Ongoing roles of Phox2 homeodomain transcription factors during neuronal differentiation

During the proliferation of neuroblasts and the differentiation of neurons, many developmental transcription factors (TFs) are expressed in defined time-windows, the functional significance of which is largely unknown. Indeed, simple knockouts – from which, most often, we infer the actions of developmental TFs – cannot reveal the phases of the expression window during which the inferred actions are carried out. In case of cell death or fate switch, they altogether obscure later actions. Documented cases where a neuronal TF, first required in progenitors or early postmitotic precursors, also carries later roles, are few and far between. They include: Nkx6.1, the forced maintenance of which in Nkx6.1^OFF somatic motoneurons endows them with the axonal projections of their Nkx6.1^ON counterparts (De Marco Garcia and Jessell, 2008), indirectly arguing for a post-mitotic role executed in the latter; Nkx2.1, the conditional inactivation of which shows that, after determining the fate of progenitors in the medial ganglionic eminence (Sussel et al., 1999), it guides the migration of some of their post-mitotic progeny (Nobrega-Pereira et al., 2008); Isl1, which has unique early and late embryonic roles in primary sensory neurons (Sun et al., 2008); Nurr1, the inactivation of which in the adult causes a loss of dopaminergic neurons (Kadkhodaei et al., 2009); and Gata3, which keeps an anti-apoptotic role in sympathetic ganglionic cells throughout embryonic and adult life (Tsarovina et al., 2010).

The homeobox gene Phox2b and its paralog Phox2a are switched on in most classes of visceral neurons, starting either at the progenitor stage or soon after exit from the cell cycle (Brunet and Pattyn, 2002), and stay expressed in differentiating precursors for various lengths of time, depending on the cell type. The constitutive knockout of Phox2b leads to a rapid demise of most Phox2b-positive neuronal precursors, and that of Phox2a to the agenesis of the locus coeruleus (LC) and of oculomotor and trochlear motoneurons (Brunet and Pattyn, 2002), thus precluding assessment of all but their earliest roles. These roles, in the branchiomotor, visceromotor (BM/VM) and noradrenergic neurons (the focus of the present study), can be summarized as follows.

In the domain of the hindbrain ventricular zone (VZ), which gives rise to BM/VM neurons, Phox2b is expressed during the production of these neurons, then downregulated, which allows the production of serotonergic (5-HT) neurons (Pattyn et al., 2003a). In progenitors, Phox2b blocks the 5-HT fate while specifying the BM/VM fate (Pattyn et al., 2000b; Pattyn et al., 2003a), and promotes the exit of neuroblasts from the cell cycle by acting like a proneural gene (Dubreuil et al., 2000; Pattyn et al., 2004). Thereafter, Phox2b stays on in post-mitotic BM/VM precursors for periods that range from a couple of days (e.g. in trigeminal motoneurons, E.C. and J.-F.B., unpublished) to the lifetime of the animal [e.g. in vagal motoneurons (Kang et al., 2007) for rat] and nothing is known of the potential roles of this second phase of expression. Phox2a is induced in early postmitotic BM/VM neurons but its knockout has no phenotype in these cells.

In the precursors of sympathetic neurons, Phox2b is expressed as soon as they aggregate at the dorsal aorta and before all sympathoadrenergic markers, including the transcription factors Hand2, Gata2/3, Insm1 and Phox2a, and the effector genes tyrosine hydroxylase (Th), dopamine-β-hydroxylase (Dbh), Scg10 (Stmn2 – Mouse Genome Informatics), peripherin (Prph) and β-III tubulin (Pattyn et al., 2006; Wildner et al., 2008). The knockout of Phox2b prevents sympathoblasts from switching on any of these markers and kills them, leading to agenesis of the sympathetic chain (Pattyn et al., 1999). Again, the significance of the continued expression of Phox2b in sympathetic ganglia, which persists at least up to late embryonic stages (Hendershot et al., 2008) (E.C., C. Goridis and J.-F.B., unpublished data), is unknown. Phox2a is also expressed at least throughout embryogenesis, but its knockout has only a weakly penetrant morphological phenotype in the superior cervical ganglion (Morin et al., 1997).

Finally, in the main noradrenergic center of the hindbrain, the LC, Phox2a is expressed first and required for the expression of Phox2b, which in turn is required for the expression of Dbh, before being sharply downregulated, while Phox2a stays on for the life-time of the animal (Pattyn et al., 2000a). Again, nothing is known of the role of the ongoing expression of Phox2a.

