An ENU-induced splice site mutation of mouse Col1a1 causing recessive osteogenesis imperfecta and revealing a novel splicing rescue

GU-AG consensus sequences are used for intron recognition in the majority of cases of pre-mRNA splicing in eukaryotes. Mutations at splice junctions often cause exon skipping, short deletions, or insertions in the mature mRNA, underlying one common molecular mechanism of genetic diseases. Using N-ethyl-N-nitrosourea, a novel recessive mutation named seal was produced, associated with fragile bones and susceptibility to fractures (spine and limbs). A single nucleotide transversion (T → A) at the second position of intron 36 of the Col1a1 gene, encoding the type I collagen, α1 chain, was responsible for the phenotype. Col1a1 seal mRNA expression occurred at greatly reduced levels compared to the wild-type transcript, resulting in reduced and aberrant collagen fibers in tibiae of seal homozygous mice. Unexpectedly, splicing of Col1a1 seal mRNA followed the normal pattern despite the presence of the donor splice site mutation, likely due to the action of a putative intronic splicing enhancer present in intron 25, which appeared to function redundantly with the splice donor site of intron 36. Seal mice represent a model of human osteogenesis imperfecta, and reveal a previously unknown mechanism for splicing “rescue.”

type I collagen synthesis or assembly resulting in osteogenesis imperfecta (OI), a genetic disorder characterized by bone fragility and deformity, blue sclera, short stature, dentinogenesis imperfecta, and hearing loss. To date, more than 1,500 mutations in COL1A1 and COL1A2 have been identified in patients with OI, among which nonsense, frame shift, and splicing mutations often cause quantitative deficiency in the pro-α chains, whereas missense mutations lead to aberrant pro-α chains that exert a dominant negative effect on collagen synthesis [2][3][4] . The most common missense mutations are glycine substitutions within the Gly-X-Y in the triple helix. More severe clinical phenotypes manifest in OI patients with helical glycine mutations than those with other mutations causing quantitative procollagen deficiencies 5,6 .
COL1A1 and COL1A2 contain approximately 52 intronic sequences that need to be precisely excised to generate mature mRNA, making these genes particularly susceptible to splicing mutations 7,8 . Accurate splicing depends on splice sites, conserved sequence elements positioned at the 5′ ends (5′ss or splice donor sites) and 3′ ends of introns (3′ss or splice acceptor sites) that are recognized by components of the spliceosome 9 . Greater than 99% of mammalian introns, including all of those in COL1A1, are spliced by the major (U2-dependent) spliceosome and have as their terminal dinucleotides GU (5′ end) and AG (3′ end) 10 . These dinucleotides are invariant in splice sites of major (U2-type) introns and are therefore considered critical elements of them. In contrast, flanking nucleotides (up to 3 bp into the adjacent exon and 8 bp into the intron for 5′ss) may deviate from consensus sequences resulting in variation in splice site strength 11 . Splice donor and acceptor site mutations can lead to exon skipping, use of cryptic splice sites, and/or insertions/deletions; many human diseases stem from such defects caused by splice site mutations, and their mechanisms have been well documented 12 . Importantly, the effects of splice site mutations are determined in part by the order and rate of intron removal from the pre-mRNA 13,14 . Exon skipping mutations causing lethal or moderate phenotypes of OI have been identified in both splice donor and acceptor consensus sequences in COL1A1 and COL1A2 genes 15,16 . Novel causative mutations for OI continue to be identified 17 .
Here we report the identification and characterization of a novel N-ethyl-N-nitrosourea (ENU)-induced Col1a1 mutation named seal that causes OI in homozygous mice. Analysis of the effects of the donor splice site mutation led to the discovery of a regulatory element for splicing within intron 25, acting on the excision of intron 36.

