Region-specific endodermal signals direct neural crest cells to form the three middle ear ossicles

The peripheral auditory system in mammals comprises three distinct parts: the outer, middle and inner ear. Sound waves collected by the outer ear are converted into vibrations at the tympanic membrane and these vibrations, amplified by the chain of three ossicles in the middle ear, the malleus, incus and stapes, are relayed to the cochlea of the inner ear (Anthwal and Thompson, 2016). At the cochlea, these mechanical signals are converted by the sensory hair cells into chemical signals, which activate the innervating cochlear nerves to generate electrical impulses that are transmitted to the brain. Each ossicle has a unique structure and function to efficiently transmit and amplify sound from the environment (Ozeki-Satoh et al., 2016). A break in this transmission, such as ossicular malformations or stapes ankyloses, can result in conductive hearing loss (Quesnel et al., 2015).

Middle ear ossicles are derived from neural crest cells (NCCs; Anthwal and Thompson, 2016; Chapman, 2011). In the craniofacial region, NCCs migrate from the dorsal neural tube and populate the mesenchymal area between the ectodermal and endodermal layers of the pharyngeal arches (PAs; also known as branchial arches). NCCs that originate from the posterior midbrain and the 1st and 2nd rhombomeres of the hindbrain migrate to PA1 and condense to a single cartilage structure that is later separated to form the malleus and incus as well as Meckel's cartilage (Fig. 1H). In contrast, NCCs that originate from the 4th rhombomere migrate to PA2 and give rise to the stapes and hyoid cartilage (Fig. 1H; Bhatt et al., 2013; Minoux and Rijli, 2010). It has been shown that the anterior and posterior identity of the NCCs is pre-determined by their origin in the hindbrain (Minoux and Rijli, 2010). Thus, NCCs that form malleus-incus and stapes are thought to be fated prior to migration into PA1 and PA2, respectively (Bhatt et al., 2013; Minoux and Rijli, 2010).

Signals from the pharyngeal endoderm such as SHH, BMP4 and fibroblast growth factors (FGFs) play important roles in instructing the size, shape and location of NCCs during development of cartilage and bone in the craniofacial region (Abu-Issa et al., 2002; Chapman, 2011; Couly et al., 2002; Jeong et al., 2004; Kanzler et al., 2000; Moore-Scott and Manley, 2005). Some of these molecules have also been implicated in middle ear ossicle development. Notably, disturbance of SHH signaling in NCCs causes developmental defects in nearly all NCC-derived craniofacial cartilage and bone, including middle ear ossicles (Billmyre and Klingensmith, 2015; Jeong et al., 2004). These defects resulted from increased cell death and decreased cell proliferation in the mesenchymal layers of PAs, where NCCs are populated. Inactivation of FGF signaling in mice has also been demonstrated to cause severe craniofacial defects, including loss of middle ear ossicles due to massive cell death in the PAs (Abu-Issa et al., 2002; Trumpp et al., 1999). Although both SHH and FGF signaling have broad effects on PA-derived structures, their specific requirements in ossicle formation are not well defined. It is also unclear how the ossicular NCCs find their proper locations within the two PAs and initiate mesenchymal condensation to form specific ossicles that are in perfect alignment with each other as well as the outer and inner ears.

In this study, we investigated the roles of SHH and BMP4 signaling in middle ear ossicle formation in mouse. We generated endodermal-specific knockouts of Shh or Bmp4 as well as NCC-specific knockouts, which are compromised in their ability to respond to these signaling molecules. Our results show that endodermal SHH signaling is essential for NCCs to initiate mesenchymal condensation to form the malleus and incus in the PA1, whereas endodermal BMP4 signaling is required for stapedial condensation in the PA2.