We set out to determine whether the persistent expression of Phox2 genes has any function in noradrenergic and BM/VM neurons by inactivating the genes after their onset of expression. We show that this leads in all cell types to the downregulation of a whole set of differentiation markers after an initial spike of expression, and that cell behaviors such as caudal migration of the facial motor neurons are severely disrupted. In contrast to what is observed in the constitutive knockouts, in noradrenergic cells, delayed inactivation of one of the two paralogs can be partially (sympathetic chain) or fully (LC) compensated by the persistent expression of the other – the full phenotype being revealed by the conditional inactivation of both genes. Together, the results show a requirement for Phox2 genes beyond the initial phase of neuronal specification previously revealed by the simple knockouts.

The mutant lines used in this study are the conditional Phox2a (Phox2a^lox/lox) and Phox2b (Phox2b^lox/lox) lines (Coppola et al., 2010; Dubreuil et al., 2009), and the Isl1^Cre (Srinivas et al., 2001) and the CMV-βactin-Cre-ERT2 lines (Santagati et al., 2005) (hereafter designated as CreERT2) lines. Mice were mated overnight and noon of the day of the vaginal plug was considered to be E0.5. Embryos were dissected out in PBS and fixed at 4°C in 4% paraformaldehyde in PBS. All animal studies were carried out in accordance with the guidelines issued by the French Ministry of Agriculture and have been approved by the Direction départementale des services vétérinaires de Paris.

Tamoxifen (Sigma, USA) was dissolved (20 mg/ml) in pre-warmed corn oil (Sigma) and stored at 4°C. It was administered to pregnant females by intraperitoneal injections. Females were treated for 2 consecutive days (E13.5 and E14.5), twice a day, using 2 mg of tamoxifen in the morning and 1 mg in the afternoon. The embryos were collected at gestational day (E) 18.5.

BrdU (Sigma, USA) was injected intraperitoneally into pregnant mice (100 μg/g body weight) 1 hour before dissection.

Fixed embryos or embryo tissues were cryoprotected overnight in 20% sucrose in PBS and embedded in OCT. Sections (12 μm) were cut in the transversal or sagittal planes. The methods for single in situ hybridization or immunostaining or the combination of both have been described previously (Tiveron et al., 1996). All probes were synthesized using the DIG RNA labeling kit (Roche) as specified by the manufacturer. The primary antibodies used were as following: rabbit anti-Phox2a (1/500) (Tiveron et al., 1996), rabbit anti-Phox2b (1/500) (Pattyn et al., 1997), guinea pig anti-Phox2b (1/500) (Dubreuil et al., 2009), mouse anti-Isl1/2 (1/100) (40.2D6 and 39.4D5, Developmental Study Hybridoma Bank, Iowa city, USA), rabbit anti-Th (1/200) (Chemicon, USA), mouse anti-βIII tubulin (1/200) (Sigma, USA) and rat anti-BrdU (AbCam, Cambridge, UK). The primary antibodies were revealed for bright field observation by biotin-labeled secondary antibodies using the Vectastain ABC kit, or for fluorescent observation using secondary antibodies of the appropriate specificity labeled with Cy3 or Cy5 (Jackson ImmunoResearch Laboratories, West Grove, USA) or Alexa 488 (Invitrogen, Oregon, USA). For whole-mount in situ hybridization, fixed embryos were dehydrated in ethanol in PBS-0.1% Tween (from 25% up to 100%) after fixation in 4% PFA. The method was as described by Tiveron et al. (Tiveron et al., 1996).

Apoptotic cells were detected on cryosections by using the Apoptag in situ apoptosis detection kit (Millipore).

Single crystals of the carbocyanine tracer dyes DiI (1,1′-dioctade-cyl-3,3,3′,3′-tetramethylindocarbocyanine perchlorate) or DiA [4-(4-dihexadecylaminostyryl)-N-methylpyridinium iodide (Molecular Probes, Invitrogen, Oregon, USA)] were applied in the first or second branchial arch of E11.5 embryos previously fixed in 4% buffered paraformaldehyde. The dye was left to diffuse for 1 week at room temperature in 4% paraformaldehyde, then the embryos were dissected and the hindbrains photographed under fluorescent light.

