Msx1 deficiency interacts with hypoxia and induces a morphogenetic regulation during mouse lip development

Clefts of the lip and palatal structures are common developmental defects in humans and are part of the phenotypic spectrum in various Mendelian syndromes (Tolarová and Cervenka, 1998). However, most cleft lips and/or cleft palates (summarized as orofacial clefting) are nonsyndromic malformations and have a multifactorial aetiology, including strong contributions by interacting genetic components and environmental factors (Mossey et al., 2009). The incidence of orofacial clefting varies among different ethnic groups, gender and socioeconomic status and occurs in 1 out of 700 births on average (Dixon et al., 2011). Based on the results of genetic and embryological studies, nonsyndromic orofacial clefts are traditionally classified as isolated, nonsyndromic cleft palate (equivalent to nonsyndromic cleft palate only, nsCPO) and nonsyndromic cleft lip that occurs with (nsCLP) or without (nsCLO) a cleft palate, the latter two collectively referred to as nsCL/P. The multifactorial aetiology and concealed Mendelian inheritance patterns underlying nsCPO and nsCL/P complicate the identification of genetic loci (Dixon et al., 2011; Setó-Salvia and Stanier, 2014); however, recent genome-wide association studies (GWAS) and meta-analyses identified at least 40 genetic risk loci for these defects (Birnbaum et al., 2009; Grant et al., 2009; Mangold et al., 2010; Beaty et al., 2010; Ludwig et al., 2012, 2016, 2017; Sun et al., 2015; Leslie et al., 2016; Yu et al., 2017; Butali et al., 2019). The studies provide an essential framework for disentangling the complex processes regulating lip and palate development (Mangold et al., 2011). However, the identity of the vast majority of causal variants and critical gene–gene (GG) and gene–environment (GE) interactions underlying the multifactorial aetiology of orofacial clefting remain to be determined.

The formation of the lip requires continuous fusion and merging of craniofacial processes (or prominences); that is, the medial nasal process (Mnp) and the maxillary process (Mxp), which both initially develop independently of each other (Gritli-Linde, 2012). A cleft of the lip occurs when this process is impaired, and a cleft may extend further involving the nose when merging is also disrupted between the Mnp and the lateral nasal process (Lnp). A crucially important step in lip formation is the timely regulated disintegration of the epithelium at the sites of fusion, a prerequisite for intermingling of neural crest cell-derived mesenchymal cells to establish stable junctions. Epithelial disintegration involves programmed cell death and occurs at the so-called lambdoidal junction, the region at which the tips of Mnp, Mxp and Lnp make contact and fuse with each other. Crucially important developmental pathways that regulate this tightly controlled process have been identified, showing that a Pbx1/2-controlled Wnt signalling pathway regulates p63/Irf6-mediated cell death in the epithelial seam at the lambdoidal junction (Ferretti et al., 2011; Kousa and Schutte, 2016). Interestingly, expanded Shh signalling negatively affects the Wnt signalling and p63/Irf6 pathways at the lambdoidal junction and results in cleft lip formation (Kurosaka et al., 2014). Expanded Shh signalling causes ectopic canonical Wnt inhibitor activity, which may involve displaced Vax1 expression (Kurosaka et al., 2014). Although the functional roles of Vax1 in lip and palate development remain to be investigated (Geoghegan et al., 2017), several reports of variants in the human VAX1 gene highlight its involvement in human CL/P (Slavotinek et al., 2012; Butali et al., 2019).

The MSX1 gene encodes a homeodomain-containing transcription factor, and MSX1 haploinsufficiency in humans causes familial oligodontia (Vastardis et al., 1996). CL/P is occasionally seen in affected patients (van den Boogaard et al., 2000) and a contribution of MSX1 mutations to nsCL/P and nsCPO was first identified by a candidate-gene linkage-disequilibrium strategy (Lidral et al., 1998). A number of studies have subsequently supported a role for MSX1 in nonsyndromic orofacial clefting in a broad ethnic range (Fallin et al., 2003; Schultz et al., 2004; Suzuki et al., 2004; Vieira et al., 2005; Tongkobpetch et al., 2006; Otero et al., 2007; Jagomägi et al., 2010; Gowans et al., 2016). One study suggested that about 2% of nsCL/P cases might involve mutations of the MSX1 gene in some populations (Jezewski et al., 2003). A genome-wide significant association between genetic variants at the MSX1 locus and the sub-phenotype nsCLP has recently been identified in the Chinese population (Yu et al., 2017). Interactions between DNA variants in MSX1 with maternal smoking and with alcohol consumption have been suggested to increase the risk for orofacial clefting (Romitti et al., 1999; Beaty et al., 2002; Fallin et al., 2003; van den Boogaard et al., 2008), although this association was not found in a cohort of Danish origin (Mitchell et al., 2001). Likewise, it has been demonstrated that a link between maternal smoking and MSX1 variants increased the risk for developmental limb malformations (Hwang et al., 1998), but this could not be confirmed by another study (Carmichael et al., 2004).

