Mouse Hoxa2 mutations provide a model for microtia and auricle duplication

The external ear is composed of the auricle (or pinna) and the external auditory canal (EAC). A variety of factors can affect its morphogenesis, inducing a wide range of abnormalities such as microtia and partial auricle duplications (Alasti and Van Camp, 2009; Baschek et al., 2006; Gore et al., 2006; Hunter and Yotsuyanagi, 2005; Ku et al., 1998; Mishra and Misra, 1978; Pan et al., 2010). Microtia, which is characterised by a small, abnormally shaped auricle, is one of the most common external ear abnormalities; the estimated prevalence ranges from 0.8 to 4.2 per 10,000 births depending on the population (Alasti and Van Camp, 2009). External ear abnormalities can occur as the only clinical defect, but in most cases they appear in complex syndromes in which others organs and structures are affected. External ear anomalies are indeed commonly associated with internal and/or middle ear dysplasia, resulting in hearing loss and problems in speech and language development (Kountakis et al., 1995). Deciphering the molecular mechanisms involved in external ear morphogenesis is crucial not only to better understand the defaults affecting the external ear, but also to gain a more comprehensive view of the mechanisms involved in the genetic syndromes that encompass external ear abnormalities.

Genetic diagnostics in humans, as well as loss-of-function experiments in mice, have begun to identify signalling factors and transcriptional regulators involved in external ear morphogenesis. Among these factors, Hoxa2 plays a crucial role. In humans, a HOXA2 mutation induces a bilateral microtia associated with abnormal shape of the auricle (Alasti et al., 2008). Moreover, hearing impairment and partial cleft palate have been reported (Alasti et al., 2008). A role for Hoxa2 in external ear morphogenesis is also observed in mice, where its inactivation induces duplication of the EAC and absence of the pinna (Gendron-Maguire et al., 1993; Mallo and Gridley, 1996; Mark et al., 1995; Rijli et al., 1993). This phenotype is associated with other craniofacial abnormalities, notably a cleft palate and middle ear structure malformations. Indeed, Hoxa2 inactivation results in morphological transformation of the neural crest-derived skeletal elements of the second pharyngeal arch, including the middle ear ossicle stapes, into a duplicated set of first arch-like elements, including duplication of the middle ear ossicles malleus and incus (Gendron-Maguire et al., 1993; Rijli et al., 1993). By temporally controlled inactivation, we have further shown that pinna morphogenesis requires Hoxa2 function through advanced developmental stages (Santagati et al., 2005). Whereas Hoxa2 inactivation before E11.5 results in the absence of the pinna, Hoxa2 inactivation at later stages (between E12.5 and E13.5) results in a hypomorphic pinna (Santagati et al., 2005), thus mimicking the human HOXA2 mutant condition. Altogether, these data emphasize the major role of Hoxa2 in auricle morphogenesis. However, the molecular programme that is regulated by Hoxa2 is only beginning to be elucidated (Donaldson et al., 2012) and remains largely unknown. Furthermore, it is unclear whether Hoxa2 is not only necessary but also sufficient to induce and orchestrate the whole developmental programme underlying pinna morphogenesis.

By genetic fate mapping, we show here that the mouse pinna derives from the second pharyngeal arch Hoxa2-expressing neuralcrest cells (NCCs), and not from a combined contribution of first and second arch neural crest-derived mesenchyme, as proposed by some authors based on morphological observations of human embryos (reviewed by Hunter and Yotsuyanagi, 2005; Alasti and Van Camp, 2009; Klockars and Rautio, 2009; Passos-Bueno et al., 2009; Porter and Tan, 2005; Schoenwolf and Larsen, 2009). We further show that the mouse EAC is entirely lined by Hoxa2-negative first arch mesenchyme, and does not develop, as previously proposed, at the first pharyngeal cleft (Jakubíková et al., 2005; Schoenwolf and Larsen, 2009). By a conditional gain-of-function approach, we show that ectopic Hoxa2 expression in first arch NCCs is alone sufficient to induce the transformation of the neural crest-derived first arch mesenchyme lining the EAC into a mirror-image duplication of the pinna. Functional and molecular analyses revealed that Hoxa2 controls the formation of the pinna through the BMP signalling pathway by regulating the expression of bone morphogenetic protein 5 (Bmp5), Bmp4 and twisted gastrulation (Tsg; Twsg1 - Mouse Genome Informatics). Chromatin immunoprecipitation and parallel sequencing (ChIP-Seq) on second arch cells additionally shows that Hoxa2 binds to Bmp4 and Bmp5 non-coding regions, suggesting that they are direct targets. Bmp5 inactivation results in a small pinna [known as the short ear mutation (King et al., 1994; Kingsley et al., 1992)], and we further show that Bmp4 conditional inactivation also partially impairs pinna development. Moreover, Hoxa2 regulates the expression of eyes absent 1 (Eya1), which in humans is involved in the branchio-oto-renal syndrome (Abdelhak et al., 1997; Kochhar et al., 2007). Thus, Hoxa2 is a fundamental transcriptional regulator orchestrating the morphogenesis of the auricle. This genetic approach in the mouse might therefore represent a suitable model with which to understand the aetiology of human auricle abnormalities.

