Gene-teratogen interactions influence the penetrance of birth defects by altering Hedgehog signaling strength

Six percent of newborns suffer from structural birth defects, leading to 8 million cases per year worldwide (Christianson et al., 2005). Many of these structural defects require surgical intervention early in life and lead to adverse long-term health consequences. The underlying mechanisms driving birth defects remain unknown in a majority of cases. Complex interactions between genetic and environmental factors are thought to shift morphogen signaling beyond the threshold required for normal developmental patterning (Beames and Lipinski, 2020; Finnell, 1999; Krauss and Hong, 2016). However, in most cases the specific molecular mechanisms remain poorly understood. Penetrance and expressivity of birth defects, both between embryos and between tissues, remains unpredictable and confounds identification of causal factors. Improved understanding of molecular mechanisms is crucial to developing strategies to alleviate the significant public health burden of birth defects.

The Hedgehog (Hh) pathway is one of a handful of signaling systems that regulate developmental patterning and morphogenesis of many tissues, including the face, limbs, heart, lungs, brain and spinal cord (McMahon et al., 2003). Developing tissues are often exquisitely sensitive to the precise amplitude of Hh signaling. Even small changes in signaling strength can cause birth defects in mice and humans (Nieuwenhuis and Hui, 2005). Hh ligands are considered to be classical morphogens: secreted molecules that direct cell-fate choices in a dose-dependent manner (Lee et al., 2016). Temporal and spatial gradients of Hh ligands are translated into intracellular gradients of activity of the GLI transcription factors in target cells (Harfe et al., 2004; Jacob and Briscoe, 2003; Stamataki et al., 2005). Varying Hh signaling strength leads target cells to adopt different cell fates (Dessaud et al., 2008). Given the centrality of morphogen gradients in developmental patterning, considerable research effort has focused on understanding how they are established in tissues. However, this ligand-centric view of patterning is incomplete. Specific signaling mechanisms function in target cells to regulate their sensitivity to morphogens. Indeed, cell fate decisions often depend on both the extracellular concentration of ligands and the reception sensitivity of target cells to these ligands. A prominent example of such a mechanism can be found in the WNT pathway: cell-surface levels of frizzled receptors (and consequently WNT sensitivity) is controlled by the transmembrane E3 ubiquitin ligases ZNRF3 and RNF43 (de Lau et al., 2014). These E3 ligases are themselves controlled by R-spondins, secreted ligands that play central roles in both pattern formation during development and in postnatal tissue homeostasis.

Our analysis of a gene called Mosmo (modulator of Smoothened) led us to uncover a cell-surface pathway that regulates the sensitivity to Hh ligands, and consequently the development of multiple tissues. Mouse genetic analysis revealed that Mosmo uniquely functions to tune the Hh signaling gradient in target cells by promoting the degradation of Smoothened (SMO), a 7-pass transmembrane protein that carries the Hh signal across the plasma membrane. Interestingly, mutations in Mosmo (which increase SMO protein abundance) influence penetrance, expressivity and tissue specificity of birth defects caused by an exogenous teratogen that directly inhibits SMO activity. These findings show that the penetrance of birth defects can be modulated by gene-environment interactions that alter ligand sensitivity in the Hh pathway.

Mosmo (previously named Atthog) is a previously unannotated gene we initially identified in a loss-of-function CRISPR screen conducted in NIH/3T3 fibroblasts designed to find negative attenuators of Hh signaling (Pusapati et al., 2018). Mosmo encodes an 18.2 kDa four-pass transmembrane (4TM) protein that defines a distinct branch of a large superfamily of eukaryotic 4TM proteins that include the claudins, which are known to function at tight junctions. Depletion of MOSMO in cultured cells results in increased accumulation of the Hh transducer SMO on the plasma membrane and the primary cilium membrane, resulting in hyper-responsiveness to Hh ligands. Mosmo is widely expressed in mouse embryos, based on in situ hybridization (Fig. S1A) and the analysis of published single-cell RNAseq data (Pijuan-Sala et al., 2019) (Fig. S1B). To understand the developmental roles of MOSMO, we used CRISPR/Cas9 genome editing to generate mice carrying null alleles of Mosmo (Fig. S2A and S2B). Although Mosmo^+/− mice developed normally, no live Mosmo^−/− pups were recovered from heterozygous intercrosses (Fig. 1A and Table S1). Most Mosmo^−/− embryos die by gestational day 14.5 (E14.5) (Fig. 1A and Table S1). We conclude that the function of MOSMO is essential for embryonic development.

Mosmo deficiency results in developmental defects across many organ systems. Mosmo^−/− embryos have preaxial polydactyly in both forelimbs and hindlimbs (Fig. 1B,C). Whole-mount skeletal staining revealed additional skeletal defects, including a split sternum (Fig. 1C, arrows) and truncated tibia (Fig. 1C, arrowheads). A subset of Mosmo^−/− embryos exhibited exencephaly (Fig. 1D). Detailed necropsy examination of the internal anatomy revealed that all Mosmo^−/− embryos had heterotaxy: discordant patterning of the left-right body axis manifested as abnormalities in lung lobation and abnormal left-right positioning of multiple visceral organs, including the heart, stomach, spleen and pancreas (Table S2). Analysis of the early lung branching pattern indicated that most Mosmo^−/− embryos have either complete or partial right pulmonary isomerism: a duplication of the right lung morphology on the left side (Fig. 1E and Fig. S3A, Table S3). Analysis of the developing heart using episcopic confocal fluorescence microscopy (ECM) revealed that all Mosmo^−/− embryos have complex congenital heart defects (CHDs). The most common CHDs observed are transposition of the great arteries (TGA) and atrioventricular septal defects (AVSDs) (Fig. 1F and Table S2). TGA and AVSDs are classified as ‘critical’ heart defects in humans as they require surgical intervention soon after birth. Mosmo^−/− embryos likely die in utero due to these complex structural heart defects.

To understand the etiology of the birth defects observed in Mosmo^−/− embryos, we focused on the Hh signaling pathway because Mosmo was originally identified as an attenuator of Hh signaling in our CRISPR screens (Pusapati et al., 2018), and many of the Mosmo^−/− phenotypes (i.e. polydactyly and exencephaly) can be caused by elevated Hh signaling (Hui and Joyner, 1993). To assess Hh signaling activity in Mosmo^−/− cells, primary mouse embryonic fibroblasts (pMEFs) were isolated and treated with varying concentrations of Sonic hedgehog (SHH), a secreted ligand that initiates Hh signaling in target cells. Compared with cells from wild-type littermate controls, Hh signaling was elevated in Mosmo^−/− pMEFs (Fig. 2A). A low concentration of SHH (1 nM) that failed to fully activate expression of the Hh target gene Gli1 in wild-type pMEFs was sufficient to maximally activate Gli1 in Mosmo^−/− pMEFs.

To assess levels of Hh signaling activity in Mosmo^−/− embryos, we crossed Mosmo^+/− mice with Gli1^lacZ/+ mice, an extensively used Hh reporter line in which a lacZ transgene (encoding β-galactosidase) was inserted into the first coding exon of Gli1 (Bai et al., 2002). The Gli1-lacZ expression pattern recapitulated endogenous Gli1 expression and the expression of other Hh target genes such as Ptch1 (Goodrich et al., 1997; Guzzetta et al., 2020; Hui et al., 1994). At E8.5 and E9.5, Gli1-lacZ was expressed in the neural tube, somites and secondary heart field (SHF) (Fig. 2B and Fig. S3B) (Guzzetta et al., 2020). At E11.5, Gli1 reporter activity was observed in the brain (telencephalon and diencephalon), spinal cord, limbs, lungs, frontonasal processes and pharyngeal endoderm (Fig. 2B and Fig. S3B). Compared with littermate controls (Mosmo^+/+ and Mosmo^+/−), Gli1-lacZ expression was elevated in Mosmo^−/− embryos at all embryonic ages analyzed. The expansion of Gli1 expression is notable in the developing limb and cardiac outflow tract (Fig. 2B and Fig. S3C), consistent with the finding of polydactyly and outflow tract-related heart defects in Mosmo^−/− embryos (Table S2) (Goddeeris et al., 2007). Consistent with the elevation of Gli1 expression, we observed an increase in PTCH1 (encoded by a direct Hh target gene) and a decrease in GLI3R (the major Hh transcriptional repressor) in Mosmo^−/− whole embryo lysates, showing that Mosmo deficiency results in increased Hh signaling activity in vivo (Fig. 2C).

