HOXA5 plays tissue-specific roles in the developing respiratory system

In mammals, successful transition to air breathing at birth requires harmonious development of the respiratory system, a highly orchestrated process that involves the synchronized formation of trachea, lung and diaphragm. In the mouse, respiratory development initiates around embryonic day (E) 9 with the outpocketing and elongation of the foregut endoderm into the surrounding mesenchyme to form the laryngotracheal groove and primary lung buds (Morrisey and Hogan, 2010). Next, a coordinated program of dichotomous branching gives rise to the airway tree. Transition from branching morphogenesis to differentiation of the respiratory epithelium into specialized cell types, regionally distributed along the proximo-distal axis, leads to a patterned respiratory tract (Chang et al., 2013). Functional air–blood barriers form via differentiation of the distal epithelium in close association with the expansion of the vasculature. Lung development ends after birth with formation of alveoli, which increase the gas-exchange surface area to meet the respiratory requirements of the growing organism.

Physical forces and mechanical interactions between the thorax and respiratory tract are involved in lung growth and function. In fluid-filled fetal lungs, distention is maintained by breathing-like movements and by upper airway resistance during apnea (Kotecha, 2000). Genetic defects that perturb the thoracic skeleton, or space-occupying lesions that compress the lungs, like congenital diaphragmatic hernia, can cause pulmonary hypoplasia; this in turn compromises respiratory function and can be lethal (Jay et al., 2007). Lung development also requires innervation of the diaphragm by the phrenic nerves. Diaphragm denervation abolishes fetal breathing movements, which impairs lung growth (Wigglesworth and Desai, 1979; Liggins et al., 1981).

Diaphragm muscle formation involves contributions from several embryonic structures: somites, pleuroperitoneal folds (PPFs), septum transversum and neural tube (Merrell and Kardon, 2013; Merrell et al., 2015). Around E10.5, muscle progenitors delaminate from the hypaxial dermomyotome of the cervical (C) somites C3 to C5 to reach the PPFs, proliferate and spread ventrally and dorsally on the septum transversum to colonize the entire diaphragm, except the tendinous central region. Muscle progenitors then undergo two waves of myogenesis to differentiate into myofibers (Bentzinger et al., 2012). Meanwhile, the phrenic motor neurons exit the neural tube, also from the C3-C5 region, reach the PPFs and split into three primary trunks, the sternocostal, dorsocostal and crural branches, to innervate the entire diaphragm (Merrell and Kardon, 2013). Muscle progenitors and nerves reach the PPFs at the cervical level, and together descend toward the lower thorax as the lung and heart grow within the thoracic cavity (Allan and Greer, 1997). The PPFs, which constitute a transient embryonic structure, ultimately contribute to the diaphragm's central tendon and connective tissue fibroblasts (Merrell et al., 2015).

Respiratory tract morphogenesis relies on the functional integration of several transcriptional regulators and signaling pathways. Together, they direct reciprocal interactions between epithelium and mesenchyme, and control cell proliferation, differentiation and branching (Morrisey and Hogan, 2010; Hogan et al., 2014). Inappropriate expression of regulatory genes can cause malformations and diseases.

In vertebrates, Hox genes occupy a crucial position in the developmental hierarchy. They encode transcription factors that control the formation of body segment-specific structures via downstream effectors that, in turn, direct region-specific morphogenetic events in numerous tissue types along the embryonic axes. In mammals, 39 Hox genes are organized in four clusters and classified into 13 paralog groups. The spatiotemporal profile of Hox expression reflects the organization of clusters, the 3′ most genes being expressed earlier and in more anterior domains than the 5′ located ones (McGinnis and Krumlauf, 1992). Several Hox genes, predominantly from the 3′ half of clusters, are expressed along the respiratory tract, each with distinct proximal-distal distribution (Herriges et al., 2012). Previous work has shown the requirement for the three Hox5 paralog genes in lung development. They are expressed in a partially overlapping pattern: Hoxa5 and Hoxb5 are strongly expressed in lung mesenchyme, whereas Hoxc5 expression is barely detected (Boucherat et al., 2013; Hrycaj et al., 2015). Hoxa5 and Hoxc5, but not Hoxb5, are expressed in the phrenic motor neurons, which innervate the diaphragm (Philippidou et al., 2012). Finally, only Hoxa5 is expressed in tracheal mesenchyme. In agreement with its numerous expression domains in the developing respiratory system, most Hoxa5^−/− pups die at birth from respiratory failure, and show tracheal occlusion, lung hypoplasia and cell misspecification, and diaphragm innervation defects (Jeannotte et al., 1993; Aubin et al., 1997; Boucherat et al., 2013). In contrast, Hoxb5 and Hoxc5 single mutants are viable. The lung phenotypic traits of Hoxb5 mutants are less severe than for Hoxa5 mutants, whereas Hoxc5 mutants lack respiratory defects (Boucherat et al., 2013; P.P. and L.J., unpublished). These data unveil the functional predominance of Hoxa5 in formation of the respiratory system. They also show that Hoxa5 can compensate for the lack of Hox5 paralog function, as further supported by the increased Hoxa5 expression detected in Hoxb5^−/− lungs (Boucherat et al., 2013). Compound mutants for the three Hox5 genes display an exacerbated lung phenotype, likely reflecting functional redundancy due to expression overlap (Hrycaj et al., 2015).

The functions of Hoxa5 in trachea, lung and diaphragm may separately or in combination affect lung formation and neonate survival. To dissect the tissue-specific role(s) of Hoxa5 in respiratory system development, we used a Hoxa5 conditional allele with Cre recombinase deleter strains to ablate Hoxa5 in epithelium, mesenchyme, or motor neurons. Epithelial deletion revealed no role for Hoxa5 in this tissue, but deletion in either of the latter tissues resulted in hypoplastic lungs. Mesenchymal deletion of Hoxa5 also reproduced the abnormal patterning of tracheal cartilage and lung cell misspecification seen in Hoxa5 null mutants. However, only deletion of Hoxa5 in motor neurons caused death at birth owing to the abnormal innervation and formation of the diaphragm. Thus, the functional prevalence of Hoxa5 in the formation of the respiratory system results from distinct functions of Hoxa5 in its different expression sites.

