Actomyosin contractility and a threshold of cadherin cell adhesion are required during tissue fusion

Tissue fusion, wherein two distinct embryonic primordia come together to become one continuous structure, is a common theme in embryonic development across the animal kingdom and requires an assemblage of intricate cell behaviors coordinated in space and time. For example, in Drosophila dorsal closure, the flanking epithelia are pulled together and fused to cover an epidermal gap on the dorsal side (Franke et al., 2005), and in chordate spinal cord formation, closure of the neural tube requires fusion of the dorsal neuroepithelium (Galea et al., 2017). Multiple tissue fusion events are required for normal mammalian craniofacial development, and their disruption is a meaningful contributor to congenital disease (Ray and Niswander, 2012).

Failed tissue fusion results in cleft lip, a common congenital anomaly occurring with or without cleft palate affecting roughly one in 700 live births worldwide (Dixon et al., 2011). Static views and histological sections of mammalian embryos during development have provided a gross understanding of the embryology of upper lip formation in mammals (Jiang et al., 2006; Cox, 2004) (Fig. 1 A). Upper lip morphogenesis is a multistep process that initiates with the formation of the medial and lateral nasal processes (MNP and LNP, respectively). The MNP and LNP are joined at their caudal ends, where they also meet with the maxillary process (MXP), at a point sometimes referred to as the lambdoidal junction (Tamarin and Boyde, 1977; Ferretti et al., 2011; Losa et al., 2018). Fusion is initiated between the MNP and LNP in a posterior-to-anterior direction in a zippering-like process that brings the epithelia of the two structures into contact followed by regression of the intervening epithelium to fully unite them (Trasler, 1968; Gaare and Langman, 1977a; Gaare and Langman, 1977b). All of this occurs concomitantly to overall tissue expansion, making the development of the upper lip a three-dimensionally complex and dynamic tissue morphogenesis. Static cytological and immunohistochemical studies in combination with studies of tissue fusion in other contexts have enabled postulation of some of the cellular behaviors that may be involved in lip development. Cell cycle arrest and high levels of apoptosis within the epithelium of the MNP/LNP junction are correlated with successful upper lip fusion (Jiang et al., 2006; Ferretti et al., 2011; Losa et al., 2018; Ji et al., 2020; Qu et al., 2025, Preprint). The cellular drivers of this process have not been directly functionally assessed, however, and the origins of cell-based forces powering this remarkable tissue fusion remain largely mysterious.

Generally, actomyosin contractility driven by non-muscle myosin II (NMII) is thought to generate mechanical force in mammalian cells and is key for cell movement and tissue shape changes during embryonic development (Vicente-Manzanares et al., 2009). Single-nucleotide variants in MYH9, encoding non-muscle myosin IIA, are associated with non-syndromic cleft lip with or without cleft palate in humans (Martinelli et al., 2007; Birnbaum et al., 2009; Chiquet et al., 2009). Cell–cell adhesion mediated by cadherins at adherens junctions connects to the actomyosin cytoskeleton and contributes to mechanical coupling during morphogenesis (Mège and Ishiyama, 2017; Heisenberg and Bellaïche, 2013). Notably, in humans, numerous mutations in genes encoding proteins of the adherens junction complex, including core components such as E-CADHERIN (CDH1) and P120-CATENIN (CTNND1), have been implicated in causing cleft lip as part of blepharocheilodontic syndrome and non-syndromic cleft lip (Kievit et al., 2018; Cox et al., 2018). A recent study found that CTNND1 was recurrently mutated in a study of 841 orofacial cleft (OFC) cases, making it the most common genetic cause of cleft lip in that population (Diaz Perez et al., 2023). P120-catenin is notable for its function as part of adherens complexes, regulating cadherin levels by suppressing their degradation by the proteasome (Ireton et al., 2002; Davis et al., 2003). Additionally, p120-catenin can regulate the cytoskeleton by interacting with Rho GTPases (Anastasiadis et al., 2000; Grosheva et al., 2001; Anastasiadis and Reynolds, 2001), can interact with the transcription factor Kaiso to regulate transcription, and can both facilitate and antagonize Wnt signaling (Daniel and Reynolds, 1999; Kelly et al., 2004; Spring et al., 2005; Hernández-Martínez et al., 2019). To date, there is only limited delineation of the context-specific cellular functions of p120-catenin deployed in vivo in mammalian development.

In this study, we interrogate cellular mechanisms of lip formation, focusing particularly on actomyosin contractility, apoptosis, and cell–cell adhesion. We use a combination of live imaging to examine the local dynamics and function of actomyosin contractility and a series of new genetic alleles in mice to dissect the in vivo molecular functions of p120-catenin and cadherin-mediated adhesion in this process. Our studies reveal locally enriched F-actin during lip fusion and indicate that actomyosin forces are required to pull embryonic prominences together. This fusion morphogenesis demands an elevated cadherin adhesion threshold satisfied through combined E-cadherin and P-cadherin levels.

The dynamic cellular processes underlying upper lip development remain unknown. As forces generated by the actomyosin cytoskeleton are a major driver of morphogenesis generally (Vicente-Manzanares et al., 2009), we developed a strategy for the live imaging of actomyosin dynamics during upper lip fusion in mouse embryos. We dissected the embryonic midface, comprising the MNP, LNP, MXP, and eyes, and cultured it in the presence of SiR-Actin, a fluorogenic probe for live cells based on F-actin binding jasplakinolide conjugated to a bright and photostable silicon-rhodamine fluorophore (Lukinavičius et al., 2014) (Fig. 1 B). We live-imaged wild-type embryos at multiple stages defined by the number of tail somites (ts), to capture the entire process of lip fusion. We identified three phases of actin dynamics, which we refer to as “initiation,” “expansion,” and “zipping.” At the initiation phase, roughly 6ts or embryonic day (E) 10.0, wild-type embryos exhibit dramatic accumulation of F-actin in the epithelium at the MNP and LNP junction with a particular enrichment in the LNP epithelium, which was highlighted by quantification of F-actin signal intensities (Fig. 1 C; Video 1; and Fig. S1, A and B). While the overall pattern of actin enrichment varied meaningfully from embryo to embryo, we consistently observed enrichment in the fusion junction–adjacent LNP. This pattern was also observed using a fluorogenic FastAct_X probe, which binds to but does not stabilize F-actin (Nasufovic et al., 2025, Preprint) (Fig. S1 C); all subsequent experiments were performed with SiR-Actin only. Later, at roughly 10ts or E10.5, F-actin remains enriched at the fusing seam and expands laterally through the epithelium of the MNP and LNP in long multicellular cables appearing to pull together the two nasal processes initiating fusion, leading us to consider this the “expansion phase” (Fig. 1 D and Video 2). Finally, at roughly 16ts or E11.0, when a substantial amount of fusion has already occurred, we detected supracellular actin cables that continue to zipper up the nasal processes to shorten the nasal slit, which we named the “zipping phase” (Fig. 1 E and Video 3). These results from using our novel live-imaging approach unveil a dynamic actin network that is patterned at the fusing nasal processes.