Inactivation of Msx1 in mice results in the complete absence of teeth and cleft secondary palate (Satokata and Maas, 1994). Msx1 and its close homologue Msx2 are co-expressed in various craniofacial regions derived from neural crest cells, a multipotent migratory cell population that originates from the dorsal neural fold and generates a major source of the skeletal and connective tissue of the developing face (Trainor, 2005; Chai and Maxson, 2006; Suzuki et al., 2016). Although a cleft lip does not form in the absence of Msx1, the Msx1^−/−;Msx2^−/− double homozygous mouse mutants exhibit severe craniofacial defects, including exencephaly and a particularly dysmorphic midface (Satokata et al., 2000; Ishii et al., 2005). In addition, defects in cardiac outflow tract formation and a severe limb phenotype in Msx1^−/−;Msx2^−/− mutants are not seen in either single mutant, showing that the functions of Msx1 and Msx2 can compensate for each other at various sites during development (Ishii et al., 2005; Lallemand et al., 2005). Cleft lip formation also occurs in Msx1^−/−;Pax9^−/− double homozygous mouse mutants (Nakatomi et al., 2010); however, this genetic interaction has not been investigated and the impact of combined Msx1 and Pax9 deficiency on lip development is currently unknown.

Multiple lines of evidence have shown that episodes of embryonic hypoxia during the first trimester of human pregnancy are associated with developmental defects and craniofacial malformations, including cleft lip and palate (reviewed by Webster and Abela, 2007). More recently, an increasing number of medically relevant substances have been shown to cause embryonic hypoxia through a significant disruption of the embryonic heart beat rate (Danielsson et al., 2003, 2005a,b, 2007). Although these adverse effects are linked to cleft lip and palate formation, their interactions with genetic predispositions in orofacial clefting are largely unknown.

Embryonic hypoxia has been implicated as an environmental risk factor for orofacial clefting (Millicovsky and Johnston, 1981a; Bronsky et al., 1986; Nagaoka et al., 2012; Smith et al., 2013). Among other causes, embryonic hypoxia might be a consequence of constriction of uterine blood vessels or of bradycardia/arrhythmia of the embryonic heart, conditions that have been widely described as side effects of several commonly prescribed drugs (Webster and Abela, 2007). To investigate a possible interaction between Msx1 deficiency and hypoxia, we mated heterozygous Msx1 mutant mice and exposed pregnant females to 10% O₂ during the period of early lip formation [i.e. from embryonic day (E) 10.5 to E12.5]. Gross inspection of embryos at E15.5 revealed that 72% (13/18) of Msx1 homozygous mutant embryos (Msx1^−/−) developed a bilateral or unilateral cleft lip, whereas none of the Msx1^+/− embryos (0/39; P=1.54⁻⁰⁹) or Msx1^+/+ embryos (0/8; P=6.75⁻⁰⁴) showed this defect (Fig. 1A,B).

Embryonic hypoxia in the mesenchyme of the branchial arches has been identified as a side effect of phenytoin, an anticonvulsant drug with known teratogenic effects in both humans and mice, in particular for cleft lip and cleft palate formation (Azarbayjani and Danielsson, 2001; Danielsson et al., 2005a; Webster et al., 2006). Thus, to model pharmacologically induced hypoxic stress during lip development, we exposed pregnant Msx1^+/− females to different concentrations of phenytoin from E10.5 to E11.5. Notably, the exposure strongly increased the incidence of cleft lip formation in Msx1^−/− embryos in a dose-dependent manner and reached 91.7% at the highest dosage (Fig. 1C), at which all affected mutants exhibited a bilateral cleft lip. The defects involved severe growth retardation of the developing Mnp and Mxp at E11.5 (Fig. 1D). Moreover, affected Msx1^−/− embryos exposed to phenytoin at the mid-range concentration exhibited a spectrum of upper lip deformities typically seen in human patients, including hypoplastic lip, unilateral cleft lip and bilateral cleft lip (Fig. 1E-H; Table S1). At the highest dosage, the frequency of cleft lip was also increased in heterozygous Msx1 mutants (9.2%; P<0.027; Table S1), a condition reminiscent of cleft lip formation in human patients with heterozygous mutations in MSX1. Consistent with redundant roles of Msx1 and Msx2 in craniofacial development (Ishii et al., 2005), the sensitivity to phenytoin-induced cleft lip was significantly increased in Msx1/Msx2 compound mutants (Table S1).

To confirm the hypoxia-inducing effects of phenytoin in the developing lip, we carried out immunohistochemical staining of pimonidazol (Hypoxyprobe-1), which forms covalent adducts in hypoxic tissues, including mouse embryonic tissues (Arteel et al., 1995; Danielsson et al., 2003). The staining revealed a strongly increased intensity of hypoxia in the Mnp at E11.0 in both phenytoin-treated wild-type and Msx1-deficient embryos, whereas weak staining was observed in untreated samples (Fig. S1). In contrast, upregulated expression of Hif1, a key factor mediating vital cellular responses to embryonic hypoxia (Dunwoodie, 2009), and of the Hif1 target Vegf, was less markedly upregulated in Msx1^−/− embryos (Fig. S1). To test whether it is possible to offset the adverse effects of phenytoin during lip formation (Millicovsky and Johnston, 1981a,b), we kept pregnant females in a hyperoxic environment (50% O₂) for the duration of phenytoin treatment (48 h). In this experimental group, only 7 out of 22 (31.8%) of the lip fusion sites were affected in Msx1^−/− mutant embryos, whereas 8 out of 14 (57.1%) fusion sites showed a cleft at normoxia. Although the calculated P-value (P=0.13) did not reach statistical significance, the data suggests that simultaneous exposure to elevated oxygen levels can mitigate the adverse effects of phenytoin on lip development in the absence of Msx1.