We previously showed that Hoxa5 mRNA is expressed in the mesenchyme along the entire embryonic respiratory tract (Boucherat et al., 2013). Immunofluorescence assays confirmed HOXA5 protein expression in mesenchymal nuclei of upper airways and distal lung at E12.5 during branching morphogenesis, and also in the developing diaphragm (Fig. 1A,B). Further, descendants of Hoxa5-expressing cells showed restricted fates within the diaphragm. We bred a Hoxa5-Cre transgenic line with the R26^mT reporter mouse. R26^mT-positive cells, corresponding to descendants of Hoxa5-expressing cells, were detected in the PPF region of the diaphragm (Bérubé-Simard and Jeannotte, 2014) (Fig. 1C). In the trachea and primary bronchi, HOXA5 protein expression was observed in the ventrolateral mesenchyme (Fig. 1D). HOXA5 distribution in lung and trachea remained similar at later stages (Fig. 1E-F,H-K). As previously shown, HOXA5 immunoreactivity was detected in the prevertebra (pv)3-pv10 region encompassing the domain where diaphragm muscle progenitors originate and phrenic motor column axons exit and project to contact the diaphragm (Fig. 1E,E′) (Coulombe et al., 2010). Hoxa5-expressing cells were also present throughout the diaphragm at E14.5 (Fig. 1G-G″). At the end of gestation, Hoxa5 expression was reduced in the diaphragm, and descendants of Hoxa5-expressing cells were observed in the costal region and central tendon (Fig. 1L-M). No co-labeling between Hoxa5-expressing or descendant cells and cells positive for the myogenic transcriptional regulator PAX7 was detected in the developing diaphragm, indicating that Hoxa5 was not expressed in the diaphragm muscle lineage (Fig. S1). In trachea, HOXA5 expression was detected in cartilage condensations and the surrounding mesenchyme at E16.5 and E18.5. In lung, HOXA5 mesenchymal expression became sparser (Fig. 1H-K). HOXA5 protein was also detected in pulmonary endothelial cells (Fig. 1K). HOXA5 expression was not detected in and Hoxa5 descendants were absent from respiratory epithelium at all stages.

Although HOXA5 expression appeared to be absent from lung epithelium, it remains possible that expression is too low or restricted for a small cell population to be detected. To exclude a cell-autonomous function in lung epithelium, we ablated Hoxa5 in epithelium with the Shh^cre line (Harfe et al., 2004; Boucherat et al., 2014). Immunohistochemistry (IHC) confirmed that HOXA5 expression was not altered in trachea, lung and diaphragm of Hoxa5^flox/flox; Shh^+/Cre specimens (Fig. S2D-I). Further, no phenotype was apparent in these specimens. Offspring were obtained at expected Mendelian ratios at E18.5 and this ratio was maintained to adulthood (Fig. S3A). No lung hypoplasia was observed (Fig. S3B). Epithelial cell differentiation was normal, as assessed by the abundance and distribution of secretory club cells and type I pneumocytes, two cell types that are reduced in Hoxa5^−/− mutants (Fig. S3C-F). Other Hoxa5 null phenotypes, including goblet cell metaplasia and enlarged lung airspaces caused by perturbed myofibroblast localization, were also absent following Hoxa5 epithelial deletion (Fig. S3G-J) (Mandeville et al., 2006). Altogether, these data confirmed that Hoxa5 has no cell-autonomous role in respiratory epithelium.

To assess the requirement for Hoxa5 in respiratory mesenchyme and phrenic motor neurons, we used the Dermo1^+/cre and Olig2^+/cre deleter mouse lines, respectively (Yu et al., 2003; Dessaud et al., 2007; Dermo1 is also known as Twist2). Dermo1Cre activity is specific to lung and tracheal mesenchyme (Boucherat et al., 2014). However, the Dermo1^Cre allele was not effective in the entire developing diaphragm, demonstrated by the lack of reporter staining in the PPFs (Fig. S2A). Cre activity of the Olig2^+/cre line was restricted to the ventral spinal cord with no expression in skeletal muscle or lung (Fig. S2B,C) (Park et al., 2010a). Olig2Cre efficiently removes Hoxa5 from motor neurons (Philippidou et al., 2012). Here, we confirmed by IHC and quantitative RT-PCR (qRT-PCR) that it did not alter Hoxa5 expression in other respiratory tissues (Fig. S2J-L,X,Y). In Hoxa5^flox/flox; Dermo1^+/Cre embryos, HOXA5 expression was reduced but not abolished in trachea, lung and diaphragm mesenchyme, indicating that the Dermo1Cre was inefficient or not expressed in all mesenchymal cell types (Fig. S2M-O,X,Y). To reduce mesenchymal Hoxa5 expression further, we generated Hoxa5^flox/−; Dermo1^+/Cre animals. Near-complete loss of HOXA5 protein expression was observed in the trachea and lung from these embryos, but it was still detected in the diaphragm (Fig. S2P-Y).

Hoxa5^flox/flox; Dermo1^+/Cre and Hoxa5^flox/−; Dermo1^+/Cre individuals were obtained at expected Mendelian ratios at embryonic and adult ages, suggesting either that the mortality of Hoxa5^−/− pups did not result from the loss of Hoxa5 mesenchymal expression or that residual Hoxa5 expression was sufficient to allow survival (Table 1A,B). Conversely, 45% of the Hoxa5^flox/flox;Olig2^+/Cre newborns became cyanotic, developed respiratory distress and died within 24 h of birth, indicating that Hoxa5 deletion in motor neurons was sufficient to reproduce the mortality rate observed for Hoxa5^−/− pups (Table 1C).

