Hox3 genes coordinate mechanisms of genetic suppression and activation in the generation of branchial and somatic motoneurons

An evolutionarily conserved function of Hox genes is to assign positional information to body segments along the anteroposterior (AP) axis of developing organisms (Ferrier and Holland,2001; Hafen et al.,1984; Lewis, 1978; McGinnis and Krumlauf, 1992; Rozowski and Akam, 2002; Weatherbee et al., 1998). In vertebrates, segments or rhombomeres (r) of the developing hindbrain have been used as a model to understand the role of Hox genes in segmental patterning(Barrow et al., 2000; Bell et al., 1999; Carpenter et al., 1993; Davenne et al., 1999; Gaufo et al., 2000; Gavalas et al., 1998; Goddard et al., 1996; Mark et al., 1993; McClintock et al., 2002; Pata et al., 1999; Studer et al., 1998; Studer et al., 1996). A plethora of genetic and embryological experiments have established that Hox genes are necessary and sufficient to confer a unique identity to rhombomeres in early stages of hindbrain segmentation, as well as cell identity in later stages of development. The general phenomenon of Hox gene induced homeosis -transformation of one body part to another - initially observed in Drosophila, appears to be equally applicable in complex structures such as the vertebrate hindbrain. In the mouse, for example, absence of Hoxb1 function in r4 results in a lack of facial branchiomotoneurons(BMNs) owing to an apparent segmental transformation of r4- to an r2-like rhombomere identity (Goddard et al.,1996; Studer et al.,1996). Conversely, after segmentation has established an apparent normal r2 in chick embryos, gain of Hoxb1 function results in the ectopic differentiation of r4-like facial BMNs in r2(Bell et al., 1999). Thus,although studies of loss of Hox gene function has revealed an important role in early segmental patterning, the latter study also implicates a significant role for Hox genes in the differentiation of unique cell types later in hindbrain development.

There is also evidence that combinations of various Hox genes can elicit emergent phenotypes beyond the contributions of individual genes(Condie and Capecchi, 1994; Gavalas et al., 1998; Rossel and Capecchi, 1999; Studer et al., 1998). For example, the absence of either Hoxa1 or Hoxb1 function leads to some degree of rhombomere identity transformation and cell loss is observed(Carpenter et al., 1993; Gaufo et al., 2000; Goddard et al., 1996; Mark et al., 1993; Pattyn et al., 2003; Studer et al., 1996). However,the combined absence of both genes leads to severe abnormal programmed cell death resulting in the deletion of multiple rhombomeres and subsequent reorganization of the hindbrain (Gavalas et al., 1998; Rossel and Capecchi, 1999). Therefore, the interaction of these genes is equally important for the determination of segmental specification as is the contribution of each individual Hox gene to individual rhombomere identity. It remains to be determined how the various Hox genes expressed in the hindbrain,r2 to r7, interact with each other to determine distinction of each rhombomere as well as integration of function across rhombomeres.

Similarly, given that Hox genes are co-expressed with molecules involved with generic neural specification programs in later stages of development, it is likely that they can also interact with these programs to establish unique cell identities within each rhombomere(Davenne et al., 1999; Gaufo et al., 2000; Osumi et al., 1997; Pattyn et al., 2003; Takahashi and Osumi, 2002). For example, neural progenitors expressing the homeodomain proteins Nkx2.2 and Phox2b give rise to all BMNs present in various rhombomeres of the hindbrain. The interaction between these molecules with different combinations of Hox proteins expressed at different rhombomeres may give rise to functionally distinct BMNs, such as facial BMNs unique to r4 where Hoxb1 is expressed (Gaufo et al., 2000; Goddard et al., 1996; Pattyn et al., 2003; Studer et al., 1996). Likewise, somatic motoneurons (SMNs) derive from neural progenitors expressing the homeodomain proteins Pax6 and the bHLH Olig2 are also present at various rhombomeres. In this case, however, the Hox genes involved in their differentiation into functionally unique SMNs within each rhombomere have not been determined (Guidato et al.,2003).

