Identification of fibronectin 1 as a candidate genetic modifier in a Col4a1 mutant mouse model of Gould syndrome

Collagens are the most abundant proteins in the body, making up ∼30% of the dry weight. The collagen superfamily of extracellular matrix (ECM) molecules comprises 28 members encoded by 46 genes (Ricard-Blum, 2011). Among these, type IV collagens are primordial ECM molecules and are fundamental constituents of specialized structures called basement membranes (Fidler et al., 2017). In mammals, six genes encode the type IV collagens (Col4a1 to Col4a6) and their protein products assemble into three distinct heterotrimers [α1α1α2(IV), α3α4α5(IV) or α5α5α6(IV)]. The α1α1α2(IV) network is ubiquitous throughout development and in most adult tissues and its absence results in embryonic lethality (Pöschl et al., 2004). Pathogenic mammalian Col4a1 mutations were first identified using forward mutagenesis screens in mice with variable forms of ocular pathology (Favor et al., 2007; Gould et al., 2005; Thaung et al., 2002), and human mutations were identified in individuals with severe inherited or de novo porencephaly and early-onset intracerebral hemorrhages (ICHs) (Breedveld et al., 2006; Gould et al., 2005, 2006; Sibon et al., 2007; Vahedi et al., 2007). Studies in humans and mice have subsequently expanded the phenotypic spectrum, and it is now well established that mutations in COL4A1 and COL4A2 cause a multisystem syndrome (Jeanne and Gould, 2017; Labelle-Dumais et al., 2019; Mao et al., 2015; Meuwissen et al., 2015; Yoneda et al., 2013; Zagaglia et al., 2018).

This paper marks the first published use of the name Gould syndrome. Over the years, a number of abbreviations have been used to describe the various pathologies associated with COL4A1 and COL4A2 mutations, including RATOR, HANAC, PADMAL, BSVD1 and BSVD2 [Online Mendelian Inheritance in Man (OMIM) 1800009, 611773, 618564, 175780 and 614483, respectively]. However, the use of clinically based descriptions is impractical and ineffective for a syndrome for which even affected members within a family can have highly variable involvement of different organs. The use of multiple abbreviations fragments the literature and creates confusion for researchers, physicians and families. In efforts to establish a comprehensive and unified identity, over the past 2 years, family support groups and patient advocacy groups have coalesced on Facebook under the name Gould syndrome, and the name has been used in human interest stories on multiple local news broadcasts. Importantly, a non-profit foundation has been established and is currently enrolling patients in a global registry under the name Gould syndrome (https://www.gouldsyndromefoundation.org), which we have now adopted in response.

Gould syndrome is highly clinically heterogeneous and can have variable penetrance and severity across many organs. Cerebrovascular, ocular, renal and neuromuscular pathologies are among the most commonly described manifestations. We originally described a mouse model of Gould syndrome with a Col4a1 splice acceptor mutation that leads to exclusion of exon 41 (Δex41) from the mature transcript and 17 amino acids from the triple-helical domain of the protein (Gould et al., 2005). However, the majority of disease-causing variants reported in humans are missense mutations in highly conserved glycine residues of the triple-helical domain (Jeanne and Gould, 2017). In order to faithfully replicate human disease, we compared the effects of different mutations in an allelic series composed of nine distinct Col4a1 and Col4a2 mutant mouse strains (Favor et al., 2007; Jeanne et al., 2015; Kuo et al., 2014). We established that allelic heterogeneity has important implications for penetrance and severity of various pathologies including ICH and myopathy. We found that mutations can differ in degree or kind, and that allelic differences contribute to clinical variability, in part, by tissue-specific mechanistic heterogeneity (Labelle-Dumais et al., 2019). However, variable clinical manifestations including age of onset and severity among organs have been reported in people with the same recurrent or inherited mutations, and reduced penetrance and asymptomatic carriers exist in some families, indicating the involvement of factors in addition to allelic heterogeneity (Coupry et al., 2010; de Vries et al., 2009; Rødahl et al., 2013; Shah et al., 2012).