Hh signal transmission across the plasma membrane requires the SHH-induced accumulation of SMO in the membrane of the primary cilium (Corbit et al., 2005; Rohatgi et al., 2007). The loss of Mosmo led to the constitutive, high-level accumulation of SMO in the ciliary membrane in pMEFs (Fig. 2D) and all embryonic tissues analyzed (Fig. 2E). We also observed an increase in SMO protein abundance in Mosmo^−/− embryo lysates (Fig. 2C). This increase was dramatic for the SMO protein band that migrated more slowly in the SDS-PAGE gel, which represents the population of SMO that has transversed the endoplasmic reticulum (post-ER) and acquired glycan modifications attached in the Golgi. We conclude that Mosmo functions to attenuate Hh signaling strength both in cells and embryos by reducing SMO levels in the ciliary membrane. Re-expression of MOSMO into Mosmo^−/− cells restored both wild-type Hh signaling and ciliary SMO levels (Fig. S4A,B).

The cellular phenotypes (elevated ciliary SMO and sensitivity to SHH) and developmental defects (polydactyly, heterotaxy and CHDs) seen in Mosmo^−/− embryos were reminiscent of those caused by the loss of either Megf8 or Mgrn1 and Rnf157, components of a membrane-tethered E3 ubiquitin ligase complex that ubiquitylates SMO and accelerates its endocytosis and degradation (Kong et al., 2020) (Table 1 and Fig. S4C, Table S2). To determine whether MEGF8 and MOSMO are part of the same pathway, we compared Hh signaling activity and SMO abundance in cells lacking each gene individually (Mosmo^−/− and Megf8^−/− single knockouts) to cells lacking both (Mosmo^−/−;Megf8^−/− double knockouts). The increase in cell-surface SMO and SHH sensitivity seen in Mosmo^−/−;Megf8^−/− double knockout cells was comparable with that seen in Mosmo^−/− and Megf8^−/− single knockout cells (Fig. S4D). Similarly, the birth defects observed in Mosmo^−/−;Megf8^m/m double mutant mouse embryos were comparable in both penetrance and expressivity to Mosmo^−/− and Megf8^m/m single mutant embryos (Table 1 and Table S2). Taken together, analysis in both cells and embryos suggested that Mosmo and Megf8 belong to the same epistasis group, and thus the proteins encoded by these genes likely belong to the same pathway.

A clue to the biochemical function of MOSMO came from the observation that the abundance of cell-surface MEGF8 was reduced in Mosmo^−/− cells. Cell-surface biotinylation analysis demonstrated that the loss of MOSMO reduced MEGF8 (and concomitantly increased SMO) at the plasma membrane; Fig. 3A). These results are consistent with the model that MOSMO promotes ubiquitylation of SMO via the MEGF8-MGRN1 complex by increasing MEGF8 levels at the cell surface. Indeed, co-expression of MOSMO increased the ubiquitylation of SMO by the MEGF8-MGRN1 complex in an assay reconstituted in HEK293T cells (Fig. 3B,C). The influence of MOSMO on MEGF8 activity and localization led us to test the possibility of a physical interaction between the two proteins. Epitope-tagged MOSMO stably expressed in Mosmo^−/− cells (Fig. S4A,B) was co-immunoprecipitated with both endogenous MEGF8 and MGRN1 (Fig. 3D).

The interaction between MEGF8 and MOSMO mapped to the previously unrecognized β-strand-rich MEGF8-Stem (hereafter ‘M-Stem’) domain of MEGF8, positioned at the extracellular end of the transmembrane helix (Fig. 3E,F). Using sensitive deep-learning-based structure prediction methods (Yang et al., 2020), the M-Stem domain is predicted to adopt a novel β-jellyroll topology, but is not closely related to any previously described domain with such a structure (Fig. S5A,B). We propose that the M-Stem domain of MEGF8 docks to the compact extracellular β-sheet surface of MOSMO. As MOSMO is distantly related to the claudins (Pusapati et al., 2018), we modeled this interaction based on the binding of the Clenterotox domain of Clostridium perfringens enterotoxins (which adopt a fold related to CUB domains) to claudin 3, claudin 4 and claudin 19 (Suzuki et al., 2017) (Fig. S5C, left). Taken together, we propose that MOSMO, MEGF8 and MGRN1 together form a membrane-tethered E3 ligase complex (hereafter the ‘MMM complex’, Fig. S5C, right) that modulates the strength of Hh signaling by regulating levels of SMO at the cell surface and primary cilium.

As a component of the MMM complex, MOSMO attenuates Hh signaling activity in the developing embryo (Fig. 2A-C) by clearing SMO from the cell surface and primary cilium (Fig. 2D,E). However, MOSMO may also regulate other cellular pathways and processes. Thus, we sought to investigate whether the developmental defects (i.e. polydactyly and CHDs) observed in Mosmo^−/− embryos (Figs 1 and 3A) were caused by elevated Hh signaling. We took the unconventional approach of using an FDA-approved small-molecule SMO antagonist (vismodegib) to reduce Hh signaling strength at crucial periods of embryonic development. There are many advantages to using small-molecule inhibitors in an embryonic system. First, previous studies have shown that the SMO inhibitors cyclopamine and vismodegib are potent placenta-permeable teratogens that can induce embryonic defects in Hh-dependent tissues when delivered orally to the pregnant mother (Binns et al., 1963; Lipinski et al., 2008). Second, a small-molecule strategy allows us to selectively reduce Hh signaling during defined developmental periods to target events like limb digitation and heart looping. Last, by changing the treatment dose and frequency, a small-molecule inhibitor allows us to experimentally adjust Hh signaling activity as needed, based on the phenotypic outcomes (Heyne et al., 2015). In a proof-of-concept experiment, we found that treatment of pregnant mice with vismodegib for about 2 days (e9.75-e11.5) reduced Gli1-lacZ expression in wild-type embryos when compared with untreated controls (Fig. 4A).

Shh is transiently expressed along the posterior margin from ∼E9.5-E12 in the murine forelimb and ∼E10-E12.5 in the hindlimb (Büscher et al., 1997; Zhu et al., 2008). Hh signaling plays an established role in the anterior-posterior patterning of digits. Oligodactyly (digit loss) can be caused by exposure to a Hh antagonist or to loss of Shh expression during a crucial period of limb development (Heyne et al., 2015; Zhu et al., 2008). Conversely, preaxial polydactyly can arise from elevated Hh signaling activity caused by an increase in Shh or reduction in Gli3 expression (Hill and Lettice, 2013; Hui and Joyner, 1993). To determine whether the fully penetrant preaxial polydactyly observed in the Mosmo^−/− embryos was due to elevated Hh signaling activity, pregnant dams from Mosmo^+/−×Mosmo^+/− crosses were exposed to vismodegib for varying durations of time and E14.5 embryos were collected to examine digit patterning. There were no defects in limb patterning in embryos that received no treatment or embryos that were treated with vismodegib before E9.5 (prior to Shh limb expression). However, when embryos were treated with vismodegib after E9.5, increasing the duration of drug treatment resulted in progressively greater oligodactyly (Fig. 4B). The exquisitely graded nature of Hh signaling was vividly demonstrated by the striking dose-response relationship between vismodegib exposure and digit number. Interestingly, vismodegib treatment had a profoundly different impact on limb patterning, even between embryos in the same litter (Fig. 4B). Although vismodegib caused severe oligodactyly in control (Mosmo^+/+ and Mosmo^+/−) embryos, the same dose often corrected the polydactyly in Mosmo^−/− embryos, resulting in embryos with normal limbs bearing five digits (Fig. 4B,C). We conclude that the polydactyly observed in Mosmo^−/− embryos is indeed due to an elevation of SMO activity as it can be reversed by a direct SMO antagonist (Fig. S6A).

