Differing contributions of the first and second pharyngeal arches to tympanic membrane formation in the mouse and chick

The ear is divided into three anatomical compartments: the external, middle, and inner ear. The external ear consists of the external auditory meatus (EAM) and (only in mammals) the pinna, which collect sound. The inner ear contains the cochlea, a sensory organ for hearing that converts sound into neural pulses. The middle ear, which comprises the middle ear ossicle(s) and tympanic membrane (TM), amplifies air vibrations (sound) and transmits them to the inner ear. The morphology of the middle ear is highly associated with the lifestyle of terrestrial tetrapods (Clack and Allin, 2004; Wever, 1978). The mammalian middle ear has three ossicles: the malleus, incus, and stapes, whereas diapsids (modern reptiles and birds) have only one ossicle, the columella auris. Comparative morphological analyses have concluded that the incus and malleus have been derived from jaw-joint skeletal elements – the quadrate and articular elements, respectively – of a pre-mammalian ancestor (Anthwal et al., 2013; Maier and Ruf, 2016; Manley and Sienknecht, 2013; Reichert, 1837; Takechi and Kuratani, 2010). The stapes is homologous with the columella auris. Paleontological studies have suggested that the middle ear evolved independently several times in tetrapods (Clack, 2002; Laurin, 1998, 2010; Lombard and Bolt, 1979; Müller and Tsuji, 2007). However, precisely how this transformation from jaw-joint elements to middle ear ossicles took place has remained an intriguing question for many years.

The middle ear ossicles and TM are produced from all three germ layers, including the cranial neural crest cells (CNCCs) derived from the ectoderm, in the pharyngeal arches (PAs). The incus (or quadrate) and malleus (or articular) are mostly derived from CNCCs in the first pharyngeal arch (PA1), and the stapes (or columella auris) originates from CNCCs in the second pharyngeal arch (PA2) and from mesodermal cells (Köntges and Lumsden, 1996; Noden, 1986; Rijli et al., 1993; Thompson et al., 2012). The TM comprises three components both in mammals and diapsids: the external auditory meatus (EAM), which is derived from the surface ectoderm; the first pharyngeal pouch (PP1), which comprises the pharyngeal endoderm between PA1 and PA2; and fibrous mesenchymal cells (Chapman, 2011; Chin et al., 1997). The EAM and PP1 contact each other and form a thin layer, sandwiching the middle ear ossicle (Mallo, 2001, 2003). In the mouse, the EAM invaginates toward the forming tympanic ring, an intramembranous bone surrounding the EAM, and then contacts with PP1 to form the TM (Fig. 1A-C) (Mallo and Gridley, 1996; Mallo et al., 2000). The goosecoid (Gsc) knockout mouse exhibits loss of the EAM and tympanic ring, resulting in lack of the TM (Yamada et al., 1995). Therefore, development of the EAM and tympanic ring are considered to be interdependent in the mouse. Studies in mouse and chick have suggested that development of the middle ear requires histologically and molecularly coordinated interactions between the epithelium and mesenchyme in PA1 and PA2 (Fuchs and Tucker, 2015; Mallo, 2001, 2003; Takechi et al., 2016; Zou et al., 2012). Thus, the different patterns of the middle ears in mammals and diapsids should be ascribed to changes in the developmental programs for PA1 and PA2, and we consider that the evolutionary relationship between the TM and skeletal elements is key to solving morphological evolution of the middle ear in amniotes (Takechi and Kuratani, 2010; Kitazawa et al., 2015a).

Endothelin 1 (Edn1) signaling works as a selector of lower jaw development in jawed vertebrates (Clouthier et al., 2013; Medeiros and Crump, 2012). Inhibition of Edn1 signaling results in transformation of the lower jaw into an upper-jaw-like structure in mouse (Ozeki et al., 2004; Ruest et al., 2004; Sato et al., 2008) and chick (Kempf et al., 1998; Kitazawa et al., 2015a). We previously found that inhibition of Edn1 signaling causes loss of the TM in mouse, but its duplication in chick (Kitazawa et al., 2015a). Therefore, the TM is dependent on different mechanisms underlying developmental patterning along the dorsal-ventral axis; Edn1-dependent lower jaw formation influences TM formation in mammals, whereas Edn1-independent upper jaw formation is relevant to the diapsid TM (Kitazawa et al., 2015a). These results strongly suggest independent origins of the TM during amniote evolution. In other words, although the TM is formed by similar processes in both mammals and diapsids, the acquisition of the TM is likely to result from parallel evolution in these lineages. Thus, it is of interest to elucidate similarities and differences in the developmental mechanisms of the TM in the mammal and diapsid lineages to understand the genetic bases underlying this type of morphological evolution.

