Successful therapeutic intervention in new mouse models of frizzled 2-associated congenital malformations

Frizzled 2 (FZD2) encodes a transmembrane receptor that is an essential component in Wnt signal transduction (Chan et al., 1992; Zhao et al., 1995; Clevers, 2006; Clevers and Nusse, 2012; Rim et al., 2022). FZD2 is part of a diverse family of 10 frizzled receptors that interact with 19 Wnt ligands. Wnt signaling regulates a variety of developmental processes throughout all tissues in the body through a combination of spatio-temporal regulation of signaling molecule expression and differential ligand-receptor interactions. Many human syndromes are caused by the disruption of WNT or FZD genes, including autosomal dominant omodysplasia (ADO) (Borochowitz et al., 1991; Nagasaki et al., 2018; Saal et al., 2015; Tan et al., 2005; Turkmen et al., 2017; Warren et al., 2018) and autosomal dominant Robinow syndrome (AD-RS) (Birgmeier et al., 2018; Hosseini-Farahabadi et al., 2017; Person et al., 2010; Roifman et al., 1993, 2015; White et al., 2018); yet the mechanisms underlying these syndromes need further investigation in model organisms.

Our previous findings demonstrated the first evidence of FZD2 variants in human pathology (Saal et al., 2015). In two related individuals, a single nucleotide variant leading to a premature stop allele in FZD2 (W548*) resulted in ADO, a syndrome characterized by small stature due to the shortening of long bones, craniofacial dysmorphism, including cleft lip/palate, and genitourinary abnormalities. Subsequently, other reports have supported the conclusion that heterozygous variants in FZD2 can result in limb and craniofacial anomalies. Turkmen et al. reported a missense variant, G434V, in an individual with ADO (Turkmen et al., 2017). Shortly after, Nagasaki et al. identified an individual with a premature stop variant at the amino acid position directly upstream of W548: S547* (Nagasaki et al., 2018). Additional W548* and G434V variants were reported by White et al. (White et al., 2018) and Warren et al. (Warren et al., 2018) in individuals with AD-RS and ADO. White et al. (White et al., 2018) also identified G434S and W377* variants, and Zhang et al. reported another three related individuals with the W548* variant, as well as a F130Cfs*98 variant and G434S (Zhang et al., 2022).

Owing to the highly overlapping phenotypes seen in individuals classified with ADO and AD-RS, there has been some confusion as to how persons with FZD2-related skeletal phenotypes should be classified. Zhang et al. performed a Human Phenotype Ontology analysis for 16 subjects with FZD2 variants and concluded all individuals with FZD2 variants grouped with individuals that had known variants in other genes associated with Robinow syndrome (Zhang et al., 2022). They suggested that individuals with FZD2 variants that result in characteristic limb shortening and craniofacial malformations be classified as FZD2-associated AD-RS. We agree with this suggestion and use this nomenclature hereafter.

Two previous studies investigated FZD2 loss of function in a mouse model (Yu et al., 2010, 2012) and reported a partially penetrant phenotype of recessive cleft palate (50%) as well as an incompletely penetrant failure to thrive in the animals without cleft palate. Although these studies are consistent with the human FZD2-associated AD-RS phenotype, there are differences. The Fzd2 knockout animals were reported to have hypognathia but were not reported to have limb reductions. In addition, the homozygous knockout animals displayed more mild craniofacial phenotypes than the ROR2 knockout animals, and these phenotypes have been reported to phenocopy Robinow syndrome (Schwabe et al., 2004). We hypothesized that these differences might be due to the comparison of a Fzd2 null mouse model with a dominant human syndrome. These studies were unable to definitively determine whether the effects on Wnt signaling primarily compromised the canonical and/or non-canonical regions of the pathway. We therefore sought to further investigate the role of the FZD2 W548* patient variant in mice to identify whether disruption of only the C-terminal region of Fzd2 was phenotypically different than a complete null. The C-terminal region of FZD2 is highly conserved between mouse and human, which makes modeling the pathogenic allele in mouse particularly relevant to human FZD2 biology. Here, we present the generation of the mouse ortholog of this patient allele of FZD2, as well as two additional lines containing small deletions in the same C-terminal region of the protein. We present data to show that the orthologous W553* allele in mice recapitulates the autosomal dominant inheritance pattern seen in human, and that both the skeletal and craniofacial abnormalities from the human were observed in mouse. We also present data in which the canonical and noncanonical Wnt pathways are perturbed in these mouse mutants. Augmenting the canonical Wnt signaling pathway resulted in some rescue of several skeletal phenotypes.

To model the human FZD2^W548* (NP_001457.1) variant, we used CRISPR/Cas9-mediated genome editing to create the orthologous W553* change in mice (Fzd2^W553*; NP_065256.1). C57BL6/J zygotes were injected with sgRNAs and a DNA oligonucleotide donor. Sanger sequencing of the targeted region in mosaic founder animals revealed the presence of editing for the desired W553* knock-in as well as multiple indels (Fig. 1A,B). Crossing the mosaic founders to wild-type C57BL6/J mice resulted in the generation of stable lines of two small deletions. Fzd2^em1Rstot(D3) is a 3 bp deletion precisely removing the codon for serine 552 (S552) and is hereafter referred to as Fzd2^D3 (Fig. 1A,B). Fzd2^em2Rstot(D4) is a 4 bp deletion that results in an immediate frameshift at amino acid 552 and a stop codon following 60 residues of nonsense sequence and is hereafter referred to as Fzd2^D4 (Fig. 1A,B). Although there was some evidence of low levels of editing for W553* in founders, we were unable to recover this specific allele in any offspring to generate a stable line. All the alleles we generated are predicted to disrupt a region of the canonical dishevelled binding domain (KTxxxW, Fig. 1B) (Umbhauer et al., 2000).