Studies using inbred strains of mice clearly demonstrate that a mutation can have variable outcomes in different genetic contexts, underscoring the important role that genetic modification plays in clinical heterogeneity. When compared to other mutations in the allelic series, Col4a1^Δex41 tends to have the most severe phenotypes across multiple organs when maintained on a pure C57BL/6J (B6) genetic background (Gould et al., 2007; Jeanne et al., 2015; Kuo et al., 2014). However, when the Col4a1^+/Δex41 B6 mice were crossed for a single generation to the CAST/EiJ (CAST) genetic background, ocular anterior segment dysgenesis (ASD), ICH and skeletal myopathy were all significantly reduced in the F₁ progeny (CASTB6F1) (Gould et al., 2007; Jeanne et al., 2015; Labelle-Dumais et al., 2011). Interestingly, when Col4a1^+/Δex41 B6 mice were crossed for a single generation to the 129SvEvTac (129) genetic background, ASD was ameliorated (albeit to a lesser extent than in CASTB6F1 mice) but ICH was not (Gould et al., 2007; Jeanne et al., 2015). Collectively, these data suggest that the B6 genetic background confers susceptibility to develop severe Gould syndrome-related pathologies and that CAST and 129 strains have one or more locus/loci that can genetically suppress pathology in a tissue-specific manner.

Identification of genetic modifier genes can help reveal disease mechanisms and potential therapeutic targets. Phenotype-driven mapping studies using mice and other model organisms are a powerful approach to identify genetic interactions that may be difficult to discover in humans even when large pedigrees are available (Ceco and McNally, 2013; Meyer and Anderson, 2017; Vieira et al., 2015). We previously performed a genetic modifier screen for pathology caused by the Col4a1^Δex41 mutation to provide insight into the underlying pathogenic mechanisms. Because (1) ASD is relatively severe, easily screened and robustly suppressed by two strains, and (2) the magnitude and breadth of phenotypic rescue was greater in CAST compared to 129 mice, we selected ASD and CAST as the phenotype and strain, respectively, to perform a pilot suppressor screen in a small number of mice (Gould et al., 2007). We generated CASTB6F1 mice and iteratively crossed mutant mice back to the B6 background. In each backcross, we applied selective pressure to retain one or more genetic suppresser locus by choosing the progeny with the mildest phenotype to breed for the next generation. Using a crude genome-wide scan, we identified a dominant modifier locus in CAST Chromosome (Chr) 1 that suppressed ASD. However, the approach had significant limitations, including screening on a single phenotype, the inability to identify potential recessive loci and the possibility that we may have incidentally overlooked other dominant loci by imposing bottlenecks at each generation.

Here, we performed independent genetic modifier screens for ASD and skeletal myopathy on a large-scale F₂ cross with the potential to identify multiple dominant or recessive loci that enhanced or suppressed pathology. Both screens identified a single suppressor locus on CAST Chr 1 that we called modifier of Gould syndrome 1 (MoGS1). Surprisingly, although the CAST background also suppressed ICH, the effect was not attributable to the MoGS1 locus. These data suggest that ASD and skeletal myopathy may be mechanistically related to each other but distinct from ICH, which further supports the notion of tissue-specific mechanistic heterogeneity contributing to the clinical variability of Gould syndrome (Labelle-Dumais et al., 2019). Furthermore, molecular analyses suggest that the ECM protein fibronectin 1 (FN1) is a strong candidate as the genetic suppressor at the MoGS1 locus and provide evidence for a role of altered integrin signaling in Gould syndrome pathogenesis.

To identify modifiers of Gould syndrome, we first performed a genome-wide screen with the early-onset and easily observed ASD phenotype. We phenotyped and genotyped 192 mutant F₂ progeny from (CAST X B6-Col4a1^+/Δex41) F₁ intercrosses using a mouse medium-density linkage panel with 646 informative single-nucleotide polymorphisms (SNPs) (Fig. 1A-C). We identified a region of interest on CAST Chr 1 with a logarithm of the odds (LOD) score of 11.2 and Bayesian confidence interval extending from 51.3 to 73.0 Mb (Fig. 1C). To determine whether smaller effects of other loci might have been masked, we performed a second genome-wide scan conditioned on this locus, but no additional loci reached statistical significance (Fig. 1D). This observation suggests that the major modifying effect of the CAST background on ASD is imparted through this single dominant locus on Chr 1.