The positional specifications of PAs along the anterior-posterior (A-P) axis are mediated by a collinear, nested expression pattern of Hox genes in the CNCCs. No Hox genes are expressed in the CNCCs of PA1. By contrast, Hoxa2 is expressed in the CNCCs of PA2 and more posterior regions of the body, and this gene expression is considered to provide the morphological identity of PA2. In Hoxa2 knockout mice, the PA2-derived components are transformed into PA1 derivatives such as the malleus, incus and the tympanic ring (Fig. 1D-F) (Gendron-Maguire et al., 1993; Rijli et al., 1993). By contrast, ectopic expression of Hoxa2 in CNCCs (the NCC-Hoxa2 mouse) results in duplication of skeletal elements derived from PA2, such as the stapes, styloid process and pinna, in the PA1 region, as well as complete loss of the EAM and tympanic ring (Kitazawa et al., 2015b; see also Minoux et al., 2013). Furthermore, although it was formerly generally believed that the EAM invaginates at the first pharyngeal cleft, which is the groove located at the interface between PA1 and PA2 (Grevellec and Tucker, 2010; Jakubíková et al., 2005; Schoenwolf and Larsen, 2009), the mouse EAM has been found to originate in PA1 territory (Minoux et al., 2013). Therefore, in the mouse both the EAM and tympanic ring are considered to derive from PA1, and PA1 is crucial for TM formation.

An analysis that is analogous to establishing the Hoxa2 knockout mouse was performed in avian embryos. Noden (1983) grafted the quail neural fold of rhombomere (r) 1-2 from the hindbrain heterotopically into the otic levels of the chick. In the chimera, the PA2 neural crest-derived structures are absent and replaced with PA1 elements; it exhibits a duplicate of the jaw and loss of the columella auris (Noden, 1983; also see Couly et al., 1998). Duplication of an EAM-like structure is also described, suggesting that PA1 plays an important role in EAM formation in the chick, as in the mouse, and that the TM is formed by similar developmental mechanisms in terms of A-P patterning of the PAs in mammals and diapsids. However, detailed analysis of TM formation has not previously been performed in the chick-quail chimera (Noden, 1983).

In this study, we examine EAM and TM formation, focusing on the A-P patterning of the PAs in the mouse and chick. We visualize the boundary between PA1 and PA2 by examining Hoxa2 expression to compare the position of the forming TM in mouse and chick embryos. We also seek to elucidate differences in the developmental mechanisms of TM formation in the mouse and chick by producing a chick-quail chimera in which PA2 derivatives are transformed into PA1 derivatives. Our data shed light on similarities and differences of the developmental mechanisms of TM formation as well as developmental events which led to differing paths of middle ear evolution in mammals and diapsids.

In the Hoxa2 knockout mouse, the EAM and tympanic ring, in addition to the malleus and incus, are duplicated in a mirror image of the original components (Rijli et al., 1993) (Fig. 1A-F). In this study, we also confirm that both the original and duplicated EAM attach to PP1, resulting in duplication of the TM at E18.5 in the Hoxa2 knockout mouse (Hoxa2^EGFP/EGFP; Minoux et al., 2013) (Fig. 1D-F) (also see Mallo and Gridley, 1996). These results strongly suggest that the EAM and tympanic ring derive from PA1, and that in the mouse a PA1-specific genetic program also plays an important role in TM formation (Gendron-Maguire et al., 1993; Rijli et al., 1993). Indeed, the mouse EAM begins to invaginate in PA1 territory (Minoux et al., 2013), but whether the TM also forms in PA1 has not been directly examined. In addition, as the original and duplicated EAM-tympanic rings and TMs in the Hoxa2 knockout mouse are incomplete in shape and size compared with wild type (Fig. 1A,B,D,E), a partial contribution of PA2 mesenchyme to EAM-tympanic ring formation could not be ruled out.