In summary, given that Hox gene expression persists throughout hindbrain development, it is plausible that it plays at least two distinct roles in neuronal specification. First, Hox genes are required for assignment of rhombomere identity and in their absence, for example Hoxa1 and Hoxb, such patterning is highly perturbed(Gavalas et al., 1998; Rossel and Capecchi, 1999; Studer et al., 1998). Second,they may play a role in differentiating functionally unique motoneurons at a later stage of development (Bell et al.,1999; Guidato et al.,2003). The goal of this study was to identify what combination of Hox genes are necessary to distinguish rhombomeres r4 to r6, and how these genes effect the production of motoneuron subtype programs unique to these rhombomeres. To achieve this, we first focused on Hox3 paralogous genes(Hoxa3, Hoxb3 and Hoxd3), expressed in caudal rhombomeres,and analyzed the affects of different combinations of Hox3 mutations on the segmental identities of r4, r5 and r6. Then, to test the role of these Hox genes in neural specification, we looked for the presence or absence of specific SMNs in r5 with relation to different combinations of null mutations of the Hox3 genes. We observed that loss of any of the Hox3 genes results in a r6- to r4-like change in cell fate; different double mutation showed a graded increase in cell fate change. We also observed a gene dosage dependence of SMN specification of Hoxa3 and Hoxb3 in r5. This specification appears to be mediated through the control of the Pax6/Olig2regulatory pathway for SMN formation, suggesting a direct influence of Hox3 genes on SMN fate. Together, these observations reveal two functions for Hox3 paralogous genes in the developing hindbrain in defining segmental identity across multiple rhombomeres and controlling cell fate within an individual rhombomere.

Mice harboring mutations for Hoxb1, Hoxa3, Hoxb3, Hoxd3 and Pax6 were previously reported(Ericson et al., 1997; Gaufo et al., 2000; Greer et al., 2000; Manley and Capecchi, 1998). The Hoxa3^CFP reporter mouse expresses an in-frame fusion between the C terminus of the Hoxa3 protein and the N terminus of ECFP. This allele was generated by the targeted insertion of the ECFP gene (Clontech)to the second exon of Hoxa3. This fusion has been shown not to affect Hoxa3 function (K.R.T. and M.R.C, unpublished). Single- or compound-mutant embryos were generated from intercrossings of either single-or double-heterozygote mice bearing the mutant allele(s) of interest. Mice were mated overnight and the detection of plugs the following morning was considered 0.5 days post-coitum [embryonic day 0.5, (E0.5)]. Genotyping was performed as previously described (Gaufo et al., 2000; Greer et al.,2000; Manley and Capecchi,1998). A minimum of two embryos were analyzed for each time point and experimental group.

For whole-mount in situ hybridization, E11.5 neural tubes were dissected in cold PBS and fixed in cold 4% formaldehyde for 2-3 hours and processed for Hoxa3, Hoxb3 and Hoxd3 RNA in situ hybridization as previously described (Manley and Capecchi,1998). Detection of Hoxb1^GFP was detected by confocal microscopy on E11.5, live-dissected flat-mount hindbrain preparations. E9.25-E11.5 embryos were harvested as described above and processed for cryostat sectioning. Frozen, 10 μm transverse and coronal sections were labeled with rabbit anti-Hoxb1 (1:250, Covance), rabbit anti-Phox2b (1:1000, a gift from C. Goridis), rabbit anti-Olig2 (1:5000; a gift from H. Takebayashi),mouse anti-TuJ1 (1:1000, Covance), rat anti-Hoxb4 (1:25; Developmental Studies Hybridoma Bank, DSHB), mouse anti-Nkx2.2 (1:25, DSHB), mouse anti-Isl1/2(1:25; DHSB), rabbit anti-Chx10 (1:2500; a gift from S. Morton and T. Jessell), mouse anti-MPM2 (1:8000; Upstate Biotechnology), rabbit anti-activated Caspase-3 (1:50; NEB Cell Signaling), TUNEL (manufacturer's protocol, Roche), mouse anti-NeuN (1:250; Chemicon), rabbit anti-ChAT (1:1000;Chemicon) and αBTX (1:500; Molecular Probes). Immunolabeled sections were developed with Alexa-fluor (1:1000; Molecular Probes) and Cy5 (1:1000;Jackson Immunoresearch)conjugated secondary antibodies. Images of fluorescent-labeled sections were captured with the BioRad MRC 1024 confocal microscope and processed with Adobe Photoshop and Microsoft Powerpoint software.