We analyzed the earliest stage of middle ear development using expression of Sox9 as an indicator for mesenchymal condensation and aggrecan (Acan) as a cartilage marker from embryonic day (E) 10.5 to E15.5 (Fig. 1A-G). In addition, Hoxa2 was used to demarcate the border between PA1 and PA2 (Minoux and Rijli, 2010), and Bapx1 (Nkx3-2) was used to follow the condensation in PA1 for the malleus-incus and Meckel's cartilage (Tucker et al., 2004). The Sox9-positive condensation of NCCs to form the middle ear ossicles is first evident at E10.5 in the mesenchymal regions of PA1 and PA2 (Fig. 1A,B; Fig. S1A-C). The border between PA1 and PA2 was evident by the restricted expression of Hoxa2 in the mesenchyme of PA2 from E10.5 to E12.5 (Fig. S1B,F,I). The single condensation in PA1 gradually separates to form the malleus and incus by E13.5 (Fig. 1B-E; Fig. S1). Bapx1 was expressed in the mesenchyme medial to the malleus-incus condensation at E11.5 and later in the mesenchyme surrounding the malleus at E13.5 (Fig. S1G,J,M). The condensation in PA2 for the stapes starts to acquire its unique ‘stirrup’ shape at E11.5 (Fig. 1C; Fig. S1E) and is connected to the otic capsule of the inner ear by E13.5 (Fig. 1E). Hoxa2 expression in PA2 partially overlapped with stapes condensation at E10.5 and E11.5 and was downregulated in the stapes at E12.5 (Fig. S1B,F,I,L). The Sox9-positive ossicular condensations become Acan positive by E15.5 (Fig. 1G).

To confirm that these ossicle-related condensations observed in PA1 and PA2 are derived from NCCs, we compared the Sox9 expression domains with NCC lineage tracing by crossing Wnt1-Cre mice with a Cre reporter, Rosa26-lacZ (R26R) mice (Chai et al., 2000) (Fig. 1I-P). In coronal sections (Fig. 1I,M), Sox9 expression in the malleus-incus condensation was observed in the mesenchyme lateral to the pharyngeal endoderm in the PA1 region (Fig. 1J), whereas Sox9 expression for stapedial condensation was observed in the mesenchyme dorsal to the pharyngeal endoderm in the PA2 region (Fig. 1N). The lacZ-positive NCC lineage cells (identified by X-gal staining) were populated broadly in the mesenchymal regions of PA1 and PA2, which encompassed the Sox9-positive condensation areas for both the malleus-incus and stapes (Fig. 1J-L,N-P). Together, these results show that the identified condensations of middle ear ossicles at E10.5 in PA1 and PA2 are part of the NCC lineage.

SHH signaling has been shown to play important roles in normal development of craniofacial structures, including the middle ear ossicles, by promoting proliferation and survival of NCCs in the mesenchymal layers of PAs (Jeong et al., 2004); however, it remains unclear whether SHH signaling contributes to early stages of middle ear development, such as migration and condensation of NCCs in the prospective ossicular regions. We thus examined the spatial relationship between the initial condensation of middle ear ossicles and SHH signaling at E10.5 (Fig. 2). In a cross-section of the PA1 region (Fig. 2K), Shh is broadly expressed in the pharyngeal endoderm (Fig. 2A), and Ptch1, a readout of SHH signaling, is expressed in the surrounding mesenchyme in a graded pattern (Fig. 2B). The malleus-incus condensation overlapped with the graded Ptch1 expression domain (Fig. 2C). In contrast, Shh expression was specifically absent in the pharyngeal endoderm just beneath the stapes condensation (Fig. 2L, red bracket). Ptch1 expression was also much weaker in the stapes condensation compared with the mesenchyme surrounding the endoderm (Fig. 2M,N, red bracket). These results indicate that endodermal SHH signaling is much stronger neighboring the condensation for malleus-incus than for stapes, raising the possibility that SHH signaling plays differential roles in the condensation of the middle ear ossicles.

We tested this hypothesis by inactivating the ability of NCCs to respond to SHH by deleting the transducer of SHH signaling, Smo, in NCCs using Wnt1-Cre. In Wnt1-Cre; Smo^lox/lox mutants, Ptch1 expression was greatly reduced in the mesenchyme where NCCs normally populate both PA1 and PA2 at E10.5 (Fig. 2G,R, red asterisks), whereas Shh expression in the pharyngeal endoderm was unaffected (Fig. 2F,Q), suggesting that SHH signaling is specifically inactivated in NCC derivatives in Wnt1-Cre; Smo^lox/lox mutant embryos. Interestingly, Sox9 expression in PA1 was selectively abolished in the area where malleus-incus condensation normally occurs (Fig. 2H, red asterisk). In contrast, Sox9 expression associated with stapes condensation, though reduced, remained in PA2 at E10.5 (Fig. 2S, arrow). These results suggest that NCCs require SHH signaling to initiate mesenchymal condensation for the malleus-incus but not for the stapes.