To fate map the external ear, the Z/AP (Lobe et al., 1999) and Rosa-CAG-LSL-tdTomato (Ai14) (Madisen et al., 2010) reporter mouse lines were crossed with the R4::Cre (Oury et al., 2006) mouse line. The Hoxa2^EGFP and Hoxa2^{EGFP(lox-neo-lox)} knock-in mouse lines were described previously (Pasqualetti et al., 2002). Wnt1::Cre;Hoxa2^{EGFP(lox-neo-lox)/+} embryos were obtained by crossing Hoxa2^{EGFP(lox-neo-lox)/+} mice with Wnt1::Cre mice (Danielian et al., 1998). The following lines were also used: CMV::CreER^T2 (Santagati et al., 2005), Hoxa2^lox (Ren et al., 2002), Bmp4^loxP-lacZ [(Kulessa and Hogan, 2002); hereafter referred to as Bmp4^lox]. Hoxa2^del/+ or Bmp4^del/+ alleles were generated by Cre-mediated deletion, mating CMV::Cre (Dupé et al., 1997) with Hoxa2^lox/lox or Bmp4^lox/lox mice, respectively. To generate CMV::CreER^T2;Hoxa2^lox/del embryos, Hoxa2^lox/lox or Hoxa2^lox/+ were crossed with CMV::CreER^T2;Hoxa2^del/+ mice. To generate Bmp4^lox/del embryos, Bmp4^lox/lox were crossed with Bmp4^del/+ mice. To generate Wnt1::Cre^{Hoxa2-IRES-EGFP} mice, the Rosa^{(lox-stop-lox)Hoxa2-IRES-EGFP/+} line (Miguez et al., 2012) was crossed with the Wnt1::Cre line. All animal experiments were approved by the Basel Cantonal Veterinary Authorities and conducted in accordance with the Guide for Care and Use of Laboratory Animals.

Tamoxifen (TM) (Sigma) was dissolved at 20 mg/ml in pre-warmed corn oil (Sigma) and stored at 4°C. Three successive TM administrations (10 mg at E12.5, E13.0 and E13.5) were administered to pregnant females by oral gavage with 12-hour intervals.

E14.5 mouse fetuses were fixed overnight in 4% paraformaldehyde (PFA), rinsed in PBS, equilibrated in 20% sucrose and embedded in Cryomatrix (Thermo Electron Corporation). Cryostat sections (20 μm) were cut in the frontal and horizontal planes (see Fig. 1A). Sections were fixed for 1 hour in 4% PFA, rinsed in PBS at room temperature and incubated for 1 hour in PBS at 65°C to inactivate endogenous alkaline phosphatase. Sections were rinsed in a solution of NTMT [0.1 M NaCl, 0.1 M Tris-HCl (pH 9.5), 0.05 M MgCl₂, 0.1% Tween 20]. For the staining, 3.5 μl NBT (Roche 1383213) and 3.5 μl BCIP (Roche 1383221) were used per ml of NTMT. Sections were rinsed in water, then in 100% ethanol, and mounted onto slides.

Immunostaining for EGFP on cryosections was performed using a polyclonal rabbit anti-EGFP antibody (Molecular Probes) and a peroxidase-conjugated goat anti-rabbit IgG (Beckman Coulter) or an Alexa 488-conjugated secondary antibody (Invitrogen). For peroxidase-conjugated goat anti-rabbit IgG, detection was performed with DAB chromogen (DAKO). Immunostaining for RFP on cryosections was performed using a polyclonal rabbit anti-RFP antibody (Rockland) and an Alexa 568-conjugated secondary antibody (Invitrogen). Immunostaining for Ki67 was performed on paraffin sections using a rabbit anti-Ki67 antibody (Novocastra NCL-Ki67p) and a biotinylated anti-rabbit antibody. After incubating for 30 minutes at room temperature with the VECTASTAIN ABC reagent (Vector Labs), peroxidase activity was revealed with DAB chromogen. For phosphorylated Smad (phosphoSmad) staining, cryostat sections were incubated with phosphoSmad1/5/8 antibody (Cell Signaling) at 37°C for 3 hours, followed by Alexa 488-conjugated secondary antibody (Invitrogen).

Consecutive cryostat sections (25 μm) of E14.5 Hoxa2^EGFP/+ and Hoxa2^EGFP/EGFP fetuses, immunostained for EGFP, were imaged with a Leica fluorescence macroscope. Between 49 and 68 sections were aligned using Bitplane AutoAligner 6.0.1 (manual alignment). Structures were artificially labelled in separated channels in Adobe Photoshop CS5.1. The artificially labelled structures, as well as the GFP channel, were transformed into surfaces in Bitplane Imaris 7.5.2 (surface area detail level: 35 μm; thresholding: absolute intensity).

In situ hybridisation on frontal and horizontal sections (see Fig. 1A) were performed as described (Santagati et al., 2005). The following RNA probes were used: Hoxa2 (Ren et al., 2002), Eya1 (Xu et al., 1999), Bmp5 (Solloway and Robertson, 1999), Bmp4 (Hogan et al., 1994), Tsg (Zakin and De Robertis, 2004), Prrx1, Prrx2 (ten Berge et al., 1998) and Tshz2 (Caubit et al., 2000).

