Hoxb1 controls effectors of sonic hedgehog and Mash1 signaling pathways

The vertebrate hindbrain coordinates multiple complex functions, including somatic and visceral motor activities and the processing of sensory information. Formation of the neural circuits underlying these functions depends on the generation of distinct populations of neuronal subtypes that constitute these circuits (Altman and Bayer, 1980; Carpenter and Sutin, 1983; Paxinos, 1995; Ramón y Cajal, 1995). The molecular and cellular mechanisms that spatially and temporally specify these neurons are beginning to be identified. This advance has largely been made possible by assigning the plethora of neurally expressed transcription factors and receptor-ligand signaling systems to coherent molecular pathways that control the formation of the neuronal subtypes (Lumsden, 1990; Tanabe and Jessell, 1996; Kageyama and Nakanishi, 1997; Flanagan and Vanderhaeghen, 1998; Sasai, 1998; Edlund and Jessel, 1999).

Genes belonging to the Hox complex constitute one component of this molecular network. Gain- and loss-of-function analyses have identified Hox genes that are important to the regional specification of the hindbrain into compartmental units called rhombomeres (r) (Lumsden and Keynes, 1989; Lumsden and Krumlauf, 1996; Capecchi, 1997). Hox genes have been implicated both in the process of rhombomere formation as well as in the subsequent specification of cell identities within rhombomeres. For example, while loss-of-function mutations in Hoxa1 result in the failure to form specific rhombomeres, disruption of Hoxb1 leads to a failure to specify distinct motoneurons within r4 (Carpenter et al., 1993; Mark et al., 1993; Goddard et al., 1996; Studer et al., 1996; Pata et al., 2000). Gain-of-function experiments have also been informative. For example, ectopic expression of either Hoxb1 or Hoxa2 in r1 results in the apparent mis-specification of neurons within this non-Hox-expressing rhombomere to acquire characteristics normally associated with r4 or r2 branchiomotor neurons, respectively (Bell et al., 1999; Jungbluth et al., 1999). Although the data linking Hox genes to hindbrain neurogenesis is compelling, the ways in which these genes interact with the other molecular pathways that define neuronal subtypes has not been delineated.

We have examined the interaction of the Hoxb1 mutation with two principal molecular pathways mediating neuronal subtype specification, the sonic hedgehog (Shh) and the Mash1 (Ascl1 – Mouse Genome Informatics)/Ngn signaling pathways. The former is required for motor and interneuron specification in the ventral neural tube (Ericson et al., 1996, 1997; Pierani et al., 1999), whereas the latter transcription factors are involved in the specification of very early neural progenitors, principally at the ventricular surface of the neural tube (Gradwohl et al., 1996; Lee, 1997; Ma et al., 1997). We have also established an epistatic relationship between Hoxb1 and Phox2b/Phox2a (Arix/Pmx2b – Mouse Genome Informatics) in r4, the latter of which are required for specification of the branchial and visceral motoneurons in the brainstem (Pattyn et al., 2000).

Mash1 and Ngns are homologues of the basic helix-loop-helix (bHLH) Drosophila acheate-scute complex and atonal-like proneural genes (Jan and Jan, 1994; Ma et al., 1996, 1997; Anderson and Jan, 1997). In the mouse peripheral nervous system, Mash1 and Ngns foster developmental programs that specify distinct neuronal subtypes. Thus Mash1-deficient mice lack peripheral neurons of the noradrenergic lineage (Guillemot et al., 1993; Hirsch et al., 1998), whereas Ngn1 (Neurod3 – Mouse Genome Informatics) and Ngn2 (Atoh4 – Mouse Genome Informatics) mutants lack sensory ganglia (Fode et al., 1998; Ma et al., 1998, 1999). Within the central nervous system (CNS), these mammalian proneural genes are expressed in complementary as well as overlapping neural progenitor domains, suggesting that in the CNS they function in a more complex, combinatorial fashion with each other, compared to their distinct roles in peripheral neurogenesis.

The Shh-signaling pathway has been elegantly demonstrated to play a pivotal role in dorsoventral (DV) patterning of neurons within the spinal cord (Tanabe and Jessell, 1996; Davenne et al., 1999). In the ventral neural tube, the activity of Shh is graded ventral^high to dorsal^l°^w, and is absolutely required for the induction of floor plate cells, motoneurons and interneurons (Chiang, 1996; Ericson et al., 1996, 1997; Briscoe et al., 1999; Pierani et al., 1999). In response to Shh, cells of the ventral neural tube differentiate into progenitors expressing the homeodomain transcription factors Nkx2.2 (Nkx2-2 – Mouse Genome Informatics) or Pax6. The progeny of these ventral progenitors express specific Lim-homeodomain transcription factors during their progressive differentiation into motor neurons or ventral interneurons, thus acquiring distinct Lim-homeodomain codes (Tsuchida et al., 1995; Pfaff, 1996; Tanabe and Jessell, 1996; Briscoe et al., 2000). The spatiotemporal expression pattern of Mash1 and Ngns suggests that they may coordinate with Nkx2.2 or Pax6 to specify distinct neural progenitors along the DV axis of the neural tube.

