Functional analysis of a hypomorphic allele shows that MMP14 catalytic activity is the prime determinant of the Winchester syndrome phenotype

Abstract Winchester syndrome (WS, MIM #277950) is an extremely rare autosomal recessive skeletal dysplasia characterized by progressive joint destruction and osteolysis. To date, only one missense mutation in MMP14, encoding the membrane-bound matrix metalloprotease 14, has been reported in WS patients. Here, we report a novel hypomorphic MMP14 p.Arg111His (R111H) allele, associated with a mitigated form of WS. Functional analysis demonstrated that this mutation, in contrast to previously reported human and murine MMP14 mutations, does not affect MMP14’s transport to the cell membrane. Instead, it partially impairs MMP14’s proteolytic activity. This residual activity likely accounts for the mitigated phenotype observed in our patients. Based on our observations as well as previously published data, we hypothesize that MMP14’s catalytic activity is the prime determinant of disease severity. Given the limitations of our in vitro assays in addressing the consequences of MMP14 dysfunction, we generated a novel mmp14a/b knockout zebrafish model. The fish accurately reflected key aspects of the WS phenotype including craniofacial malformations, kyphosis, short-stature and reduced bone density owing to defective collagen remodeling. Notably, the zebrafish model will be a valuable tool for developing novel therapeutic approaches to a devastating bone disorder.


Introduction
In 2007, we reported two Dutch brothers with an autosomal recessive disorder consisting of dysmorphic facial features, mitral valve prolapse, severe acne and reduced bone density (1). We diagnosed them with Borrone syndrome, as their phenotype strongly resembled that of two patients first described by Borrone et al. (2). Symptoms in our patients were less severe, which we attributed to their younger age compared with Borrone's patients at the time of diagnosis. However, in the intervening years since diagnosis their phenotype has not appreciably worsened, suggesting that it is intrinsically milder.
In the patients originally reported by Borrone et al., we recently identified a homozygous splice site mutation in SH3PXD2B (3). Thus, Borrone syndrome is no longer considered as a separate entity, but as allelic to Frank-Ter Haar syndrome (FTHS, MIM #249420) (4). SH3 and Phox-homology (PX) Domaincontaining Protein 2B (SH3PXD2B, also known as TKS4) is an adapter protein required for functionality of podosomes (5). These are actin-rich membrane structures that mediate adhesion and invasive motility in a variety of cell types. Specifically, upon phosphorylation by c-SRC, SH3PXD2B recruits the membrane-bound matrix metalloprotease 14 (MMP14, also known as MT1-MMP) to the nascent podosome membrane (6). Here, MMP14 hydrolyzes intact fibrillar collagen and activates downstream effectors, including the gelatinase matrix metalloproteinase 2 (MMP2) that in turn can further degrade fragmented collagen fibrils (7)(8)(9). MMP14's collagenolytic activity is thought to be one of its most important functions in vivo, for which homodimerization is required (10). Loss of either MMP2 or MMP14 results in a spectrum of recessive skeletal dysplasias with osteolysis, encompassing multicentric osteolysis, nodulosis and arthropathy (MONA, MIM #259600) and Winchester syndrome (WS, MIM #277950). These disorders exhibit significant clinical overlap. Notably, WS is associated with mutations in MMP2 as well as in MMP14 (11,12).
In our patients, we found no mutation or deletion of SH3PXD2B. Subsequent homozygosity mapping and nextgeneration sequencing identified a novel homozygous missense mutation in MMP14, c.332G>A (www.LOVD.nl/MMP14), which was predicted to result in the substitution p.Arg111His (R111H; see Supplementary Material, Methods and Fig. S1) and assessed as possibly damaging by multiple algorithms (data not shown). Accordingly, we revised our patients' diagnosis to WS, pending confirmation of the mutation's pathogenicity.
MMP14 encodes a membrane-bound metalloprotease that requires removal of an N-terminal pro-domain sequence for its activation and presentation at the cell surface (13). The prodomain has two furin cleavage motifs, R 89 -R-P-R-C 93 and R 108 -R-K-R-Y 112 . Previously published work suggests that the latter motif is cleaved to generate the active enzyme (13,14). Therefore, we reasoned that the R111H mutation might interfere with cleavage and thereby impair MMP14 membrane localization and activation.
To test our hypothesis, we analyzed the consequences of the R111H change for MMP14's intracellular processing and functionality, comparing with known mutations associated with WS and similar mouse phenotypes. To better understand the connection between loss of MMP14 activity and the clinical manifestations of WS, we additionally generated a knockout (KO) zebrafish model. Our findings provide novel insights into the pathogenesis of the WS phenotype, with potential consequences for therapy.