In addition to suppressing ASD, the CAST genetic background significantly suppresses skeletal myopathy in Col4a1 mutant mice (Labelle-Dumais et al., 2011). To identify potential modifier loci for this phenotype, we performed an independent genome-wide mapping analysis using skeletal myopathy as the primary phenotype. We quantified myopathy severity as the percentage of muscle fibers containing non-peripheral nuclei (NPN) in histological sections of quadriceps (Fig. 1E). This assay is more quantitative with greater statistical power but also more labor intensive, and therefore we analyzed a randomly generated subset (n=49) of the mutant (CAST X B6-Col4a1^+/Δex41) F₂ progeny. The genome-wide scan revealed an interval on Chr 1 defined by the same markers as the ASD modifier locus (max. LOD score=5.61) (Fig. 1F).

To validate the biological effect of the MoGS1 locus, we independently generated an incipient congenic strain by iterative backcrossing the CAST-derived locus onto the susceptible B6 genetic background for five generations (N5). At N5, the genetic background of the mice is ∼97% pure B6 on average, except that they carry the CAST Chr 1 interval that includes the MoGS1 locus. Incipient congenic mice were intercrossed to generate Col4a1^+/Δex41 mice with zero (B/B), one (C/B) or two (C/C) copies of the CAST-derived chromosomal interval (Fig. 2). Of 24 eyes from Col4a1^+/Δex41 mice that were homozygous for the B6 allele (MoGS1^B/B) at the congenic interval, 29% (7) were moderate, and 71% (17) were severe (Fig. 2A). In contrast, of 28 eyes from the heterozygous group (MoGS1^C/B), 21.5% (6), 46.5% (13) and 32% (9) of the eyes were scored with mild, moderate and severe ASD, respectively. In mice homozygous for the CAST allele (MoGS1^C/C), 50% (10), 35% (7) and 15% (3) of the eyes were mild, moderate and severe, respectively. In a parallel experiment, we tested the effect of this congenic locus on myopathy and observed a dosage effect, whereby Col4a1^+/Δex41 mice that were heterozygous at the MoGS1 locus (MoGS1^C/B) showed a trend toward reduced myopathy, while homozygous MoGS1^C/C incipient congenic mice had significantly milder myopathy compared to mice that were homozygous for the B6 allele (MoGS1^B/B). Together, these results support the existence of one or more semi-dominant modifier genes at the MoGS1 locus on Chr 1.

Individuals with Gould syndrome have highly penetrant and clinically variable cerebrovascular diseases that include porencephaly and ICH (Bilguvar et al., 2009; de Vries et al., 2009; Giorgio et al., 2015; Vilain et al., 2002). ICH was previously reported to be significantly reduced in Col4a1 mutant mice on a CASTB6F1 background compared to those maintained on a B6 background, suggesting that the CAST genetic background also suppresses this phenotype (Jeanne et al., 2015). To test whether the MoGS1 locus might also genetically modify cerebrovascular diseases, we assessed porencephaly penetrance and ICH severity of the incipient congenic mice. We found that porencephaly penetrance and ICH severity were similar among Col4a1^+/Δex41 mice, irrespective of their genotypes at the MoGS1 locus (Fig. 2C,D).

To refine the interval of interest at the MoGS1 locus, we generated subcongenic lines of the CAST-derived chromosomal fragments on a B6 background (N5) and tested which line suppresses ASD and skeletal myopathy in Col4a1^+/Δex41 mice (Fig. 3A). Line 1, which included an interval of ∼10 Mb from the CAST genome, did not have a modifying effect on either phenotype (Fig. 3B,C). In contrast, a subcongenic line (Line 2) containing the distal portion of the MoGS1 locus (68.7-73.0 Mb) showed a significant protective effect for both ASD (Fig. 3D) and myopathy (Fig. 3E). Although ∼87% of eyes from Col4a1^+/Δex41;MoGS1^B/B mice had severe ASD (1 mild, 4 moderate and 33 severe), only ∼35% of the eyes from Col4a1^+/Δex41;MoGS1^C/C mice were severe (8 mild, 14 moderate and 12 severe). Likewise, Col4a1^+/Δex41;MoGS1^C/C mice had significantly fewer NPN compared to Col4a1^+/Δex41;MoGS1^B/B mice. Heterozygosity for this distal interval had intermediate effects for both phenotypes.