The molecules involved in neural determination and Shh signaling pathways are expressed along the full extent of the anteroposterior (AP) axis. This makes unlikely the involvement of these pathways in conferring distinct identities on the neural progenitors at specific AP levels. However, the AP-restricted expression patterns of Hox genes (Dollé et al., 1989; Graham et al., 1989), their ability to specify distinct neural subtypes and their correct spatiotemporal expression patterns during hindbrain neurogenesis make them ideal candidates for interacting with these pathways to confer AP identity on the neuronal progenitor cells (Davenne et al., 1999; this study). In this study we provide evidence that within r4 there is a requirement for Hoxb1 by both the Mash1/Ngn and Shh signaling pathways. Hoxb1 is needed to correctly specify early progenitor cells along the DV axis of r4 and subsequently for normal neuronal differentiation along the ventriculopial (VP) axis, the third coordinate of the developing neural tube.

In order to allow continuous monitoring of Hoxb1 expression in live embryos and embryonic tissues, we generated an allele of Hoxb1, in which the reporter gene for green fluorescent protein (GFP), was targeted inframe into the Hoxb1-coding sequence (Fig. 1a and b). The genomic DNA used for vector construction was isolated from a 129Sv mouse library in lambda FIX II (Stratagene) (Goddard et al., 1996). A 12.3 kb segment of genomic sequence containing the Hoxb1 locus was included in the targeting vector. A 2.25 kb SalI fragment, derived from the vector pEGFPKT1LOXNEO (Godwin et al., 1998), including the gene coding for GFP sequence (EGFP, Clontech), followed by a loxP-flanked neomycin-resistance gene, was cloned into the unique EagI site in exon I by blunt-end ligation, placing EGFP in frame with the Hoxb1 gene (Fig. 1a). The targeting vector was linearized with XhoI and electroporated into R1 ES cells (Nagy et al., 1993). After double selection with G418 and FIAU (Mansour et al., 1988), surviving clones were analyzed by Southern blot. To identify homologous recombinants, a 0.6 kb 5′ flanking probe was used with BamHI-digested DNA (Fig. 1b). To show that no random integration of the targeting vector had occurred and that only one copy of the neo gene was present, as a result of homologous recombination, a neo probe (a 0.75 kb PstI fragment) was used to hybridize digested genomic DNA (data not shown). Positive clones were subsequently injected into C57BL/6J (BL6) blastocysts and the resulting chimeric males bred with BL6 females. Offspring harboring the targeted allele was identified by Southern blot using the same probe as described above (data not shown). A neo-resistance cassette, flanked by loxP sites, was located 3′ to the fusion gene.

Previously, we have encountered several examples where removal of the neo gene has been critical for appropriate expression of the reporter gene and/or to obviate problems associated with interference of the neo gene with the expression of neighboring genes (Barrow and Capecchi, 1996; Olson et al., 1996). For this reason, the loxP-flanked neo selection marker in the Hoxb1^GFPne° allele was removed by Cre/loxP-mediated recombination in vivo. This was accomplished by crossing females heterozygous for the Hoxb1^GFPne° allele with male mice transgenic for the Cre-deleter gene (Schwenk et al., 1995). The offspring from these intercrosses were screened for excision of the neo gene by PCR. The following neo primers, resulting in a 355 bp product, were used: 5′GTGCTCGACGTTGTCACTGAAG3′ (forward primer) and 5′CCATGATATTCGGCAAGCAGGC3′ (reverse primer). The PCR conditions were one cycle at 94ºC for 1 minute, then 94ºC, 30 seconds; 60ºC, 20 seconds; 72ºC, 1 minute for 28 cycles and finally 72ºC for 7 minutes. The neo-less Hoxb1^GFP mice were then backcrossed to BL6 mice for two generations. Heterozygous Hoxb1^GFP mice were subsequently intercrossed, which produced wild-type, heterozygous, and homozygous littermates in the expected Mendelian ratio. The mice were genotyped by PCR using a set of three primers. These were as follows: Hoxb1 sense primer, 5′AGCGCCTACAGCGCCCCAACCTCTTTT3′ (nucleotides 153-179, upstream of the EagI cloning site); Hoxb1 antisense primer, 5′CTTGACCTTCATCCAGTCGAAGGTCCG3′ (nucleotides 615-64, downstream of the EagI cloning site) (Frohman et al., 1990) and GFP antisense primer, 5′ATGGTGCGCTCCTGGACGTAGCCTT3′. The PCR conditions were one cycle at 94°C for 1 minute and then 94ºC, 30 seconds; 60ºC, 30 seconds; 68ºC, 2 minutes for 30 cycles. The wild type allele product was 489 bp and the Hoxb1GFP allele product was 353 bp (Fig. 1c). The phenotype of mice homozygous for the Hoxb1^GFP mutation was indistinguishable from mice homozygous for our previously described mutant alleles of Hoxb1 (Barrow and Capecchi, 1996; Goddard et al., 1996; Rossel and Capecchi, 1999).

