A ubiquitin-based mechanism for the oligogenic inheritance of heterotaxy and heart defects

The etiology of congenital heart defects (CHDs), amongst the most common human birth defects, is poorly understood partly because of its complex genetic architecture. Here we show that two genes previously implicated in CHDs, Megf8 and Mgrn1, interact genetically and biochemically to regulate the strength of Hedgehog signaling in target cells. MEGF8, a single-pass transmembrane protein, and MGRN1, a RING superfamily E3 ligase, assemble to form a transmembrane ubiquitin ligase complex that catalyzes the ubiquitination and degradation of the Hedgehog pathway transducer Smoothened. Homozygous Megf8 and Mgrn1 mutations increased Smoothened abundance and elevated sensitivity to Hedgehog ligands. While mice heterozygous for loss-of-function Megf8 or Mgrn1 mutations were normal, double heterozygous embryos exhibited an incompletely penetrant syndrome of CHDs with heterotaxy. Thus, genetic interactions between components of a receptor-like ubiquitin ligase complex that tunes morphogen signaling strength can cause a birth defect syndrome inherited in an oligogenic pattern.


Introduction
Morphogens are secreted ligands that influence developmental outcomes (differentiation, patterning or morphogenesis) in a dose-dependent manner. For example, temporal and spatial gradients of Hedgehog (Hh) ligands (like Sonic Hedgehog, SHH) pattern the spinal cord and limb during development. Varying concentrations or durations of morphogen exposure produce different cellular outcomes by changing the strength or persistence of signaling in target cells (Harfe et al., 2004;Stamataki et al., 2005). Much of the focus in morphogen signaling has been on understanding how ligands like SHH are produced and distributed across tissues to form gradients. However, signaling strength in target cells is a function of both ligand exposure and ligand sensitivity. Much less is known about mechanisms that function in target cells to modulate ligand reception and whether such mechanisms are damaged in developmental diseases.
In recent CRISPR screens for regulators of Hh signaling, we discovered several proteins that function in target cells to attenuate signaling strength. Because of similarities in their lossof-function phenotypes, we focus here on three of these proteins: Multiple Epidermal Growth Factor-like Domains 8 (MEGF8), a type I single-pass transmembrane protein, and two paralogous RING superfamily E3 ubiquitin ligases, Mahogunin Ring Finger 1 (MGRN1) and RNF157. Megf8 was identified as a regulator of both left-right patterning and cardiac morphogenesis in mouse genetic screens (Aune et al., 2008;Li et al., 2015;Zhang et al., 2009).
Human mutations in MEGF8 result in Carpenter syndrome, an autosomal recessive syndrome similarly characterized by heterotaxy (defects in left-right patterning), severe congenital heart defects (CHDs), preaxial digit duplication, and skeletal defects (Twigg et al., 2012). Unlike many other genes associated with heterotaxy, loss of Megf8 does not result in any detectable defects in either primary or motile cilia (Aune et al., 2008;Pusapati et al., 2018a;Zhang et al., 2009).
Loss of MGRN1 was also previously shown to cause CHDs and heterotaxy with partial penetrance in mice (Cota et al., 2006). How MEGF8 and MGRN1 regulate these critical developmental events has remained unknown for over a decade.
We investigated the biochemical and biological functions of MEGF8, MGRN1 and RNF157 using a combination of mechanistic studies in cultured cells and mouse genetics.
MEGF8, MGRN1 and RNF157 anchor a novel ubiquitination pathway that regulates the sensitivity of target cells to Hh ligands. They assemble into an unusual transmembrane E3 ubiquitin ligase complex that functions as a traffic control system for signaling receptors, including the Hh transducer Smoothened (SMO). Mouse studies revealed striking genetic interactions and gene dosage effects involving Megf8, Mgrn1 and Rnf157 that impact the penetrance of a wide spectrum of birth defects, including congenital heart defects (CHDs), heterotaxy, skeletal defects, and limb anomalies. Our work shows how genetic interactions can arise from biochemical mechanisms that regulate signaling strength, a conclusion with broader implications for understanding the molecular pathophysiology of human diseases in which oligogenic interactions are emerging as an important mechanism of heritability.

Megf8 and Mgrn1 are negative regulators of Hedgehog signaling
Amongst the top gene hits identified in our genome-wide screen for attenuators of Hh signaling , we pursued a detailed analysis of Megf8 and Mgrn1 because of similarities in their loss of function phenotypes (Fig. S1A). In both NIH/3T3 fibroblasts and cultured neural progenitor cells (NPCs), CRISPR-mediated loss-of-function mutations in Megf8 and Mgrn1 resulted in an elevated response to Sonic hedgehog (SHH) ligands caused by the accumulation of SMO at the cell surface and primary cilium . To determine if MEGF8 and MGRN1 can attenuate Hh signaling in a more physiological context, we isolated primary mouse embryonic fibroblasts (pMEFs) from embryos homozygous for previously characterized mutant alleles of Megf8 (C193R) or Mgrn1 (md-nc) (Fig. S1A) (He et al., 2003;Phillips, 1963;Zhang et al., 2009). As we observed in NIH/3T3 cells, Megf8 C193R/C193R and Mgrn1 md-nc/md-nc (hereafter referred to as Megf8 m/m and Mgrn1 m/m ) pMEFs were more sensitive to SHH. When exposed to a sub-saturating concentration of SHH (1 nM), Gli1 (a direct Hh target gene) was only partially induced in wild-type pMEFs, but this same low concentration induced Gli1 to maximum levels in Megf8 m/m and Mgrn1 m/m pMEFs (Figs. S1B and S1C).
Heightened SHH sensitivity was likely caused by an elevated abundance of SMO in the primary cilia of Megf8 m/m and Mgrn1 m/m pMEFs, both in the absence of SHH and when signaling was activated by pathway agonists (Figs. S1D and S1E).
The accumulation of ectopic ciliary SMO was also observed in multiple tissues within Megf8 m/m and Mgrn1 m/m embryos (Fig. S2). In wild-type embryos, Hh signaling activity is restricted to early embryonic development and by e12.5 is turned off in most tissues, resulting in ciliary SMO restricted to cells that were exposed to only the highest concentrations of SHH, like the progenitor cells within the ventral neural tube (Fig. S2) (Corbit et al., 2005;Rohatgi et al., 2007a). In contrast, SMO was concentrated in the primary cilia of nearly all Megf8 m/m embryonic tissues, regardless of whether it had been exposed to Hh ligands (Fig. S2). Tissues from Mgrn1 m/m embryos did not have widespread accumulation of ciliary SMO (Fig. S2). However, ciliary SMO was inappropriately present sporadically in the dorsal neural tube and brain of Mgrn1 m/m embryos (Fig. S2), consistent with our observation that Mgrn1 -/-NPCs exhibited a moderately elevated response to SHH . The difference in the severity of the Hh signaling phenotype between Megf8 m/m and Mgrn1 m/m pMEFs and embryos (Figs. S1 and S2) is consistent with our previous results comparing Mgrn1 -/and Megf8 -/-NIH/3T3 cells . As reported previously, the loss of MEGF8 or MGRN1 did not cause any defects in the number or the structure of primary cilia (Pusapati et al., 2018b;Zhang et al., 2009).