In the maintenance of the Fzd2^D3 and Fzd2^D4 alleles by mating Fzd2 carriers with wild-type mice, we noted normal proportions of Fzd2^D3/+ mice but a significant reduction in Fzd2^D4/+ mice at weaning (Fig. 1C). Heterozygous matings of both the Fzd2^D3/+ and Fzd2^D4/+ mice revealed that both Fzd2^D3/D3 and Fzd2^D4/D4 homozygous embryos were recovered in approximately expected Mendelian ratios at late embryonic time points (E16.5-18.5) (Fig. 1D). We saw a slight reduction in the number of Fzd2^D4/D4 homozygote embryos (Fig. 1D); however, no Fzd2^D3/D3 or Fzd2^D4/D4 homozygous animals survived to weaning (Fig. 1D).

We were not able to create a stable mouse line that recapitulates the orthologous human variant Fzd2^W553*/+. This was expected, as heterozygotes for the W553* variant were predicted to have cleft palate, which leads to perinatal lethality in mice because of an inability to feed. We therefore performed an additional round of CRISPR/Cas9 zygotic injections in addition to several rounds of improved genome editing of oviductal nucleic acid delivery (i-GONAD) (Ohtsuka et al., 2018; Gurumurthy et al., 2019) to generate additional animals for phenotypic analyses. In these ‘F0’ experiments, we collected the resulting embryos at E16.5-18.5 to investigate phenotypes in animals that otherwise would be unavailable at weaning age due to the cleft palate phenotype.

From the second round of zygote injections, 20 embryos were recovered at E17.5 and Sanger sequencing showed evidence of genome editing for seven. Of these seven, five were mosaic (50-90%) for the desired W553* heterozygous knock in (Fig. 2A), and two were mosaic for multiple other indels that did not create the W553* edit.

All Fzd2^W553*/+-edited embryos had short frontonasal prominences (Fig. 2C,G,I) relative to wild-type littermate controls (Fig. 2B,F,H). The mouth opening was wider in Fzd2^W553*/+-edited embryos (Fig. 2E) when compared with littermates (Fig. 2D,O). Examination of the palates of these embryos revealed that all five embryos with the W553* variant had cleft palates, whereas none of the unedited embryos or embryos with other indels had cleft palates (Fig. 2J-M). Thus, we conclude that the presence of the Fzd2^W553*/+ allele is sufficient to cause a ‘dominant’ cleft palate phenotype despite the presence of some wild-type Fzd2 sequence via Sanger sequencing (Fig. 2A). The lower face was significantly wider in Fzd2^W553*/+-edited embryos relative to interorbital width (Fig. 2N).

In addition to facial dysmorphism and cleft palate, Fzd2^W553*/+-edited embryos weighed significantly less than their littermate controls (Fig. 2O). Because limb shortening is the most dramatic phenotype seen in human FZD2-associated AD-RS, we performed skeletal preparation staining on all ‘F0’ embryos to measure limb lengths. This skeletal analysis revealed that the lengths of the humeri, radii, ulnae, femora, tibiae and fibulae were significantly shorter in Fzd2^W553*/+-edited embryos when compared with littermate controls, even if normalized to embryo weight to control for size differences (Fig. 2P-AB). Although all Fzd2^W553*/+-edited embryos had shortened limbs, there was some variability in the severity of the shortening, with some embryos having very severe loss of bone structure (Fig. 2R). Overall, skeletal staining showed that the Fzd2^W553*/+ allele has a dominant effect on bone length. Although some individuals with Robinow syndrome have cardiac malformations, individuals with variants in FZD2 have not been reported to have heart defects. Consistent with this, visual inspection of the hearts from Fzd2^W553*/+-edited embryos did not reveal any grossly observable phenotypes (data not shown).

To efficiently generate additional embryos with the W553* variant to further confirm the penetrance of phenotypes, we used i-GONAD, which allows the direct injection of CRISPR/CAS9 reagents into the oviduct of a pregnant female mouse followed by oviductal electroporation. Using this technique, we generated an additional seven animals containing the Fzd2W553* modification and all animals had cleft palate, further confirming complete penetrance of the phenotype. Additionally, several embryos with unintended insertion/deletion alleles were generated and had cleft palate (Table S1). In-frame insertions did not affect phenotypes; however, small to large deletions near amino acid residue 553 resulted in embryos with cleft palate. Additionally, several missense variants led to cleft palate.

To determine the cause of the lethality in Fzd2^D3/D3, Fzd2^D4/D4 and Fzd2^D4/+ animals, we examined embryos throughout embryonic development (Fig. 3A-I, Fig. S1A-C). Histological examination of embryos revealed an incompletely penetrant cleft palate in Fzd2^D3/D3 embryos (Fig. S1C-E, n=25/31). However, the cleft palate was fully penetrant in all Fzd2^D4/D4 embryos (Fig. 3I-K, n=59/59). None of the wild-type, Fzd2^D4/+ or Fzd2^D3/+ heterozygous embryos exhibited palate closure or elevation defects (summarized in Fig. 3K, Fig. S1D). Measurement of the palate width when normalized to interorbital distance revealed that Fzd2^D4/D4 and cleft Fzd2^D3/D3 embryos had significantly wider palates than Fzd2^D4/+, Fzd2^D3/+, Fzd2^D4/+ or non-cleft Fzd2^D3/D3 embryos (Fig. 3J, Fig. S1E). This is grossly visible in some Fzd2^D4/D4 embryos at E13.5 (Fig. 3C, arrows). This finding suggests that the widening of the lower face is highly correlated with cleft palate status. As cleft palate results in perinatal lethality in mice, it is unsurprising that there are no surviving Fzd2^D4/D4 animals at weaning. However, another mechanism must be responsible for the postnatal lethality of the few non-cleft Fzd2^D3/D3 mice as well as the fraction of Fzd2^D4/+ animals that do not survive to weaning. In addition to palatal defects, Fzd2^D4/D4 had significantly shorter mandibles compared with wild-type controls (Fig. 3L-O). Interestingly, Fzd2^D3/D3 non-cleft animals had shorter mandibles, but animals with cleft palates had similar length mandibles relative to controls (Fig. S1F-I).