The ability of vismodegib to rescue the polydactyly phenotype led us to test its effects on the complex CHD phenotypes contributing to the lethality of Mosmo^−/− embryos (Fig. 1A,F). Conotruncal heart defects are a group of malformations that arise due to defects in outflow tract development. All of the Mosmo^−/− embryos had conotruncal heart defects, most commonly transposition of the great arteries (TGAs) (Table 1A and Table S2). Hh signaling plays a crucial role in multiple aspects of outflow tract development, including: (1) maintenance of cardiac progenitor proliferation and identity within the secondary heart field (which contributes to the outflow tract) (Dyer and Kirby, 2009; Rowton et al., 2020 preprint); (2) survival of migratory cardiac neural crest cells (Washington Smoak et al., 2005); and (3) proper septation of the outflow tract (Goddeeris et al., 2007). To determine whether the conotruncal heart defects observed in the Mosmo^−/− embryos are due to elevated Hh signaling, we first needed to identify the critical time window during gestation when outflow tract development is sensitive to vismodegib. Loss-of-function mutations in Shh result in a failure of the primitive truncus to properly divide into the aorta and pulmonary artery (persistent truncus arteriosus, PTA) (Washington Smoak et al., 2005). Building on this information, we found that vismodegib administered from E7.25/E8.25 to E11.25 caused PTA in all control (Mosmo^+/+ and Mosmo^+/−) embryos (Fig. 5A,B). In contrast, Mosmo^−/− embryos exposed in utero to the same vismodegib regimen did not develop PTA, suggesting that these mutant embryos were resistant to the SMO antagonist because of elevated SMO abundance (Fig. 5B).

Indeed, vismodegib administration actually improved the CHD phenotypes characteristically seen in Mosmo^−/− embryos. Instead of the predominant TGA phenotype seen in untreated Mosmo^−/− embryos, SMO inhibition shifted the phenotype to DORV, an overlapping conotruncal malformation also associated with defective ventriculoarterial alignment but of reduced severity compared with TGA (Fig. 5A-C and Table S4). Vismodegib treatment in Mosmo^−/− embryos also had a corrective effect on the position of the heart, reducing the incidence of mesocardia, dextrocardia and right aortic arch (RAA). Notably, these defects are all reflective of improper left-right cardiac morphogenesis (Fig. 5A,B and Table S4). Perhaps the most compelling evidence that vismodegib treatment improved Mosmo^−/− heart development and function came from the analysis of embryo survival. The number of Mosmo^−/− embryos that survived to E14.5 was 16-fold higher in vismodegib-treated litters compared with vehicle-treated litters (Fig. 5D and Table S5). We speculate that the incomplete rescue of CHDs may be due to the challenges of delivering the correct dose of vismodegib at the correct time to impact the multiple (temporally different) points in development when Hh signaling is required for heart development.

The loss of MOSMO did not cause defects in all tissues that are known to require SHH for their patterning. Dorsal-ventral patterning of the developing spinal cord is coordinated by a gradient of SHH that is secreted initially from the notochord and later from the floor plate. A large body of literature has shown that neural progenitors adopt different cell fates depending on the magnitude of their exposure to SHH. Increasing the concentration of SHH or the duration of SHH exposure results in an expansion of ventral neural cell fates (Dessaud et al., 2007). Unexpectedly, staining with a panel of cell-type specific markers revealed normal neural tube patterning in E8.5 and E10.5 Mosmo^−/− embryos (Fig. S6B and data not shown). The integrity of neural patterning was maintained despite the fact that the loss of MOSMO resulted in markedly elevated levels of ciliary SMO along the entire dorsal-ventral axis of the developing spinal cord (Fig. S6C). In wild-type embryos, SMO accumulates in the cilia of only the ventral-most progenitor cells (floor plate and p3 progenitors) of the developing spinal cord, cells that are exposed to the highest concentrations of SHH (Kong et al., 2015). However, ciliary SMO levels were elevated in all progenitor domains in Mosmo^−/− embryos, even those distant from the SHH source at the floor plate (Fig. S6C).

While the patterning of ventral spinal progenitors was indistinguishable in control (Mosmo^+/+ and Mosmo^+/−) and Mosmo^−/− embryos, a dramatic difference was uncovered when these embryos were exposed to vismodegib. Similar to previous experiments, pregnant dams from Mosmo^+/−×Mosmo^+/− crosses were exposed to vismodegib from E7.75 to E11.25 and then collected at E11.5 for spinal cord analysis. In control (Mosmo^+/+ and Mosmo^+/−) embryos, vismodegib caused the loss of ciliary SMO (Fig. 6A) and a profound reduction of the ventral-most progenitor domains (OLIG2⁺ and NKX2.2⁺ neural progenitors) that require SHH for their specification (Fig. 6B). These same ventral cell types are also lost in Smo^−/− and Shh^−/− embryos, confirming that vismodegib mimics the genetic disruption of Hh signaling (Chiang et al., 1996; Wijgerde et al., 2002). In contrast, Mosmo^−/− embryos from the same litter (exposed to the same vismodegib regimen), maintained OLIG2⁺ and NKX2.2⁺ progenitors (Fig. 6B-D). Although the OLIG2⁺ and NKX2.2⁺ progenitors were present, they were shifted to more ventral positions within the spinal cord in vismodegib-treated Mosmo^−/− embryos, consistent with partial SMO inhibition. Thus, the loss of a single gene, Mosmo, can influence the impact of an established teratogen (vismodegib) on neural tube patterning, likely by increasing the abundance of SMO, the protein target of the teratogen.

As MOSMO was an uncharacterized protein, we ablated Mosmo in the mouse and discovered that it is essential for embryonic development. In the absence of MOSMO, SMO is enriched in the primary cilia of all tissues, rendering cells hypersensitive to endogenous SHH. Consequently, Mosmo^−/− embryos suffer from severe developmental defects, including heterotaxy, skeletal abnormalities and congenital heart defects (CHDs). The MMM complex composed of MOSMO, MEGF8 and MGRN1 anchors a signaling pathway that regulates the sensitivity of target cells to Hh morphogens (Fig. 7). All three components of the MMM complex were originally identified as attenuators of Hh signaling in our genome-wide CRISPR screens (Pusapati et al., 2018). The phenotypic similarities between Mosmo^−/−, Megf8^−/− and Mgrn1^−/−;Rnf157^−/− cells and mouse embryos supports the model that the MMM proteins function in the same pathway. We previously found that MGRN1 interacts with MEGF8, forming a membrane-tethered ubiquitylation complex that targets SMO for degradation (Kong et al., 2020). Here, we report that MOSMO interacts with MEGF8 to facilitate its accumulation at the cell surface. In general terms, this action of Mosmo might be compared with that of other members of the claudin-like superfamily of 4-pass transmembrane (4TM) proteins to which it belongs. These include the calcium channel γ subunits that play a role in the localization of transmembrane calcium-channel AMPA receptors to the synapse (Chen et al., 2007). More specifically, another related protein, LHFPL4, forms a ternary complex with ionotropic GABA_A receptors and neuroligin 2 (NL2) and helps localize the former receptor to synaptic membranes (Yamasaki et al., 2017). Similarly, another related protein, LHFPL5, forms a complex with the transmembrane channel-like protein isoform 1 (TMC1) in the auditory stereocilia and might play a role in its localization to that structure (Beurg et al., 2015; Yu et al., 2020). Thus, beyond their roles in tight junctions, members of the claudin-like superfamily, including MOSMO, might play a general role in the accumulation of specific complexes at the membrane.

Mosmo is widely expressed in the mouse embryo (Fig. S1) and MOSMO deficiency results in a dramatic increase in ciliary SMO in all tissues we examined (Fig. 2D,E). However, Mosmo^−/− embryos did not show indiscriminate Hh signaling activation in all tissues. The pattern of elevation in Hh signaling activity due to loss of Mosmo is very different from the pattern seen in embryos lacking the well-studied Hh negative regulators PTCH1 and SUFU (Cooper et al., 2005; Goodrich et al., 1997; Svärd et al., 2006). Although Hh signaling is fully activated in most tissues in Ptch1^−/− and Sufu^−/− embryos, Hh signaling is amplified selectively in SHH-exposed tissues in Mosmo^−/− embryos. We propose that the purpose of the MMM complex is to attenuate the gradient of Hh signaling strength in tissues, rather than to suppress basal signaling activity.