Rnf157 is a genetic modifier of Mgrn1
In both mice and cultured cells, loss of MEGF8 consistently produced stronger phenotypes than the loss of MGRN1 (Figs. S1 and S2), suggesting the involvement of additional genes. The reported penetrance and expressivity of CHDs, heterotaxy, and preaxial digit duplication was much higher in Megf8 m/m embryos compared to Mgrn1 m/m embryos (Cota et al., 2006;Zhang et al., 2009) (Table S1). Similarly in NIH/3T3 cells, when compared to the loss of MGRN1, the loss of MEGF8 resulted in more Hh signaling activity at baseline and a greater abundance of SMO at the plasma and ciliary membranes (Figs. 1A and 1C). Evolutionary sequence analysis indicated that RNF157, which also encodes a RING superfamily E3 ligase, is a vertebratespecific paralog of MGRN1 ( Fig. 1B)  . Although MGRN1 is more widely distributed, found amongst almost all major eukaryotic lineages, MGRN1 and RNF157 share a RING domain and a distinctive predicted substrate-binding domain that is unique amongst other members of the RING superfamily ( Figs. 1B and S3A).
To assess the relationship between RNF157 and MGRN1 in vivo, we generated Rnf157 -/mice (hereafter referred to as Rnf157 m/m mice) using CRISPR methods (Fig. S3B). Consistent with data collected by the International Mouse Phenotyping Consortium (IPMC) using a different knockout strategy (Dickinson et al., 2016), the Rnf157 m/m mice were viable, fertile, and without obvious developmental defects (Figs. 1D and 1F). Rnf157 m/+ mice were mated to Mgrn1 m/m mice to generate compound heterozygotes, which were intercrossed to obtain 1A, 1C-1F, and S3C-E). This compensation is asymmetric, as the loss of RNF157 alone had few developmental consequences (Figs. 1D and 1F), presumably because MGRN1 can fully cover RNF157 functions. In conclusion, Rnf157 is a modifier gene: mutations in Rnf157 are insufficient to cause a phenotype alone, but they increase the penetrance of phenotypes caused by mutations in a different gene (Mgrn1).

MEGF8 binds to MGRN1
Mouse embryos and cells that lack MEGF8 are indistinguishable from those that lack both MGRN1 and RNF157 (Figs. 1A and 1C-F), leading us to speculate that MEGF8, MGRN1, and RNF157 may work together to regulate SMO trafficking. We were encouraged to look for a physical interaction by the BIOGRID and BioPlex databases (Huttlin et al., 2017;Oughtred et al., 2019), which predict that MEGF8 is an interaction partner of both MGRN1 and RNF157 ( Fig.   2A). To validate this prediction, we transiently expressed MEGF8 in HEK293T cells and observed that it could be co-immunoprecipitated (co-IP) with either endogenous or overexpressed MGRN1 ( Figs. 2A and 2C). Deleting the ~170 amino acid (a.a.) long cytoplasmic tail (hereafter called the "Ctail") of MEGF8 (MEGF8 ΔCtail ), but not its large ~2500 a.a. extracellular domain (MEGF8 ΔN ), abolished the interaction with MGRN1 ( Figs. 2A-C). The MEGF8 Ctail contains a peptide motif (with the sequence "MASRPFA") that is highly conserved across a family of single-pass transmembrane proteins found in Filozoa, animal-like eukaryotes including Filasterea, Choanoflagellatea, and Metazoa (Figs. 2B and S4A) (Gunn et al., 1999;Haqq et al., 2003;Nagle et al., 1999). The deletion of this motif (MEGF8 ΔMASRPFA ) abrogated the interaction between MEGF8 and MGRN1 (Fig. 2C), establishing an E3 ligase recruitment function for this mysterious sequence element.
To test if the association between MEGF8 and MGRN1 was relevant for the regulation of Hh signaling, we stably expressed wild-type MEGF8 or the interaction-defective MEGF8 ΔCtail mutant in Megf8 -/-NIH/3T3 cells (Fig. 2D). Stably expressed MEGF8, but not its truncated MEGF8 ΔCtail variant, bound to endogenous MGRN1 (Fig. 2D) and suppressed the elevated basal GLI1 and ciliary SMO seen in Megf8 -/cells (Figs. 2D and 2E). The MEGF8-MGRN1 interaction was unchanged when signaling was activated by the addition of SHH (Fig. 2D).
These data establish that MGRN1 in the cytoplasm stably associates with the Ctail of MEGF8 and this interaction is required to suppress ciliary SMO levels and attenuate Hh signaling.
The ubiquitin ligase activity of MGRN1 is required to attenuate Hh signaling MGRN1 regulates processes ranging from skin pigmentation to spongiform neurodegeneration by directly ubiquitinating multiple substrates (Chakrabarti and Hegde, 2009;Gunn et al., 2013a;Jiao et al., 2009). We constructed two variants of MGRN1 (MGRN1 Mut1 and MGRN1 Mut2 ) carrying mutations in highly conserved residues of the RING domain (Fig. S4B). These mutations are known to abolish binding between RING domains and their cognate E2 partners, thereby preventing ubiquitin transfer to substrates (Garcia-Barcena et al., 2020;Gunn et al., 2013a). We stably expressed wild-type MGRN1, MGRN1 Mut1 , or MGRN1 Mut2 in Mgrn1 -/-;Rnf157 -/-NIH/3T3 cells and measured the abundance of GLI1, post-ER SMO, and ciliary SMO (Figs. 2F and 2G). In all three assays, wild-type MGRN1 was able to fully attenuate Hh signaling and SMO levels, but the MGRN1 Mut1 and MGRN1 Mut2 variants were inactive. Importantly, MGRN1 Mut1 and MGRN1 Mut2 were expressed at equivalent levels as MGRN1 (Fig. 2F) and maintained their stable interaction with MEGF8 ( Fig. S4C), demonstrating their integrity. These results support the conclusion that both the stable interaction of MGRN1 with MEGF8 and its E3 ligase function are required to attenuate Hh signaling.