Results
Identification of the seal phenotype and its genetic cause. The recessive seal phenotype was initially recognized as a defect of hind limb movement induced by grasping the loose skin over the nape of the neck, as is commonly practiced during routine handling of laboratory mice. Once triggered in this manner, the hind legs become paralyzed for a period of about 8 days before they regain function. However, most of the mice still display a residual "seal-like" gait after the recovery. Homozygous seal mice show shortened limbs due to a reduction in the length of the long bones relative to that of wild-type littermates ( Fig. 1A-C). About 50% of seal homozygotes also have swollen heels and foot pads, occasionally with deformed feet due to pathologic fracture (Fig. 1D). Necropsy of seal mice that showed abnormal locomotion revealed spinal bone fracture that presumably caused hind limb paralysis. Body weight of seal mice was reduced 8% compared to those of wild-type mice throughout the period of rapid growth between 6 and 12 weeks of age (Fig. 1E). All described phenotypes were transmitted in a recessive manner and heterozygotes were indistinguishable from wild-type mice.
The mutation causing the seal phenotype was mapped to chromosome 11 by genome-wide linkage analysis using a panel of 59 microsatellite markers ( Fig. 2A); 4 additional chromosome 11 markers were used to confine the mutation to a 2.1 Mbp critical region containing 81 genes (Fig. 2B). Residing in the critical region, Col1a1 encoding the type I collagen, α1 chain was considered a promising candidate since type I collagen mutations generally result in bone fragility 15 . Sequencing of Col1a1 identified a single nucleotide transversion (T → A) in the donor splice site of intron 36 at position + 2 relative to the exon 36 boundary (Fig. 2C,D).
Defective bone structure, turnover, and collagen network in seal mice. Type I collagen is a key structural component of bone, and we therefore examined bone structure in seal homozygotes. Hematoxylin and eosin staining revealed fine trabecular bone in the metaphysis of seal tibiae compared to that in wild-type controls ( Fig. 3A-D), however, the trabecular number was not markedly different between the two groups by gross visual evaluation. Moreover, artificial fissures indicative of bone fragility were often discovered in the bone sections of seal homozygotes but not in those of wild-type mice (Fig. 3A,B). Micro-computed tomography (CT) images showed thinner cortical bone in seal femurs in comparison with wild-type femurs (Fig. 4). Assessment of trabecular and cortical parameters revealed significant defects in seal mice compared to wild-type controls. For example, cortical thickness and cortical bone area fraction were reduced 14% (P = 0.0016) and 24% (P < 0.0001), respectively, in seal homozygotes relative to wild-type mice. Seal mice also showed a significant decrease of trabecular bone mineral density (BMD; P = 0.0002), in agreement with the histological observation of fine trabeculae with reduced bone deposition (Table 1). Transmission electron microscopy (TEM) demonstrated many intact osteoblasts containing well-developed cellular organelles such as rough endoplasmic reticulum and Golgi apparatus, indicating the active form of osteoblasts in both wild-type and seal tibiae (Fig. 3E,F,I,J). The wild-type bone matrix showed densely-deposited collagen fibrils (See black fibrillar structures in Fig. 3G), while seal bone matrix contained sparsely-distributed collagen fibrils, consequently including many foci of organic materials (arrows in Fig. 3H). The effect of the seal mutation on steady state bone resorption in vivo was determined by measurement of the serum level of the type I collagen α1 chain C-terminal telopeptide (CTX), a biomarker for osteoclast activity. CTX levels in seal mice were reduced approximately 65% compared to those in wild-type mice (Fig. 5).
Type I collagen biosynthesis is a complex process that includes multiple post-translational modifications (e.g. hydroxylation of proline) leading to the formation of covalently cross-linked collagen fibrils 18 . To understand the effects of collagen mutations, it is important to analyze collagen quantity as well as collagen components, which are crucial determinants of bone mechanical properties 19 . To analyze the effect of the seal mutation on the collagen amount in bone, hydroxyproline content was evaluated in demineralized bone hydrolysate 20 . The bone samples from seal homozygotes contained significantly less hydroxyproline indicating reduced collagen content compared with wild-type bone (Fig. 6A). These data support the conclusion that the seal phenotype is due to reduced collagen, rather than accelerated bone turnover.