Individuals with FZD2-associated AD-RS have shortened limb elements. To determine whether Fzd2^D3 or Fzd2^D4 animals had limb phenotypes, E17.5 embryos were stained with Alcian Blue (for cartilage) and Alizarin Red (for bone), and the bone lengths were measured as a fraction of body weight. All limb elements from Fzd2^D3/D3 and Fzd2^D4/D4 embryos were statistically significantly decreased in length (Fig. 4A-J, Fig. S2A-J). Fzd2^D4/+ animals had limb lengths consistent with wild-type animals, whereas Fzd2^D3/+ embryos had statistically significant reduction in limb lengths, but not to the degree of Fzd2^D3/D3 animals.

The previously published Fzd2^null mouse was generated on the Sv129 background strain (Yu et al., 2010) and had a different phenotype from the CRISPR alleles generated in this study on a C57BL/6J (B6) background. We investigated whether genetic background possibly contributed to the differing phenotypic severity in the two studies. We crossed Fzd2^D4/+ mice to wild-type CD-1 mice, an outbred strain with many sequence variants when compared with B6. In this F1 generation, there were increased numbers of Fzd2^D4/+ (45%) as opposed to only 37% from the B6 experiment (Fig. 5A,B), suggesting that even a 50% contribution of the CD-1 outbred strain partially rescues the lethality phenotype seen in a significant proportion of the Fzd2^D4/+ animals on the B6 background. Upon mating the heterozygous F1 B6/CD-1 Fzd2^D4/+ mice together, Fzd2^D4/D4 homozygosity was still lethal after birth. However, embryonic evaluation revealed that 8/10 F1 B6/CD-1 Fzd2^D4/D4 animals had cleft palate (Fig. 5C), as opposed to the 100% penetrant cleft palate on the C57BL6/J background. Interestingly, a small proportion (3/26) of the F1 B6/CD-1 Fzd2^D4/+ heterozygous embryos developed cleft palate (Fig. 5C). Together, these data show that the phenotypes resulting from the Fzd2^D4 allele with a more dramatic disruption of the C-terminal region of FZD2 remain generally consistent, but with slightly increased variability, on mixed genetic backgrounds. Therefore, we conclude the differences we note in comparison with the null allele are likely due to different molecular mechanisms of action with the C-terminal modifications when compared with complete deletion of Fzd2.

To better understand whether the craniofacial phenotypes observed in the Fzd2^D4/D4 animals were related to changes in cell proliferation or cell death, we examined embryonic palatal shelves at E13.5 for proliferation (phospho-histone H3) and apoptotic cell death (cleaved caspase 3). Immunohistochemistry analysis did not identify a significant difference in either proliferation or apoptosis at E13.5 (Fig. 6A,B), suggesting that the palatal phenotype is not due to changes in cellular proliferation and/or death.

The mouse models generated in this study confirm that disruptions of the cytoplasmic tail of the FZD2 protein are sufficient to recapitulate the FZD2-associated AD-RS phenotypes first identified in human patients (Saal et al., 2015) and subsequently reported in other human patients (Warren et al., 2018; Nagasaki et al., 2018; Turkmen et al., 2017; White et al., 2018; Zhang et al., 2022). The ability to maintain the D4 line as heterozygotes allows the generation of additional embryos to explore the underlying molecular mechanisms. The effects on both canonical and non-canonical/PCP Wnt signaling are of particular interest, as previous efforts to explore this question have suggested that both pathways may be affected (Yu et al., 2010, 2012).

We first performed qRT-PCR on craniofacial tissues from E13.5 Fzd2^D4/D4 embryos for transcriptional targets of canonical Wnt signaling: Axin2 and Dkk1. We found that both are reduced (Fig. 7A,B). This finding and our previous results regarding reduced DVL recruitment with the truncated FZD2 protein upon Wnt signaling (Saal et al., 2015) are consistent with a model in which disruption of the FZD2 cytoplasmic tail prevents full binding of the DVL protein and disrupts normal effects on the β-catenin destruction complex. Interestingly, however, MEFs generated from Fzd2^W553* embryos showed no changes in canonical signaling, as measured by active (non-phosphorylated) β-catenin and phosphorylated LRP6 (Fig. 7C,D).

We then addressed the hypothesis that non-canonical Wnt signaling is affected. This is more challenging to assess as there are fewer well-established parameters to measure this form of Wnt signaling. We measured the length and width of the chondrocytes in the E13.5 limb as a measure of the ability of the cell to position itself within a field of growing cells. Previous reports have linked perturbations of non-canonical Wnt signaling with a reduction in the ability of a cell to lengthen parallel to the longitudinal axis of limb growth (Gao et al., 2011; Wang et al., 2011; Gao and Yang, 2013; Gao et al., 2018). We first measured the length and width of chondrocytes with wheat germ agglutinin and found an ∼50% reduction in the length/width ratio of the cells (Fig. 8A-C). We repeated the experiment with Safranin O staining with similar results (Fig. 8D-I). We conclude from these data that both canonical and non-canonical Wnt signaling are disrupted in FZD2 mutant tissues. Additionally, MEFs generated from Fzd2^W553* embryos showed altered baseline non-canonical signaling. Namely, phospho-VANGL2 was significantly decreased in MEFs from FZD2^W553* animals compared with wild-type controls (Fig. 8J,K).