The lack of malleus-incus condensation in Wnt1-Cre; Smo^lox/lox mutant embryos may be due to either a failure of NCC migration to the malleus-incus region or a failure of NCC condensation or survival after migration. It has been reported that SHH signaling is not essential for NCC migration but is essential for NCC condensation and survival (Billmyre and Klingensmith, 2015; Brito et al., 2006; Jeong et al., 2004). To follow the lineage of NCC derivatives, we genetically labeled NCC derivatives by crossing Wnt1-Cre; Smo^lox/lox mice with R26R-lacZ mice. Interestingly, lacZ-positive cells were present in the prospective malleus-incus region even though no clear malleus-incus condensation was detected (Fig. 2D,I). These results suggest that NCCs migrate to the malleus-incus region but fail to initiate condensation in the absence of SHH signaling.

When Sox9 expression was examined in Wnt1-Cre; Smo^lox/lox embryos at E11.5, the malleus-incus condensation remained absent (Fig. 2E,J, red asterisk), suggesting that the lack of malleus-incus condensation is not due to a developmental delay. Interestingly, the stapes condensation, which was present at E10.5 (Fig. 2N,S), had disappeared by E11.5 (Fig. 2P,U, red asterisk). Consistent with this, no middle ear cartilages were observed in Wnt1-Cre; Smo^lox/lox embryos at E15.5 (Fig. 3B, red asterisks). These results indicate that SHH signaling, which is dispensable for the initial condensation of the stapes, is required for the maintenance of this condensation.

Because SHH signaling is important for NCC proliferation and survival during craniofacial development (Ahlgren and Bronner-Fraser, 1999; Brito et al., 2006; Jeong et al., 2004), we determined whether the loss of ossicular condensation in Wnt1-Cre; Smo^lox/lox embryos was due to abnormal NCC proliferation or survival. Cell proliferation was analyzed by counting the number of cells that had incorporated 5-ethynyl-2′-deoxyuridine (EdU) in the condensation areas (Fig. 4A-B′,F-G′). No significant difference in the number of EdU-positive cells between control and Wnt1-Cre; Smo^lox/lox embryos was detected (Fig. 4K). In contrast, the number of terminal deoxynucleotidyl transferase dUTP nick end labeling (TUNEL)-positive cells was significantly increased in both the malleus-incus and stapedial condensation areas in Wnt1-Cre; Smo^lox/lox embryos compared with control embryos (Fig. 4C-D′,H-I′,L). These results indicate that although SHH signaling plays differential roles in the initial condensation of the malleus-incus in PA1 and the stapes in PA2, this signaling pathway is essential for NCC survival in both PAs regardless of condensation status.

Next, we investigated whether SHH from the pharyngeal endoderm is important for condensation of the malleus-incus and maintenance of the stapes. To delete Shh in the endodermal epithelium specifically, Shh^lox/lox mice were crossed with Foxg1^Cre mice, which have been shown to induce Cre-mediated recombination in the foregut epithelium (Hébert and McConnell, 2000). In Foxg1^Cre; Shh^lox/lox embryos, endodermal Shh as well as mesenchymal Ptch1 expression were completely abolished in both PA1 and PA2 (Fig. 5F,G,Q,R, red asterisks). Consistent with inactivation of SHH signaling in the NCCs (Fig. 2), Sox9 expression was strongly downregulated in the malleus-incus condensation area in PA1 (Fig. 5C,H, red asterisk), whereas Sox9 expression was present, although reduced, in the stapedial condensation in PA2 (Fig. 5N,S, arrows). Examination of Tfap2a expression to identify migrating NCCs indicated that this gene was not expressed in the prospective malleus and incus region in PA1 but was expressed in the stapes region in PA2 (Fig. 5D,I,O,T). Consistent with previous results (Fig. 2P,U), condensations for both malleus-incus and stapes were not present at E11.5 in these animals (Fig. 5P,U, red asterisk). It should be noted that the Cre activity of Foxg1^Cre is also detectable in areas of pharyngeal ectoderm and mesoderm (Tavares et al., 2012). In addition, we noticed that PA development is severely disrupted in Foxg1^Cre; Shh^lox/lox embryos. Thus, the middle ear phenotypes of Foxg1^Cre; Shh^lox/lox embryos may be confounded by malformed PA structures as well as contributions from other sources. Nevertheless, our results strongly suggest that SHH emanating from the pharyngeal endoderm plays an essential role in the condensation and maintenance of the ossicles.

