Hox and Pbx Factors Control Retinoic Acid Synthesis during Hindbrain Segmentation

SUMMARY

In vertebrate embryos, retinoic acid (RA) synthesized in the mesoderm by Raldh2 emanates to the hind-brain neuroepithelium, where it induces anteroposterior (AP)-restricted Hox expression patterns and rhombomere segmentation. However, how appropriate spatiotemporal RA activity is generated in the hindbrain is poorly understood. By analyzing Pbx1/Pbx2 and Hoxa1/Pbx1 null mice, we found that Raldh2 is itself under the transcriptional control of these factors and that the resulting RA-deficient phenotypes can be partially rescued by exogenous RA. Hoxa1-Pbx1/2-Meis2 directly binds a specific regulatory element that is required to maintain normal Raldh2 expression levels in vivo. Mesoderm-specific Xhoxa1 and Xpbx1b knockdowns in Xenopus embryos also result in Xraldh2 downregulation and hindbrain defects similar to mouse mutants, demonstrating conservation of this Hox-Pbx-dependent regulatory pathway. These findings reveal a feed-forward mechanism linking Hox-Pbx-dependent RA synthesis during early axial patterning with the establishment of spatially restricted Hox-Pbx activity in the developing hindbrain.

INTRODUCTION

Retinoic acid (RA), the acidic form of vitamin A, is essential for normal development and organogenesis of the vertebrate embryo. In the early mouse embryo, RA is mainly produced by the biosynthetic enzyme Raldh2 in presomitic mesoderm (PSM), paraxial mesoderm, and lateral plate mesoderm (LPM), from which it emanates to the developing central nervous system (Duester, 2008; Niederreither et al., 1999). In turn, RA binds to nuclear receptors and directly activates target gene expression (Forman and Evans, 1995; Lohnes et al., 1993; Mark et al., 2009). RA acts as a diffusible morphogen forming a posterior-to-anterior activity gradient required for normal rostrocaudal patterning of the spinal cord and hindbrain neuroepithelial segments, or rhombomeres (r) (Glover et al., 2006; Kiecker and Lumsden, 2005; Maden, 2007; Marshall et al., 1992; Niederreither et al., 2000). Rhombomere identity and patterning is mediated by the transcription factors of the Hox gene family, whose activation in the neuroepithelium is directly under RA control (Glover et al., 2006). Hox expression domains are further refined in specific rhombomeres by local RA degradation regulated by the cytochrome p450 family 26 (Cyp26) enzymes (Hernandez et al., 2007; Sirbu et al., 2005; White et al., 2007). Maintaining normal levels of RA is crucial because retinoid excess and deficiency have teratogenic effects, including abnormal hindbrain segmentation and patterning. An outstanding question is how the synthesis of RA is regulated to provide appropriate retinoid levels along the rostrocaudal axis of the developing hindbrain and achieve normal segmentation. However, little is known about how the expression of Raldh2 is coordinated at the transcriptional level to generate appropriate retinoid levels and activate nested Hox gene expression domains with specific rostral boundaries in the developing hindbrain.

In this study, we found that Raldh2 mesodermal expression is itself under the direct transcriptional control of Hox, Pbx, and Meis factors in vivo. In Pbx1/Pbx2 null mice, Raldh2 levels are not properly maintained, resulting in progressive reduction of endogenous retinoid activity. In Hoxa1/Pbx1-deficient embryos, Raldh2 is also significantly downregulated at early somite stage, resulting in caudal shift of hindbrain RA activity and an RA-deficient rhombomere phenotype that is partially rescued by exogenous RA administration. Xhoxa1 and Xpbx1 mesoderm-specific knockdowns in Xenopus embryos also resulted in Raldh2 down-regulation and induced hindbrain patterning defects similar to those of mouse compound mutants. By chromatin immunoprecipitation (ChIP) in mouse embryos, we identified a specific Raldh2 enhancer containing a Hox-Pbx bipartite element bound by a Hoxa1-Pbx1/2-Meis2 complex and required to maintain normal expression levels in the context of the endogenous Raldh2 promoter. In the Raldh2-negative (Raldh2⁻) head of early stage embryos this enhancer is selectively bound by Suz12, a member of the Polycomb Repressive Complex 2 (PRC2), correlating with an enrichment of the H3K27me3 mark associated with facultative heterochromatin. These findings reveal a molecular feed-forward mechanism linking Hox-Pbx-dependent RA synthesis in mesoderm with the establishment of Hox-Pbx neuroepithelial activity during hindbrain segmentation.

RESULTS

Pbx1/2-Dependent Maintenance of Raldh2 Expression and Retinoid Activity

Pbx genes encode three-amino-acid loop extension (TALE) class homeodomain (HD) transcription factors that form heteroligomeric complexes with a subset of Hox and Meis/Prep HD proteins and regulate a variety of developmental processes (Mann and Chan, 1996; Moens and Selleri, 2006). Compound Pbx1^−/−/Pbx2^−/− (referred to as Pbx1/2 null) embryos exhibit multiple organogenesis defects and eventually die by E10.5 (Capellini et al., 2006). Specifically, Pbx1/2 null mutants display abnormal turning, shortened bodies, abnormal development of forebrain and limb buds, a dilated heart, hypoplastic posterior branchial arches, and somite/vertebral patterning defects (Capellini et al., 2008; Selleri et al., 2001; Stankunas et al., 2008). These developmental defects are markedly similar to those described for Raldh2 deficient mouse embryos (Niederreither et al., 1999). Thus, the Pbx1/2 null mutant pleiotropic phenotype may be partly due to reduced endogenous retinoid levels.

To test this hypothesis, we first investigated Raldh2 expression in Pbx1/2 null embryos. In E7.75 single and Pbx1/2 null mutants, Raldh2 spatial distribution and expression levels did not appear to be significantly altered, as compared to wild-type controls (see Figure S1 available online). In contrast, at E8.75 Raldh2 transcript levels were significantly decreased in Pbx1/2 null embryos; moreover, Raldh2 expression was selectively absent from the LPM just posterior to the cardiac field and the anterior-most somites of the mutants (Figure 1D). By E9.0-E10.0, Raldh2 expression was strongly downregulated in the somitic mesoderm (Figures 1H, 1L, and 1L′). Accordingly, endogenous retinoid activity was severely depleted in Pbx1/2 null mutants mated to RARE::lacZ reporter mice (Rossant et al., 1991) (Pbx1/2;RARE::lacZ) (Figure 1Q). A progressive reduction of Raldh2 expression and RARE::lacZ reporter β-gal activity was already evident in single Pbx1^−/− as well as compound Pbx1^−/−/Pbx2^+/− and Pbx2^−/−/Pbx1^+/−, though not in Pbx2^−/− single mutants (Figures 1O and 1P and data not shown). Notably, treatment of Pbx1/2 null embryos with exogenous RA (10 mg/kg) at E8.5 partially rescued the mutant phenotype and yielded embryos with normal turning (3/9; 33%) (Figure S1), suggesting that at least part of Pbx1/2 function is mediated through the control of RA production. These data strongly point to a synergistic genetic interaction between Pbx1 and Pbx2 for temporal maintenance of Raldh2 transcriptional levels and control of endogenous retinoid signaling, with a main requirement for Pbx1.

