Saturation mutagenesis defines novel mouse models of severe spine deformity

Embryonic formation and patterning of the vertebrate spinal column requires coordination of many molecular cues. After birth, the integrity of the spine is impacted by developmental abnormalities of the skeletal, muscular, and nervous systems, which may result in deformities such as kyphosis and scoliosis. We sought to identify novel genetic mouse models of severe spine deformity by implementing in vivo skeletal radiography as part of a high-throughput saturation mutagenesis screen. We report selected examples of genetic mouse models following radiographic screening of 54,497 mice from 1,275 pedigrees. An estimated 30.44% of autosomal genes harbored predicted damaging alleles examined twice or more in the homozygous state. Of the 1,275 pedigrees screened, 7.4% presented with severe spine deformity developing in multiple mice, and of these, meiotic mapping implicated ENU alleles in 21% of pedigrees. Our study provides proof-of-concept that saturation mutagenesis is capable of discovering novel mouse models of human disease, including conditions with skeletal, neural, and neuromuscular pathologies. Furthermore, we report a mouse model of skeletal disease, including severe spine deformity, caused by recessive mutation in Scube3. By integrating results with a human clinical exome database, we identified a patient with undiagnosed skeletal disease who harbored recessive mutations in SCUBE3, and we demonstrated that disease-associated mutations are associated with reduced trans-activation of Smad signaling in vitro. All radiographic results and mouse models are made publicly available through the Mutagenetix online database with the goal of advancing understanding of spine development and discovering novel mouse models of human disease. D is ea se M o de ls & M ec ha ni sm s • D M M • A cc ep te d m an us cr ip t INTRODUCTION Development of the spinal column is a highly regulated process that begins early in embryogenesis. Post-natal growth and maintenance of a structurally sound spine requires coordinated integration of the vertebrae (skeleton), inter-vertebral discs, nerves, and muscles. Disruption of any of these components may lead to spinal malformations. In humans, scoliosis, defined as a lateral curvature of the spine, is the most common spine deformity and is characterized as idiopathic, congenital, neuromuscular, or syndromic (Herring, 2013). Alternatively, kyphosis manifests as an anterior-posterior curvature of the spine, sometimes in combination with scoliosis (so-called kyphoscoliosis). Unlike idiopathic scoliosis, congenital scoliosis can be associated with skeletal malformations or segmentation defects of the vertebrae, and neuromuscular scoliosis is associated with neurological or muscular disease etiologies. Each of these disease types present with different scoliotic deformities and may require different treatment paradigms. Population-based genome-wide association studies have identified several loci associated with idiopathic scoliosis, while next-generation sequencing studies have elucidated rare mutations underlying congenital scoliosis (Chen et al., 2020; Gao et al., 2007; Khanshour et al., 2018; Kou et al., 2013; Sharma et al., 2011; Sharma et al., 2015; Takahashi et al., 2011; Wu et al., 2015). Despite the success of human genetic studies, modeling spine deformities for these loci in mice has proven challenging. For example, the correlation of human scoliosis with nongenetic factors, such as the adolescent growth spurt, hormonal influences during adolescence, and the distribution of mechanical forces of a bipedal spine, may be difficult to recapitulate in mice or other model systems. Moreover, certain genes implicated in severe forms of scoliosis, such as TBX6, may be sub-viable in genetically engineered knockout mouse models (Chapman and Papaioannou, 1998; Wu et al., 2015). Alternative strategies, such as using Cre-inducible conditional knockout mouse models or engineering hypomorphic alleles, have been used successfully to investigate scoliosis-associated genes in vivo (Karner et al., 2015; Yang et al., 2019). These strategies, however, are limited due to a lack of knowing a priori the relevant cell type for conditional models or which alleles may be hypomorphic and allow for viable mouse models of spine deformity in essential genes. D is ea se M o de ls & M ec ha ni sm s • D M M • A cc ep te d m an us cr ip t To facilitate discovery of genes required for proper spine development, we performed live-animal radiography to detect severe spine deformity as part of a saturation mutagenesis skeletal screen in mice (Rios et al., 2021; Wang et al., 2015). As proof-of-concept, we report selected results from 54,497 mice from 1,275 pedigrees. As previously described (Wang et al., 2018), estimates of saturation based on the standard of detecting lethal effects in a curated collection of essential genes suggests that 30.44% of all protein encoding autosomal genes were mutated to a state of detectable hypomorphism and examined in the homozygous state twice or more in this sample of mice. Among the mutations were 5,746 putative null alleles in 4,320 genes examined twice or more in the homozygous state. Mutations in genes regulating proper skeletal, muscular, and nervous system development are all implicated in scoliotic manifestations discovered through our screen. Results from our study may inform human genetic studies, provide novel mouse models of human disease, and advance knowledge of spine development and deformity. As proof-of-principle, we cross-referenced results from our study with a database of undiagnosed subjects who underwent clinical exome sequencing, and we report a patient with a skeletal syndrome caused by recessive mutations in the SCUBE3 gene that resulted in reduced hetero-dimerization and reduced Smad signal transduction. All results and mouse models, including others not reported here, are made publicly available through the online Mutagenetix database (mutagenetix.utsouthwestern.edu).