Results
An in vitro model for assessing MMP14 processing and subcellular localization To examine MMP14 processing, we created a construct encoding either wild-type (WT) or mutant human pro-MMP14 with an N-terminal triple (3)-HA tag and a C-terminal enhanced green fluorescent protein (EGFP) (resulting in the fusion protein 3HA-MMP14-EGFP, Fig. 1A). Given correct processing of MMP14, the 3HA tag should not be detectable in a similar location to EGFP. The EGFP signal, on the other hand, should be visible at the Golgi/trans-Golgi network, the cell membrane and along the route by which the protein traffics. Aberrant processing and subsequent abnormal trafficking ought to be reflected in an altered subcellular distribution of both tags. Fusion proteins were exogenously expressed in MRC-5V1 immortalized human fetal lung fibroblasts (hereafter termed MRC5) and compared with control of 3HA-EGFP (Fig. 1B). This cell type is a good model for the disease phenotype, which includes connective tissue and lung abnormalities in humans and mouse models, respectively (15,16). Immunoblotting with a commercially available antibody demonstrated that MRC5 cells have very low endogenous MMP14 expression (Supplementary Material, Fig. S2). Thus, mutant protein localization is unlikely to be affected by dimerization with endogenous MMP14. Next, we determined that the localization of exogenously expressed double-tagged MMP14-WT was very similar to that of endogenous MMP14 (Supplementary Material, Fig. S3). Having established that our tagging strategy did not affect protein localization, we next compared and contrasted MMP14-WT and MMP14-R111H with previously characterized MMP14 missense mutants to determine why our patients' phenotype was comparatively mild.
MMP14 R111H is processed normally and is trafficked to the cell surface As shown in Figure 1C and Supplementary Material, Fig. S4, MMP14-R111H-EGFP (panel iv) localizes in a similar manner to that observed for MMP14-WT-EGFP (panel i). The lack of any HA signal at the cell membrane suggests that R111H does not markedly impact removal of the signal peptide. In western blot (WB) analysis [ Fig. 2A, lane 2 (WT) and lane 5 (R111H)] the banding pattern of R111H was similar to that of WT, with EGFP signal indicating removal of the HA tag and subsequent further processing of the resultant protein.
The human mutation p.Thr17Arg (T17R) has previously been reported as causing WS (12). This substitution is located within the conserved N-terminal signal peptide sequence of MMP14. T17R is thought to prevent proper localization to the ER and subsequent processing of MMP14. Accordingly, whilst the pro-enzyme was observed at the cell membrane by membrane fractionation and surface biotinylation, there was an absence of active MMP14 (12). We did not observe the T17R mutant at the cell membrane by immunofluorescence [IF, Fig. 1C(ii) and Supplementary Material, Fig. S4(ii)], most likely owing to differences in sensitivity between assays used. Rather, we found MMP14 T17R distributed throughout the cytoplasm. The (partial) co-localization of HA and EGFP tags observed by IF, as well as the overlapping HA and EGFP bands observed by WB ( Fig. 2A, lane 3) suggest that the T17R mutation impairs MMP14's signal sequence removal. The R92C mutation completely abolishes gelatin degradation, whereas the R111H mutant retains some residual activity. In contrast, the T17R and S466P mutations do not impair gelatin degradation. Error bars indicate SEM of biological triplicates. (D) Average migration rate of MMP14 fusion protein expressing MRC5 cells on fibronectin. Expression of MMP14-WT significantly stimulates migration compared with 3HA-EGFP expression, whereas the T17R, R92C and R111H mutations impair this stimulatory effect. In contrast, the S466P mutation does not affect MMP14-dependent migration. Significance levels: dot, P < 0.05; double dot, P < 0.01, triple dot, P < 0.001.
To further probe the relationship between MMP14 domains and disease, we generated double-tagged human versions of two MMP14 mutants associated with murine phenotypes that strongly resemble WS. The corresponding mutations are located in distinct MMP14 domains yet cause disorders that are virtually indistinguishable. The p.Arg92Cys (R92C) substitution, underlying the small and bugged-eyed (Sabe) phenotype, affects a conserved R 89 -R-P-R-C 93 furin cleavage motif (17). As seen in Figure 1C(iii) and Supplementary Material, Figure S4(iii), MMP14-R92C-EGFP was absent from the cell membrane, suggesting that R92C might impair the prodomain processing that is required for cell surface localization. The removal of the HA tag was confirmed by WB ( Fig. 2A, lane 4), however, the lack of any additional bands suggests that R92C impairs subsequent processing of MMP14.