Although Hoxa5 mesenchymal inactivation was incomplete, it recapitulated several aspects of the Hoxa5 null phenotype, indicating that a threshold of Hoxa5 expression is required in the mesenchyme. For example, defects in tracheal cartilage rings are fully penetrant in Hoxa5^−/− mice (Aubin et al., 1997). Tracheas from controls and Hoxa5^flox/flox; Olig2^+/Cre were normal (Fig. 2A,B,F,G). Conversely, all Hoxa5^flox/flox; Dermo1^+/Cre and Hoxa5^flox/−; Dermo1^+/Cre trachea specimens displayed irregular and incompletely formed cartilaginous rings with abnormal lateral fusion and merging of the cricoid cartilage with the first cartilage rings, anomalies that phenocopy those encountered in Hoxa5^−/− mice (Fig. 2C-E,H,I). Mesenchymal deletion of Hoxa5 shared additional phenotypes with Hoxa5^−/− embryos including ectopic cartilage dots in lungs (Fig. 2C,D), and reduced tracheal length, external diameter and luminal surface (Fig. 2J-L) (Boucherat et al., 2013). Thus, Hoxa5 activity in mesenchyme, but not in motor neurons, is essential for tracheal cartilage patterning.

Hoxa5^−/− embryos show disorganization of the tracheal epithelial layer at E18.5; however, no quantitative or qualitative variation in club, ciliated, basal and goblet cells was observed (Fig. S4) (Aubin et al., 1997). Smooth muscle cells of the trachealis muscle, derived from the dorsal mesenchyme, appeared normal as detected by αSMA (ACTA2) staining, consistent with lack of Hoxa5 expression in these cells (Fig. 1D, Fig. S4). Thus, Hoxa5 mesenchymal expression does not contribute to the dorso-ventral patterning of the trachea or to the epithelial-mesenchymal crosstalk that governs tracheal cell specification.

To characterize the mechanisms underlying the Hoxa5 tracheal phenotype, we examined expression of SOX9, a key transcriptional regulator of chondrogenesis (Turcatel et al., 2013). In the trachea, SOX9 is first uniformly expressed in the ventrolateral mesenchyme, and then expression becomes segmented in a pattern prefiguring the cartilage rings (Elluru and Whitsett, 2004). In lung, SOX9 expression is restricted to committed distal endodermal progenitor cells (Tian et al., 2011). In E14.5 Hoxa5^−/− embryos, mesenchymal SOX9 protein expression expanded caudally along the secondary bronchi (Fig. 3A,B). This ectopic SOX9 expression might explain the abnormal cartilage dots in mutant lungs. At E18.5, SOX9 tracheal expression was limited to Alcian Blue-positive cartilage rings in controls and Hoxa5^flox/flox; Olig2^+/Cre specimens. In Hoxa5^−/− and Hoxa5^flox/flox; Dermo1^+/Cre embryos, ectopic SOX9 expression was observed outside of the cartilaginous nodules (Fig. 3C-F′). Expression of cartilage-specific type II collagen, a direct transcriptional target of SOX9, was reduced in Hoxa5^−/− and Hoxa5^flox/flox; Dermo1^+/Cre tracheas (Fig. 3G-J) (Lefebvre et al., 1997). qRT-PCR analysis confirmed the decrease in Col2a1 mRNA levels, which was more substantial in Hoxa5^−/− specimens probably owing to the residual Hoxa5 expression in trachea of Hoxa5^flox/flox; Dermo1^+/Cre mutants (Fig. 3K). qRT-PCR also revealed a significant diminution of Sox9 expression levels, despite the spatial expansion observed. Consistent with this, expression of Sox5 and Sox6, which are genetically downstream of Sox9 in cartilage specification and constitute with Sox9 the chondrogenic regulator trio, was also decreased (Fig. 3K) (Lefebvre et al., 1998).

To characterize further the molecular mechanisms leading to Sox9 dysregulation and cartilage defects, we next examined other known regulators of tracheal cartilage patterning: Fgf10, Shh, Bmp4, Tbx4 and Tbx5. Of these, all but Fgf10 influence Sox9 expression in tracheal mesenchyme (Arora et al., 2012; Park et al., 2010b; Sala et al., 2011). qRT-PCR showed significantly decreased expression of Bmp4 in the trachea from Hoxa5^−/− and Hoxa5^flox/flox; Dermo1^+/Cre mutants, and similarly displayed reduction of Id1 and Id2 expression, which are targets of the BMP4 pathway (Fig. S5). Expression of the follistatin-like 1 (Fstl1) gene was also diminished in Hoxa5 mutants. Fstl1, which encodes a BMP antagonist, is a direct transcriptional target of HOXA5 in the trachea (Boucherat et al., 2013). No changes in Fgf10, Shh, Tbx4 and Tbx5 expression levels were detected (Fig. S5). These data suggest that Hoxa5 controls both spatial domain and expression levels of Sox9 in tracheal mesenchyme via mechanisms that involve BMP signaling. This regulation is cell-autonomous, because conditional deletion of Hoxa5 in mesenchyme recapitulates the null phenotype. Further, the incomplete deletion by Dermo1Cre indicates that a threshold level of Hoxa5 is required for its full activity in tracheal mesenchyme.

Lung hypoplasia, measured as a reduced lung weight/body weight (LW/BW) ratio, occurred in Hoxa5^flox/flox; Dermo1^+/Cre, Hoxa5^flox/−; Dermo1^+/Cre and Hoxa5^flox/flox; Olig2^+/Cre mutants and in Hoxa5^−/− embryos (Fig. 4A). Decreased proliferation contributes to Hoxa5^−/− lung hypoplasia (Boucherat et al., 2013). A similar downward trend was observed in the number of 5-bromo-2′-deoxyuridine (BrdU)-positive cells in Hoxa5^flox/flox; Dermo1^+/Cre, Hoxa5^flox/−; Dermo1^+/Cre and Hoxa5^flox/flox; Olig2^+/Cre lungs (Fig. S6). Hoxa5 is thus essential in both lung mesenchyme and phrenic motor neurons innervating the diaphragm to sustain correct proliferation of lung tissues.