INTRODUCTION
Development of the spinal column is a highly regulated process that begins early in embryogenesis. Post-natal growth and maintenance of a structurally sound spine requires coordinated integration of the vertebrae (skeleton), inter-vertebral discs, nerves, and muscles.
Disruption of any of these components may lead to spinal malformations. In humans, scoliosis, defined as a lateral curvature of the spine, is the most common spine deformity and is characterized as idiopathic, congenital, neuromuscular, or syndromic (Herring, 2013).
Alternatively, kyphosis manifests as an anterior-posterior curvature of the spine, sometimes in combination with scoliosis (so-called kyphoscoliosis). Unlike idiopathic scoliosis, congenital scoliosis can be associated with skeletal malformations or segmentation defects of the vertebrae, and neuromuscular scoliosis is associated with neurological or muscular disease etiologies. Each of these disease types present with different scoliotic deformities and may require different treatment paradigms.
Population-based genome-wide association studies have identified several loci associated with idiopathic scoliosis, while next-generation sequencing studies have elucidated rare mutations underlying congenital scoliosis (Chen et al., 2020;Gao et al., 2007;Khanshour et al., 2018;Kou et al., 2013;Sharma et al., 2011;Sharma et al., 2015;Takahashi et al., 2011;Wu et al., 2015). Despite the success of human genetic studies, modeling spine deformities for these loci in mice has proven challenging. For example, the correlation of human scoliosis with nongenetic factors, such as the adolescent growth spurt, hormonal influences during adolescence, and the distribution of mechanical forces of a bipedal spine, may be difficult to recapitulate in mice or other model systems. Moreover, certain genes implicated in severe forms of scoliosis, such as TBX6, may be sub-viable in genetically engineered knockout mouse models (Chapman and Papaioannou, 1998;Wu et al., 2015). Alternative strategies, such as using Cre-inducible conditional knockout mouse models or engineering hypomorphic alleles, have been used successfully to investigate scoliosis-associated genes in vivo (Karner et al., 2015;Yang et al., 2019). These strategies, however, are limited due to a lack of knowing a priori the relevant cell type for conditional models or which alleles may be hypomorphic and allow for viable mouse models of spine deformity in essential genes.

