Gpr63 is a modifier of microcephaly in Ttc21b mouse mutants

The primary cilium is a signaling center critical for proper embryonic development. Previous studies have demonstrated that mice lacking Ttc21b have impaired retrograde trafficking within the cilium and multiple organogenesis phenotypes, including microcephaly. Interestingly, the severity of the microcephaly in Ttc21baln/aln homozygous null mutants is considerably affected by the genetic background and mutants on an FVB/NJ (FVB) background develop a forebrain significantly smaller than mutants on a C57BL/6J (B6) background. We performed a Quantitative Trait Locus (QTL) analysis to identify potential genetic modifiers and identified two regions linked to differential forebrain size: modifier of alien QTL1 (Moaq1) on chromosome 4 at 27.8 Mb and Moaq2 on chromosome 6 at 93.6 Mb. These QTLs were validated by constructing congenic strains. Further analysis of Moaq1 identified an orphan G-protein coupled receptor (GPCR), Gpr63, as a candidate gene. We identified a SNP that is polymorphic between the FVB and B6 strains in Gpr63 and creates a missense mutation predicted to be deleterious in the FVB protein. We used CRISPR-Cas9 genome editing to create two lines of FVB congenic mice: one with the B6 sequence of Gpr63 and the other with a deletion allele leading to a truncation of the GPR63 C-terminal tail. We then demonstrated that Gpr63 can localize to the cilium in vitro. These alleles affect ciliary localization of GPR63 in vitro and genetically interact with Ttc21baln/aln as Gpr63;Ttc21b double mutants show unique phenotypes including spina bifida aperta and earlier embryonic lethality. This validated Gpr63 as a modifier of multiple Ttc21b neural phenotypes and strongly supports Gpr63 as a causal gene (i.e., a quantitative trait gene, QTG) within the Moaq1 QTL.


Introduction
Primary cilia are microtubule-based organelles known to play essential roles in proper development and function of a number of organ systems including the central nervous system (CNS) [1][2][3][4]. Ciliopathies are a class of human diseases caused by pathogenic variants in genes encoding proteins responsible for the proper form and function of cilia [5][6][7]. The frequent presentation of cognitive impairment in ciliopathy patients, in addition to severely compromised brain development in a number of ciliary mutant mouse models, clearly displays the importance of primary cilia in CNS health and development [8,9]. Primary cilia have been implicated in transducing and regulating several critical developmental pathways including SHH, WNT, PDGF, TGFβ/BMP, RTK, and Notch [3,10]. An array of cell surface receptors, including G-protein coupled receptors (GPCRs), localize to the ciliary membrane to modulate these pathways and other signaling events [11]. Intraflagellar transport (IFT) proteins are responsible for the movement of cargo within the cilium with complex-B proteins regulating anterograde transport from the basal body to the distal tip of the cilium and complex-A proteins regulating retrograde transport [12]. IFT-A proteins have also been shown to specifically play a role in the trafficking of select GPCRs in and out of the cilium [13][14][15].
Tetratricopeptide repeat domain 21b (Ttc21b; Thm1; Ift139; MGI 1920918) encodes an IFT-A protein necessary for the proper rate of retrograde trafficking within the primary cilium [16]. TTC21B is not part of the "core IFT-A complex" [15]. When TTC21B is reduced, a large portion of the remaining IFT-A protein subunits remains intact as this "core" complex and continues to interact with the dynein motor powering retrograde movement. Thus, TTC21B likely acts to facilitate binding of peripheral cargo proteins to the core proteins of the complex, thus assisting the transport of these cargo proteins through the cilium. Proper ciliary trafficking is known to be critical for the processing of GLI transcription factors, which are in turn essential for proper regulation of the SHH pathway [17,18]. Mutant mice homozygous for the alien (aln) null allele of Ttc21b display impaired processing of GLI3, the primary repressor of SHH target genes, and consequently show an increase in SHH pathway activity in multiple tissues including the developing forebrain [16,19].
The ciliopathies are only one example of a class of human diseases with broad phenotypic variability. The advent of genome sequencing is facilitating the study of genes and variants affecting disease penetrance and expressivity. The ability to specifically model such modifiers in experimental organisms like mice with genome editing should allow more in-depth studies of genetic interactions and their role in human disease. It is well known that genetic backgrounds of mouse inbred strains can affect phenotypic severity and penetrance of specific mouse mutations. These differences can be used to detect genetic interactions in complex traits and begin to understand the mechanisms leading to this variability [20,21]. One example of genetic background influencing the expressivity of a phenotype is the forebrain malformations seen in Ttc21b aln/aln mutants. Here we present a QTL analysis to further understand the underlying pathogenic molecular mechanism leading to the microcephaly. We demonstrate that Gpr63 interacts with Ttc21b to modify the size of the forebrain in Ttc21b aln/aln null mutants and provide evidence that the Gpr63 mutation is the likely causal variant for the chromosome 4 QTL affecting brain size.