Our analyses focused on the function of Hox genes in the central region of the hindbrain spanning r4 to r6, where two motoneuron subtypes are characterized by their specific distribution and the expression of distinct combinations of genes encoding Hox and non-Hox homeodomain proteins(Briscoe et al., 2000; Carpenter and Sutin, 1983; Gaufo et al., 2000; Pattyn et al., 2000; Pattyn et al., 1997; Saper, 2000)(Fig. 1A-E). The facial BMNs can be identified by their unique r4 origin and their posterolateral migration into r6 (Fig. 1B, green)(Gaufo et al., 2000). In the same panel, the relative location of the abducens somatic motoneurons (SMNs)is diagrammed to show their positions relative to the BMNs. In contrast to BMNs, SMNs are born in a ventromedial position in r5 and migrate laterally within the same rhombomere. In transverse sections of the ventral hindbrain of E11.5 control embryos labeled for Hoxb1^GFP, Phox2b and Isl1/2, the differentiation and migration of the BMNs can be further characterized by their lateral location and combined expression of Hoxb1 (green), Phox2b (blue)and Isl1/2 (red) in r4 to r6 (triple labeling seen as white) and the SMNs are labeled singly by Isl1/2 in r5 (Fig. 1C-E). In E11.5 control hindbrain flatmount, the relative high levels of Hoxa3, Hoxb3 and Hoxd3 expression are restricted to the r4/r5, r6/r7 and r5/r6 boundaries, respectively, suggesting that Hox3 genes may have a role in the specification of multiple rhombomere identities as well as the motoneuron subtypes that are derived from these rhombomeres(Fig. 1F-H).

Analysis of patterning defects along the AP axis among mice individually homozygous for Hoxa3, Hoxb3 or Hoxd3 revealed a mild ectopic expression of Hoxb1 in single mutants (data not shown). However, the ectopic expression of Hoxb1 was exacerbated in E11 embryos harboring different combinations of double mutations in the Hox3 genes, with Hoxa3^-/-d3^-/- double mutant embryos showing the greatest effect (Fig. 2A-D; data not shown). The systematic elimination of Hoxa3 and Hoxd3 alleles reveal a gene dosage-dependent effect in the severity of ectopic Hoxb1 expression(Fig. 2A-D). Labeling of Hoxb4,which demarcates the boundary between r6 and r7 suggests that the ectopic expression of Hoxb1 is restricted to r6(Fig. 2E,F)(Gould et al., 1998). Hoxb4 also labels neurons in the marginal layer of r6, presumably neurons that have migrated from r7, which are reduced in Hoxa3^-/-d3^-/- mutant embryos. The Hox3 genes may thus function to regulate the anterior migration of these neurons. With respect to this study, however, the appearance of ectopic Hoxb1-expressing cells in the inner ventricular and outer marginal neuroepithelial layers suggest that these cells are initially born and differentiate in r6.

To determine if the loss of Hoxa3 and Hoxd3 result in early segmental defects, we analyzed the expression patterns of Hoxb1, Krox20 and Hoxb4 in coronal sections of E9.25-9.5 Hoxa3^-/-d3^-/- double mutant embryos. In both control and Hoxa3^-/-d3^-/-double mutant embryos, Hoxb1 expression was normal in r4 and in neural crest cells migrating from r4 (Fig. 2G,H). Krox20 expression in r3 and r5 and the expression of Hoxb4 at the border of r6 and r7 also appear normal in Hoxa3^-/-d3^-/- double mutant embryos compared with controls (Fig. 2I,H). In r6, however, the expression of Hoxb1 persists in Hoxa3^-/-d3^-/- double mutant embryos at a period when the expression of Hoxb1 is normally down regulated posterior to r4 (Gaufo et al., 2000)(Fig. 2H, bracket). Although segmentation and rhombomere periodicity appear normal in Hoxa3^-/-d3^-/- double mutant embryos,the persistence of Hoxb1 expression in r6 indicates an early defect in segmental identity.