To further compare the stability of the FZD2 variants, we transiently transfected plasmids directing the expression of mouse FZD2 variants (wild type, W553*, D3 and D4) with a C-terminal 1D4 tag into HEK293 T cells lacking all 10 FZDs. Immunoblotting for 1D4 revealed similar levels of FZD2 in cells transfected with wild type, W553*,and D3, but reduced levels of FZD2 protein in cells transfected with the FZD2 D4 variant (Fig. S4A,B). This suggests similar stability of the mutant FZD2 alleles, except for the D4 variant. More work is needed to determine the precise biochemical properties to explain why the W553* variant causes a more-dominant phenotypic penetrance than the D3 and/or D4 alleles.

A previous study had demonstrated the power of modulating the Wnt pathway in craniofacial development to rescue the cleft palate in Pax9 mutant mice (Li et al., 2017). We reasoned that the action of the IIIC3a molecule as an antagonist of the Wnt antagonist DKK may act to augment the pathway and partially rescue the genetic lesions in Fzd2. We tested the activity of IIIC3a in vitro and found that IIIC3a increased the canonical Wnt signaling reporter TOPFlash in a dose-dependent manner in the presence of WNT3A-conditioned media (Fig. S3A,B). We tested this hypothesis with the Fzd2^D4 allele and treated pregnant dams with daily 25 mg/kg intraperitoneal injections of IIIC3a from E10.5 to E14.5 and examined embryos at E17.5 (Fig. 9A). The body weight of the embryos did not differ significantly (Fig. 9H), and all Fzd2^D4/D4 homozygotes (n=4) still had cleft palate. We measured the long bones and mandible, and normalized to body weight (Fig. 9I-O). We found that all bones measured in IIIC3a-treated Fzd2^D4/D4 homozygotes were significantly longer than untreated Fzd2^D4/D4 homozygotes and very similar to all control genotypes. Thus, we conclude that, although both canonical and non-canonical Wnt signaling are affected, modulating the canonical Wnt pathway was able to rescue long bone length significantly.

In this study, we found that disruptions of the C-terminal region of FZD2 in mice leads to highly penetrant and severe craniofacial abnormalities resulting in pre-weaning lethality. Disruption of the DVL-binding domain (Umbhauer et al., 2000; Qi et al., 2017; Wong et al., 2003), even a single amino acid deletion (Fzd2^D3), was sufficient to lead to autosomal recessive cleft palate. More substantial disruption of the DVL-binding domain via a frameshift in this region (Fzd2^D4) leads to a similar phenotype with higher penetrance of cleft palate in Fzd2^D4/D4 mice, as well as an incompletely penetrant-dominant failure to thrive in Fzd2^D4/+ animals. Intriguingly, the most severe phenotypes were associated with the ortholog of the variant identified in the human genetics studies: Fzd2^W553*. For this allele, we could not generate live heterozygous or homozygous mice at P14; however, we were able to identify embryos via a series of independently generated ‘F0’ CRISPR-modified mice with mosaic Fzd2^W553* knock-in. Such Fzd2^W553*-edited embryos exhibited cleft palate, decreased weight and significantly shorter long-bone length compared with unedited littermates. This demonstrates a dominant inheritance mechanism for Fzd2^W553* and recapitulates virtually all the phenotypes observed in the human FZD2-associated AD-RS that we previously attributed to this variant.

Fzd2^D4/+ pups are born in normal Mendelian ratios, although it is clear by 1-2 weeks of age that a subset of mice are substantially smaller than their littermates. The mechanism of the incompletely penetrant failure to thrive phenotype and ultimate cause of death in non-cleft Fzd2^D4/+ and Fzd2^D3/D3 animals remains unclear. This failure to thrive was also seen previously in non-cleft Fzd2^null mice, where a thorough investigation again failed to yield a cause of death (Yu et al., 2010).

The mechanism of dominance resulting from C-terminal disruption of FZD2 also remains unclear. One potential hypothesis is that the C-terminal intracellular domain of FZD2 is disrupted while the sequence encoding the Wnt-binding extracellular domains of FZD2 remains intact. We hypothesize that if Wnt ligands can bind the C-terminal FZD2 mutants but are unable to mediate a downstream signaling cascade, FZD2 may act as a ‘sink’ for Wnt signaling components and disrupt signaling of other Fzd family members through their overlapping expression patterns and interactions with Wnt ligands that bind multiple frizzled proteins. There is also evidence that Fzd receptors form multimers in the membrane (Dann et al., 2001; Carron et al., 2003). In this way, one FZD2 protein with a C-terminal modification may ultimately affect the signaling capacity of many more proteins and produce a ‘dominant-negative’ effect. Additional studies are necessary to identify biochemical changes in FZD2-DVL binding interactions and the impact of FZD2-signaling disruption on other members of the WNT-FZD signaling pathway.

Pathogenic variants in ROR2 are associated with two diseases: autosomal recessive Robinow syndrome (ARRS) and autosomal dominant Brachydactyly, type B1 (BDB). The pathogenicity of the variant depends on where the variant is located within the ROR2 protein (Lima et al., 2022; Kirat et al., 2020; Schwabe et al., 2000; Gui et al., 2021). For example, variants that cause ARRS preferentially occur in exons that encode regions of the extracellular domains (and occasionally in the tyrosine-kinase-like domain). In contrast, BDB-associated mutations occur in inter-domain regions, likely contributing to their gain-of-function characteristics. Interestingly, the severity of phenotypes seen in individuals with FZD2 variants also depends on the location and type of variant (missense or truncation) (Zhang et al., 2022). Specifically, individuals with missense variants (e.g. within the 5th transmembrane of the FZD2 protein; Warren et al., 2018; White et al., 2018) tend to present with more severe phenotypes relative to those with truncations in the region on which we have focused the current study (immediately after the final transmembrane domain). Future work will seek to define whether these differences reflect differential interactions with effector proteins or an alteration in the formation of Wnt signalosomes (DeBruine et al., 2017).