To test this we used the Hoxa2^EGFP/+ mouse as a readout, in which EGFP is knocked in at the Hoxa2 locus and recapitulates Hoxa2 expression (Fig. 2) (Pasqualetti et al., 2002). We first mapped the position of the boundary between PA1 and Hoxa2^EGFP/+ PA2 in relation to the development of the TM. We examined aggrecan (Acan) expression by in situ hybridization to visualize chondrogenesis and checked the anterior boundary of the EGFP signal by immunohistochemistry at E12.5 (Fig. 2A,B). The malleus and incus began forming in the EGFP-negative PA1 area (Fig. 2B), whereas the orbicular apophysis of the malleus (a PA2 derivative; Mason, 2013; O'Gorman, 2005) was located in the EGFP-positive area (Fig. 2B). Thus, the anterior border of EGFP expression domain can be referred to as the boundary between PA1 and PA2 (Minoux et al., 2013). The apex of the EAM was positioned in PA1 territory at E12.5 (Fig. 2C) (Minoux et al., 2013). At E14.5, PA2 derivatives, such as the styloid process, were included in the EGFP-positive area, whereas PA1 derivatives such as the tympanic ring were mapped in the PA1 region (Fig. 2D). The whole TM was formed in PA1 territory (Fig. 2D). We further confirmed that the forming tympanic ring and TM were generated in the PA1 region in the E14.5 and E16.5 R4::Cre;Rosa-CAG-LSL-tdTomato mouse, in which the CNCCs in PA2 are labeled by tdTomato (Fig. S1). In the E18.5 Hoxa2^EGFP/+ mouse, PA2 derivatives, such as the orbicular apophysis and the styloid process, were still labeled with EGFP (Fig. 2E). We reconstructed the middle ear region based on consecutive histological sections of the E18.5 Hoxa2^EGFP/+ fetus after EGFP immunohistochemistry. The majority of the EAM, tympanic ring and TM were found in the EGFP-negative PA1-derived territory (Fig. 2F).

We next investigated EAM and TM formation in chick embryos. In E5 and E6 embryos, three hillocks on each side of the first pharyngeal cleft were detected (a-c and d-f, Fig. 3A,B) (Moldenhauer, 1877; Hamburger and Hamilton, 1951). We examined the relative positions of the boundary between PA1 and PA2 and of invagination of the EAM using the Hoxa2 expression pattern visualized by in situ hybridization. The anterior end of the Hoxa2 expression domain corresponded to PP1 at E6 (data not shown) and E7 (Fig. 3G) and can thus be referred to as the boundary between PA1 and PA2. The first developmental event of EAM formation was an invagination of the surface ectoderm at the depressed area of hillock ʻe' in the E6 embryo (Fig. 3B,F). This first invagination was detected in the Hoxa2 expression domain, namely the PA2 territory (Fig. 3F). Subsequently, the growing apex of the EAM traveled caudally in the PA2 area at E7 and E8 (Fig. 3C,D,G,H), and formed the TM by attaching to PP1 at E10 (Fig. 3E,I). Three-dimensional (3D) reconstruction of the jaw and middle ear based on consecutive histological sections of the E11 embryo showed that the EAM was associated with the forming columella auris, sandwiching the PP1-derived middle ear cavity to form the TM (Fig. 3L,M). At this stage, Hoxa2 expression was still detected in the region where PA2 derivatives, such as the columella auris and connective tissue of the mandibular depressor muscle, form (Fig. 3J,K). Importantly, we found that the mesenchyme located between the EAM and PP1 expressed Hoxa2 at E10 and E11 (Fig. 3I,K). We identified the boundary between PA1 and PA2 based on consecutive histological sections showing Hoxa2 expression at E11, and found that the entire forming TM was entirely encompassed within the Hoxa2 expression domain (Fig. 3L, yellow dotted line).

To confirm that the Hoxa2 expression pattern at E11 marks the PA2 region, we also performed cell lineage tracing analysis by producing chick-quail chimeras (Fig. S2). The left side of the neural fold that gives rise to CNCCs migrating into PA2 (r4 together with a portion of r3 and r5; Kulesa and Fraser, 2000) was excised from the Hamburger and Hamilton (HH) stage 9 chick embryo, and the corresponding area in a quail embryo was orthotopically transplanted (Fig. S2A). The transplanted quail cells were identified by QCPN immunohistochemistry at E11 (n=2). We found that the quail cells in the chimeras mostly distributed in the same manner as Hoxa2-expressing cells around the TM-forming region (Fig. S2C,D, compare with Fig. 3J,K). The intense signals of Hoxa2 expression correspond to a high density of QCPN-positive cells (Fig. 3K, Fig. S2D). The mesenchymal cells in the TM were QCPN positive (Fig. S2C,D). These findings collectively indicate that the TM in the chick develops in the PA2 region.