We next examined the functional consequence of the ectopic expression of Hoxb1 in r6 of Hoxa3^-/-d3^-/- double mutant embryos. To address this issue, we focused on two well-defined functions of Hoxb1 in r4: (1) the expression requirement of Nkx2.2, Phox2b and Isl1/2 among BMNs; and (2) the posterior migration of Hoxb1-expressing BMNs from r4 to the ventrolateral region of r6(Gaufo et al., 2000; Goddard et al., 1996; Pattyn et al., 2000; Studer et al., 1996)(Fig. 1C-E). In control embryos, a small population of these r4-derived Hoxb1-expressing BMNs can be seen adjacent to the Nkx2.2-expressing progenitor domain in ventral r6(Fig. 3A). In Hoxa3^-/-d3^-/- double mutant embryos,Hoxb1 is ectopically expressed in an expanded domain of Nkx2.2-expressing cells, indicative of BMN progenitors (Fig. 3B) (Gaufo et al.,2000; Pattyn et al.,2000). Co-localization of Isl1/2 with these same Hoxb1-expressing cells suggests that these progenitors are differentiating into motoneurons(Fig. 3D). Furthermore, the ectopic expression of Phox2b and Isl1/2 suggest that these neurons are differentiating specifically into BMNs(Fig. 3F).

These findings suggest that Hoxb1 is capable of inducing the ectopic expression of BMN-specific genes within r6. However, it does not provide evidence for the capacity of these motoneurons to adopt functional properties,such as posterior migration, unique to BMNs in r4. To examine the possibility that the ectopic Hoxb1-expressing BMNs in r6 have adopted an r4-like migratory capacity, we looked for their presence in r7 as a consequence of migration from r6. The expression of Hoxb4 was used to demarcate the posterior boundary of r6 and also to label cells specific to r7(Gould et al., 1998). Transverse sections of E11.5 control embryos show the expression of Hoxb1 in r6 is restricted to a small population of BMNs that have migrated from r4;whereas in Hoxa3^-/-d3^-/- double mutant embryos, Hoxb1 is expressed ectopically at a high level throughout r6(Fig. 3G,H). Analysis of r7 shows no Hoxb1-expressing neurons in control embryos(Fig. 3I). By contrast,Hoxb1-expressing motoneurons, as confirmed by co-expression with Isl1/2 (data not shown), are observed in r7 of Hoxa3^-/-d3^-/- double mutant embryos(Fig. 3J). Moreover, the ectopic Hoxb1-expressing cells in r7 do not express Hoxb4, suggesting that these cells have migrated from r6 and are not derived from r7.

To address the possibility that the observed Hoxb1-expressing cells in r7 of Hoxa3^-/-d3^-/- double mutant embryos were derived from r4 and migrated aberrantly through r6, we analyzed coronal sections of younger E10.5 embryos at a period when the Hoxb1-expressing BMNs have initiated their migration from r4(Fig. 3K,L). In both control and Hoxa3^-/-d3^-/- double mutant embryos, the expression of Hoxb1 in r4 is normal. In r5, the initial migration of Hoxb1-expressing cells is also normal in both groups. In r6, the ectopic expression of Hoxb1 in Hoxa3^-/-d3^-/-double mutant embryos is evident compared with control embryos. In r7 of Hoxa3^-/-d3^-/- double mutant embryos,small clusters of Hoxb1-expressing neurons, negative for Hoxb4, are clearly visible (Fig. 3L). This observation thus precludes the possibility that the ectopic Hoxb1-expressing cells in r7 have migrated from the BMN population derived from r4. Together,these observations provide strong evidence that the ectopic expression of Hoxb1 in r6 is sufficient to activate a BMN differentiation and migration program normally unique to r4. These findings are consistent with a previous report demonstrating that the localized or global misexpression of Hoxb1 in r2 of the chick embryo is sufficient to transform cells in this rhombomere into r4-like BMNs (Bell et al.,1999). Moreover, the posterior migration of r4-like BMNs from r6 into r7 suggests that r7, like r5, produces local environmental cues required for the migration of even-numbered derived `facial' BMNs(Garel et al., 2000; Studer, 2001).

Among the Hox3 genes, only Hoxa3 and Hoxb3 are expressed in the ventral r5 region from which SMNs are derived(Fig. 1D,F-H). The loss of Hoxa3 alone, results in a reduction in the number of SMNs (data not shown), whereas the elimination of both Hoxa3 and Hoxb3 in E11.25-E11.5 mutant embryos lead to the complete loss of SMNs, as identified by expression of the homeodomain HB9 (Fig. 4A,B) (Arber et al.,1999). By contrast, r4-derived Phox2b-expressing BMNs and r5-derived visceromotoneurons labeled by Isl1/2 and Phox2b are unaffected(data not shown). The specific loss of SMNs in Hoxa3^-/-b3^-/- double mutant embryos closely resembles the phenotype of embryos bearing independent mutations for Olig2 and Pax6 (Fig. 4C) (Lu et al.,2002; Novitch et al.,2001; Takebayashi et al.,2002; Zhou and Anderson,2002). However, in the case of the Hoxa3^-/-b3^-/- double mutant embryos,the loss of SMNs is restricted to r5.