Findings from an independently made mouse model with a single base pair insertion after FZD2-S553 resulting in a frameshift supports our conclusions that canonical and non-canonical signaling are altered in mouse models with frameshift mutations in the DVL-binding domain (Zhu et al., 2022 preprint). Our collective findings that both canonical and non-canonical Wnt signaling are affected by these Fzd2 variants suggests that pharmacological intervention targeted at the canonical arm of the pathway might ameliorate some of the phenotypes. Indeed, we saw that IIIC3a addition could restore more normal length to the long bones, but the cleft palate incidence was refractory to this intervention. This suggests that non-canonical effects on convergent extension during craniofacial development may be too profound for this treatment or that alternate doses should be attempted. Regardless of that potential result, the long-bone deficits are much more difficult to treat in these individuals than surgical repair of the cleft palate. Medical intervention would appear to be a very attractive option for therapeutic treatment, especially in a case where the genetic diagnosis could potentially be made years before long-bone growth is concluded in human development.

Overall, this study demonstrates an essential role for the signaling functions of the C-terminal region of FZD2 and validates the findings of FZD2-associated AD-RS we previously observed in humans. This work also continues to illustrate the power of using the approach of CRISPR/Cas9-mediated knock-in of human variants as an important tool for investigating presumably causative sequence variants. Importantly, the i-GONAD technique allows for the rapid production of these alleles and decreases the number of animals needed to generate embryos harboring variants of interest. In the future, when developing mouse models in which generating a stable line is impossible (similar to the FZD2-W553* variant), one could increase the precision and efficiency of editing by exploiting the wobble position(s) in codon use. By changing the codon without changing the amino acid on one allele, it is possible to generate a colony of heterozygous animals with an allele that is resistant to editing by repair templates targeting the wild-type allele. With regards to this study, it could be interesting to create inducible mouse models of FZD2 variants to allow maintenance of an animal colony and temporal- and tissue-specific expression of the FZD2 variant alleles.

As the availability of human exome/genome sequencing becomes higher due to ever-decreasing costs, many previously unreported rare variants will be identified. Although large-scale mouse knockout projects provide valuable information as to the function of understudied genes, most human sequence variation does not comprise highly-orchestrated deletions of entire coding sequences that lead to fully null alleles, but rather small single nucleotide variants (Collins et al., 1997; International HapMap Consortium, 2003; Kruglyak and Nickerson, 2001). As such, there is tremendous value in studying specific patient variants encoding either hypomorphic alleles or gain-of-function alleles that would not be adequately modeled by a complete loss of function. For some genes in which loss of function is well tolerated, studying such gain-of-function alleles may be the most efficient way to increase our understanding of the normal function of a particular gene.

All experiments using mice in this study were performed using ethically acceptable procedures as approved by the Institutional Animal Care and Use Committee at Cincinnati Children's Hospital Medical Center and Van Andel Institute. Mice were fed LabDiet 5021 mouse breeder diet and housed in ventilated cages with a 12 h light/12 h dark cycle.

Identification of optimal sgRNA targets was performed using the MIT genome engineering tool (http://crispr.mit.edu). The top three guide sequences were cloned into PX458 M; MK4 cells cultured under standard conditions (DMEM+10% FBS+1% P/S) and then transfected with each plasmid using Lipofectamine 3000 according to the manufacturer's protocol (Invitrogen). After 24 h, genomic DNA was isolated using NaOH lysis of cells. A region flanking the target region was PCR amplified, and the Surveyor mutation detection kit (IDT) was used according to the manufacturer's instructions to digest the re-annealed PCR product. The cleaved products were separated on an agarose gel that was quantified using Gel-Imager for PC. The sgRNA showing the highest levels of editing was used for all subsequent experiments.

Mouse zygotes (C57BL6/N strain) were injected with 200 ng/µl Cas9 protein (IDT and ThermoFisher), 100 ng/µl Fzd2-specific sgRNA (GCAAGACACTGCACTCGTGG) and 75 ng/µl single-stranded donor oligonucleotide (GTGGGCATCACGTCGGGCTTCTGGATCTGGTCCGGAAAAACTCTTCATTCTTGATGATAGTTCTACACTCGTCTCACCAACAGCCGGCATGGCGAGACCACTGTGTGAAGC; IDT, Iowa) followed by surgical implantation into pseudo-pregnant female (CD-1 strain) mice. Two rounds of microinjection were performed. The resulting live-born pups from the first round were weaned and used for additional screening and mating. Pregnant females from the second round were euthanized at E17.5 and embryos were screened for phenotypic changes and genetic modifications.