The refined MoGS1 locus is 4.3 Mb and contains 18 protein-coding genes and 38 predicted non-protein coding genes, including 13 long non-coding RNAs (lncRNAs), six small nuclear RNA (snRNA) genes, pseudogenes and others (Table S1). Using publicly available databases (Keane et al., 2011), we examined the refined MoGS1 interval for sequence differences between the B6 and CAST genomes. Among all variations, there were 40,237 SNPs, 6579 small insertions and deletions (indels), and 166 large structural variations including large insertions or deletions. Thirty-five SNPs were predicted to be missense variants, affecting ten protein-coding genes (Table 1). Among those genes, Ankar had three missense variants predicted to be deleterious by Sorting Intolerant From Tolerant (SIFT; Vaser et al., 2016) and phyloP (Pollard et al., 2010), and Spag16 had one predicted stop-gain variant and one missense variant that was predicted to be damaging by SIFT. However, neither gene is an obvious functional candidate. In addition, Ikzf2, Atic, Xrcc5 and Smarcal1 each have one SNP predicted to be damaging by either phyloP or PROVEAN (Choi and Chan, 2015) (Table S2). None of these SNPs in the MoGS1 locus was consistently predicted to be damaging by all three tools. Based on the Gene Expression Database (GXD) from Mouse Genome Informatics (informatics.jax.org/expression), 13 of the 18 protein coding genes have been shown to be expressed in eyes and in skeletal muscles. Two genes, Vwc2l and Bard1, are only expressed in eyes, and three genes, Ankar, Spag16 and Abca12, have no reported expression in either tissue. Because both ocular and muscular defects in Col4a1^+/Δex41 mice were suppressed by the MoGS1 locus, we hypothesized that the underlying pathogenic mechanism(s) may be shared between these two tissues. Therefore, despite the presence of possible pathogenic variants, we excluded Ankar, Spag16, Abca12, Vwc2I and Bard1 from further analysis.

Next, we tested whether any of the 13 genes with reported expression in ocular and muscular tissues are differentially expressed in B6 mice with or without the CAST interval at the MoGS1 locus. Four genes, Mreg, Fn1, Pecr1 and Igbp2, showed significantly increased expression in MoGS1^C/C eyes at postnatal day (P)0 compared to MoGS1^B/B eyes by quantitative PCR (qPCR) (Table 1 and Fig. 4A). FN1 is an ECM protein with multiple important roles in development and tissue homeostasis by interacting with cell surface receptors, ECM proteins and growth factors (Zollinger and Smith, 2017). Notably, FN1 has binding sites for type IV collagen (Laurie et al., 1986, 1982) and localizes adjacent to basement membranes in many tissues (Laurie et al., 1982). Moreover, like collagens, FN1 interacts with cells via integrin receptors (Johansson et al., 1997; Vandenberg et al., 1991), making it a strong functional candidate as a genetic modifier of COL4A1-related pathology.