In the outbred genetic background (CD1) of the mice used in this work, the interaction caused a phenotypic spectrum ranging from clefts of the soft palate only to complete cleft secondary palate (Fig. 1I-K). The interaction affected the morphogenesis of the palatal shelves (Fig. 1L) and the incidence was highly reproducible in heterozygous Msx1 mutant embryos at higher dosages (P<0.004 at 60 mg/kg and P<0.0007 at 85 mg/kg) and reached 100% when Msx1 heterozygosity was combined with the absence of Msx2 (Fig. 1M, Table S1). Importantly, Msx1 heterozygosity of the pregnant mother did not influence the frequency of cleft palate (Fig. S2), indicating that the genotype of the embryo is the key genetic determinant for the observed GE interactions. A GE interaction between phenytoin and Msx1 heterozygosity that caused a cleft secondary palate was recently also observed in an inbred (C57BL/6) genetic background (M.N., unpublished).

Msx1 and Pax9 are co-expressed in neural crest cell-derived dental mesenchyme (Mackenzie et al., 1991; Neubüser et al., 1995) and both genes are essential for tooth formation (Satokata and Maas, 1994; Chen et al., 1996; Peters et al., 1998). Moreover, Msx1 and Pax9 cooperatively regulate downstream target genes in the tooth mesenchyme, and preliminary observations showed that about 40% of Msx1;Pax9 double homozygous mouse mutants exhibited a cleft lip (Ogawa et al., 2006; Nakatomi et al., 2010). Based on these observations we speculated that the anticonvulsant drug phenytoin might interact with Pax9 deficiency in cleft lip formation in a similar manner to that revealed for Msx1. Surprisingly, however, the incidence of cleft lip in Pax9 homozygous mutant embryos was not significantly increased at any dosage tested and there was no significant increase in phenytoin-induced cleft palate in heterozygous Pax9 mutant embryos (Fig. 1C; Table S1).

The strong interaction of phenytoin with Msx1, but not with Pax9, suggests that Msx1 and Pax9 could regulate different aspects during lip formation. To understand this, we first compared their expression patterns in the developing Mnp, Lnp and Mxp. Whole-mount in situ hybridization on mouse embryonic heads at E10.5 revealed that Msx1 and Msx2 were expressed in all three processes, whereas expression of Pax9 was restricted to the Mnp and Lnp (Fig. 2A,C,E). In situ hybridization on sections showed that the expression domains of Msx1 and Msx2 were restricted to the anterior-distal compartment of the Mnp, whereas Pax9 was expressed in the posterior-proximal region of the Mnp (Fig. 2B,D,F). In situ hybridization of MSX1 and immunohistochemical analyses of PAX9 expression in human embryonic tissue revealed that both genes were expressed during human Mnp formation in patterns similar to those observed during mouse Mnp development (Fig. 2G-M). In addition, we found that Msx2 and Pax1, the paralogous genes of Msx1 and Pax9, respectively, were not upregulated in Msx1 and Pax9 single homozygous mutants (Fig. S3).

Next, we compared Mnp morphology of Msx1 and Pax9 single homozygous mutants at E11.5. At this developmental stage, the distal part of the Mnp is normally oriented towards the Lnp and Mxp (Fig. 3A), a transient and characteristic feature during mammalian upper lip formation that is also seen in primates, including humans (Senders et al., 2003). In Msx1^−/− embryos, we consistently observed growth retardation of the distal compartment of the Mnp, but the contact between Mnp and Mxp was always well established in these mutants (Fig. 3A). Importantly, this condition was associated with a considerably pronounced curvature of the Mnp (Fig. 3A,B). In contrast, the Mnp curvature was attenuated in Pax9^−/− mutants and we did not observe significant growth retardation of the distal Mnp region in the absence of Pax9 alone (Fig. 3A,B).

It was noticeable that growth retardation of the Mnp caused by Msx1-deficiency was consistently associated with a significant increase in Mnp curvature. Interestingly, although the distal compartment of the Mnp was also smaller in Msx1^−/−;Pax9^−/− mutants, the Mnp curvature was significantly flattened, resulting in an irregularly long nasal pit (Fig. 3A,B). Moreover, the area of contact between Mnp and Mxp was noticeably smaller in the double homozygous mutants (Fig. 3C), resulting in a wide gap between left and right Mxp in embryos affected by cleft lip (Fig. 3D). These findings indicate that insufficient contact between Mnp and Mxp in Msx1^−/−;Pax9^−/− mutants resulting in a cleft lip is a consequence of combining two Mnp defects that occur independently of each other: growth retardation caused by Msx1 deficiency and reduced curvature caused by the absence of Pax9.

Because Pax9 deficiency leads to a decrease in Mnp curvature, we hypothesized that the establishment of a Pax9-dependent and pronounced Mnp curvature in Msx1^−/− mutant embryos could involve upregulation of Pax9 expression. In normal Mnp development, Pax9 was initially expressed distally but gradually became restricted to more proximal regions between E10.5 and E11.5 (Fig. 4A-C; Fig. S4). Interestingly, Pax9 expression was upregulated and distally expanded in the Mnp of Msx1^−/− and Msx1/Msx2 compound mutant embryos between E10.5 and E11.5 (Fig. 4D-F,L; Fig. S5). In contrast, Msx1 expression levels were not altered in the Mnp of Pax9^−/− mutants (Fig. 4G,H). Together, these results show that the gradual restriction of Pax9 expression to a proximal domain in the Mnp is Msx1-dependent, revealing that Msx1 acts genetically upstream of Pax9 in the developing Mnp.