To rapidly generate additional embryos with the Fzd2^W553* variant, the improved genome editing of oviductal nucleic acid delivery (i-GONAD) technique was used (Gurumurthy et al., 2019; Ohtsuka et al., 2018). The night before surgery, B6C3F1/J (JAX stock 100010) males (8-16 weeks old) and females (7-12 weeks old) were mated. Vaginal plugs were detected the following morning at 09:00. Complete sgRNAs from IDT (CCGGCAAGACACTGCACTCG; crRNA and tracrRNA) were incubated with Cas9-mSA (Michalski et al., 2021 preprint) protein to generate ribonucleoprotein (RNP) complexes for 10 min at room temperature before the biotinylated template (ACATGATCAAATACCTCATGACGCTCATCGTGGGCATCACGTCGGGCTTCTGGATCTGGTCCGGCAAGACACTGCATTCATGAAGGAAGTTCTACACTCGTCTCACCAACAGCCGGCATGGCGAGACCACTGTGTGAAGCGGTCTCGCCTGCCTGCCGGGCTT; IDT) was added (Harms et al., 2014). The final i-GONAD mix contained 30 µM sgRNA: 0.6 µg/µl Cas9-mSA protein, 2 µg/µl BIO-template [in filtered 1 mM Tris-HCl (pH 7.5); 0.1 mM EDTA]. Females that were identified with vaginal plugs underwent the i-GONAD technique at ∼E0.7 (16:00). Anesthesia (1× Avertin, intraperitoneal, ∼45 units per 20 g mouse bodyweight) and analgesia (5 mg/kg carprofen, subcutaneous) were administered, and deep anesthesia tested via the toe pinch reflex. Under a surgical microscope, the mouse was placed dorsal side up, sprayed with 70% ethanol and wiped with a Kimwipe. A 10 mm incision was made just inferior to the ribs, through the skin centrally, and sprayed with 70% ethanol and wiped with a Kimwipe. The incision was adjusted to expose the right or left flank and a 5 mm incision was made through the muscle layer. A sterile surgical drape was placed over the incision, and the ovary was pulled out of the body cavity by grasping the fat pad. The fat pad was then clamped with an Aorta-Klemme to stabilize the tissue. Approximately 1.5 µl of appropriate CRISPR reagents were administered via mouth pipette to the oviduct just upstream of the ampulla containing the fertilized eggs. A 5 mm×5 mm Kimwipe strip soaked in 1×PBS was wrapped around the injected oviduct. The oviduct was placed firmly within the probes of the electrokinetic tweezers (NepaGene, CUY652P2.5×4) and electroporated with eight pulses at 50 V, 5 ms per pulse and 1 s interval in between pulses (BTX T820 square wave electroporator). The Kimwipe was removed after electroporation, and the ovary was placed back in the body cavity. The procedure was repeated on the other ovary/oviduct and the skin incision closed with two surgical staples. Animals were placed back in their cage on a warmer and monitored until completely recovered.

The resulting embryos or live born pups were screened for evidence of editing via PCR/Sanger sequencing of embryonic yolk sacs or tail clips. Genomic DNA was amplified using forward 5′-GTAAGCCAGCACTGCAAGAG-3′ and reverse 5′-GTGAAGGAGGGCACGGTG-3′ primers. Pups exhibiting editing of interest were then crossed to wild-type (C57BL6/J) mice and the resulting progeny were Sanger sequenced (CCHMC DNA Sequencing and Genotyping Core) to confirm the alleles generated. Sanger sequences were analyzed using ‘4-Peaks’ software (Mac). In silico analysis (Benchling.com) was used to analyze the consequences of each allele on the predicted putative proteins.

Fzd2^W553* animals generated via i-GONAD were genotyped by Next Generation Sequencing. Briefly, the targeted region (including ∼250 bp regions at both ends) was PCR amplified with primers, including unique 8 bp barcodes per sample (Table S2). The PCR products were run on a gel to confirm the purity and pooled before being sent for Amplicon EZ Sequencing (Genewiz/Azenta, please see the manufacturer's website for more technical details, www.genewiz.com). The reads were then aligned to the wild-type FZD2 DNA sequence with the aligner BWA-MEM. After demultiplexing the reads based on the barcodes, reads were visualized with the software IGV (BROAD Institute) to identify the modifications at the target region.

Embryos generated via the second round of microinjection or i-GONAD were collected at E16.5-E18.5 and analyzed for cleft palate. Embryos were imaged in PBS before preparing tissue for downstream histological assays, skeletal preparations or micro-CT analyses.

To identify potential survival defects in CRISPR-generated alleles, a het×het breeding scheme was used. Resulting pups were genotyped via PCR followed by either restriction enzyme digest/agarose gel electrophoresis or Sanger sequencing. Survival versus expected Mendelian ratios were scored at late embryonic stages (E15.5-E18.5), weaning age (P18-P28) and analyzed for significance using a χ² test (Excel) with P<0.05 as a threshold for significance. Examination and imaging of embryonic phenotypes was performed via whole-mount imaging in PBS, histological analyses and skeletal preparations.

H&E fixation in Bouin's solution followed by washes in 70% ethanol and H&E staining. For H&E staining, embryos were embedded in paraffin and cut to 10 µm sections before staining using standard techniques (Behringer et al., 2014). All images were taken via Zeiss Discovery.V8 Stereoscope. Paired images are shown at the same magnification.

Embryos were collected at E13.5, fixed in 4% PFA for 2 h followed by cryoprotection in 30% sucrose and embedding in OCT compound. 10 µm frozen sections were cut and transferred to glass slides. Antibody staining for phospho-histone H3 and cleaved-caspase 3 was performed using standard immunohistochemistry techniques. Briefly, sections were blocked in 4% normal goat serum (NGS) for 1 h, incubated overnight with anti-cleaved caspase 3 (CC-3; Cell Signaling 9661S; 1:300) or anti-phospho-Histone H3 (pHH3; Sigma H0412; 1:500) diluted in blocking buffer. After washing with PBS, slides were incubated for 1 h with Alexa Fluor 488 goat anti-rabbit IgG (Invitrogen, A11008, 1:500) diluted in blocking buffer. Sections were briefly stained with DAPI and coverslipped. Images were taken on a Nikon C2 confocal microscope. Paired images are shown at the same magnification.

