motif, functionally redundant with the R1-R3 elements(Jacobs et al., 1999; Ferretti et al., 2000). In addition, it should be noted that the Hoxb1, Hoxb2 and Phox2b enhancers differ in the spacing and relative orientations of their P/M and PH motifs. Although the Hoxb2 and Hoxb1 (R2)P/M sites are located close to the 5′ and 3′ ends of their PH sites, respectively, the mouse Phox2b P/M element is located 147 nucleotides 3′ to the PH motif (Fig. 2). Different configurations and spacing might correlate with distinct spatial and/or levels of activity of Hox-regulated r4 enhancers(Jacobs et al., 1999). In this respect, the organization of PH-P/M modules in the Hoxb1 and Hoxb2 enhancers vary among vertebrate species that show fine regulatory differences in r4 and its derivatives(Popperl et al., 1995; Scemama et al., 2002). The permissivity of the different PH-P/M spacing could be explained by looping-out or bending of the intervening DNA (Fig. 6I), to allow formation of a trimeric complex. In the Phox2b enhancer, the unusual spacing of the P/M site might introduce further constraint on the ability of the low-affinity PH site to be activated in vivo, despite high endogenous Hoxb1, Hoxb2 and co-factor levels.

In conclusion, unlike the Hoxb1 or the Hoxb2 r4 enhancers that contain multiple and/or high-affinity PH sites readily activated by threshold levels of endogenous Hox proteins and their co-factors, the low-affinity Phox2b PH motif must integrate additional inputs in order to be fully functional in vivo. It is tempting to speculate that similar low-affinity PH sites are present in the enhancers of Hox target genes, the expression of which is tightly regulated in sharp columns in the hindbrain and the activation of which outside their normal domains would have deleterious consequences for neuronal patterning.

Our data strongly suggest that the Phox2b PH-P/M module is involved in the maintenance of high Phox2b expression levels in ventral r4. First, inactivation of the Phox2b PH site mirrors the effect of Hoxb1 or Hoxb2 loss-of-function in mice, i.e. the lack of maintenance of Phox2b expression in the r4 pMNv domain(Pattyn et al., 2003a). Second, other conserved PH-P/M cis-regulatory modules in Hoxb1, Hoxb2,Hoxa3 and Hoxb4 enhancers are all involved in Hox-dependent maintenance of rhombomere-restricted expression(Popperl et al., 1995; Jacobs et al., 1999; Ferretti et al., 2000; Maconochie et al., 1997; Manzanares et al., 2001; Gould et al., 1997). Clearly,other elements must then be required for the initiation of Phox2bexpression in the pMNv domain. In this respect, forced expression of Hox and Nkx2 factors is sufficient to induce ectopic Phox2b expression and generates ectopic BM neurons in the chick hindbrain(Fig. 7), indicating that these factors could also mediate Phox2b activation through additional sequences other than the identified PH or P/M sites.

In conclusion, three key r4 targets of Hox paralog 1 and 2 genes,i.e. Phox2b, Hoxb1 and Hoxb2 bear conserved PH-P/M modules,arguing for a cis-regulatory signature that could be shared by a more ample collection of Hox direct targets requiring temporal maintenance in r4.

We discussed how formation of a Hox-Pbx-Prep ternary complex results in transcriptional cooperation and contributes to overcome insufficient activation at the low-affinity PH site(Fig. 3A). However, as Hox, Pbx and Prep factors are present throughout r4(Ferretti et al., 1999; Popperl et al., 1995; Schnabel et al., 2001), this model cannot solely explain how Phox2b expression is sharply restricted in ventral r4 to the pMNv progenitor domain. Our results indicate that cooperation with Nkx2.2 is an additional component of the regulation of the Phox2b enhancer.

How does Nkx2.2 contribute to the Phox2b enhancer regulation?First, Nkx2.2 binding to the Phox2b enhancer is not required(Fig. 6A; data not shown). Second, transcriptional activation by Hox and co-factors is further enhanced by the activity of Nkx2.2 as a repressor(Fig. 6A). Third, an intact PH site is an essential component of the Hox and Nkx2.2 cooperation on the Phox2b enhancer (Fig. 6B). Moreover, a trimerized PH site is sufficient to respond to Nkx2.2 activity in the presence of Hox factors, both in P19 cells and chick hindbrain (Fig. 6C,F,G). Nkx2.2 activity is mediated in part through association with the Gro/TLE class of co-repressors, as deletion of the TN interacting domain impairs Nkx2.2 activity on the Phox2b enhancer(Fig. 6C). One possibility is that Nkx2.2/Gro could transcriptionally repress, or sequester, a putative repressor (R) normally bound at, or in the vicinity of, the PH site(Fig. 6H). In the absence of Nkx2 proteins, i.e. dorsal to the pMNv domain, R could prevent the formation of a Hox-Pbx-Prep ternary complex and consequently the activation of high Phox2b expression levels. Within the pMNv progenitor domain, the presence of Nkx2.2 would relieve repression by blocking the activity or the expression of R acting on the Phox2b enhancer(Fig. 6I). After recruitment of Hoxb1 or Hoxb2 by Pbx and binding of Prep1 to the P/M site(Ferretti et al., 2000; Jacobs et al., 1999), a ternary complex would then form and stimulate high levels of transcription(Fig. 6I).

Nonetheless, in the chick hindbrain reporter expression driven by the Phox2b enhancer was not restricted in a columnar pattern(Fig. 3C), unlike endogenous Phox2b. Thus, although the PH and P/M sites embedded within the 376 bp enhancer are required for ventral r4 regulation (Figs 3, 5), additional inhibitory inputs from regulatory regions outside the enhancer are also needed to achieve columnar regulation. In this respect, ventral restriction of reporter expression is obtained with the 2.8 kb construct(Fig. 1C). Therefore, the proposed repressor (Fig. 6H),although required, may not be sufficient to restrain the Phox2benhancer activity outside the pMNv domain, when the enhancer is tested in isolation from its genomic context. Although an important site of integration of AP and DV regulatory inputs, the Phox2b enhancer may require interaction with distant regulatory elements for precise columnar regulation.

In conclusion, our results take a first significant step in understanding how the transcriptional activity of repressors and activators converges on a specific target gene promoter to direct expression in a specific progenitor domain in the mammalian central nervous system.

We thank Jean-François Brunet, Yasunori Murakami, Franck Oury,Massimo Pasqualetti and Chris Rose for discussions and critical reading of the manuscript. We are grateful to Elisabet Andersson for plasmids, M. Poulet for excellent technical assistance and M. Kmita for gift of lacZ probe. O.A.S. was supported by La Ligue Nationale Contre Le Cancer and Association pour la Recherche sur le Cancer (ARC). M.J.G. was supported by the Luxembourg Ministère de la Culture, de l'Enseignement Supérieur et de la Recherche. F.M.R. was supported by grants from the ARC, Ministère de la Recherche (ACI program), and institutional funds from CNRS, INSERM and Hôpital Universitaire de Strasbourg. F.M.R., C.G. and J.E. were supported by the EEC `Brainstem Genetics' Program (#QLG2-CT01-01467). V.Z. was supported by the Italian Association for Cancer Research (AIRC) and Italian Telethon.