We next tested if and how actomyosin contractility is crucial to lip fusion. As the enriched actin observed in live imaging appeared localized to the epithelium, we reasoned that an epithelial Cre driver would be appropriate to ablate non-muscle myosin function during lip fusion. Therefore, we conditionally deleted Myh9 and Myh10, which encode the non-muscle myosin heavy chain IIA (NMHCIIA) and NMHCIIB isoforms, respectively, using the Crect craniofacial ectoderm Cre. Whole-mount nuclear staining of embryos at E12.5 revealed that while Myh9^f/f; Myh10^f/+; Crect and Myh9^f/+; Myh10^f/f; Crect mutants are comparable to controls, Myh9^f/f; Myh10^f/f; Crect double mutants exhibited bilateral cleft lip (Fig. 2 A). This result undoubtedly reveals a role of actomyosin contractility in proper lip formation; however, the extent to which it specifically drives fusion as opposed to other processes was unclear. To evaluate the impact of actomyosin contractility perturbation on the fusion process per se, we took advantage of our live imaging setup and commonly used pharmacological inhibitors. We added blebbistatin, a myosin II inhibitor, or dimethyl sulfoxide (DMSO) as a control condition to the culture media of tissue explants from 9ts wild-type embryos prior to imaging. Over the course of 10 h in controls, F-actin accumulated at the fusion junction as described above and fusion was readily observed (Fig. 2, B and C; and Video 4). While there was initial F-actin enrichment at the fusion junction in the blebbistatin-treated condition, the propagation of F-actin was dissipated and the MNP and LNP were separated (Fig. 2, B and C; and Video 5). Treatment with Y-27632, a ROCK inhibitor, also results in the lack of multicellular actin cables characteristic of the fusion junction and overtly appeared to disrupt fusion similar to blebbistatin (Fig. S1 D; and Videos 6 and 7). Thus, perturbation of actomyosin contractility results in stagnant fusion of nasal processes and ultimately cleft lip, revealing a critical role of actomyosin contractility in bringing the MNP and LNP together in the fusion process.

Actomyosin-generated cellular forces are transduced by mechanical coupling by cell–cell adherens junctions. We therefore next sought to dissect the cellular requirements of adherens junctions for fusion during lip morphogenesis. Others have previously noted that epithelial loss of p120-catenin results in cleft lip in mice (Cox et al., 2018), and we recapitulated this in embryos with conditional deletion of Ctnnd1 driven by the craniofacial ectoderm Cre, Crect, which we confirmed mediated loss of p120-catenin specifically within the epithelium in Ctnnd1^f/f; Crect^Tg/0 mutant embryos (Fig. S2). The majority of whole-mount nuclear–stained E12.5 Ctnnd1^f/f; Crect^Tg/0 mutant embryos exhibited cleft lip, compared with littermate controls that displayed completed fusion and a normal upper lip (Fig. 3 A and Table S1). Histological analysis of Ctnnd1^f/f; Crect^Tg/0 mutant embryos revealed epithelial disorganization at 9ts (E10.5) that was strikingly limited to the zone where we observed actomyosin contractility enrichment when the MNP and LNP were first making contact in controls (Fig. 3 B). Whereas controls formed an extended epithelial seam by 17ts (E11), Ctnnd1^f/f; Crect^Tg/0 mutant embryos failed to do so, leaving a gap between the MNP and LNP with disorganized epithelium specifically within fusing regions where actomyosin contractility would normally be elevated (Fig. 3 C). Notably, the epithelium remains intact and apparently normal in regions away from the actomyosin-enriched fusion zone.

Previous research on upper lip fusion has shown enrichment of apoptosis in the fusing epithelial seam, also known as the nasal fin (Jiang et al., 2006; Gaare and Langman, 1977b; Song et al., 2009). Analysis of cleaved caspase-3 by immunostaining in control embryos revealed apoptosis in the prefusion epithelium of the MNP and LNP at 9ts (Fig. 3 D), and also in the fusing nasal fin at 17ts (Fig. 3 E). Apoptosis appeared reduced in Ctnnd1^f/f; Crect^Tg/0 mutants at both 9ts and 17ts, but when quantified, this difference was only statistically significant at 17ts when fusion had failed, with cleaved caspase 3(+) cells remaining adjacent to but not a part of the epithelium (Fig. 3, D and E).

While apoptosis is thought to be a major cellular mechanism for lip fusion, its requirement has not been fully established. Thus, we interrogated whether apoptosis is necessary for proper lip fusion by abrogating intrinsic apoptosis in the epithelium through disruption of the pro-apoptotic regulators Bax and Bak (Westphal et al., 2011) in Bak^−/−; Bax^f/f; Crect^Tg/0 mutants. We observed near-complete loss of all cell death at the nasal fin at E11.5 through cleaved caspase-3 staining and TUNEL assay (Fig. 4 A), which was confirmed by quantification of cleaved caspase-3 surface volumes (Fig. 4 B), indicating that apoptosis in this context occurs through the intrinsic apoptotic pathway. At E17.5, whole-mount nuclear–stained Bak^−/−; Bax^f/f; Crect^Tg/0 mutants appear indistinguishable from control littermates (Fig. 4 C), suggesting that apoptosis in the epithelium is dispensable for successful lip fusion, though it may still contribute in combination with other cell behaviors. This result is also congruent with our recent finding that cell death is not needed for mouse secondary palate fusion, despite the prevailing idea that apoptosis is required for resolving the epithelial seam (Teng et al., 2022).

Given the prominent role of p120-catenin in adherens junctions, and recent findings that show mutations in CDH1 in human cleft lip, we anticipated that E-cadherin, a core component of epithelial adherens complexes, would be required for lip fusion. Immunostaining cryosections of 10ts embryos for E-cadherin revealed high expression and localization to the epithelial cell membrane of control embryos, whereas Ctnnd1^f/f; Crect^Tg/0 mutants displayed a substantial decrease of E-cadherin in the epithelium (Fig. 5 A), consistent with the role of p120-catenin in preventing E-cadherin endocytosis and degradation. Epithelial-specific disruption of Cdh1 led to E-cadherin loss (Fig. 6 B), but, astonishingly, did not result in cleft lip, and E12.5 Cdh1^f/f; Crect^Tg/0 mutant embryos were indistinguishable from littermate controls (Fig. 5 B).