The complementary expression patterns of Msx1 and Pax9 in the wild-type Mnp (Fig. 2) indicate that Msx1-dependent restriction of Pax9 expression is regulated by a non-cell autonomous mechanism. A potential candidate for mediating this process is Bmp4, a secreted growth factor that functions in a positive feedback loop with Msx1 and Msx2 at various sites during craniofacial development (Zhang et al., 2002) and has been shown to suppress mesenchymal Pax9 expression in early tooth and palate development (Neubüser et al., 1997; Liu et al., 2005). In the developing Mnp, Bmp4 expression shifts from the epithelium to the mesenchyme (Gong and Guo, 2003), where it overlaps with that of Msx1 and Msx2 and in which the expression pattern is complementary to that of Pax9 at E11.0 (Fig. 4I,J). Consistent with redundant roles of Msx1 and Msx2 in activation of mesenchymal Bmp4 transcription, expression of Bmp4 in the Mnp mesenchyme was strongly reduced in Msx1;Msx2 compound mutants (Fig. 4K), whereas that of Pax9 was correspondingly expanded (Fig. 4L). In the absence of Msx1 alone, Bmp4 expression was mostly affected in a medial domain within the posterior Mnp region (Fig. 4N) and expression of Pax9 was expanded particularly into this domain (Fig. 4P). Importantly, Msx2 expression was barely detectable in the medial domain within the posterior Mnp in both wild-type and Msx1^−/− mutants (Fig. 4Q,R), suggesting that the strong reduction in Bmp4 expression results from a near-complete absence of Msx1/Msx2-regulated expression of Bmp4 in this region. To verify a downstream effect of reduced Bmp4 signalling, we analysed the distribution of phosphorylated Smad proteins (pSmad1/5/8; He et al., 2010). Quantification of pSmad1/5/8-positive nuclei revealed significantly reduced expression in the Mnp of Msx1^−/− embryos, with 76.95% of the nuclei stained in controls and 36.73% stained in Msx1^−/− embryos (P=0.0007; Fig. 4S-V; Fig. S6, Table S2). These results indicate that the Bmp4 downstream pathway is significantly affected in the medial area of the Mnp and supports the view that Msx2 does not fully compensate for the absence of Msx1 in this region.

To directly demonstrate an inhibitory effect of Bmp4 on Pax9 expression, we implanted Bmp4 protein-soaked beads into the mesenchyme of cultured nasal processes dissected at E10.5. Following organ culture, in situ hybridization revealed a strong downregulation of Pax9 in both Mnp and Lnp (Fig. 5A). This finding demonstrates that Bmp4 is sufficient for inhibiting Pax9 expression in the Mnp mesenchyme and thus provides a plausible explanation for the expansion of Pax9 expression seen in Msx1^−/− and Msx1^−/−;Msx2^−/− mutant embryos.

In addition to the distal expansion of Pax9 expression seen in the Mnp of Msx1^−/− and Msx1^−/−;Msx2^−/− mutant embryos, we also observed an increase in Pax9 promoter activity in the absence of a functional Pax9 gene (Fig. 5B). This was assessed by using a combination of the nonfunctional Pax9 alleles Pax9^lacZ (Peters et al., 1998) and Pax9^del (Kist et al., 2007). The finding indicates the presence of a negative, auto-regulatory Pax9 feedback loop that contributes to restricting Pax9 expression to proximal regions of the Mnp. Although downregulated Bmp4 expression probably accounts for the expanded Pax9 expression in Msx1^−/−;Msx2^−/− mutant embryos, we did not detect altered Bmp4 expression in the Pax9-deficient Mnp at E10.5 and E11.5 (Fig. 5C). In addition, the expression of Shh and its receptor Ptch1 was not changed (M.N., unpublished). However, using a candidate gene approach we found that gremlin1 (Grem1), a secreted Bmp4 antagonist (Hsu et al., 1998), was ectopically expressed in the Mnp of Pax9^−/− mutants (Fig. 5D). This suggests a mechanism by which upregulated Grem1 expression neutralizes Bmp4-mediated inhibition of Pax9 expression in the MNP mesenchyme of Pax9^−/− mutants.

Collectively, the data suggest a model that integrates Msx1 and Pax9 functions into two parallel acting pathways that are connected through a relay mechanism involving Bmp4 (Fig. 5E). Whereas Msx1 is primarily required for Mnp growth, a loss of Msx1 function also causes decreased Bmp4 expression, which lowers Bmp4-mediated repression of Pax9. Our data suggest that the resulting increase in Pax9 activity induces morphogenetic regulation and adjustment of Mnp shape that facilitates contact with the Mxp.

To investigate whether genetic data from GWAS in human nsCL/P samples support our model of morphogenetic regulation derived from the experimental animal work (Fig. 5E), we turned to large-cohort data from human patients with nsCL/P and its subtypes nsCLP and nsCLO. We retrieved GWAS data from a previously published meta-analysis (Ludwig et al., 2017) and performed both single-gene and pathway analyses. First, we tested for association in aggregates of common variants in the genes investigated in this study, and each of the three traits as listed above. A set-based test as implemented in VEGAS (Liu et al., 2010) was applied to imputed data of both a European (Euro) and a multiethnic (All) meta-analysis. Consistent with previous results (Ludwig et al., 2016), GREM1 reached genome-wide significance for nsCL/P and nsCLP (Table 1). We also observed a test-wide significant enrichment for MSX1 in nsCL/P, and additional nominally significant results for all remaining genes and nsCL/P, except for BMP4. Although the significant associations of MSX1 and MSX2 were reflected by contributions from both nsCLP and nsCLO, the significance in nsCL/P for PAX9 was primarily driven by nsCLP only (with non-significant results for nsCLO; Table 1).