Disease Models & Mechanisms • DMM • Accepted manuscript
To facilitate discovery of genes required for proper spine development, we performed live-animal radiography to detect severe spine deformity as part of a saturation mutagenesis skeletal screen in mice (Rios et al., 2021;Wang et al., 2015). As proof-of-concept, we report selected results from 54,497 mice from 1,275 pedigrees. As previously described (Wang et al., 2018), estimates of saturation based on the standard of detecting lethal effects in a curated collection of essential genes suggests that 30.44% of all protein encoding autosomal genes were mutated to a state of detectable hypomorphism and examined in the homozygous state twice or more in this sample of mice. Among the mutations were 5,746 putative null alleles in 4,320 genes examined twice or more in the homozygous state. Mutations in genes regulating proper skeletal, muscular, and nervous system development are all implicated in scoliotic manifestations discovered through our screen. Results from our study may inform human genetic studies, provide novel mouse models of human disease, and advance knowledge of spine development and deformity. As proof-of-principle, we cross-referenced results from our study with a database of undiagnosed subjects who underwent clinical exome sequencing, and we report a patient with a skeletal syndrome caused by recessive mutations in the SCUBE3 gene that resulted in reduced hetero-dimerization and reduced Smad signal transduction. All results and mouse models, including others not reported here, are made publicly available through the online Mutagenetix database (mutagenetix.utsouthwestern.edu).

Radiographic screen for spine deformity
The breeding scheme for generating and screening mice harboring ENU-induced alleles is shown in Fig. 1. C57BL/6J male mice are mutagenized with ENU and out-crossed to nonmutagenized C57BL/6J females, resulting in male pups (termed the G1 generation) heterozygous for germline ENU-induced alleles. G1 male mice are out-crossed to C57BL/6J female mice, and G2 pups are subsequently back-crossed to their G1 sire to produce G3 mice homozygous for the reference allele, heterozygous for the ENU allele, or homozygous for the ENU allele. All G3 mice undergo radiographic screening, including both dorsal and lateral radiography (Fig. 1).
Phenotypic scoring is performed from radiographs as either presence or absence of any rigid spine deformity or malformation, such as kyphosis, scoliosis, or kinked/curly tail.

Disease Models & Mechanisms • DMM • Accepted manuscript
ENU-induced alleles are detected by exome sequencing G1 male mice, as previously described (Wang et al., 2015). Nonsynonymous variants identified relative to the C57BL/6J reference genome are genotyped by targeted capture and sequencing of all G2 and G3 mice.
Automated meiotic mapping tests the null hypothesis that ENU-induced alleles segregating within the pedigree are not associated with risk for spine deformity. Significantly associated loci are identified following Bonferroni correction for the number of ENU-induced alleles in the pedigree.
Of the 1,275 pedigrees evaluated, 94 (7.3%) pedigrees were identified with at least 2 mice scored with a spine deformity phenotype. Of these, 20 (20.6%) were mapped to ENUinduced alleles (Fig. 1, Table S1), including 19 recessive and 1 dominant allele. We sought to evaluate variables limiting detection of significantly associated loci, such as the total number of mice in the pedigree and the total number of affected mice in the pedigree. Pedigrees with mapped alleles had significantly higher numbers of affected mice (Wilcoxon p=1.59e -5 ) compared to unmapped pedigrees (Fig. S1A), and this remained significant among only recessive alleles (Wilcoxon p=4.41e -5 ). Total numbers of mice were unchanged between pedigrees with mapped and unmapped alleles (Fig. S1B). Most (12/19; 63%) pedigrees with mapped recessive alleles had at least 80% penetrance of the spine deformity phenotype among homozygous G3 mice (Table S1).

Mouse model and case report of SCUBE3-associated skeletal disease
We identified the Scube3 C301Y allele associated with recessive spine deformity, including severe kyphosis and kinked tail ( Fig. 2A). While the homozygous Scube3 C301Y allele cosegregated with homozygous Rhot2 V604A and Nrde2 D607E alleles, we considered these to be less likely candidates based on the presence of skeletal phenotypes in mice homozygous for a previously-reported Scube3 N294K allele (Fuchs et al., 2016). Additionally, the Scube3 C301Y allele was predicted to be more damaging by PolyPhen analysis than the Rhot2 V604A and Nrde2 D607E alleles.