Thus, we investigated which functions of p120-catenin are necessary for lip development. P120-catenin is a multifunctional protein that interacts with numerous other proteins and can regulate the cytoskeleton through RhoA association and facilitate WNT signaling through repression of the Kaiso transcriptional repressor, in addition to its role in stabilizing cadherins. As such, we sought to molecularly dissect its association with Kaiso, RhoA, and cadherins to determine the function of p120-catenin in lip development. We generated four new mouse lines harboring amino acid substitutions previously established in cell culture studies to affect specific functions of p120-catenin. These amino acid substitutions in the p120-catenin protein include (1) Y112E that results in decreased p120-catenin/RhoA interaction (Castaño et al., 2007); (2) S288A that results in decreased p120-catenin/Kaiso interaction (Zhang et al., 2011); (3) K622A,K623A that results in decreased p120-catenin interaction with both RhoA and Kaiso (Anastasiadis et al., 2000; Kelly et al., 2004); and (4) K401M that results in decreased p120-catenin/cadherin interactions (Yu et al., 2016; Ishiyama et al., 2010) (Fig. 5 C). To generate Ctnnd1 alleles that harbor genetic edits that would produce the above amino acid changes, we utilized the CRISPR/Cas9-based technique known as improved-Genome editing via Oviductal Nucleic Acids Delivery (i-GONAD) (Takahashi et al., 2015; Ohtsuka et al., 2018) (Fig. 5 D). To interrogate these functions of p120-catenin specifically in lip fusion, we generated compound heterozygous mice harboring one Ctnnd1 floxed allele with the other allele carrying one of these four function-ablating point mutations and examined them at E12.5 by whole-mount nuclear staining. Notably, Ctnnd1^Y112E/f; Crect^Tg/0 mutants, Ctnnd1^S288A/f; Crect^Tg/0 mutants, and Ctnnd1^{K622A,K623A/f}; Crect^Tg/0 mutants did not exhibit cleft lip (Fig. 5, E–G), indicating that RhoA and Kaiso binding are not required for p120-catenin function in this context. Cleft lip was only observed in Ctnnd1^K401M/f; Crect^Tg/0 mutants (Fig. 5 H), pointing to a crucial role of p120-catenin/cadherin interactions in the epithelium during lip development. Beyond craniofacial development, we were surprised to observe that E12.5 Ctnnd1^Y112E/Y112E, Ctnnd1^S288A/S288A, and Ctnnd1^{K622A,K623A/K622A,K623A} homozygous mutants also did not exhibit cleft lip and were viable and healthy as adult mice (Fig. S3, A–C), indicating that these interaction domains are not absolutely required for the many crucial in vivo functions of p120-catenin. Of note, we did detect low penetrant exencephaly in Ctnnd1^{K622A,K623A/K622A,K623A} mutants (Fig. S3 C), which is reminiscent of the phenotype caused by brain- and neural crest–specific loss of p120-catenin (Pieters et al., 2020). We did not recover any Ctnnd1^K401M/K401M embryos at E12.5, pointing to the importance of this interaction domain for general embryo development. Overall, these results indicate that p120-catenin binding to RhoA and Kaiso does not have a significant in vivo role in mammalian development, in contrast to the crucial p120-catenin regulation of cadherins. We therefore hypothesize that adherens junction regulation by p120-catenin is specifically vital for lip fusion.

As epithelial-specific loss of E-cadherin does not result in cleft lip (Fig. 5 B), we reasoned that a different cadherin may be involved in lip fusion. Although E-cadherin is the primary epithelial cadherin, P-cadherin is also present at the 6ts prefusion epithelium and somewhat lowered in Ctnnd1^f/f; Crect^Tg/0 mutants (Fig. 6 A). Further, we find both E and P-cadherin localized to the epithelium at 8–20ts, stages that span initiation, expansion, and zipping phases of lip fusion (Fig. S4). Intriguingly, P-cadherin has been shown to partially compensate for E-cadherin loss in some contexts (Tinkle et al., 2004). Immunostaining on cryosections shows P-cadherin persists in the fusion zone of 10ts Cdh1^f/f; Crect^Tg/0 mutants lacking E-cadherin (Fig. 6 B), hinting at a role of P-cadherin. Thus, to identify if P-cadherin is a key cadherin in lip fusion, we used i-GONAD to delete the first two exons of the P-cadherin–coding gene, Cdh3, generating a null mutant (Fig. 6 C). These Cdh3^−/− mutant mice do not exhibit cleft lip, and immunostaining shows E-cadherin persists in the fusion zone of 20ts Cdh3^−/− mutants lacking P-cadherin (Fig. 6 D). These mice are viable and fertile, consistent with previous findings that P-cadherin is not individually required for lip or general development (Radice et al., 1997).

We next sought to test whether the combined loss of E-cadherin and P-cadherin results in cleft lip. Complete loss of E-cadherin leads to early lethality at the blastocyst stage (Larue et al., 1994), and Cdh1 and Cdh3 are 46 kb apart, making generation of compound mutant embryos challenging through recombination during breeding (Hatta et al., 1991) (Fig. 6 C). Thus, we performed i-GONAD gene editing to generate a Cdh3 loss-of-function mutation linked to the Cdh1 floxed allele (Boussadia et al., 2002). Cdh3^−/−; Cdh1^f/f; Crect^Tg/0 mutants exhibited catastrophic loss of cell adhesion and dramatic craniofacial phenotypes, dying by E9.5, thereby revealing their shared contributions to embryonic epithelial development, but also precluding our analysis of lip development (Fig. 6 E). To more finely reduce overall cadherin levels, we combined the linked Cdh1 mutant; Cdh3 floxed alleles with unlinked Cdh1 and Cdh3 alleles to generate an allelic series of cadherin loss. We discovered that a subset of Cdh3^+/−; Cdh1^f/f; Crect^Tg/0 mutants exhibit cleft lip, while Cdh3^−/−; Cdh1^f/+; Crect^Tg/0 mutants did not (Fig. 6 E and Table S1). Together, these data reveal a new role of P-cadherin in lip fusion and demonstrate that there is a critical threshold of cadherin-based cell adhesion needed for proper lip fusion.

Previous studies indicate that CDH1, CTNND1, and MYH9 gene disruption may contribute to cleft lip in humans. Though a CDH3 homozygous truncating mutation has been previously implicated in ectodermal dysplasia, ectrodactyly, and macular dystrophy syndrome, these patients were not reported to exhibit orofacial clefting (Kjaer et al., 2005). We sought to examine CDH3 and extend our understanding of the extent of involvement of actomyosin contractility and cell adhesion genes in cleft lip. In people with an OFC, we found 22 rare, protein-altering variants in CTNND1, 20 such variants in CDH1, 21 in CDH3, 28 in MYH9, and 24 in MYH10 (Fig. 7 and Table S2). The combined annotation-dependent depletion (CADD) framework evaluates the potential pathogenicity of genetic variants (Kircher et al., 2014), and the majority of the variants we found (89.4%) have a CADD score >15 and thus would be predicted to be deleterious (Table S2). Seven variants are de novo (4 missense, 1 in-frame deletion, 1 nonsense, and 1 predicted splicing); three de novo variants (DNVs) were located in CDH1, two in CTNND1, one in CDH3, and one in MYH10. Seven of the inherited variants are inherited from a parent who also had an OFC. In CDH1, this included a missense variant and a synonymous variant predicted by SpliceAI to result in a loss of the splice donor site. In CDH3, this included a missense and nonsense variant. Notably, these CDH1 and CDH3 variants were absent from the exomes and genomes in the Genome Aggregation Database (gnomAD). The remaining variants inherited from affected parents were missense variants in CTNND1 (1 variant) and MYH9 (2 variants). All other rare variants were inherited from unaffected parents and may be considered variants of unknown significance, given the documented variable expressivity and incomplete penetrance among families with OFCs. When we tested for a burden of DNVs, we found that these five genes collectively had an enrichment of DNVs (P = 0.001), including an enrichment of missense (P = 0.008) and loss-of-function DNVs (P = 0.003) (Table S3). Most of this signal came from CTNND1 (P = 0.006) and CDH1 (P = 0.005), which were the only genes that also individually had a burden of DNVs (Table S3). However, even with those two genes removed, there was still a small enrichment of missense (0.04) and loss-of-function DNVs (P = 0.03) from variants in CDH3, MYH9, and MYH10. Of note, a CDH3 variant found in three unrelated individuals with cleft lip is predicted to result in premature termination c.212G>T[p.Glu708Ter] within the p120-catenin interaction domain of P-cadherin (Fig. 7 C). For comparison, we selected 3 control genes (MYH14, CDH5, and NKX2-1) that we did not predict to be involved in craniofacial development based on known mouse phenotypes and expression patterns. Of these, only MYH14 had a single DNV, which was not more than expected by chance (P = 0.22). In these control genes, there is no enrichment in missense variants (P = 0.32), protein-altering variants (P = 0.34), or all DNVs (P = 0.45) in this list. Together with other recently published human genetic findings (Cox et al., 2018; Diaz Perez et al., 2023; Alharatani et al., 2020), these data strongly indicate that disruption of the cell biological mechanisms we have uncovered here correlates with human cleft lip.