We also found that a depression formed near the EAM opening on the outside surface of the body (Fig. 3E). It was located in PA1 territory, the Hoxa2-negative mesenchymal region (Fig. 3E and arrowheads in Fig. 3I,J). Based on these results, the chick EAM can be divided into two parts: the depression of the skin in the PA1 region; and the meatus contributing to the TM in the PA2 region. In this study, these components are referred to as the wall part of the EAM (wEAM) and the TM part of the EAM (tEAM), respectively (Fig. 3E,L-N). These two EAM parts were still observed in the postnatal chick (Fig. 3O). Interestingly, the wEAM was located near the otic process of the quadrate (Fig. 3L).

Taking these findings together, we found that the chick EAM does not begin to develop from the surface ectoderm at the first pharyngeal cleft, but from the ectoderm in the PA2 region (Fig. 3F), and the entire developing TM is housed in PA2 territory (Fig. 3L,M). These results contrast with mouse TM formation, in which the EAM begins to develop from the ectoderm in the PA1 region (Minoux et al., 2013), and the developing TM is located at PA1 (Fig. 2).

The fact that the chick TM is formed in PA2 territory suggests that PA2 plays a crucial role in TM development in the chick. To test this, we attempted to generate a chick-quail chimera that exhibits loss of PA2 identity and duplication of PA1 derivatives, as performed by Noden (1983). The neural fold that gives rise to the CNCCs migrating into PA2 was excised from the HH9 chick embryo, and the neural fold corresponding to the midbrain-hindbrain boundary (MHB) in the HH9 quail embryo was heterotopically transplanted (Fig. 4A). Since CNCCs derived from the transplanted quail graft do not express Hoxa2, even when transplanted heterotopically (Couly et al., 1996), the CNCCs should exert a developmental program for establishing PA1 identity in the transplanted location. We confirmed that in the chimera the majority of the CNCCs in PA2 of the operated side were replaced by QCPN-positive quail cells, but the mesodermal cells, distributed in the lateral area of the arch artery, were derived from QCPN-negative chick cells (Fig. 4B). We collected the chimeras at E9-12 (n=5). The operated side of the chimera exhibited an ectopic beak, as in previous reports (n=4; Fig. 4C-F) (Noden, 1983). 3D reconstruction of consecutive histological sections of the chimeras showed that the endochondral components of the jaw, such as the quadrate, Meckel's cartilage and articular, together with intramembranous components, such as the squamosal, quadratojugal and surangular, were duplicated in a mirror image of the original components (n=5; Fig. 4G, compare with Fig. 3L).

Most of the duplicated and ectopic skeletal elements were derived from QCPN-positive quail cells (Fig. 4L), although a portion of the quadrate included QCPN-negative chick cells. This might be because the boundary between PA1 and PA2, which is normally constructed when CNCCs are migrating, was not established in the chimera and so the CNCCs in these PAs intermingled to some extent.

The columella auris was not formed on the operated side in all chimeras (n=5; Fig. 4H), although the footplate of the columella auris, which is considered to derive from mesoderm (Noden, 1986; Köntges and Lumsden, 1996), was detected in one chimera (Fig. 4L). Interestingly, as Noden (1983) described, ectopic cartilages were detected near the area where the columella auris is normally formed (n=5; Fig. 4I). These ectopic cartilages derived from both QCPN-positive and QCPN-negative cells (data not shown). The cartilages were fused with the duplicated quadrate, pterygoid or squamosal, which are skeletal elements that the columella auris is not connected to in normal development (Fig. 4I), and they were not considered to be portions of the original columella auris.