To define the stage-specificity of the defects in Hoxa3^-/-b3^-/- double mutant embryos,we first characterized the normal progenitors and neuronal subtypes in ventral r5 at the onset of SMN specification, prior to the migration of r4-derived BMNs into this region (Fig. 4D-I; data not shown). In E10.25 control embryos, three distinct progenitor domains and the neuronal subtypes that arise from them can be identified along the DV axis by their expression of specific homeodomain proteins (Fig. 4D-I)(Briscoe et al., 2000). In the ventral region of r5 just dorsal to the floor plate, the progenitors for V3 interneurons (pV3), SMN (pSMN) and V2 interneurons (pV2) can be identified by their expression of Nkx2.2, low levels of Pax6 and high Pax6, respectively(Fig. 4D). The pSMN can be further characterized by the expression of Olig2, the high expression of which appears to encompass the low Pax6-expressing region(Fig. 4E). These progenitors give rise to SMNs of the abducens nucleus, which are specifically labeled by HB9 (Fig. 4F). The more ventral V3 interneurons can be identified by their expression of Phox2b(Fig. 4G). These interneurons presumably give rise to serotonergic neurons(Hendricks et al., 2003; Pattyn et al., 2003). The SMNs and V3s in r5 are both labeled by the pan motoneuron marker Isl1/2(Fig. 5J,K). The V2 interneurons, as in the spinal cord, are located immediately dorsal to SMNs and can be identified by the expression of Chx10(Fig. 4H). Using a Hoxa3^CFP reporter, the expression of Hoxa3^CFP and HB9 were found to be co-localized in the outer marginal layer of the neuroepithelium (Fig. 4I). Although the relative high expression levels of Hoxa3^CFP are found amongst differentiated cells, its ubiquitous expression throughout the neuroepithelium, from progenitors to differentiated neurons, suggests a much broader regulatory role in neuronal patterning.

As described earlier, the restricted loss of motoneurons in r5 of Hoxa3^-/-b3^-/- double mutant embryos was specific to SMNs labeled by HB9, with no apparent loss of Phox2b expression amongst the more ventral V3 interneurons(Fig. 4A,B). To determine the onset of these defects, we analyzed E10.25 control and Hoxa3^-/-b3^-/- double mutant embryos for the expression of Olig2, which specifically labels pMNs(Fig. 5A-D). The expression of Olig2 and HB9 show a gene dose-dependent response to the sequential loss of Hoxa3 and Hoxb3 alleles. The complete loss of Olig2 expression in Hoxa3^-/-b3^-/- double mutant embryos, therefore suggests that Hoxa3 and Hoxb3 are upstream of the Olig2-dependent specification of SMN progenitors.

To determine the cause of the Olig2-specific loss in Hoxa3^-/-b3^-/- double mutant embryos,we examined two possibilities: (1) programmed cell death; and (2)transformation in neuronal fate. To address the first possibility, we analyzed E10.25 embryos for the expression of TUNEL and activated caspase 3, both markers for programmed cell death (Fig. 5E-H). No differences were observed across experimental groups. Analysis of older Hoxa3^-/-b3^-/- double mutant embryos (E10.5-E11.25) also showed no differences in TUNEL and caspase 3 expression (data not shown). An assay for cell proliferation using the mitosis marker MPM2, showed normal cell division in Hoxa3^-/-b3^-/- double mutant embryos compared with controls (Fig. 5E-H). The absence of aberrant cell death among the experimental groups left the possibility that the loss of SMNs in Hoxa3^-/-b3^-/- double mutant embryos may be the result of a change in neuronal fate. To address this issue, we examined the expression pattern of Pax6, which identifies progenitors of SMNs and V2s. In E10.25 Hoxa3^-/-b3^-/-double mutant embryos, the normal low Pax6 expression in the pSMN domain is qualitatively similar to the more dorsal high Pax6 expression in the pV2 domain (Fig. 5I,M). The molecular change in the pSMN domain to that of the pV2 domain is substantiated by appearance of Chx10-expressing V2 interneurons immediately dorsal to V3 interneurons, a region normally occupied by SMNs(Fig. 5K,O). The ectopic appearance of Chx10-expressing V2 interneurons in the region normally occupied by HB9-expressing SMNs was also confirmed in E11.5 Hoxa3^-/-b3^-/- double mutant embryos compared with controls (Fig. 5L,P). In a gene dose-dependent manner, Hoxa3 and Hoxb3 are thus required for the specification of progenitors that will give rise to SMNs of the abducens nucleus.