The loss of Mosmo clearly has tissue-specific effects. Although Hh signaling is required for the patterning of many tissues, development of the limbs, heart and skeleton was more severely affected by this elevation in pathway activity than the ventral spinal cord. These differences may reflect whether patterning in a tissue depends on transcriptional de-repression or activation. The patterning of the limb bud is driven by the de-repression of downstream target genes due to a loss of GLI transcriptional repressors (GLIR) (Litingtung et al., 2002; te Welscher et al., 2002). In contrast, the patterning of the ventral spinal cord is primarily driven by the activation of downstream target genes by GLI transcriptional activators (GLIA) (Stamataki et al., 2005). We speculate that loss of the MMM complex potentiates Hh signaling by reducing GLIR levels, rather than by elevating GLIA (Kong et al., 2019; Niewiadomski et al., 2014). In support of this notion, Mosmo^−/− mouse embryos have some of the same phenotypes as Gli3^−/− embryos. GLI3 is proteolytically processed to generate GLI3R, the predominant transcriptional repressor in Hh signaling. As seen in Mosmo^−/− embryos, a loss of GLI3 results in polydactyly (Hui and Joyner, 1993; Johnson, 1967), but no changes in the patterning of the ventral spinal cord (Persson et al., 2002). Overall, our results suggest that limb, heart and skeleton development are particularly susceptible to subtle changes in Hh signaling strength caused by either genetic perturbations or environmental exposures to teratogens such as vismodegib. This heightened sensitivity could underlie the sporadic nature of CHDs and contribute to its variable penetrance and expressivity.

Left-right patterning defects, which manifest as heterotaxy with randomization of visceral organ situs, are frequently associated with severe CHDs in humans, suggesting that the signals that specify the left/right body axis also play a role in regulating heart development (Li et al., 2015; Lin et al., 2014). Left-right patterning defects and complex heart malformations are prominent phenotypes common to all the MMM mutant mouse lines (Cota et al., 2006; Zhang et al., 2009). Although vismodegib treatment was able to fully rescue the Mosmo^−/− polydactyly phenotype (Fig. 4), demonstrating that digit duplication is a product of elevated Hh signaling, it failed to fully rescue heterotaxy phenotypes (Fig. 5B). With regard to the heart, vismodegib partially rescued cardiac situs and improved outflow tract malalignment defects, with DORV seen instead of TGA. This finding is intriguing as there is an ongoing debate about whether DORV and TGA are related phenotypes arising from a disturbance in left-right patterning. Our findings suggest that these two phenotypes are indeed developmentally related and may be influenced by changes in Hh signaling strength.

Our failure to rescue heterotaxy phenotypes with vismodegib could be because (1) the MMM complex regulates receptors involved in other signaling pathways or (2) we did not deliver vismodegib during the correct time window in development. Interestingly, a study recently found that the conditional deletion of Megf8 in all known cardiac cell lineages did not reproduce the heart defects observed in the global Megf8 knockout (Wang et al., 2020). These data suggest that Megf8 is required for cardiac development at a time point earlier than cardiac organogenesis and supports the possibility that the heart defects seen in MMM mutant mice are a consequence of disrupted left-right patterning earlier in development. Further studies are needed to identify both the cell types and the critical time periods that are relevant to the function of the MMM complex during development of various tissues.

A central principle of teratology holds that susceptibility to birth defects depends on interactions between the genotype of the embryo and environmental exposures (Finnell, 1999). However, the molecular mechanisms that underlie these gene-teratogen interactions remain largely unknown. Our work on the MMM complex provides one molecular mechanism that explains how mutations and small molecules can interact to influence birth defect outcomes by their combined effects on Hh signaling. Hh ligands function as classical morphogens that direct developmental processes in a manner dependent on the strength of signal in target cells. This simple and elegant mechanism for regulating the patterning and morphogenesis of target tissues, however, also leaves the development of these tissues vulnerable to even small shifts in signaling strength.

Mutations in MMM complex genes cause elevated sensitivity of cells to Hh ligands by increasing the abundance of SMO on the cell surface. Even between embryos in the same litter, Mosmo mutations have a profound influence on the teratogenicity of vismodegib, a direct small molecule SMO antagonist. Vismodegib can cause a variety of structural birth defects: neural tube patterning errors, oligodactyly and cardiac outflow tract abnormalities such as PTA (Figs 4C and 5B). Remarkably, the developing neural tube and cardiac outflow tracts of Mosmo^−/− embryos are resistant to vismodegib exposure compared with wild-type embryos (Figs 5 and 6), likely because the elevated SMO abundance protects these embryos from its teratogenic effects. In the limb and the heart, vismodegib has an even more striking effect on Mosmo^−/− embryos: it rescues structural birth defect phenotypes and improves overall embryo survival, likely by reducing SMO activity to levels that allow normal development. We propose that total SMO activity in target cells, which is influenced by both SMO protein abundance and exposure to a SMO antagonist, determines birth defect outcomes. Elevated SMO abundance in MMM-mutant embryos can be overcome by reducing SMO activity with vismodegib. Conversely, the reduction in SMO activity caused by vismodegib can be overcome by increasing SMO protein abundance. Thus, gene-environment interactions can arise when genetic factors change the abundance of the protein target of a teratogen.

Previous studies have provided evidence that mutations and environmental exposures can influence development of the face and brain by their combined effects on Hh signaling. Ethanol and the pesticide component piperonyl butoxide (PBO) are exogenous agents that suppress Hh signaling. Early embryonic exposure to high concentrations of ethanol and PBO can cause craniofacial abnormalities and holoprosencephaly (an incomplete division of the forebrain, HPE): phenotypes associated with reduced Hh signaling activity (Ahlgren et al., 2002; Everson et al., 2019; Wang et al., 2012). Embryos exposed to low concentrations of ethanol and PBO develop normally. However, even at low concentrations the teratogenic effects of these agents begin to emerge when they are administered to mice carrying mutations in Hh pathway components (Shh^+/−, Gli2^+/− or Cdon^−/− embryos) (Everson et al., 2019; Hong and Krauss, 2017; Kietzman et al., 2014).

In summary, the graded dose-dependent influence of Hh signaling on developmental patterning and morphogenesis explains how gene-teratogen interactions can conspire to modulate the penetrance and expressivity of birth defects by tuning the strength of Hh signaling. A provocative corollary that follows from this idea is that it may be possible to rescue structural birth defect phenotypes by using small molecules (e.g. SMO agonists or antagonists) to re-calibrate Hh signaling strength to the optimal levels required to support normal development. As these drugs are teratogens themselves, they would need to be delivered during defined time periods in development, at precise doses and to embryos of defined genotypes.

Flp-In-3T3 (referred to as ‘NIH/3T3’ cells throughout the text) and HEK293T cell lines were purchased from Thermo Fisher Scientific and the American Type Culture Collection (ATCC), respectively. Information on the gender of the cell lines is not available. As previously described (Kong et al., 2020), NIH/3T3 and HEK293T cells were cultured in Complete Medium: Dulbecco's Modified Eagle Medium (DMEM) containing high glucose (Thermo Fisher Scientific, Gibco) supplemented with 10% fetal bovine serum (FBS) (MilliporeSigma), 2 mM L-glutamine (Gemini Bio-Products), 1 mM sodium pyruvate (Thermo Fisher Scientific, Gibco), 1× MEM non-essential amino acids solution (Thermo Fisher Scientific, Gibco), and penicillin (40 U/ml) and streptomycin (40 µg/ml) (Gemini Bio-Products). The NIH/3T3 and HEK293T cells were rinsed once with sterile PBS and then passaged using 0.05% Trypsin/EDTA (Gemini Bio-Products). Cells were housed at 37°C in a humidified atmosphere containing 5% CO₂. Cell lines and derivatives were free of mycoplasma contamination, as determined by PCR using the Universal Mycoplasma Detection Kit (ATCC).

Primary mouse embryonic fibroblasts (pMEFs) were generated using a modified published protocol (Durkin et al., 2013). Briefly, E13.5 embryos were harvested from Mosmo^+/−×Mosmo^+/− crosses and rinsed thoroughly with sterile PBS. Using forceps, the head and internal organs were removed. The embryos were then separated into individual dishes and the tissue was physically minced into small bits in 0.25% Trypsin/EDTA (Thermo Fisher Scientific, Gibco) using a sterile razor blade. Using initially a 5 ml serological pipette and later a P1000 pipette tip, the minced tissue was pipetted up and down several times to further break up the tissue, and the dishes were placed in a 37°C tissue culture incubator for 10-15 min. If there were still large tissue pieces present, the minced tissue was pipetted further and the dish was placed in the incubator for an additional 5-10 min. The trypsin was then deactivated using Complete Medium (containing 10% FBS). The cells were then centrifuged, resuspended in fresh Complete Medium and plated. Each clonal cell line represents pMEFs generated from a single embryo. Analysis of Mosmo^−/− pMEFs was always performed with pMEFs prepared from Mosmo^+/+ and Mosmo^+/− littermate controls. The gender of the embryos were not determined. Cells were housed at 37°C in a humidified atmosphere containing 5% CO₂.