We further examined the hyoid region in the E11 and E12 chimeras, in which the hyoid bones were well formed (n=4). The entoglossum, a PA1 derivative (Couly et al., 1996), was duplicated on the operated side in all chimeras (Fig. 4J,K). However, the proximal part of the basihyal, a PA2 derivative (Couly et al., 1996), did not disappear and exhibited normal development (Fig. 4K). We also found that QCPN-positive quail cells contributed to the operated side of the basihyal in all chimeras (Fig. 4J,K, qc). From these results, it is likely that although the transplanted quail CNCCs migrated into the original PA2 territory on the operated side, normal chick cells migrating from the control side were able to exert normal development of the basihyal at the midline of the embryo even in the presence of the quail cells. Taken together, the phenotypes detected in the jaw and hyoid regions described above clearly showed that the PA2 derivatives were transformed into PA1 components in the chimeras.

The original EAM on the operated side is much smaller than that of the control side (n=5; Fig. 4D,F). The smaller EAM was formed in proximity to the otic process of the original quadrate (Fig. 4G,M), in PA1 territory where the mesenchyme is composed of QCPN-negative chick cells (Fig. 4N). Therefore, this incomplete EAM appeared to correspond to the wEAM (Fig. 4D,F). 3D reconstruction of consecutive histological sections of the E11 chimeras revealed that the wEAM and PP1 did not contact each other, resulting in an absence of TM formation (n=5; Fig. 4H, arrow).

In addition to formation of only the wEAM part of the original EAM, we found an ectopic depression without feathering at the caudal area of the original EAM position (n=4; Fig. 4D), as described by Noden (1983). The depression formed in the region occupied by QCPN-positive mesenchymal cells (Fig. 4O). Furthermore, the ectopic depression exhibited the morphology of the quail wEAM, with a teardrop shape (n=2; Fig. 4D, Fig. S3). Since CNCCs can impose species-specific morphology on the epithelium (Schneider and Helms, 2003), these results indicate that the ectopic depression was formed by inductive signaling from the transplanted quail CNCCs. On the control side, feathering was not seen on the skin around the tEAM (Fig. 4F), but ectopic feathers surrounded the ectopic depression on the operated side, as in the original wEAM (Fig. 4D, Fig. S3). Importantly, the depression was located in the proximity of the duplicated otic process of the quadrate (Fig. 4G,M), suggesting that it is a duplicated structure corresponding to the wEAM. Taking these findings together, it is reasonable to think that the tEAM was lost and the wEAM was duplicated in the chimera. These results strongly suggest that specific genetic programs in PA1 and PA2 are involved in the wEAM and tEAM, respectively, and thus PA2 plays a crucial role in TM formation in the chick.

Evolution of the jaw and middle ear in mammals includes substantial morphological changes, which have attracted considerable research interest (e.g. Takechi and Kuratani, 2010). We propose that the key to understanding the evolution of the middle ear lies in developmental relationships between the TM and skeletal elements (Takechi and Kuratani, 2010; Kitazawa et al., 2015a). We have reported non-homology of the TM in mammals and diapsids (Kitazawa et al., 2015a). In addition, contrary to the classical view, we found that the mouse EAM begins to form at the surface ectoderm in the PA1 region (Minoux et al., 2013). These previous findings prompted us to perform a more detailed comparative analysis of TM formation, especially focusing on A-P patterning of the PAs in mouse and chicken embryos. Our present data suggest that a PA1-specific genetic program plays a crucial role in the formation of the TM in the mouse. By contrast, in the chick, a PA2-specific genetic program is necessary for TM formation, albeit with a certain contribution from PA1 to a portion of EAM formation. Furthermore, we propose that in the chick, the PA2 parts of the EAM and the columella auris are interdependent in TM development, as in the relationship between the EAM and tympanic ring in the mouse.

In this study we mapped the relative position of the TM with respect to the boundary between PA1 and PA2 territories using the Hoxa2^EGFP/+ knock-in mouse as a readout (Fig. 2). We found that the TM and tympanic ring are formed in the EGFP-negative PA1 territory (Fig. 2F, Fig. 5A) (Minoux et al., 2013), strongly suggesting that the TM is induced by a PA1-specific genetic program in the mouse.