Prompted by these results we investigated whether GWAS association signals are overrepresented in aggregate in Gene Ontology (GO) pathways that include MSX1, PAX9, BMP4 and GREM1. Using the gene-based P-values generated above, we carried out individual VEGAS pathway analyses for nsCL/P, nsCLP and nsCLO. In the analysis of nsCLP, a genome-wide significant result was identified for ‘organ morphogenesis’ (GO:0009887), with P=1×10⁻⁰⁶ after 10⁶ permutations. Repeating the analysis with 10⁷ permutations, revealed P=2×10⁻⁰⁷. When performed for nsCLO, the pathway-based result for GO:0009887 was not significant (P>0.5). The morphogenesis pathway GO:0009887 also includes MSX2, and a total of 557 genes directly matched results of the nsCLP-specific VEGAS pathway analysis (Table S3); however, 75 genes identified by the VEGAS pathway analysis did not produce a match in GO:0009887 (Table S4). Interestingly, a substantial proportion of these genes (185 out of 632; 29.3%) have previously been reported to play a role in human or mouse orofacial clefting, or for facial development (Tables S3,S4). Moreover, GO:0009887 was recently also identified through a database search approach for genes involved in jaw morphogenesis and disorders (Manocha et al., 2019).

We also investigated whether the morphogenesis pathway GO:0009887 plays a role in normal facial variation. Based on a list of 112 genes located at genetic loci previously identified through GWAS to be associated with diverse facial traits (Liu et al., 2012; Paternoster et al., 2012; Adhikari et al., 2016; Cole et al., 2016; Pickrell et al., 2016; Shaffer et al., 2016; Cha et al., 2018; Claes et al., 2018; Indencleef et al., 2018; Xiong et al., 2019), we identified 37 genes that were among the 632 genes included in GO:0009887 (33%, Table S5). Statistical assessment of whether the overlap between both gene sets was significant revealed a highly significant P-value (P<10⁻¹⁶, hypergeometric test), suggesting that the morphogenesis pathway also contributes to variation in facial traits in the general population.

Our data show that transient exposure of pregnant mice to reduced respiratory oxygen levels during a crucial time-period of lip development is sufficient to cause cleft lip in Msx1^−/− mouse embryos at an incidence of about 72% (0% cleft lip in Msx1^−/− embryos at normoxia), whereas none of the wild-type embryos showed this defect. In addition, more than 50% of Msx1^−/− embryos developed a cleft lip when the pregnant mouse was injected with 60 mg/kg phenytoin (90% cleft lip at 85 mg/kg). In contrast, the incidence of cleft lip was extremely low in wild-type embryos (0% at 60 mg/kg phenytoin and 2% at 85 mg/kg phenytoin), consistent with previous results showing that CD1 mice are particularly resistant to the teratogenic effects of phenytoin (Hansen and Hodes, 1983; Azarbayjani et al., 2006). Together, the data reveals a strong, and specific, GE interaction involving deficiency of Msx1 and transient hypoxia, which could also be relevant in the aetiology of orofacial clefting in humans.

Interestingly, low oxygen partial pressure is present at high altitude, which has been suggested to underlie an increased risk for cleft lip formation in Bolivia (Castilla et al., 1999) and might be a factor contributing to the increased risk for orofacial clefting in certain regions in China (Fan et al., 2018). Other conditions linked to orofacial clefting in humans might induce embryonic hypoxia during pregnancy at normoxia, including constriction of uterine vessels, placental insufficiency and life style factors such as smoking (Webster and Abela, 2007). Moreover, various pharmacologically relevant substances used in the treatment of epilepsy have been suggested to cause hypoxia directly in the developing embryo (Danielsson et al., 2007). The anticonvulsant drug phenytoin is an example of this group and our results clearly demonstrate that Msx1 deficiency in mouse embryos greatly enhances susceptibility to the cleft lip-inducing effects of phenytoin. A high sensitivity to reduced Msx1 gene dosage is also illustrated in Msx1^+/− mutants with different genetic backgrounds, which form a cleft secondary palate at a high frequency when exposed to phenytoin (Fig. 1; M.N., unpublished). Previous work showed that combined Msx1/Msx2 deficiency is associated with impaired maturation of the endothelial layer of blood vessels, a defect secondary to the abnormal development of neural crest-derived vascular smooth muscle cells (Lopes et al., 2011). Similarly, Msx1^−/− embryos show transiently impaired blood vessel formation during facial development (Medio et al., 2012). Our data showing reduced upregulation of the hypoxia markers Hif1 and Vegf in Msx1-deficient Mnp exposed to phenytoin are consistent with these findings.

Additional research is warranted to investigate the molecular basis for the interaction between Msx1 deficiency and phenytoin-induced hypoxia. Phenytoin is considered to cause irregular phases of embryonic hypoxia, which are followed by episodes of re-oxygenation and the formation of reactive oxygen species (Danielsson et al., 2007). Msx1 could be involved in oxidative stress response; however, the strong interaction between Msx1 deficiency and continuous exposure to reduced maternal oxygen levels argues that additional mechanism should also be considered.