Hoxa5^−/− embryos display abnormal and dense lung histology with airway epithelium differentiation defects (Boucherat et al., 2012, 2013). At E18.5, the most affected Hoxa5^flox/flox; Dermo1^+/Cre and Hoxa5^flox/−; Dermo1^+/Cre embryos similarly exhibited narrower air spaces and thicker mesenchyme (Fig. 4B-D). Control and Hoxa5^flox/flox; Olig2^+/Cre lung specimens presented a normal structure with dilated peripheral saccules and thin mesenchyme (Fig. 4B,E). At postnatal day (P) 0, thicker lung mesenchyme persisted in Hoxa5^flox/flox; Dermo1^+/Cre and Hoxa5^flox/−; Dermo1^+/Cre specimens but pulmonary structures were expanded (Fig. 4F-H). In contrast, lungs from Hoxa5^flox/flox; Olig2^+/Cre newborns found dead were collapsed compared with controls, indicating that neuronal Hoxa5 expression was essential for the expansion of the lung at birth (Fig. 4F,I).

Hoxa5 null mutation causes lung epithelial cell differentiation defects, including fewer club cells and type I pneumocytes (Boucherat et al., 2012, 2013). Both cell types were reduced in Hoxa5^flox/flox; Dermo1^+/Cre and Hoxa5^flox/−; Dermo1^+/Cre lungs, but not in control or Hoxa5^flox/flox; Olig2^+/Cre specimens (Fig. 4J-Q). Goblet cell metaplasia was also observed in Hoxa5^flox/flox; Dermo1^+/Cre and Hoxa5^flox/−; Dermo1^+/Cre adult mice whereas mucus-producing cells were scarce in controls and Hoxa5^flox/flox; Olig2^+/Cre mutants (Fig. 4R-U). Variable expressivity of the lung phenotype in specimens carrying the Dermo1Cre allele likely reflects the incomplete activity of the Dermo1Cre. Overall, the requirement for Hoxa5 in lung epithelial organization and differentiation appears to be mediated by Hoxa5 activity in mesenchyme, and does not require expression in motor neurons. This further illustrates that Hoxa5 possesses tissue-specific roles that differentially participate in respiratory tract morphogenesis.

Deletion of all three Hox5 paralogs in the lung was reported to affect the canonical Wnt/Bmp4 signaling axis but not the Fgf and Shh pathways (Hrycaj et al., 2015). To assess the specific contribution of Hoxa5 in this context, we evaluated gene expression in Hoxa5^−/− lungs by qRT-PCR. At E12.5, expression of Wnt2 and Wnt7a was reduced in Hoxa5^−/− specimens (Fig. S7A). These are ligands of the canonical Wnt pathway expressed in lung mesenchyme and epithelium, respectively (Miller et al., 2012; Hrycaj et al., 2015). Wnt/β-catenin signaling activates the airway smooth muscle program, and acts upstream of Bmp4 and Fgf signaling to regulate differentiation of airway epithelium (Shu et al., 2005; Goss et al., 2011). In Hoxa5^−/− lungs, expression of downstream targets of the canonical Wnt pathway was significantly decreased, including Lef1, a transcriptional effector of the pathway, Sm22a (Tagln), an early marker of the smooth muscle lineage, and Pdgfra, Fgf10, Bmp4 and Shh, key players in airway smooth muscle development and branching morphogenesis (Fig. S7A) (Morrisey and Hogan, 2010). At E18.5, expression of Wnt2, Wnt7a, Sm22a, Pdgfra and Fgf10 remained significantly diminished in Hoxa5^−/− specimens (Fig. S7B). Analysis was also performed on E18.5 Hoxb5^−/− and compound Hoxa5^−/−; Hoxb5^−/− mutant embryos. Similar reductions in Wnt2 and Fgf10 expression were observed in Hoxa5 and Hoxb5 single mutants and in compound mutants. Wnt7a expression was not affected in Hoxb5^−/− specimens but was decreased in Hoxa5^−/−; Hoxb5^−/− mutants suggesting a predominant contribution of Hoxa5 to Wnt7a expression. In contrast, Shh and Lef1 expression was significantly diminished in Hoxb5^−/− and Hoxa5^−/−; Hoxb5^−/− mutants. As in Hoxa5^−/− lungs, Bmp4 expression was unaffected in Hoxb5^−/− and Hoxa5^−/−; Hoxb5^−/− mutants. Finally, Pdgfra, Sm22a and Acta2 (encoding αSMA) expression was reduced in Hoxa5^−/− and Hoxa5^−/−; Hoxb5^−/− but not in Hoxb5^−/− mutant lungs. Thus, Hoxa5 and Hoxb5 genes contribute differentially to fine-tune the signaling networks controlling lung development. They act similarly on the Fgf10 pathway, their influence differs on Wnt/β-catenin and Shh signaling, and Hoxa5 specifically controls the lung airway smooth muscle program via regulation of Pdgfra.

Almost half of the Hoxa5^flox/flox; Olig2^+/cre pups died at birth from respiratory distress with compact lungs and a significantly reduced LW/BW ratio, suggesting that lungs failed to grow and inflate normally as a result of impaired fetal breathing movements caused by abnormal diaphragm function. To confirm that this phenotype was associated with abnormal innervation of diaphragm, we characterized phrenic nerves of Hoxa5^flox/flox; Olig2^+/Cre embryos. The phrenic nerve arises bilaterally from motor neurons in the ventral horns of the cervical spinal cord and contacts the developing diaphragm near the mid-costal region before branching into three primary trunks, of which two extend to innervate the ventral and dorsal parts of diaphragm (Merrell and Kardon, 2013). Whole-mount staining of nerves and neuromuscular junctions revealed that, at E18.5, Hoxa5^−/− and Hoxa5^flox/flox; Olig2^+/Cre embryos presented a severely compromised motor innervation pattern. Axons consistently failed to reach the ventral-most part of the costal muscle, and fewer branches were observed in the dorsal region. Diaphragm innervation was normal in controls and Hoxa5^flox/flox; Dermo1^+/Cre specimens (Fig. 5A-D). The phenotype was similar at E14.5 suggesting that altered axon patterning, rather than neuronal degeneration, was the cause of the defective innervation (Fig. 5E-G).