Embryos used for detection of GFP fluorescence were the progeny of Hoxb1^GFP-heterozygous intercrosses or crosses between Hoxb1^GFP homozygous males and C57BL/6J females. Embryos were harvested between embryonic days 8.5 and 14.5 (E8.5-E14.5), and maintained at room temperature in Leibovitz’s L-15 medium during imaging with a Bio-Rad MRC 1024 Laser Scanning Confocal Imaging System connected to a Leitz Aristoplan microscope. Hoxb1^GFP homozygotes do not produce GFP protein in the hindbrain after E8.0 because maintenance of Hoxb1 expression in r4 requires Hoxb1 autoregulation (Pöpperl et al., 1995).

Embryos were fixed in 4% paraformaldehyde in PBS and processed for single- or double-immunolabeling as transverse sectioned tissues. After immunolabeling, sections were then analyzed by confocal microscopy. Antibodies were used at the following dilutions: rabbit polyclonal anti-Hoxb1 (Covance, Berkeley, CA), 1:200; mouse monoclonal anti-Shh, anti-HNF3β, anti-Pax6, anti-Nkx2.2 and anti-Isl1 (Developmental Hybridoma), 1:20; mouse monoclonal anti-Mash1, 1:20 (D. J. Anderson); and rabbit polyclonal anti-Phh3 (Upstate Biotechnology, Waltham, MA). Fluoroscein-, Texas Red-, or Cy5-conjugated secondary antibodies were obtained from Jackson Immunoresearch (Westgrove, PA).

The TUNEL assay was used to detect apoptotic cell death in fixed, frozen transverse sections of E9.0-E12.0 embryos following the manufacturer’s protocol (Roche).

Whole-mount RNA in situ hybridization was performed as previously described (Manley and Capecchi, 1995; Goddard et al., 1996).

The dynamic changes in Hoxb1 expression were monitored in live mouse embryos using a targeted allele of Hoxb1 in which the vital reporter gene for GFP, was fused, in frame, with the first protein encoding exon of this gene (Fig. 1; for details see Materials and Methods). The sensitivity of the Hoxb1^GFP reporter allowed detection of cell body migratory processes and axon-specific labeling of r4-derived neurons in live embryonic tissues. In the present study we used the Hoxb1^GFP allele to monitor the spatiotemporal appearance of Hoxb1- positive neuronal columns during hindbrain development.

At E8.5 (six somites), Hoxb1^GFP was highly expressed along the entire AP neural axis up to the level of the presumptive r3-r4 boundary. Dorsal views show that Hoxb1^GFP was expressed uniformly in r4 (Fig. 2a, arrowheads). At this stage, reconstruction of the hindbrain from 5 μm sections showed that all discernible nuclei in r4, approximately 1000, express Hoxb1 (data not shown). Hoxb1^GFP could also be seen in a more anterior domain (r3) than has been previously reported (Fig. 2a, arrow) (Murphy et al., 1989; Frohman et al., 1990; Murphy and Hill, 1991). At the posterior boundary of r4, a transverse band of cells not expressing Hoxb1^GFP could be seen (Fig. 2a, lower arrowhead). This reflects the onset of the r4-restricted expression of Hoxb1 seen in later stages.

By E9.5, the hindbrain neural tube has fully closed. In order to visualize Hoxb1 expression in the ventricular neuroepithelium, a dorsal midline dissection along the AP axis of the hindbrain was made, and the neural tube was splayed open (flat mount) exposing the ventricular surface. At E9.5 (21 somites), two distinct Hoxb1^GFP-expressing columns, corresponding to the basal (ventral) and the alar (dorsal) plates of r4, were visible on each side of the neural tube (Fig. 2b). In slightly older embryos (E9.5, 27 somites), the formation of a Hoxb1^GFP-expressing intermediate column could be seen (Fig. 2c). To confirm that the Hoxb1^GFP signal accurately reflects the expression of endogenous Hoxb1 protein, immunolabeling of E10.5 and E11.5 wild-type mice was performed using a Hoxb1-specific polyclonal antibody (Goddard et al., 1996). At E10.5, the ventral, intermediate and dorsal columns, containing early differentiating neurons, were condensed, as compared with younger embryos (Fig. 2d and data not shown). Intense Hoxb1 labeling could also be seen in the anterior and posterior boundaries of r4. By E11.5, numerous Hoxb1-expressing columns containing later-born neurons had formed along the DV axis of r4 (Fig. 2e). Hoxb1^GFP was similarly expressed in live E10.5 and E11.5 Hoxb1^GFP heterozygous mice (data not shown).