Forced expression of Shh in the chondrocytes disrupts normal limb cartilage development by affecting cell proliferation and apoptosis, resulting in impaired joint formation (Tavella et al., 2006, 2004). We thus investigated whether constitutive activation of SHH signaling in NCCs affects middle ear development. To activate SHH signaling constitutively in NCCs, we used Wnt1-Cre; SmoM2 mutant embryos, in which a constitutively active form of Smo was expressed in NCC derivatives (Jeong et al., 2004). In Wnt1-Cre; SmoM2 embryos, Ptch1 expression was expanded in the entire mesenchymal region where NCCs normally reside (Fig. S2D,K, arrowheads). Sox9 expression in the ossicular condensation was also increased in Wnt1-Cre; SmoM2 embryos compared with control embryos at E10.5 (Fig. S2E,L); however, the number of EdU-positive cells in the middle ear condensation regions were not significantly different between control and Wnt1-Cre; SmoM2 embryos (Fig. S2C,F,J,M,O). This finding suggests that the enlarged condensation is not due to increased proliferation but rather an upregulation of Sox9 expression (Tavella et al., 2004). In E15.5 Wnt1-Cre; SmoM2 embryos, the middle ear cartilages were fused together and displaced from the prospective oval window of the otic capsule (Fig. 3C, arrowheads). These results suggest that tight regulation of SHH signaling in the NCCs is important for normal development and localization of middle ear ossicles.

The presence of the stapedial condensation in the absence of SHH signaling suggests that other signaling pathways are at play in the PA2 region for mediating NCC condensation. In addition to Shh, it has been shown that Fgf8 and Bmp4 are expressed in the pharyngeal endoderm and play essential roles in patterning the PA-derived structures (Moore-Scott and Manley, 2005). However, in Fgf8-null embryos, the stapes was found to be normal or only slightly smaller than in control embryos (Abu-Issa et al., 2002), excluding a requirement for FGF8 in the initial condensation of stapes. In contrast, BMP signaling is important for chondrogenesis and functions by regulating the induction and maintenance of Sox9 expression in vitro and in vivo (Bandyopadhyay et al., 2006; Kumar et al., 2012; Semba et al., 2000). Interestingly, we found that Bmp4 was specifically expressed in the endodermal epithelium just beneath the stapes condensation in PA2 (Fig. 6E,F, bracket) but not adjacent to the malleus-incus or hyoid condensation (Fig. 6A,B,I,J). In addition, Shh, which is broadly expressed in the entire pharyngeal endoderm, was specifically absent in the Bmp4 expression domain (Fig. 6C,F,G,K, brackets).

These spatial relationships suggest a possible role for endodermal BMP4 signaling in NCC migration and condensation for stapes development. We tested this possibility by inactivating the ability of NCCs to respond to TGFβ/BMP signaling by deleting Smad4 in NCCs using Wnt1-Cre (Fig. 7). In E10.5 Wnt1-Cre; Smad4^lox/lox embryos, Sox9 expression was completely downregulated in the stapedial condensation area (Fig. 7M,R, red asterisk), whereas expression in the malleus-incus condensation area was unaffected (Fig. 7B,G, arrows). These results indicate that TGFβ/BMP signaling is selectively required for NCCs to initiate the mesenchymal condensation for the stapes.

Analysis of cell proliferation and cell death demonstrated that the numbers of EdU- or TUNEL-positive cells in the condensation regions of the malleus-incus (Fig. 7C,D,H,I) and the stapes (Fig. 7N,O,S,T) were not significantly different between control and Wnt1-Cre; Smad4^lox/lox embryos (Fig. 7W,X). Thus, the failure of stapedial condensation of NCCs in the absence of TGFβ/BMP signaling is unlikely to be due to abnormal cell proliferation or cell death.

As TGFβ/BMP signaling has been shown to be important for NCC formation and migration (Kanzler et al., 2000), it is possible that the loss of stapes condensation in Wnt1-Cre; Smad4^lox/lox embryos is due to a failure of NCC migration into the stapes region. To test this idea, we first examined normal NCC migration patterns into the prospective ossicular condensation regions from E9.5 to E10.5 (Fig. 8). In PA1, lacZ-positive NCC derivatives were present in the prospective malleus-incus region from E9.5 (Fig. 8D,H,L, arrows); however, condensation for the malleus-incus was not evident until E10.5 (Fig. 8A,E,I, arrow). In PA2, lacZ-positive NCC derivatives were not detected at E9.5 (Fig. 8Q, asterisk), began to localize in the lateral mesenchyme at E10.0 (Fig. 8U, arrowheads), and expanded medially to the stapes region at E10.5 (Fig. 8Y, arrow) concurrent with initiation of Bmp4 expression in the endoderm (Fig. 8O,S,W, bracket). Interestingly, in E10.5 Wnt1-Cre; Smad4^lox/lox embryos, lacZ-positive NCC derivatives were restricted in the lateral mesenchyme of the PA2 but not detected in the region where NCCs normally condense to form the stapes (Fig. 7P,U, arrowheads and red asterisk). These results suggest that NCCs fail to migrate to the location where the stapes condensation normally occurs when the TGFβ/BMP signaling pathway is disrupted.