ChIP experiments were performed on second pharyngeal arches isolated from E11.5 CD1 embryos as described (Donaldson et al., 2012). Immunoprecipitated DNA was subjected to qPCR using the following primers: Bmp4, forward 5′-TGTGGGATAAAACAGGAGTGC-3′ and reverse 5′-GCTCCCTCAGTTTGGCTAGA-3′; Bmp5, forward 5′-TATGCAGTCTAGGGCCACCT-3′ and reverse 5′-CATTTGGGATAAAAGAACCTCAA-3′.

External ear morphogenesis occurs early in development, starting at E12.0 in mouse or at the sixth week in humans. In humans, the auricle was proposed to derive from tubercles, or hillocks, which originate from spatially segregated NCC populations of both first and second pharyngeal arches (reviewed by Hunter and Yotsuyanagi, 2005). To map second arch neural crest contribution to the mouse pinna, we crossed a rhombomere (R)4::Cre mouse line (Oury et al., 2006) with the Z/AP (Lobe et al., 1999) or Rosa-CAG-LSL-tdTomato (Ai14) (Madisen et al., 2010) reporter lines. Upon Cre-mediated recombination, Alkaline phosphatase (AP) or tdTomato gene expression is permanently activated in R4-derived NCC progenitors, thus allowing second arch NCC contribution to the external ear to be assessed. Although the presence of a few spared unstained cells cannot be ruled out, careful observation of consecutive sections (Fig. 1D,E), as well as high magnifications of co-stainings of tdTomato with DAPI to identify cell nuclei (supplementary material Fig. S1), revealed that the whole mouse pinna is contributed by R4-derived second arch NCCs, and not by a composite of spatially segregated first and second arch NCC-derived mesenchyme.

Hoxa2 is expressed in second, but not in first, arch NCCs (reviewed by Minoux and Rijli, 2010). To map the contribution of Hoxa2-expressing (Hoxa2⁺) NCCs to the pinna, we crossed the Wnt1::Cre driver (Danielian et al., 1998), which expresses Cre in NCC progenitors, to the Hoxa2^{EGFP(lox-neo-lox)} allele (Pasqualetti et al., 2002), in which EGFP is knocked in at the Hoxa2 locus and is conditionally induced by Cre-mediated excision. In E14.5 Wnt1::Cre;Hoxa2^{EGFP(lox-neo-lox)/+} fetuses, the pinna is composed of EGFP-expressing NCCs (Fig. 1F,I,J), indicating that this structure mainly originates from Hoxa2⁺ NCC progenitors. Moreover, high magnification of DAPI/EGFP co-stainings (supplementary material Fig. S1) revealed that virtually all cells in the pinna are EGFP⁺, i.e. they derive from Hoxa2⁺ NCC progenitors. Together with the R4 fate mapping, these results indicate that the mouse pinna mainly originates from Hoxa2⁺ second arch NCCs.

In humans, the EAC, an ectodermal structure lined by NCC-derived mesenchyme, has been proposed to originate at the first pharyngeal cleft, i.e. at the interface between the first and second arches, and to be surrounded by NCCs contributed by both arches (Jakubíková et al., 2005; Schoenwolf and Larsen, 2009). Contrary to this prediction, we found that in E14.5 Wnt1::Cre;Hoxa2^{EGFP(lox-neo-lox)/+} mouse fetuses the EAC and its surrounding mesenchyme are entirely composed of Hoxa2^-/EGFP^- cells, and therefore not derived from the second arch (Fig. 1G,H,J). Moreover, we observed a sharp spatial segregation between Hoxa2^-/EGFP^- and Hoxa2⁺/EGFP⁺ cell populations contributing to the EAC and the pinna, respectively (Fig. 1J). A similar cell sorting was observed in E14.5 Hoxa2^EGFP/+ fetuses in which EGFP is knocked in at the Hoxa2 locus and faithfully recapitulates the Hoxa2 expression domain (Pasqualetti et al., 2002) (Fig. 1K-N). Analysis at E12.5 confirmed that mouse EAC begins to form entirely anterior to the second arch-derived Hoxa2⁺/EGFP⁺ cells that form the pinna, within the Hoxa2^-/EGFP^- first arch-derived tissue (Fig. 1O). Thus, at E12.5 the EAC and the pinna develop opposite to and almost equidistant from the vestigial first pharyngeal cleft that maps at the border between EGFP⁺ and EGFP^- territories (arrowhead, Fig. 1O); notably, their respective cell contributions from the first and second arches never intermingle despite the complex morphogenetic cell movements that occur during their formation.

To investigate the spatial distribution of second arch cells in the absence of Hoxa2 function, we used Hoxa2^EGFP/EGFP homozygous mutant embryos (Fig. 1P). By 3D reconstruction of tissue sections, we confirmed the previously described duplication of the EAC (Gendron-Maguire et al., 1993; Mallo and Gridley, 1996; Mark et al., 1995; Rijli et al., 1993) (EAC^*, Fig. 2) and found that the EAC^* develops within the Hoxa2⁺/EGFP⁺ domain (Fig. 2, see also Fig. 1Q-T). By contrast, the orthotopic EAC fully develops into the Hoxa2^-/EGFP^- territory, both in E14.5 Hoxa2^EGFP/+ control and Hoxa2^EGFP/EGFP homozygous mutant fetuses (Fig. 2, see also Fig. 1L,M,T). The border between EGFP⁺ and EGFP^- territories appears to be the axis of symmetry, on either side of which the EAC and its ectopic EAC^* counterpart develop (Fig. 1T). Tissue sections additionally show that in E14.5 Hoxa2^EGFP/EGFP homozygous mutant fetuses, the EGFP⁺ cells maintain their normal spatial segregation and do not ectopically intermingle with EGFP^- cells, despite the fact they have acquired a first arch-like identity (Fig. 1T).