In silico analysis comparing the CAST and B6 alleles of Fn1 revealed three missense SNPs, ten splice region SNPs, 13 SNPs in the UTR and 639 other non-coding SNPs, and 135 small indels and two structural variations within the locus (Table S3). Murine FN1 has 12 isoforms (White et al., 2008), making it difficult to predict the functional consequence(s) of a particular variant. The primary consequence of Col4a1 mutations is impaired secretion of mutant α1α1α2(IV) heterotrimers into the basement membranes (Kuo et al., 2014) and it is possible that increased FN1 confers partial compensation for extracellular α1α1α2(IV) deficiency. Consistent with this hypothesis, we observed increased Fn1 expression in developing eyes from MoGS1^C/C mice compared to MoGS1^B/B mice (Fig. 4A). Furthermore, we found that FN1 levels were higher in P10 quadriceps from Col4a1^+/Δex41 compared to Col4a1^+/+ mice and in MoGS1^C/C compared to MoGS1^B/B mice (Fig. 4B,C). FN1 plays important roles in cell adhesion, migration and signaling during tissue morphogenesis, which is primarily mediated through its interaction with integrins (Miyamoto et al., 1998; Sakai et al., 2003). Therefore, we tested the effect of MoGS1 on integrin signaling by evaluating the protein levels and phosphorylation status of two downstream effectors, integrin linked kinase (ILK) and focal adhesion kinase (FAK; also known as PTK2) (Hu and Luo, 2013; Humphries et al., 2019) (Fig. 4B,C). Western blot analysis of P10 quadriceps revealed that ILK levels were higher in Col4a1^+/Δex41 mice irrespective of whether they carried the MoGS1^B/B or MoGS1^C/C interval. Similarly, FAK phosphorylation (pFAK) and pFAK/FAK ratio were significantly higher in Col4a1^+/Δex41;MoGS1^B/B compared to Col4a1^+/+;MoGS1^B/B mice, and were reduced by the MoGS1^C/C interval. No difference was detected for the broadly used beta subunit of the integrin receptor, ITGB1, between genotypes. Taken together, these data suggest that elevated integrin signaling may contribute to Gould syndrome and that ASD and myopathy may be partially rescued by compensatory Fn1 expression.

Here, we performed a genetic modifier screen in Col4a1 mutant mice to gain insight into the pathogenic mechanisms that contribute to Gould syndrome. To this end, we used a large-scale F₂ cross of CAST×B6 and conducted independent screens for ASD and skeletal myopathy – two phenotypes that are commonly observed in individuals with Gould syndrome. Independent analyses for both phenotypes revealed a single, shared, semi-dominant locus on CAST Chr 1 that corroborates a previously identified locus (Gould et al., 2007). To validate these genetic data, we iteratively backcrossed the locus onto the B6 background and showed that a refined interval, referred to as MoGS1, suppressed ASD and myopathy in Col4a1 mutant mice. However, the MoGS1 locus had no effect on porencephaly penetrance and ICH severity – two phenotypes associated with Gould syndrome and previously characterized in Col4a1 mutant mice. The failure of MoGS1 to suppress ICH severity was unexpected because the CASTB6F1 genetic context significantly suppresses this phenotype. These data suggest that the effect of the MoGS1 locus is tissue specific and that other modifiers of ICH exist in the CAST background.

Using subcongenic strains, we refined the minimum critical interval for the suppressor locus to a 4.3 Mb region containing 18 protein-coding genes. Of the 18 positional candidate genes, 13 are expressed both in eyes and muscles, of which six contained missense variants between the CAST and B6 genomes; however, none were predicted to be deleterious by at least two of the three software prediction tools. Mutations in one of these six genes, Smarcal1, causes a rare multisystem disorder (Schimke Immunoosseous dysplasia; OMIM 242900) characterized by spondyloepiphyseal dysplasia, nephrotic syndrome and T-cell immunodeficiency, and a portion of patients develop corneal opacity, myopia, astigmatism and optic atrophy (Boerkoel et al., 2000). SWI/SNF-related matrix-associated, actin-dependent regulator of chromatin, subfamily a-like 1 (SMARCAL1) is a chromatin-remodeling protein involved in transcriptional regulation and DNA replication, repair and recombination (Bansbach et al., 2010). Compared to the B6 reference genome, CAST has two missense variations in Smarcal1. One was predicted to be tolerated by all three tools and the other was predicted to be damaging in one transcript isoform by PROVEAN, and no change of expression was detected in P0 eyes. Of the four positional candidate genes that were differentially expressed, Fn1 is a strong functional candidate. FN1 is a ubiquitously expressed extracellular glycoprotein that plays important roles in multiple processes by interacting with cell surface receptors, growth factors and other ECM proteins. In cell culture, FN1 promotes collagen IV deposition, assembly and incorporation into the ECM, and rescued some cellular phenotypes caused by a COL4A2 mutation (Filla et al., 2017; Ngandu Mpoyi et al., 2020). Mice deficient in Fn1 have aberrant lens placode formation and develop microphthalmia and cataracts (Hayes et al., 2012; Huang et al., 2011). Moreover, lack of FN1 in the skeletal muscle stem cell niche impairs muscle regeneration in aged mice (Lukjanenko et al., 2016). Collectively, these findings support our genetic and molecular evidence suggesting that FN1 is a strong candidate as the gene responsible for the effects conferred by the MoGS1 locus.