Axons typically regulate diaphragm muscle development from the time of initial contact, by facilitating myotube formation via electrically mediated effects and/or diffusible substances. We therefore analyzed the structure of diaphragm muscle fibers in E18.5 embryos by electron microscopy (Greer et al., 1999). Controls and Hoxa5^flox/flox; Dermo1^+/Cre diaphragms showed aligned sarcomeres. Conversely, Hoxa5^−/− and Hoxa5^flox/flox; Olig2^+/Cre mutant specimens presented misaligned Z-discs, damaged sarcomeric proteins and much wider actin-only I-bands, indicative of a high proportion of non-contracted sarcomeres. These defects were observed in the entire muscularized diaphragm (Fig. 6). Further, at E18.5, skeletal muscle was significantly thinner in Hoxa5^−/− and Hoxa5^flox/flox; Olig2^+/Cre specimens compared with controls and Hoxa5^flox/flox; Dermo1^+/Cre embryos (Fig. 7A-E). Both ventral and dorsal regions were equally affected, consistent with electron microscopy data. We also examined muscle fiber size by immunostaining for α-laminin, an extracellular matrix protein of basal lamina, which allows delineation of myofibers. Muscle fiber number per surface area was significantly augmented in Hoxa5^−/− and Hoxa5^flox/flox; Olig2^+/Cre diaphragms, reflecting a decrease in myofiber size (Fig. 7F). Moreover, Hoxa5^−/− and Hoxa5^flox/flox; Olig2^+/Cre diaphragms showed a decreased number of BrdU-positive nuclei compared with controls and Hoxa5^flox/flox; Dermo1^+/Cre specimens, indicating that decreased proliferation contributed to muscle atrophy (Fig. 7G).

Muscle atrophy is associated with increased RUNX1 expression, a transcription factor induced in denervated muscle that helps to prevent myofibrillar disorganization and muscle wasting (Zhu et al., 1994; Wang et al., 2005). In Hoxa5^−/− and Hoxa5^flox/flox; Olig2^+/Cre diaphragms, Runx1 expression was significantly increased in response to the incorrect innervation of the diaphragm (Fig. 7H). Thus, even if phrenic nerves reached the dorsal region of the diaphragm in these mutants, neuromuscular contacts appeared to be insufficient to properly support muscular development and growth, causing muscle atrophy. The decreased mechanical forces of the atrophic myofibers in turn likely caused a reduction in fetal breathing-like movements, contributing to lung hypoplasia and resulting in neonatal mortality.

The defective sarcomeres, reduced diaphragm thickness and smaller cross-sectional area of the myofibers in Hoxa5^−/− and Hoxa5^flox/flox; Olig2^+/Cre embryos raised the question of whether aberrant muscle differentiation contributes to the Hoxa5 phenotype. We thus analyzed expression of myogenic transcriptional regulators in these two mutants. Somitic myogenic precursors express Pax3 and Pax7 and the myogenic regulatory factor (MRF) Myf5 (Sambasivan and Tajbakhsh, 2007; Bentzinger et al., 2012). Expression of these three genes was significantly reduced in E18.5 Hoxa5^−/− diaphragms, suggesting defective expansion, migration and/or maintenance of the resident myogenic progenitor population (Fig. 7I). Surprisingly, although Pax3 and Myf5 are genetically upstream of the MRF Myod (Myod1), its expression was significantly augmented in Hoxa5^−/− specimens (Tajbakhsh et al., 1997). Abnormal expression of both Myod and Myf5 in Hoxa5^−/− specimens might reflect aberrant progenitor fate decisions. In contrast, expression of the MRFs myogenin (Myog) and Mrf4 (Myf6), regulators of myocyte fusion and myotube formation, was unaffected (Fig. 7I). A negative-feedback loop exists between Myod and Igf2 in the diaphragm muscle, and double deletion of these genes causes diaphragm atrophy (Borensztein et al., 2013). Consistently, E18.5 Hoxa5^−/− diaphragms also showed decreased Igf2 expression (Fig. 7I). Notably, none of these changes in myogenic regulatory expression was detected in Hoxa5^flox/flox; Olig2^+/Cre specimens, suggesting an additional role for Hoxa5 outside of motor neurons in diaphragm myogenesis.

Both Hoxa5^flox/flox; Olig2^+/Cre and Hoxa5^−/− diaphragms showed a high proportion of non-contracted sarcomeres. We thus assessed myofiber type composition in these animals. During mouse embryonic diaphragm formation, two populations of fibers are produced in the first wave of myogenesis expressing either Myh3 and Myh7 or Myh3 and Myh8 myosin heavy chain isoforms, and giving rise to slow-twitch and fast-twitch muscle fibers, respectively. In the second wave, the developing myofibers only express Myh3 and Myh8 (Schiaffino and Reggiani, 2011). Myh7, Myh3 and Myh8 expression was significantly decreased in both Hoxa5^flox/flox; Olig2^+/Cre and Hoxa5^−/− diaphragms (Fig. 7J). In contrast, expression of the adult heavy chain isoforms Myh1 and Myh2, which give rise to fast-twitch muscle fibers, presented an upward trend in these specimens. Together, these results suggested that innervation defects resulting from Hoxa5 deletion in phrenic motor neurons promoted the incorporation of fast-twitch fibers in diaphragm.

Hoxa5 is important for the formation of several organs and tissues, consistent with its tissue-specific expression in embryos and adults (Jeannotte et al., 2016). Hoxa5 is crucial for respiratory system development, and the null mutation disrupts formation of the trachea, lung and diaphragm causing respiratory failure and neonatal death (Aubin et al., 1997; Boucherat et al., 2013). HOXA5 expression is restricted to the mesenchymal layer of the respiratory tract and to the phrenic motor neurons, which are the sole source of diaphragm innervation. The lack of phenotype in Hoxa5^flox/flox; Shh^+/Cre specimens coupled to immunofluorescence data confirmed that Hoxa5 is not expressed and has no cell-autonomous function in the respiratory epithelium.