Figure 1. Pbx1 and Pbx2 Are Required for Maintenance of Mesodermal Raldh2 Expression and Retinoid Activity.

(A–L′) Whole-mount in situ hybridization showing Raldh2 expression in control (A, E, I, and I′), Pbx1^−/− (B, F, J, and J′), Pbx1^+/−/Pbx2^−/− (C, G, K, and K′), and Pbx1^−/−/Pbx2^−/− (D, H, L, and L′) embryos at E8.75 (A–D), E9.0 (E–H), and E9.5 (I–L and I′–L′).

(M) Normal Uncx4.1 expression in E9.5 Pbx1^−/−/Pbx2^−/− mutants.

(N–Q) β-gal staining of E10.0 RARE::lacZ (N), Pbx1^−/−;RARE::lacZ (O), Pbx1^+/−/Pbx2^−/−;RARE::lacZ (P), and Pbx1^−/−/Pbx2^−/−;RARE::lacZ (Q). h, heart; lpm, lateral plate mesoderm; s, somites.

See also Figure S1.

Hoxa1/Pbx1-Dependent Regulation of Raldh2 Expression and Retinoid Activity

Pbx factors are essential DNA-binding partners of Hox transcription factors (Mann and Chan, 1996; Moens and Selleri, 2006; Popperl et al., 2000; Remacle et al., 2004). Hoxa1 is activated in the posterior primitive streak and, in turn, in presomitic, somitic, and lateral plate mesoderm (Deschamps and van Nes, 2005; Murphy and Hill, 1991). RA-mediated activation of Hoxa1 in the overlying neuroectoderm is induced through a 3′ retinoic acid responsive enhancer (3′RARE) (Dupe et al., 1997). Under RA response, the Hoxa1 expression domain spreads up into the presumptive r3 territory and subsequently sets its border at the r3/r4 boundary, providing the earliest sign of molecular segmentation in the mouse hindbrain (Makki and Capecchi, 2010). Hoxa1 inactivation resulted in hindbrain segmentation and rhombomere patterning defects (Carpenter et al., 1993; Mark et al., 1993) that resemble vitamin A partial deficiency phenotypes. Hoxa1 mesodermal expression in the early embryo precedes Raldh2 activation (Murphy and Hill, 1991; Niederreither and Dolle, 2008). Thus, Pbx factors may cooperate with Hoxa1 in mesoderm to regulate the early phase of Raldh2 expression, and, in turn, the early availability of retinoids diffusing into the developing hindbrain prior to segmentation.

To test this hypothesis, we analyzed Raldh2 expression in compound Hoxa1^−/−/Pbx1^−/− (referred to as Hoxa1/Pbx1 null) mutants. Several 3′ Hox genes are sequentially activated in a temporally collinear manner in the posterior end of the embryo (Deschamps and van Nes, 2005; Soshnikova and Duboule, 2009), thus resulting in potential functional redundancy for the regulation of Raldh2 levels. Thus, we focused on early somitogenesis stages, when the lack of Hoxa1 and its cofactor Pbx1 may be expected to have a major functional impact. In E8.0–E8.25 (3–4 somite stage) Hoxa1/Pbx1 null embryos, Raldh2 transcript levels were significantly lower than in controls, specifically in PSM and somitic mesoderm (Figures 2A and 2B). Moreover, the rostral Raldh2 expression domain in LPM, just posterior to the heart field and adjacent to the presumptive hindbrain, was selectively downregulated (arrowhead, Figure 2B).

Figure 2. Hoxa1- and Pbx1-Dependent Regulation of Mesodermal Raldh2 Expression and Retinoid Activity Boundary in the Early Hindbrain.

(A and B) Whole-mount in situ hybridization shows Raldh2 expression in E8.25 control (A) and Hoxa1^−/−/Pbx1^−/− (B) embryos (ventral view).

(C–F) β-gal staining of RARE::lacZ (C and E) and Hoxa1^−/−/Pbx1^−/−;RARE::lacZ (D and F) embryos at E8.0 (C and D) and E8.25 (E and F).

(G and H) Double in situ hybridization for Otx2 and Hoxb1 in E8.25 control (G) and Hoxa1^−/−/Pbx1^−/− (H). lpm, lateral plate mesoderm; psm, presomitic mesoderm; s, somites.

We next analyzed endogenous retinoid activity in Hoxa1/Pbx1;RARE::lacZ null mutants by β-gal staining. At 0–2 somite stage, β-gal activity was severely downregulated in Hoxa1/Pbx1 null embryos (Figure 2D). By the 3–4 somite stage, the overall β-gal staining levels were still significantly lower than in controls (Figures 2E and 2F). Moreover, the rostral LPM domain of retinoid activity diffusing into the heart field was either missing or severely reduced (double arrowheads), in agreement with Raldh2 in situ hybridization results (compare Figures 2B and 2F). Notably, the anterior boundary of β-gal staining in the presumptive hindbrain was posteriorly shifted in Hoxa1/Pbx1 null mutants, as compared to controls (single arrowhead, Figures 2E and 2F). These data revealed that Hoxa1 and Pbx1 genetically interact in regulating early mesodermal Raldh2 expression and setting the RA activity boundary in the presumptive hindbrain neuroepithelium.

RA-Deficient Hindbrain Phenotype of Hoxa1/Pbx1 Null Mutants

We next analyzed the hindbrain segmentation pattern in Hoxa1/Pbx1 null embryos (Figures 3A–3N). In Pbx1 null mutants, the Hoxb1⁺ r4 territory was normally positioned (Figure 3C). In contrast, Hoxa1 inactivation results in a smaller and slightly caudally displaced Hoxb1⁺ r4 (Figure 3B) (Barrow et al., 2000; Carpenter et al., 1993; Mark et al., 1993) because of an early requirement of Hoxa1 for Hoxb1 transcription in presumptive r4 (Di Rocco et al., 2001). Late Hoxb1 expression in r4 is instead maintained by an autoregulatory mechanism (Ferretti et al., 2005; Popperl et al., 1995). Hoxb1 is also a direct RA target through specific RAREs (Marshall et al., 1992, 1994; Studer et al., 1994, 1998). RA emanating from the mesoderm is required to position the anterior limit of Hoxb1 expression at the r3/r4 border and to repress Hoxb1 in r3 and r5 (Marshall et al., 1994; Studer et al., 1994, 1998).