Embryos were collected at E12.5, fixed in 4% PFA overnight and forelimbs embedded in OCT (Sakura) before cryosectioning. Samples were sagittal sectioned (10 µm) and stained for WGA (5 µg/ml in HBSS) for 3 min at room temperature. Slides were washed in PBS followed by staining with DAPI (ThermoFisher, 1 µg/ml) and mounting with ProLong Gold (Invitrogen). Paired wild-type and Fzd2^em1Rstot(D4) homozygote images were taken at the same magnification using a Nikon C2 confocal microscope. Length-to-width ratios were calculated for chondrocytes from the midshaft of the forelimb from three Fzd2⁺ and three Fzd2^em1Rstot(D4) homozygote embryos.

Femur and ulna from E12.5 were fixed in 4% PFA and processed using standard histology techniques. Sections (10 µm) were stained for Safranin O following a previously published protocol with modifications (Schmitz et al., 2010). Briefly, slides were deparaffinized, washed in distilled H₂O, and stained with Hematoxylin (Hematoxylin Solution, Harris Modified, Millipore Sigma) for 5 min. Slides were washed with tap water followed by five dips in acid ethanol (1:400 HCl in 70% ethanol) and washed again with tap water. Slides were then stained with 0.001% Fast Green FCF (C.I. 42053) for 5 min followed by 1% acetic acid for 10-15 s and 0.1% Safranin O for 5 min. Slides were then dehydrated and paired wild-type and Fzd2^em1Rstot(D4) homozygote images were taken at the same magnification using a Zeiss Discovery.V8 Stereoscope. Length-to-width ratios were calculated for chondrocytes from the midshaft of the forelimb or femur from three wild-type and three Fzd2^em1Rstot(D4) homozygote embryos.

Fzd2^D4/+ (C57BL/6J; Jackson Lab, strain 000664) background animals were crossed with wild-type CD-1 mice (Charles River, strain 022). Heterozygous D4/+ males and females from the F1 generation (50% B6, 50% CD-1) were crossed to produce an F2 generation that was phenotypically evaluated and genotyped to identify survival defects at both late embryonic time points (E15.5-E18.5) and at weaning (P18-P28).

Skeletons from E17.5 animals were stained for Alcian Blue and Alizarin Red to visualize cartilage and bone, respectively. Briefly, embryos were eviscerated and fixed for 2 days in 95% ethanol. They were stained overnight at room temperature in 0.03% (w/v) Alcian Blue solution (Sigma-Aldrich, A3157) containing 80% ethanol and 20% glacial acetic acid. Samples were destained in 95% ethanol for 24 h followed by pre-clearing in 1% KOH overnight at room temperature. Skeletons were then stained overnight in 0.005% Alizarin Red solution (Sigma-Aldrich, A5533) containing 1% KOH. A second round of clearing was performed by incubating tissues in 20% glycerol/1% KOH solution for 24 h. Finally, they were transferred to 50% glycerol/50% ethanol for photography. Skeletal preparations were imaged using a Zeiss Discovery.V8 Stereoscope and length measurements were recorded for mandibular bones, humeri, ulnae, radii, femora, tibiae and fibulae.

Skulls and limbs from E18.5 Fzd2^W553* animals were examined using the SkyScan 1172 micro-computed tomography (µCT) system (Bruker MicroCT; Kontich, Belgium). Tissues were fixed in 10% neutral-buffered formalin (NBF) at room temperature for 48 h, stored in 70% ethanol and kept in the 70% ethanol solution for scanning. Specimens were scanned using an X-ray voltage of 50 kV, current of 201 µA and 0.5 mm aluminum filter. The samples were imaged using 2000×1200 pixel resolution and 8 µm image pixel size, and reconstructed using NRecon 1.7.4.6 (Bruker MicroCT). For each sample, the volume of interest (VOI) was defined using DataViewer 1.5.6.3 (Bruker MicroCT) and a region of interest (ROI) was defined using CTAn 1.18.8.0 (Bruker MicroCT). The defined ROI was used to generate a 3D surface-rendered model in CTAn that was then visualized and manipulated in CTVol 2.2.1.0 (Bruker MicroCT).

qRT-PCR was performed using standard techniques. RNA was extracted from craniofacial tissues of E13.5 Fzd2⁺ and Fzd2^D4/D4 littermate embryos using Qiagen RNA extraction kit. RNA was reverse transcribed using Superscript III First-Strand Synthesis System for RT-PCR (Invitrogen), and cDNA was amplified and detected using TaqMan Universal PCR master mix (Applied Biosystems) and TaqMan probes, including mouse Axin2 (Mm00443610_m1), mouse Dkk1 (Mm00438422_m1) and mouse Gapdh (Mm99999915_g1). Real-time PCR was analyzed on Applied Biosystems QuantStudio 6 (ThermoFisher).

HEK293-Super TOPFlash (STF) [a gift from Jeremy Nathans (Xu et al., 2004)] or C3H10T1/2-STF (ATCC CCL-226) cells were treated with 1 µM LGK-974 (Cayman Chemical 14072) for 48 h to inhibit endogenous Wnt secretion. Dkk inhibitor II (IIIC3a; Calbiochem; EMD_BIO-317701) was added at varying concentrations (0-250 µM) with CTRL CM (1/10 dilution) or WNT3A CM (1/10 dilution) made from L cells (ATCC L Cells CRL-2648, ATCC L-Wnt-3A CRL-2647, ATCC L Wnt-5A CRL-2814). Transactivation of a β-catenin/TCF-responsive reporter construct was measured using the Promega Dual-Luciferase Reporter (DLR) Assay to identify activated canonical Wnt signaling. Each treatment had four technical replicates and the experiments were performed twice with consistent results.