In general, the presence of mutant COL4A1 or COL4A2 leads to intracellular accumulation of mutant heterotrimers at the expense of their secretion. Intracellular heterotrimer accumulation represents a potential cell-autonomous (proximal) insult (Jeanne and Gould, 2017; Mao et al., 2015). Impaired heterotrimer secretion can also lead to a cell non-autonomous (distal) insult caused by extracellular heterotrimer deficiency, which can simultaneously perturb any number of presently unidentified cellular pathways. Intracellular accumulation and extracellular deficiency can be addressed simultaneously by targeting the proximal defect of protein misfolding. For example, pharmacologically promoting heterotrimer secretion using a chemical chaperone, 4-phenylbutyrate (4PBA), alleviates diverse pathologies in Col4a1 mutant mice (Hayashi et al., 2018; Jeanne et al., 2015; Jones et al., 2019; Labelle-Dumais et al., 2019). However, even in the absence of 4PBA, secretion of mutant heterotrimers is not completely abolished, and the presence of mutant heterotrimers in the basement membrane represents a third and distinct class of insult. We demonstrated this proof of concept by showing that mice with a Col4a1^G394V mutation have disproportionately severe myopathy, which is exacerbated when they are treated with 4PBA (Labelle-Dumais et al., 2019). The affected residue, glycine 394, is adjacent to a putative integrin-binding domain (Parkin et al., 2011), implicating impaired integrin binding in skeletal myopathy. Thus, for mutations that impair subdomains of the heterotrimer responsible for executing specific extracellular functions, promoting secretion of mutant heterotrimers would still be predicted to ameliorate many extracellular functions. However, the increased levels of mutant heterotrimers in basement membranes may also exacerbate specific disease pathway(s) related to the function(s) of the impacted subdomain (Labelle-Dumais et al., 2019). Identifying the multitude of extracellular roles of COL4A1/2 and how they are executed will be key in understanding fundamental aspects of matrix biology and the tissue-specific pathogenic mechanisms in Gould syndrome. Mapping functional subdomains on the α1α1α2(IV) heterotrimers will be important for genetically stratifying patients and may influence the prognosis and potential therapeutic approaches.

The unbiased nature of genetic screens leaves open the possibility for finding modifiers of either proximal or distal insults. By independently analyzing two phenotypes using an inbred strain (CAST) with broad phenotypic suppression, we sought to find modifiers that might represent therapeutic targets for the breadth of Gould syndrome phenotypes. MoGS1, the only significant locus that we identified, conferred tissue-specific effects by suppressing ASD and myopathy but not ICH. This observation is consistent with a modifier that has an extracellular function and suggests that the two phenotypes have at least partially overlapping distal pathogenic mechanisms. In a previous study using the Col4a1^Δex41 mutation, we found that, compared to B6, the CASTB6F1 background appeared to reduce intracellular accumulation but did not increase extracellular COL4A1 levels (Jeanne et al., 2015). This observation implicated intracellular accumulation in pathogenesis but did not rule out the potential of compensation by other extracellular factors. The failure of MoGS1 to suppress ICH severity suggests a mechanism independent of Fn1 expression levels, and intracellular accumulation of mutant proteins may indeed be the primary pathogenic insult for that phenotype. Further experiments are required to determine the molecular basis for the suppression of ICH severity.