Figure 3. Retinoic Acid-Deficient Phenotype of Hoxa1/Pbx1 Mutant Hindbrain and Rescue by Exogenous RA.

(A–H) Whole-mount in situ hybridization for Hoxb1 in E9.0-untreated (A–D) and RA-treated (E–H) control (A and E), Hoxa1^−/− (B and F), Pbx1^−/− (C and G), and Hoxa1^−/−/Pbx1^−/− (D and H) embryos.

(I–J′) Hoxa2 and En2 expression in E9.0 control (I and I′) and Hoxa1^−/−/Pbx1^−/− (J and J′) embryos.

(K–N) Krox20 and Kreisler expression in E9.0 control (K and M) and Hoxa1^−/−/Pbx1^−/− (L and N) embryos.

(O–T) Summary diagrams illustrating the Hoxa1/Pbx1 mutant RA-deficient phenotype and its RA-mediated rescue. BA2, second branchial arch; NCC, neural crest cells; ot, otic vesicle; r, rhombomere; RA, retinoic acid.

In E8.25 Hoxa1/Pbx1 null mutants, the Hoxb1 expression border in presumptive hindbrains retreated caudally (compare Otx2 and Hoxb1 expression domains; Figure 2H), whereas the remainder of its expression domain appeared normal. At E9.0, the Hoxb1 expression domain was shifted posteriorly beyond the otocyst, lacked sharp anterior and posterior borders, and was larger in Hoxa1/Pbx1 than Hoxa1 null mutants (r4*, compare Figures 3B and 3D). In keeping with the marked posterior shift of r4, r4-derived Hoxa2⁺ neural crest cells (NCC) migrated caudally to the otic vesicle, rather than rostrally (Figures 3I′ and 3J′). A posterior shift of r3, which abnormally faced the otocyst, and lack of r5 were also observed, as assessed with the r3/r5- and r5/r6-specific markers Krox20 and Kreisler, respectively (Figures 3K–3N; note that the r3-specific Krox20⁺ expression domain is severely downregulated). Lastly, Hoxa1/Pbx1 null mutants displayed a prominent posterior expansion of r2, as assessed with a Hoxa2 probe, that was not present in either single Hoxa1 or Pbx1 null embryos (Figures 3I and 3J).

In summary, in the hindbrain of Hoxa1/Pbx1, though not single, null mutants we observed a posterior displacement of anterior rhombomere identities at the expense of the r5-r6 territory, similar to the phenotype observed in partial RA-deficiency, as posterior rhombomeres require higher RA signaling than rostral ones to be positioned and specified (Dupe and Lumsden, 2001; Gavalas, 2002; Niederreither et al., 2000). In this respect, the early reduction of mesodermal Raldh2 expression and hind-brain RA activity in Hoxa1/Pbx1 null embryos (Figure 2) strongly predicts that at least part of the above hindbrain phenotype may result from partial RA deficiency induced by early roles of Hoxa1 and Pbx1 in mesoderm, distinct from their later roles in neuroepithelium.

Mesoderm-Specific Xhoxa1 and Xpbx1b Knockdown Results in Xraldh2 Downregulation and an RA-Deficient Phenotype in Xenopus Embryo Hindbrain

To test Hox-Pbx-dependent conservation of Raldh2 regulation across species, we used a mesoderm-specific morpholino (MO)-mediated knockdown approach in Xenopus embryos. We took advantage of the established fate map of individual blastomeres of the Xenopus embryo (Hirose and Jacobson, 1979; Moody, 1987; Moody and Kline, 1990) and injected antisense MOs against Xpbx1b (Maeda et al., 2002) and/or Xhoxa1 (McNulty et al., 2005) selectively in the left V2.2 blastomere of the 16-cell stage embryo and compared to the uninjected side as internal control. The V2.2 blastomere and its progeny largely contribute to somitic mesoderm and LPM, though not, or only marginally, to hindbrain and nervous system (Hirose and Jacobson, 1979; Moody, 1987; Moody and Kline, 1990). To further screen for injected embryos devoid of morphant cells in the nervous system, we coinjected MOs in V2.2 with mRNAs for Red Fluorescent Protein (RFP) and nuclear lacZ (Figure 4). Embryos were sorted at the late neurula stage (stage 17–18) for the distribution of RFP fluorescence (n = 634; Figures 4A–4D) by selecting only those displaying significant unilateral RFP expression in paraxial mesoderm and LPM, but not in nervous system, as confirmed on tissue sections (e.g., Figure 4D). Moreover, a subset of the selected embryos was additionally stained by salmon-gal prior to further processing for in situ hybridization (n = 61/634), and consistently confirmed the lack of injected cells in the nervous system (Figure 4J and data not shown).

Figure 4. Mesoderm-Specific Xpbx1 and Xhoxa1 Knockdown in Xenopus Embryos.

(A) Diagram of marginal zone V2.2 Xenopus blastomere injection at 16-cell stage, and fate of injected blastomere in stage 17 neurula.

(B) Whole-mount detection of RFP in left V2.2 blastomere-injected embryo.

(C) Merge of bright field and fluorescence pictures of the embryo in (B).

(D) Cross-section showing selective RFP labeling in somites (s) and lateral plate mesoderm (lpm).

(E–I) Whole-mount double in situ hybridization for Xraldh2 and Xkrox20 in control-MO (E), Xpbx1b-MO (F and G), and Xpbx1b-MO;Xhoxa1-MO (H and I) left V2.2 blastomere-injected embryos.

(J) Nuclear-salmon-gal staining (red cells) of Xpbx1b-MO;Xhoxa1-MO V2.2-injected embryos indicate morphant cell localization in mesoderm. Xkrox20 expression shows r3* posteriorization and r5* loss on the injected side.

(K and L) Correlation between Xraldh2 downregulation and hindbrain phenotype in Xpbx1b-MO and Xpbx1b;Xhoxa1-MO injected embryos (mosaic plots). Relative phenotype severity is color-coded. Phenotype frequencies (y axis) are compared to levels of Raldh2 (x axis). Synergistic action of Xpbx1b-MO;Xhoxa1-Mo results in rising the frequencies of Xraldh2 severe reduction or loss and hindbrain patterning defects. MO, morpholino; r, rhombomere; RFP, Red Fluorescent Protein.

See also Figure S2.

At stage 17–18, Xraldh2 is mainly expressed in presomitic and somitic mesoderm, and LPM in the middle part of the trunk (Chen et al., 2001). Although embryos injected with control MO did not display molecular changes (n = 68; Figure S2; Figure 4E), the majority of embryos injected with a Xpbx1b MO in the mesoderm displayed a reduction of Xraldh2 expression (n = 92/128; 72%) that ranged from significant (n = 80/128; 63%) to severe in some cases (n = 12/128; 9%) (Figure S2; data not shown; see also Figures 4F and 4G). Such variability is likely due to mosaicism of MO distribution inherent to the knockdown approach and/or to potential functional redundancy with other Pbx factors. At any rate, these findings underscored an important role of Xpbx1b in maintaining normal Xraldh2 levels in the mesoderm, similarly to mouse (Figure 1).