Ttc21b aln microcephaly is background dependent
The phenotypes observed in Ttc21b aln/aln mutants were first identified as part of an ENU mutagenesis experiment to identify recessive developmental phenotypes in the perinatal mouse [22]. A/J mice were treated with the ENU mutagen and the subsequent breeding and mapping of the causative mutation involved an outcross to the FVB/NJ (FVB) strain [16,22]. Ttc21b aln/ aln homozygous mutants on this mixed A/J;FVB background exhibited multiple forebrain and craniofacial phenotypes (Fig 1A-1D) [16,19]. While the microcephaly phenotype was completely penetrant, craniofacial phenotypes such as cleft lip and palate were only present in a minority of embryos. To increase the frequency of craniofacial phenotypes and facilitate future molecular analyses, the Ttc21b aln allele was serially backcrossed onto the C57BL/6J (B6) strain. The B6 genetic background has previously been shown to be more susceptible to craniofacial phenotypes in a number of studies [23][24][25][26]. This initial, exploratory backcross was done informally and was not genotyped. After the Ttc21b aln allele was serially backcrossed to the B6 background for at least five generations, we noticed the relative size of the forebrain tissues in B6.Cg-Ttc21b aln/aln mutants increased substantially as compared to the mutants previously analyzed, although they remained smaller in all Ttc21b aln/aln mutants than control brains and still lacked olfactory bulbs (Fig 1E and 1F). To confirm this was indeed an effect of varying the genetic background, the Ttc21b aln allele was again serially backcrossed to the FVB background for at least three generations. Embryonic analysis showed the severity of microcephaly initially seen was recovered in the new FVB.Cg-Ttc21b aln/aln mutants (Fig 1C and 1D). We measured the forebrain surface area of Ttc21b aln/aln mutants on both FVB and B6 genetic backgrounds ( Fig 1G) and saw a 60% decrease in forebrain size in FVB.Cg-Ttc21b aln/aln animals as compared to littermate controls (n = 7 wt and n = 4 aln mutants; p<0.0001), but only a 30% decrease in B6.Cg-Ttc21b aln/aln animals (n = 4 wt and n = 3 aln mutants; p = 0.00033). The interaction between genetic background and the Ttc21b aln allele was analyzed with a 2-way ANOVA (p = 0.0053).
Given all of this preliminary data, we hypothesized the genetic background effect on Ttc21b aln/aln mutant forebrain size would allow a Quantitative Trait Locus (QTL) analysis to generate insight into the underlying molecular mechanism(s). In order to generate mice as genetically uniform as practical prior to performing a QTL analysis, we further backcrossed both strains of Ttc21b aln mice. After at least three further generations of backcrossing onto FVB (FVB;B6-Ttc21b aln/aln ) and four generations onto B6 (B6;FVB-Ttc21b aln/aln ), we performed a genome-wide high-density MegaMUGA SNP scan and confirmed that each strain was approximately 90% pure (S1 Table). We further purified the FVB and B6 advanced backcross lines with at least two additional backcrosses combined with targeted microsatellite marker screening of chromosomal areas not yet homozygous for FVB or B6 as identified in the initial genome scan. This allowed us to create FVB.Cg-Ttc21b aln/wt animals that were >99% FVB and B6.Cg-Ttc21b aln/wt animals purified to~97% B6 background, as assayed by a GigaMUGA genome SNP scan providing strain-specific sequence information at over 143,000 SNPs [27] (S1 Table).
As we performed this breeding, we continued to note an incompletely penetrant exencephaly phenotype in Ttc21b aln/aln animals, but the frequency did not appear to significantly differ between backgrounds (Table 1). We calculated the incidence of exencephaly and found that maintaining the Ttc21b aln allele on either the FVB or B6 genetic backgrounds yielded exencephaly in~40% of Ttc21b aln/aln mutants (24/58, 41% in FVB;B6-Ttc21b aln/aln and 30/65, 46% in B6;FVB-Ttc21b aln/aln ). Note these embryos were collected for other experimental  �� exencephaly or earlier lethality purposes as the incipient congenics were being further purified and are on a mixed genetic background. We first generated F2 Ttc21b aln/aln mutant progeny from a B6.FVB F1 (Ttc21b aln/wt ) intercross for the QTL analysis. Fifty-four F2 litters were generated, which identified 148 Ttc21b aln/ aln mutant embryos from 607 total embryos. We excluded 35 mutants (24%) with exencephaly, which precluded a measurement of forebrain area. Ninety-six brains from the remaining F2 mutants were micro-dissected and the dorsal surface area of the forebrain was measured. Whole genome SNP genotyping of each F2 recombinant was carried out using a GigaMUGA panel and QTL mapping was performed using R/qtl [28] to identify chromosomal regions linked to differential brain sizes in Ttc21b aln/aln mutants. The dataset used in the QTL analysis is presented in S2 Table and an overview of the QTLs identified and their effects on the phenotype is provided in Table 2.
A significant QTL (genomewide p < 0.05; LOD 6.1) was mapped to chromosome 4 with its peak at 27.8 Mb (marker UNC6953268) and a 95% confidence interval spanning 19.5 Mb to 46 Mb (Fig 2A and 2B). This QTL explained 25.1% of the phenotypic variance ( Table 2). We have named this chromosome 4 locus modifier of alien QTL 1 (Moaq1). A second QTL was identified with a LOD score suggesting significance (genomewide p < 0.63; LOD 4.7) and mapped to chromosome 6 at 93.6 Mb (UNCHS018175) with a 95% confidence interval from 0 to 117.5 Mb (Fig 2A and 2C). This chromosome 6 QTL explained 20.1% of the trait variance and was named modifier of alien QTL 2 (Moaq2). A third peak was found on chromosome 2 at the Ttc21b locus, which likely represents the A/J passenger background on which the initial mutagenesis was performed (Fig 2A). To assess the contribution of each locus to the trait, we determined the change in LOD score and percent variance explained for the three possible 2-loci models and the 3-loci model ( Table 2). A significant increase in LOD score (>+3) was found for all comparisons, indicating that all three loci (aln and the 2 modifiers) significantly added to the phenotype, and together they explained 44.4% of the phenotypic variance. To assess whether the three loci interacted with each other and/or were additive, we performed a scantwo analysis in R/qtl (Table 2), using 1,000 permutations to establish the threshold levels for significance. Results indicated that none of the loci interacted (lod.int for all 3 pairings, not significant). However, the main effects of both Moaq1 and Moaq2 were independently additive with aln. The additive effect of Moaq1 and aln was highly significant (lod.add = 9.13) and that We confirmed a correlation between the genetic identity of Moaq1 and Moaq2 and an effect on forebrain size by comparing the genotype of the peak SNP marker for each QTL with forebrain size (Fig 2D and 2F). We first independently assessed F2 mice for genotype at the peak marker of the Moaq1 locus and found an incremental decrease in brain size among Ttc21b aln/aln -Moaq1 FVB/FVB animals as compared to Ttc21b aln/aln -Moaq1 FVB/B6 or Ttc21b aln/aln -Moaq1 B6/B6 animals ( Fig 2D). An overall ANOVA was performed to identify significant differences among these groups (p<0.0001) followed by a subsequent Tukey's multiple comparison test analysis for comparison of each group (Ttc21b aln/aln -Moaq2 FVB/FVB , Ttc21b aln/aln -Moaq2 FVB/B6 and Ttc21b aln/aln -Moaq2 B6/B6 ) with individual p-values ranging from 0.0145 to <0.0001 ( Fig 2D). To control for a potential litter-based difference, we also compared only Ttc21b aln/aln -Moaq1 FVB/FVB mutants from litters with Ttc21b aln/aln -Moaq1 FVB/B6 or Ttc21b aln/aln -Moaq1 B6/B6 littermates ( Fig 2E). Although this analysis necessarily used a smaller sample size than the non-litter matched comparison (Fig 2D), we again observed a 1.94mm 2 reduction in Ttc21b aln/aln -Moaq1 FVB/FVB animals (p = 0.012).
A similar analysis testing just the Moaq2 locus alone showed a significant variation among