To examine the composition of type I collagen in bone from seal mice, collagen components from femurs were extracted, separated by SDS-PAGE, and visualized by CBB staining (Fig. 6B and Supplementary Fig. S1). Band intensities were higher for seal bone samples, reflecting increased collagen extractability relative to wild-type bone samples, which was supported by quantitation using densitometric image analysis (Fig. 6C). The quantitation also revealed an α1 (I)/α2 (I) chain ratio of 2.0 in control mice, reflecting the formation of a heterotrimer of two α1 (I) chains and one α2 (I) chain. Notably, the α1 (I)/α2 (I) chain ratio in seal mice was elevated to 3.6, suggesting the formation of some α1 (I) homotrimers (Fig. 6D).
β-chains refer to α-chain dimers in which the two α-chains are linked by intra-or intermolecular covalent cross-links, which remain intact under the conditions of SDS-PAGE while non-covalent bonds of the type I collagen triple helix are destroyed. We analyzed the composition of such dimers based on the distinct migration in SDS-PAGE of each possible dimer. In control mice, β12 chains [heterodimers of α1 (I) and α2 (I)] predominated, whereas β11 chains were the major form in seal mice, consistent with the possible formation of α1 (I) homotrimers in seal mice (Fig. 6E). These findings indicate that both the amount and the composition of collagen were altered in seal homozygous mice.
Reduced Col1a1 transcripts in seal mice but normal splicing. Donor splice site mutations often result in exon skipping, and the seal mutation was predicted to cause skipping of the 108-nucleotide exon 36, resulting in an in-frame deletion of 36 amino acids near the middle of the α1 chain helical domain. Using quantitative RT-PCR, we confirmed that Col1a1 transcripts were significantly reduced in total RNA, nuclear RNA, and cytoplasmic RNA Sequencing of Col1a1 transcripts across the junction between exon 35 and the next exon revealed normal splicing of exon 35 to exon 36 in RNAs isolated from seal femurs; normal splicing between exon 36 and exon 37 was also observed (Fig. 8A), although we cannot exclude the possibility that minor amounts of abnormal transcripts were present below the level of detection. The recessive nature of the seal phenotype suggests that the quantity of aberrant type I collagen α1 chains produced, if any, is insufficient to exert a dominant negative effect. These data suggest that the seal mutation slows the rate of splicing such that Col1a1 mRNA levels are diminished in seal bone compared to those in wild-type bone, despite correct splicing of exon 35 to 36 and exon 36 to 37.
To examine the effect of the seal mutation on Col1a1 pre-mRNA splicing, we constructed four Col1a1 minigenes containing the seal mutation and assessed mRNA splicing following transfection into HEK293 cells (Fig. 8B). Minigenes exon 26-39 (Fig. 8C, right) and exon 34-43 (Fig. S2, right) yielded transcripts in which exon 36 was mostly skipped; we noted that normally spliced transcripts of these minigenes were detected by DNA sequencing but not by staining in gels, likely due to a low abundance of normally spliced transcripts below the level of detection by ethidium bromide staining. In contrast, minigene exon 25-45 (Fig. S2, left) produced properly spliced transcripts. We hypothesized that the presence of intron 25 promoted inclusion of exon 36 in the spliced minigene mRNAs. To test our hypothesis, we constructed a minigene in which intron 25 was included in minigene exon 26-39 (exon 26-39 + intron 25; Fig. 8B), and examined splicing upon expression in HEK293 cells. Addition of intron 25 promoted inclusion of exon 36 in the majority of transcripts, although residual transcripts lacking exon 36 were still produced (Fig. 8C, left). These data suggest that Col1a1 intron 25 may contain elements that compensate for the splicing error caused by the splice site mutation and support normal splicing of exon 36.