When we compared disease severity across a murine allelic series of nine Col4a1 and Col4a2 mutations, we found that the Col4a1^G394V mutation had relatively little intracellular accumulation and that Col4a1^+/G394V mice had mild ICH but severe ASD and myopathy (Kuo et al., 2014). The proximity of the Col4a1^G394V mutation to a predicted integrin-binding domain suggests that impaired integrin binding contributes to ocular dysgenesis and myopathy. Therefore, one possibility is that FN1 ameliorates ASD and skeletal myopathy caused by Col4a1 mutations through compensatory integrin-mediated interactions. Consistent with this hypothesis, we detected evidence for increased integrin signaling in Col4a1^+/Δex41 mice that was reduced by the MoGS1 locus from CAST in the context of increased Fn1 expression. Moreover, ILK and FAK are both previously implicated in lens development (Cammas et al., 2012; Harburger and Calderwood, 2009; Hayes et al., 2012; Samuelsson et al., 2007; Teo et al., 2014) and myopathy (Boppart and Mahmassani, 2019; Gheyara et al., 2007). Although we detected increased Fn1 expression, this does not preclude the potential contributions of alternative splicing. Understanding the precise mechanism of action and functionally validating this pathway with experimental manipulation and preclinical interventions would represent a significant advance in the understanding of the pathogenic mechanisms that contribute to Gould syndrome.

All experiments were compliant with the ARVO Statement for the Use of Animals in Ophthalmic and Vision Research and approved by the Institutional Animal Care and Use Committee at the University of California San Francisco (UCSF). Col4a1^+/Δex41 mice were originally identified in a mutagenesis screen conducted at The Jackson Laboratory (Bar Harbor, ME, USA) (Gould et al., 2007, 2005). Since then, Col4a1^+/Δex41 mice have been backcrossed to B6 for at least 20 generations. CAST mice were obtained from The Jackson Laboratory. All animals were maintained in full-barrier facilities free of specific pathogens on a 12-h light/dark cycle with food and water ad libitum. Both male and female mice were used in this study.

Ocular anterior segment examinations were performed on mice at 1.0-1.5 months of age using a slit-lamp biomicroscope (Topcon SL-D7; Topcon Medical Systems, Oakland, NJ, USA) attached to a digital SLR camera (Nikon D200; Nikon, Melville, NY, USA).

At 2 months of age, mice were subjected to treadmill exercise and their quadriceps were harvested 2 days later. Exercise was 30 min with a 15° downhill grade on a treadmill equipped with a shock plate (Columbus Instruments, Columbus, OH, USA). Animals were started at 6 m/min and increased by 3 m/min every 2 min until a maximum of 15 m/min speed was reached. Quadriceps were dissected and frozen in liquid nitrogen-cooled isopentane. Cryosections (10 μm) were collected at regular intervals and stained with Hematoxylin and Eosin (H&E) for histopathology. Then, the number of NPN were evaluated on a total of 12 sections for each muscle, and the percentage of muscle fibers with NPN was quantified. The observers were masked to genotypes and counted between 2000 and 5000 muscle fibers per animal.

ICH was assessed at 2 months of age by Perl's Prussian Blue staining as previously described (Jeanne et al., 2015). Briefly, mice underwent transcardial perfusion with saline followed by 4% paraformaldehyde (PFA) and then were fixed in 4% PFA overnight. Brains were dissected, cryoprotected in 30% sucrose and embedded in optimal cutting temperature compound (Sakura Finetek, Torrance, CA, USA). Coronal cryosections (35 μm) regularly spaced along the rostro-caudal axis were stained with Prussian Blue/Fast Red and imaged. On each section, the percentage brain area with Prussian Blue staining was calculated using ImageJ software (National Institutes of Health). ICH severity was expressed as the average percentage of hemosiderin surface area on 28 sections for each brain. The presence or absence of porencephaly on sections used in ICH analysis was also recorded.

Genomic DNA were obtained from 192 (CAST X B6) F₂ mice carrying the Col4a1^+/Δex41 mutation and genotyped at the UCSF Genomic Core Facility with a commercial SNP panel (Illumina, San Diego, CA, USA), which contains 646 informative SNPs for (CAST X B6) F₂ progeny. A genome-wide suppressor screen was performed in R (R Core Team, 2019) using the package R/qtl (Broman et al., 2003), treating the trait as binary. Genome-wide significance was established using 1000 permutation testing (Churchill and Doerge, 1994). A confidence interval for each quantitative trait locus (QTL) location was then calculated using Bayesian confidence sets (Sen and Churchill, 2001). The ASD score for each animal was averaged from both eyes and treated as a dichotomous variable; a score of 1 was mild, and a score of 1.5-3 was severe. A subset (49) of these 192 mice was randomly selected in a separate mapping analysis for the muscle modifier as a quantitative trait.