In summary, continuous analysis of Hoxb1 expression along the neural tube DV axis has shown that early in development, E8.0-E8.5, all of the cells in r4 express Hoxb1. Subsequent to this stage, Hoxb1 expression first becomes concentrated to two zones within r4, the alar and basal plates, and then to a third, the intermediate zone. As neural development continues, more Hoxb1-expressing columns, of increasing refinement, become apparent. The segregation of increasing numbers of Hoxb1- expressing columns parallels the concomitant formation of increasing numbers of neuronal subtypes within r4 (Taber Pierce, 1973; Marin and Puelles, 1995).

A very prominent population of neurons that are specified within r4 consist of those that innervate the muscles of facial expression, the facial branchiomotor (FBM) neurons (Goddard et al., 1996; Studer et al., 1996). However, in newborn animals these motoneurons are located within the ventrolateral region of the r6-derived upper medulla. The migration pattern of these motoneurons from r4 to the ventrolateral region of r6 is shown in Fig. 2f-i by a series of transverse sections of the hindbrain of E11.5 heterozygous Hoxb1^GFP embryos. Sections from levels spanning r4 to r6 were immunostained with anti-Shh to distinguish Hoxb1^GFP-expressing cells (green) from the Shh-expressing floor plate (red). In ventral-r4, Hoxb1^GFP expression was primarily observed in the ventricular and mantle layers of the neuroepithelium, juxtaposed to the floor plate (Fig. 2f). This region contains proliferating progenitor and early differentiating neurons, respectively. However, a small population of Hoxb^GFP-expressing neurons was visible in the marginal layer that contained more differentiated neurons. As Hoxb1^GFP-expressing neurons migrated posteriorly into r5, they formed a distinct cluster in the mantle layer immediately lateral to the ventral progenitor domain (VPD). The VPD is a region containing progenitor cells juxtaposed to the Shh-expressing floor plate (Fig. 2g,h; defined molecularly in the next section). At the level of r6, the Hoxb^GFP-expressing FBM neurons took a lateral course into the marginal layer of the neuroepithelium to contribute to the formation of the facial nucleus (VIIn, Fig. 2i).

At E12.5, the expression of Hoxb1^GFP persisted in columns arrayed along the DV axis of the ventricular neuroepithelium (Fig. 2j). Furthermore, Hoxb1^GFP expression could still be seen among posteriorly migrating FBM neurons located in the deeper mantle layer. The morphology of migrating neurons in r5 was pyramidal, with their apical dendritic ends facing in the direction of their migratory path (data not shown). The extensive migration patterns of neurons derived from r4 can be visualized by exposing the pial surface of the hindbrain of E12.5, Hoxb1^GFP heterozygous embryos (Fig. 2k). In contrast to the ventricular layer, where Hoxb1^GFP-positive progenitor cells were restricted to an r4 region, the marginal layer contained differentiated Hoxb1^GFP-positive neurons, which were seen in r4 as well as in more anterior and posterior regions of the hindbrain. The migration pattern of the FBM neurons and their lack of migration in Hoxb1 mutant homozygotes, has been well documented (Goddard et al., 1996; Studer et al., 1996; Pata et al., 2000). Here, we show that in addition to the FBM neurons, there was also extensive migration of non-branchiomotor, Hoxb1-expressing neurons rostral to r4. This observation provides an explanation of how disruption of Hoxb1 results in perturbation of the organization of neurons in r3, a region that is rostral to the majority of Hoxb1-expressing cells (data not shown). By E14.5, the expression of Hoxb1, as detected by Hoxb1^GFP and confirmed by Hoxb1 immunostaining, continues to be observed in the developing hindbrain (data not shown). This continued expression suggests functions for Hoxb1 during later periods of hindbrain development.