Isl1/2-positive cells were counted in rhombomere 4 (r4) on one section out of five on each side of three mutant and wild-type embryos.

To evaluate the effect of the different genotypes on the LC, the number of Th-immunoreactive cells, as well as the number of those co-immunoreactive for Phox2a or Phox2b, were counted. Measurements were made on one 12 μm transverse section out of four, throughout the LC. Both sides of three control, three Phox2a/b^CreERT2-KO, and two Phox2a^CreERT2-KO fetuses were analyzed.

Quantitative analysis of the stellate ganglion was carried out by measuring the area enclosed in the outline of the βIII-tubulin in situ hybridization signal corresponding to the ganglion, on one section out of four throughout the ganglion, using the Image J software. Both ganglia of six controls (12 ganglia), three Phox2b^Islet-CKO (six ganglia) and three Phox2a/b^Islet-CKO mutants (six ganglia) were measured. For BrdU analysis, BrdU-positive cells were counted on one section out of four throughout one ganglion in three embryos for each genotype (wild-type and Phox2a/b^Islet-CKO mutants), within the limits defined by the βIII-tubulin immunofluorescence signal, and the number divided by the corresponding surfaces.

Means were compared using Student's t-test. Equality of variances was analyzed with an F-test and Welch's correction was employed when variances of populations was significantly different.

To uncover a possible role for Phox2b after progenitor specification in BM/VM motoneurons, we conditionally inactivated the gene by partnering a floxed Phox2b allele (Dubreuil et al., 2009) with a Cre recombinase driven by the promoter of Isl1 (Srinivas et al., 2001), a homeobox gene switched on after cell-cycle exit in BM/VM motoneuron precursors (Ericson et al., 1992; Varela-Echevarría et al., 1996). In Isl1^Cre; Phox2b^lox/lox embryos (hereafter designated as Phox2b^Islet-CKO), at embryonic day (E) 16.5, in situ hybridization for peripherin, an intermediate filament gene, on transverse sections of the hindbrain revealed only a few scattered cells at the place where the facial (nVII) and trigeminal (nV) motor nuclei are found in the wild type (Fig. 1). Thus, continued expression of Phox2b past the progenitor stage is required for the development of the facial and trigeminal motor nuclei.

To elucidate the mechanism for the abnormal development of facial and trigeminal motor nuclei, we examined earlier stages of ontogeny. For most purposes, we focused our study on facial motoneurons, which are easier to track, owing to their abundance and protracted period of generation [E10-E13 (Taber Pierce, 1973)] from the ventral-most progenitor domain of the fourth rhombomere (r4) (Pattyn et al., 2000b). At E11.5, in r4 of Phox2b^Islet-CKO embryos, Phox2b protein was present in facial BM progenitors of the VZ – as expected, because they have not expressed Isl1 yet – and in the post-mitotic Isl1/2⁺ (hereafter Isl1⁺) precursors closest to VZ (presumably the youngest, which have just switched on Isl1). However, it was lost from the Isl1⁺ precursors further away from the VZ (Fig. 2A,A′), which were presumably born earlier. This rapid and complete disappearance of the Phox2b protein after expression of the Isl1^Cre transgene attests the efficacy of action of the Cre recombinase, and the short half-life of both Phox2b protein and mRNA in that cell type.

We then assessed the consequence of the secondary loss of the Phox2b protein on the expression of a battery of markers for facial BM neuronal precursors: the TFs Tbx20 (Dufour et al., 2006), Tbx2 (Song et al., 2006), Ebf1 (Garel et al., 2000), Nkx6.1 (Pattyn et al., 2003b), the intermediate filament peripherin and the pan-neuronal microtubule protein βIII-tubulin. These markers fell into two categories according to their expression in the mutants: those whose expression was either abrogated (Tbx2) or, after an initial spike in early post-mitotic precursors, abruptly downregulated, alongside the disappearance of Phox2b protein (Ebf1, Tbx20) (Fig. 2B-D′); and those that were maintained in the mutant at their wild-type levels – Nkx6.1, βIII-tubulin and peripherin (Fig. 2E-F′; not shown), and Isl1 (Fig. 2A,D′). Thus, there is a transcriptional program downstream of Phox2b that requires continuous input from Phox2b for its implementation.