Since, in the Hoxa2 knockout mouse, PA1 derivatives are duplicated in a mirror-image manner (Fig. 1D-F), patterning signals determining the morphological identity of PA1 derivatives might be provided by PP1, which is located between PA1 and PA2 (Rijli et al., 1993; Mallo, 2001). Based on the phenotype of the Hoxa2 knockout mouse, the axis of symmetry of the mirror-image duplication lies on the manubrium of the malleus (Fig. 1D,E, Fig. 2F). By contrast, the normal EAM-tympanic ring is formed towards a posterior region of the axis and lateral to the region of the malleus and incus in the wild-type embryo (Fig. 1A,B, Fig. 2F). Thus, it is reasonable to propose that, in the Hoxa2 knockout mouse, the developing original and duplicated EAM-tympanic rings might interfere with each other, resulting in failure to form the posterior aspect of the components and consequent deficiencies in their shape and size (Fig. 1D,E, Fig. 5B).

We found that the chick EAM begins to invaginate at the PA2 ectoderm and grows directly toward the columella auris in the PA2 region, and a portion of the EAM is also seen in the PA1 region (Fig. 3, Fig. 5D). The boundary between PA1 and PA2 appears to be consistent with the morphology of the EAM (Fig. 3). We thus refer to the two components as wEAM and tEAM, respectively. We then produced chick-quail chimeras in which PA2 is transformed into a PA1-like structure, as previously studied (Fig. 4) (Noden, 1983; Couly et al., 1998). The chimera lost the columella auris and lacked the TM (Fig. 4H). These results suggest that tEAM formation in the chick is largely dependent on a PA2-specific genetic program, and the formation processes of the tEAM and columella auris are interdependent during development, as in the relationship between the EAM and tympanic ring in the mouse (Fig. 5A,D). We also found that wEAM was duplicated in the chimeras, suggesting that wEAM formation is dependent on a PA1-specific genetic program (Fig. 5D,E).

Middle ear development is achieved by interaction between epithelia (PP1 and EAM) and mesenchyme (CNCCs) (Mallo, 2001, 2003). In the chick, the tEAM contacts the columella auris, which mostly derives from the PA2 CNCCs. Therefore, it is likely that interaction between the surface ectoderm and CNCCs in PA2 is important as a primary step of TM formation in the chick (Fig. 5D). Similarly, mesenchyme that generates the otic process of the quadrate appears to be related to wEAM formation (Fig. 5D,E). Since the surface ectoderm in the PA2 territory is not replaced by quail cells in the chimera (Fig. 4B), the mesenchyme that forms the otic process of the quadrate appears to be crucial for initiation of wEAM formation. It is worth noting that the duplicated EAM is a shallow depression, less developed than the original (Fig. 4D,N,O). This difference might be because growth of the wEAM also requires a developmental signal from PP1, and the duplicated wEAM was far from PP1 and unable to receive the signal (Fig. 4G,M, Fig. 5E).

In this study, we did not try to examine avian embryos that exhibit loss of PA1 identity, but rather duplication of PA2, because previous studies have shown that it is difficult to obtain such avian embryos by experimental manipulation. The chick-quail chimera in which the neural crest at the PA1 level (r1-2 or mesencephalon) is transposed to the PA2 level (r4-6) fails to transform PA1 into PA2 identity and results in formation of a normal jaw (Couly et al., 1998). This is because the Hox-negative CNCCs derived from rostral to the r1-2 area compensate for the lost area of the neural crest (Couly et al., 1998). When the broad area from the forebrain to hindbrain level is ablated in chick, and the quail r4-6 level is transplanted, the embryo exhibits loss of the jaw structure, but duplication of PA2 identity is not observed (Couly et al., 1998). Similarly, when a retroviral vector or active virus that expresses Hoxa2 ectopically is introduced to the chick, the jaw structure is lost but duplication of the columella auris does not occur (Grammatopoulos et al., 2000). Thus, further experimental modification might be needed to confirm whether the columella auris and tEAM are duplicated when PA2 is duplicated and PA1 is lost in avian embryos.