At E14.5, Hoxa2 expression extensively overlaps with the developing pinna. On horizontal and frontal sections, the Hoxa2⁺ domain contains an outer territory, which includes the pinna, and an inner territory (Fig. 3B,H, the dashed line delimits the territories). The inner domain is strongly Hoxa2⁺, whereas the outer domain displays sparser Hoxa2 transcript distribution (Fig. 3B,H). The two Hoxa2⁺ territories roughly abut at the base of the bending pinna (arrows, Fig. 3B,H). Hoxa2 is additionally expressed in the mesenchyme at the tip and below the ectoderm on both the dorsal and ventral sides of the pinna (Fig. 3B,H).

The spatial distribution of Hoxa2 transcripts suggests distinct proliferation/differentiation states of cellular subsets. In E14.5 control fetuses, Ki67⁺ proliferating cells are present at the tip and are orderly aligned on both the dorsal and ventral sides of the pinna, below the ectoderm, whereas the mesenchymal core of the pinna is only contributed by Ki67^- postmitotic differentiating cells (Fig. 3A,C,G). At the pinna distal edge (tip), the Ki67⁺ cell pattern overlaps with the Hoxa2 transcript distribution (Fig. 3A,B), indicating that Hoxa2 is mainly expressed in proliferating NCCs. By contrast, the inner Hoxa2 expression domain encompasses both Ki67⁺ and Ki67^- cell subsets (Fig. 3A,B,G,H), with Ki67⁺ cells orderly aligned (arrows, Fig. 3A,C,G) and adjacent to Ki67^- differentiating mesenchymal cells. In addition, at E14.5 a subset of spatially organised Ki67⁺ proliferating cells at the base and all along the dorsal aspect of the pinna (arrowheads, Fig. 3A,C,G) does not express Hoxa2 (Fig. 3A,B,G,H). This latter population of highly proliferating cells selectively expresses Eya1 (Fig. 3G,I), which is essential for pinna development in mice (Xu et al., 1999). Interestingly, at the beginning of pinna morphogenesis (E12.0-12.5), Eya1 and Hoxa2 appear to be co-expressed in a subset of second arch-derived NCCs at the base of the future pinna (Fig. 3M-O), which are likely to correspond to the precursors of the Eya1⁺ population observed at later stages (Fig. 3I). In E12.5 Hoxa2 homozygous mutant embryos, this early Eya1 expression domain is lacking (Fig. 3P,Q).

In control Hoxa2^lox/del fetuses, which bear both a fully deleted and a floxed Hoxa2 allele (Ren et al., 2002), the pinna is of normal appearance (Santagati et al., 2005) (Fig. 3A-C), indicating that Hoxa2 haploinsufficiency in mouse does not result in gross external ear defects. By contrast, tamoxifen (TM) administration to CMV::CreER^T2;Hoxa2^lox/del fetuses at NCC postmigratory stages (E12.5, E13.0 and E13.5) induces a hypomorphic pinna (Santagati et al., 2005) (supplementary material Fig. S2), thus bypassing an early role of Hoxa2 and supporting a late requirement in pinna morphogenesis. The smaller and dysmorphic pinna of TM-treated CMV::CreER^T2;Hoxa2^lox/del mutant newborns (supplementary material Fig. S2) is reminiscent of the HOXA2 mutant phenotype in humans (Alasti et al., 2008). Analysis by in situ hybridisation confirmed the almost complete loss of Hoxa2 expression in TM-treated mutants when compared with control embryos (supplementary material Fig. S3).

At E14.5, no apoptotic cells were observed in control or in TM-treated CMV::CreER^T2;Hoxa2^lox/del mutant fetuses (not shown), suggesting that the smaller dysmorphic pinna is not the result of increased cell death. This was also indirectly supported by the persistence of EGFP⁺ cells in the outer ear region of Hoxa2^EGFP/EGFP full knockout fetuses, which lack a normal pinna (Fig. 1P-T). The hypomorphic pinna of E14.5 TM-treated CMV-CreER^T2;Hoxa2^lox/del fetuses lacks the normal spatial segregation between Ki67⁺ and Ki67^- mesenchymal cells (arrows, Fig. 3E,F). Moreover, a mass of proliferating cells abnormally accumulates at the dorsal base of the pinna (asterisks, Fig. 3D-F,J), and the inner Hoxa2⁺ domain appears disorganised, with fewer Ki67⁺ cells and an accumulation of unpatterned Ki67^- differentiating cells (arrowheads, Fig. 3D-F,J). Lastly, in E14.5 TM-treated CMV-CreER^T2;Hoxa2^lox/del fetuses, the population of spatially organised proliferating cells that expresses Eya1 is still present (Fig. 3L), suggesting a role of Hoxa2 in establishing, but not maintaining, the proliferative Eya1⁺ territory.