Dkk Inhibitor II (IIIC3a; Calbiochem; EMD_BIO-317701) was administered (25 mg/kg; intraperitoneal) to pregnant females on E10.5, E11.5, E12.5, E13.5 and E14.5. A stock solution of IIIC3a was made at 25 mg/ml in DMSO and diluted 1:10 in 1× Dulbecco’s PBS (2.5 mg/ml) for a working solution on the day of injections.

Following i-GONAD surgery to induce the W553* variant in mice, E12.5 embryos were harvested to generate MEFs according to standard protocols (Qiu et al., 2016), and yolk sacs were collected for Sanger sequencing and next-generation sequencing. Briefly, E12.5 embryos were minced in 0.05% Tryspin-EDTA (Gibco) and incubated for 30 min at 37°C. Cells were then neutralized with DMEM with 10% fetal bovine serum (FBS) and 1% penicillin/streptomycin, and cultured for 48 h in T75 flasks. Cells used for experiments were passage 3 or less. To test canonical and noncanonical signaling changes in MEFs, cells were first treated with LGK-974 (1 µM) for 48 h to inhibit endogenous Wnt ligand secretion. MEFs were then serum starved for 24 h (with LGK-974), followed by a 24 h treatment with CM made from L cells. CM was diluted 1/5 in DMEM with 10% FBS. Lysates were harvested for immunoblot analyses.

HEK293T cells that lack all 10 FZDs (FZD^less; Eubelen et al., 2018) were transfected with constructs encoding chimeric mouse FZD2 cDNAs (wild-type, W553*, D3 or D4) with a C-terminal 1D4 tag or control plasmid (eGFP). Using pRK5-mFzd2 (Addgene 42254) as the backbone plasmid and the Q5 Site-Directed Mutagenesis Kit (NEB), we designed primers using https://nebasechanger.neb.com/ to generate corresponding mutations in the Fzd2 gene. FZD2-1D4 plasmid DNA (0.25 µg) was diluted with serum-free Opti-MEM media (Gibco, 31985062). X-tremeGENE HP DNA Transfection Reagent (Roche, 06366236001) was added to the diluted DNA at a 1:1 ratio and incubated at 25°C for 15 min. The transfection complex was added to freshly seeded HEK293T FZD^less cells in a drop-wise manner. Cell lysates were harvested 48 h post-transfection.

HEK293T FZD^less cells or MEFs were rinsed once with ice-cold 1×PBS and lysed with lysis buffer [50 mM Na₂HPO₄, 1 mM sodium pyrophosphate, 20 mM NaF, 2 mM EDTA, 2 mM EGTA, 10 mM NaCl, 1% Triton X-100, 1 mM DTT and cOmplete Mini EDTA-free Protease Inhibitor Cocktail (Roche)]. Cells were centrifuged at 21,000 g for 20 min at 4°C and the supernatant collected. Samples were resolved on Mini-PROTEAN TGX Stain-Free Gels (Bio-Rad, 456026) and transferred using the Trans-Blot Turbo transfer system (Bio-Rad). The membranes were blocked with EveryBlot Blocking Buffer (Bio-Rad, 12010020) for 30 min.

MEF blots were probed with non-phospho (active) β-catenin (Cell Signaling Technologies, 8814, Lot 6; 1:1000) or phospho-LRP6 (Cell Signaling Technologies, 2568, Lot 6; 1:1000) primary antibodies followed by secondary antibody detection with anti-rabbit IgG, HRP-linked antibody (Cell Signaling Technologies, 7074, Lot 29; 1:1000). To assess non-canonical signaling, blots were probed with phospho-VANGL2 (Invitrogen, MA5-38242, Lot XB3512685A; 1:1000) antibody followed by anti-mouse IgG HRP-linked antibody (Cell Signaling Technologies, 7076, Lot 22; 1:1000) and chemiluminescent visualization on a Bio-Rad ChemiDoc. Β-ACTIN-HRP conjugate (Cell Signaling Technologies, 5125; Lot 6; 1:2000) was used as an endogenous control. Total protein was used for normalization of active β-catenin (Fig. S5).

HEK293T FZD^less transfection blots were probed with anti-Rhodopsin antibody [1D4] (Abcam, ab5417, Lot GR3426077-3; 1:1000) followed by secondary antibody detection with anti-mouse IgG HRP-linked antibody (Cell Signaling Technologies, 7076, Lot 38; 1:1000) and chemiluminescent visualization on a Bio-Rad ChemiDoc.

All statistical analyses were performed using GraphPad Prism 9.3.1. Ordinary one-way ANOVA with Dunnett's post-hoc multiple comparisons test was performed for comparison of D3/D4 heterozygotes and homozygotes to wild-type samples. Ordinary one-way ANOVA with Tukey's post-hoc multiple comparison test was performed for IIIC3a treatment studies. An unpaired t-test was performed for comparison of W5553* heterozygotes or D4 homozygotes with wild-type samples. P values are indicated on all graphs. The data are mean±95% confidence interval. No samples were excluded from analyses.

All experiments using mice in this study were performed according to ethically acceptable procedures, as approved by the Institutional Animal Care and Use Committees at Cincinnati Children's Hospital Medical Center and Van Andel Institute.

We thank the Cincinnati Children's Hospital Medical Center (CCHMC) Transgenic Core (RRID:SCR_022642) for assistance with zygote injections of CRIPSR reagents for the Fzd2 W553*, D3 and D4 alleles. We thank the VAI Transgenic Core (RRID:SCR_022914) for providing training on the i-GONAD injections used to generate additional Fzd2 W553* embryos, and the VAI Vivarium staff for their assistance with animal care and husbandry.