The loss of SMN precursors of the abducens nucleus was also confirmed in late stage, E18 Hoxa3^-/-b3^-/- double mutant embryos. In the upper medulla, the r5-derived abducens nucleus can be easily identified by its stereotypic relationship with axons from the r4-derived facial nucleus. The axons of the facial nucleus, known as the genu,circumscribe the abducens nucleus in a medial to lateral pattern(Carpenter and Sutin, 1983). The genu of the facial nerve is clearly visible in both E18 control and Hoxa3^-/-b3^-/- double mutant embryos in the region devoid of NeuN expression, a nuclear marker for differentiated neurons (Fig. 6A,B, arrow). The early embryonic loss of r5-derived progenitors and precursors of the abducens nucleus is substantiated by the absence of choline acetyltransferase (ChAT)expression in Hoxa3^-/-b3^-/- double mutant embryos (Fig. 6C,D). These findings are supported by an associated reduction in the expression of TuJ1, a pan-neuronal marker, and acetylcholine receptors (AChR), as visualized by α-bungarotoxin (αBTX), in transverse sections of the lateral rectus muscle, innervated normally by the abducens nerve(Fig. 6E-J). The binding ofαBTX in the lateral rectus of Hoxa3^-/-b3^-/- double mutant embryos is consistent with the finding that prepatterning of AChRs occurs in the target muscle independent of motor innervation(Yang et al., 2001). The remaining expression of TuJ1 in the lateral rectus of Hoxa3^-/-b3^-/- double mutant embryos may represent peripheral processes from sensory neurons (i.e. trigeminal ganglia). However, the close proximity of these processes with αBTX suggests a contribution from a motoneuron source, perhaps the aberrant innervation by axons from other cranial motoneurons.

In this report, we provide genetic evidence that the paralogous Hox3 genes influence neuronal identity in two crucial ways. In r6, the Hox3 genes are required to genetically suppress the expression of the r4-specific determinant Hoxb1 (Fig. 7A,B). In ventral r5, the combined functions of Hoxa3 and Hoxb3 are necessary for the specification of Pax6- and Olig2-dependent somatic motoneuron progenitors that give rise to the abducens nucleus(Fig. 7C,D). In their absence,the domain normally occupied by somatic motoneuron progenitors appears to be programmed for a V2 interneuron fate. Together, these data provide evidence that the coordinated activities of Hox3 genes along the AP axis are required to establish the unique identities of r5 and r6.

Hox gene loss-of-function and transgenic regulatory analyses have so far revealed positive regulatory roles for Hox genes in neuronal subtype specification, segmental patterning and crossregulatory interactions between Hox genes during hindbrain patterning(Gavalas et al., 1998; Maconochie et al., 1997; Manzanares et al., 2001; Mark et al., 1993; McClintock et al., 2002; Popperl et al., 1995; Rossel and Capecchi, 1999; Studer et al., 1998; Studer et al., 1996). Single and compound mutations in Hox genes have demonstrated their requirement for the survival of AP-restricted populations of cells to patterning of entire segments of the hindbrain. These experiments provide ample evidence for the role of Hox genes in general cell survival and the cellular diversification within AP-restricted segments. However, they do not sufficiently explain how the Hox genes regulate identity amongst functionally related groups of neurons(i.e. motoneurons) along the AP axis of the neural tube.