We then investigated the potential impact of mesodermal knockdown of Xpbx1b on hindbrain patterning. We predicted that the variable impairment of Xraldh2 expression could result in a range of RA-deficient rhombomere phenotypes. In situ hybridization with Xkrox20 of Xpbx1b MO-injected embryos (n = 135) revealed rhombomere abnormalities that ranged from r5 reduction to absence in some cases (n = 74/135; 55%; data not shown; see also Figures 4F and 4G). A fraction of the injected embryos (n = 14/135; 10%) additionally displayed a one-rhombomere posterior shift of r3 and r5 (data not shown; see also Figures 4H–4J), indicating a more severe RA deficiency. Direct correlation between the extent of Xraldh2 reduction and the severity of the observed hindbrain defects was demonstrated by simultaneous in situ hybridization with Xraldh2 and Xkrox20 of an additional set of Xpbx1b MO-injected embryos (n = 100; Figures 4F and 4G; see also Figure 4K for a mosaic plot of the distribution of hindbrain phenotypes versus Xraldh2 expression levels). In summary, Xpbx1b selective knockdown in mesoderm has a direct impact on Xraldh2 expression and induces hindbrain abnormalities that are expected features of partial RA deficiency.

We next asked whether Xpbx1b could functionally synergize with Xhoxa1 for mesodermal regulation of Xraldh2. A previously described Xhoxa1 MO induced only subtle hindbrain defects when injected at the four-cell stage throughout the left side of the embryo, likely because of functional redundancy with other Hox1 paralogs (McNulty et al., 2005). The injection of this Xhoxa1 MO in mesoderm (n = 26) did not significantly alter Xraldh2 expression or hindbrain patterning (n = 21/26; data not shown). A mild decrease of Xraldh2 expression was scored in the remaining 5 of 26 injected embryos, as compared to the uninjected side, that was not, however, sufficient to induce hindbrain patterning defects (data not shown).

Coinjection of MOs against both Xhoxa1 and Xpbx1b in mesoderm strongly enhanced the effects of the Xpbx1b knockdown (Figure S2; Figures 4H and 4I). Increasingly severe downregulation of Xraldh2 expression was observed in the vast majority of coinjected embryos (n = 98/110; 89%; Figure S2; see also Figures 4H and 4I). Moreover, the fraction of embryos displaying a drastic Xraldh2 reduction was notably increased (n = 46/110; 42%; Figure S2; see also Figures 4H and 4I), as compared to singly Xpbx1b MO-injected embryos. These findings reveal a synergistic role of Xhoxa1/Xpbx1b in Xraldh2 regulation in Xenopus embryo mesoderm.

Accordingly, we scored more frequent and/or penetrant hind-brain abnormalities in Xpbx1b/Xhoxa1 MO-coinjected embryos (n = 93) than in singly Xpbx1b or Xhoxa1 MO-injected embryos. These phenotypes ranged from strong reduction/absence of r5 (n = 52/93; 56%) to a 1–2-rhombomere posterior shift of r3 with loss of r5 (n = 18/93; 19%), as assessed with the Xkrox20 probe (data not shown; see also Figures 4H–4J). Importantly, the specimen in Figure 4J was additionally costained with salmon-gal to directly detect the injected cells (red cells), thus demonstrating that a strong hindbrain phenotype (e.g., a posterior shift of r3 with loss of r5; Figure 4J) can be induced by the selective injection of Xpbx1b/Xhoxa1 MOs in mesoderm.

Direct correlation between the extent of Xraldh2 reduction and the severity of the hindbrain phenotype was further demonstrated by simultaneous in situ hybridization with Xraldh2 and Xkrox20 of Xpbx1b/Xhoxa1 MO-injected embryos (n = 42; Figures 4H and 4I). Comparison of the mosaic plots in Figures 4K and 4L allows us to directly assess the synergistic role of Xpbx1b/Xhoxa1 and the impact of their knockdowns in mesoderm on Xraldh2 expression and hindbrain phenotype, as compared to singly Xpbx1b MO-injected embryos.

In sum, our mesoderm-specific knockdown approach demonstrated the requirement for Xpbx1b to maintain normal levels of Xraldh2, and its synergistic functional interaction with Xhoxa1. It also demonstrated that selective downregulation of Pbx-Hox factors in the mesoderm, independently of their roles in neuroepithelium, is sufficient to induce abnormal hindbrain segmentation. Such hindbrain defects phenocopied the effects of RA-deficiency (Dupe and Lumsden, 2001; Gavalas, 2002), and were notably similar, at least in part, to those observed in Hoxa1/Pbx1 null mice (Figure 3).

Rescue of Mouse Hoxa1/Pbx1 Mutant Hindbrain by Exogenous RA Treatment

We then asked which features of the Hoxa1/Pbx1 null mouse phenotype could be specifically ascribed to the mesodermal Hoxa1/Pbx1-dependent decrease of retinoid synthesis (Figure 2), as opposed to those resulting from a direct role of Hoxa1 and Pbx1 in hindbrain neuroepithelium. The administration to double mutants of a subteratogenic dose of exogenous RA (5 mg/kg at E8.0 [5RA-8]; Pasqualetti et al., 2001) may be expected to rescue the former, though not the latter, aspects of the Hoxa1/Pbx1 null hindbrain phenotype.

Remarkably, 5RA-8 treatment of Hoxa1/Pbx1 null mutants was sufficient to rescue the AP position of the r4 Hoxb1⁺ domain and shift it beyond the rostral aspect of the otocyst (4/4; 100%) (Figures 3E–3H). Moreover, the Hoxb1⁺ domain was caudally shortened in 5RA-8-treated E9.5 mutant embryos, as compared to untreated double mutants (Figure 3H), indicating partial rescue of RA-mediated repression, which normally restricts Hoxb1 expression caudal to r4 (Studer et al., 1994), and of posterior rhombomere patterning. However, 5RA-8 treatment was not able to ectopically induce Hoxb1 in r2 of single Hoxa1 or Hoxa1/Pbx1 null mutants (Figures 3F and 3H), unlike in wild-type or single Pbx1 mutant embryos (Figures 3E and 3G), confirming that ectopic Hoxb1 activation specifically requires Hoxa1 function in the neuroepithelium (Zhang et al., 1994; Di Rocco et al., 2001).