Among the potential downstream genes of Phox2b is Phox2b itself, which might even be a direct transcriptional target owing to three auto-regulatory sites validated by transient transfection assays (Cargnin et al., 2005). To test whether ongoing expression of Phox2b relies on an auto-activation loop, we used a Phox2b RNA probe, which covers the first exon and should thus detect both the native and recombined alleles. In Phox2b^Islet-CKO, a Phox2b mRNA was detectable in facial motoneuronal precursors at E11.5 (Fig. 3A-B′) and still at E13.5 (Fig. 3C-D′, arrowheads), albeit at a lower level than in wild type. The signal is unlikely to come from a long-lived pre-recombination full-length mRNA, as the protein that such an intact mRNA would produce disappears within hours of Isl1^Cre expression. Therefore, the maintenance of expression of Phox2b does not entirely depend on auto-activation in BM/VM neurons during the time scale investigated.

Finally, we checked whether premature extinction of Phox2b redirected BM/VM post-mitotic precursors towards the fate of 5-HT neurons, as it does for the dividing progenitors of the p3 neuroepithelial domain, common to BM/VM and 5-HT neurons (Pattyn et al., 2003a). The expression of Pet1, a determinant of 5-HT differentiation, was unchanged in the mutants (see Fig. S1 in the supplementary material). Thus, the 5-HT fate, which is accessible to BM/VM progenitors if Phox2b is removed (Pattyn et al., 2003a), is no longer available to post-mitotic BM/VM precursors.

Two non-mutually exclusive mechanisms could account for the massive atrophy of the facial nucleus in Phox2b^Islet-CKO embryos at E16.5: depletion by cell death and abnormal migration. Indeed, soon after their birth, facial BM precursors migrate caudally to r6 along the edge of the floor plate, then radially in r6 to form the facial nucleus close to the pial surface (Auclair et al., 1996). This migration is disrupted, to various extents, in a number of mutant backgrounds, including the knockouts of the transcription factors Tbx20 and Ebf1 (Garel et al., 2000), which both happen to be downregulated in Phox2b^Islet-CKO embryos (Fig. 2). To follow the migration of FBM precursors in Phox2b^Islet-CKO embryos, we first used immunofluorescence against Isl1 proteins on serial transverse sections of the hindbrain. At E11.5, in wild-type embryos, many facial BM precursors had colonized r5 and some had reached r6, commencing their radial migration towards the pia. In Phox2b^Islet-CKO, more neurons than in the wild type were found in r4 (Fig. 4A,A′,F), fewer neurons had reached r5 (Fig. 4B,B′) and practically none were radially migrating in r6 (Fig. 4C,C′). We confirmed this migration block on flatmounted hindbrains retrogradely labeled from the main branch of the facial nerve by DiI (Fig. 4D,D′). Simultaneous labeling of the trigeminal nerve with DiA showed that the migration of trigeminal BM precursors, which proceeds dorsally in r2, was slowed down (Fig. 4D,D′). To rule out the possibility that these migration defects are a mere delay that is later compensated for, we hybridized a whole-mount preparation of the hindbrain at E13.5 with a peripherin probe. Although the majority of nVII precursors had reached their final destination in the wild-type embryos, their bulk was still stuck in r4 and r5 in Phox2b^Islet-CKO embryos. The anlage of the trigeminal nucleus appeared smaller and scattered along the migratory path (Fig. 4E,E′). Thus, the caudal migration of facial BM precursors and dorsal migration of trigeminal BM precursors is disrupted in Phox2b^Islet-CKO mutants. Although this defective migration is sufficient to explain the agenesis of the respective nuclei, it does not rule out the concomitant occurrence of cell death. The lack of proper nucleogenesis made it difficult to count the cells. Staining for apoptotic cells by the TUNEL method did not reveal excess cell death in the area occupied by facial precursors, although their lack of histological coherence could make it difficult to detect.