The most prominent feature of Hoxb1 mutant homozygous mice is a failure to specify at least two pools of motoneurons: the FBM and contralateral vestibuloacoustic efferent (CVA) neurons (Goddard et al., 1996; Studer et al., 1996; Pata et al., 2000). In the spinal cord, specification of motoneurons is crucially dependent on Shh signaling (Chiang et al., 1996; Ericson et al., 1996, 1997; Briscoe et al., 1999; Pierani et al., 1999). To determine whether this signaling pathway is affected by the Hoxb1 mutation, we examined the cellular organization of the floor plate in r4, using two floor plate-specific markers Shh and HNF3β (Foxa2 – Mouse Genome Informatics; Ruiz i Altaba et al., 1993; Weinstein et al., 1994; Chiang, 1996; Ericson et al., 1996). At all stages examined, the pattern of Shh expression in the r4 floorplate of homozygous mutants was indistinguishable from that of wild-type or heterozygous controls (Fig. 3a,b and data not shown). This is in contrast to what was observed for HNF3β expression. In the spinal cord and most regions of the hindbrain, HNF3β expression was restricted to the floor plate. In ventral r4, however, HNF3β was detected in two distinct domains (Fig. 3c-f and data not shown): the most ventral domain, corresponding to the Shh-expressing floor plate, and a more dorsal domain that co-expresses Hoxb1. We will refer to the more dorsal of the two HNF3β-expressing domains as VPD. This domain appears to be unique to r4 and may represent an important signaling center for this rhombomere. Fig. 3c,g shows that this domain also expresses phosphorylated histone H3 (Phh3) and Mash1. Mash1 labels early neuronal progenitor cells, whereas Phh3 is a marker for actively dividing cells (Johnson et al., 1990; Gradwohl et al., 1996; Lee, 1997; Ma et al., 1997; Wei et al., 1999). As expected in wild-type embryos, Mash1 and Phh3-expressing cells were found close to the inner ventricular layer (i.e., the proliferative layer of the developing neural tube). Fig. 3d shows that the integrity of the VPD, the more dorsal HNF3β expression domain, requires Hoxb1 function, since in the absence of Hoxb1, the cytoarchitecture of this zone was disorganized. In Hoxb1 mutant homozygous embryos, cells expressing Mash1 and Phh3, which are normally restricted to the ventricular layer, had expanded into the mantle layer, a region normally occupied by postmitotic, early differentiating neurons (Fig. 3b,d,f,h). This aberrant behavior of cells in Hoxb1 mutant homozygotes suggests a deficiency in neuronal specification, such that cells leaving the ventricular layer have not appropriately turned off expression of these genes. Furthermore, as indicated by the continued expression of Phh3, these cells continue to divide aberrantly.

In the ventral spinal cord, the activities of the transcription factors Nkx2.2, Pax6 and Isl1 are required for interpretation of the floor plate intercellular signal, Shh (Ericson et al., 1996, 1997; Briscoe et al., 1999, 2000). These molecules are required for proper formation of motoneurons and ventral interneurons. As a further indication that the motoneuron population in r4 is not properly specified in Hoxb1 mutant embryos, the normal expression pattern of these transcription factors is markedly perturbed by this mutation (Fig. 4).

In normal embryos at E11.5, approximately 18-20 layers of cells expressing Nkx2.2 could be viewed in transverse section (Fig. 4a). This figure shows such a section through the ventral region of r4. In Hoxb1 mutant homozygotes, there was a loss of Nkx2.2 expression in the most dorsal aspect of this expression domain, which now occupies only approx. 12 cell layers (Fig. 4b). As described for Mash1 and Phh3, this transcription factor also remained active in cells that were leaving the ventricular surface and progressing towards the mantle layer. Moreover, the Pax6-expression domain expanded ventrally in Hoxb1 mutant embryos into the region that formerly expressed Nkx2.2 (Fig. 4c,d). This observation is consistent with the reported role of Nkx2.2 as a negative regulator of Pax6 expression (Briscoe et al., 1999, 2000).

In the r4-region of E11.5 mouse embryos, Isl1 prominently labelled both the FBM and CVA (contralateral vestibuloacoustic) neurons, and a population of ventrolateral (VL) neurons scattered along a ventral-dorsal region of the mantle layer (Fig. 4e). Expression of this protein is one of the earliest postmitotic markers for motoneurons along the entire extent of the neural tube (Ericson et al., 1996, 1997; Pfaff, 1996). In Hoxb1 mutant homozygous embryos, the two Isl1-labeled motor pools, FBM and CVA, were not observed, whereas the Isl1-labeled VL neuron population had increased in number and was displaced more dorsally (Fig. 4f and data not shown).

To determine how the two motoneuron pools (FBM and CVA) are lost in Hoxb1 mutant mice, we performed TUNEL assays to assess possible loss via programmed cell death. In Hoxb1 mutant mice, we observed a dramatic increase in apoptosis in r4 (Fig. 5a,b). This wave of ectopic cell death began at E9.5 and was finished by E10.5. This increase in apoptosis in r4 of Hoxb1 mutant embryos was not restricted to just the ventral aspect of the neural tube, where motoneurons are normally formed, it was also seen in intermediate and dorsal zones that coincide with the three zones normally occupied at this stage by cells expressing high levels of Hoxb1 (Fig. 2c,d). Fig. 5c-h illustrate parts of transverse sections through r4 at higher magnifications. These figures highlight the cellular detail of apoptosis occurring in these three zones in Hoxb1 control and mutant embryos. The sections were also immunolabeled for Phh3 to delineate the distribution of dividing cells within these regions. From these figures, it is apparent that in Hoxb1 mutants, cells are aberrantly dying in r4 throughout the extent of the neuroepithelium, from the inner ventricular to the outer pial surfaces. In particular, note that cells, at the position of the very early, still proliferating neural progenitors, are dying. This may explain the absence of Isl1-positive FBM and CVA neurons normally seen at later stages (Fig. 4e,f).