In contrast to the traditional view, our present data suggest that the EAM does not begin to form at the first pharyngeal cleft in either mouse or chick (Fig. 5A,D). The traditional view postulated that signals derived from the interface of PA1 and PA2 are involved in EAM formation, but our data suggest that such signals are not relevant to the beginning of EAM formation. However, although the EAM does not invaginate at the boundary between PA1 and PA2, in the chick the EAM straddles both PA1-derived and PA2-derived territories (Fig. 5D). Interestingly, the boundary between PA1 and PA2 may partition the chick EAM into its components, namely the wEAM and tEAM (Fig. 3). Since both PA1 and PA2 are involved in EAM formation in the chick (Fig. 4), the developmental mechanisms of EAM formation might utilize the interface of the PA1 and PA2 to generate morphological differences. These constitute differences to mouse external and middle ear development: only PA1 contributes to EAM and TM formation, whereas PA2 is involved in pinna formation in mouse (Fig. 5B,C) (Kitazawa et al., 2015b; Minoux et al., 2013; Rijli et al., 1993). It will be of interest to elucidate if other diapsid species also exhibit EAM-TM formation similar to the chick embryo.

We found that the TM is formed in PA1 in the mouse, but the entire TM is housed in the PA2 region in the chick (Figs 2 and 3). Therefore, regional contributions of PP1 to the TM may differ between mouse and chick; the contribution of PP1 in the PA2 territory is much larger in the chick than in the mouse (Fig. 5A,D). Although the prevailing view has been that PP1 is a comparable component for TM formation in mammals and diapsids, our observation suggests that different PP1 regions along the A-P axis contribute to the TM in these animals. In this regard, the epithelium of the middle ear cavity appears to originate from both endoderm and CNCCs in mammals, but not in diapsids (Thompson and Tucker, 2013). Future study elucidating the precise regional contribution of PP1 to the TM during mouse and chick development will be needed in order to provide further insight into the differences in TM formation in these lineages.

Loss of PA2 and duplication of PA1 results in duplication of the EAM in mouse and chick (Noden, 1983; Rijli et al., 1993), suggesting that PA1 is crucial for TM formation in both animals. However, we found that the original and duplicated EAMs form the TMs in the mouse (Fig. 5B) but not in the chick (Fig. 5E). Our findings suggest that TM formation in the mouse is highly dependent on a PA1-specific genetic program, whereas the chick TM is associated with a PA2-specific genetic program. Since the EAM finally reaches PP1 to form the TM, signals induced from PP1 to the surface ectoderm might be involved in TM formation. However, previous reports in mouse (Mallo et al., 2000; Mallo and Gridley, 1996; Yamada et al., 1995) and our present data in chick (Fig. 4) suggest that one of the main interactions for TM formation between epithelium and mesenchyme occurs between the cells that give rise to the EAM and tympanic ring (a PA1 derivative) in mammals, but the EAM and columella auris (a PA2 derivative) in diapsids. Therefore, it is likely that the EAM begins to form in proximity to the skeletal component in the PA1 region in mouse but in the PA2 region in chick. The tympanic ring and columella auris are a lower jaw and an upper jaw component, respectively. The idea of a close relationship between the EAM and these skeletal elements supports our previous report that the TM is formed as a lower jaw component by Edn1 signaling in mammals, but as an upper jaw component in diapsids (Kitazawa et al., 2015a).

The present study shows that although developmental interactions between epithelium and mesenchyme are crucial for TM formation in both mammals and diapsids, one of the main interactions appears to occur with different developmental origins in different PAs in these lineages. Further research should elucidate precisely how amniotes independently evolved functionally similar systems with different developmental origins and mechanisms.

Hoxa2^EGFP/EGFP, Hoxa2^EGFP/+ and R4::Cre;Rosa-CAG-LSL-tdTomato mice (Mus musculus) were described previously (Minoux et al., 2013; Pasqualetti et al., 2002). These transgenic and ICR mice were housed in an environmentally controlled room at 23±2°C, under a 12-h light:12-h dark cycle. Embryos were staged according to Theiler (1989). Fertilized eggs of the chicken Gallus gallus domesticus and quail Coturnix coturnix japonica were incubated at 38°C and staged according to Hamburger and Hamilton (1951). Animal care was entirely in accordance with the guidelines provided by the Tokyo Medical and Dental University, The University of Tokyo, and Friedrich Miescher Institute for Biomedical Research, and approval for the experiments was obtained from the institutions.

Embryos of wild-type and Hoxa2 knockout mice were fixed with Bouin's fixative overnight at room temperature (RT), dehydrated, penetrated by xylene, and embedded in paraffin. The samples were cut into 8 µm horizontal (transverse) sections. The sections were stained with Hematoxylin and Eosin.