The MEGF8-MGRN1 complex ubiquitinates SMO
At this point our data suggested that MGRN1 functions as a membrane-tethered ubiquitin ligase complex that attenuates Hh signaling by reducing SMO abundance at the cell surface and primary cilium. This mechanism is reminiscent of a prominent membrane-localized ubiquitination system that attenuates WNT signaling by decreasing cell-surface levels of Frizzled (FZD) proteins, receptors for WNT ligands that are the closest relatives of SMO in the GPCR superfamily (Bjarnadóttir et al., 2006). Two transmembrane RING-family E3 ubiquitin ligases, ZNRF3 and RNF43, attenuate WNT responsiveness by directly ubiquitinating FZD and promoting its clearance from the cell surface (Hao et al., 2012;Koo et al., 2012).
To examine if a similar ubiquitination system regulates Hh signaling sensitivity, we measured the stability of SMO at the plasma membrane using a non-cell permeable biotinylation reagent that only labels proteins at the cell surface in wild-type, Megf8 -/-, and Mgrn1 -/-;Rnf157 -/cells (Fig. 3A). Both the steady state abundance and the stability of cellsurface SMO were markedly greater in both mutant cell lines compared to wild-type cells (Figs. 3B and 3C). The increase in ciliary SMO abundance (Fig. 1C) is likely a secondary consequence of elevated SMO at the plasma membrane, because plasma membrane-localized SMO can enter the cilia by a lateral transport pathway (Milenkovic et al., 2009). These results are analogous to how the stability of cell-surface FZD is enhanced when the ligases ZNRF3 or RNF43 are inactivated (Hao et al., 2012;Koo et al., 2012), prompting us to consider whether SMO is a substrate for the MEGF8-MGRN1 ubiquitin ligase complex.
We established an assay to measure SMO ubiquitination by expressing SMO and Hemagglutinin (HA)-tagged ubiquitin (UB) together in HEK293T cells and then measuring the amount of HA-UB conjugated to SMO (Figs. 4A and 4B). Lysates were prepared under denaturing conditions to ensure only covalent HA-UB conjugates would survive. SMO was isolated by immunoprecipitation and the attached UB chains detected (as a smear) by immunoblotting with an anti-HA antibody. Co-expression of MGRN1 alone had no effect on SMO ubiquitination, co-expression of MEGF8 alone modestly increased SMO ubiquitination, but the co-expression of both MEGF8 and MGRN1 dramatically increased levels of ubiquitinated SMO and concomitantly reduced SMO abundance (Fig. 4A). A ubiquitin mutant lacking all lysine residues (UB K0 ) was poorly conjugated to SMO, suggesting that SMO is attached to poly-UB chains, rather than to a single ubiquitin (Fig. S5A). Inactivating mutations in the RING domain of MGRN1 (MGRN1 Mut1 and MGRN1 Mut2 ) failed to promote SMO ubiquitination (Fig. 4A), indicating the E3 ligase activity of MGRN1 was required. SMO was a selective substrate for MGRN1 and MEGF8 because their co-expression did not change the abundance of a different ciliary GPCR, SSTR3 (Fig. S5B). SMO contains 21 lysine (K) residues exposed to the cytoplasm that could function as acceptors for ubiquitin. Changing all of these lysines to arginines (R) impaired MGRN1-mediated ubiquitination (Fig. S5C), but changing specific clusters of lysines in each of the cytoplasmic loops or the tail of SMO did not reduce ubiquitination (Fig. S5C). Thus, MGRN1 does not seem to favor a particular lysine residue or set of lysine residues on the cytoplasmic surface of SMO, at least in this over-expression based HEK293T assay.
Efficient SMO ubiquitination required both MEGF8 and the E3 ligase function of MGRN1.
The small increase in SMO ubiquitination seen in the presence of MEGF8 alone is likely due to presence of endogenous MGRN1 in HEK293T cells (see asterisks in the MGRN1 panel in Fig.   4A). To directly test whether the physical interaction between MEGF8 and MGRN1 was required to mediate SMO ubiquitination, we co-expressed MGRN1 with one of three MEGF8 variants (diagrammed in Fig. 2A): (1) MEGF8 ΔCtail , (2) MEGF8 ΔMASRPFA (both of which cannot bind to MGRN1, Fig. 2C), or (3) MEGF8 ΔN , which lacks the large extracellular domain of MEGF8 but retains its transmembrane (TM) domain and Ctail. MEGF8 ΔCtail and MEGF8 ΔMASRPFA failed to support SMO ubiquitination (Fig. 4B). In contrast, MEGF8 ΔN , which can still bind to MGRN1 (Fig. 2C), efficiently promoted SMO ubiquitination and degradation (Fig. 4B).
Interestingly, MEGF8 ΔN was much more active than full-length MEGF8 (despite both proteins being expressed at comparable levels), suggesting that the extracellular domain of MEGF8 may negatively regulate the function of the Ctail or interfere with its interaction with SMO. In addition to recruiting MGRN1 to the plasma membrane, the association between MEGF8 and MGRN1 promoted the intrinsic E3 ligase activity of MGRN1, evident through the ability of MEGF8 ΔN to reduce the abundance of co-expressed wild-type MGRN1 (Fig. 4B). Most E3 ligases catalyze their own ubiquitination and de-stabilization, a property that reflects their intrinsic catalytic activity.
Unexpectedly, MEGF8 ΔN , which includes only the TM domain and Ctail of the protein (232 out of the 2778 amino acids in the full-length protein), was sufficient to promote SMO ubiquitination (Fig. 4B). To further narrow down the region of MEGF8 required for SMO recognition, we constructed a set of chimeric proteins that fused the MEGF8 Ctail, TM domain, or both to heterologous extracellular and transmembrane domains from CD16 and CD7, respectively (diagrammed in Fig. S5D). In the HEK293T assay, both the TM domain and the Ctail of MEGF8 were required to promote SMO ubiquitination; simply tethering the isolated Ctail to the plasma membrane by fusing it to a CD16-CD7 hybrid protein was not sufficient.
Abrogating the interaction with MGRN1 by deleting the "MASRPFA" motif abolished the function of these chimeric proteins, demonstrating that they still require MGRN1 to promote SMO ubiquitination (Fig. S5D).
If the biochemical function of MEGF8 in Hh signaling is to ubiquitinate SMO, a key prediction is that the CD16 ECD -MEGF8 TM+Ctail chimera, a minimal engineered protein that is sufficient to carry out this function, should be able to reverse the enhanced Hh signaling phenotype in Megf8 -/cells. To test this prediction, we stably expressed CD16 ECD -CD7 TM -MEGF8 Ctail , CD16 ECD -MEGF8 TM+Ctail , and CD16 ECD -MEGF8 TM+CtailΔMASRPFA (a variant carrying the MASRPFA deletion) in Megf8 -/cells (Fig. 4C). All three chimeras were expressed and localized properly to the cell surface as measured by flow cytometry using an antibody against the CD16 ECD (Fig. S5E). However, only the CD16 ECD -MEGF8 TM+Ctail chimera could completely suppress Gli1 expression and both post-ER and ciliary SMO abundance (Figs. 4C and 4D