Discussion
The helical domain of murine type I collagen α1 chain is encoded by 43 of the 51 total exons of Col1a1; these 43 exons encode the repeating sequence Gly-X-Y, and each begins with a glycine codon and ends with a Y-position codon. Because a glycine residue at every third position of the chain is critical to the formation of the triple helix of mature type I collagen 21 , frameshifts or premature termination caused by aberrant splicing can be detrimental for type I collagen synthesis 15,16 . We speculate that multiple redundant mechanisms have evolved to ensure proper splicing of collagen mRNAs. In support of this hypothesis, only normal Col1a1 transcripts were detected in bone tissue of seal mice, despite the presence of a mutation in the invariant GU dinucleotide of the intron 36 donor splice site. Results of splicing analyses using minigenes suggested that either an intact 5′ss in intron 36 or the presence of a putative intronic splicing enhancer in intron 25 was necessary for proper splicing of exon 36 in minigene 26-39. An implication of this finding is that removal of intron 36 precedes removal of intron 25. However, the intron 25 regulatory element appeared to be less efficient than the 5′ splice site of intron 36 in directing splicing, as evidenced by the overall reduction in Col1a1 transcript abundance in bone from seal mice relative to wild-type mice, which was also observed in the minigene analysis. The observation of a low frequency of spliced transcripts lacking exon 36 among intron 25-containing seal minigene transcripts (Fig. 8C, left), as well as a low frequency of normally spliced seal minigene transcripts lacking intron 25 (Fig. 8C, right), suggests that other cis-acting splicing regulatory elements outside intron 25, or trans-acting regulators, also contribute to normal exon 36 splicing.  Human OI manifests a wide spectrum of severity as well as variability of causative mutations. Administration of bisphosphonates was shown to effectively increase vertebral areal bone mineral density and height. However, concerns have been raised as to its efficacy for fracture reduction [22][23][24][25] . This suggests that understanding the relationship between clinical manifestation and underlying pathogenesis is necessary for the development of effective therapy, and animal models mimicking various types of human OI are pivotal to these studies. Seal mutant mice, showing short limbs with short undermineralized long bones and sporadic limb deformity, model human type III OI. Oim, a spontaneous mutation of Col1a2 encoding the type I collagen pro-α2 chain 26 , causes a phenotype similar to seal. However, the underlying mechanisms are different. The oim mutation causes aberrant pro-α2 (I) collagen synthesis that inhibits assembly of a normal type I collagen trimer 26,27 . In contrast, type I collagen in seal mice was greatly reduced due to a decrease in transcription of the α1 chain, and that which was produced consisted, at least in part, of α1 (I) homotrimers, which have been associated with impaired bone strength leading to increased risk of bone fracture 26,28-31 . In normal type I collagen, the hydrophobicity of the α2 (I) chain is thought to promote the stability of the heterotrimer by increasing the hydrophobic interactions between the heterotrimeric molecules, and increasing the binding of the molecules in the fiber 32 . Therefore, the elevated α1 (I)/α2 (I) chain ratio of type I collagen from seal mice may signify a reduced efficiency of self-assembly and loose-packing collagen fibers. In addition, it has been shown that each tissue has a unique collagen cross-link pattern that supports the tissue′s mechanical features. An abnormal pattern of collagen cross-linking is often observed in aged and diseased bone, making it brittle or fragile 33 . Our results showed that the composition of β-chains differed between type I collagen in bone from seal versus wild-type mice; this abnormal collagen cross-link pattern may also contribute to decreased fracture strength of bone in seal mice. Seal mice provide a valuable disease model of human OI, in which Col1a1 splicing regulation and its effects on transcript and protein abundance, and on type I collagen fiber formation may be investigated.

Methods
Mice. C57BL/6 J and C3H/HeN mice were obtained from The Jackson Laboratory (Bar Harbor, ME) and Taconic Biosciences (Germantown, NY) and maintained under specific pathogen-free conditions in The Scripps Research Institute vivarium and Niigata University animal facility. All male mice used in the experiments were  4-12 weeks in age. Animals were to be excluded from analysis only if they displayed obvious illness or death; these conditions were not observed and no animals were excluded. No randomization of the allocation of animals to experimental groups was performed.