Pattyn et al. (2000) have recently demonstrated that the formation of the branchial and visceral motoneurons in the hindbrain is crucially dependent on the function of the paired-like homeodomain transcription factor, Phox2b. In mouse embryos homozygous for Phox2b loss-of-function mutations, the branchial and visceral motoneurons of the brainstem are not properly specified. These neurons do not express early postmitotic molecular markers common to motoneurons, such as Isl1 and Phox2a. Moreover, they do not turn off early neuronal progenitor markers such as Mash1 and Nkx2.2. Progenitors continue to divide as they migrate from the ventricular to the mantle layer, and they are programmed for cell death at E10.5. The aberrant cellular phenotype of the branchial motor neurons in Phox2b mutants is very similar to the mutant cellular phenotype described above for the r4-component of this motoneuron system in Hoxb1 mutant embryos, suggesting possible involvement of these two transcription factors in a common genetic pathway.

Fig. 6 shows that Hoxb1 is epistatic to Phox2b for the formation of the FBM neurons. At the onset of Phox2b expression, E9.5 (Pattyn et al., 1997), the dorsal region of the r4 VPD domain was significantly depleted in mutant compared with control embryos, whereas the more ventral VPD maintained the expression of Phox2b (Fig. 6a,b). At this stage, the dorsal VPD was occupied by a significant number of TUNEL-positive cells (Fig. 6b). At a slightly later time point, E10.0, the entire VPD was almost completely devoid of Phox2b-expressing progenitor cells with significant numbers of TUNEL-positive cells occupying this region (Fig. 6c,d). After E10.5, few TUNEL-positive cells were detected in r4 of control and mutant embryos (data not shown). By E11.0-E11.5, the expression of Phox2b in the r4 VPD of mutant embryos was completely absent compared with control embryos (Fig. 6e-h). However, an expanded population of cells expressing Isl1 and/or Phox2b was observed in the marginal layer of ventral r4. The loss of Phox2b expression in the dorsal VPD was supported by a similar loss of Nkx2.2 expression in this progenitor pool (Fig. 6g,h). The expanded population of Nkx2.2-expressing progenitor cells in the ventral region of the VPD corresponded to an increase of Isl1- and Phox2b-expressing VL neurons (Fig. 6e-h). The overall requirement along the DV axis of r4 neuronal columns expressing Phox2b and its effector, Phox2a, on Hoxb1 is demonstrated in Fig. 6i-l. From these flatmount preparations, it is apparent that by E11.5 the entire Phox2b and Phox2a expressing, FMB neural population has gone in the Hoxb1 mutant embryos (Fig. 6j and l), as well as the r4 components of restricted, more dorsal neural columns. Taken together, these experiments demonstrate that Hoxb1 function is required to activate and maintain cells expressing one of the earliest known transcription factors needed for specification of the r4 components of the branchiomotor system.

Examination of Fig. 6 shows that the Hoxb1 mutation not only affects the formation of the r4-component of the branchiomotor system, but also the r4 component of intermediate and dorsal neural columns that express Phox2a and Phox2b (arrow), respectively. As an example, the r4-component of the most dorsal Phox2b-expressing column is absent in Hoxb1 mutant homozygous embryos (Fig. 6j). Similarly, the r4 component of the intermediate column that expresses Phox2a is also absent in Hoxb1 mutant embryos (Fig. 6l). These results show that Hoxb1 is not only involved in the specification of motoneurons within r4, but in the specification of neurons throughout the DV extent of r4. These results are entirely consistent with the report recently published by Davenne et al. (1999) that demonstrated a role for Hoxa2 in the specification of neurons along DV extent of r2 and r3. Consistent with the early mis-specification of neurons throughout the DV extent of r4, Hoxb1 mutants showed extensive aberrant cell death in E9.5 mutant embryos in the dorsal, intermediate and ventral regions of r4 (Fig. 5).

We have shown that the pattern of Hoxb1 expression in r4 is very dynamic. Initially Hoxb1 is expressed uniformly in r4. Subsequently, its expression becomes confined to more and more refined columns within r4. The early neural specification defects seen in Hoxb1 mutants suggest interactions of Hoxb1 with homologues of the Drosophila proneural genes achaete-scute and atonal, the bHLH transcription factors Mash1, Ngn1 and Ngn2 (Jan and Jan, 1994; Anderson and Jan, 1997; Fode et al., 1998; Ma et al., 1996, 1997, 1998, 1999). These genes are expressed in broad columns along the length of the AP axis in proliferating and early differentiating neurons as early as E8.5 in the mouse, approx. 1 day after Hoxb1 expression is first detected in r4 (Guillemot and Joyner, 1993; Lee, 1997).