Genetic interactions between Megf8 and Mgrn1
After identifying the MEGF8-MGRN1 interaction and elucidating the ubiquitination based mechanism through which it regulates the sensitivity of target cells to Hh ligands, we sought to investigate the role of this protein complex in embryonic development using the previously published Megf8 m/+ and Mgrn1 m/+ mouse lines Phillips, 1963;Zhang et al., 2009). Notably, both Megf8 m/m and Mgrn1 m/m mutant embryos display CHDs, heterotaxy, and preaxial digit duplication. While these phenotypes are fully penetrant in the Megf8 m/m mutants, they show lower penetrance in the Mgrn1 m/m mutants (likely due to partial redundancy with Rnf157) (Fig. 1F) (Cota et al., 2006;Zhang et al., 2009). To determine whether the developmental defects exhibited by these two mutants are a product of the same pathway (as predicted by our biochemical studies), we assessed for a genetic interaction by intercrossing the Megf8 m/+ and Mgrn1 m/+ mice and examining the phenotypes of the resultant double heterozygous Megf8 m/+ ;Mgrn1 m/+ embryos.
In addition to reduced penetrance, the CHDs seen in Megf8 m/+ ;Mgrn1 m/+ double heterozygous embryos were also milder compared to Megf8 m/m embryos. All Megf8 m/m embryos suffered from transposition of the great arteries (TGA), a severe outflow tract (OFT) malalignment defect in which the aorta emerges from the right ventricle and the pulmonary artery from the left ventricle (Figs. 6A and 6B, Given the known co-occurrence of heterotaxy with severe CHDs in clinical data from human birth registries (Lin et al., 2014;Pradat et al., 2003), we examined the correlation between these two types of birth defects in our mutant mouse embryos. All Megf8 m/m embryos had both heterotaxy and TGA (Figs. 1F and 6C, Table S1). In Megf8 m/+ ;Mgrn1 m/+ embryos, heterotaxy was associated 100% of the time with CHDs and, conversely, CHDs were associated 64% of the time with heterotaxy (Fig. 6C, Table S4). Interestingly, the presence of heterotaxy was also correlated with more severe CHDs: ~60% of these embryos also had TGA ( Fig. 6A). In contrast, Megf8 m/+ ;Mgrn1 m/+ embryos with normal left-right patterning (situs solitus) did not have TGA and instead had the milder DORV in ~20% of cases (Fig. 6A).
These correlations are remarkably similar to data from human birth registries, which report that ~85% of heterotaxy cases are associated with CHDs that include DORV, TGA and AVSD (Lin et al., 2014;Pradat et al., 2003). The tight association between CHD and heterotaxy is also supported by the observation that all seven embryos with only preaxial digit duplication (but no CHD) had normal situs solitus (Table S4). Thus, the double heterozygous Megf8 m/+ ;Mgrn1 m/+ embryos recapitulate the known association between severe CHD and heterotaxy seen in human clinical data. The wider spectrum of CHDs seen in these embryos, including DORV, compared to homozygous Megf8 m/m embryos resembles the more diverse range of CHDs seen in human patients with heterotaxy ( Fig. 6A) (Lin et al., 2014;Pradat et al., 2003).

Gene dosage effects involving Mgrn1, Megf8 and Rnf157
Our comparison of double heterozygous Megf8 m/+ ;Mgrn1 m/+ embryos to homozygous Megf8 m/m embryos suggested that both the penetrance and expressivity of birth defect phenotypes may be determined by precise magnitude of ubiquitin ligase activity, which in turn determines the abundance of SMO and the strength of Hh signaling. This hypothesis predicts that the dosage of Megf8, Mgrn1 and Rnf157 should influence the penetrance of birth defect phenotypes.
We analyzed embryos carrying varying numbers of loss-of-function Megf8 m , Mgrn1 m , and Rnf157 m alleles (Fig. 6C). Megf8 m/m and Mgrn1 m/m ;Rnf157 m/m embryos have a 100% penetrance of CHDs, heterotaxy, and preaxial digit duplication, presumably because the functions of both the transmembrane adaptor (MEGF8) and the cytoplasmic E3 ligases (MGRN1 or RNF157) are essential for SMO ubiquitination. Loss of one allele of Megf8 (Megf8 m/+ embryos), one allele of Mgrn1 (Mgrn1 m/+ embryos) or both alleles of Rnf157 (Rnf157 m/m embryos) did not lead to birth defects, likely because the abundance of the MEGF8-MGRN1/RNF157 complex remains above the threshold required for normal development.
However, between these two extremes, decreasing the cumulative gene dosage (by increasing the number of mutant alleles) of Mgrn1 and Megf8 led to a progressive increase in the penetrance of CHDs, heterotaxy and preaxial digit duplication (Fig. 6C). In addition, the incidence of TGA (Table S6), the most severe CHD, and the co-occurance of heterotaxy ( Fig.   6C) increased with decreasing gene dosage. These striking gene dosage effects support the model that a progressive decrease in ubiquitin ligase function leads to a progressive increase in the penetrance and expressivity of birth defects, perhaps by driving a graded increase in Hh signaling strength.
The exquisite sensitivity of heart development to mutations in Megf8, Mgrn1 and Rnf157 seen in mouse embryos prompted us to look for potentially damaging variants in these genes in patients with CHDs. Using whole exome sequencing data from a cohort of 652 CHD patients, we searched for missense variants in all three genes with a Combined Annotation Dependent Depletion (CADD) score >10. We additionally used a stringent mean allele frequency (MAF) filter of < 0.5% for MEGF8 and MGRN1, but a more relaxed MAF filter (< 5%) for RNF157, since the Rnf157 m/m mouse has no phenotype. Using these criteria, we identified one patient (7501) with two mutations each in MEGF8 and MGRN1 and one mutation in RNF157 (Figs. S6A and S6B). Genotyping the parents of patient 7501 revealed that the two mutations in MEGF8 and MGRN1 were both present in the same allele, with the former transmitted from the mother and the latter from the father (along with the RNF157 variant). Patient 7501 clinically presented with OFT anomalies: pulmonary atresia (Fig. S6C), a severely hypoplastic right ventricle with an intact interventricular septum (Fig. 7B), an atrial septal defect (Fig. 7B) and patent foramen ovale (Fig. 7B). Primary fibroblasts from patient 7501 displayed increased abundance of ciliary SMO (Fig. 7C) and elevated Gli1 expression (Fig. 7D), both at baseline and in response to SHH, when compared to fibroblasts generated from a subject without CHD. Collectively, our mouse and human data support a model where disruption of the MEGF8-MGRN1/RNF157 ubiquitin ligase complex can lead to elevated SMO, increased Hh signaling strength and, consequently, to the emergence of CHDs.