Congenic lines
We produced independent congenic lines to replicate the Moaq1 and Moaq2 loci. We noted in our original F2 population of 96 mice for the QTL analysis that no Ttc21b aln/aln mutants were homozygous for B6 SNPs over a 10-Mb region on chromosome 14. We hypothesized this represented a region of the B6 genome that, in combination with homozygosity for the Ttc21b aln allele, may lead to either early embryonic lethality and/or exencephaly. Thus, we attempted to avoid this complication and create FVB congenic lines with Moaq1 and Moaq2 genomic regions derived from the B6 strain with the hypothesis that the B6 genomic identity of Moaq1 and Moaq2 would increase the size of the Ttc21b aln/aln over the size of the congenic FVB Ttc21b aln/aln mutants. B6;FVB-Ttc21b aln/wt F1 mice, (obligate heterozygotes across their genome), were repetitively backcrossed with FVB mice to produce a congenic strain that approached inbred FVB except for the heterozygous Moaq1 or Moaq2 B6 QTLs (i.e., FVB.Cg-Moaq1 FVB/B6 and FVB.Cg-Moaq2 FVB/B6 ). Mice from each congenic line were then intercrossed to determine if the B6-derived Moaq1 and/or Moaq2 QTLs could influence forebrain size on their own. During the process of creating these congenics, we noted that successive generations of mating onto the FVB background was leading to lethality of Ttc21b aln/aln mutants prior to E17.5 or exencephaly. We observed Ttc21b aln/aln mutants on a >99% FVB background and 21/22 either died prior to E17.5 or had exencephaly precluding measurement of brain size ( Table 1). Note that while we had previously created mice that were largely enriched for FVB, even as much as 90%, (e.g., Fig 1), the incipient congenics created here for homozygote analysis are the purest FVB-Ttc21b aln/wt population created to date. (While creating the mice used to perform the initial QTL analysis, we had not been intercrossing and dissecting embryos to create Ttc21b aln/aln homozygotes so had not previously identified this effect on the phenotype.) This suggests a possibility for future work identifying a separate exencephaly modifier from this population. Given that the increasing incidence of exencephaly precluded an analysis of forebrain size in the congenic lines (N5 and greater), we analyzed animals from the earlier backcross generations. Littermate brains were analyzed using paired t-tests to account for potentially confounding variables, such as length of gestation and/or background differences based on the generation of backcross. The Moaq1 QTL was found to affect brain size in FVB;B6-Ttc21b aln/aln mutants in the N2 to N5 generations. Ttc21b aln/aln -Moaq1 FVB/B6 and Ttc21b aln/aln -Moaq1 B6/B6 animals had larger brains as compared to Ttc21b aln/aln -Moaq1 FVB/FVB (Fig 3A, p = 0.0177), with an average decrease of 2.24mm 2 (Fig 3B). The Moaq2 QTL was also validated using this strategy and Ttc21b aln/aln -Moaq1 FVB/B6 and Ttc21b aln/aln -Moaq1 B6/B6 animals again had larger brains than Ttc21b aln/aln -Moaq1 FVB/FVB (Fig 3C, p = 0.0201), with an average decrease of 3.40mm 2 ( Fig 3D). Thus, we conclude the QTL analysis of just 96 B6;FVB-Ttc21b aln/aln F2 forebrains revealed two validated, novel QTLs that modify the effect of the Ttc21b aln/aln mutation on forebrain size.