Work in Drosophila, however, has provided insight into how the Hox genes may use the mechanism of genetic suppression to determine segmental identity (Hafen et al., 1984; Lewis, 1978; Struhl and White, 1985; Weatherbee et al., 1998). For example, Ubx normally suppresses multiple genes in the wing developmental pathway within the metameric segment that gives rise to the haltere such that in the Ubx mutant these genes are de-suppressed and the more posterior haltere is transformed into the more anterior wing(Weatherbee et al., 1998). The suppression of the wing program by Ubx is analogous to our current observation of the r6-specific suppression of the facial BMN specification program by the Hox3 genes. This important crossregulatory phenomenon among members of the Hox3 and Hox1 paralogous members ensures the individual identity of rhombomeres. Interestingly, the observation that Hoxb1 is suppressed specifically in r6 may reflect a developmental ground state common amongst even-numbered rhombomeres. Thus, as in other serially homologous structures, the periodicity of even- and odd-numbered rhombomeres exhibits common functions (Casares and Mann,2001; Lumsden,1990; Trainor and Krumlauf,2000; Waskiewicz et al.,2002).

The phenotypes of mice harboring independent mutations for the Hox3 paralogs Pax6 and Olig2 suggest that these genes are part of a common regulatory network necessary for determining the fate of SMNs(Ericson et al., 1997; Lu et al., 2002; Osumi et al., 1997; Takahashi and Osumi, 2002; Zhou and Anderson, 2002). Although these genes are necessary for the specification of SMNs, the regulatory process by which they attain this goal is qualitatively different. As described in this study and in a previous report, Hox3 genes and Pax6 are upstream of Olig2 expression(Novitch et al., 2001). In the loss of Hoxa3 and Hoxb3, however, Pax6 is expressed at ectopically high levels in the pSMN domain, suggesting that Hoxa3and Hoxb3 genetically suppress Pax6 expression levels in the pSMN domain. Our previous report also demonstrated the de-suppression of Pax6 expression in the Nkx2.2-expressing BMN progenitor domain in r4 of Hoxb1 mutant embryos(Gaufo et al., 2000). These observations are analogous to the role of the Drosophila Hox gene, Antp, in suppressing the activity of eyeless, the homologue of Pax6 in vivo and in vitro(Plaza et al., 2001). However,a direct interaction between the mammalian Hox and Pax genes remains to be tested.

In contrast to the combined loss of Hoxa3 and Hoxb3, the Pax6 mutation results in the expansion or de-suppression of the ventral Nkx2.2 pV3 domain into the more dorsal SMN progenitor domain in r5 (data not shown) (Ericson et al.,1997; Takahashi and Osumi,2002). Consequently, the loss of Hox3 and Pax6 genes leads to ectopic differentiation of V2 and V3 interneurons, respectively. The comparison of the Hox3 and Pax6 genes demonstrates that they are necessary for the formation of SMNs, but they differ significantly at the level of specifying SMN progenitors. Genetically, Hox3 genes suppress the dorsal Pax6 pV2 domain in the more ventral pSMN domain, whereas Pax6 suppresses the ventral Nkx2.2 pV3 domain in the more dorsal pSMN domain. In contrast to Pax6, the functions of Olig2 appear similar to Hoxa3 and Hoxb3 in its capacity to specify pSMNs. In both Olig2^-/- and Hoxa3^-/-b3^-/- mutant embryos, the loss of SMNs is associated with the de-suppression of Pax6 and the subsequent ectopic differentiation of V2 interneurons(Lu et al., 2002; Takebayashi et al., 2002; Zhou and Anderson, 2002). In both mutants, the more ventral Nkx2.2-expressing V3 progenitor domain appears unaffected, unlike that observed in the Pax6 mutant. The functional similarities between the Hox3 and Olig2 genes with respect to the specification of SMN progenitors in r5 suggest interactions between the pathways mediated by these genes.

The observation that the gene-dose dependent loss of Hoxa3 and Hoxb3 strongly correlates with the loss of Olig2/Hb9 and ectopic Chx10 expression suggests a direct role in SMN fate decisions. Furthermore,the expression of the Hox3 genes among progenitors and differentiating neurons in r5 (Fig. 1F-H, Fig. 4I) supports a possible role for Hox3 genes at later stages of motoneuron differentiation. This hypothesis has received additional support by the recent report that Hoxa3 gain-of-function in chick r1-r4 is sufficient to generate SMNs in these rhombomeres (Guidato et al.,2003). Roles for the Hox3 genes at different stages of motoneuron specification would be consistent with the observation for the multi-level developmental regulation of the Drosophila wing by the Hox gene, Ubx (Weatherbee et al.,1998). However, the question remains whether the effect of Hox3 gene mutations is direct or results from a gradual titration of r5 to a more anterior rhombomere - a hallmark of a homeotic transformation. The present use of conditional mouse models should clarify more precisely the role of the Hox3 genes on distinct stages of motoneuron patterning.