Discussion
Using a combination of mechanistic studies, mouse genetics, and deep anatomical phenotyping, we uncovered a unique membrane-tethered ubiquitination pathway that regulates developmental patterning in a variety of tissues by controlling the trafficking of signaling receptors. MEGF8 functions as a transmembrane substrate adaptor that recruits a cytoplasmic E3 ligase (MGRN1) to catalyze the ubiquitination of SMO, leading to its endocytosis and degradation (Fig. 7E). This ubiquitination reaction reduces the abundance of SMO at the cell surface and primary cilium and, consequently, dampens Hh signaling in target cells.

Receptor-like ubiquitin ligases attenuate signaling strength
The architecture of the MEGF8-MGRN1 complex is notable for the presence of a membranespanning component with an extracellular or luminal domain (Fig. 7E). This feature suggests a receptor-like function, conceptually analogous to receptor kinases, to transmit extracellular or luminal signals across the membrane to alter the ubiquitination of substrates in the cytoplasm.
Interestingly, FZD receptors for WNT ligands are regulated by transmembrane E3 ligases (RNF43 and ZNRF3) in which the RING-containing domain is directly fused to the membranespanning component (rather than being non-covalently associated as in the MEGF8-MGRN1 system) (Fig. 7E). While a ligand for MEGF8 remains unknown, ZNRF3 and RNF43 are regulated by R-Spondin ligands, critical regulators of progenitor cells during development and stem cells in adult tissues (Hao et al., 2012;Koo et al., 2012). The ubiquitination of receptors by membrane-tethered E3 ligases represents an attractive post-transcriptional mechanism to control the sensitivity of tissues to signaling ligands during development or tissue renewal.
Evolutionary sequence analysis supports a widespread (but under-appreciated) role for transmembrane E3 ligase complexes in ubiquitin signaling. In animals and their immediate sister lineages, MGRN1 and RNF157 likely function as common components of multiple membrane-tethered E3 ligase complexes featuring members of the MEGF8 family of cellsurface proteins, all of which contain an equivalent of the cytoplasmic MASRPFA motif ( Fig.   S4A) (Gunn et al., 1999;Haqq et al., 2003;Nagle et al., 1999). For example, MGRN1 and a different member of this family, Attractin (ATRN), have been implicated in regulation of melanocortin receptor levels by ubiquitination ( Fig. 7E) (Cooray et al., 2011;Walker, 2010).
However, members of the MGRN1 family of RING finger proteins are more widely distributed across eukaryotes (Fig. 1B) compared to MEGF8/ATRN-like proteins (which are restricted to animal-like eukaryotes) (Pusapati et al., 2018b). Thus, MGRN1-family proteins may associate with other adaptors beyond the MEGF8/ATRN family to form comparable membrane-localized E3 ligase complexes in different eukaryotic lineages. Indeed, a plant ubiquitin ligase, LOG2, which belongs to the MGRN1 family, associates with and ubiquitinates a single TM protein Glutamine dumper-1 (GDU1) which in turn regulates amino acid transport ( Fig. 7E) (Guerra et al., 2013). Strikingly, human MGRN1 can functionally replace LOG2 in plants ( Guerra et al., 2013). Like the MEGF8/ATRN family in animals, the extensive GDU1 family, conserved across all land plants, has a cytoplasmic tail which features a conserved motif. This motif has a central "MAs" signature (where s is a small amino acid typically G or S) reminiscent of the MASRPFA sequence in the MEGF8/ATRN family (Fig. 2B). Accordingly, we propose that the MGRN1 family of RING E3 ligases can associate more generally across eukaryotes with single-pass TM proteins with conserved cytoplasmic motifs, each of which function as a substrate adaptor to target the ubiquitination of specific receptors or transporters (Fig. 7E).

Role of Hh signaling in left-right patterning and heart development
Several lines of evidence point to the Hh signaling pathway as being important for the phenotypes seen in mice carrying mutant alleles of Megf8 and Mgrn1. Both genes were identified in our unbiased genetic screen as attenuators of Hh signaling  and the key Hh transducer SMO is a direct substrate of the MEGF8-MGRN1 complex. Hh signaling plays a key role in each of the developmental events where we see prominent birth defect phenotypes: left-right patterning, cardiac morphogenesis, and limb development.
Hh signaling has previously been implicated in sustaining left-right patterning signals, a very early event in development that directs the correct asymmetric development of the heart and other visceral organs (Levin et al., 1995;Tsiairis and McMahon, 2009;Zhang et al., 2001).
The genetic deletion of SMO, which reduces Hh signaling strength, disrupts left-right patterning and causes a midline heart tube that fails to loop to the right and an embryo that fails to turn (Zhang et al., 2001). Distinct from prior work showing that the loss of Hh signaling impairs proper left-right patterning in the mouse, we find that mutations (in Megf8 and Mgrn1) that increase Hh signaling strength also result in heterotaxy and CHDs. These results support the view that left-right patterning (and cardiac and limb development) depend on a just-right "goldilocks" level of Hh signal amplitude or duration.
Hh signaling also influences multiple aspects of heart development: development of the secondary heart field and proper septation of the atria and outflow tract (Dyer and Kirby, 2009;Hoffmann et al., 2009;Washington Smoak et al., 2005). We expect that CHDs seen in our mutant mice are caused by both heterotaxy and by later defects in Hh-mediated patterning of the cardiac septa and outflow tract. The common link of both processes to precisely-tuned level of Hh signaling may explain the tight association between heterotaxy and CHDs that has been long-noted in clinical studies and is recapitulated in our mutant mouse embryos (Pradat et al., 2003).
We acknowledge that MEGF8 and MGRN1/RNF157 may regulate signaling receptors other than SMO and some of the birth defect phenotypes we observe may be related to disruption of other signaling pathways. Future priorities include the identification of other substrates targeted for ubiquitination by the MEGF8-MGRN1 complex and the assessment of where and when these proteins function during embryonic development.