Candidate genes for Moaq1 QTL
In order to identify potential candidate causal genes within the~26.5 Mb Moaq1 QTL, the 253 genes within the 95% confidence interval were analyzed (Fig 4; S3A Table). We initially focused our analysis on strain-specific missense coding variants. We recognize that this initial analysis of missense variants, rather than a parallel approach first examining the strain-specific insertions and/or deletions, introduces some ascertainment bias towards different candidate genes. We also could include RNA expression analysis of mutants from both background and look for non-coding regulatory variants within the QTL. However, we initially reasoned that missense coding variants might be best suited to act as modifiers than more severe protein perturbations resulting from insertion/deletion events. We first identified 33 genes with missense polymorphisms when comparing B6 and FVB genomic sequences (S3B Table). The list was further refined to 21 (S3C Table) by selecting only those genes in which the A/J and FVB SNP were identical, given the phenotypic similarities in Ttc21b aln/aln mutants on those genetic backgrounds. The remaining missense mutations were analyzed with the SIFT prediction algorithm to assess pathogenicity [29], as well as information on known phenotypes in mouse models or human disease (OMIM), expression data and other information from the literature (S3C Table). Four genes were found with predicted deleterious missense mutations (<0.05 SIFT value). We then performed a literature review of these four remaining genes (Table 3,  S3D Table) to identify those previously known to have a role in developmental neurobiology and narrowed the list to 3 candidates: gamma-glutamyl hydrolase (Ggh, MGI 1329035), G-protein coupled receptor 63 (Gpr63, MGI 2135884), and origin recognition complex subunit 3 (Orc3, MGI 1354944). Ggh is a lysosomal enzyme with a role in folate metabolism, which is provocative given the exencephaly noted in Ttc21b aln/aln mutants [30]. Furthermore, Ggh is the only gene in the interval with 2 potentially deleterious missense mutations. Orc3 is a nuclear localized component of the Origin Recognition Complex [31,32]. While conditional ablation of Orc3 in the brain revealed a requirement in proper neural progenitor and radial glial development, germline loss of Orc3 led to early embryonic lethality [33]. We next considered any evidence for these remaining candidate genes within the Moaq1 QTL to have a role in primary cilia biology. Unlike Ggh or Orc3, Gpr63 is a membrane protein, making it the more attractive candidate to be a modifier of a mutation in a plasma membrane-based organelle, the primary cilium. An siRNA screen to identify novel regulators of ciliogenesis in vitro did provide some evidence Gpr63 may have a role in cilia biology [34]. Gpr63 has been reported to have brainspecific expression and to be enriched in the forebrain [35,36]. The predicted deleterious polymorphic SNP within Gpr63 (rs13477613; NM_030733 c.G1635T; NP_109658 p.R407S; SIFT  score of 0.01) is a missense variant altering the coding of amino acid 407 from Arginine, an amino acid with an electrically charged side chain, in the B6 reference genome to Serine, an amino acid with a polar uncharged side chain, in the A/J and FVB genomes (Table 3). This Arginine residue is conserved from frogs to chimpanzees (Fig 5A), although humans have a Lysine at the orthologous residue in GPR63, an amino acid with an electrically charged side chain like Arginine. The loss of Arginine from the mouse GPR63 cytoplasmic tail alters a predicted RxR endoplasmic reticulum (ER) retention signal (Fig 5) [37]. The C-terminus of GPCRs is understood to play critical roles in protein-protein interactions and regulation of the receptor [38]. As shown in Fig 6A, the c.G1635T SNP changes the amino acid sequence of the coat protein complex I (COPI) binding sequence. COPI is a coatomer responsible for the retrograde transport of vesicles from the trans-Golgi network to the cis-Golgi and from the cis-Golgi to the ER [39,40]. COPI has been implicated in the retrieval of proteins with exposed RxR motifs [41,42]. These motifs function to retain proteins in the ER until the RxR containing protein is completely processed and ready to be transported out through the Golgi [37]. Often these proteins escape the RxR retrieval mechanism by masking the motif through interaction with another protein, a critical step in preparing the protein for its function which may include assembling a hetero-multimeric complex [37]. There are candidate motifs for protein domain interactions on the GPR63 C-tail which could serve in this masking role including: LPRLPGH, a predicted SH3 binding motif of DLG4, and EHRTV, a predicted PDZ binding domain of RGS3 [43][44][45][46]. We also note a BBSome binding motif which may serve as a potential ciliary targeting sequence [47].