Comparison of the columnar expression pattern of Mash1 in Hoxb1 mutant and control embryos suggests that one role of Hoxb1 is to restrict the boundaries of these columns within r4. This is illustrated in Fig. 7. Fig. 7a,b show flatmount preparations of Hoxb1 control and mutant hindbrains labeled with a Mash1 RNA probe. Figs 7c-l show transverse sections through r4 immunostained with Mash1 and Hoxb1 antibodies. It is apparent from these figures that in control embryos the Mash1-expressing, early neural progenitor columns are juxtaposed by columns of cells expressing high levels of Hoxb1. In Hoxb1 mutant homozygotes, the widths of the Mash1-expressing columns in r4 have expanded. Very similar changes in the widths of the r4-component of the Ngn1 and Ngn2 neuronal expression columns are observed in Hoxb1 mutant embryos (data not shown).

Individual members of the Hox complex are differentially expressed along the embryonic AP axis. As such, they are ideally suited to provide the positional cues to reiterated cell types and structures, such as neurons, vertebrae and muscles, that are needed to guide their distinction along this major axis. Indeed, it appears that a major role for Hox genes during embryogenesis is to provide such positional values to these cells (Chisaka and Capecchi, 1991; Krumlauf, 1994; Lumsden and Krumlauf, 1996; Duboule, 1998). The question then arises, when and through which molecular circuits are Hox positional values contributed to these differentiating cells? From the studies described here, we would argue that with respect to Hoxb1 and the neurons derived from r4, this specification occurs very early during hindbrain neurogenesis. Specifically, Hoxb1-mediated neuronal specification appears to occur at the ventricular or proliferative layer of the r4 neuroepithelium, working in parallel with molecules required for early DV patterning, Shh and HNF3β, and those involved in neural determination, Mash1, Ngn1 and Ngn2.

The most prominent mutant phenotype associated with disruption of Hoxb1 is the absence of a functional FBM nucleus (Goddard et al., 1996; Studer et al., 1996; Pata et al., 2000). As a consequence, Hoxb1 mutant homozygous adults show complete paralysis of the muscles of facial expression, which are normally innervated by this nerve (Goddard et al., 1996). Consistent with the hypothesis that the absence of a FBM nucleus results from a failure to specify FBM neurons, early postmitotic molecular markers that normally label these neurons, such as Isl1 and Phox2a, fail to do so in r4 and r5 of these mutant embryos. Even more informative, one of the earliest transcription factors known to be required for specification of all branchial and visceral motor neurons of the brainstem, Phox2b, is not expressed in a distinct pool of progenitor cells in the ventral progenitor domain of r4 of Hoxb1 mutant embryos, at any stages that we have examined. Instead, we observe that cells expressing effector molecules of Shh, such as HNF3β and Nkx2.2, that are normally associated with early, dividing neural progenitors, are reduced and continue to be expressed ectopically in the mantle layer normally occupied by postmitotic neurons. This aberrant cellular phenotype is very similar to that described by Brunet and his colleagues for mis-specified FBM neurons resulting from disruption of Phox2b (Pattyn et al., 2000). This is accompanied in Hoxb1 mutant embryos by induction of a wave of ectopic apoptosis that begins at E9.5, and corresponds directly to the time of normal onset of FBM neuron generation and Phox2b expression (Taber Pierce, 1973; Pattyn et al., 1997). The ectopic apoptosis is, however, not restricted to the ventral region of r4 in Hoxb1 mutant embryos, but extends across the three regions of high Hoxb1 expression. And indeed, failure to specify the r4-component of specific neuronal columns is observed throughout the DV extent of the neural tube.

An alternative to the hypothesis that failure to specify the FBM neurons leads to their aberrant death in Hoxb1 mutants, is that as a consequence of this mutation, these neurons acquire a different fate. For example, in the absence of Hoxb1 gene product, these neurons could now behave as r2-like, trigeminal-branchiomotor neurons. This hypothesis would predict that these mis-specified neurons should still express Phox2b, Phox2a and Isl1, reflecting their motoneuron character. However, the failure to detect populations of such cells in Hoxb1 mutant embryos that express these markers, either in the ventral or in ectopic regions of r4, argues against this alternative hypothesis. There still remains the possibility that in Hoxb1 mutant mice, later-born neurons derived from the ventral r4 VPD, such as the VL neuron population, may be mis-specified, owing to their dependence on interactions with earlier-born motoneurons (McConnell, 1995; Sockanathan and Jessell, 1998).

Although the fate of cells within r4 is affected by the Hoxb1 mutation, the overall cytoarchitecture of this rhombomere is not dramatically altered by this mutation (Goddard et al., 1996; Studer et al., 1996). This observation suggests that the Hoxb1 mutation affects selective cell populations within r4 and that the wave of ectopic apoptosis observed in r4 does not dramatically alter the final cell number within this rhombomere. Interestingly, in Hoxb1 mutant embryos we also observe ectopic expansion of cell proliferation identified by Phh3 expression, extending from the ventricular to the mantle neuroepithelial layers. This increase in cell proliferation may, in part, compensate for the loss of cells via aberrant apoptosis early in neurogenesis.