Thus, the hindbrain analysis of RA-rescued Hoxa1/Pbx1 mutants provided strong additional evidence that the changes in rhombomere rostrocaudal position in untreated mutants may in part result from the Hoxa1/Pbx1-dependent decrease of mesodermal RA synthesis (Figure 2), independently of Hoxa1/Pbx1 role in neuroepithelium.

A Specific Raldh2 Regulatory Element Is Bound In Vivo by a Hoxa1-Pbx1/2-Meis2 Complex

Hox overexpression in chicken micromass cultures indicated the potential for direct regulation of Raldh2 (Kuss et al., 2009). Therefore, we sought to assess whether Raldh2 transcription in early embryonic mesoderm is directly regulated by Hoxa1 and Pbx1/2 factors.

In silico analysis revealed four conserved regions—namely, E1 (334 bp), E2 (980 bp), E3 (514 bp), and E4 (938 bp)—located upstream of the Raldh2 transcription start site (E1 and E2) and in its first intron (E3 and E4), respectively (Figure 5A). Putative Pbx-Hox (PH) binding sites were identified in all such regions, whose sequences shared high conservation with previously described PH sites from Hox-Pbx target genes (Figure 5B; Figure S3). In vitro binding electrophoretic mobility shift assays (EMSAs) using different combinations of in vitro translated Hoxa1, Hoxb1, Pbx1a long isoform (Monica et al., 1991), Pbx1b short isoform (Monica et al., 1991), and Pbx2 proteins revealed that all PH sites could bind Pbx-Hox paralog group 1 (PG1) heterodimers (Figure 5C; Figure S3). Competition assays with cold wild-type or point-mutated oligonucleotides, or with specific antibodies against Pbx or Hox PG1 proteins, further demonstrated PH site specificity and Pbx-Hox requirement for in vitro binding (Figure S3).

Figure 5. In Vivo Direct Regulation of Raldh2 by Hoxa1-Pbx1/2-Meis2.

(A) Mouse Raldh2 locus (chr9:71,055,462-71,092,461, UCSC Mouse Browser). Conservation plot across vertebrate species (green peaks); blue peaks indicate highest conservation. Blue boxes (E1, E2, E3, and E4) highlight conserved regions containing Pbx-Hox (PH) binding sites.

(B) Sequence comparison of PH elements from known targets and Raldh2 E1PH, E2PH, E3PH, E4PH1, and E4PH2; blue letters indicate divergency from PH consensus, and variable bases are in green.

(C) In vitro binding EMSA with translated Pbx1a, Pbx1b, Pbx2, Meis2, Hoxa1, and Hoxb1 on the E3PH-containing oligonucleotide (red sequence).

(D) A ternary complex (TC) comigrating band binds to E3PH probe in nuclear extracts from E8.5 embryo posterior part (red box inset, E8.5 NE). Binding specificity is assessed using specific antibodies. Hoxa1(b1)/Pbx1a/Meis2 in vitro-translated proteins were used as molecular weight control of TC.

(E–J) In vivo chromatin immunoprecipitation (ChIP) from “head” (red box inset, E, G, and I) and “body” (red box inset, F, H, and J) of E8.5 embryos. Specific antibodies against Pbx, Meis2 (E and F), Hoxa1 (F), trimethylated histone H3 lysine 4 (H3K4me3), trimethylated histone H3 lysine 27 (H3K27me3) (G and H), and Suz12 (I and J) were used. In all ChIP assays, specificity was tested by nonspecific primers outside the conserved regions (OUT, in (A). Rabbit IgG is a control for amplification specificity. Fold enrichment over IgG is plotted. Bars represent mean ± SEM; *p < 0.01, t test. See also Figure S3.

(K and L) eGFP expression from transgenic E8.5 embryos carrying a Raldh2 BAC construct (not in scale) recapitulates endogenous Raldh2 expression pattern (K). Mutation of E3PH (E3ΔPH) in the Raldh2 BAC causes eGFP downregulation (L). DC, dimeric complex; Lys, reticulocyte lysate endogenous binding activity; SS, supershifted band. See also Figure S4.

Pbx-Hox binding and transcriptional activity can be enhanced by Prep or Meis proteins, which facilitate the formation of transcriptionally active ternary complexes on PH sites (Ferretti et al., 2000; Jacobs et al., 1999). Thus, we assessed the ability of Hox-Pbx-Meis/Prep ternary complexes to bind the Raldh2 PH sites in vitro. Specific Hoxa1-Pbx1a(b)-Meis2 and Hoxb1-Pbx1a(b)-Meis2 ternary complexes formed only on the PH element within the E3 region (E3PH), though not on those in E1, E2, and E4 (Figure 5C; Figure S3). The establishment of a ternary, as opposed to multimeric, complex on the E3PH site was confirmed by using a combination of Pbx1a and Pbx1b, together with Hoxa1 (or Hoxb1) and Meis2 (Figure S3). Mutations in Hoxa1 DNA-binding or hexapeptide (Pbx-binding) domains disrupted the formation of the ternary complex, showing that complex assembly requires Hoxa1 binding to both Pbx1 and E3PH (Figure S3). Lastly, binding of nuclear extracts from posterior part of E8.5 embryos (inset, red box) to the E3PH oligonucleotide also resulted in the formation of a specific ternary complex containing Pbx, Meis2, and Hoxa1, that was super-shifted by specific antibody competition (Figure 5D and data not shown).

To assess whether Hoxa1, Pbx1, and their Meis2 cofactor could bind the E3PH site in vivo, we carried out chromatin immunoprecipitation (ChIP) on E8.5 mouse embryos. We compared the posterior “body” that includes the mesodermal Raldh2⁺ domains (inset, Figure 5F) to the anterior “head” (inset, Figure 5E) that is Raldh2⁻ at this stage. qPCR from “body” immunoprecipitated chromatin using anti-Pbx (pan-Pbx) and anti-Meis2 antibodies, demonstrated Pbx1 and Meis2 enrichment at Raldh2 E3 and E4, though not E2 and E1, PH site-containing elements, respectively (Figure 5F and data not shown). ChIP with anti-Hoxa1 antibody demonstrated Hoxa1 enrichment only at Raldh2 E3 (Figure 5F), supporting the in vitro data showing the formation of a ternary complex only on E3PH (Figure 4C; Figure S3; data not shown). In summary, despite the presence of multiple potential regulatory elements containing PH-binding sites at the Raldh2 locus, the ChIP data in E8.5 embryos revealed in vivo selectivity for binding of all three Hoxa1, Pbx1/2, and Meis2 proteins only to E3 (Figure 5).

The E3 Pbx-Hox Element Is Necessary for In Vivo Transcriptional Regulation of Raldh2

The ChIP data revealed that in the “head” part of the embryo, Hoxa1, Pbx1/2, or Meis2 were not bound to any of the E1–E4 regions (Figure 5E; see below). This suggested that the E3 element accessibility may be related to the transcriptionally active or inactive state of Raldh2, which in turn may be determined by distinct epigenetic configurations of the chromatin in Raldh2⁺ versus Raldh2⁻ tissues at this specific locus.