Oligogenic interactions and gene dosage effects underlie birth defects
Our biochemical studies provide an explanation for the strong genetic interactions observed between the genes that encode subunits of the MEGF8-MGRN1/RNF157 complex. While single heterozygous Megf m/+ and Mgrn1 m/+ embryos are normal, double heterozygous Megf m/+ ;Mgrn1 m/+ embryos display CHDs with heterotaxy. This phenomenon has been called "synthetic haploinsufficiency" and can result in an oligogenic pattern of inheritance, where mutations in one gene affect the phenotypic outcome of mutations in a different gene (Kousi and Katsanis, 2015;Veitia et al., 2013). Synthetic haploinsufficiency is most commonly seen between genes that encode subunits of a protein complex, like MEGF8 and MGRN1 (Veitia, 2010). Pioneering studies of Bardet-Biedl Syndrome (BBS) and other inherited retinopathies have demonstrated the importance of oligogenic interactions for understanding the genetic etiology of human diseases (Badano et al., 2006;Katsanis et al., 2000).
Beyond binary genetic interactions, the penetrance and expressivity of birth defect phenotypes progressively increases as an inverse function of the gene dosage of Megf8, Mgrn1 and Rnf157. We propose that this quantitative effect of mutations in this pathway is explained by the dependence of proper left-right patterning and cardiac development on the precise amplitude of Hh signaling and by the central role of the MEGF8-MGRN1 pathway in setting this amplitude in target cells. The inheritance of increasing numbers of Megf8, Mgrn1 and Rnf157 mutant alleles will lead to a progressive decrease in the abundance (and hence activity) of the MEGF8-MGRN1/RNF157 complex. Decreasing E3 ligase activity will result in progressive increases in cell surface and ciliary SMO and thus increases in target cell sensitivity to Hh ligands. More generally, our results show that developmental patterning events can be tightly regulated by mechanisms in target cells that function to precisely tune sensitivity to extracellular morphogens.
We finish by noting that our genetic analyses highlight how interactions between a small number of genes can produce a complex inheritance pattern (common to many human diseases) that features sporadic occurrence, incomplete penetrance, and variable expressivity.

Declaration of Interests
The authors declare no competing interests.    (A) Immunoblots of wild-type, Mgrn1 -/-, Mgrn1 -/-;Rnf157 -/-, and Megf8 -/-NIH/3T3 cells treated with no SHH (0 nM), low SHH (1 nM), or high SHH (25 nM). Immunoblots show GLI1 (a metric for signaling strength), SMO, and α-Tubulin (α-TUB) as a loading control. The fraction of SMO that has traversed the Endoplasmic Reticulum (ER) and acquired glycan modifications in the golgi (labeled "post-ER SMO") is distinguished by its slower migration on the gel. An analysis of additional clones of double mutant Mgrn1 -/-;Rnf157 -/cell lines is shown in Fig. S3C. (B) Unrooted maximum-likelihood tree topology showing the evolutionary relationship between MGRN1 and RNF157. Branches of the tree are colored according to the major eukaryotic lineages, with the vertebrate-specific RNF157 lineage highlighted in purple. The circle on the node separating MGRN1 and RNF157 denotes 100% confidence support (1000 replicates). The scale bar indicates phylogenetic distance as the number of amino acid substitutions per site. The full Newick tree file is provided in Supplemental File 1. (C) Violin plots (left) and corresponding representative confocal fluorescence microscopy images (right) of SMO (red) at primary cilia (green, marked by ARL13B) in wild-type, Mgrn1 -/-, Mgrn1 -/-;Rnf157 -/-, and Megf8 -/-NIH/3T3 cells (n~70 cilia/condition). The DNA in the nucleus is stained blue with DAPI. Wild-type cells were analyzed after treatment with either no SHH or saturating SHH (25 nM). Arrowheads identify individual cilia captured in the zoomed images above each panel. Statistical significance was determined by the Kruskal-Wallis test; **p-value ≤ 0.01 and ****p-value ≤ 0.0001. Scale bars, 10 µm in merged panels and 2 µm in zoomed displays. An analysis of additional clones of Mgrn1 -/-;Rnf157 -/cell lines is shown in Fig. S3D.  Table S1.  Wild Type  (A) To measure the degradation rate of cell-surface SMO in wild-type, Megf8 -/-, and Mgrn1 -/-;Rnf157 -/-NIH/3T3 cells, SMO was labeled with non-cell-permeable biotin at 4°C. After warming cells to 37°C to allow for SMO internalization and degradation for various periods of time, the amount of biotinylated SMO remaining was isolated on streptavidin beads and measured by immunoblotting (B). While the steady state abundance of SMO (at t=0) was much higher in Megf8 -/and Mgrn1 -/-;Rnf157 -/cells compared to wild-type cells, the fraction of initial SMO remaining at various times after cell-surface labeling is quantified in (C). Error bars represent the standard error of two independent replicates.  (A and B) SMO ubiquitination was assessed after transient co-expression with HA-tagged ubiquitin (HA-UB) and the indicated MEGF8 and MGRN1 variants in HEK293T cells (see Fig.  2A for a summary of protein variants). Cells were lysed under denaturing conditions, SMO was purified by IP, and the amount of HA-UB covalently conjugated to SMO assessed using immunoblotting with an anti-HA antibody. An asterisk (*) indicates endogenous MGRN1 present in HEK293T cells. (C and D) Total GLI1 and SMO abundances were measured by immunoblotting (C) and ciliary SMO by fluorescence confocal microscopy (D) in Megf8 -/cells expressing various CD16/CD7/MEGF8 chimeras (diagrammed in Fig. 4C). Chimeras were composed of the extracellular domain (ECD) of CD16, followed by a transmembrane (TM) helix from either CD7 or MEGF8, and finally by the Ctail of MEGF8, either with or without the MGRN1-interacting MASRPFA motif. The ability of these chimeras to support SMO ubiquitination is shown in Fig.  S5D and the abundances of chimeras at the cell surface is shown in Fig. S5E. (D) Horizontally positioned violin plots summarize the quantification of SMO fluorescence (red) at ~50 individual cilia (green, ARL13B) per cell line from representative images of the type shown immediately to the left. Arrowheads identify individual cilia captured in the zoomed images above each panel. Statistical significance was determined by the Kruskal-Wallis test; not-significant (ns) > 0.05, **p-value ≤ 0.01, and ****p-value ≤ 0.0001. Scale bars, 10 µm in merged panels and 2 µm in zoomed displays.          ) that causes heart defects and heterotaxy in mice. In the loss-of-function md-nc allele of Mgrn1, a thymidine to adenine mutation in intron 9 disrupts splicing, leading to a premature stop (at the position shown by the arrow above MGRN1) and nonsense mediated decay of the transcript .
(B and C) Hh signaling strength was assessed using qRT-PCR to measure Gli1 mRNA in primary mouse embryonic fibroblasts (  Confocal fluorescence microscopy images of ciliated tissues (forelimb bud, spinal cord, lateral ventricles of the brain, and heart) collected from e12.5 control (Mgrn1 m/+ or Megf8 m/+ ), Megf8 m/m , and Mgrn1 m/m embryos. Arrows denote regions enlarged in insets. In control embryos, ciliary SMO was only present in the floor plate of the ventral spinal cord. In contrast, SMO was detected at cilia in all Megf8 m/m tissues analyzed. SMO was sparsely detected at cilia in the Mgrn1 m/m spinal cord and brain. Red, SMO; green, ARL13B (primary cilia); blue, DAPI (nuclei). Scale bars, 10 µm. Wild Type