Phenytoin was shown to inhibit a voltage-dependent K⁺ channel (I_kr) that is expressed in the developing rodent heart. I_kr is encoded by an ether-a-go-go-related gene (Kcnh2, also known as Erg) and is particularly sensitive during the period of heart development that coincides with crucial stages of lip and palate formation (Danielsson et al., 2005a,b). Interestingly, I_kr blocking activities have been demonstrated for several anti-arrhythmics, but also for some antihistamines, antibiotics and antipsychotic drugs (Tamargo et al., 2004). With regards to orofacial clefting, epidemiological analyses have not yet produced conclusive data on the roles of these drugs. However, because these analyses were carried out on samples collected from a genetically highly heterogenous population, it is conceivable that inconclusive results and low risk estimates might mask significantly increased risks in those patients that are genetically predisposed. Nevertheless, our results revealed that mutations in Msx1 strongly interact with hypoxic stress during mouse lip development and the findings might therefore be useful for an improved risk assessment in patient groups affected by heterozygous mutations in MSX1: these patients are regularly affected by oligodontia (Fournier et al., 2018), a phenotype that is readily diagnosed and thus could be helpful to aid the development of targeted prevention strategies (e.g. during the early period of pregnancy).

Identifying causative GE and GG interactions that underlie nonsyndromic forms of orofacial clefting in humans is a complex challenge. Although mutations in MSX1 are frequently associated with nsCL/P, these birth defects are incompletely penetrant and phenotypes are highly variable across individuals within affected families (Gu et al., 2018). Previous work has shown that absence of Msx1 and Msx2 in mouse embryos results in severely growth-retarded craniofacial processes (Ishii et al., 2005), defects that clearly predispose to cleft palate and cleft lip. Here we show that a reduction in Mnp size is associated with a moderate but significant alteration of Mnp shape (Fig. 2), a finding that has not been described before. The enhanced curvature of the Mnp appears to facilitate contact with the Mxp to complete lip formation. We propose that the capacity of the Mnp to react to growth retardation by inducing a favourable alteration of Mnp shape represents a ‘morphogenetic regulation’. Compared with the midfacial development of other mammals, the evolutionary shortening of the maxilla and depressed midfacial anatomy typically seen in primates, in particular in the hominid clade (Lacruz et al., 2019), requires that the growth of their midfacial area is naturally suppressed during development. It is therefore tempting to speculate that a morphogenetic regulation that compensates for a genetically caused growth defect might be crucially important, specifically in humans. Our data suggest that morphogenetic regulation could be a significant component of the ‘developmental robustness’ of lip formation in humans; its disruption may therefore be considered a novel pathogenic mechanism in the aetiology of cleft lip.

The morphogenetic regulation that enhances Mnp curvature in Msx1^−/− embryos is Pax9 dependent and provides a mechanistic explanation for the negative epistasis between Msx1 and Pax9 in lip development. The lack of a distinct Mnp curvature in Pax9-deficient embryos is reminiscent of morphogenetic disruption of the secondary palatal shelves in these mutants. Similar to the Mnp, the secondary palatal shelves are not growth retarded but show a distinct absence of indentations on the oral side (Peters et al., 1998; Zhou et al., 2013; Jia et al., 2017). Both Mnp and palatal shelves of Pax9^−/− mutant embryos appear bulky and uniformly shaped, defects that clearly reveal an important role for Pax9 in the morphogenesis of both structures during a crucial period of development. With respect to lip formation, we show that the expression domain of Pax9 in the Mnp is controlled by Msx1 and involves Msx1-regulated expression of Bmp4. Interestingly, the shift of Bmp4 expression from the epithelium to the mesenchyme (Gong and Guo, 2003) correlates with the gradual restriction of Pax9 expression to the proximal region of the developing Mnp by Bmp4. Together, the interactions involving Msx1, Pax9, Bmp4 and Grem1 might contribute to a molecular network that continuously coordinates the growth and shape of the Mnp during a crucial period of lip development (Fig. 5E). Although the functions of Grem1 in Mnp morphogenesis remain to be tested in vivo using appropriate genetic modelling, important roles for GREM1 in human nose and lip development have been indicated by a strong association between polymorphisms at the GREM1 locus and nose width (Boehringer et al., 2011), and by genome-wide significant associations between variants at the GREM1 locus and nsCLP (Ludwig et al., 2016) and other forms of orofacial clefting (Gowans et al., 2018; Wang et al., 2018; Rafighdoost et al., 2019).

The mechanism by which Pax9 controls morphogenesis of the Mnp remains to be elucidated. Pax9 might be involved in regulating differential growth directly within different areas in the Mnp. The function of Pax9 might also affect Mnp development by regulating growth of the primary palate, which could contribute to lateral displacement of the distal regions of the Mnp and is affected in Pax9^−/− embryos (Jia et al., 2017; Li et al., 2017). Moreover, the irregular and extended nasal pit in Pax9^−/− mutant embryos might indicate that Pax9 is also required for regulating efficient epithelial fusion between Mnp and Lnp at the lambdoid junction. Interestingly, Wnt signalling has been shown to play a crucial role in this process (Ferretti et al., 2011; reviewed by Reynolds et al., 2019), and inhibition of secreted antagonists of Wnt signalling are sufficient to rescue cleft palate in Pax9-deficient mouse embryos (Jia et al., 2017; Li et al., 2017).

GWAS have established significant associations of regions around MSX1, PAX9 and GREM1 in the aetiology of orofacial clefts in humans of different ethnic origins (Yu et al., 2017; Huang et al., 2019; Ludwig et al., 2016; Table 1). Although genome-wide significance for loci around BMP4 and MSX2 in human orofacial clefting remains to be demonstrated, analyses of mouse mutant embryos have shown that, except for Grem1, all five genes regulate the development of both secondary palate and lip (Satokata and Maas, 1994; Liu et al., 2005; Suzuki et al., 2009; Nakatomi et al., 2010; Parada and Chai, 2012; Ludwig et al., 2016; this work). Interestingly, our pathway analysis identified a genome-wide significant overrepresentation of common DNA variants in the ‘organ morphogenesis’ for nsCLP but not nsCPO, which points to the importance of the genes (in particular those analysed in this study) in this specific orofacial clefting subphenotype. The genome-wide significance of the MSX1 locus for nsCLP (Yu et al., 2017) supports this view.