To test whether Bmp4 expressed in the endoderm is the ligand specifically required for stapes condensation, we deleted Bmp4 in the endodermal epithelium using Foxg1^Cre; Bmp4^lox/Tm1 mice (Fig. 9K, red bracket). In these embryos, Sox9 expression in the prospective stapes condensation was lost (Fig. 9I,L, red asterisk), whereas this expression in the malleus-incus condensation was unaffected (Fig. 9B,E, arrow). Consistent with this, Tfap2a expression in the migrating NCCs was absent in the stapes condensation area in Foxg1^Cre; Bmp4^lox/Tm1 embryos (Fig. 9J,M, red asterisk). These results demonstrate that endodermal Bmp4 is specifically required for NCC migration and initial condensation for the stapes but not for the malleus and incus.

We next examined whether the loss of stapes condensation in the absence of TGFβ/BMP signaling leads to a specific loss of stapes in later development. Previous studies demonstrated that Foxg1^Cre; Bmp4^lox/Tm1 and Wnt1-Cre; Smad4^lox/lox mutants exhibit early embryonic lethality at E13.5 and E12.5, respectively (Chang et al., 2008; Ko et al., 2007). Thus, we examined the middle ear phenotypes at the latest stage that we could harvest the mutant embryos. In Foxg1^Cre; Bmp4^lox/Tm1 embryos, the stapes was specifically absent, whereas development of the malleus and incus appeared normal at E12.5 and E13.5 (Fig. 10G,H,J,K, red asterisks), confirming the specific requirement of endodermal Bmp4 in stapes development. Similarly, a specific loss of the stapes was observed in Wnt1-Cre; Smad4^lox/lox embryos at E12.5 (Fig. 10G,I, red asterisk). We also noticed, however, that the malleus-incus condensation was much reduced compared with that in control embryos (Fig. 10I, arrow). This result is consistent with a report that Smad4-mediated TGFβ/BMP signaling plays an important role in developmental progression of NCCs during craniofacial development (Ko et al., 2007).

In this study, we investigated how migratory NCCs find their destinations and develop into specific middle ear ossicles. Our results suggest that pharyngeal endoderm provides region-specific signaling molecules to initiate NCC condensation in PA1 or guide NCCs to specific locations within PA2 to differentiate into the three ossicles.

The loss of SHH function by either making NCCs unable to mediate SHH signaling (Wnt1-Cre; Smo^lox/lox) or abolishing the expression of Shh in the endoderm (Foxg1^Cre; Shh^lox/lox) resulted in similar phenotypes: failure of initial condensation of the malleus-incus but not of stapes (Fig. 11C,D). Notably, although Shh is broadly expressed in the endoderm and specifies malleus-incus condensation, Shh expression is specifically absent from the region of the endoderm closely associated with stapedial condensation (Fig. 2 and Fig. 11B). These results suggest that SHH signaling emanating from the endoderm is specifically required for the condensation of NCCs for malleus-incus generation in PA1 but not for the stapes in PA2. However, all three ossicles were eventually lost in the absence of SHH signaling (Figs 2, 3 and 5) (Billmyre and Klingensmith, 2015; Jeong et al., 2004), indicating that, although the requirements for SHH in condensation differs for each ossicle, the SHH requirement for survival is common for all ossicles.