Although Hoxa5 is not required in lung epithelium, derivatives of both epithelium and mesenchyme are affected by the Hoxa5 mutation. Indeed, most lung phenotypes are epithelial, indicating that Hoxa5 acts primarily via mesenchymal-epithelial inductive signaling in this context. Hox5 compound mutants showed that Hox5 genes mainly control the canonical Wnt/Bmp4 signaling axis (Hrycaj et al., 2015). Here, analysis of Hoxa5 single mutants establishes the specific contribution of Hoxa5. Hoxa5 positively regulates the canonical Wnt pathway, and also controls expression of Fgf10, Shh and Bmp4, indicating a broad role in signal integration during lung morphogenesis. Hoxa5 action is also time dependent, as its influence on Wnt, Shh and Bmp4 lung expression decreases in older embryos. We previously reported that canonical Wnt activity was increased in adult Hoxa5^−/− lungs, revealing a difference in how Hoxa5 controls the Wnt pathway in lung development compared with maintenance (Boucherat et al., 2012). Decreased Wnt activity in the developing lung might underlie the reduced proliferation and resulting lung hypoplasia in Hoxa5 mutants. Conversely, augmented Wnt activity in adults could explain the increased cell proliferation in lungs from Hoxa5 surviving mutants, allowing recovery of lung growth (Mandeville et al., 2006).

Hoxa5 positively regulates Fgf10 lung expression throughout embryogenesis. Hoxb5 also positively controls Fgf10 consistent with reduced lung branching in Hoxb5 mutants (Bellusci et al., 1997; Boucherat et al., 2013). Consequently, Fgf10 expression was also reduced in Hoxa5; Hoxb5 double mutants. The previously reported lack of Fgf10 expression change in E14.5 Hox5 triple mutants is therefore surprising (Hrycaj et al., 2015). One possibility could be negative regulation by the barely expressed Hoxc5 gene. Alternatively, variation in temporal Fgf10 expression might exist in Hox5 triple mutants. Finally, the presence of residual trachea tissue in lung samples from the triple mutants could explain the difference, as Hoxa5 does not regulate Fgf10 expression in upper airways.

Hoxa5 is the predominant Hox5 paralog to act in the lung airway smooth muscle program, where it regulates Pdgfra expression. Alveolar myofibroblasts are interstitial contractile cells responsible for elastin deposition and alveolar formation, two processes affected in Hoxa5^−/− mutants (Mandeville et al., 2006). They are derived from embryonic lung mesenchyme, which expresses Pdgfra and is co-labeled with HOXA5 protein (Lindahl et al., 1997; Hamilton et al., 2003; Mandeville et al., 2006). This suggests a cell-autonomous function for Hoxa5 in alveolar myofibroblast progenitors (Fig. 8).

Hoxa5 is also involved in lipofibroblast cell fate, as revealed by the reduced expression of peroxisome proliferator-activated receptor gamma (Pparg), a master regulator of adipogenesis, and adipose differentiation-related protein (Adrp; Plin2), a trafficking protein in lipofibroblasts (Fig. S7C, Fig. 8). Both genes are downstream of Fgf10 signaling during lipofibroblast formation (Al Alam et al., 2015). Moreover, the decreased expression of Tbx4, an early marker of lung mesenchymal cells, indicates that Hoxa5 may have a broad impact on lung mesenchymal cell fate, in addition to its instructive role on lung epithelium (Fig. S7C, Fig. 8) (Arora et al., 2012).

Tracheal cartilage malformations are fully penetrant in Hoxa5; Dermo1^+/Cre mice, as in Hoxa5^−/−, indicating that, as with lung development, it is the mesenchymal expression of Hoxa5 that regulates tracheal development. Incomplete Hoxa5 deletion by the Dermo1Cre, also establishes the requirement for a minimal level of Hoxa5 expression. Hoxb5 and Hoxc5 cannot rescue the Hoxa5 tracheal phenotype as they are not expressed in the developing upper airways (Boucherat et al., 2013). Therefore, Hoxa5 is the only Hox5 paralog gene involved in the tracheal developmental program.

In Hoxa5 mutants, tracheal cartilage rings form, thus mesenchymal cells become committed towards the chondrogenic lineage. SOX9 and HOXA5 are co-expressed in ventrolateral tracheal mesenchyme. Later, SOX9 becomes restricted to the condensing cartilage rings, whereas HOXA5 is present in ring cartilage and perichondrium and in surrounding mesenchyme (Fig. 1) (Arora et al., 2012). Hoxa5 mutant tracheas show both altered spatial expression of SOX9, and reduced Sox9 expression levels. Together, this suggests that Hoxa5 acts cell-autonomously to control mesenchymal Sox9 expression, which in turn permits the correct chondrogenesis of cartilage rings. This is consistent with the action of Hoxa5 on Sox9 expression during acromion formation and in chick somites, suggesting that it could be a conserved regulatory axis to direct cartilage development and morphology (Aubin et al., 2002; Chen et al., 2013).

Expression of Bmp4, and its target genes Id1 and Id2, was decreased in Hoxa5 mutant tracheas. BMP4 is a pro-chondrogenic signal that promotes chondrogenesis through Sox9 activation (Park et al., 2010b). Reduced BMP4 signaling might thus underlie the observed Sox9 downregulation. Moreover, Hoxa5 is co-expressed with Bmp4, making it a potential direct HOXA5 target (Li et al., 2008). Interestingly, expression of the BMP antagonist Fstl1 is also reduced in Hoxa5 mutant specimens, and Fstl1^−/− mice present a malformed trachea with fewer cartilage rings and a disorganized epithelium reminiscent of the Hoxa5^−/− tracheal phenotype (Geng et al., 2011). We previously showed that Fstl1 is a direct HOXA5 transcriptional target (Boucherat et al., 2013). The decreased expression of both Bmp4 and Fstl1 following the loss-of-Hoxa5 function appears to be contradictory. However, it underscores the complexity of the regulatory network in tracheal-bronchial chondrogenesis, which involves extensive cross-regulation between signaling pathways and transcriptional regulators by mechanisms that remain to be fully elucidated (Fig. 8).