CRISPR-Cas9 genome editing of Gpr63
A standard approach to reduce the number of candidate genes within a QTL is to narrow the genetic interval carrying the trait of interest using meiotic recombination. For this, QTL-interval specific recombinants are identified and overlapping sub-congenics produced and tested to determine the minimal region of effect. This process can be repeated as needed to critically narrow the candidate interval. We initially pursued this strategy, but the results above show that the combination of exencephaly and early lethality of congenic FVB.Cg-Ttc21b aln/aln mice will preclude successful implementation unless we could also identify the region contributing to the exencephaly. An alternative approach is to directly test hypotheses about candidate sequence variant(s) with genome editing to create novel transgenic models. We chose to pursue this and targeted FVB mice to change the rs13477613 polymorphism from FVB sequence (c.1635-T; coding for Gpr63 Ser(FVB) ) to the reference, non-pathogenic, B6 sequence (c.1635-G; coding for Gpr63 Arg(B6) ). This allele Gpr63 em1Rstot , is hereafter referred to as FVB-Gpr63 Arg(B6) . An added benefit of CRISPR-Cas9 genome editing is the potential to recover an allelic series from the initial set of transgenic founders. We recovered and maintained a second allele Gpr63 em2Rstot (Gpr63 Del ), with an 8-bp deletion resulting in a frameshift mutation that leads to the translation of 5 missense amino acids and an early stop, thus preventing translation of the last 21 amino acids of the GPR63 C-terminus (Fig 6A). This Gpr63 Del mutation may disrupt both an ER retention signal and an RGS3 PDZ binding motif, creating a potentially more deleterious mutation than Gpr63 Ser(FVB) [37,[43][44][45][46].
We first determined that the Gpr63 Arg/Ser polymorphism did not affect survival or brain size at E17.5 (Fig 7A, ANOVA p = 0.1743). We also did not see an effect at postnatal day (P) 42 on either forebrain dorsal surface, brain weight or total body weight (Fig 7B-7D, ANOVA p = 0.9639, 0.7753, 0.9420, respectively). This was expected, given that these are sequence polymorphisms found between two commonly maintained inbred strains. We hypothesized the  Gpr63 Del allele may be more deleterious than the Gpr63 Ser(FVB) polymorphism alone. We first determined if this allele affects survival and created animals homozygous for the Gpr63 Del allele. We recovered Gpr63 Del/Del mutants in approximately Mendelian ratios at weaning (8/ 35) and they survived to adulthood (n = 8) with no discernible phenotypes seen to date. We measured forebrain dorsal surface, brain weight and body weight at both E17.5 and P42 and found no change in the Gpr63 Del/Del mutants (Fig 7E-7L; ANOVA p = 0.2569, 0.6846, 0.6344, 0.6655, respectively). Consistent with the predicted deleterious effects of the SNP at amino acid 407 seen in both viable and fertile FVB and A/J animals, this suggests that Gpr63 does not have an essential requirement in development. Gpr63 is a modifier of Ttc21b microcephaly