The second murine mutation, p.Ser466Pro (S466P), causes the Cartoon phenotype (18). Serine 466 is a highly conserved residue in blade 4 of MMP14's hemopexin (Hx)-like domain, which is required for enzyme maturation and trafficking as well as for homodimer interactions (19,20). Figure 1C

MMP14 R111H retains partial pro-MMP2 hydrolyzing activity
Since MMP14-R111H seemed to be processed and trafficked normally, we next assessed the functionality of this mutant with respect to pro-MMP2 activation, utilizing gelatin zymography (7). First, we determined that medium conditioned by 3HA-EGFP expressing MRC5 cells did not activate pro-MMP2 (Fig. 2B, lane 6), consistent with low endogenous MMP14 levels in these cells. Subsequently, we assessed the pro-MMP2 activating potential of media conditioned by cells expressing the tagged WT or mutant MMP14 fusion proteins. As shown in Figure 2B MMP14 R111H retains partial gelatinolytic activity MMP14 has a key role in mediating invasive cell behavior (7). Thus, we determined the effect of MMP14 mutations on invasive extracellular matrix (ECM) degradation. We used Millipore's QCM TM Gelatin Invadopodia Assay system (Cat. No. ECM671) combined with live cell imaging. The results are summarized in Figure 2C, indicating the extent to which MRC5 cells expressing WT or mutant tagged MMP14 were capable of digesting the gelatin matrix on which they were seeded. In line with our previous localization and zymography data, the expression of MMP14-WT resulted in significantly more gelatin degradation than expression of 3HA-EGFP. R92C abolished all gelatinase degradation, whereas R111H resulted in a significant reduction yet retained partial activity. Intriguingly, in this assay the S466P mutant retained full activity when compared with the WT protein, despite membrane localization for this mutant being impaired.

MMP14 R111H has a reduced cell migration stimulatory potential
The gelatin invadopodia assay could not be used to assess MMP14's ability to promote cellular migration. We repeatedly observed that cells expressing tagged MMP14 seemed to adhere strongly to the poly-L-lysine coated coverslip in areas where gelatin was degraded, such that when the cells finally did move, fragments of cell membrane were left behind (data not shown) resulting in cell death. We hypothesized that abnormal cell migration might contribute to the disease phenotype, as knockdown of sh3pxd2a was previously observed to reduce the migrational velocity and podosome formation of zebrafish neural crest cells, resulting in craniofacial malformations (21). We therefore subjected cells expressing WT or mutant MMP14 fusion proteins to a phagokinetic assay as previously described (22). Given the cells' strong adherence to gelatin, we used fibronectin (1 mg/ml) as the substrate. As shown in Figure 2D, the T17R, R92C and R111H mutants demonstrated significantly reduced migratory behavior when compared with MMP14-WT, however, without showing a clear correlation with the severity of the associated clinical phenotypes. Unexpectedly, motility of cells expressing MMP14-S466P was not impaired compared with MMP14 WT expressing cells.

KO of mmp14a/b in Danio rerio recapitulates key aspects of the WS phenotype
Our observations suggest that loss of MMP14's catalytic activity might be a prime determinant of the WS phenotype. To better understand this link, we decided to generate a zebrafish (Danio rerio) KO model of WS. Danio rerio is a well-established vertebrate model system for the study of ossification and has several advantages over mice, such as rapid external embryonic development, the ability to generate large numbers of offspring and relatively low cost (23,24). MMP14 is well conserved between humans and zebrafish, consistent with significant functional overlap even though the latter have two copies, mmp14a and mmp14b, owing to genomic duplication (25,26). Previous work using morpholino oligonucleotide (MO)-mediated knockdown suggested that loss of mmp14a expression results in defective craniofacial morphogenesis, and hinted at abnormal matrix composition of craniofacial cartilage Anlagen. Observations on the consequences of mmp14b knockdown are conflicting. It has been proposed that mmp14a and mmp14b might have a role in cellular migration during gastrulation (25,27). However, it was later shown that this was probably a p53-mediated off-target MO effect, although additional non-p53-dependent off-target apoptosis might have contributed as well (28,29).
In light of these considerations, and because our data indicate that the mutation in our patients causes a loss of function, using CRISPR/Cas9 we knocked out mmp14a and mmp14b in zebrafish (30). Targeting Exon 4 in both mmp14a and mmp14b genes resulted in the introduction of 2bp and 8bp deletions, respectively (Supplementary Material, Fig. S5A), which are predicted to cause a frameshift and premature stop (p.S183Rfs23X and p.F190X, respectively). Using qPCR, we showed that the resulting mutant mRNA undergoes nonsense-mediated decay (Supplementary Material, Fig. S5B and C). We further confirmed that any residual protein product would not contain the WT sequence C-terminal to the mutation site (Supplementary Material, Fig. S5D).