Altogether, these results indicate that, after E12.5, Hoxa2 contributes to the normal size and shape of the pinna by organising spatially restricted local patterns of cell proliferation.

The hypomorphic pinna in TM-treated CMV::CreER^T2;Hoxa2^lox/del fetuses is reminiscent of the phenotype of the short ear mutation, which inactivates Bmp5 (King et al., 1994; Kingsley et al., 1992). Thus, Bmp5 is a suitable candidate to be regulated by Hoxa2. In E14.5 control pinna, Bmp5 expression is restricted to the mesenchymal cores of the inner and outer Hoxa2⁺ domains, although not at the tip or in most of the proliferating mesenchyme of the pinna (Fig. 4A,B,E,F). In TM-treated CMV::CreER^T2;Hoxa2^lox/del fetuses, Bmp5 expression levels were strongly reduced in the pinna (arrowheads, Fig. 4M,N). We also observed a reduction in the spatial extent of Bmp5 expression in the inner Hoxa2⁺ domain (arrows, Fig. 4M,N), with an accumulation of Bmp5 residual expression at the base of the pinna (asterisk, Fig. 4M). Hoxa2 temporal inactivation selectively downregulates Bmp5, although not paired related homeobox 1 (Prrx1), which encodes a transcription factor involved in craniofacial development (Martin et al., 1995) (Fig. 4I,J,Q,R).

We next analysed Bmp4 expression. At E12.0-12.5, Bmp4 is expressed at the distal edge (tip) of the forming pinna, as well as in a small population of mesenchymal cells at its base (not shown). This pattern is maintained and extended at E13.5 and E14.5, becoming similar to the Hoxa2 expression pattern. Specifically, Bmp4 is highly expressed in the mesenchymal core of the inner Hoxa2⁺ domain (Fig. 4C,D,G,H). In the pinna, Bmp4 is expressed at the tip and in the proliferating mesenchyme below the dorsal and ventral ectoderm (Fig. 4G,H). This transcript distribution is complementary to that of Bmp5 (supplementary material Fig. S4A,B); notably, the Bmp4 and Bmp5 expression patterns together recapitulate the Hoxa2 expression pattern in the pinna (Fig. 4A-H). In E14.5 TM-treated CMV::CreER^T2;Hoxa2^lox/del mutant pinna, Bmp4 expression is absent or strongly reduced (Fig. 4O,P), whereas Prrx1 (Fig. 4J), Prrx2 (Fig. 4L) (ten Berge et al., 1998) and teashirt zinc finger family member 2 (Tshz2) (Caubit et al., 2000) (Fig. 4K) are still expressed in the inner Hoxa2⁺ domain and at the tip of the pinna (Fig. 4S,T; data not shown).

In summary, the above results indicate that Hoxa2 is required to maintain the normal levels and spatial distribution of Bmp4 and Bmp5 transcripts during pinna growth and morphogenesis.

The activity of BMPs is modulated in the extracellular space through their interaction with secreted agonists or antagonists. Tsg is a secreted BMP modulator involved in mouse head development (Zakin and De Robertis, 2004). Although no external ear defects are observed in Tsg^-/- or Bmp4^+/- mice, compound Tsg^-/-;Bmp4^+/- newborns display, in severe cases, low-set ears associated with an abnormal shape of the pinna (Zakin and De Robertis, 2004), indicating that Tsg is involved in modulating Bmp4 activity during pinna development. In E14.5 wild-type fetuses, Tsg is strongly expressed in a stripe of Ki67⁺ highly proliferating cells in the inner Hoxa2⁺ domain (supplementary material Fig. S5K, arrow; Fig. 3G, arrow), directly adjacent to the Bmp4⁺ mesenchymal core (supplementary material Fig. S5A-O). In E14.5 TM-treated CMV::CreER^T2;Hoxa2^lox/del pinna, Tsg is severely downregulated (supplementary material Fig. S5P-R). These results, together with the downregulation of Bmp4 and Bmp5 expression and the effects on cell proliferation patterns (Figs 3, 4), further support the notion that Hoxa2 is involved in BMP signalling regulation during pinna formation.

To further investigate the downstream effects induced by Bmp5, Bmp4 and Tsg downregulation in Hoxa2 mutants, we analysed the phosphorylation of Smad1, 5 and 8, a hallmark of canonical BMP signal transduction (Heldin and Moustakas, 2012; Horbelt et al., 2012), by anti-phosphoSmad1/5/8 immunostaining. In E14.5 control fetuses, the phosphoSmad1/5/8⁺ cell distribution is spatially restricted. PhosphoSmad1/5/8⁺ cells are observed throughout the Bmp5⁺/Bmp4⁺ territory of the inner Hoxa2⁺ domain extending to the base of the pinna (white arrows, Fig. 5A,C; supplementary material Fig. S4C), although not in the Bmp5⁺/Bmp4^- mesenchymal core of the developing pinna (Fig. 5A,C; supplementary material Fig. S4A-C). PhosphoSmad1/5/8⁺ cells are also detected in the Ki67⁺/Bmp4⁺ proliferating domain at the tip of the pinna (green arrowheads, Fig. 5A,C; supplementary material Fig. S4A,C), and in the Eya1⁺/Ki67⁺ highly proliferative domain abutting the pinna mesenchymal core (white arrowheads, Fig. 5A,C; supplementary material Fig. S4C).