Another Phox2b-dependent structure that was altered in Isl1^Cre; Phox2b^lox/lox embryos was the sympathetic chain, in which Isl1 is switched on concomitantly with Phox2a (see Fig. S2 in the supplementary material) and noradrenergic differentiation. In E16.5 Phox2b^Islet-CKO embryos, the sympathetic chain, the outline of which could be detected by in situ hybridization for βIII-tubulin, was 64% atrophic at the level of the stellate ganglion (Fig. 5A,B,D). Together with Phox2b, sympathetic ganglion cells express its paralogous gene, Phox2a, which has an identical homeodomain. We found that Phox2a, although strictly under the control of Phox2b for its onset of expression (Pattyn et al., 1997), is maintained at a normal level in Phox2b^Islet-CKO at E13.5 (see Fig. S3 in the supplementary material), showing that Phox2a secondarily escapes dependence on Phox2b, whether through auto-activation, or another mechanism. We thus tested the possibility that Phox2a could partially compensate for the deletion of Phox2b by conditionally inactivating both genes. [No such compensation was expected for facial or trigeminal BM motoneurons, since they normally downregulate Phox2a soon after its onset of expression (see Fig. S3 in the supplementary material).] In Isl1^Cre; Phox2a^lox/lox; Phox2b^lox/lox (hereafter designated as Phox2a/b^Islet-CKO) embryos, the atrophy of the stellate ganglion was significantly more pronounced than in Phox2b^Islet-CKO embryos [∼77% (Fig. 5A,C,D)]. This shows that the Phox2a protein, although unable to replace Phox2b at the onset of sympathoadrenergic differentiation in vivo (Coppola et al., 2005) can compensate for the absence of Phox2b to a moderate extent, at later stages of sympathetic development. To be sure to evaluate the full extent of late Phox2 function, we mostly restricted our subsequent study to double Phox2a/b^Islet-CKO mutants.

The atrophy of the sympathetic chain could be due to apoptosis or decreased cell proliferation. To assess these possibilities, we analyzed Phox2a/b^Islet-CKO embryos at E13.5, a time point intermediate between the onset of ganglion formation and the hypoplasia observed at E16.5. We could not detect any apoptotic cell in the anlage of the stellate ganglion of E13.5 Phox2a/b^Islet-CKO embryos (see Fig. S4 in the supplementary material). This was unexpected, given that sympathetic precursors die in large numbers at around E11.5 in Phox2b constitutive knockouts (Pattyn et al., 1999). This shows that after an initial spike of activity, Phox2b becomes dispensable for cell survival. However, BrdU incorporation was decreased by 44% in the sympathetic ganglia of Phox2a/b^Islet-CKO embryos (Fig. 5E-G), which is likely to account for the eventual atrophy. Thus, Phox2b expression promotes sympathoblast proliferation during embryogenesis.

Sympathoblasts derive from neural crest cells, and form a chain by aggregating alongside the dorsal aorta. Soon after aggregation, ganglion cells switch on Phox2b. In Phox2a/b^Islet-CKO embryos at E10.5, Phox2b was, as expected, normally expressed at E10.5, and every Phox2b-positive ganglionic cell had switched on a whole battery of markers downstream of it (see Introduction) at a normal level: Phox2a, Isl1, Hand2, Gata3, Insm1, βIII-tubulin and the effectors of the noradrenergic phenotype Dbh and Th (Fig. 6; data not shown). Thus, in Phox2a/b^Islet-CKO, the full program of cell-type and generic neural differentiation is triggered in sympathetic progenitors. At E11.5, the Phox2b protein was absent from most sympathoblasts, showing that, as in BM/VM neurons, the recombination of the floxed Phox2b locus by Isl1^Cre and the disappearance of the corresponding protein and mRNA takes 1 day at most (see Fig. S5A,A′ in the supplementary material). Unexpectedly, Phox2a protein was still detected (see Fig. S5B,B′ in the supplementary material), owing either to its longer half-life or to the lesser accessibility of the locus to recombination. Other markers were still expressed at levels only slightly diminished compared with wild type, including Hand2, Gata3, Dbh or Th (see Fig. S5C-F′ in the supplementary material) and the Phox2b locus itself (see Fig. S6 in the supplementary material).

By contrast, at E16.5 (at which stage Phox2a protein was still detectable but only in a minority of neurons, not shown), most markers tested were downregulated to undetectable levels in the vast majority of ganglion cells, including Gata3, Dbh and peripherin, with the notable exception of βIII-tubulin (Fig. 5), Hand2 and Ret (Fig. 7) which stayed on in most ganglionic cells. In single Phox2b^Islet-CKO embryos, noradrenergic marker expression was only marginally higher than in the double knockout (see Fig. S7 in the supplementary material), showing that Phox2a has only a modest capacity to enforce the maintenance of the noradrenergic phenotype. The sparse expression of vasoactive intestinal polypeptide (Vip) and VAChT (Slc18a3 – Mouse Genome Informatics) in the stellate ganglion was unchanged by the Phox2a/b^Islet-CKO mutation (see Fig. S8 in the supplementary material). The Phox2b locus was expressed but markedly weaker than in the wild type at E13.5 (see Fig. S6 in the supplementary material). Altogether, these data show that part of the differentiation program of sympathetic ganglionic cells requires continuous input from Phox2 TFs for maintenance.