Regarding lip development, our data suggest that all five genes studied in this work participate in a genetic network that controls Mnp morphogenesis during a crucial period of lip formation in mice (Fig. 5E). Although overt clefts of the lip represent a clinically relevant outcome of impaired lip development in humans, more subtle genetic alterations are increasingly recognized to manifest as milder phenotypes affecting lip and mid-facial morphology in the unaffected population (Boehringer et al., 2011; Liu et al., 2012; Miller et al., 2014; Wilson-Nagrani et al., 2018; Xiong et al., 2019). Our comparison identified a considerable overlap between genes identified in the ‘organ morphogenesis’ pathway and those associated with facial variations, thus supporting the concept of shared genetic effects on orofacial clefting and variations of facial development.

All experimental procedures were carried out under a project licence evaluated by the Newcastle University Local Ethics Committee and approved by the Home Office (UK), the Institutional Animal Care and Use Committee of Niigata University and the Kyushu Dental University Animal Care and Use Committee. Single heterozygous mice of Msx1 (Satokata and Maas, 1994) were maintained on a CD1 (Newcastle University) or C57BL/6 (Niigata Univ. and Kyushu Dental Univ.) genetic background. Wild-type or Msx1^+/− mice were used as controls. Single heterozygous mice of Pax9 (Peters et al., 1998) and Msx2 (Satokata et al., 2000) and combined double heterozygous mice of Pax9;Msx1 and Msx1;Msx2 were maintained on a CD1 background. Pax9^flox and PGK-Cre mice (Kist et al., 2007; Lallemand et al., 1998) were propagated on a C57BL/6 background. The day of vaginal plug detection was designated as E0.5. At least three samples of each genotype were collected and analysed.

Human foetal tissue sections were obtained from the MRC/Wellcome Trust-funded Human Developmental Biology Resource at Newcastle University (HBDR; http://www.hdbr.org), with appropriate maternal written consent and approval from the Newcastle and North Tyneside NHS Health Authority Joint Ethics Committee. HDBR is a licensed research tissue bank with the UK Human Tissue Authority.

Embryos were dissected and fixed with 4% paraformaldehyde (PFA) in phosphate buffered saline (PBS) at 4°C for overnight. For whole-mount in situ hybridization, samples were dehydrated through a graded series of methanol. For section in situ hybridization and immunohistochemistry, samples were dehydrated through a graded series of ethanol and xylene, embedded in paraffin and cut into 4 µm frontal sections. Non-radioactive whole-mount and section in situ hybridizations were carried out as previously described (Nakatomi et al., 2010). The MSX1 in situ probe was generated by PCR amplification using SP6 and T7 tagged primers 5′-AATACGATTTAGGTGACACTATAGAATACGTGCCTCTGGCCCCTTCCAGCGCG-3′ and 5′-TAAGTTAATACGACTCACTATAGGGCGAACATAGTACACACAATCCCTTCCA-3′.

For immunohistochemical analysis, sections were treated with anti-phospho-Smad1/5 primary antibody (rabbit monoclonal, #9516, Cell Signaling Technology, Danvers, MA, USA; 1:200 dilution), anti-rabbit secondary antibody (#BA-1000, Vector Laboratories, Burlingame, CA, USA; 1:500 dilution) and Vectastain Elite ABC kit (Vector; 1:50 dilution), essentially according to the manufacturers' instructions. Sigmafast 3, 3′-diaminobenzidine (DAB) tablets (Sigma-Aldrich, St Louis, MO, USA) were used for colour reaction, followed by counterstaining with diluted haematoxylin. A negative control experiment replacing primary antibody with PBS did not yield significant staining. For cell counting, a 200×50 µm rectangle was set at the mesial corner of the mesial nasal process, where Msx1 is intensely expressed, of both right and left sides of three serial sections (in total, six rectangles per individual) using Adobe Photoshop CS software (Adobe Systems, San Jose, CA, USA). The ratio of DAB-positive cells to total cells in the rectangle was measured using ImmunoRatio free web application (https://biii.eu/immunoratio; Tuominen et al., 2010). The average of six rectangles was calculated to obtain a pSmad score for each individual area; Student's t-test (two-tailed) was carried out to compare controls with Msx1^−/− mutant embryos (n=4 each). PAX9 immunostaining on human sections was carried out as previously described (Kist et al., 2014) using DAB as a chromogen.

Phenytoin injection was performed as previously reported (Azarbayjani and Danielsson, 2001). Briefly, phenytoin sodium salt (Sigma-Aldrich) was freshly dissolved in 0.9% NaCl (pH 11.7) and intraperitoneally injected into pregnant Msx1^+/− mice (CD1 background) at 11 am at both E10.5 and 11.5 at different concentrations (35, 60 and 85 mg/kg body weight). Embryos were dissected at E15.5 and fixed with Bouin's fixative after recording facial cleft state. Chi-square test was carried out for statistical analysis regarding the occurrence of cleft lip between wild-type (n=98, 85 mg/kg; 39, 60 mg/kg; 28, 35 mg/kg) and Msx1^−/− (n=12, 85 mg/kg; 19, 60 mg/kg; 18, 35 mg/kg) embryos. To label hypoxic cells, pimonidazole (60 mg/kg body weight) contained in a Hypoxyprobe-1 Plus Kit (Hypoxyprobe, Burlington, MA, USA) was intraperitoneally injected into pregnant mice 1 h before sacrifice at E11.5, after phenytoin injection as described above. Labelled cells in frontally cut paraffin sections were immunohistochemically visualized using primary antibody (anti-FITC-MAb1; 1:100 dilution), secondary antibody (HRP-conjugated rabbit anti-FITC; 1:400 dilution) and DAB tablets (Sigma-Aldrich), followed by counterstaining with diluted haematoxylin.