In E14.5 TM-treated CMV::CreER^T2;Hoxa2^lox/del pinna, co-staining between phosphoSmad1/5/8 antibody and DAPI shows that phosphoSmad1/5/8⁺ domains at the base and at the tip of the pinna are either strongly reduced or absent (white arrows and green arrowheads, Fig. 5E-H). These data indicate that Hoxa2 temporal inactivation has strong effects on the local regulation of canonical BMP signalling. By contrast, the Eya1⁺ subpopulation of phosphoSmad1/5/8⁺ cells is still present (white arrowheads, Fig. 5E,G), in keeping with the observation that the late Hoxa2 inactivation only partially affects this Eya1⁺ cell population (see above; Fig. 3I,L).

Bmp4^lox/+ (Kulessa and Hogan, 2002) heterozygous mutant mice survive and do not show external ear defects (not shown). To investigate the consequences of further reducing Bmp4 activity during pinna morphogenesis, we generated Bmp4^lox/del hypomorphic mutants bearing both a fully deleted and a floxed Bmp4 allele. Unlike null mutants, Bmp4^lox/del fetuses survive until E16.5, a stage at which the pinna has normally already folded over the meatus (Fig. 6A,D). In severe cases, Bmp4^lox/del fetuses display a hypomorphic pinna (Fig. 6B,F), albeit at low frequency (n=5/22). When present, this phenotype is always associated with polydactyly (arrows, Fig. 6E, compare with 6C), which also occurs at a low frequency (n=9/22) in these hypomorphic mutants. Bilateral microphthalmia, smaller body size and subcutaneous edema are also observed in most of the Bmp4^lox/del hypomorphic mutants (Fig. 6A,B). The weak penetrance of the pinna phenotype is likely to be due to partial functional redundancy with other BMP family members, such as Bmp5, and/or the persistence of residual Bmp4 function in Bmp4^lox/del hypomorphic mutants. Bmp4 homozygous null mutants die around gastrulation (Winnier et al., 1995), preventing analysis of its role in external ear morphogenesis. Nonetheless, these results highlight Bmp4 involvement, at least to some extent, in the morphogenesis of the pinna.

By mining datasets from a genome-wide map of Hoxa2 binding to chromatin (ChIP-Seq) from second pharyngeal arch tissue dissected just prior to external ear formation (Donaldson et al., 2012), we found binding of Hoxa2 in proximity to Bmp4 [72 kb downstream of the Bmp4 transcription start site (TSS)] and within the third intron of Bmp5 (52 kb upstream of the Bmp5 TSS). We confirmed Hoxa2 binding enrichment on both Bmp4 and Bmp5 regions by performing conventional ChIP on second arches dissected from E11.5 wild-type embryos (supplementary material Fig. S6A). The region bound by Hoxa2 in the third intron of Bmp5 is the result of an insertion that is present only in mice and rat. By contrast, the region located downstream of the Bmp4 TSS is widely conserved in vertebrates and contains putative Pbx/Hox and Hox binding sites (supplementary material Fig. S6B). Moreover, in E11.5 Hoxa2 mutant second arch the expression of both Bmp4 and Bmp5 is downregulated (Donaldson et al., 2012). Together with the expression pattern changes observed in temporally induced Hoxa2 mutants (Fig. 4), these results suggest that Hoxa2 might directly maintain Bmp4 and Bmp5 expression in spatially restricted domains of the developing external ear.

To address whether Hoxa2 is not only necessary but also sufficient for the formation of the pinna we established a conditional overexpression system in the mouse allowing ectopic Hoxa2 expression in Hox-negative NCCs anterior to the second arch. Conditional Hoxa2 overexpression in NCCs was induced by mating the Wnt1::Cre line (Danielian et al., 1998) with a Rosa^{(lox-stop-lox)Hoxa2-IRES-EGFP} allele (Miguez et al., 2012) to produce Wnt1::Cre;Rosa^{(lox-stop-lox)Hoxa2-IRES-EGFP} (hereafter designated Wnt1^{Hoxa2-IRES-EGFP}).

Wnt1^{Hoxa2-IRES-EGFP} fetuses die at birth and display craniofacial defects, including face, middle ear and skull bone malformations, as well as a reduction in lower jaw size (Fig. 7; data not shown). Most notably, all of them (n=25/25) display an ectopic, fully formed pinna, which is a mirror-image duplication of its normal orthotopic counterpart (compare Fig. 7A with supplementary material Fig. S7B-E). The ectopic pinna replaces the EAC, and thus is likely to form from first arch NCCs. The orthotopic pinna, which is derived from the second pharyngeal arch, appears normally shaped, indicating that Hoxa2 overexpression from the Rosa^{(lox-stop-lox)Hoxa2-IRES-EGFP} allele within its own expression domain does not result in overt pinna morphological abnormalities (Fig. 7B-E). It is also noteworthy that Wnt1^{Hoxa2-IRES-EGFP} fetuses occasionally display multiple additional ectopic structures around the eye that morphologically resemble small ectopic pinnae (arrows, Fig. 7C,D).