We therefore analyzed the ChIP patterns of trimethylated histone H3 lysine 4 (H3K4me3) and trimethylated histone H3 lysine 27 (H3K27me3) at the Raldh2 locus from E8.5 “body” and “head” embryonic regions, respectively (Figures 5G and 5H). H3K4me3 is catalyzed by trithorax-group (trxG) proteins and primarily associated with transcriptionally active chromatin regions at the start of transcription, whereas H3K27me3 is mainly catalyzed by Polycomb-group (PcG) proteins and associated with stable transcriptional repression (Ruthenburg et al., 2007). In ChIP qPCR assays from embryonic “bodies,” we found a selective enrichment of H3K4me3 at the E3 element (Figure 5H and Figure 6E). Indeed, E3 is located proximal to (within about 1 kb of) the Raldh2 transcription start site. In contrast, in chromatin obtained from “heads” the E3 element, though not E1, E2, or E4, was significantly enriched with H3K27me3 (Figure 5G, Figure 6E, and data not shown). This suggested that in tissues not actively expressing Raldh2, its transcription may be silenced by PcG activity. Accordingly, we found a strong enrichment of Suz12, a core PRC2 member (Pasini et al., 2004), at E3, though not at E1, E2, or E4 (Figure 5I and data not shown), correlating with the distribution of the H3K27me3 mark (compare Figures 5G and 5I). Moreover, Suz12 was not significantly enriched at E1–E4 in chromatin from Raldh2⁺ “bodies” (Figure 5J), thus correlating with the transcriptionally active status of Raldh2.

Figure 6. Transcriptional Feed-Forward Model of Pbx-Hox-Mediated Raldh2 Mesodermal Regulation and Induction of Hox Expression in Early Hindbrain Neuroepithelium.

(A) Pbx and Hox factors regulate Raldh2 expression levels and maintenance in presomitic, somitic, and lateral plate mesoderm (red boxed inset) in early stage embryo. The resulting graded RA activity (red triangle) diffuses to neuroepithelium, drives Hox paralog group 1 activation, and sets their rostral boundary in the hindbrain. Retinoid signaling feedback on Hox/Pbx expression maintenance in mesoderm may also occur (e.g., Lohnes et al., 1994) (red curved arrow).

(B) In Hoxa1/Pbx1 null embryos, Raldh2 expression is lower in presomitic and somitic mesoderm (light blue) and absent from anterior lateral plate mesoderm. Decreased expression results in diminished RA activity (small triangle, inset) and posterior shift of Hoxb1 anterior boundary in the neuroepithelium (dashed line indicates normal position, compare with A).

(C and D) RA- and Hox/Pbx-dependent regulation of Hoxa1 and Hoxb1 in presumptive hindbrain neuroepithelium and r4. Panels integrate known pathways of transcriptional regulation (e.g., Dupe et al., 1997; Popperl et al., 1995; Studer et al., 1998) with current findings. In (C), Pbx1/Hoxa1 mesodermal activity controls Raldh2 expression levels and in turn the production of RA diffusing to the adjacent neuroepithelium (1). RA directly activates Hoxa1 and Hoxb1 in neuroepithelium up to r4 through retinoic acid response elements (RAREs) (2). In turn, Hoxa1/Pbx1- and Hoxb1/Pbx1-mediated cross- and auto-regulatory transcriptional mechanisms, respectively, maintain Hoxb1 r4 expression levels (3 and 4). In (D), Hoxa1/Pbx1 loss in mesoderm results in lower Raldh2 expression levels, consequent lower RA activity (1), and reduced Hoxb1 activation in neuroepithelium (2). Lack of Hoxa1 and Pbx1 in mutant neuroepithelium further impairs the establishment of normal levels of Hoxb1 expression (3 and 4).

(E) In Raldh2-negative (Raldh2⁻) tissue in the early embryo, the H3K27me3 mark on the Raldh2 E3 enhancer and Suz12 binding reveal PcG-mediated repression (PRC2 presence at E3, but not at E1, E2, or E4, is depicted). In Raldh2-positive (Raldh2⁺) tissue, the repressive mark at E3 is replaced by H3K4me3, associated with an active chromatin state. The locus is accessible for direct Pbx1/2, Hoxa1 and Meis2 binding and transcriptionally active. The binding of E4 by Pbx-Meis, though not Hoxa1, suggests the involvement of additional transcription factors, including Hox members from other paralog groups, in Raldh2 regulation.

Next, we tested the potential of E3 to drive transcriptional activity in vivo and its dependence on the PH site. Coelectroporation in chick embryos of Hoxa1 with lacZ constructs carrying either wild-type E3 (E3::lacZ) or E3 with a mutated PH site (E3ΔPH::lacZ) demonstrated Hoxa1-mediated E3 trans-activation in an in vivo heterologous system and the requirement for PH site integrity for such trans-activation (Figure S4). To investigate the spatial pattern driven by the Raldh2 E3 enhancer in the mouse, we generated mouse transgenic embryos carrying the E3::lacZ construct (Figure S4). In E8.5 embryos, the reporter expression driven by the E3 element was spatially restricted to the posterior part of the embryo, similar to endogenous Raldh2. Reporter expression was detected in the PSM and strongly throughout the dorsoventral extent of the neural tube, up to a rostral border in the posterior hindbrain (Figure S4). Even though ectopic, as compared to the endogenous Raldh2 expression pattern, the observed spatial domain of enhancer activity indicated that, in isolation from its surrounding genomic sequences, E3 behaves as a “Hox-regulated” enhancer driving spatially restricted AP expression. In transgenic mouse embryos carrying the PH mutated construct (E3ΔPH::lacZ), reporter expression in PSM and neural tube was almost abolished (Figure S4), thus demonstrating that E3 in vivo transcriptional activity is strictly dependent on PH site integrity.

To assess the role of the E3PH element within the intact Raldh2 promoter, we made suitable constructs for transgenic mouse analysis by BAC (bacterial artificial chromosome) recombineering (Liu et al., 2003). We first generated a construct that carried the eGFP reporter in-frame to the Raldh2 ATG translation start codon (BAC Raldh2::eGFP), containing a 160 kb DNA insert spanning the entire murine Raldh2 locus, thus likely containing all the regulatory elements to achieve normal in vivo Raldh2 transcriptional regulation. Indeed, when tested in transient transgenic assays in E8.5 (5–6 somite stage) mouse embryos, the BAC Raldh2::eGFP displayed an eGFP expression pattern faithfully reproducing endogenous Raldh2 expression (n = 5/5; Figure 5K). We then mutated the E3PH element and generated the BAC Raldh2-E3ΔPH::eGFP construct. In E8.5 transgenic embryos, the E3ΔPH mutation resulted in severe eGFP downregulation (n = 5/7; Figure 5L; eGFP decrease, albeit less severe, was also observed in the two remaining embryos; data not shown).