F F S A F F L F A G V V I G A S K I H V V L S E R A T T I Q E Q T Q M Q T L A R R P M G K L L L A W G P 2 3 P I A S Q P T R G N Q S S L V T M L V E L P S 5 N L S L A T A -2332 XP_004343315.2_Capsaspora_owczarzaki 1112 N L I H F F L T F F A C F C A L L L A A F I V W R I R V R F E H N Q F V R E I T L A F E Q M S T R P S A S I T L W D F C 1 3 P L A I E P F R D G K A A V M T F A L E V P A 6 Q V C L A S A L 1221 consensus/70%
-     . Immunoblots indicate that coexpression of MEGF8 and MGRN1 complex reduced the abundance of SMO, but had no effect on SSTR3. (C) Ubiquitination of wild type SMO or variants carrying mutations in cytoplasmically-exposed lysine residues by the MEGF8-MGRN1 complex. The following five SMO mutants were tested:

HA-UB
(1) SMO-K0 (all 21 cytoplasmic lysine residues were changed to arginines), (2) Smo-Ctail K0 (all 16 lysines in the C-terminal cytoplasmic tail were changed to arginines), (3) SMO-ICL K0 (all 5 lysines in the three cytoplasmic loops were changed to arginine), (4) SMO-ICL2 K0 (both lysines in the second intracellular loop were changed to arginines), and (5) SMO-ICL3 K0 (all three lysines in the third intracellular loop were changed to arginines). Cells were lysed under denaturing conditions, native SMO was purified by IP using beads covalently linked to an anti-SMO antibody, and the amount of HA-UB covalently conjugated to SMO was assessed using immunoblotting with an anti-HA antibody (bottom panel). (D) Chimeric proteins were used to identify the minimal region of MEGF8 sufficient to support SMO ubiquitination. Using the assay shown in Fig. 4A, SMO ubiquitination was assessed after transient co-expression of the following in 293T cells: SMO-eGFP, HA-UB, MGRN1 (or the inactive mutant MGRN1 Mut1 ), and the MEGF8 mutant or chimera shown above the blot. See Fig.  2A and associated text for a description of these chimeras. (E) Flow cytometry was used to measure cell surface labeled CD16 in live Megf8 -/cells stably expressing various CD16/CD7/MEGF8 chimeras (diagrammed in Fig. S5D). These are the same stable cell lines analyzed in Figs. 4C and 4D. 4000 cells were analyzed per condition.

LEAD CONTACT AND MATERIALS AVAILABILITY
Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Rajat Rohatgi (rrohatgi@stanford.edu). All unique/stable reagents generated in this study are available from the Lead Contact with a completed Materials Transfer Agreement.

NIH/3T3 and HEK293T cell culture
Flp-In-3T3 (a derivative of NIH/3T3 cells and referred to as "NIH/3T3" cells throughout the text) and HEK293T cell lines were purchased from Thermo Fisher Scientific and ATCC, respectively.
Information on the gender of the cell lines is not available. NIH/3T3 and HEK293T cells were

Generation of primary mouse embryonic fibroblasts
Primary mouse embryonic fibroblasts (pMEFs) were generated using a modified published protocol (Durkin et al., 2013). Briefly, e12.5-14.5 embryos were harvested and rinsed thoroughly with PBS to remove any excess blood. Using forceps, the head and internal organs (heart and liver) were removed. The embryos were then separated into individual dishes and a sterile razor blade was used to physically mince the tissue in 0. 25% Trypsin/EDTA (Thermo Fisher Scientific,Gibco). After pipetting the minced tissue up and down several times to further break up the tissue, the dishes were placed in a 37 °C tissue culture incubator for 10-15 minutes. 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 minutes. 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. The gender of the embryos were not determined prior to generating the pMEF cultures.
Cells were housed at 37 °C in a humidified atmosphere containing 5% CO2.

Patient recruitment and nasal sampling for patient derived fibroblast cultures
Patients and parents were recruited from the Children's Hospital of Pittsburgh with informed consent obtained under a human study protocol approved by the University of Pittsburgh Institutional Review Board. Control, CHD patient, and parents recruited had blood drawn for DNA extraction. CHD diagnosis was confirmed with examination of the patient's medical records. Nasal tissue was obtained from the patient by curettage of the inferior nasal turbinate using a rhino probe. The nasal epithelial tissue was plated in RPMI medium (Thermo Fisher Scientific, Gibco) with 10% FBS (MilliporeSigma) and the fibroblast outgrowths that emerged were expanded and used for Hh signaling assays (below). Both primary fibroblast cell lines (control and patient 7501) were derived from cells collected from female patients.

Hh signaling assays in NIH/3T3 cells and primary fibroblasts
For Hh signaling assays, NIH/3T3 cells, pMEFs, and primary human fibroblasts were first grown to confluence in Complete Medium (containing 10% FBS) and then ciliated by changing the cell medium to Low Serum Medium (complete medium containing 0.5% FBS) for 24 hours. Cells were then treated with either no SHH, a low concentration of SHH (1 nM), a high concentration of SHH (25 nM), or SAG (100 nM) for at least 4 hours prior to fixation (for NIH/3T3 immunofluorescence assays), 24 hours prior to lysis (for NIH/3T3 Western blot assays), or 48 hours prior to experimentation (for pMEF and primary human fibroblast immunofluorescence and western blot assays).
Hh signaling activity was measured using real-time quantitative reverse transcription PCR (qRT-PCR). RNA was extracted from NIH/3T3 cells and mouse pMEFs using TRIzol reagent (Thermo Fisher Scientific, Invitrogen) as previously described (Rio et al., 2010). Equal amounts of RNA were used as template for cDNA synthesis using the iScript Reverse

Reagents and antibodies
Recombinant SHH was expressed in bacteria and purified in the lab as previously described (Bishop et al., 2009). SAG was purchased from Thermo Fisher Scientific (Enzo Life Sciences).
The selection antibiotic puromycin was purchased from MilliporeSigma and hygromycin B from protein and affinity purified before use (Cocalico Biologicals, Inc., 1:2000). Hoechst 33342 and secondary antibodies conjugated to horseradish peroxidase (HRP) or Alexa Fluor dyes were obtained from Jackson Laboratories and Thermo Fisher Scientific.