To understand the interplay between actomyosin contractility and cell adhesion in lip fusion, we investigated the effects of adhesion loss on the dynamic and patterned actin network that we observed through live imaging of the fusing nasal processes. We dissected multiple stages of control and Ctnnd1^f/f; Crect^Tg/0 mutant embryos to capture different phases of the fusion process (Fig. 1 B). These live samples were labeled with SiR-Actin and/or Hoechst for F-actin and nuclei, respectively, and imaged. Prior to the initiation of fusion, actin enrichment was not yet observed, and control and mutant embryos were indistinguishable (Fig. 8 A). During the initiation phase, controls showed enriched actin organized into cables at and around the fusion site including the LNP as previously observed (Fig. 8 B). In Ctnnd1^f/f; Crect^Tg/0 mutants, however, F-actin accumulated at the fusion region lacked filamentous architecture, and quantification revealed significantly fewer organized cables in mutants (Fig. 8, B and E). Instead, in regions corresponding to those exhibiting F-actin multicellular cables in controls, we observed rounded cells enriched in F-actin that were detaching from the epithelium (Fig. 8 B). This is consistent with the spatially specific epithelial disorganization we observed through histological analysis of the mutants (Fig. 3, B and C). Some of these rounded cells exhibited cleaved caspase-3 staining and appeared to be separating from the epithelium, suggesting that these cells may be undergoing anoikis possibly as a consequence of loss of adhesion coupled with high cortical actomyosin contractility (Fig. 8 C). Taken together, we find that p120-catenin is not required for actin enrichment at the fusion site but is required for its organization into the multicellular actin cable patterns characteristic of active fusion.

While Cdh1^f/f; Crect^Tg/0 mutants have lowered adherens junctions, the lack of cleft lip suggests that a critical adhesion threshold required during fusion is satisfied. To test this hypothesis, we challenged Cdh1^f/f; Crect^Tg/0 mutants by hyperactivating myosin contractility by treating with calyculin A, an inhibitor of myosin light chain phosphatase. Actin organization in tissue explants from 8ts Cdh1^f/f; Crect^Tg/0 mutants treated with DMSO and stained with SiR-Actin appeared comparable to littermate controls (Fig. S5). We then added calyculin A to tissue explants of 8ts control and Cdh1^f/f; Crect^Tg/0 mutant embryos 15 min prior to imaging. With this treatment, controls exhibited multicellular actin cables, while mutants had fewer and shorter cables and also rounded cells at the fusion region (Fig. 8, D and E), reminiscent of Ctnnd1^f/f; Crect^Tg/0 mutants. These results support the idea that a critical adhesion threshold must be met during lip fusion, possibly to withstand high actomyosin contractility forces.

Uncovering the cellular forces and mechanisms driving tissue fusion is essential to a comprehensive understanding of mammalian tissue morphogenesis. While principles of morphogenesis have been established in powerful non-mammalian model systems, our understanding of mammalian morphogenesis, and therefore the detailed cellular basis of human congenital anomalies, has lagged due to challenges intrinsic to live imaging of the viviparous mouse embryo. Here, we reveal dynamic cellular processes driving initiation of a mammalian tissue fusion through the establishment of a novel live-imaging platform for the upper lip. We demonstrate that F-actin is enriched in a highly localized pattern and that actomyosin-generated forces are crucial for drawing together embryonic prominences and zippering the upper lip. This is particularly notable in light of the fact that several studies have shown human cleft lip associated with mutations in MYH9, which encodes NMHCIIA (Martinelli et al., 2007; Birnbaum et al., 2009; Chiquet et al., 2009). Dorsal closure tissue fusion in Drosophila, ventral enclosure in Caenorhabditis elegans hypodermis, and spinal neural tube closure in mice are driven, at least in part, by the uniform contraction of a biomechanically coupled multicellular actomyosin purse-string, while neural tube closure in Ciona robusta occurs by sequential junctional actomyosin contraction that results in progression of a fusion zipper (Williams-Masson et al., 1997; Pyrgaki et al., 2010; Hashimoto et al., 2015; Wernike et al., 2016; Galea et al., 2017; Hayes and Solon, 2017; Hashimoto and Munro, 2019). We find that the zipping stage of upper lip fusion also exhibits purse-string-like actomyosin zippering to shorten the nasal slit, though the earlier initiation and expansion steps of upper lip fusion involve a unique pattern of F-actin enrichment that appears to pull the epithelia of the MNP and LNP together. This early and dramatic function of actomyosin contractility may be more akin to the requirement of actomyosin contractility in amnioserosa cell constriction that pulls the epidermal fusion fronts together in Drosophila dorsal closure. Our novel imaging approach establishes upper lip fusion as a system for future detailed studies to untangle the relative contributions of each of these actomyosin and other cellular mechanisms.

Through a battery of newly generated mouse mutants, we also reveal the detailed molecular requirements for p120-catenin during upper lip fusion. Previous studies have proposed that p120-catenin acts as a crucial scaffold protein that regulates numerous molecular functions (Anastasiadis et al., 2000; Anastasiadis and Reynolds, 2001; Daniel and Reynolds, 1999; Spring et al., 2005; Castaño et al., 2007; Zhang et al., 2011; Yu et al., 2016; Noren et al., 2000; McCrea and Park, 2007). P120-catenin can bind directly to RhoA in the cytoplasm and can act as a Rho guanine nucleotide dissociation inhibitor (RhoGDI) to inhibit RhoA activity (Anastasiadis et al., 2000; Castaño et al., 2007). Phosphorylation of the Tyr112 residue in p120-catenin prevents p120-catenin/RhoA binding, while mutation of two lysines, K622 and K623, disrupts the ability of p120-catenin to act as a RhoGDI (Anastasiadis et al., 2000; Castaño et al., 2007). In our study, mice harboring the Y112E phosphomimetic mutation exhibited no discernible phenotype, with adult Ctnnd1^Y112E/Y112E mice viable and healthy. This was also the case for most Ctnnd1^{K622A,K623A/K622A,K623A} homozygous mutant mice. While we have not confirmed that these mutations also disrupt p120-catenin/RhoA interactions in vivo, based on published work in cell culture, we interpret this to mean that binding of p120-catenin to RhoA and its ability to act as a GDI are both dispensable for normal development and viability in mice. These lysine mutations have also been shown to impact the nuclear localization of p120-catenin and therefore its inhibitory effect on Kaiso transcriptional activity (Kelly et al., 2004). Phosphorylation of Ser288 is also required for p120-catenin nuclear accumulation and its ability to shuttle Kaiso to the cytoplasm, which are disrupted by the S288A mutation (Zhang et al., 2011). As Ctnnd1^S288A/S288A mice also did not exhibit developmental phenotypes, we conclude that Kaiso regulation is likely dispensable for development, though a role of this regulatory modality in cancer cannot be ruled out (Zhang et al., 2011; Mayerle et al., 2003; Vermeulen et al., 2012; Thoreson and Reynolds, 2002). In one out of six of the E12.5 Ctnnd1^{K622A,K623A/K622A,K623A} embryos we examined, we observed exencephaly that was similar to that observed in Ctnnd1 neural-specific mutants (Pieters et al., 2020), which may indicate an additive effect of the loss of RhoA and Kaiso regulation by p120-catenin. However, Ctnnd1^{K622A,K623A/K622A,K623A} mice are born at a Mendelian ratio, viable, and healthy in our colony, suggesting that such an additive function for these interactions is minor in comparison with the function of p120-catenin in binding to cadherins to stabilize cell–cell adhesion.