The 13 mammalian Hox paralogs, each containing two to four genes, are expressed in a nested pattern along the AP axis of the neural tube, from the caudal-most tip of the spinal cord to the level of r2 of the hindbrain(Davenne et al., 1999; Economides et al., 2003). To date, the only published reports addressing the role of paralogous Hox genes on neuronal patterning have been the knockouts of the paralogous Hox1 and Hox2 genes (Davenne et al., 1999; Gavalas et al., 1998; Rossel and Capecchi, 1999; Studer et al., 1998). The characterization of the Hox1 paralogous mutants, Hoxa1 and Hoxb1, highlights the basic requirements for these genes in normal hindbrain patterning (Gavalas et al.,1998; Rossel and Capecchi,1999; Studer et al.,1998). During early hindbrain patterning, prior to the closure of the roof plate, Hoxa1 and Hoxb1 are involved in establishing the AP-restricted identities of at least rhombomeres 3, 4 and 5. Equally important to this role is the requirement for Hoxa1 and Hoxb1 in cell proliferation and survival. In the absence of Hoxa1 and Hoxb1, the hindbrain undergoes abnormal programmed cell death associated with deletions of multiple rhombomeres and subsequent reorganization (Rossel and Capecchi,1999). The deletion of multiple hindbrain segments precludes analysis of later patterning events associated with the specification of a multitude of neuronal subtypes along the DV axis.

In contrast to the paralogous Hox1 mutants, the numbers and periodicity of rhombomeres appear to be normal in paralogous Hox2, Hoxa2 and Hoxb2 mutant embryos (Davenne et al., 1999). The normal features of the rhombomeres thus allow for the examination of later events in neuronal patterning. Indeed, the analysis of Hox2 mutant embryos led to the discovery that the program mediated by Hox genes along the AP axis appear to influence later patterning events along the DV axis. The present study, however, defines the possible mechanisms by which Hox genes may control distinct aspects of AP and DV patterning. For example,the functions of Hoxa3 and Hoxb3 appear to influence early neuronal fate decisions by regulating a developmental program common with a SMN-specific determinant, Olig2(Lu et al., 2002; Novitch et al., 2001; Takebayashi et al., 2002; Zhou and Anderson, 2002). However, owing to the extended expression of the Hox3 genes during embryogenesis, their precise role in neuronal specification remains to be characterized. Nevertheless, the direct or indirect regulation of Olig2 expression, a putative transcriptional repressor, by the early actions of Hoxa3 and Hoxb3 highlights the complex interplay between mechanisms of activation and suppression in the progressive specification of motoneurons.

The role of the paralogous Hox1, Hox2 and Hox3 groups in the positive regulation of segmental formation and DV patterning programs explain only in part the genetic mechanism by which segmental neuronal identity is achieved(Davenne et al., 1999; Gavalas et al., 1998; Rossel and Capecchi, 1999; Studer et al., 1998). Combined with the observations revealing the role of Hox genes in genetic suppression along the AP axis, a general mechanism emerges of how neuronal identity is acquired. Remarkably, the fundamental principles of Hox gene function involved in this developmental process are programmed within a single paralogous Hox group (where a distinct combination, a `Hox code', may perform distinct roles). The functions mediated by the Hox3 genes in r5 and r6 may represent a phenomenon reiterated by other Hox paralogs along the entire AP axis of the neural tube. Ultimately, this developmental process could lead to the unique identities of neurons along the entire AP axis of the embryo.

We are grateful to members of the Capecchi laboratory; animal and tissue staff; A. Boulet, M. Hockin, B. Arenkiel and N. Wicha for critical review of the manuscript; L. Oswald for administrative assistance; C. Goridis, H. Takebayashi, T. Jessell and S. Morton for kindly providing reagents; and A. Mansouri and P. Gruss for generously providing Pax6^Sey/+ mice. G.O.G., K.R.T. and M.R.C. are research associate, senior associate and investigator, respectively, of the Howard Hughes Medical Institute.