Disease Models & Mechanisms • DMM • Accepted manuscript
The human SCUBE3 gene, encoding the Signal peptide, CUB domain, and EGF like domain containing 3 protein, is highly expressed in osteoblasts (Wu et al., 2004). The SCUBE3 protein was identified as a transforming growth factor-beta (TGF-β) receptor ligand and activator of Smad signaling (Wu et al., 2004;Wu et al., 2011). Proper regulation of the TGF-β signaling cascade is essential for normal skeletal development (Crane and Cao, 2014). Disruption of TGFβ signal transduction leads to defects in endochondral ossification via dysregulation of osteogenesis, which has been studied extensively using germ-line and conditional mouse models (Wu et al., 2016). Furthermore, disruption of TGF-β signaling affects the cross-talk between bone forming osteoblasts and bone resorbing osteoclasts during bone remodeling (Erlebacher and Derynck, 1996;Filvaroff et al., 1999). Mice homozygous for a Scube3 N294K allele developed mild phenotypic abnormalities, including shorter hindlimbs, rib defects, and auditory abnormalities, while Scube3 -/mice were recently shown to develop a post-natal dwarfism that was not apparent during embryonic development (Fuchs et al., 2016;Lin et al., 2021;Xavier et al., 2013).
Although all three Scube genes (Scube1-3) are implicated in regulating proper skeletal development in mice (Fuchs et al., 2016;Lin et al., 2015;Tu et al., 2008), mutations in any one of three human SCUBE gene (SCUBE1-3) have yet to be associated with human disease.
However, during the course of this study, multiple subjects with recessive SCUBE3 mutations were reported with short stature, oral-facial abnormalities, skeletal abnormalities, and, although variable, spine deformity, suggesting a novel SCUBE3-associated skeletal disease (Lin et al., 2021). To identify additional patients with skeletal disease potentially attributable to mutations in SCUBE3, we queried the Baylor Genetics Laboratory database that includes nonsynonymous variants identified from clinical exome sequencing and clinician-provided clinical synopses. We sought patients described with skeletal abnormalities yet remaining without a genetic diagnosis following evaluation by clinical exome sequencing among whom rare nonsynonymous variants in SCUBE3 were identified. We identified a now 16-year-old male patient presenting with GI malrotation, hearing loss, short stature, joint hyperextensibility, dilated aortic root with abnormal aortic valve, and oral-facial differences including cupped ears, high-arched palate, and micrognathia ( Table 1). Skeletal abnormalities included asymmetric fusion of a pseudoepiphysis of the 2 nd metacarpal bilaterally (Fig. 2B). His feet were notable for an unusual fusion of the 2 nd metatarsal to the 2 nd cuneiform (Fig. 2B) differences in trans-activation were not due to differences in SCUBE3 secretion (Fig. 2G, S2D).

Proteasome dysfunction
The 20S proteasome is a multimeric complex composed of seven alpha (PSMA1-7) and seven beta (PSMB1-7) subunits (Demartino and Gillette, 2007 (Fig. 4C,D). These mice also variably developed ataxia and a wobbling gait (P REC =1.15e -6 ; Supplementary Movie), suggestive of a neurologic deficit, and were significantly smaller (Fig.   S3). It remains to be determined whether the Psma5 V146G alele reported here is a complete lossof-function or hypomorphic, as Psma5 is likely an essential gene, and knockout mice have not been characterized.

Myotonic dystrophy
Neuromuscular conditions, such as muscular or myotonic dystrophies, present increased risk for spine deformities. For example Becker-type myotonic dystrophy is a skeletal muscle disorder caused by recessive mutations in the CLCN1 gene, which encodes the chloride voltagegated channel 1 protein (Lorenz et al., 1994). In contrast, dominant-negative mutations in the same gene cause Thomsen-type myotonic dystrophy, though the recessive Becker-type is more common and more severe. Both diagnoses present with a general inability to relax skeletal muscles throughout the body, and some patients with CLCN1 deficiency have been noted to develop spine deformity (Skalova et al., 2013). The CLCN1 homodimer functions as a chloride ion channel chiefly responsible for opposing excitatory signals in resting skeletal muscle. In mice, a spontaneous Clcn1 mutant line (Clcn1 adr ) was described with myotonia, muscle weakness, reduced growth, and brittle bones (Watkins and Watts, 1984). We identified significant association of the homozgygous Clcn1 V292A allele with spine deformity (Fig. 5A,B).
The Clcn1 V292A allele co-segregated with a missense allele in the olfactory receptor 13 gene (Olfr13 Y59H ); however, the phenotypic resemblance of these mice with the Clcn1 adr line and the lack of any published characterization of this olfactory receptor led us to implicate the  Grewal et al., 2001;Lane et al., 1976). We identified significant association of the Large1 Q359X nonsense allele with recessive spine deformity (Fig.   5D,E). Interestingly, the Large1 Q359X allele was not significantly associated with post-natal lethality. The variant is located in the middle of the first glycosyltransferase domain and is predicted to result in complete loss of function.