Currently, how endodermal SHH signaling directs NCCs to initiate condensation in the prospective malleus-incus region remains unknown. In our SHH gain-of-function model (Wnt1-Cre; SmoM2), SHH signaling was activated in all Wnt1-positive NCC derivatives, resulting in expansion of the Ptch1 expression domain to mesenchymal areas at a distance from the Shh-expressing endoderm (Fig. S2). The Sox9 expression domain delineating malleus-incus condensation, however, was only slightly expanded, and this expanded region was restricted to areas around the original condensation domains (Fig. S2). Thus, activation of SHH signaling alone is not sufficient to promote mesenchymal condensation in the migratory NCCs. In addition, Shh is expressed broadly throughout the endodermal epithelium of PAs, suggesting that endodermal SHH signaling may not be specific enough to recruit NCCs to the prospective malleus-incus region, and other factors may cooperate with endodermal SHH signaling to direct NCCs to the proper location and promote mesenchymal condensation. One such factor could be FGF8, as Fgf8 is expressed in the ectoderm of PA1 and in the endoderm of PA2, PA3 and PA4 (Abu-Issa et al., 2002; Moore-Scott and Manley, 2005; Trumpp et al., 1999). However, the expression domain of Fgf8 in the ectoderm of PA1 is not in close proximity to where the malleus-incus develops (data not shown). Therefore, it is not apparent how FGF8 signaling could play a role in directing NCC migration to the malleus-incus region. Nevertheless, mice lacking Fgf8 also show increased apoptotic cell death in PA1, resulting in the loss of PA1-derived craniofacial structures, including the malleus and incus (Abu-Issa et al., 2002; Moore-Scott and Manley, 2005; Trumpp et al., 1999). The similarity of cell death in the prospective malleus-incus area in both Fgf8 and Shh mutants suggests that ectodermal FGF8 and endodermal SHH are both survival factors for NCCs but whether they function together to specify NCC migration to form malleus and incus remains to be determined.

In contrast to the broad expression of Shh, Bmp4 was specifically expressed in the endodermal epithelium of PA2 just below the site of stapes condensation and not in other endoderm epithelia, including the vicinity of malleus-incus condensation (Fig. 6 and Fig. 11B). Interestingly, Shh, which is broadly expressed in the endoderm, is not expressed in the Bmp4-expressing region (Fig. 6 and Fig. 11B). The reciprocal expression patterns of Shh and Bmp4 may result from negative regulation between the two pathways, as shown in the endoderm of PA3 where Bmp4 expression is expanded in Shh^−/− mutants (Moore-Scott and Manley, 2005). Bmp4 expression in the PA2 endoderm, however, was not changed in Shh^−/− mutants (Moore-Scott and Manley, 2005) or in Foxg1^Cre; Shh^lox/lox mutants, suggesting that SHH exerts differential effects on Bmp4 expression in different PA positions. The mechanism driving the pattern of Bmp4 expression in the endoderm at the site of stapes condensation is currently unknown.

Nevertheless, consistent with its definite expression domain, deletion of endodermal Bmp4 expression (Foxg1^Cre; Bmp4^lox/Tm1) or rendering NCCs unable to mediate BMP signaling (Wnt1-Cre; Smad4^lox/lox) resulted in loss of condensation for the stapes but not for the malleus-incus (Fig. 11E,F). Previous studies using transgenic mice or chicken explant cultures have shown that BMP2/4 signaling plays crucial roles in NCC migration and also directly or indirectly provides positional information to NCCs and induces Sox9 expression (Kanzler et al., 2000; Kumar et al., 2012; Saito et al., 2012; Semba et al., 2000). Xnoggin transgenic mice, in which BMP2/4 signaling is inactivated in the NCCs populating the PA2 and more caudal PAs, fail to form the skeletal structures derived from the targeted PAs including the stapes (Kanzler et al., 2000). Implantation of BMP4 protein-soaked beads into chicken PA2 mesenchyme or into mouse mandibular explants induced Sox9 expression, indicating that BMP4 is sufficient to induce mesenchymal condensation (Kumar et al., 2012; Semba et al., 2000). The transplantation of pharyngeal endoderm into chicken embryos altered the morphology of the columella (a structure analogous to middle ear ossicles in mammals) (Zou et al., 2012). Based on these chicken explants results and our mouse studies here, we postulate that endodermal Bmp4 has a conserved role in middle ear ossicle formation.

The lineage-tracing analysis that we performed in Wnt1-Cre; Smad4^lox/lox; R26R mice showed that lacZ-positive NCC derivatives were specifically depleted from the prospective stapes condensation area, whereas extensive lacZ-positive NCC derivatives populated other mesenchymal areas of PAs (Fig. 7). Thus, the loss of stapes condensation observed in Wnt1-Cre; Smad4^lox/lox embryos appears to result from a failure of NCC migration into the prospective stapes region. Similarly, upon deletion of endodermal Bmp4 expression, Tfap2a expression, which indicates NCC migration, was abolished in the prospective stapes condensation area but not in the prospective malleus-incus condensation area (Fig. 9). These results suggest that BMP4 signaling emanating from the endodermal epithelium attracts migratory NCCs to populate the prospective stapes region and to initiate mesenchymal condensation.