In summary, these data demonstrate a fundamental role of the E3PH element in the context of the entire Raldh2 promoter in maintaining normal Raldh2 expression levels in the early embryo.

DISCUSSION

We demonstrate that Hoxa1 and Pbx1 synergistically regulate the levels of mesodermal Raldh2 expression in the early embryo and, in turn, control endogenous RA levels available for normal hindbrain segmentation. This conclusion is supported by: (1) the decrease of Raldh2 expression and endogenous RA activity observed in Pbx1/2 and Hoxa1/Pbx1 null embryos (Figure 2; model in Figures 6A and 6B); (2) the decrease of Xraldh2 expression in frog embryos with mesoderm-specific knockdowns of Xpbx1b and Xhoxa1/Xpbx1b (Figure 4); (3) the induction of abnormal segmentation in frog embryo hindbrains following mesoderm-specific knockdown of Xpbx1b and Xhoxa1/Xpbx1b, independently of their roles in the neuroepithelium (Figure 4); (4) the posterior shift of endogenous RA activity in the hindbrain of Hoxa1/Pbx1 null embryos (Figure 2; model in Figures 6A and 6B); (5) the partial rescue of Hoxa1/Pbx1 null hindbrain phenotype by exogenous RA treatment (Figure 3); (6) the in vivo binding of the E3 Raldh2 enhancer by Hoxa1/Pbx1(2)/Meis2 in E8.5 embryos (Figure 5); and (7) the demonstration of the E3PH element requirement for normal expression of Raldh2 in vivo (Figure 5). Thus, Hoxa1 and Pbx1 are necessary in the mesoderm to generate sufficient levels of retinoids that, in turn, induce positionally appropriate gene activation in the early hindbrain neuroepithelium to begin normal segmentation.

Collinear Hox patterns are already established in mesoderm precursors before their ingression through the primitive streak (Deschamps and van Nes, 2005; Iimura and Pourquie, 2006), thus before Raldh2 activation and independently of RA activity (Lloret-Vilaspasa et al., 2010). This early phase of Hox expression in turn controls ordered paraxial mesoderm ingression and its positioning along the AP axis (Iimura and Pourquie, 2006). Blocking RA activity only impairs the neuropithelial, though not the early mesodermal, Hox expression domains (Lloret-Vilaspasa et al., 2010). Feedback retinoid regulation of Hox expression maintenance in vertebral somite precursors may in turn take place at later stages (e.g., Lohnes et al., 1994). RA produced in the paraxial mesoderm is thought to act as a diffusible morphogen that patterns the hindbrain in a concentration-dependent manner by inducing spatially restricted Hox expression patterns in the overlying neuroepithelium. Based on our results, such “homeogenetic induction” (De Robertis et al., 1989) of Hox expression across germ layers may indeed be controlled by the Hox genes themselves through a feed-forward transcriptional mechanism that induces their own expression and sets their rostral boundary in the hindbrain neuroepithelium, through the direct control of mesodermal Raldh2 expression and production of RA diffusing to the neuroectoderm (Figures 6A–6D). In this respect, Hox expression boundaries in mesoderm and hindbrain may not need to be in register, because of local control of retinoid levels and RA responsiveness along the hind-brain AP axis by degradation enzymes (e.g., Sirbu et al., 2005; Hernandez et al., 2007). Our current findings provide a conceptual framework to support such a model and reveal its potential importance for proper hindbrain segmentation.

Using a mesoderm-specific MO-mediated approach in Xenopus embryos, we showed that the functional knockdown of Xhoxa1/Xpbx1b in mesoderm results in decrease of Xraldh2 expression. In turn, this correlates with a range of hindbrain segmentation defects, including posterior shift of rhombomere identities, a feature of partial RA deficiency. Notably, the hind-brain abnormalities observed in mesoderm MO-injected frog embryos phenocopied the hindbrain defects of Hoxa1/Pbx1 null mutant mice, indicating that at least part of the mouse mutant hindbrain phenotype can be accounted for by the lack of Hoxa1/Pbx1 in the mesoderm and its early effect on Raldh2 downregulation. Previous work has shown that blocking lzr(pbx4)/pbx2 or pbx4/hox1 function in the zebrafish embryo results in extensive posterior expansion of r1 at the expense of more caudal rhombomere identities (Waskiewicz et al., 2002). It is tempting to speculate that such an extreme phenotype may also be partly contributed by a partial RA deficiency induced by the lack of these factors in the mesoderm. Finally, although our results do not claim to rule out additional Hox-independent roles for Pbx factors, the synergistic effects of Xhoxa1/Xpbx1b knockdowns in the frog mesoderm, as well as the molecular and in vivo analyses of the mouse Raldh2 locus demonstrate the functional impact of Hox-Pbx-mediated Raldh2 regulation.

The above observations provide strong support for evolutionary conservation of this Hox-Pbx-dependent retinoid regulatory pathway. Moreover, by identifying Hox-Pbx factors as regulators of RA levels in the early vertebrate embryo, together with their known role as RA targets, these results help to better rationalize how the retinoid signaling pathway could have been evolutionary co-opted for vertebrate AP patterning and integrated into the Hox positional system. More broadly, given the pleiotropic and instructive functions of RA in vertebrate development (Duester, 2008; Niederreither and Dolle, 2008), including axial patterning, regional segmentation of the nervous system, regulation of early organogenesis, and differentiation of stem and progenitor cells, our results establish a mechanism for the transcriptional control of the synthesis of appropriate RA activity in the embryo.

EXPERIMENTAL PROCEDURES

Retinoic Acid Administration

Retinoic acid administration was performed as described elsewhere (Pasqualetti et al., 2001). Mice were mated for 2 hr. A vaginal plug at the end of the mating was scored as E0.0 +1 hr. Pregnant mice were treated at the following gestational stages: E8.0 +2 hr (Hoxa1/Pbx1) and E8.5 (Pbx1/Pbx2). Pregnant females were administered a final RA concentration of 5 mg/kg (Hoxa1/Pbx1) or 10 mg/kg (Pbx1/Pbx2) body weight by oral gavage.

In Situ Hybridization

Whole-mount mouse in situ hybridization (ISH) was performed as described elsewhere (Studer et al., 1998). Each probe was hybridized on at least three single or compound mutant embryos. As for frog ISH, Digoxigenin (DIG)-labeled antisense RNA probes were generated for Xraldh2 and Xkrox20. Whole-mount ISH was performed as described elsewhere (Pasqualetti et al., 2000). After color development, embryos were post-fixed and bleached under fluorescent light to remove the pigment. For histological examination, whole-mount ISH processed embryos were embedded in a gelatin-albumin solution and then sectioned at 50 μm using a Leica VT1000S vibratome.