Variants between B6 and CAST were exported from the Sanger Mouse Genome Project website (Keane et al., 2011; Yalcin et al., 2011) and annotated using the Ensembl Genome Browser (Hunt et al., 2018; Yates et al., 2020). SNPs with high and moderate levels of impact were prioritized. Non-synonymous SNPs were further analyzed using phyloP (Pollard et al., 2010), PROVEAN (Choi and Chan, 2015), in addition to SIFT (Ng and Henikoff, 2003), a built-in feature in Ensembl. PhyloP scores more than 1, PROVEAN scores less than −2.5 and SIFT scores less than 0.05 were considered to be likely damaging.

Eyes from P0 mice were dissected in PBS, immediately transferred in RNAlater (ThermoFisher Scientific, USA) and stored at −80°C until use. Total RNA was extracted using TRIzol Reagent (ThermoFisher Scientific) according to the manufacturer's instructions and reverse transcribed to complementary DNA (cDNA) using an iScript cDNA synthesis kit (Bio-Rad, Hercules, CA, USA). qPCR was performed on a Bio-Rad CFX96 real-time system using SsoFast Evagreen mix (Bio-Rad) and primers are listed in Table S4. Briefly, 10 ng cDNA and 1.25 μM primers were used per reaction in a final volume of 10 μl. Each cycle consisted of denaturation at 95°C for 5 s, followed by annealing and extension at 60°C for 5 s, for a total of 45 cycles. All experiments were run with technical duplicates and four to six biological replicates were used per group. The relative expression of each gene was normalized to Hprt1 or Gapdh and analyzed using the 2^−ΔΔCT method (Livak and Schmittgen, 2001).

Quadriceps from P10 pups were dissected and total proteins extracted using radioimmunoprecipitation assay (RIPA) buffer (VWR, Radnor, PA, USA) supplemented with Halt Protease and Phosphatase Inhibitor Cocktail (ThermoFisher Scientific), EDTA and 2 mM phenylmethylsulfonyl fluoride. Total proteins (10 μg) were separated on 4-15% gradient SDS-PAGE gels (Bio-Rad) and transferred to polyvinylidene fluoride (PVDF) membranes (Bio-Rad). Membranes were blocked for 1 h at room temperature in 5% non-fat milk in TBS containing 0.1% Tween-20 (TBST), incubated overnight at 4°C in primary antibodies diluted in 2% non-fat milk in TBST. Primary antibodies and dilutions used are as follows: rabbit anti-FN1 (1:10,000; Abcam, ab2413), rabbit anti-pFAK (1:1000; Invitrogen, 44624G), rabbit anti-FAK (1:1000; Santa Cruz Biotechnology, sc-557), rabbit anti-ILK (1:1000; Cell Signaling Technology, 3862), mouse anti-ITGB1 (1:1000; BD Biosciences, 610467) and mouse anti-GAPDH (1:500,000; Millipore, MAB374). Following washes in TBST, membranes were incubated for 1 h at room temperature with species-specific horseradish peroxidase-conjugated secondary antibodies (1:10,000; Jackson ImmunoResearch, West Grove, PA, USA) diluted in 2% non-fat milk in TBST. Immunoreactivity was visualized using chemiluminescence (ECL or Luminata Forte substrate, ThermoFisher Scientific) and imaged using a ChemiDoc MP Imaging System (Bio-Rad) or exposed to X-ray films. Densitometric analyses were performed on low-exposure images using the Quantity One software (Bio-Rad). To compare samples loaded on different gels, all samples were normalized to the same calibrator sample (a wild-type B6 sample) that was loaded on all gels.

Statistical analyses and graphs were prepared using Prism v8.0 software (GraphPad, La Jolla, CA, USA). Multiple comparisons were carried out using one-way ANOVA followed by Tukey post-test or Kruskal–Wallis test followed by Dunn's post-test. P-values less than 0.05 were considered statistically significant.

We thank members of the Gould Laboratory for their critical reviews of the manuscript. We also thank Cassandre Labelle-Dumais for helpful discussion and guidance with the muscle studies.