Indeed, we find that generation of a K401M amino acid substitution in p120-catenin that specifically disrupts binding to cadherins results in a cleft lip phenotype reminiscent of epithelial loss of p120-catenin (Yu et al., 2016). Previous studies in MDCK cells showed that this mutation or complete loss of p120-catenin resulted in a dramatic change in epithelial architecture and lateral expansion of active RhoA, which could not be rescued by mutations that force stabilization of E-cadherin at the membrane. These results contributed to the interpretation that p120-catenin has important functions suppressing RhoA and actomyosin contractility at cadherin-based cell–cell contacts to regulate epithelial architecture. While that study did not identify a relevant RhoA interaction mechanism, and we did not observe any effect of disruption of the RhoA-binding domains of p120-catenin in vivo, we still cannot rule out that p120-catenin may regulate RhoA through some other unknown mechanism. We find, however, that the cadherin interaction domain of p120-catenin is crucial in vivo, leading us to favor the concept that the principal, and possibly sole, function of p120-catenin during development relates to cadherin stabilization. As MDCK cells also express P-cadherin, p120-catenin may also have an important role in its stabilization in that context, potentially explaining the discrepancy (Wu et al., 1993). Our findings are also consistent with recent studies in vascular endothelial development, where loss of p120-catenin, which is crucial for vascular integrity, can be rescued by a VE-cadherin mutation that impairs VE-cadherin endocytosis, indicating that stabilization of VE-cadherin is the main function of p120-catenin in that context (Grimsley-Myers et al., 2020).

While p120-catenin is crucial for tissue fusion in the upper lip, compound reduction of E-cadherin and P-cadherin is required to reveal their role in this process. While compensatory or synergistic roles of P-cadherin and E-cadherin have been demonstrated in the skin (Tinkle et al., 2004) and lung organoids (Hirai et al., 1989), in vivo interrogation of their combined function has been stymied by the genetic proximity of the Cdh1 and Cdh3 genes. Our gene editing strategy generates a new Cdh1^lox; Cdh3⁻ allele for untangling this question in vivo. We reveal a catastrophic loss of cell adhesion and embryonic lethality in Cdh3^−/−; Cdh1^f/f; Crect^Tg/0 mutant embryos prior to midface formation, supporting important and widespread collaboration of E- and P-cadherin during development. Importantly, we observed a highly specific cleft lip phenotype when E- and P-cadherin were reduced in Cdh3^+/−; Cdh1^f/f; Crect^Tg/0 mutants, indicating that bolstered adherens junctions are specifically required during this epithelial fusion process. As we do not see changes in P-cadherin abundance upon loss of E-cadherin or vice versa, we do not think that one is compensating for the loss of the other. Therefore, these findings suggest that there is not a single key cadherin or a cadherin code functioning in the epithelium for fusion and instead support the idea that a critical elevated threshold of reinforced adherens junctions is required specifically at the fusion site. This cooperative role of E- and P-cadherin may parallel that seen between Nectin1 and Nectin4 during palatogenesis, where loss of both adhesion proteins resulted in a severe cleft palate while cleft palate was seen with low penetrance in single knockdowns (Lough et al., 2020).

Cell adhesion may play at least two distinct roles during lip fusion. First, a minimum threshold of cadherins may be required for nascent adherens complexes between cells of apposed nasal processes during initial contact. Second, high tension forces generated by actomyosin contractility needed to pull the apposing processes together may necessitate a minimum threshold of cadherin function that is higher than the surrounding non-fusing epithelium (Fig. 9). We find that this threshold requires contributions from both E-cadherin and P-cadherin, as well as reinforcement through p120-catenin stabilization. While p120-catenin is dispensable during normal Drosophila development, abnormalities in cell arrangement and actin organization were observed in dorsal closure tissue fusion. Notably, levels of p120-catenin junctional localization were highest in regions of the embryo undergoing morphogenetic movements (Myster et al., 2003). Together with our findings, these results suggest a conserved role of p120-catenin in strengthening cell adhesion in tissues experiencing elevated mechanical stress. We also infer, from loss of the multicellular cable patterns upon disruption of p120-catenin, that adherens junctions are crucial in propagating actomyosin tensional forces to neighboring cells during fusion. Future studies analyzing the changes in forces during lip development in mutants will be needed to understand the intricate interplay between cadherin-mediated adhesion and actomyosin tension. In addition, adherens complexes are mechanosensors that convert mechanical forces into biochemical signals (Borghi et al., 2012). A mechanotransduction role of the cadherin threshold may also be pertinent in the context of lip fusion, for example, by strengthening junctions under actomyosin contractility through catch bonds (Buckley et al., 2014). One fascinating possibility is that mechanotransduction through cadherins could result in the initiation of other cell behaviors, such as cell cycle arrest or apoptosis, which may also contribute to completion of fusion potentially in combination with other cellular behaviors (Losa et al., 2018; Qu et al., 2025, Preprint).

Apoptosis has often been considered as a key terminal cell behavior facilitating tissue fusion through the removal of epithelial cells. We find that regionalized cell death at the fusion junction coincides with regionally enriched F-actin. This, together with the finding that Bak; Bax mutants lacking apoptosis at the fusion site do not have cleft lip, insinuates that cell death may indeed be a consequence of increased actomyosin-generated tension and not, by itself, a predominant driver of fusion as currently thought. This is consistent with findings of abundant apoptosis that correlate with high F-actin during mammalian neural tube closure (Yamaguchi et al., 2011) and enhanced actomyosin dynamics in secondary palate fusion (Teng et al., 2022). In both cases, apoptosis has been demonstrated not to be required to complete tissue fusion (Massa et al., 2009; Teng et al., 2022). In the secondary palate, attenuated actomyosin contractility through genetic perturbation of NMII also resulted in fewer apoptotic cells. This further suggests that apoptosis is a consequence of the high actomyosin contractility that initiates tissue fusion events, though the relative functional contributions of apoptosis remain unclear (Teng et al., 2022).