Hh signaling assays were performed as previously described (Kong et al., 2020). Briefly, NIH/3T3 cells and pMEFs were first grown to confluence in Complete Medium (containing 10% FBS) and then ciliated by changing to Low Serum Medium (Complete Medium containing 0.5% FBS) overnight. NIH/3T3 cells were treated with either no SHH, a low concentration of SHH (1 nM) or a high concentration of SHH (25 nM) prepared in Low Serum Medium. SHH treatment durations varied based on application: 12 h prior to fixation for NIH/3T3 immunofluorescence assays, 24 h prior to lysis for NIH/3T3 western blot assays or NIH/3T3 RNA extraction, and 48 h prior to pMEF experimentation (immunofluorescence, western blot and RNA extraction).

Hh signaling activity was measured using real-time quantitative reverse transcription PCR (qRT-PCR). RNA was extracted from NIH/3T3 cells and pMEFs using TRIzol reagent (Thermo Fisher Scientific, Invitrogen) as previously described (Rio et al., 2010). Equal amounts of RNA were used as a template for cDNA synthesis using the iScript Reverse Transcription Supermix (Bio-Rad Laboratories). qRT-PCR for mouse Gli1 and mouse Gapdh was performed on a QuantStudio 5 Real-Time PCR System (Thermo Fisher Scientific) with the following custom designed primers: mouse Gli1 (fwd, 5′-CCAAGCCAACTTTATGTCAGGG-3′; rev, 5′-AGCCCGCTTCTTTGTTAATTTGA-3′) and mouse Gapdh (fwd, 5′-AGTGGCAAAGTGGAGATT-3′; rev, 5′-GTGGAGTCATACTGGAACA-3′). For all qRT-PCR experiments, Gli1 transcript levels were calculated relative to Gapdh and reported as a fold change across conditions using the comparative C_T method (ΔΔC_T method).

Clonal Mosmo^−/−, Megf8^−/−, Mgrn1^−/− and Mgrn1^−/−;Rnf157^−/−NIH/3T3 lines were previously generated and validated (Kong et al., 2020; Pusapati et al., 2018). Clonal double knockout Mosmo^−/−;Megf8^−/− NIH/3T3 cell lines were generated using the dual sgRNA strategy to target Megf8 in Mosmo^−/− NIH/3T3 cells as previously described (Pusapati et al., 2018). Briefly, sgRNAs targeting Megf8 (5′-TGCCTTCTCTGCCCGAATTG-3′ and 5′-ATAACTTCTCCACGAACACC-3′) were cloned into pSpCas9(BB)-2A-GFP (Addgene) (Ran et al., 2013) and pSpCas9(BB)-2A-mCherry, and transfected into NIH/3T3 cells using X-tremeGENE 9 DNA transfection reagent (Roche Molecular Systems). Five days post transfection, GFP and mCherry double-positive single cells were sorted into a 96-well plate using a FACSAria II at the Stanford Shared FACS Facility. To detect the GFP, a 488 nm (blue) laser was used with a 530/30 filter and B530 detector. To detect the mCherry, a 561 nm (yellow) laser was used with a 616/23 filter and G616 detector. Clonal lines were screened by PCR (forward primer, 5′-CCTCATGCTTGTCCCTTGTT-3′; reverse primer, 5′-GGAGTGTGGGCAAGAAGAAG-3′) to detect excision of the genomic DNA (196 bp) between the two sgRNA cut sites. Knockout of MEGF8 was further confirmed by immunoblotting (Fig. S4D).

Mosmo^−/− NIH/3T3 cells with stable addback of tagged MOSMO (featured in Fig. S4A,B) were generated using the lentiviral expression system as previously described (Kong et al., 2020). Briefly, to generate lentivirus, four million HEK293T cells were seeded onto a 10 cm plate and transfected 24 h later with 1 µg pMD2.G (Addgene), 5 µg psPAX2 (Addgene) and 6 µg of the Mosmo-1D4 pLenti CMV Puro DEST construct using 36 µl of 1 mg/ml polyethylenimine (PEI) (Polysciences). Approximately 48 h post transfection, the lentivirus was harvested and filtered through a 0.45 µm filter. 2 ml of the filtered lentivirus solution was mixed with 2 ml of Complete Medium containing 16 µg/ml polybrene (MilliporeSigma). The diluted virus was then added to NIH/3T3 cells seeded on 6-well plates. Approximately 48 h post infection, cells were split and selected with puromycin (2 µg/ml) for 5-7 days or until all the cells on the control plate were dead.

All mouse studies were conducted using animal study protocols approved by the Institutional Animal Care and Use Committee (IACUC) of Stanford University, the University of Pittsburgh and the McLaughlin Research Institute for Biomedical Sciences. Gli1^tm2Alj mice (referred to in the paper as Gli1^lacz) (MGI:2449767) and Megf8^C193R/C193R mice (referred to in the paper as Megf8^m/m) (MGI:3722325) have been described previously (Bai et al., 2002; Zhang et al., 2009). Gli1^lacz mice were genotyped using the following primers: Fwd (common), 5′-GGGATCTGTGCCTGAAACTG-3′; Rev (wild type), 5′-AGGTGAGACGACTGCCAAGT-3′; and Rev (mutant), 5′-TCTGCCAGTTTGAGGGGACGAC-3′.

Mosmo^+/− mice were generated by the Stanford Transgenic Knockout and Tumor Model Center using CRISPR/Cas9-mediated genome editing. In brief, four gRNAs were designed to delete exon 1 of Mosmo to create a downstream reading frame shift in exon 2 and exon 3, and ultimately remove its function (Fig. S2A). The four guide RNAs (5′-CCGGCGCGCGGTTTCGCTTC-3′, 5′-CGCGGTTTCGCTTCCGGGTG-3′, 5′-CCCCGGGTCGGCGATCCCGA-3′ and 5′-CCCTCGGGATCGCCGACCCG-3′) and CAS9 protein were obtained from Integrated DNA Technologies. A ribonucleoprotein (RNP) injection mix was prepared, consisting of gRNAs (15 ng/μl) and CAS9 protein (30 ng/μl), and introduced into C57BL/6 mouse zygotes via pronuclear microinjection. DNA from the 30 pups born was amplified using primers flanking Mosmo exon 1 and sequenced. One male founder that had a 386 bp deletion, including all of exon 1, was backcrossed with C57BL/6 females for two generations, then heterozygotes were intercrossed. Mosmo knockout mice were genotyped using two sets of primers to detect the wild-type or KO allele: Fwd (wild-type set 1), 5′-GATAAACTGACCATCATCTCAGGATG-3′; Rev (wild-type set 1), 5′-ACTTCAAAGGGGAAAGGGGGAG-3′; Fwd (wild-type set 2), 5′-GGGCGATGGATAAACTGACC-3′; Rev (wild-type set 2), 5′-CGCCTCTTTCTTGAGGACAC-3′; Fwd (mutant set 1), 5′-CCAGTTCCTTCCCATTGCATCT-3′; Rev (mutant set 1), 5′-GCAGTTCAAATACAAGACCGTTCC-3′; Fwd (mutant set 2), 5′-CCGAGAGCTGGGATTCGTAG-3′; and Rev (mutant set 2), 5′-CCACAGACACTTCAAAGGGGA-3′ (Fig. S2B).

We used single-cell transcriptome data from the MouseGastrulationData R package for mouse gastrulation at E7.5 and E8.5. The raw single cell RNA-seq data we analyzed can be found in ArrayExpress under accession number E-MTAB-6967. We used Seurat (version 3.2.3) to analyze the scRNA-seq data (Pijuan-Sala et al., 2019). Our workflow for processing the scRNA-seq data involves data pre-processing, centered log ratio transformation across features, scaling with a linear model, dimensionality reduction and visualization. The annotation of cell types was based on metadata labels included in the data. Cells were plotted based on their euclidean coordinates after a UMAP dimensionality reduction, and Mosmo expression is given in normalized read counts.