Gpr63 expression in vitro shows ciliary localization
We directly tested the hypothesis that Gpr63 might, at least partially, localize to the primary cilium by measuring in vitro expression of a myc-flag-epitope tagged construct in NIH 3T3 cells (pCMV6-GPR63-myc). Indeed, we found colocalization of the GPR63 signal with Arl13b as a marker of the ciliary axoneme (Fig 8A-8C). We were not able to identify a completely reliable antibody to verify these findings, but saw similar results from an independent GPR63 expression construct (GPR63-TANGO-FLAG [48], S2 Fig). Site-directed mutagenesis was used to generate constructs recapitulating the variants seen in vivo (Fig 8D-8I). Overexpression of either polymorphism of Gpr63 or the Gpr63-deletion construct did not affect the proportion of ciliated cells (Fig 8M, ANOVA p = 0.067). However, when we expressed either the myc-Gpr63 Ser or the myc-Gpr63 Del construct we saw a reduction in the number of transfected cells with GPR63 in the primary cilia (Fig 8N, ANOVA p = 0.004). We saw a 65% decrease with Gpr63 Ser407 as compared with Gpr63 Arg407 (p = 0.011) and a 68% decrease with the Gpr63 Del construct (p = 0.008). We conclude that the different alleles changing the protein composition of the GPR63 cytoplasmic tail can alter the subcellular (i.e., ciliary) localization of GPR63.
To further address a possible interaction between Ttc21b and Gpr63, we aimed to create Ttc21b aln/aln ;Gpr63 Del/Del double homozygous mutant animals. Note this experiment was done in a way that necessitated crosses with differing mixtures of the FVB and B6 background. Therefore, the Gpr63;Ttc21b double mutant analysis is best considered on a hybrid background with Gpr63 alleles being the only relevant variable. We recovered 8 Ttc21b aln/aln ; Gpr63 Del/Del embryos at E17.5 and E18.5 from this cross and found that 75% (6/8) were embryonic lethal and had died around E12.5. The remaining two had exencephaly precluding measurement of forebrain area.
We further explored this phenotype at~E12.5 (E10.5-14.5) and recovered 9 Ttc21b aln/aln -Gpr63 Del/Del double homozygous null embryos (Fig 10F). We saw many of the phenotypes previously noted in our studies of Ttc21b aln/aln mutants. These included 89% (8/9) with exencephaly, and 100% with polydactyly in both the forelimb and hindlimb (9/9). Surprisingly, we also noted phenotypes never seen in any Ttc21b aln/aln mutant. One of the double mutant embryos had quite dramatic spina bifida aperta (Fig 10G), a severe neural tube defect. 62.5% (5/8) of the remaining embryos had morphology consistent with the considerably milder spina bifida occulta. Both of the previously mentioned~E17.5 double mutant embryos also had obvious spina bifida occulta (Fig 10J). We then rigorously examined all the embryos recovered from this cross for spinal neural tube defects (NTD) as this is not a phenotype we had seen in previous analyses of the Ttc21b aln/aln mutants [16,19]. We did not note any spinal NTDs in Gpr63 Del/Del; Ttc21b wt/wt (Fig 10B; n = 5) or Gpr63 Del/Del; Ttc21b aln/wt embryos (Fig 10D;  n = 13). In Gpr63 wt/wt; Ttc21b aln/aln embryos, we did note a previously unappreciated thickened neural tissue in one mutant from E10-E14 (Fig 10C; n = 1/9), and a disruption of the overlying ectoderm possibly consistent with spina bifida occulta from E15-17 (Fig 10H; n = 1/4). Based on our previous studies with the Ttc21b aln/aln mutant where the NTDs were not noticed in studies performed on similar genetic backgrounds, we believe these to be relatively uncommon phenotypes in mice with just the Ttc21b aln allele [16,19]. Gpr63 Del/wt; Ttc21b aln/aln embryos with an additional deleterious Gpr63 allele however, showed a significantly increased incidence of NTDs (Fig 10E and 10I, n = 7/14). However, we never see the spina bifida aperta phenotype in embryos other than Gpr63 Del/Del; Ttc21b aln/aln double mutants. Even in this genotype, we have seen it once in this study. We therefore suggest it is a relatively rare phenotype, but is still restricted to the Gpr63 Del/Del; Ttc21b aln/aln double mutant genotype.  Thus, we conclude that an allele of Gpr63 with a truncation of the cytoplasmic tail genetically interacts with Ttc21b, such that double mutants are mostly embryonic lethal at stages prior to any lethality seen in either single mutant. The remaining animals have an increased incidence of NTDs unique from either single mutant. The mechanism(s) leading to this lethality will be the subject of future study. We further suggest that Gpr63 is likely to be a causal gene within the Moaq1 QTL and reduced function of GPR63 in the FVB and A/J genetic backgrounds leads to the more severe forebrain phenotype of the Ttc21b aln/aln homozygous mutants on these backgrounds, as compared to those recovered on a B6 inbred background.

Discussion
In this study, we sought to understand the genetic basis for the influence of mouse inbred strain background on the severity of the Ttc21b aln/aln homozygous mutant microcephaly phenotype. Our QTL analysis identified two loci (Moaq1, Moaq2) which interact with Ttc21b to control brain size in Ttc21b aln/aln mutants. These were validated by constructing QTL-containing congenic lines for each. The 95% confidence interval of Moaq1 was small enough to allow a candidate gene approach, and Gpr63 was selected for further study. Experiments directly testing Gpr63 function with novel mouse alleles generated with CRISPR-Cas9 genome editing support our hypothesis that reduced Gpr63 function further compromises embryos lacking functional Ttc21b. We further demonstrated that Gpr63;Ttc21b double mutants had earlier lethality than either single mutant as well as novel spina bifida phenotypes.