For hypoxia experiments, pregnant Msx1^+/− mice (C57BL/6 background) were kept in an airtight chamber (Deuce Co., Tokyo, Japan) in which the O₂ concentration was maintained at 10% for 48 h from E10.5 to E12.5. The oxygen concentration was continuously monitored with an oxygen analyser (JKO-25MT II R, Ichinen Jikco, Tokyo, Japan). Then, embryos were dissected at E15.5 and fixed with Bouin's fixative to analyse lip phenotype. Chi-square test was carried out for statistical analysis regarding the occurrence of cleft lip between wild-type (n=3) and Msx1^−/− (n=8) embryos.

To examine whether a hyperoxic environment can rescue phenytoin-induced cleft lip, pregnant Msx1^+/− mice (C57BL/6 background) were injected with phenytoin (60 mg/kg) once a day at E10.5 and E11.5 and kept at 50% O₂ concentration in the airtight chamber for 48 h from E10.5 to E12.5. The oxygen concentration was continuously monitored with an oxygen analyser. Then, embryos were dissected at E15.5 and fixed with Bouin's fixative to analyse lip phenotype. Phenytoin injection on pregnant Msx1^+/− mice (C57BL/6 background) under the same conditions in a normoxic environment was performed as control. Chi-square test was carried out for statistical analysis regarding the occurrence of cleft lip between normoxic (n=7) and hyperoxic (n=11) groups of Msx1^−/− mice.

Dissected embryos were fixed with 1% formaldehyde in PBS containing 0.02% NP-40 (PBS-NP) at 4°C for 20 min. Samples were washed three times with PBS-NP at room temperature for 15 min and incubated with staining buffer (2 mM MgCl₂, 5 mM K₄Fe(CN)₆·3H₂O, 5 mM K₃Fe(CN)₆ and 1 mg/ml X-gal in PBS-NP) at 37°C overnight, followed by several rinses with PBS.

Embryos were dissected and fixed with 2% glutaraldehyde in Sorenson's phosphate buffer at 4°C overnight. Subsequent tissue processing and scanning electron microscopy (SEM) were carried out according to a standard protocol at the Electron Microscopy Research Services, Newcastle University, UK. The oral and nasal angles of the medial nasal process were measured using Adobe Photoshop CS software (Adobe Systems). Student's t-test (two-tailed) was performed for the comparison of the oral or nasal angles; n=42 (WT), 16 (Msx1^−/−), 14 (Pax9^−/−) and 17 (Msx1^−/−;Pax9^−/−).

Affi-gel blue beads (Bio-Rad Laboratories, USA) were soaked with 500 µg/ml human recombinant BMP4 protein (R&D Systems, USA) or control PBS. Mouse embryos (E10.5) were harvested and the facial regions dissected prior to bead implantation into the Mnp. Samples were cultured by Trowel's standard method with Dulbecco's modified Eagle's medium (Sigma-Aldrich, USA) including antibiotics and 10% foetal bovine serum. After 8 h, cultured wild-type samples were fixed with 4% PFA for whole-mount in situ hybridization; Pax9^+/lacZ samples were fixed with 1% PFA in PBS for X-gal staining.

To test for association between each trait (nsCL/P, nsCLO and nsCLP) and common SNPs in each of the genes considered in this study (MSX1, MSX2, BMP4, PAX9, GREM1), a set-based test implemented in VEGAS2 (Liu et al., 2010) was performed (script accessed at https://vegas2.qimrberghofer.edu.au/, January 2017). To this end, we used our previous in-house meta-analysis data (Ludwig et al., 2017), which contained imputed data for nsCL/P, nsCLP and nsCLO. For each of the five genes, we extracted P-values for all common variants at the respective genetic loci (±50 kb, minor allele frequency >1%, info-score >0.4) and calculated whether there was an enrichment of SNPs with significant association P-values, while accounting for linkage disequilibrium (LD) between markers. An empirical P-value for the 10% top SNPs was calculated as the proportion of permutated test statistics less than the observed test statistics in each gene, with the numbers of permutations increasing depending on the P-value obtained.

To identify pathways in which genes with significant P-values from this gene-based test were overrepresented, an adapted version of VEGAS2 (VEGAS2-pathway analysis) was used. For this, gene-based P-values were calculated for all RefSeq genes and used as input. This tool additionally requires a ‘gene-pathway annotation’ file, in which GO pathways are defined and genes are assigned to those pathways. For the present analysis, we used the ‘Biosystems gene-pathway annotation file’, as available at the VEGAS2 homepage (January 2017). PubMed and OMIM databases available at the National Centre for Biotechnology Information platform (www.ncbi.nlm.nih.gov/) were used to search for genes with documented roles in lip, palate and facial development.

We thank J. Currie for PAX9 immunohistochemical staining on human embryo sections, L. Weinhold for statistical counselling and the team at the Electron Microscopy Research Services of Newcastle University for carrying out the SEM. We also thank L. Wolpert and J. Davies for earlier discussions about morphogenetic regulation. The human embryonic and foetal material was provided by the Joint MRC/Wellcome (MR/R006237/1) Human Developmental Biology Resource (www.hdbr.org).