Davenne et al. (1999) have recently shown that Hoxa2 plays an important role in the DV patterning of neurons within r2 and r3. As observed for the Hoxb1 mutation, disruption of Hoxa2 selectively affects the formation of the r2/r3-component of neuronal columns that express transcription factors, such as Pax6 and Phox2b, that are in turn involved in the specification of neuronal subtypes. Together, these observations emphasize that the neuronal columns that extend longitudinally across multiple rhombomeres and even into the spinal cord, are built in modules, with different Hox genes being responsible for the formation of the separate modules. Concomitant with this early role of Hox genes in neuronal specification, the progenitor cells automatically acquire a positional value along the AP axis, that allows them to be distinguished from similar cells within a contiguous longitudinal functional column. These observations also emphasize that these Hox genes are epistatic to the set of transcription factors that are used to specify neuronal subtype differentiation. The obvious advantage of this strategy is that positional value can be assigned to multiple neuronal subtypes within an AP region, rather than having to ascribe positional cues individually to each subtype subsequent to its specification.

Interestingly, Hoxb1 mutant mice also show defects in the function of the lacrimal and salivary glands (Goddard et al., 1996). These glands are innervated by postganglionic parasympathetic neurons of the pterygopalatine and submandibular ganglia, respectively (Carpenter and Smith, 1988). These ganglia are in turn innervated by preganglionic parasympathetic neurons of the superior salivatory nucleus. The source of these visceral efferent neurons has not been well established. They may arise in r5, or they may be born in r4 and migrate into r5. In either case, their function is affected by the Hoxb1 mutation. However, the expressivity of this defect in Hoxb1 is variable. Generally, variability in expressivity of a mutant phenotype in Hox mutants is associated with the participation of more than one Hox gene in that function. Therefore, it may be possible that Hox genes expressed in r5, such as members of the Hox3 paralogous family, are good candidates for such a shared function with Hoxb1.

To examine the role of Hoxb1 during early neural differentiation, the effects of the Hoxb1 mutation on cells expressing the neural-specific bHLH transcription factors were studied. From an examination of in situ hybridization and immunohistochemical patterns of flat-mount and transverse section preparations, respectively, it is apparent that the broad longitudinal columns expressing Mash1, Ngn1 and Ngn2 are juxtaposed in r4 with columns of cells expressing high levels of Hoxb1. In the absence of Hoxb1, the width of these columns expands in r4, suggesting that Hoxb1 normally restricts the domains of these early neural progenitor cells. Since it has been shown in multiple laboratories that in both Drosophila and vertebrates the early neural progenitor domains are restricted and reinforced by the Delta/Serrate/Notch signaling pathways (Jan and Jan, 1994; Heitzler et al., 1996; Anderson and Jan, 1997; Panin et al., 1997), it is attractive to consider that the restriction of the Mash1, Ngn1 and Ngn2 domains within r4 by Hoxb1 is also mediated by the same signaling pathway. On the basis of this hypothesis, it will be of interest to determine whether the production of successively more refined Hoxb1- expressing columns within r4 is involved in restricting and/or reinforcing increasing numbers of neuronal subtype columns within this rhombomere.

In Hoxb1 mutant mice, there is a loss of r4-dorsal and intermediate columns expressing Phox2b and Phox2a, respectively. Interestingly, in Mash1 mutant mice, there is also a loss of the same Phox2b-expressing column, but in contiguous dorsal columns along the hindbrain (Hirsch et al., 1998). Together, these data suggest that these two transcriptional systems work in parallel with each other during neural determination of common progenitors, and provide further support for Hoxb1 contributing the AP-specific information to progenitor cells that may otherwise be similar along the length of the hindbrain.

In conclusion, the present study provides evidence that Hoxb1 is required for the formation of multiple neuronal subtypes along the full extent of the DV axis of r4. The role of Hoxb1 appears to be required very early during hindbrain neurogenesis in parallel with molecules required for DV patterning and neural determination, and prior to the activation of the transcription factors such as Nkx2.2, Isl1, Phox2b and Phox2a, which are used to specify neuronal subtypes. In the absence of Hoxb1, the r4-component of multiple neuronal subtypes fails to be properly specified and is then destined for aberrant programmed cell death.

We are deeply grateful to members of the Capecchi laboratory for useful discussion and comments, to members of the mouse and tissue culture facilities for technical support, and to L. Oswald for preparation of the manuscript. We also thank the following scientists for providing valuable reagents: D. Anderson, J. Brunet, C. Goridis, F. Guillemot, R. Kageyama and Q. Ma. Monoclonal antibodies against SHH, HNF3β, Pax6, Nkx2.2, and Isl1 were obtained from the Developmental Hybridoma Bank under the contract N01-HD-62915. P. F. was supported by the Wenner-Gren Foundations. G. O. G. is a research associate of the Howard Hughes Medical Institute.