In summary, we show that actomyosin contractility provides forces driving tissue fusion in the mammalian upper lip. Sensitivity of fusion to reduced cadherin-mediated cell adhesion is localized to a region of high actomyosin contractility, which manifests as a biomechanical fault point that requires p120-catenin–mediated reinforcement of adherens junctions and transduction of actomyosin-generated forces (Fig. 9). A similar mechanism may explain cleft secondary palate observed in epithelial-specific p120-catenin mouse mutants (Cox et al., 2018) and exencephaly observed in neural-specific p120-catenin mouse mutants (Pieters et al., 2020). Patients with cleft lip have previously been shown to harbor variants in CDH1, while variants in CTNND1 are among the most common genetic causes of cleft lip in humans (Cox et al., 2018; Caetano da Silva et al., 2024; Diaz Perez et al., 2023). Here, we reveal novel genetic variants that further suggest that this pathway may comprise a crucial functional module that reinforces cell adhesion to withstand and transduce morphogenetic force during formation of the mammalian upper lip.

All animal experiments were performed in accordance with the protocols approved by the University of California, San Francisco Institutional Animal Care and Use Committee. The following alleles were backcrossed to, and maintained on, a coisogenic C57BL/6J genetic background: Crect (MGI:4887352) (Reid et al., 2011), Bax^lox/lox (MGI: 3589203) (Takeuchi et al., 2005), and Bak^−/− (MGI:1097161) (Lindsten et al., 2000). Cdh1^f/f (MGI:2389020) (Boussadia et al., 2002), Ctnnd1^f/f (MGI:3640772) (Elia et al., 2006), Myh9^lox/lox (MGI:4838521) (Jacobelli et al., 2010), and Myh10^lox/lox (MGI:4443039) (Ma et al., 2009) alleles were used on a mixed background.

Point mutations in Ctnnd1 and deletion mutations in Cdh3 were generated using the CRISPR/Cas9-based method called i-GONAD (Ohtsuka et al., 2018; Gurumurthy et al., 2019). For generating Ctnnd1 variants, C57BL/6J wild-type males were crossed to CD1 wild-type females, and a 2 μl mixture comprised of the following was injected into each oviduct of female mice pregnant with E0.7 embryos and electroporated: 0.7 μg CRISPR RNA (crRNA), 0.7 μg trans-activating crRNA (tracrRNA), 2 µg donor single-stranded oligodeoxynucleotides (ssODN), and 2 ng Cas9 protein. For generating an independent Cdh3 mutant allele, we used CD1 wild-type females crossed to C57BL/6J wild-type males. In generating a Cdh3 allele linked to the floxed Cdh1 allele, we used Cdh1^f/f females intercrossed with Cdh1^f/f males. For both Cdh3 alleles, we used the following injection mixture: 0.7 μg each of two crRNAs, 1.4 μg tracrRNA, and 2 ng Cas9 protein. All crRNA, tracrRNA, ssODN, and Cas9 were acquired from IDT. Germline transmission of desired gene edits was confirmed by sequencing the progeny of candidate founders. We screened all off-target sites predicted by CHOPCHOP (Montague et al., 2014; Labun et al., 2016; Labun et al., 2019) that had fewer than three mismatches with the target sequences or were on the same chromosome with the target gene.

The Ctnnd1^Y112E allele includes a missense mutation that changes codon 112 from TAT (tyrosine) to GAG (glutamic acid). The Ctnnd1^S288A allele includes a missense mutation that changes codon 288 from AGC (serine) to GCA (alanine). No additional edits were made at these loci. The Ctnnd1^K401M allele includes a missense mutation changing codon 401 from AAG (lysine) to ATG (methionine) and a silent mutation in the protospacer adjacent motif sequence changing codon 405 from CGG (arginine) to CGT (arginine). The Ctnnd1^K622A,K623A allele includes a missense mutation that changes codons 622-623 from AAGAAG (lysines) to GCAGCA (alanines). The Cdh3^Δ404 allele contains a 404-base pair deletion, while the Cdh3^Δ398 allele contains a 398-base pair deletion that is linked with Cdh1^flox. Both Cdh3 deletions encompass exon 1, intron 1, and exon 2. Two independent founders were used to establish independent lines for each allele, except for Cdh3^Δ398 that only had one successful founder. In all cases with two independent lines, both lines produced the same results. Detailed sequences of guide RNAs and genotyping primers are listed in Table S4.

The confocal live-imaging approach was adapted from our previous work (Kim et al., 2015; Kim et al., 2017). Specifically, embryo heads were removed in phosphate-buffered saline (PBS) and transferred to culture media [20% fetal bovine serum (FBS), 100 U/ml penicillin, 100 μg/ml streptomycin, 200 μg/ml L-ascorbic acid, Dulbecco’s modified Eagle’s medium/F12 media that contain L-glutamate and HEPES and no phenol-red]. Heads were dissected leaving intact nasal processes, MXPs, mandible, and eyes. This facial tissue was then embedded in 0.6% low-melting agarose in culture media that contained SiR-Actin (1:2,000 or 0.5 μM, CY-SC001; Spirochrome) or SPY-650-FastAct_X (1:700, CY-SC500; Spirochrome). After incubation at room temperature for 15 min for the agarose to set, the tissue was incubated for another 45 min at 37°C before imaging. To inhibit actomyosin contractility, culture media–agarose mixture also contained blebbistatin (B0560; Sigma-Aldrich, 20 μM) or Y-27632 (SC-281642; Santa Cruz Biotechnology, 20 μM). Thus, treatment began 1 h prior to and continued throughout imaging. Focusing on the fusing MNP and LNP, we imaged at 10- or 20-min intervals using an inverted Zeiss Observer.Z1 spinning disk confocal microscope, a Plan-Apochromat 20× air objective with 0.8 numerical aperture, Zeiss Axiocam 506 Mono Camera, and a 37°C incubation chamber. For Y-27632 treatments and controls, samples were imaged using an inverted Zeiss LSM900 Airyscan 2 laser scanning confocal microscope, a Plan-Apochromat 40× oil objective with 1.3 numerical aperture, and a 37°C incubation chamber. Both microscopes utilize Zeiss Zen acquisition software. In experiments looking at actin in Ctnnd1 mutants and controls, Hoechst 33342, trihydrochloride trihydrate (1:2,000, H1399; Thermo Fisher Scientific) was included in the agarose/media/SiR-Actin mixture. For these samples and Cdh1 mutants, we acquired image stacks using a Zeiss LSM900 microscope and Plan-Apochromat 25× objective with 0.8 numerical aperture and water without continuing the time lapse. Cdh1 mutant and respective control tissue explants were incubated in culture media containing SiR-Actin for 45 min at 37°C, then embedded in 0.6% low-melting agarose in culture media that contained SiR-Actin and calyculin A (ab141784; Abcam, 40 nM) for 15 min at room temperature prior to imaging. All images were rendered in 3D and time-lapse data made into movies using Imaris software (Oxford Instruments). As the SiR-Actin signal intensity increases over time, threshold levels were adjusted throughout the movie to circumvent signal oversaturation.