How does endodermal BMP4 signaling direct the migrating NCCs to populate the future stapes condensation region? Thus far, several receptor/ligand complexes, such as Eph/ephrin, Nrp/Sema and Cxcl12/Cxcr4, have been implicated in directional NCC migration through roles as chemoattractant or -repellent factors (Minoux and Rijli, 2010). For example, ephrin B1 expressed in the pharyngeal cleft between PA1 and PA2 directs NCCs expressing EphA4 and EphB1/B3 into PA2 (Adams et al., 2001). The interaction between the chemokine CXCL12 and its receptor CXCR4 is of particular interest. Cxcl12 is expressed in the ectoderm and pharyngeal endoderm at the time of NCC migration into the PAs, whereas Cxcr4 is expressed in the migrating NCCs (Escot et al., 2016). Defective CXCR4 signaling in both zebrafish and chicken embryos has been shown to result in craniofacial and neural anomalies due to aberrant NCC migration defects (Escot et al., 2016; Olesnicky Killian et al., 2009). In addition, CXCL12 acts as a chemoattractant for NCC migration in sympatho-adrenal specification (Saito et al., 2012). Interestingly, BMP signaling emanating from the dorsal aorta is required for Cxcl12 expression in the para-aortic mesenchyme (Saito et al., 2012). It will be interesting to determine whether endodermal BMP4 signaling in PA2 dictates NCC migration by promoting expression of Cxcl12 in the prospective stapes region.

Our results demonstrate that endodermal SHH signaling is crucial for NCCs to initiate mesenchymal condensation for the malleus and incus in PA1. Based on the lineage and TUNEL analyses, we postulate that in the absence of SHH, NCCs reach the appropriate location in PA1 but fail to thrive and condense to form the malleus and incus. By contrast, BMP4 emanating from the endoderm of PA2 is required for NCCs to initiate stapes formation. Owing to the absence of NCC lineage cells in the prospective stapes region in PA2 and the lack of apparent cell death in the region when NCCs become unresponsive to BMP signaling, we postulate that the lack of BMP4 signaling fails to attract NCCs to the correct location and thereby affects stapes formation. Additionally, although SHH is not required for the mesenchymal condensation of the stapes, it is required for the maintenance of the condensed mesenchyme. Together, these results demonstrate that region-specific interactions between the migratory NCCs and the pharyngeal endoderm play a crucial role in guiding NCCs to migrate to the proper locations and differentiate into specific middle ear ossicles. Further studies aiming to identify other signaling molecules that cooperate with these endodermal signals in the formation of middle ear ossicles will help to elucidate the pathology of ossicular anomalies leading to conductive hearing loss.

The generation of Wnt1-Cre; Smo^lox/lox, Wnt1-Cre; SmoM2 (Jeong et al., 2004), Wnt1-Cre; Smad4^lox/lox (Ko et al., 2007), Foxg1^Cre; Shh^lox/lox (Bok et al., 2013) and Foxg1^Cre; Bmp4^lox/Tm1 (Chang et al., 2008) mice was previously described. A ROSA26 conditional reporter line (R26R) was used to label NCC derivatives (Soriano, 1999). Littermates carrying genotypes other than the above-mentioned conditional knockouts were used as controls. Wnt1-Cre; Smad4^lox/lox; R26R embryos were generated by mating Wnt1-Cre; Smad4^lox/+ mice with Smad4^lox/lox; R26R mice. Similarly, Wnt1-Cre; Smo^lox/lox; R26R embryos were generated by mating Wnt1-Cre; Smo^lox/+ mice with Smo^lox/+; R26R mice. All animal protocols were approved by the Institutional Animal Care and Use Committee at Yonsei University College of Medicine.

Antisense RNA probes for Bmp4 (Morsli et al., 1998), Sox9 (Wright et al., 1995), Ptch1 (Goodrich et al., 1996), Shh (Echelard et al., 1993), Acan (Sandell et al., 1991), Tfap2a (+513-+1287, NM_001122948.2), Bapx1 (+720-1438, NM_007524.3) and Hoxa2 (+671-+1470, NM_010451.2) were labeled with digoxigenin. In situ hybridization was performed as previously described (Morsli et al., 1998). The micrographs of gene expression patterns were acquired using Olympus BX40 and Leica DM2500 optical microscopes. All in situ hybridization figures are representative of at least three different samples in two or more independent experiments.