The 184 amino acid M-Stem domain from human MEGF8 comprises amino acids 2463-2647, and is tightly sandwiched between an EGFL module and the hydrophobic transmembrane helix. The Hhpred program in the MPI bioinformatics toolkit (Gabler et al., 2020) was used to define the boundaries and β-sheet nature of this elusive domain by sequence and structural profile matching to MEGF8 orthologs, and Attractin (ATRN) and Attractin-like 1 (ATRNL1) paralogs. The structure of the isolated MEGF8 M-Stem domain (where a 32 residue, disordered Pro and Gly rich insert, amino acids 2530-2562, relative to a compact ATRN loop, was replaced by a Gly) was predicted and modeled by trRosetta based on its de novo folding algorithm in a template-free fashion, guided by deep learning-derived restraints of residue distances and orientations (Yang et al., 2020). The confidence of the predicted model is high with an estimated TM-score of 0.567. Similar folds present in other PDB structures were revealed by PDBeFOLD searches (Krissinel and Henrick, 2004) with the trRosetta-derived models of MEGF8 and related ATRN and ATRNL1 Stem domains. Residue conservation profiles were mapped to the Stem domain structure with the ConSurf program (Ashkenazy et al., 2016).

MEGF8-1D4, MEGF8ΔN-1D4, MEGF8ΔC-1D4, Mgrn1-3xFLAG, Mgrn1^Mut1-3xFLAG, Smo-EGFP and Mosmo-1D4 have been previously described (Kong et al., 2020; Pusapati et al., 2018). Mosmo-3xHA was synthesized as a gBlock (Integrated DNA Technologies) and used as a template for the PCR amplification step. To generate MEGF8ΔN^Stem, MEGF8 (NM_001410.3) nucleotide sequence coding for amino acids 2313-2778 was PCR amplified using full-length MEGF8 as a template. All constructs were cloned initially into the pENTR2B plasmid (Thermo Fisher Scientific, Invitrogen) and then transferred into pEF5/FRT/V5-DEST (Thermo Fisher Scientific, Invitrogen) or pLenti CMV PURO DEST (Campeau et al., 2009) using Gateway recombination methods (Thermo Fisher Scientific, Invitrogen).

Recombinant SHH was expressed in bacteria and purified in the lab as previously described (Bishop et al., 2009). Briefly, His-tagged SHH-N (C24II followed by human SHH amino acids 25-193) was expressed in Escherichia coli [BL21 strain; Rosetta2 (DE3)pLysS]. Cells were lysed in 10 mM phosphate buffer (pH 7.5), 500 mM NaCl, 1 mM 2-mercaptoethanol, 1 mM PMSF and 1× protease inhibitor cocktail, followed by centrifugation at 20,000 g for 30 min at 4°C. Clarified samples were incubated with Ni-NTA resin (Qiagen) for 1 h at 4°C. The resin was washed with 20 column volumes of wash buffer A (lysis buffer without protease inhibitors), followed by wash buffer B (wash buffer A+10 mM Imidazole), and bound proteins were eluted with elution buffer (wash buffer A+250 mM Imidazole). Peak fractions were pooled, concentrated using a 5 kDa cut-off VIVASPIN 15R (Life Technologies), and loaded onto a Superdex 75 gel filtration column (Amersham Biosciences) equilibrated with column buffer [10 mM HEPES (pH 7.5), 150 mM NaCl and 1 mM DTT]. The recombinant protein was >98% pure, as assessed from Coomassie staining, and stored at −80°C. The selection antibiotic puromycin was purchased from MilliporeSigma. The transfection reagent XtremeGENE 9 was purchased from Roche Molecular Systems and polybrene from MilliporeSigma. Bafilomycin A1 was purchased from Cayman Chemical. Vismodegib and Bortezomib were purchased from LC labs. The following primary antibodies were purchased from the following vendors: mouse anti-1D4 (The University of British Columbia, 56504; 1:5000); rat anti-E-cadherin (clone ECCD-2, Thermo Fisher Scientific, 13-1900; 1:1000); mouse anti-FLAG (clone M2, MilliporeSigma, F1804; 1:2000); goat anti-GFP (Rockland Immunochemicals, 600-101-215; 1:1000); rabbit anti-GFP (Novus Biologicals, NB600-308; 1:5000); mouse anti-GLI1 (clone L42B10, Cell Signaling, 2643; 1:1000); mouse anti-HA (clone 2-2.2.14, Thermo Fisher Scientific, 26183; 1:2000); rabbit anti-p38 (Abcam, ab7952; 1:2000); and rabbit anti-RNF156 (anti-MGRN1, Proteintech, 11285-1-AP; 1:500); mouse anti-ɑ-Tubulin (Clone DM1A, MilliporeSigma, T6199; 1:10,000); mouse anti-acetylated-Tubulin (MilliporeSigma, T6793; 1:10,000). The following primary antibodies were generated in the lab or received as a gift: guinea pig anti-ARL13B (1:1000) (Dorn et al., 2012); rabbit anti-SMO (designed against an intracellular epitope, 1:2000) (Rohatgi et al., 2007); and rabbit anti-MEGF8 (1:2000) (Kong et al., 2020). Hoechst 33342 and secondary antibodies conjugated to horseradish peroxidase (HRP) or Alexa Fluor dyes were obtained from Jackson ImmunoResearch Laboratories and Thermo Fisher Scientific as follows: Peroxidase AffiniPure donkey anti-mouse IgG (Jackson ImmunoResearch Laboratories, 715-035-150, 1:10,000); Peroxidase AffiniPure donkey anti-rabbit IgG (Jackson ImmunoResearch Laboratories, 111-035-144, 1:10,000); Peroxidase AffiniPure donkey anti-goat IgG (Jackson ImmunoResearch Laboratories, 705-035-003, 1:10,000); donkey anti-rabbit IgG (H+L) Highly Cross-Adsorbed Secondary Antibody, Alexa Fluor 488 (Thermo Fisher Scientific, A-21206, 1:1000); donkey anti-rabbit IgG (H+L) Highly Cross-Adsorbed secondary antibody, Alexa Fluor 594 (Thermo Fisher Scientific, A-21207, 1:1000); donkey anti-mouse IgG (H+L) secondary antibody, Alexa Fluor 488 (Thermo Fisher Scientific, A-21202, 1:1000); donkey anti-mouse IgG (H+L) secondary antibody, Alexa Fluor 647 (Thermo Fisher Scientific, A-31571, 1:1000); Alexa Fluor 488 AffiniPure donkey anti-guinea Pig IgG (H+L) (Jackson ImmunoResearch Laboratories, 706-545-148, 1:1000); and Alexa Fluor 647 AffiniPure donkey anti-guinea pig IgG (H+L) (Jackson ImmunoResearch Laboratories, 706-605-148, 1:1000).

Whole-cell extracts from HEK293T and NIH/3T3 cells were prepared in immunoprecipitation (IP) lysis buffer: 50 mM Tris (pH 8.0), 150 mM NaCl, 1% NP-40, 0.25% sodium deoxycholate, 1 mM DTT and 1× SIGMAFAST protease inhibitor cocktail (MilliporeSigma). Cells were lysed for 1 h on a shaker at 4°C, supernatants were clarified by centrifugation (20,000 g for 30 min at 4°C), and 1D4 tagged MOSMO or MEGF8 was captured by a 1D4 antibody (The University of British Columbia) covalently conjugated to Protein A Dynabeads (Thermo Fisher Scientific, Invitrogen). Immunoprecipitates were washed once with IP Wash Buffer A [50 mM Tris (pH 8.0), 150 mM NaCl, 1% NP-40, 0.25% sodium deoxycholate and 1 mM DTT], once with IP Wash Buffer B [50 mM Tris (pH 8.0), 500 mM NaCl, 0.1% NP-40, 0.25% sodium deoxycholate and 1 mM DTT], and finally with IP Wash Buffer C [50 mM Tris (pH 8.0), 0.1% NP-40, 0.25% sodium deoxycholate and 1 mM DTT]. Proteins were eluted by resuspending samples in 2× NuPAGE LDS sample buffer (Thermo Fisher Scientific, Invitrogen) supplemented with 100 mM DTT, incubated at 37°C for 1 h and subjected to SDS-PAGE (Fig. 3D,F).

Whole-cell extracts were prepared in RIPA lysis buffer: 50 mM Tris (pH 8.0), 150 mM NaCl, 2% NP-40, 0.25% sodium deoxycholate, 0.1% SDS, 0.5 mM TCEP, 10% glycerol, 1× SIGMAFAST protease inhibitor cocktail (MilliporeSigma) and 1× PhosSTOP phosphatase inhibitor cocktail (Roche). The resolved proteins were transferred onto a nitrocellulose membrane (Bio-Rad Laboratories) using a wet electroblotting system (Bio-Rad Laboratories) followed by immunoblotting.