The presence of an internal ribosome entry site (IRES) in the Rosa^{(lox-stop-lox)Hoxa2-IRES-EGFP} allele allows the cells that are ectopically expressing Hoxa2 to be traced by immunohistochemistry with an anti-EGFP antibody. We therefore compared the EGFP⁺ cell distribution in E14.5 Wnt1::Cre;Hoxa2^{EGFP(lox-neo-lox)/+} fetuses, in which EGFP is conditionally expressed in NCCs expressing the endogenous Hoxa2 (Fig. 8A), with Wnt1^{Hoxa2-IRES-EGFP} fetuses, which additionally ectopically express Hoxa2-IRES-EGFP in rostral Hoxa2-negative territories. In Wnt1^{Hoxa2-IRES-EGFP} mutants, the ectopic pinna is entirely contributed by Hoxa2-IRES-EGFP-expressing cells (Fig. 8B-D).

Moreover, the ectopic pinna of Wnt1^{Hoxa2-IRES-EGFP} fetuses displays mirror-image duplications of the Bmp5, Bmp4 and Tsg expression patterns (Fig. 8E-L; supplementary material Fig. S7). Namely, Bmp4 is expressed at the tip of the duplicated pinna as well as in the mesenchyme at its base (Fig. 8I-L). Bmp5 and Tsg are expressed in the mesenchymal core and at the base of the duplicated pinna (Fig. 8E-H; supplementary material Fig. S7). Notably, Bmp4 is also expressed in the ectopic structures that form around the eye (Fig. 8L), while Bmp5 is expressed in the underlying mesenchyme (Fig. 8F-H). Strikingly, proliferating Ki67⁺ and differentiating DAPI⁺ cell patterns are spatially organised in the ectopic pinna as a faithful mirror image of the cellular organisation of the orthotopic pinna (Fig. 8M-Q).

These results not only confirm the molecular identity of the duplicated pinna but also demonstrate that Hoxa2 is sufficient to induce and maintain the genetic programme that underlies the formation of the pinna.

The human auricle cartilage has classically been proposed to derive from six nodular masses of mesenchyme, termed the hillocks of His, that appear during the sixth week of development, three in the first (mandibular) and three in the second (hyoid) pharyngeal arch, respectively (reviewed by Hunter and Yotsuyanagi, 2005; Schoenwolf and Larsen, 2009). On the other hand, the ectodermal EAC has been proposed to originate at the first pharyngeal cleft and to be lined by NCCs contributed by both first and second arches (Jakubíková et al., 2005; Schoenwolf and Larsen, 2009). The hillocks eventually fuse to form the different parts of the human external ear. Although there is consensus that the tragus, projecting in front of the EAC, derives from mandibular hillocks, the respective contributions of mandibular or hyoid hillocks to the auricle are less clear (Hunter and Yotsuyanagi, 2005; Kagurasho et al., 2012). For instance, according to some authors, but not others, first arch-derived hillocks not only form the tragus but also give rise to the helix, which is the auricle folded edge, and the antihelix (reviewed by Hunter and Yotsuyanagi, 2005).

Our genetic fate mapping in the mouse provides novel insights into the embryological origin of the external ear. We show that the mesenchyme just anterior to the EAC (possibly, the mouse homologue of the human tragus) is contributed by Hoxa2^-/EGFP^- cells (Fig. 1), indicating that, as proposed in humans, it might originate from first pharyngeal arch NCCs. By contrast, genetic fate mapping and Hoxa2 functional analysis reveal for the first time that the mouse pinna is entirely contributed by second arch Hoxa2⁺ neural crest-derived mesenchyme, and not by both first and second arch NCCs. Moreover, genetic fate mapping shows that the ectoderm-derived EAC invaginates into, and is entirely surrounded by, Hoxa2-negative mesenchyme, which is both anterior and spatially segregated from second arch-derived mesenchyme (Fig. 1), and is thus most likely of first arch origin. Therefore, the mouse EAC appears to be a first arch-derived structure that does not originate at the border between first arch Hoxa2-negative and second arch Hoxa2-positive mesenchyme, i.e. at the first pharyngeal cleft, as previously assumed (Jakubíková et al., 2005; Schoenwolf and Larsen, 2009).

Auricle morphology is complex in humans, suggesting that additional components might have been recruited from the developing first arch mesenchyme, as compared with mouse. Nonetheless, it is possible that, as in mouse, the human auricle is mostly contributed by second arch NCCs. This conclusion may be indirectly supported by the following arguments. First, there is agreement among clinicians that the trigeminal (first arch) nerve supply is limited to the tragus and anterior part of the EAC (reviewed by Hunter and Yotsuyanagi, 2005; Wood-Jones and I-Chuan, 1934). Second, the auricle is not affected in otocephaly, which is a genetic syndrome resulting from failure of first arch development. Interestingly, only the tragus, which is known to derive from the first arch, is absent in this syndrome (Hunter and Yotsuyanagi, 2005; Wood-Jones and I-Chuan, 1934). Lastly, based on the analysis of the recurrent localisation in the neighbourhood of the tragus of frequent human congenital abnormalities such as pre-auricular fistulae and appendages, Wood-Jones and I-Chuan (Wood-Jones and I-Chuan, 1934) already came to the conclusion that the human pinna is mainly of hyoid origin. This latter conclusion is now fully supported by our current fate mapping in the mouse, supporting evolutionary conservation of the pharyngeal arch contribution to the definitive external ear in mammals. Overall, our results could be of interest from a clinical standpoint because they might facilitate our understanding of human syndromes, notably concerning the correlation between the embryological origin of the external ear, gene expression patterns, and the interpretation of the phenotypic outcome of their disruption.