Digoxigenin (DIG)-labeled riboprobes for mouse Acan and chick Hoxa2 were generated using the DIG RNA Labeling Kit (Roche) based on GenBank nucleotide sequences NM_007424 and NM_205150, respectively. Hoxa2^EGFP/+ mouse embryos were fixed with Carnoy's fixative overnight at RT. Chick E6 and E7 embryos were fixed with 4% paraformaldehyde (PFA) in PBS overnight (or for ∼2 days) at RT. The samples were then dehydrated, penetrated by xylene, and embedded in paraffin. Chick E8, E10 and E11 embryos were directly embedded in O.C.T. compound (Sakura Finetek) without fixation. These samples were sliced into 8-16 µm horizontal (transverse) or frontal (coronal) sections.

A Ventana Discovery XT system (Roche-Ventana Medical Systems) was used to perform in situ hybridization for mouse Acan. In situ hybridization for chick Hoxa2 was performed as previously described (Schaeren-Wiemers and Gerfin-Moser, 1993; Sockanathan, 2015) with slight modifications. Briefly, the sections were fixed in 4% PFA in PBS for 5-10 min followed by 10 min acetylation (21 μM HCl, 100 μM triethanolamine in water, 0.25% acetic anhydride) at RT. The sections were hybridized with hybridization buffer (50% formamide, 10 mM Tris HCl pH 7.6, 200 µg/ml E. coli tRNA, 1× Denhardt's solution, 5% dextran sulfate, 600 mM NaCl, 0.25% SDS, 1 mM EDTA) with riboprobe overnight at 70°C. The hybridized sections were incubated with anti-DIG antibody (Roche, 11093274910; 1/5000) overnight at 4°C. DIG was detected by nitro blue tetrazolium chloride (NBT)/5-bromo-4-chloro-3-indolyl phosphate (BCIP) (Roche). The sections were detected using a lightfield microscope (Olympus BX50) with a camera system (Olympus DP80).

Hoxa2^EGFP/+ and R4::Cre;Rosa-CAG-LSL-tdTomato mouse embryos and chick-quail chimera embryos were fixed with Carnoy's fixative overnight (or for ∼2 days) at RT, then dehydrated, penetrated by xylene, and embedded in paraffin. The samples were sliced into 8-10 µm horizontal or sagittal sections, which were then treated with 0.3% or 3% H₂O₂ in methanol for 20 min. For detection of EGFP or tdTomato, activation of antigen in sodium citrate buffer (1.8 μM citric acid monohydrate, 8.2 μM trisodium citrate dihydrate) at 97°C for 20-40 min was performed. The sections were incubated with anti-GFP antibody (ab6673, Abcam; 1:500), anti-RFP antibody (600-401-379, Rockland; 1/1000 or 1/2000) or QCPN antibody (Developmental Studies Hybridoma Bank; 1:10). The sections were then incubated with secondary antibody conjugated with biotin (Vector Laboratories; 1:500) for 30 min, followed by avidin-biotin complex (Vector Laboratories, Vectastain Elite ABC Kit; 1/100) for 30 min, and visualized using the Metal-Enhanced DAB Substrate Kit (Thermo Fisher Scientific). Counterstaining was performed with Hematoxylin. The sections were detected using the BX50-DP80 system.

Chick embryos at HH9 were prepared for in ovo surgery. The left side of r4 together with a portion of r3 and r5, which constitutes the PA2-forming level of the hindbrain crest (Kulesa and Fraser, 2000), was ablated from the chick with a tungsten needle. The stage-matching quail embryos were obtained as donors, and the left side of the neural crest corresponding to the ablated region of the chick or to MHB was isolated with a tungsten needle in sterile saline. The isolated graft was transplanted into the ablated region of the chick host (Fig. S2A, Fig. 4A). Chimeric embryos were incubated at 38°C and collected at E3-12. The unoperated right-hand side was used as a control in this chimera experiment. Imaging of the whole-mount samples was by stereoscopic microscope (Olympus SZX16) with DP80.

Consecutive histological sections after staining were recorded with the BX50-DP80 system. The images of the sections were reconstructed with Avizo v6.3 (FEI).

We thank Tatsuya Hirasawa and Sachiko Miyagawa-Tomita for helpful discussion; Norisuke Yokoyama, Jie Gu and Miyuri Kawasumi for helpful support; and Maryline Minoux for valuable comments on the manuscript.