Hoxa5^flox/flox; Dermo1^+/Cre and Hoxa5^flox/−; Dermo1^+/Cre adult mice were recovered at the expected Mendelian ratio indicating that tracheal malformations and compact lung architecture are not the cause of lethality of Hoxa5^−/− newborns. In contrast, Hoxa5 deletion in phrenic motor neurons reproduced this lethality, demonstrating the crucial importance of Hoxa5 in diaphragm development and function. Only Hoxa5 mesenchymal expression appeared to be necessary for lung epithelial cell differentiation. However, both Hoxa5 mesenchymal and motor neuron mutations caused lung hypoplasia due to decreased proliferation. Thus, Hoxa5 has tissue-specific functions in mesenchyme and motor neurons that have both converging and distinct effects on respiratory development.

Hoxa5 expression in phrenic motor neurons is essential for diaphragm formation. Hoxa5^flox/flox; Olig2^+/Cre and Hoxa5^−/− diaphragms were incorrectly innervated and showed abnormal musculature that evokes diaphragmatic eventration, a condition involving diaphragm relaxation that allows superior displacement of abdominal viscera into the thoracic cavity and impedes lung growth (Ackerman et al., 2012). Abnormal innervation was detected at E14.5, after the division of the phrenic nerve into the three primary branches. The failure of axons to reach the ventral part of the costal muscle, and the absence of recovery during diaphragm development, suggested that the defective innervation originates from axon guidance alterations rather than axonal degeneration and loss of synaptic contacts (Philippidou et al., 2012).

Profound alterations of the muscular component of diaphragm were also observed in Hoxa5^−/− and Hoxa5^flox/flox; Olig2^+/Cre specimens, including thinner diaphragm, smaller myofibers and misaligned sarcomeres with damaged proteins, all characteristics of muscle atrophy (Wang et al., 2005; Borensztein et al., 2013). Increased Runx1 expression, a marker for damaged and regenerating muscle, further supported the notion of muscle atrophy and attempted recovery in Hoxa5 mutants (Umansky et al., 2015). Hoxa5^−/− and Hoxa5^flox/flox; Olig2^+/Cre adult diaphragms presented an irregular distribution of costal muscles restricted to the ventral region (K.L.-T. and L.J., unpublished). This suggests that axon projections present in the dorsal diaphragm, although less abundant, were sufficient to generate synaptic contacts with muscle fibers and allow some muscle rescue.

Diaphragm muscles originate from myogenic precursor cells in the hypaxial dermomyotome of cervical somites C3 to C5, which express the myogenic transcriptional regulators Pax3, Pax7 and Myf5 (Bentzinger et al., 2012). Myf5^−/− mice do not present diaphragm defects but Myf5; Myod double mutants, in which Mrf4 expression is compromised, lack diaphragm (Braun et al., 1992; Rudnicki et al., 1993; Kassar-Duchossoy et al., 2004). Splotch (Pax3^−/−) mutants also lack diaphragm muscle, and that of Pax7^−/− mice is notably thinner (Tremblay et al., 1998; Seale et al., 2000). Therefore, decreased expression of Pax3, Pax7 and Myf5 in Hoxa5^−/− diaphragms might reflect a reduced contribution of myogenic progenitors to the diaphragm musculature.

Interestingly, Hoxa5^flox/flox; Olig2^+/Cre diaphragms do not show variation in Pax3, Pax7 or Myf5 expression. Thus, despite morphological similarity between both phenotypes, the diaphragm phenotype of Hoxa5^−/− mutants might reflect Hoxa5 activity in cell types other than motor neurons. This could include the diaphragm connective tissue or somite-derived muscles. Hoxa5 is expressed in the somites from the cervical-thoracic transition domain C3 to T2. However, Hoxa5 somitic expression is confined to the lateral sclerotome, a source of cartilage and perichondrium, and no expression was detected in the dermomyotome of chicks or myotome of mice (Chen et al., 2013; J.H.M., unpublished). This suggests that the decreased expression of Pax3, Pax7 and Myf5 in Hoxa5^−/− diaphragms is unlikely to result from a cell-autonomous action of Hoxa5 in myotome. However, we cannot exclude an indirect role for Hoxa5 somitic expression in the selective proliferation, migration and/or survival of myogenic precursors (Fig. 8).

Hoxa5 is also expressed in the developing diaphragm. However, no phenotype was detected in Hoxa5^flox/flox; Dermo1^+/Cre mutants, possibly owing to the incomplete activity of the Dermo1^Cre allele in diaphragm. Cell lineage and co-labeling data indicated that Hoxa5 is not present in muscle cells and is likely expressed in the PPF lineage (Fig. 1, Fig. S1). Thus, other Cre deleter mouse lines will have to be tested to determine the role of Hoxa5 expression in the diaphragm.

Myod expression is positively regulated by both Pax3 and Myf5 (Tajbakhsh et al., 1997). However, Myod expression was increased in diaphragms from Hoxa5^−/− mutants. One possible explanation is the decreased expression of Igf2 also observed in Hoxa5^−/− diaphragms. Myod and Igf2 show reciprocal negative regulation in this context: Myod overexpression compensates for Igf2 downregulation and Igf2 overexpression compensates for Myod downregulation (Borensztein et al., 2013). The cause of reduced Igf2 expression in the diaphragm of Hoxa5^−/− mutants remains unknown.