Another group of noradrenergic cells that differentiate under the control of Phox2 genes is the LC (Pattyn et al., 2000a). Here, the dynamic of expression and the roles of Phox2 genes diverge from those in sympathoblasts (see Introduction): the expression of Phox2a both precedes and outlasts that of Phox2b, which is switched on by Phox2a, then downregulated. We therefore asked whether, in this case, Phox2a instead of Phox2b maintains the noradrenergic phenotype during late gestation. We inactivated Phox2a while circumventing its early requirement in LC development, by crossing the Phox2a^lox/lox allele in a CreERT2 background (Santagati et al., 2005), producing a Phox2a^CreERT2-KO progeny, and examined the brains of E18.5 pups whose mother was treated with tamoxifen from E13.5 to E14.5. A LC of normal size and location was present in Phox2a^CreERT2-KO embryos on the basis of Dbh (Fig. 8A,A′) and Th expression (Fig. 8B,B′), despite the fact that, as expected, Phox2a was lost from 67% of LC cells (Fig. 8C). The fact that some LC cells had kept detectable levels of Phox2a is probably due to the mosaic action of the Cre recombinase, evident in other parts of the brain (not shown) – another, non mutually exclusive explanation being a long half-life of the Phox2a protein or mRNA. Unexpectedly, Phox2b, which, by this time, was completely extinguished from the LC of wild-type embryos (Fig. 8B, inset), was maintained in the LC of the mutants, specifically in those cells that had lost Phox2a expression (Fig. 8B′, inset). This shows that the normal extinction of Phox2b at E13.5 (Pattyn et al., 2000a) results from an active repression by Phox2a, which is relieved in inducible Phox2a mutants. To circumvent the possibility that this upregulation of Phox2b compensates for the loss of Phox2a, we examined the fate of the LC in a CreERT2; Phox2a^lox/lox; Phox2b^lox/lox (hereafter designated as Phox2a/b^CreERT2-KO) background. In contrast to Phox2a^CreERT2-KO, in Phox2a/b^CreERT2-KO embryos treated with tamoxifen at E13.5-E14.5 and examined at E18.5, the LC was markedly atrophic, as judged from Dbh (Fig. 8A′) and Th expression (Fig. 8B′), the number of Th+ cells being reduced by 58% (Fig. 8D). All Th-expressing cell spared in Phox2a/b^CreERT2-KO embryos were Phox2a positive and Phox2b negative (Fig. 8B′, inset). The most likely explanation is that, in Phox2a/b^CreERT2-KO mutants, recombination is mosaic and LC cells are either recombined for both genes (and lose Th and Dbh expression or die) or recombined for neither, in which case Phox2b is repressed by Phox2a, as in the wild type. Owing to a lack of additional markers for LC cells, which are embedded in neuronal tissue, we could not determine whether the depletion of Th⁺ cells reflects the loss of Th expression or loss of the cells themselves. Altogether, these data show that ongoing expression of a Phox2 gene is required for the survival of LC cells and/or the maintenance of their noradrenergic phenotype, at least until late gestation. Phox2a normally plays that role, but Phox2b can compensate for its loss.

In this study we have investigated whether the homeodomain proteins Phox2b and Phox2a were still required once they have launched the program of neuronal differentiation of BM/VM neurons and noradrenergic neurons. We find that many, but not all, of the genes that lie downstream of Phox2b in these cell types (based on gain and loss-of-function studies), including transcription factors and effector genes, are downregulated if Phox2b and Phox2a are prematurely silenced by time-controlled recombination. These results illustrate a case where the differentiation program downstream of a neuronal fate determinant is not readily locked in an `on' mode, but requires continuous transcriptional input. The source of this input has been conceptualized (Hobert, 2008) under the term `terminal selector gene', of which Phox2b might be an example in vertebrates.