Meiotic recombination or genome editing
Traditionally, the identification of a QTL is followed by a laborious mapping process and the creation of congenic strains to validate the QTL and begin to hunt for the truly causal sequence difference [49,50]. The combination of deeply sequenced inbred mouse strains, combined with the capabilities conferred by new genome editing tools such as CRISPR-Cas9 to rapidly generate sequence specific alterations in novel mouse lines, can potentially greatly facilitate the identification of causal genetic variants within a QTL. In this case, we identified a potentially deleterious SNP in Gpr63 on the FVB and A/J genetic backgrounds which may cause reduced function as compared to the canonical sequence seen in B6 mice. We did pursue a traditional congenic backcross approach by continued backcrossing to FVB towards creating a true congenic line which would be "pure" FVB with only dilution by the Ttc21b aln locus itself and the B6 Moaq1 region. However, this was confounded by the increased incidence of exencephaly and/or early embryonic lethality seen in Ttc21b aln/aln mutants when maintained on the >90% FVB genetic background. Presumably this increased frequency in exencephaly and/or death was due to the remaining 10% of the heterozygous genome that was further purified to FVB during this backcrossing. This is an intriguing finding that could potentially facilitate mapping those modifiers independently. In parallel, we attempted to bypass the meiotic recombination and QTL refinement entirely by recreating the specific Gpr63 SNP and determining if it affected brain size in the Ttc21b mutant background. In the F2 recombinants generated for the QTL analysis, we did not recover embryos both without exencephaly and homozygosity for B6 at a region on chromosome 14. This suggested to us that the Ttc21b aln allele on a pure B6 background may be early embryonic lethal; therefore, we chose to test our Gpr63 hypothesis by creating the B6 SNP on an FVB background (FVB Gpr63 Arg(B6)/Ser(FVB) ). Further study has shown this lack of homozygosity on chromosome 14 was a statistical anomaly with no apparent biological significance as we have since recovered Ttc21b aln/aln mutants with B6 identity across this region. In contrast, the consequences of creating a pure inbred FVB strain with the Ttc21b allele was a strain with a high incidence of exencephaly preventing us from testing this Gpr63 hypothesis directly. We are now carrying out a further validation of the Moaq1 and Moaq2 QTL by constructing pure B6 congenics with FVB Moaq1 and Moaq2 intervals.

Gpr63 can localize to the primary cilia but is not essential for development
Our data suggest that the predicted deleterious Gpr63 407Arg(B6)/Ser(FVB) SNP is not essential for development. This is consistent with the fact this SNP is present in the reference sequence for A/J and FVB mice. Furthermore, the deletion of the extreme C-terminus of the GPR63 GPCR also did not seem to have an effect on survival (Fig 7). This is consistent with data from the International Mouse Phenotyping Consortium and Knockout Mouse Project that Gpr63 deletion alleles do not affect survival (https://www.mousephenotype.org/data/genes/ MGI:2135884).
We did use two different epitope-tagged expression constructs to demonstrate GPR63 at least partially localizes to the primary cilium (Fig 8). This is consistent with the BBSome binding motif located in the C-terminus. We further studied the ciliary localization of different alleles of Gpr63 in these overexpression experiments and have provided evidence that the Arg/ Ser transition and the cytoplasmic truncation reduced ciliary localization of GPR63 (Fig 8). Neither of these residues affect the presumed BBSome binding motif, suggesting that other residues contribute to the ciliary targeting of GPR63 through currently unknown mechanisms. The detailed analysis of this trafficking motif and potential physical interaction with TTC21B should be the subject of future study.

Gpr63 and Ttc21b genetic interactions
The alleles we generated with CRISPR-Cas9 editing did allow us to demonstrate a genetic interaction between Gpr63 and Ttc21b. The confounding effects of FVB genetic purity on exencephaly and the apparent chromosome 14 lethality interval did hamper some of the initially designed experiments. However, the truncated Gpr63 allele we recovered from our CRISPR-Cas9 studies allowed us to address this hypothesis and we do see that Ttc21b aln/aln ; Gpr63 Del/Arg(B6) embryos have smaller forebrain tissues when compared to Ttc21b aln/aln ; Gpr63 Ser(FVB)/Arg(B6 (Fig 9). The best evidence for a genetic interaction between Gpr63 and Ttc21b, however, is the novel phenotypes seen in the Ttc21b aln/aln ;Gpr63 Del/Del embryos. The lethality of these embryos is earlier than what we observed in Ttc21b aln/aln mutants and we have never previously seen some of the spina bifida phenotypes observed in the Ttc21b aln/aln ; Gpr63 Del/Del embryos.
Here we identify regions of the genome linked to the differential effect of inbred strain background on the severity of the microcephaly phenotype seen in Ttc21b aln/aln mutants. We follow this up with a new allelic series of Gpr63 and experiments in vitro and in vivo to demonstrate these polymorphisms do affect Ttc21b phenotypes and Gpr63 function. Although loss of Gpr63 alone does not seem to cause a significant phenotype, this work suggests Gpr63 may be a risk factor underlying some of the variability seen in the ciliopathies. The approach taken here may be quite useful in understanding modifier effects in any number of structural birth defects.

Ethics statement
All animals were housed under a protocol approved by the CCHMC Institutional Animal Care and Use Committee (#2016-0098) in standard conditions. All euthanasia and subsequent embryo or organ harvests were preceded by Isoflurane sedation. Euthanasia was accomplished via dislocation of the cervical vertebrae.