For the preparation of whole-embryo extracts, e12.5 embryos were collected and rinsed thoroughly in chilled PBS. Each embryo was then individually submerged in liquid nitrogen and pulverized using a mortar and pestle. The crushed tissue was then lysed in modified RIPA lysis buffer: 50 mM Tris (pH 8.0), 150 mM NaCl, 1% NP-40, 0.25% sodium deoxycholate, 0.1% SDS, 5 mM EDTA, 1 mM sodium fluoride, 1 mM sodium orthovanadate and 1× SIGMAFAST protease inhibitor cocktail (MilliporeSigma). The resolved proteins were then transferred onto a nitrocellulose membrane (Bio-Rad Laboratories) using a wet electroblotting system (Bio-Rad Laboratories) and immunoblotted.

Cell surface levels of MEGF8 (Fig. 3A) were determined by a biotinylation assay as described previously (Kong et al., 2020). Briefly, wild-type, Mosmo^−/− and Megf8^−/− NIH/3T3 cells were plated on 10 cm plates in Complete Medium. Once the cells were confluent, they were switched to Low Serum Medium for 24 h. On biotinylation day, the cells were removed from the 37°C incubator and placed on an ice-chilled metal rack in a 4°C cold room. The medium was removed and cells were quickly washed three times with ice-cold DPBS+ buffer (Dulbecco's PBS supplemented with 0.9 mM CaCl₂, 0.49 mM MgCl₂.6H₂O, 5.6 mM dextrose and 0.3 mM sodium pyruvate). Biotinylation of cell surface proteins using a non-cell permeable and thiol-cleavable probe was initiated by incubating cells with 0.4 mM EZ-Link Sulfo-NHS-SS-Biotin (Thermo Fisher Scientific) in DPBS+ buffer for 30 min. Unreacted Sulfo-NHS-SS-Biotin was quenched with 50 mM Tris (pH 7.4) for 10 min. Cells were then washed three times with 1× Tris-buffered saline [25 mM Tris (pH 7.4), 137 mM NaCl and 2.7 mM KCl] and whole-cell extracts were prepared in Biotinylation Lysis Buffer A [50 mM Tris (pH 8.0), 150 mM NaCl, 2% NP-40, 0.25% sodium deoxycholate, 1x SIGMAFAST protease inhibitor cocktail (MilliporeSigma) and 1×PhosSTOP phosphatase inhibitor cocktail (Roche)]. Biotinylated proteins from clarified supernatants were captured on a streptavidin agarose resin (TriLink Biotechnologies), washed once with Biotinylation Lysis Buffer A, once with Biotinylation Wash Buffer A (Biotinylation Lysis Buffer A+0.5% SDS), once with Biotinylation Wash Buffer B (Biotinylation Wash Buffer A+150 mM NaCl) and finally once again with Biotinylation Wash Buffer A. Biotinylated proteins captured on streptavidin agarose resin were eluted in 1× NuPAGE-LDS sample buffer (Thermo Fisher Scientific, Invitrogen) containing 100 mM DTT at 37°C for 1 h and assayed by immunoblotting for MEGF8.

Ubiquitylation assays were performed as previously reported (Kong et al., 2020). Briefly, 8 million HEK293T cells were plated onto a 15 cm plate. 24 h after plating, the cells were transfected using PEI. 6 μg of each construct was transfected into the HEK293T cells at a DNA:PEI ratio of 1:3. An empty plasmid construct was used as filler DNA to ensure that each plate was transfected with the same amount of DNA. To enrich for ubiquitylated proteins, 36 h post-transfection, cells were pre-treated with 10 µM Bortezomib (a proteasome inhibitor) and 100 nM Bafilomycin A1 (a lysosome inhibitor) for 4 h. Cells were washed twice with chilled PBS and lysed in Ubiquitylation Lysis Buffer A comprising 50 mM Tris at pH 8.0, 150 mM NaCl, 2% NP-40, 0.25% sodium deoxycholate, 0.1% SDS, 6 M urea, 1 mM DTT, 10 µM Bortezomib, 100 nM Bafilomycin A1, 20 mM N-Ethylmaleimide (NEM, MilliporeSigma) and 1×SIGMAFAST protease inhibitor cocktail (MilliporeSigma). Clarified supernatants were diluted tenfold with Ubiquitylation Lysis Buffer B (Ubiquitylation Lysis Buffer A prepared without urea) to adjust the urea concentration to 600 mM. For these assays, we assessed ubiquitylation on GFP-tagged SMO. Ubiquitylated GFP tagged SMO (Fig. 3B,C) was captured using a GFP-binding protein (GBP) covalently conjugated to carboxylic acid decorated Dynabeads (Dynabeads M-270 carboxylic acid, Thermo Fisher Scientific). Immunoprecipitates were washed once with Ubiquitylation Wash Buffer A (Ubiquitylation Lysis Buffer B+0.5% SDS), once with Ubiquitylation Wash Buffer B (Ubiquitylation Wash Buffer A+1 M NaCl), and finally once again with Ubiquitylation Wash Buffer A. Proteins bound to dynabeads were eluted in 2× NuPAGE-LDS sample buffer (Thermo Fisher Scientific, Invitrogen) containing 30 mM DTT at 37°C for 1 h and assayed by immunoblotting.

Mouse tissue was prepared for immunofluorescence imaging as previously described (Kong et al., 2020). Briefly, mouse embryos of various ages were harvested and fixed in 4% (w/v) paraformaldehyde (PFA) in PBS at 4°C on a nutator. Fixation time varied depending on the age of the embryo (30 min for E8.5-E9.5, 1 h for E10.5-E11.5 and 2 h for E12.5). The embryos were then rinsed thoroughly in chilled PBS. To cryopreserve the tissue, the embryos were transferred to 30% sucrose in 0.1 M PB (pH 7.2) and allowed to equilibrate overnight. The embryos were then carefully dissected, then the desired tissues were mounted and frozen into Tissue-Plus OCT (optimal cutting temperature) compound (Thermo Fisher Scientific) and 12-14 µm sections were collected on a Leica CM1800 cryostat. Prior to staining, the tissues were blocked for 1 h at room temperature in immunofluorescence (IF) Blocking Buffer: 1% normal donkey serum (NDS) and 0.1% Triton-X diluted in PBS. In a humidified chamber, the sections were then incubated with primary antibodies overnight prepared in IF Blocking Buffer at 4°C, rinsed three times in PBST (PBS+0.1% Triton-X), incubated with secondary antibodies and Hoechst prepared in IF Blocking Buffer for 1 h at room temperature, rinsed three times in PBST, and then mounted in ProLong Gold antifade mountant (Thermo Fisher Scientific, Invitrogen).

NIH/3T3 cells and pMEFs were fixed in chilled 4% (w/v) PFA in PBS for 10 min and then rinsed thoroughly with chilled PBS. Cells were incubated in IF Blocking Buffer for 30 min, primary antibodies for 1 h and secondary antibodies for 30 min. Coverslips were mounted in ProLong Gold antifade mountant (Thermo Fisher Scientific, Invitrogen).

Fluorescent images were acquired on an inverted Leica SP8 confocal microscope equipped with a 63× oil immersion objective (NA 1.4). Z-stacks (∼4 µm sections) were acquired with identical acquisition settings (laser power, gain, offset, frame and image format) within a given experiment. An 4-8× optical zoom was used for imaging cilia to depict representative images. For the quantification of SMO at cilia, images were opened in Fiji (Schindelin et al., 2012) with projections of the maximum fluorescent intensities of z-stacks. Ciliary masks were constructed based on ARL13B images and then applied to corresponding SMO images to measure the fluorescence intensity of SMO at cilia.

Vismodegib treatment was performed as described previously (Heyne et al., 2015). Briefly, Mosmo^+/−×Mosmo^+/− and wild-type×Gli1^lacz/+ mouse crosses were set up and monitored daily. Time E0 was defined as midnight prior to the visualization of the copulation plug. Female mice were weighed at ∼E0.25 (the morning the plug was visualized) and ∼E7.25. Only mice that gained 1.75 g over 7 days were deemed ‘likely pregnant’ and treated with either vehicle or vismodegib. For vismodegib treatment, a 3 mg/ml vismodegib solution was prepared in 0.5% methyl cellulose (MilliporeSigma) with 0.2% Tween. Vismodegib (40 mg/kg) was administered via oral gavage every 12 h (∼7am and 7pm) for a total of 3 days (E8.25-E11.25) or 4 days (E7.25-E11.25) (Figs 4C, 5A-D and Table S4). Embryos were harvested at E14.5, fixed in 4% (w/v) PFA in PBS for at least 24 h and then analyzed.