By reconsidering the embryological origin of the EAC, our fate map as well as 3D tissue reconstruction allow a better understanding of the previously described mouse Hoxa2 knockout phenotype (Rijli et al., 1993). The finding that the mouse EAC and its surrounding mesenchyme derive from the first pharyngeal arch and do not develop at the first pharyngeal cleft better explains the observed partial duplication of the EAC in Hoxa2 mutants (Rijli et al., 1993). Our analysis indicates that, rather than forming the EAC, the first pharyngeal cleft maps at the border between EGFP⁺ and EGFP^- NCCs and could represent the axis of symmetry on either side of which a mirror-image duplication of the EAC occurs in Hoxa2 mutant mice.

Hoxa2 inactivation in the mouse results in a mirror-image homeotic transformation of the second arch-derived stapes, styloid process of the temporal bone and lesser horn of the hyoid bone into a duplicated set of first arch-like structures, namely the proximal part of the jaw (Meckel’s) cartilage, the incus, malleus, tympanic bone and proximal gonial bone (Gendron-Maguire et al., 1993; Rijli et al., 1993). These structures are normally derived from the Hox-negative NCCs arising from the rostral hindbrain (Köntges and Lumsden, 1996; Kuratani, 2005; Minoux and Rijli, 2010; Rijli et al., 1993; Santagati and Rijli, 2003; Takechi and Kuratani, 2010). This subset of first arch NCCs and second arch NCCs share a common Hox-free ground (default) patterning molecular programme upon which Hoxa2 expression selects second arch identity (Köntges and Lumsden, 1996; Kuratani, 2005; Minoux et al., 2009; Minoux and Rijli, 2010; Rijli et al., 1993; Santagati and Rijli, 2003; Takechi and Kuratani, 2010). However, the possible extension of this model to other NCC-derived structures, such as the external ear, was not previously assessed.

In Wnt1^{Hoxa2-IRES-EGFP} embryos, ectopic Hoxa2 expression in Hox-negative cranial NCCs induces abnormalities in structures derived from midbrain and forebrain Hox-negative NCCs, such as most of the jaw, facial and skull bones, supporting the notion that ectopic Hox gene expression interferes with normal craniofacial development (e.g. Couly et al., 1998). However, we found that conditional ectopic Hoxa2 expression is also sufficient to repattern mandibular arch mesenchyme, derived from rostral hindbrain NCCs, and to generate a mirror-image transformation into a second arch-like structure, the pinna, at the expense of the EAC. This further extends the proposal of a Hox-free ground patterning programme shared by hindbrain NCCs, and suggests that the first arch-derived mesenchyme normally lining the EAC has been homeotically transformed into a duplicated pinna, normally a second arch derivative. Lastly, the ability of Hoxa2 to repattern (a subset of) first arch NCCs appears to be conserved (Grammatopoulos et al., 2000; Pasqualetti et al., 2000). In this respect, we additionally observed in Wnt1^{Hoxa2-IRES-EGFP} newborns skeletal morphological changes that could be interpreted as partial transformations of middle ear first arch skeletal elements into second arch-like structures (not shown).

Several cases in the literature have reported partial or total duplication of the human pinna (Baschek et al., 2006; Gore et al., 2006; Hunter and Yotsuyanagi, 2005; Ku et al., 1998; Mishra and Misra, 1978; Pan et al., 2010). Although the genetic basis of such a phenotype has not been investigated, it is tempting to speculate that ectopic HOXA2 expression might underlie the ‘polyotia’ or ‘mirror ear’ phenotype observed in humans. As Hoxa2 alone is able to induce the whole developmental programme underlying the morphogenesis of the pinna, our data moreover suggest that numerous genes involved in human auricle abnormalities are HOXA2 targets.

Our study provides novel insights into the largely unknown molecular programme involved in external ear morphogenesis. The identification of such a programme could improve our understanding of the HOXA2 mutant phenotype in humans. Indeed, we have shown that Hoxa2 regulates the expression of Eya1, which is involved in the branchio-oto-renal syndrome in humans (Abdelhak et al., 1997; Kochhar et al., 2007). Our functional and molecular analyses also reveal that Hoxa2 is involved in the regulation of Bmp5, Bmp4 and Tsg expression and Smad1/5/8 activity. Thus, Hoxa2 acts upstream of the BMP canonical signalling pathway during external ear morphogenesis. A recent study has reported a role for Hoxa2 in activating the Wnt-β-catenin signalling pathway in the second arch (Donaldson et al., 2012), in part through the regulation of Wnt5a expression, a gene that when inactivated in mouse affects external ear development (Qian et al., 2007). Altogether, these data uncover roles for BMP and Wnt signalling in instructing external ear morphogenesis downstream of Hoxa2; how these pathways interact remains to be determined. More generally, understanding how multiple genes integrate into functional networks is becoming key to a full comprehension of the molecular processes that underlie normal and defective external ear morphogenesis.

We thank F. Santagati and A. Vitobello for experimental assistance and discussion; B. Hogan for the kind gift of the Bmp4 conditional allele; and L. Fasano for the Tshz2 probe.