DISCUSSION
Discovering the genetic causes of human spine deformities has been elusive until recently, due, in part, to technological advances such as genome-wide association studies and next-generation sequencing (Grauers et al., 2016;Wise et al., 2020). Understanding the genetic and mechanistic basis of human spinal disorders may be accelerated by forward genetic screens in model systems. To identify novel genetic mouse models of spine deformity, we performed a live-animal radiography screen in 54,497 mice collectively harboring predicted damaging or loss-of-function N-ethyl-N-nitrosourea (ENU)-induced mutations in at least two homozygous mice across 30.44% of autosomal genes. ENU mutagenesis in mice provides unique advantages.
First, the saturation mutagenesis approach employed here enables high-throughput generation and testing for associated phenovariance with predicted deleterious and loss-of-function alleles throughout the mouse genome (Wang et al., 2018;Wang et al., 2015). By integrating results with a clinical exome database of undiagnosed patients, we report a patient with bi-allelic SCUBE3 mutations presenting with skeletal abnormalities. The implication of SCUBE3 as a skeletal disease-causing gene was supported by our identification of mice homozygous for the Scube3 C301Y allele developing severe skeletal defects. In the course of this study, multiple patients were reported with recessive SCUBE3 mutations presenting with skeletal and non-skeletal manifestations similar to the index patient reported here (Lin et al., 2021).
Using two independent experimental assays, we show that missense variants identified in our mouse model and the undiagnosed patient reported here, as well as other missense variants recently reported (Lin et al., 2021), abrogated the interaction between SCUBE3 and SCUBE1 and reduced trans-activation of Smad signaling. These results demonstrate that the characterization of novel mouse models may inform as-yet undiscovered rare human genetic diagnoses when integrated with patient-based exome sequencing studies.
Results from our study also suggest novel mouse models of human Mendelian skeletal diseases may be readily identified from the Mutagenetix database. As proof-of-concept, we present novel mouse models for multiple known human diseases representing different developmental etiologies, including segmentation defects (Hes7), leukodystrophy (Galc), and Disease Models & Mechanisms • DMM • Accepted manuscript neuromuscular disease (Clcn1 and Large1). Though selected examples of severe spine deformity are presented here, the Mutagenetix database facilitates discovery of, and public access to, ENUinduced alleles and CRISPR-generated mouse lines associated with a spectrum of severe and mild (i.e., kinked tail) spine deformity phenotypes.
Other mutagenesis efforts have been undertaken to identify alleles associated with spine deformity in non-rodent model systems, such as zebrafish (Gray et al., 2020;Henke et al., 2017).
Such large-scale mutagenesis efforts can rapidly resolve causal alleles with the integration of next-generation sequencing (Andrews et al., 2012;Wang et al., 2015). Furthermore, in vivo gene editing using CRISPR/Cas9 to simultaneously engineer knockout and allele-specific knock-in models provides for rapid confirmation of suspected disease-causing alleles. By integrating with ongoing large-scale human genome sequencing efforts, in vivo saturation mutagenesis screens have potential to advance discovery of novel rare and ultra-rare human disease-causing genes.