Mesoderm-Specific Morpholino Injections in Xenopus Embryos

Xenopus laevis embryos were obtained by hormone-induced laying and were staged according to Nieuwkoop and Faber (1956). Capped mRNAs were synthesized in vitro from template cDNAs: Nuclear-β-galactosidase (n-β-gal) and Red Fluorescent Protein (RFP, NotI/Sp6), using the SP6 mMESSAGE mMACHINE Kit (Ambion, catalog number AM1340). A morpholino antisense oligonucleotide (Gene Tools, LLC) was designed against the Xpbx1b mRNA (Gene Bank a.n. AF480430.1) complementary to −8 to +17 nucleotides: Xpbx1b-MO 5′-CTGGGCTGATCGTCCATTTCCAAGA-3′ (the ATG complementary sequence is underlined). A 5-base-mismatch control-MO was designed from the Xpbx1b-MO sequence (5′ CTGGcCTcATCGTCgATTTCg AAcA 3′, small caps indicate mismatched nucleotides) (Gene Tools, LLC). The control MO did not induce molecular alterations or patterning defects in injected embryos (n = 78; data not shown). The MO against the Xhoxa1 mRNA was described elsewhere (McNulty et al., 2005). MOs (Xpbx1b 20ng/embryo, Xhoxa1 10 ng/embryo, control-MO 20–30 ng/embryo) or capped mRNAs (RFP and/or n-β-gal, 300 pg/embryo each) were injected into the marginal zone of V2.2 left blastomere of 16-cell stage embryos in 3% Ficoll-400 (Fluka, catalog number 46327) in 0.1× MMR. After injection, embryos were transferred in 0.1× MMR and incubated at 18°C until the desired developmental stage. The injected side was visualized by the RFP presence.

β-Galactosidase Staining

Mice carrying the RARE::lacZ reporter transgene (Rossant et al., 1991) (Jackson Laboratory) mated into the Pbx1/2 or Hoxa1/Pbx1 null backgrounds were used for endogenous retinoid detection by X-gal staining as described elsewhere (Rossant et al., 1991). Xenopus embryos injected with Nuclear-β-galactosidase (n-β-gal) capped mRNA were fixed and stained with salmon-gal substrate (BIOSYNTH AG, catalog number B-7200) before further processing for whole-mount ISH.

Constructs

A 514 bp fragment containing the E3 Pbx-Hox site was amplified by PCR from mouse genomic DNA. The amplicon was cloned into pCRII-TOPO plasmid (Invitrogen), generating the pCR-E3 construct. A SpeI-NotI fragment of pCR-E3 was subcloned into the pBGZ40 plasmid (Itasaki et al., 1999) generating the E3::lacZ construct. pCR-E3 was also used as template to obtain the pCR-E3ΔPH construct by PCR-mediated site-directed mutagenesis, replacing the Pbx-Hox binding site with a SacII site for diagnostic restriction. The SpeI-NotI fragment of pCR-E3ΔPH was subcloned into pBGZ40 plasmid (E3ΔPH::lacZ).

In Ovo Electroporation

In ovo electroporation was performed as described elsewhere (Itasaki et al., 1999). Construct concentrations were as follows: 1.0 mg/ml for E3::lacZ reporter construct, E3ΔPH::lacZ and pCMV::Hoxa1 expression vector, and 0.2 mg/ml pCMV::eGFP coinjected as positive control of electroporated cells. Embryos were harvested 24 hr after electroporation and processed for β-galactosidase staining.

Transient Mouse Transgenic Analysis

NotI-XhoI fragments containing either the E3::lacZ or its mutated version E3ΔPH::lacZ constructs were used for pronuclear injection. Embryos were harvested at E8.5 and X-gal staining was used for β-galactosidase activity detection. For coronal vibratome sectioning, stained embryos were fixed in 4% PFA overnight, rinsed in PBT, embedded in 3% agarose/PBS, and then cut. Nuclear fast red solution (Sigma) was used as counterstaining.

Generation of BAC Transgenic Mouse Embryos

The BAC clone RP24-159G6 (BACPAC Resources Center at Children’s Hospital Oakland Research Institute, Oakland, CA), spanning from 23 kb upstream to 47 kb downstream the Raldh2 locus, was used as template for bacterial recombineering (Lee et al., 2001). The plasmid pN21-eGFP-SV40 polyA was used to amplify an eGFP-frt-Kanamycin-frt cassette by using 70-mer primers containing 50 nt of homology surrounding the ATG codon of the Raldh2 coding sequence. For the mutated construct, the plasmid pL452 (Liu et al., 2003) was used as template to amplify a LoxP-Kanamycin-LoxP cassette by using 70-mer primers containing 50 nt of homology surrounding the E3PH element. Following homologous recombination and resistance cassette removal, the E3PH element was replaced by a single LoxP site. Correct recombination and removal of resistance genes in the Raldh2::eGFP and Raldh2-E3ΔPH::eGFP BACs were tested by PCR, restriction enzyme digestion, and sequencing. Before microinjection, the modified BACs were linearized by PI-SceI digestion.

Chromatin Immunoprecipitation Assay

About 1600 E8.5 mouse embryos were manually dissected in posterior “body” and anterior “head” regions. Chromatin was prepared, and ChIP was performed as described elsewhere (Frank et al., 2001). Samples were immunoprecipitated overnight at 4°C with the following antibodies: Pbx1/2/3 (C20 sc888X, Santa Cruz), Meis2 (N17 sc10600, Santa Cruz), Hoxa1 (N20 sc17146X, Santa Cruz); Suz12 (ab12073, Abcam), H3K27me3 (9756, Cell Signaling Technology), H3K4me3 (9751, Cell Signaling Technology), and Rabbit IgG (Sigma). Conserved Raldh2 fragment E1 to E4 containing the Hox-Pbx binding sites and control regions outside of the conserved regions (OUT) were amplified by real-time qPCR using specific primer pairs.

Electrophoretic Mobility Shift Assays

EMSAs were performed as described elsewhere (Ferretti et al., 2000) using nuclear extracts purified from E8.5 “body” embryonic region or in vitro translated proteins. Labeled oligonucleotide probes contain the putative Hox-Pbx binding sites and their mutated forms within the distinct Raldh2 conserved regions (E1 to E4). The antibodies used to assess DNA binding specificity are the same used for ChIP assay.

Sequence Conservation Analysis

Comparisons of mouse Raldh2 genomic sequences to other vertebrates were performed using the UCSC algorithm (http://genome.ucsc.edu/).

Statistical Methods

Results are expressed as mean ± SEM from triplicate qPCRs. Student’s t test was used when appropriate.