Myosin heavy chain fast isoform expression is augmented in Hoxa5^−/− and Hoxa5^flox/flox; Olig2^+/Cre diaphragms. The fiber composition of a given muscle results from several factors including innervation, mechanical conditions and usage patterns. Moreover, when breathing rate increases, recruitment of fast muscle fibers is favored (Polla et al., 2004). Surviving Hoxa5^−/− mice present respiratory adaptations including higher breathing frequency and augmented overall minute ventilation to circumvent morphological respiratory system defects (Kinkead et al., 2004). Together, our data suggest that they are compensatory mechanisms to maintain and adapt diaphragm muscle activity in order to allow the survival of Hoxa5 mutants.

In conclusion, characterization of Hoxa5 conditional mutants led us to establish that Hoxa5 regulates respiratory system development via several distinct and tissue-specific functions. These independent expression domains and functions are likely to reflect the stepwise origin and elaboration of respiratory structures in terrestrial vertebrates. For instance, the diaphragm and phrenic motor neurons that innervate it are an evolutionary novelty of mammals, and Hoxa5 was probably independently recruited for its various functions in the respiratory system as these structures arose (Jung and Dasen, 2015).

The Hoxa5 null (Hoxa5^tm1Rob), Hoxa5^flox/flox (Hoxa5^tm1Ljea), 14.5 kb Hoxa5-cre [Tg(Hoxa5-cre)447BLjea] and Olig2^Cre [Olig2^tm1(cre)Tmj] mouse lines were previously described (Jeannotte et al., 1993; Dessaud et al., 2007; Tabariès et al., 2007; Bérubé-Simard and Jeannotte, 2014). The Dermo1^Cre [Twist2^tm1(cre)Dor] line was obtained from Dr Ornitz (Washington University, St. Louis, MO, USA; Yu et al., 2003). The R26^LacZ, R26^mTmG and R26^mT reporter lines [Gt(ROSA)26Sor^tm1Sor, Gt(ROSA)26Sor^{tm4(ACTB−tdTomato,−EGFP)Luo} and Gt(Rosa)26Sor^{tm9(CAG−tdTomato)Hze})] and the Shh^Cre Washington University deleter strain [Shh^{tm1(EGFP/cre)Cjt}] were purchased (Jackson Laboratory; Soriano, 1999; Harfe et al., 2004; Muzumdar et al., 2007; Madisen et al., 2010). As only individuals carrying the Hoxa5^−/−, Hoxa5^flox/flox; Dermo1^+/Cre, Hoxa5^flox/−; Dermo1^+/Cre and Hoxa5^flox/flox; Olig2^+/Cre genotypes presented defects, all other genotypes were referred to as controls except when specified. Mouse lines were maintained in the 129/Sv background. Age of embryos was estimated by considering the morning of the day of the vaginal plug as E0.5. Experimental specimens were genotyped by PCR and Southern blot analyses. Trachea, lung and diaphragm from control and mutant embryos were collected at E12.5, E14.5, E18.5, P0 and P30. The wet lung and body weights were obtained by direct measurement. For RNA extraction, organs were snap-frozen in N₂. Experiments were performed according to the guidelines of Canadian Council on Animal Care and approved by the Institutional Animal Care Committee.

Specimens were processed for paraffin (4 µm) or frozen (5-10 µm) sections. Experiments were performed as described (Boucherat et al., 2017). Slides were counterstained with Nuclear Fast Red, Alcian Blue or Hematoxylin. For IF, nuclei were visualized by DAPI staining. The Cyanine 3 Tyramide Signal Amplification Kit (Perkin Elmer) was used at 1/50 dilution for HOXA5 IF detection. Antibodies are listed in Table S1.

Detection of β-galactosidase activity in frozen sections (10 µm) from E12.5 embryos was performed as described (Bérubé-Simard and Jeannotte, 2014).

Pregnant females were injected intraperitoneally with 100 µg BrdU/g body weight and embryos were collected 1 h later. Proliferation index corresponds to the number of BrdU-immunoreactive cells divided by total cell number for each section analyzed. Four to eight random fields were taken for an average number of 200 diaphragm and 750 lung cells per field, from three to five embryos per genotype.

Diaphragms from E18.5 embryos were divided into dorsal-left, dorsal-right, ventral-left and ventral-right regions, fixed in 0.1 M sodium cacodylate, pH7.3, containing 2.5% glutaraldehyde for 24 h, washed in sodium cacodylate buffer, post-fixed in 1% OsO4, dehydrated in ethanol and then propylene oxide, embedded in Epon, and sectioned at 60-80 nm. Sections were stained in 0.4% lead citrate for 8 min and in 3% uranyl acetate solution for 5 min. Pictures were taken with a transmission electron microscope JEOL, JEM1230 (Tokyo, Japan) used at 80 kV and equipped with two CCD cameras: (1) wide-field Gatan Dual Vision and (2) high-resolution Gatan Ultrascan 1000XP.

Dissected respiratory tracts from E18.5 embryos were stained overnight at 37°C in 0.015% Alcian Blue, 20% acetic acid prepared in 95% ethanol. Specimens were processed 30 min in 1% KOH for better visualization of cartilage rings.

ImageJ software was used to calculate (1) the tracheal length between the cricoid cartilage and the carina, (2) the trachea external diameter along the most linear portion of trachea, and (3) the number of diaphragm muscle fibers/area (μm²) at eight dorso-ventral locations in the right costal muscle. Using the Leica SCN 400 F SlideScanner and SlidePath Gateway Software, we measured the tracheal luminal surface, and the diaphragm thickness at eight dorso-ventral locations in the right costal muscle. Three to eleven specimens were used for each genotype tested.

Total RNA was isolated from trachea, lung and diaphragm of individuals. qRT-PCR experiments were performed as described (Boucherat et al., 2012). Three to eight specimens were used for each genotype tested. Primer sequences are listed in Table S2.

Student's t-test was performed for comparative studies. A significance level below 5% (P<0.05) was considered statistically significant. Values are expressed as mean±s.e.m. *P<0.05, **P<0.01, ***P<0.001.

We thank Drs Jean Charron and Luisa Dandolo for comments, Dr David Ornitz for Dermo1^Cre mice, Drs Normand Marceau and Gurmukh Singh for antibodies, and Richard Janvier for electron microscopy analysis.