The constitutive knockout of Phox2b had revealed that the BM neuronal fate depends on Phox2b, in the sense that BM precursors are not born from Phox2b^KO ventral progenitors. However, the transcriptional architecture of the BM differentiation program is not known. The conditional inactivation of Phox2b soon after exit from the cell cycle reveals two subsets of BM precursor markers with respect to Phox2b-dependence: one that has an ongoing requirement for Phox2b (including Tbx20, Tbx2, Ebf) and one that has not (Nkx6.1, Isl1, peripherin and βIII-tubulin). Genes of the second subset could either never depend on Phox2b in the first place, or emancipate themselves from this dependence after cell cycle exit. The first hypothesis is likely for Nkx6.1, the expression of which in progenitors is not affected in Phox2b constitutive knockouts (Pattyn et al., 2000b). As to the other genes that are maintained after Phox2b inactivation, they tend to be less specific than those that are downregulated, expressed as they are in all neurons (βIII-tubulin), all motoneurons (Isl1) or all neurons that project to or are situated in the periphery (peripherin).

Consistent with the downregulation of Tbx20, the migration of motoneurons is blocked in a manner comparable with that of Tbx20 knockouts – the slightly less dramatic phenotype of the Phox2b^Islet-CKO embryos being attributable to a transient expression of Tbx20. This shows that the machinery for facial motoneuronal migration – which mobilizes the planar cell polarity pathway (Carreira-Barbosa et al., 2003; Jessen et al., 2002; Vivancos et al., 2009), and involves Sema3A (Schwarz et al., 2004) and the chemokine receptor CXCR4 (Cubedo et al., 2009), but is still largely elusive – requires continuous input from Phox2b.

Downstream of Phox2b, several TFs [among which are Gata3, Hand2 (reviewed by Apostolova and Dechant, 2009) and, more recently, Insm1 (Wildner et al., 2008)] have been implicated in the noradrenergic differentiation of sympathetic ganglion neurons by constitutive knockout. Among those, Hand2 was recently found to be required also for the maintenance of noradrenergic traits (Schmidt et al., 2009): knockdown of Hand2 in cultured post-mitotic sympathetic neurons, which express both Th and Dbh, led to the downregulation of both genes – which require Hand2 also for their onset of expression (Hendershot et al., 2008; Lucas et al., 2006; Morikawa et al., 2007). Moreover, in vivo inactivation of Hand2 with a Dbh::Cre, i.e. after the onset of noradrenergic differentiation, resulted in the secondary extinction of Th, demonstrating a role for Hand2 in maintaining Th expression, already suggested from a study on parasympathetic ganglia (Muller and Rohrer, 2002). Our results establish that, like Hand2, the ongoing expression of Phox2b is required for noradrenergic differentiation to be maintained. They also indicate that Hand2 is not sufficient for the maintenance of the noradrenergic phenotype, as the expression of Hand2 after dual conditional inactivation of Phox2a and Phox2b, largely exceeded that of noradrenergic differentiation traits. The simplest scenario is that, in order to maintain Th and Dbh expression, Phox2b functions in tandem with Hand2, which was switched on by Phox2b in the first place (Brunet, 2008; Goridis and Brunet, 1999), in a `feedforward' transcriptional loop (Shen-Orr et al., 2002). The proposed interactions of the Dbh promoter with both Phox2 and Hand2 proteins (Rychlik et al., 2005; Seo et al., 2002; Xu et al., 2003), is compatible with such a transcriptional logic. Similarly, the decrease in cell division that we report in Phox2a/b^Islet-CKO implies that the proliferative role documented for Hand2 (Hendershot et al., 2008) is also exerted in tandem with Phox2b. The same logic might apply to Gata3, which is switched on by Phox2b, is required for Dbh expression (Apostolova and Dechant, 2009; Goridis and Brunet, 1999) and associates with the promoter of that gene, albeit indirectly (Hong et al., 2008). Finally, this type of molecular partnership might also underlie the dual ongoing requirement for Gata3 (Tsarovina et al., 2010) and Phox2b (this study) for sympathoblast proliferation.