Whole embryos were fixed in Bouin’s solution (HT101128; Sigma-Aldrich), dehydrated through an ethanol gradient, and processed through a Histo-Clear (HS-200; National Diagnostics) gradient and wax before being embedded into wax. Blocks were cut to 5-µm sections that were subsequently stained with hematoxylin and eosin. Stained sections were imaged using a Zeiss Imager.Z2, an EC Plan-Neofluar 5× air objective with numerical aperture 0.16 and Plan-Apochromat 63× oil objective with numerical aperture 1.4, an AxioCam MRc5 camera, and AxioVision software.

To assess cleft lip and other gross morphological changes, embryonic heads were fixed in 4% paraformaldehyde (PFA) in PBS, dehydrated through a methanol gradient, and processed in Dent’s bleach (4 parts 100% methanol, 1 part 30% hydrogen peroxide, 1 part DMSO). Samples were then rehydrated through a decreasing methanol gradient, washed in PBS + 0.1% Triton X-100, stained with DAPI (1:20,000) overnight at 4°C, and washed in PBS prior to imaging with a Zeiss Imager.Z2, an EC Plan-Neofluar 2.5× air objective with numerical aperture 0.075, an AxioCam MRm camera, and AxioVision software. Bak; Bax samples, Myh9; Myh10 samples, and respective controls were imaged using a Nikon AZ100M Macro Confocal, AZ Plan Apo Air 1× lens with numerical aperture 0.1, and Nikon Elements software. Details on penetrance and unilateral/bilateral cleft can be found in Table S1.

For whole-mount immunofluorescence experiments, embryonic heads were fixed in 4% PFA in PBS, dehydrated through a methanol gradient, and processed in Dent’s bleach (4 parts 100% methanol, 1 part 30% hydrogen peroxide, 1 part DMSO). Samples were then rehydrated through a descending methanol gradient, washed in PBS, incubated in blocking solution (5% normal donkey serum, 0.1% Triton X-100 in PBS) for two hours at room temperature, and treated with primary antibodies overnight at 4°C. Following five one-hour washes in PBS, samples were treated with secondary antibodies overnight at 4°C, subjected to five 20-min washes in PBS, and stained with DAPI (1:10,000) overnight at 4°C. Lastly, samples were washed five times in PBS for 20 min each, dehydrated through a methanol gradient, and cleared through a benzyl alcohol: benzyl benzoate-in-methanol gradient (Ahnfelt-Rønne et al., 2007) before imaging with a Zeiss Observer.Z1 spinning disk confocal microscope (paired with a Plan-Apochromat 10× air objective with 0.45 numerical aperture and a C-Apochromat 40× water objective with 1.1 numerical aperture) or Zeiss LSM900 microscope (paired with a Plan-Apochromat 25× objective with 0.8 numerical aperture and water).

For cryosection immunofluorescence experiments, whole embryos were fixed in 4% PFA in PBS, dehydrated through a sucrose gradient, embedded in Tissue-Tek O.C.T. Compound, and frozen in a dry ice/ethanol bath. Blocks were cut using a CryoStar NX70 HOMP (Thermo Fisher Scientific) or a CM1900 (Leica) cryostat to 12-µm sections onto SuperFrost Plus slides (Thermo Fisher Scientific). Sections were immunostained according to standard methods with antibodies recognizing cleaved caspase-3 (1:300, 9661; Cell Signaling), E-cadherin (1:300, 13-1900; Invitrogen), p120-catenin (1:300, 610133; BD Biosciences), and P-cadherin (1:300, AF761-SP; R&D Systems). Sections were mounted in Antifade Mounting Medium and imaged with a Zeiss Imager.Z2 and an EC Plan-Neofluar 5× air objective with numerical aperture 0.16 or Zeiss LSM900 laser scanning microscope and Plan-Apochromat 25× objective with 0.8 numerical aperture and water or Plan-Apochromat 40× objective with 1.2 numerical aperture and water. For apoptosis experiments that include TUNEL technology, we utilized an in situ cell death detection kit (11364795910; Roche) and followed the manufacturer’s instructions.

To quantify the intensity distribution of F-actin across the nasal processes during fusion of wild-type embryos, we used Imaris to draw a line across 3D renderings of the tissue such that the line is perpendicular to and crosses the vertex of the fusion point and spans the MNP and LNP. We then used the Surface function to create surfaces based on the F-actin signal with the following parameters in conjunction with the automated wizard: 0.5 μm surface grain size; eliminate background enabled, with largest sphere diameter of 5 μm; manual threshold of 20; morphological split enabled, with region growing estimated diameter of 3 μm; and filtered surfaces above 15.0. From this, we selected surfaces that intersected the previously drawn line and graphed the intensities of such surfaces according to their spatial positioning.

To calculate the closure between nasal processes in wild-type embryos treated with DMSO or blebbistatin, we measured the distances between the MNP and LNP in 3D using Imaris. The vertex of the fusion was visually determined, and points were placed one cell distance rostral from the vertex. In regard to anterior/posterior positioning, the two opposing points were placed on the nasal pit side of the MNP and LNP in a way that they are positioned closest together, avoiding too frontal or deep in the pit. The distance between the two points was calculated by Imaris. Statistical analysis was performed using GraphPad Prism 8, and an unpaired, two-tailed Student’s t test was used to determine significance.

The amount of apoptosis was quantified at two stages in Ctnnd1; Crect^Tg/0 mutants and controls. Cleaved caspase-3-positive cells were manually counted through intact facial processes immunostained for cleaved caspase-3, E-cadherin, and nuclei with DAPI using ImageJ/FIJI software. At the earlier stage of 8–9ts prefusion, cleaved caspase-3-positive cells in and adjacent to the epithelium were counted. At the later stage of 17–18ts where a fusion seam exists, only cleaved caspase-3-positive cells in the epithelium were counted. Statistical analysis was performed using GraphPad Prism 8, and an unpaired, two-tailed Student’s t test was used to determine significance.

In Bak^−/−; Bax^f/f; Crect^Tg/0 mutants, to calculate the percentage of fusion junction cells, as measured by DAPI, that is apoptotic, as measured by cleaved caspase-3, we divided the volume of DAPI signal by the volume of cleaved caspase-3 from tissue sections. Volumes were determined from surfaces created in Imaris, including only the junction epithelium. Surfaces based on the DAPI signal were created with the following parameters: 0.25 μm surface grain size; manual threshold of 1,700; morphological split enabled, with region growing estimated diameter of 1 μm; and filtered volume threshold of 496. Surfaces based on the DAPI signal were created with the following parameters in conjunction with the automated wizard: 0.25 μm surface grain size; manual threshold of 700; morphological split enabled, with region growing estimated diameter of 1 μm; and filtered volume threshold of 496. For one Bak^−/−; Bax^f/f; Crect^Tg/0 mutant, we lowered the manual threshold to 300 to better capture cleaved caspase-3 signal. Statistical analysis was performed using GraphPad Prism 8, and an unpaired, two-tailed Mann–Whitney t test was used to determine significance.