Mouse embryo phenotyping was performed as described previously (Kong et al., 2020). Briefly, mouse embryos (E14.5) were fixed in 4% (w/v) PFA in PBS for at least 24 h. Necropsy was performed to determine visceral organ situs (i.e. lung and liver lobation, heart directionality, and positioning of the stomach, spleen and pancreas) (Table 1, Tables S2 and S4). For analysis by episcopic confocal microscopy (ECM), the samples were embedded in paraffin and processed as previously described (Liu et al., 2013). Briefly, the tissue block was sectioned using a Leica sledge microtome and serial images of the block face were captured with a Leica confocal microscope. The serial two-dimensional (2D) image stacks generated were then three-dimensionally (3D) reconstructed using Osirix software (Rosset et al., 2004) and digitally resliced in different orientations to aid in the analysis of intracardiac anatomy and the diagnosis of congenital heart defects (Liu et al., 2013) (Figs 1F and 5A).

Whole-mount skeletal stains were prepared using a modified published protocol (McLeod, 1980; Rigueur and Lyons, 2014). Briefly, E15.5-E16.5 mouse embryos were harvested and the skin and internal organs were removed to facilitate tissue permeabilization. The embryos were then fixed first in 95% ethanol overnight at room temperature followed by 100% acetone overnight at room temperature. To stain the cartilage, the embryos were incubated overnight at room temperature in 0.03% (w/v) Alcian Blue 8GX (Millipore Sigma) prepared in a solution of 80% ethanol and 20% glacial acetic acid. After visually confirming that the embryos were completely blue, the embryos were destained in a series of ethanol washes (2-3 h in 100% ethanol, 75% ethanol, 50% ethanol and 25% ethanol). To stain the bone, the embryos were then incubated overnight at 4°C in 0.005% (w/v) Alizarin Red S (Millipore Sigma) prepared in 1% potassium hydroxide (KOH). The tissue was then cleared in 0.3% KOH for 1 to 3 days (changing the solution every day). Once the embryos cleared to the desired amount, the 0.3% KOH was replaced with glycerol. The embryos were transitioned through a series of glycerol solutions (20% for 1 day, 50% for 1 day and then 80% for 1 day). The skeletons were then kept in 80% glycerol for prolonged storage.

Whole-mount lungs were prepared as previously described (Metzger et al., 2008). Briefly, E12.5 mouse embryos were harvested and the lungs were carefully excised. The lungs were fixed in 4% (w/v) PFA in PBS for 1 h and then rinsed thoroughly in PBS at room temperature. The lungs were dehydrated in a series of methanol washes: once in 25% methanol/PBS (v/v), once in 50% methanol/PBS, once in 75% methanol/PBS and twice in 100% methanol. The dehydrated lungs were then bleached for 15 min in 5% H₂O₂/methanol at room temperature and then rehydrated in a series of PBT washes (PBS with 0.1% Tween-20): once in 75% methanol/PBS, once in 50% methanol/PBT, once in 25% methanol/PBT and thrice in 100% PBT. The lungs were blocked for 1 h at room temperature in Whole-mount (WM) Blocking Buffer: 5% donkey serum and 0.5% Triton X-100 diluted in PBS. For primary antibody labeling, the lungs were incubated overnight at 4°C in rat anti-E-cadherin antibody (clone ECCD-2, Thermo Fisher Scientific, 13-1900) diluted 1:1000 in WM Blocking Buffer. The following day, the lungs were thoroughly rinsed in PBT (8×30 min). For secondary antibody labeling, the lungs were incubated overnight at 4°C in biotin-conjugated donkey anti-rat IgG (Jackson ImmunoResearch Laboratories) diluted 1:250 in WM Blocking Buffer. The lungs were then thoroughly rinsed in PBT (8×30 min), the biotin was visualized using the VECTASTAIN Elite ABC Kit (Vector, PK-16100) and the signal was amplified using a Tyramide Signal Amplification System (Cy3, Perkin Elmer). Stained lungs were mounted in Vectashield with DAPI (Vector) and imaged on a Thunder Imager Model Organism (Leica) (Fig. 1E and Fig. S3A).

Mouse embryos were processed for β-galactosidase staining using a modified published protocol (Nagy et al., 2007). Briefly, embryos were harvested from Mosmo^+/−×Mosmo^+/−;Gli1^lacZ/+ mouse crosses and fixed at 4°C in 4% (w/v) PFA in PBS for varying durations of time depending on their age (E8.5 and E9.5 for 10 min, E10.5 for 12 min, E11.5 for 15 min and E12.5 for 20 min). The embryos were then rinsed thoroughly in PBS and permeabilized for either 2 h (≤E9.5) or overnight (≥E10.5) at 4°C in permeabilization solution: 0.02% sodium deoxycholate and 0.01% NP-40 diluted in PBS. Following permeabilization, the embryos were placed in staining solution: 1 mg/ml X-gal (Thermo Fisher Scientific), 2 mM MgCl₂, 0.02% NP-40, 5 mM potassium ferricyanide, 5 mM potassium ferrocyanide, 0.01% sodium deoxycholate diluted in 0.1 M phosphate buffer (pH 7.2). The embryos were stained for 2 h at 37°C. To remove residual yellow color from the staining solution, the embryos were rinsed in permeabilization solution (2×15 min). The embryos were fixed overnight in 4% (w/v) PFA in PBS at 4°C, rinsed in PBS, and then imaged.

As previously described (Pusapati et al., 2018), to generate a Mosmo in situ probe, Mosmo specific primers were designed using the program Primer3: forward 5′-acacgtgtgtgctgaaaagc-3′ and reverse 5′-gagattaaccctcactaaagggatgagcaggtaacccatctcc-3′. The underlined sequence marks the T3 polymerase binding site that was incorporated into the reverse primer. The Mosmo probe was generated using a digoxigenin (DIG) RNA Labeling Kit (Roche). Briefly, the probe was generated from the in vitro transcription of PCR products amplified from mouse neural progenitor cell cDNA. After overnight hybridization at 65°C, the signal was visualized using Anti-DIG-alkaline phosphatase (AP) Fab fragments (Roche) and NBT/BCIP (Roche).

Most data were analyzed using GraphPad Prism 9. Violin plots (Fig. 2D and Fig. S4A) were created using the ‘Violin plot (truncated)’ appearance function. In Prism 9, the frequency distribution curves of the violin plots are calculated using kernal density estimation. By using the ‘truncated’ violin plot function, the frequency distributions shown are confined within the minimum to maximum values of the data set. On each violin plot, the median (central bold line) and quartiles (adjacent thin lines, representing the first and third quartiles) are labeled. In Prism 9, the statistical significance between two groups was determined using either Mann–Whitney (Fig. 2D) or an unpaired t-test (Fig. 6C,D) and the significance between three or more groups was determined using the Kruskal–Wallis test (Fig. 4C, Fig. S4A). For each of these figures, P-values were calculated using Prism 9 and reported in the figure legend using the following key: not-significant (ns)>0.05, *P<0.05, **P<0.01, ***P<0.001 and ****P<0.0001 Additional figure details regarding the n value and statistical test applied are reported in the individual figure legends.

Disruptions in offspring viability due to genotype (Fig. 1A and Table S1) or treatment (Fig. 5D and Table S5) were determined using the chi-squared test. Briefly, Mosmo^+/−×Mosmo^+/− crosses were set up, live embryos were collected and deviation from the expected Mendelian ratio of 1:2:1 was calculated [not-significant (ns)>0.05, **P<0.01 and ***P<0.001].

All cell biological and biochemical experiments were performed two to three independent times, with similar results. To validate Mosmo^−/− primary mouse embryonic fibroblasts (pMEFs), three independent cell lines were generated (each from a single embryo) and compared against control (Mosmo^+/+ and Mosmo^+/−) pMEFs generated from embryos within the same litter (Fig. 2A). Similarly, two whole embryo lysate samples were prepared (each from a single Mosmo^−/− embryo) and compared against control (Mosmo^+/+ and Mosmo^+/−) lysates prepared from embryos within the same litter (Fig. 2C).

We thank Jill Helms, Xue Yuan and Bo Liu for their assistance with the skeletal stains; Alex Joyner for fruitful discussions; Derek Silvius and Janet Peters for genotyping; and the McLaughlin Research Institute Animal Resource staff for animal care.