Skeletal mouse screen
The mutagenesis and breeding strategy are previously described (Wang et al., 2018;Wang et al., 2015). Briefly, male mice (designated G0 generation) are mutagenized with Nethyl-N-nitrosourea (ENU) and out-crossed to C57BL/6J female mice. Transmitting ENUinduced alleles are detected in the resulting G1 male mice by exome sequencing. G1 male mice are out-crossed to C57BL/6J females, resulting in G2 mice heterozygous for ENU-induced alleles. Female G2 mice are back-crossed to their G1 sire, and all G2 and resulting G3 mice are genotyped by high-throughput targeted capture and next-generation sequencing.
All G3 mice are screened for spine deformity by live-animal radiography. Mice are anesthetized using inhaled isoflurane and imaged using a Faxitron UltraFocusDXA instrument.
Spine deformities are scored as a binary variable by visualizing dorsal and lateral X-ray images.
Following phenotype analysis, automated meiotic mapping identifies ENU-induced alleles associated with the spine deformity phenotype using dominant and recessive inheritance models (Wang et al., 2015).

Disease Models & Mechanisms • DMM • Accepted manuscript
All procedures were approved by the UT Southwestern Medical Center Institutional Animal Care and Use Committee.

Automated meiotic mapping of ENU alleles
ENU alleles are genotyped in G2 and G3 mice using a targeted capture and massively parallel sequencing approach. Genotypes of G2 and G3 mice are integrated with skeletal phenotypes, and the Linkage Analyzer tool tests the null hypothesis that each ENU allele within a pedigree has no association with the spine deformity phenotype using dominant and recessive models, as previously described (Wang et al., 2015). Statistical significance is assessed using logistic regression and p-values are reported using Wald tests. Results are visualized with Manhattan plots, and significantly associated alleles are those exceeding Bonferroni correction for the number of ENU alleles tested in the pedigree.

Exome sequencing and analysis
Clinical exome sequencing was performed as previously described (Yang et al., 2014).
Briefly, sequencing libraries were generated following capture using the VCRome version 2.1 capture system. Paired-end sequencing was performed using the Illumina HiSeq system with 100 basepair reads, resulting in mean >100X coverage of targeted bases. Sequence reads were analyzed using software designed and implemented in the clinical laboratory. Variants in SCUBE3 were confirmed by Sanger sequencing (Fig. S2, Table S2).
Per Institutional guidelines, informed consent was not obtained nor required for the case report included here.
Western blots from three replicate experiments were quantified using ImageJ. For each experiment, the amount of SCUBE3 and SCUBE1 protein used for immunoprecipitation (by volume) was calculated based on the INPUT sample, and the proportion of protein recovered following immunoprecipitation was calculated. The proportion of SCUBE3 protein recovered after immunoprecipitation was normalized to the proportion of SCUBE1 protein recovered after immunoprecipitation. These values were evaluated for statistically significant differences using ANOVA, and, to account for inter-experiment variation, the experimental replicate was included as covariate. ANOVA p-values for each SCUBE3 variant are shown. For plotting, each variant SCUBE3 was then normalized to the SCUBE3 WT control from the same experiment, and the mean and standard error of the mean across the three replicate experiments were plotted.

Smad activation assays
The coding sequence of human SCUBE3 (NM_152753.2) was cloned into pcDNA3. E1910), following the manufacturer instructions. Firefly luciferase activity was normalized by the Renilla luciferase activity (ratio F-Luc/R-Luc). Statistically significant differences were determined by one-way ANOVA and pairwise differences compared to NTC were determined by Dunnett's test.

Disease Models & Mechanisms • DMM • Accepted manuscript
For testing SCUBE3 secretion, 5mL of media was concentrated to 200µl by centrifugation in Amicon Ultra-4 filters (Sigma) then treated with Aurum serum protein mini kit (Bio-Rad). For each sample, 60µl was evaluated by Western blotting using an anti-Myc antibody (1:1,000 clone 9E10; Sigma, cat# OP10). Western blots from three replicate experiments were quantified using ImageJ. These values were evaluated for statistically significant differences using ANOVA, and, to account for inter-experiment variation, the experimental replicate was included as covariate. For plotting, each clone was normalized to the SCUBE3 WT control from the same experiment, and the mean and standard error of the mean across the three replicate experiments were plotted.