In subsequent crosses, fish with all possible combinations of WT and mutant mmp14a/b alleles were generated. Independent of the parental genotype, mmp14a D/D ; mmp14b D/D fish (hereafter referred to as mmp14a/b KO) developed a gradually worsening phenotype, first noticed around 30 days post-fertilization (dpf) and most striking at adult age. Other genotypes were not associated with any overt abnormalities. The double KO phenotype includes decreased total body length (20.6 versus 32.5 mm for WT at 90 dpf, P < 0.0001), a relatively small head showing dorsal hyperextension in most individuals (86% versus 0%, P < 0.0001), exophthalmos, a short operculum, and thoracic hyperkyphosis ( Fig. 3A and B). Although mmp14a/b KO fish are born at Mendelian ratio, they have a shortened average life span of about 4.5 months and fail to reproduce. Microcomputed tomography (mCT) at 90 dpf revealed that mmp14a/b KO fish have a significantly reduced skull bone mineral density (BMD, 891.5 versus 959.0 mg/cc, P < 0.05) compared with WT fish (Fig. 3C and D), and confirmed Weberian-prehemal hyperkyphosis (angle formed by line connecting center of Weberian vertebral bodies with line connecting prehemal vertebral bodies is 40.7 in mmp14a/b KO fish versus 13.8 in WT, P ¼ 0.0002). Overall, our mmp14a/b KO fish recapitulates key aspects of WS and previous murine Mmp14 mutant and KO models (1,(16)(17)(18)(31)(32)(33).

mmp14a/b KO fish have abnormal endochondral and membranous ossification
Next, we sought to identify the disturbed process(es) underlying the mmp14a/b KO skeletal phenotype. At 5 dpf, larval craniofacial cartilage elements are qualitatively normal in size and shape (Supplementary Material, Fig. S6A and B). Subsequent mineralization of the larval axial skeleton proceeded at a normal pace (Supplementary Material, Fig. S6C-E). At 30 dpf, the first differences in calvaria shape became apparent (Fig. 4A).
Mid-sagittal sections of 90 dpf fish stained with hematoxylin and eosin (H&E) revealed irregular undulation of the membranous ossifying frontal and parietal bones of mmp14a/b KO fish, containing cell clusters giving these bones a threadbare appearance (Fig. 4B). The membranous ossifying jawbones contain a relatively small amount of peripheral bone matrix and a disorganized cartilage core (Fig. 4C). Similar abnormalities are seen in the endochondral ossifying supraoccipital bone (SOC), with additional cell-free areas in the cartilage core, devoid of proteoglycans ( Fig. 4E) (34). The relative absence of bone matrix likely accounts for the low skull BMD of mmp14a/b KO fish. Another striking finding in mmp14a/b KO fish is the ventral extension of the SOC and second supraneural, focally compressing the spinal cord against the first Weberian vertebral body (Fig. 4D). The resulting compromise of spinal cord integrity might contribute to the shortened lifespan of KO fish. This interpretation is supported by observations of an abnormal swimming pattern characterized by tumbling movements, typically observed in KO fish 1-2 days prior to death (Supplemental Material, Movie S1). Picrosirius red (PSR) staining of additional sections revealed abnormal collagen content of jawbones and the SOC of mmp14a/b KO fish, those bones lacking the collagen-rich cortex observed in WT fish ( Fig. 4E and Supplementary Material, Fig. S7). Polarized differential interference contrast imaging of PSR stained sections revealed no overt differences in birefringence, suggesting collagen deposition is unaffected by mmp14a/b KO (Supplementary Material, Fig. S7) (35). Finally, membranous ossifying Weberian vertebral bodies in mmp14a/b KO fish were irregularly shaped, with clusters of multinucleated cells in their dorsal aspect (Fig. 4G). Taken together, these findings highlight the importance of Mmp14-dependent collagen remodeling in bone formation.

Discussion
Our in vitro results, summarized in Table 1, provide evidence that a mild manifestation of WS as observed in our patients is caused by a novel homozygous missense mutation that partially impairs the catalytic activity of MMP14. Our observations  suggest that MMP14's ability to activate pro-MMP2 may be the most important determinant of disease severity in humans, as T17R seemed to affect both gelatinase activity and cell migration to a lesser extent than R111H, whereas it completely abrogated pro-MMP2 activation. The importance of MMP2 is underscored by previous reports that its loss can also cause WS and highly similar phenotypes (11,36,37).