Embryos were fixed in 2% paraformaldehyde, 2 mM MgCl₂ and 5 mM EGTA for 2 h at 4°C, dehydrated in 30% sucrose overnight at 4°C, and frozen in OCT compound (Tissue-Tek). Frozen embryos were sectioned at a thickness of 12 μm using a cryostat (HM 525, Thermo Scientific) and stained with 5 mM K₃Fe(CN)₆, 5 mM K₄Fe(CN)₆, 2 mM MgCl₂, 0.02% NP-40, 1 mg/ml X-gal and 1× PBS at 37°C overnight. Micrographs of the X-gal-stained sections were acquired using Olympus BX40 and Leica DM2500 optical microscopes. All lacZ detection experiments were performed in at least three different embryos.

Cell proliferation was performed using a Click-iT EdU imaging kit (C10337, Thermo Fisher Scientific). Each pregnant mouse was injected three times with 10 mg/kg body weight EdU at 2-h intervals. Two hours after the final injection, embryos were harvested and fixed in 4% paraformaldehyde (PFA) in 1× PBS overnight at 4°C. Embryos were then dehydrated in a 30% sucrose solution overnight at 4°C and finally frozen in OCT compound (Tissue-Tek). The embryos were sectioned at a thickness of 12 μm onto Superfrost Plus slides (Tissue-Tek) using a cryostat (HM 525, Thermo Scientific) and then stored at −20°C until use. The sections were incubated with 500 μl freshly prepared EdU detection solution, which was prepared according to the manufacturer's instructions, for 30 min. This solution contained 430 μl of 1× Click-iT reaction buffer, 20 μl of CuSO₄, 1.2 μl Alexa Fluor 488 azide and 50 μl 1× freshly made reaction buffer additive in deionized water (RBA). Nuclear staining with 5 μg/ml Hoechst 33342 was performed for 30 min. After washing with 1× PBS, the slides were mounted with ProLong Gold antifade reagent (Thermo Fisher Scientific). Micrographs of EdU-positive cells were acquired with a Nikon eclipse Ti-U fluorescence microscope.

The cell death assay was performed using the ApopTag Plus Peroxidase In Situ Apoptosis Detection Kit (Millipore, S7101), which detects apoptotic cells by labeling DNA strand breaks by the indirect TUNEL method. Embryos were harvested and fixed in 4% PFA in 1× PBS overnight at 4°C, dehydrated in 30% sucrose solution overnight at 4°C, and finally frozen in OCT compound (Tissue-Tek). The embryos were sectioned at a thickness of 12 μm on Superfrost Plus slides (Tissue-Tek) using a cryostat (HM 525, Thermo Scientific) and were stored at −20°C. The staining method was performed according to the manufacturer's protocol. In brief, the sections were fixed in 1% PFA in 1× PBS, permeabilized in 2:1 pre-cooled ethanol:acetic acid, incubated in 3% H₂O₂, equilibrated in equilibration buffer, incubated with TdT enzyme in a humidifying chamber for 1 h, and then incubated for 10 min in stop/wash buffer. Then, these sections were conjugated with anti-digoxigenenin peroxidase, stained with peroxidase substrate (DAB substrate), and finally counterstained with 1% Methyl Green solution. The slides were then mounted using Permount Mounting Medium (17986, EMS). Micrographs were acquired using a Leica DM2500 optical microscope.

Adjacent sections to those used for EdU or TUNEL analysis were used for in situ hybridization for Sox9 to identify the middle ear condensation regions. Cell counting was performed within the rectangular area for malleus-incus condensation (225 μm×90 μm) and stapedial condensation (90 μm×84 μm). The investigators were blind to the genotypes of the controls and conditional knockouts until the completion of cell counting. All cell counts were determined for at least three different embryos for each genotype. Data are displayed as box plots, which were generated using the BoxPlotR web tool (http://shiny.chemgrid.org/boxplotr/). Statistical analyses were conducted using Microsoft Excel (Microsoft) and SPSS statistics 23 software (IBM). Two-tailed, unpaired Student's t-tests were used to determine statistical significance. P<0.05 was considered as significant.

We thank Dr Doris K. Wu for critical reading of the manuscript, and Mr Dong Jin Lee, Mr Duyeol Han, and Mr Iljin Bang for their technical assistance. We also thank Dr Brigid Hogan for Bmp4^lox/lox and Bmp4^Tm1/+ mice and Dr Yu Lan for arranging the shipment of Smad4^lox/lox mice.