Protein Sequence Analysis
Iterative sequence profile searches were performed using the PSI-BLAST program run against the NCBI non-redundant (NR) protein database (Altschul et al., 1997). Multiple sequence alignments were built using the Kalign2 software (Lassmann et al., 2009) and were later manually adjusted based on profile-profile, secondary structure information, and structural alignments. Similarity-based clustering for both classification and discarding of nearly identical sequences was performed using the BLASTClust program (Fig. S4A). Maximum-likelihood (ML) tree topology was derived using an edge-linked partition model as implemented in the IQ-TREE software (Nguyen et al., 2015). ModelFinder (Kalyaanamoorthy et al., 2017) was used to automatically identify the best-fit substitution model and estimated "JTT+F+R9" as the suitable model for the given dataset. Branch supports were obtained using the ultrafast bootstrap (UFBoot) approximation method (1000 replicates) (Hoang et al., 2018). To further assess the branch supports, Shimodaira-Hasegawa(SH-)aLRT branch test was also computed as implemented in the IQ-TREE software (Fig. 1B). The sequence logo was generated using the Logo software (Crooks et al., 2004) (Fig. 2B). An alignment comprising a collection of all unique members of the MEGF8-Attractin family from the RefSeq database was utilized as input. The UniProt align tool was used to compare two protein sequences with the Clustal Omega program ( Fig. S3A) (UniProt Consortium, 2019). Sequence analysis of the MGRN1 RING domain was done using ConSurf (Ashkenazy et al., 2016). Briefly, 200 MGRN1 homologs were collected from UniProt using the homolog search algorithm HMMER and a color coded multiple sequence alignment was built using ClustalW (Fig. S4B).

Immunoprecipitation and Western Blotting
Whole cell extracts from HEK293T and NIH/3T3 cells were prepared in Immunoprecipitation (IP) The resolved proteins were transferred onto a nitrocellulose membrane (Bio-Rad Laboratories) using a wet electroblotting system (Bio-Rad Laboratories) followed by immunoblotting.

Flow cytometry of live cells
As described above, a lentiviral expression system was used to stably express CD16/CD7/MEGF8 chimeras in Megf8 -/cells NIH/3T3 cells (diagramed in Fig. S5D). A modified live cell immunostaining protocol from Santa Cruz Biotechnology and Cell Signaling Technology was used to label and analyze cell surface CD16/CD7/MEGF8 chimeras (Fig. S5E)

SMO internalization assay
Cell surface internalization assay for SMO was performed as described previously for

Immunofluorescence staining of cells and tissue and image quantifications
Mouse embryos (e12.5) were harvested and fixed in 4% (w/v) paraformaldehyde (PFA) in 1x PBS for 2 hours at 4 °C and then rinsed thoroughly in chilled PBS. To cryopreserve the tissue, the embryos were transferred to 30% sucrose in 0.1M PB (pH 7.2) and allowed to equilibrate overnight. To allow for better analysis of the tissue, the embryos were further dissected into five pieces: 2 hands (forelimbs), head, upper body, and lower body. All five pieces were then mounted and frozen into Tissue-Plus OCT (optimal cutting temperature) compound (Thermo Fisher Scientific) and 12-14 µm sections were collected. Prior to staining, the tissue was blocked for 1 hour in immunofluorescence (IF) Blocking Buffer containing: 1% normal donkey serum (NDS) and 0.1% Triton-X diluted in 1x PBS. In a humidified chamber, the sections were incubated with primary antibodies overnight at 4 °C, rinsed 3 times in PBST (1x PBS + 0.1% Triton-X), incubated with secondary antibodies and Hoescht for 1 hour at room temperature, rinsed 3 times in PBST, and then mounted in Prolong Gold antifade mountant (Thermo Fisher Scientific, Invitrogen).
NIH/3T3 cells, pMEFs, and primary human fibroblasts were fixed in chilled 4% PFA in 1x PBS for 10 minutes and then rinsed with chilled PBS. Cells were incubated in IF Blocking Buffer for 30 minutes, primary antibodies for 1 hour, and secondary antibodies for 30 minutes.
Fluorescent images were acquired on an inverted Leica SP8 confocal microscope equipped with a 63X 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 4-8X optical zoom was used for imaging cilia to depict representative images. For the quantification of SMO at cilia, images were opened in Fiji 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.

Mouse embryo phenotyping analysis
Mouse embryos (e13.5-14.5) were fixed in 4% (w/v) PFA in 1x PBS for 2-3 days. Necropsy was performed to determine visceral organ situs (i.e. lung and liver lobation, heart and stomach situs, and spleen and pancreas structure). The samples were embedded in paraffin and processed for episcopic confocal microscopy as previously described (Liu et al., 2013). Briefly, this entailed sectioning of the tissue block using a Leica sledge microtome with serial images of the block face captured with a Leica confocal microscope. The serial two-dimensional (2D) image stacks generated were three-dimensionally (3D) reconstructed using the 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).

Variant Discovery and Validation
Genomic DNA was extracted from blood using the PAXgene Blood DNA kit (Qiagen). Patient genomic DNA was analyzed using whole exome sequencing performed using the Agilent V5 Exome Capture kit followed by sequencing with the Illumina HiSeq2000 with 150 base pairedend reads with 100X coverage. Reads were aligned to the human reference genome (version hg19) using Burrows-Wheeler Alignment (BWA, version 0.5.9) (Li and Durbin, 2009)

QUANTIFICATION AND STATISTICAL ANALYSIS
All data analysis and graphs were generated using GraphPad Prism 8. Violin plots were created using the "Violin Plot (truncated)" appearance function. In Prism 8, 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 8, all statistical tests were conducted using non-parametric methods, which do not assume a Gaussian distribution of the data. The statistical significance between two groups was determined using the Mann-Whitney test and the significance between three or more groups was determined using the Kruskal-Wallis test. For each figure, the error bars (representing the standard deviation) and p-values were all calculated using Prism 8 and reported in the figure legend. P-values were reported using the following key: not-significant (ns) p-value > 0.05, *p-value ≤ 0.05, **p-value ≤ 0.01, ***p-value ≤ 0.001, and ****p-value ≤ 0.0001

DATA AND CODE AVAILABILITY
The published article contains all datasets generated and analyzed during this study.
File S1: Newick tree file for the MGRN1-RNF157 family, Related to Figure 1B