Data availability. All data generated or analyzed during this study are included in this published article (and its Supplementary Information files). The seal strain (Col1a1 m1Btlr ; MGI: 3776559) is described at http:// mutagenetix.utsouthwestern.edu and is available from the Mutant Mouse Regional Resource Center (MMRRC: 030348-UCD).
Ethics Statement. All experimental procedures using mice were approved by and conducted in accordance with The Scripps Research Institute Institutional Animal Care and Use Committee, and Niigata University institutional guidelines for animal care and use. The protocol to perform euthanasia by cervical dislocation after intraperitoneal injection of chloral hydrate and to obtain specimens was approved by the animal ethics committee for animal experimentation of Niigata University (Permit Number: 39). Any unnecessary grasping of seal homozygous mice by the scruff of the neck was avoided and all efforts were made to minimize suffering.
ENU Mutagenesis, phenotypic screens, and linkage analysis. Random germline mutagenesis of C57BL/6 J mice using ENU was described previously 34 . Phenotypic screening including was applied to G3 and G1 mice. Phenotypic screens included casual inspection for immunodeficiency and dysmorphologies affecting limbs, tail, eyes, teeth, or other aspects of body form; coat color and/or coat quality anomalies, abnormal body size [35][36][37][38][39][40] . Homozygous seal mice were mated to wild-type C3H/HeN mice, and their progeny were backcrossed to the homozygous mutant stock. 34 F2 mice were scored for phenotype and genomic DNA was prepared from tail tips for genotyping. 59 microsatellite markers were used for genome-wide linkage analysis. In vitro pre-mRNA splicing assay. Col1a1 mRNA processing was analyzed using a minigene assay. Briefly,  Supplementary Table S1. HEK293 cells (DS Pharma Biomedical Co., Ltd, Osaka, Japan) transiently transfected with purified minigene plasmids were harvested 48 h post transfection, and then total RNA was extracted using TRIzol ® reagent (Invitrogen). The RNA was reverse transcribed by M-MLV reverse transcriptase (Invitrogen) in to cDNA using random primers (Takara Bio Inc., Shiga, Japan). Sequence analysis was performed using primer covering exon 36 and 37 to analyze the exon skipping.
Quantitative reverse transcription RT-PCR. Femurs RNA was extracted using TRIzol ® reagent (Invitrogen) to obtain total RNA or cytoplasmic and nuclear RNA purification kit (Norgen, ON, Canada) to obtain cytoplasmic and nuclear RNA according to the manufacturer's instructions. Total RNA (1 μg) was reverse transcribed by M-MLV reverse transcriptase (Invitrogen) using random primers (Takara Bio Inc., Shiga, Japan). Grinded specimens for femurs RNA extraction were prepared using SK-mill, after removing bone marrow and subsequent deep freezing in liquid nitrogen. Micro-CT analysis. Micro-computed tomographic scans were performed on excised femurs and morphological analysis was performed using micro-CT (SkyScan 1174, Bruker microCT, Kontich, Belgium) with an X-ray tube voltage of 50 kV and current of 800 μA, as described by Cano et al. 41 . The angular rotation was 185°, and the angular increment was 0.45° for scanning. The voxel size was set at 6.5 μm isotropically. A modified Feldkamp algorithm was used for reconstruction of data sets and segmentation into binary images (8-bit BMP images) was carried out using adaptive local thresholding. The microarchitectural properties of trabecular and cortical bone regions were evaluated within a conforming volume of interest (VOI). A VOI in the trabecular bone region was started at a distance of 1 mm from the distal growth plate, extending a further 2 mm of longitudinal distance in the proximal direction (96 image slices). The regions of trabecular bone were consisted of cylindrical segments (radius 0.86 mm). A VOI included in the middiaphyses (96 images) was selected in the cortical bone region. Cortical bone regions were selected by free drawing regions of interest. Cortical thickness (mm), mean total crossectional cortical bone area (mm 2 ), mean total crossectional tissue area (mm 2 ), mean total crossectional tissue perimeter (mm), cortical bone area fraction (%) were analyzed. Cortical bone mineral density (Cortical BMD) and trabecular bone mineral density (Trabecular BMD) were calculated using the conforming VOI. Reconstructed 8-bit BMP images have a grey value between 0 and 255 in every pixel. 255 was assumed to be white (void space), whereas 0 is black, the densest part of the image. Hydroxyapatite phantom rods (2 mm of diameter) immersed in pure water, equivalent to BMD of 0.25 g/cm 3 and 0.75 g/cm 3 were employed for calibration to express grey values as mineral content.