Animal husbandry
For embryo collections, noon of the day of vaginal plug detection was designated as E0.5. The Ttc21b aln allele used in this study has been previously published [16]. Ttc21b aln mice were serially backcrossed to C57BL/6J (JAX:000664) and FVB/NJ (JAX:001800) mice to produce strain specific Ttc21b aln mice. Genotyping was performed by PCR, Sanger Sequencing, or Taqman assays (S4 Table).

Forebrain surface area measurement
Brains were microdissected from embryos and photographed prior to fixation from the dorsal aspect on a Zeiss Discovery.V8 dissecting microscope. Zeiss Axiovision software was used to mark the boundaries of the cortical surface area for all brain examined, as shown in S1 Fig. Cortical surface area was quantified by the Zeiss Axiovision software and recorded. All images used to quantify surface area for the QTL analysis were taken on the same microscope at the same magnification.

QTL analysis
GigaMUGA and MegaMUGA genotyping of recombinants was performed by GeneSeek (Neogen; Lincoln, NE). Quality and intensity normalization of the SNP data was checked using routines provided in the R-package Argyle [51]. Linkage mapping, including scanone and scantwo analyses, was performed on informative SNP markers and brain sizes (mm 2 ) of ninety-six F 2 recombinants using the R/qtl software package (Broman 2003;Broman and Sen 2009). Threshold levels of significance were established using at least 1,000 permutations of the respective dataset.

CRISPR-Cas9 genome editing
Donor oligonucleotide and guide RNA vector sequences were designed using Benchling (Benchling, 2016). sgRNAs were validated using Benchling and CRISPRscan with the following scores: Benchling-48.5 and CRISPRscan-49 (Benchling, 2016) (Moreno-Mateos, et al., 2015). The donor oligonucleotide was supplied by Integrated DNA Technologies (sequence with lower case letters representing coding change: CCATAATCGCTCGATATGTTTCA AGAGTTCGGTATTCACAACACCGTCCGATGTTCCCCACACACGTAGACGGCAC TAGGGCGAATgCGTCGcCTTGTaTGcCCgGGGAGCCGTGGCAAGAACT. The sgRNA (sequence: TCCCTGGTCACACAAGTCGA) was synthesized and injected into FVB mouse zygotes together with donor oligonucleotide by the CCHMC Transgenic Animal and Genome Editing Core. Gpr63 mice were made by injection into FVB zygotes. The male founder mice were genotyped and those with evidence of edited GPR63 loci were bred to FVB-Ttc21b aln/wt females. Offspring were analyzed by PCR and Sanger sequencing of approximately 1000 base pairs of the Gpr63 locus around the sgRNA recognition sequence. Initial crosses were designed to rapidly assess the hypothesized interaction between Gpr63 and Ttc21b. Carriers for the SNP allele (Gpr63 Arg(B6) ) were maintained by further breeding to FVB-Ttc21b aln/wt females to immediately ascertain an effect on brain size. These were then further maintained on FVB to characterize the Gpr63 phenotype in isolation. Carriers for the Gpr63 Del allele were also bred to FVB-Ttc21b aln/wt (for at least two generations). In order to compare Gpr63 Arg with the Gpr63 Del allele, Gpr63 Del ;FVB-Ttc21b aln/wt mice heterozygous for both alleles were crossed to B6 wildtype mice.
Analysis of localization of GPR63-myc was done with Imaris 6.2.1 software. First, a surface overlying cilia axonemes labeled with Arl13b was initiated in the 488 channel, then another surface was initiated in the 596 channel to detect the cilia labeled with myc and flag. Finally, the MATLAB plugin surface-surface colocalization was utilized to detect the areas of colocalization between Arl13b stained cilia and myc-flag tagged cilia axoneme. The number of double labeled cilia were normalized against the total number of Arl13b stained cilia per field of view.
Statistical analysis and One-way ANOVA was performed using GraphPad Prism software. All samples were blinded to the investigator performing the analysis.

GPR63 protein domain analysis
Features of the GPR63 protein were identified through literature searches and the Modular Domain Peptide Interaction Server: http://modpepint.informatik.uni-freiburg.de/.

Statistical analysis
GraphPad Prism was used for all graphical analysis. Data are shown as mean ± 95% confidence interval unless otherwise noted. P values are explicitly stated in almost every instance, rather than asserting significance at a certain conventional threshold (e.g., p = 0.05). Fig 1G: Two-way ANOVA followed by unpaired t-test of each strain wild-type vs mutant; Figs 2D, 2F, 7A-7D, 7I-7L and 7O: One-way ANOVA followed by Tukey's multiple comparisons when ANOVA indicates significance; Figs 2E, 2G and 3 -paired student t-test;  [48] indicates ciliary localization similarly to the results with pCMV6-GPR63-myc. (TIF) S1 Table. Genome-wide SNP data. Genome-wide SNP data to determine background purity; used for initial congenic FVB and B6 strains carrying Ttc21b aln allele. (XLSX) S2 Table. Genome survey and brain size of F2 mutants. GigaMuga data and brain size of 96 F2 mutants used for initial QTL analysis. (XLSX) S3