To determine the length of F-actin–based segments in Ctnnd1; Crect^Tg/0 mutants, controls, and calyculin A-treated Cdh1; Crect^Tg/0 mutants and controls, we utilized the Filament function in Imaris software. Segments were traced with the following parameters in conjunction with the automated wizard: detection type of AutoPath with no loops; dendrite starting point diameter of 4.88 μm; dendrite seed point diameter of 1.50 μm; dendrite seed point threshold of 6,431 (lowered to 4,000 for a single Ctnnd1; Crect^Tg/0 mutant sample, a single calyculin A-treated control sample, and a single Cdh1; Crect^Tg/0 mutant sample to more faithfully detect the signal); diameter around starting point to remove seed points of 4.88 μm; and removing disconnected segment smooth width of 0.293 μm. The program wizard was trained to keep seed points associated with filaments/cables and to discard seed points associated with rounded cells or bright pixels unassociated with filaments/cables. Program-predicted seed points and segments were manually filtered to remove false-positive points or segments, though none were manually added. Statistical analysis was performed using GraphPad Prism 8, and an unpaired, two-tailed Mann–Whitney t test was used to determine significance.

To assess whether there were any genetic variants in CTNND1, CDH1, CDH3, MYH9, or MYH10, we used whole-genome sequencing data from the Gabriella Miller Kids First Pediatric Research Program (GMKF), which recruited 1,777 case-parent trios with cleft lip, cleft lip and palate, or cleft palate. Participants were recruited from the United States, Argentina, Turkey, Hungary, Spain, Colombia, and Taiwan. Participant recruitment was done at regional treatment or research centers after review and approval by each site’s institutional review board (IRB) and the IRB of the affiliated US institutions (e.g., the University of Iowa, the University of Pittsburgh, and Johns Hopkins University). Participants were assessed at the time of recruitment for additional clinical features and are largely classified as having an isolated cleft, but participants with additional clinical features consistent with a syndromic diagnosis were not excluded.

Samples from the GMKF were sequenced at the McDonnell Genome Institute (MGI), Washington University School of Medicine in St. Louis, MO, USA, or the Broad Institute in Boston, MA, USA. The variant calling and quality control methodology for 1,254 of the GMKF trios was done as part of previous studies (Bishop et al., 2020; Mukhopadhyay et al., 2020), and the quality control for the additional 523 trios was done similarly. Variants were called and aligned to the GRCh38/hg38 reference genome using the GATK pipeline (Conrad et al., 2011; McKenna et al., 2010; Van der Auwera et al., 2013). For variant-level QC, non-passing variants, variants with more than 2 Mendelian errors, variants with >5% missingness in individuals, and variants with a Hardy–Weinberg equilibrium P value <1 × 10^–7 were removed. Additionally, calls with a genotype quality <20 or a depth <10 were set to missing and only biallelic variants were retained using VCFtools (Danecek et al., 2011). For sample-level QC, PLINK (v1.9) (Chang et al., 2015) was used to calculate identity-by-descent to confirm all family relationships and X chromosome heterozygosity to confirm the sex of all participants.

All variants were annotated to publicly available databases using ANNOVAR (v201707) (Wang et al., 2010) and Variant Effect Predictor (McLaren et al., 2016) to assess population frequency (from gnomAD v4), pathogenicity (CADD score), or effect on protein (missense, predicted loss-of-function, splicing). Noncanonical splice variants (synonymous, splice region variants) were retained if the SpliceAI score exceeded 0.5 (Jaganathan et al., 2019).

From this dataset, we extracted rare protein-altering variants (MAF <0.1% in gnomAD v4) that were present in at least one participant with a cleft. We also noted if a variant was de novo or inherited, and if inherited, if from a parent with a cleft. We also tested if there was a burden of DNVs in these 5 genes. To ensure that high-quality variant calls were used in analyses of DNVs, additional QC and filtering were performed. For a variant to be considered de novo, the variant had to be heterozygous in the proband and homozygous reference in both parents. Additionally, the variant had to have an allele balance ratio of 0.3–0.7 in the proband and <0.05 in both parents; it could not be in a known, highly repetitive region of the genome, and it had to have a minor allele frequency (MAF) <0.3% (Itai et al., 2024) in gnomAD (v3.0). Once a list of high-quality DNVs was generated, we tested for enrichment of DNVs both by class of variant (predicted loss-of-function, missense, and protein-altering) and by gene using the package DenovolyzeR in R (Samocha et al., 2014). A P value of <0.05 was considered significant. For CTNND1, CDH1, and CDH3, lollipop plots of variants that were de novo, inherited from an affected parent, or inherited from an unaffected parent were made using ProteinPaint (Zhou et al., 2016).

Fig. S1 shows F-actin distribution and role of actomyosin contractility in wild-type embryos. Fig. S2 shows validation of epithelial-specific p120-catenin mutant. Fig. S3 shows homozygosity of p120-catenin variants. Fig. S4 shows E-cadherin and P-cadherin present in the lip epithelium throughout the fusion process. Fig. S5 shows the loss of E-cadherin does not affect F-actin enrichment in control conditions. Video 1 shows initiation of actin enrichment, related to Fig. 1. Video 2 shows expansion of actin to multicellular cables, related to Fig. 1. Video 3 shows actin zipping of nasal processes, related to Fig. 1. Video 4 shows tissue fusion proceeding in control DMSO treatment, related to Fig. 2. Video 5 shows stagnant tissue fusion with disrupted actomyosin contractility, related to Fig. 2. Video 6 shows tissue fusion in control DMSO treatment, related to Fig. S1. Video 7 shows disrupted tissue fusion with disrupted actomyosin contractility, related to Fig. S1. Table S1 shows details related to Fig. 2, 3, 4, 5, 6 and S3. Table S2 shows variants identified in CDH1, CDH3, CTNND1, MYH9, and MYH10, related to Fig. 7. Table S3 shows analysis of DNVs, related to Fig. 7. Table S4 shows additional methods data.

All data associated with this study are present in the paper or the supplemental information. This paper does not report any code or informatics dataset. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

We thank Scott Oakes for sharing Bax and Bak mutant mice and Trevor Williams for the Crect mouse line; Teng Teng for initiating breeding of Ctnnd1, Cdh1, Bak, and Bax mice; Fang-Shiuan Leung and Emily Burch for mouse colony maintenance support; and Mohammed Fayez Almulla for genotyping and quantification support. We are grateful to Kate Cavenaugh, Matthew Kutys, Licia Selleri, and members of the Bush lab for helpful conversations and suggestions. The Biological Imaging Development CoLab at UCSF was used for whole-mount imaging in Figs. 2 and 4.

This work was supported by NIH funding R35DE031926 to Jeffrey O. Bush and F32DE030367 and K01DE033989 to Camilla S. Teng.

Author contributions: Camilla S. Teng: conceptualization, data curation, formal analysis, funding acquisition, investigation, methodology, validation, visualization, and writing—original draft, review, and editing. Sarah W. Curtis: formal analysis, investigation, and writing—review and editing. Grace C. Brewer: formal analysis, investigation, and writing—review and editing. Elizabeth J. Leslie-Clarkson: conceptualization, funding acquisition, investigation, project administration, resources, and supervision. Jeffrey O. Bush: conceptualization, data curation, formal analysis, funding acquisition, methodology, project administration, resources, supervision, and writing—original draft, review, and editing.