Metabolism of bone type I collagen.
Degradation of bone type I collagen was evaluated, measuring C-terminal cross-linking type I collagen fragments (CTX) in serum by RatLaps ™ (CTX-I) ELISA kit (Immunodiagnostic Systems Limited, Boldon, UK) according to the manufacturer's protocol.
Quantification and qualification of collagen. Quantification of collagen contents in femurs was evaluated by hydroxyproline assay. Femurs were collected from 10-week-old mice. After completely remove the connective tissues, both ends of femurs were cut off, and bone marrow was washed out by ice cold phosphate buffer saline. The cleaned bone samples were demineralized with 10% EDTA for 1 week, dialyzed against water using Spectra/Por (MWC 3,500 Da, Spectrum Laboratories, Inc., Milipitas, CA) for 4 days, and lyophilized. Sample preparation was performed below 4 °C unless otherwise specified. Equal weight of samples were hydrolyzed by 12 M HCl for 20 h at 95 °C. Quantification of hydroxyproline, representing total collagen amount, was performed by a Total collagen assay kit (QuickZyme Biosciences, Leiden, Netherlands), according to the manufacturer's instruction. Collagen components were analyzed using demineralized and lyophilized bone samples. Samples of equal weight were directly resolved in sodium dodecyl sulfate (SDS) sample buffer (Life Technologies, Carlsbad, CA), heated for 10 min at 80 °C, and centrifuged at 13,000 × g for 20 min. Equal volume of supernatants were loaded onto the NuPAGE 3-8% Tris-Acetate Gel (Life Technologies), and the electrophoresis was performed at constant voltage of 150 V for 60 min. Gels were stained with Coomassie Brilliant Blue R (CBB, Sigma-Aldrich, St Louis, MO). Digital images were taken by Image Scanner GT-X970 (Epson, Nagano, Japan). Each band, corresponding α-, βand γ-chains of type I collagen were quantified by ImageJ software. Collagen extractability (α + β + γ), α1/ α2 chain ratio and composition ratio of β-chains were calculated.
Statistical analysis. Comparisons of differences were between two unpaired experimental groups in all cases. An unpaired t-test (Student's t-test) is appropriate and was used for such comparisons. The phenotypic performance of mice (C57BL/6J) is expected to follow a normal distribution, as has been observed in large datasets from numerous phenotypic screens conducted by our group. Variation within each dataset obtained by measurements from mice was assumed to be similar between genotypes since all strains were generated and maintained on the same pure inbred background (C57BL/6J); experimental assessment of variance was not performed.
The statistical significance of differences between experimental groups was determined using GraphPad Prism 5 (GraphPad Software Inc., La Jolla, CA,) and the Student's t-test (unpaired, two-tailed). P < 0.05 was considered statistically significant and indicated by *P < 0.05 and **P < 0.001. No pre-specified effect size was assumed, and in general 3-5 animals or replicates for each genotype or condition were used in experiments; this sample size was sufficient to demonstrate statistically significant differences in comparisons between two unpaired experimental groups by unpaired t-test. The investigator was not blinded to genotypes or group allocations during any experiment.