The comparison between wild type, Phox2b^Islet-CKO and Phox2a/b^Islet-CKO reveals that Phox2b is required for the maintenance of Dbh expression at their wild-type levels in sympathoblasts, and that Phox2a can only poorly compensate for Phox2b in this role (see Fig. S7 in the supplementary material). This stands in contrast with the LC in which Phox2b, although essential for the onset of Dbh expression, is later downregulated (Card et al., 2010; Pattyn et al., 2000a) while noradrenergic traits are maintained throughout life. Phox2a is the only Phox2 gene expressed at later stages in the LC and, as we show here in Phox2a/b^CreERT2-KO, it ensures the maintenance of the noradrenergic phenotype. This situation might also prevail in the other noradrenergic centers of the hindbrain, A1-A5 (Card et al., 2010) (and data not shown). This differential capacity of Phox2a to maintain noradrenergic differentiation in sympathetic versus LC neurons possibly reflects a fundamental difference between the central and peripheral noradrenergic neurons: in central neurons, noradrenergic differentiation is strictly post-mitotic, whereas in sympathoblasts, it occurs concurrently with cell division (Rothman et al., 1980). Maintenance of gene expression is likely to have different requirements in a post-mitotic cell than across cell division, making it conceivable that, in the same way as Phox2b (but not Phox2a) can trigger Dbh expression in vivo (Coppola et al., 2005), Phox2b (but not Phox2a) can ensure the reset of expression after each cell cycle. The same type of explanation could apply to peripherin, the levels of which drop sharply in sympathoblasts after dual inactivation of Phox2a and Phox2b, but are maintained in post-mitotic BM precursors at least until E16.5 (Fig. 4E and not shown) in the absence of both genes.

Although we show that proper differentiation of BM and sympathetic ganglion neurons requires ongoing expression of Phox2b, we do not know how this ongoing expression itself is implemented. An auto-activatory loop is an intuitive solution for maintaining the expression of TFs beyond the initial conditions that switch them on in the first place. The requirement for such loops, proposed as a key feature of `terminal selector genes' (Hobert, 2008), has rarely been directly demonstrated in vivo (e.g. Baumgardt et al., 2007; Kadkhodaei et al., 2009; Way and Chalfie, 1989), and most often inferred from gain of function experiments. It seemed an attractive possibility for Phox2b, owing to the existence of binding sites for the protein, which are active in vitro (Cargnin et al., 2005) and the fact that forced expression of mouse Phox2b in chicken embryos induces the endogenous gene (Dubreuil et al., 2002). Accordingly, in the sympathetic chain and BM neurons the levels of Phox2b mRNA are markedly reduced at around E13.5 and beyond. However, they remain detectable, and earlier on (at E11.5), the levels of Phox2b mRNA corresponding to the transcription of the newly invalidated locus are comparable with those of the wild type in the absence of Phox2 proteins, pointing to an unexpected degree of independence from auto-activation.

Like many TFs, Phox2a and Phox2b are eventually downregulated in several classes of neurons, including BM/VM neurons. This feature departs from the proposed property of `terminal selector genes' (largely inferred from invertebrate examples) of being expressed throughout the life of the neurons (Hobert, 2008), and might be specific to vertebrates. Once again, nothing is known about the physiological importance of this phenomenon. In BM/VM neurons, downregulation occurs as early as E11 in trigeminal motoneurons and during the first postnatal weeks in the facial motor nucleus. In the LC, where Phox2b is abruptly extinguished at E13.5, we have shown that Phox2a is required for this extinction, possibly via the autoregulatory sites in the promoter of Phox2b (Cargnin et al., 2005): Phox2a and Phox2b have identical DNA-binding domains. Such a discrete mechanism suggests developmental relevance. For example, a specific amount of Phox2 protein might be crucial for LC maintenance, and insured by repression of Phox2b by Phox2a. Alternatively, Phox2b might have different properties than Phox2a, deleterious to the maintenance of LC cells. Abnormally maintaining Phox2b expression [as occurs in Phox2a^CreERT2-KO (this study) or in Phox2a^KIPhox2b(Coppola et al., 2005)] rescues LC cells, which rules out gross toxicity, but subtle or later anomalies in their projections or physiology remain a possibility. More generally, monitoring the consequences of the maintenance of developmental TFs beyond their normal window of expression might reveal important aspects of the transcriptional control of neuronal differentiation.

We thank A. Kispert, S. Garel, M. Strehle, J. L. Rubenstein, M. Wegner, M.-M. Portier, C. Ragsdale, P. Cserjesi, D. Engle, V. Pachnis and T. Jessell for probes. This work was supported by grants from the Agence Nationale de la Recherche and Fondation pour la Recherche Médicale (to J.-F.B) and by institutional support from the Centre National de la Recherche Scientifique.