Our findings on MMP14 processing are largely in line with previous work, however, some seem contradictory. In contrast to previous observations, we did not observe T17R at the cell membrane. This apparent discrepancy might be owing to greater sensitivity of the surface biotinylation assay employed by Evans et al. when compared with our immunofluorescent detection (12). Our findings on the murine disease-causing R92C mutation seem to contradict previously published observations that the R 89 -R-P-R-C site is shielded from furin, and that the mature enzyme is generated by cleavage at the R 108 -R-K-R-Y motif (13). Rather, R92C seems to abolish prodomain cleavage. Accordingly, pro-MMP2 activation and gelatinase activity were abrogated and the protein's ability to support migration impaired.
While our observations on T17R, R92C and R111H are consistent with the prodomain's known role in MMP14 trafficking and activity, the effects of S466P are less easily explained. As expected from a damaging missense mutation in the fourth blade of the Hx domain, it had a pronounced effect on protein processing and trafficking. Yet, the mutant protein showed only reduced activity in gelatin zymography and almost normal activity in the gelatin invadopodia assay. The mutation also did not affect MMP14's ability to support cellular migration. The most straightforward explanation seems to be that there is sufficient protein still present at the cell surface for the cells to retain near-normal activity in our assays. This is consistent with previously published data showing some membrane localization of MMP14 after its Hx domain had been deleted or swapped out for that of MMP17 (38). We do not yet know how to reconcile the severe Cartoon phenotype, which so strongly resembles that of Sabe, or a full Mmp14 KO, with near normal activity of the S466P mutant. As a possible explanation, we suggest that this mutation might affect functionality of other cell types, such as macrophages and osteoblasts, to a greater extent than that of fibroblasts.
To further assess the effects of loss of MMP14's catalytic function on the organismal level, we generated a zebrafish mmp14a/b KO model. Our in vivo results support an important role for MMP14 in collagen remodeling during ossification. Similar to existing mouse models, the mmp14a/b KO fish recapitulates essential aspects of the WS skeletal phenotype including a short stature, craniofacial malformations, reduced bone density and abnormal spinal curvature (1,16,18,20,(31)(32)(33). Although the humans, mice and zebrafish all have shortened life spans, the cause of death differs between species. Where WS patients reportedly die from heart failure, the mice succumb to wasting, possibly resulting from malnutrition owing to the facial abnormalities (17,18,(31)(32)(33). Although feeding problems could contribute to the death of mmp14a/b KO fish, spinal cord impingement, not observed in humans and mice, seems to be the main cause of death.
In contrast to previous morpholino studies in zebrafish, mmp14a/b KO larvae developed normal craniofacial cartilage elements (25,27). Subsequent mineralization of the larval skeleton during metamorphosis proceeded normally as well, and the first subtle differences in skull shape only became apparent at juvenile age. The gradually worsening phenotype observed in humans, mice and zebrafish further suggests that loss of MMP14 disrupts later stages of skeletal remodeling. Consistent with our in vitro data, these observations argue against a major role for MMP14 supporting cellular migratory and invasive behavior in vivo. Although the presence and effect of maternal transcripts in our mmp14a/b KO fish cannot be ruled out, the normal skeletal patterning in WS patients and Mmp14 KO mice supports this notion.
Apart from species-specific differences in bone morphology and ossification mechanisms, the skeletal abnormalities in mmp14a/b KO fish are comparable to those observed in Mmp14 KO mice. Similar to the mice, both endochondral and intramembranous ossification is impaired in mmp14a/b KO fish (33). During these types of ossification, bone forms within, respectively in close association with, a cartilage template (39,40). It was shown in Mmp14 KO mice that these cartilage templates are not properly replaced, respectively removed, owing to impaired MMP14-dependent collagen remodeling (32,33,41). This is reflected in the mmp14a/b KO fish, where cartilage cores of skull bones are relatively large, disorganized and have altered collagen content, with bone matrix being relative sparse. The cellfilled lacunae in mmp14a/b KO fish calvariae are reminiscent of the excessive osteoclastic bone resorption in Mmp14 KO mouse parietal bone (32)(33)(34)40,41). Impairment of additional processes that depend on MMP14's catalytic activity, including stimulating osteoblast differentiation and activity while having an opposite effect on osteoclasts, might contribute to the mmp14a/b KO skeletal phenotype (33,(41)(42)(43)(44). Further studies are needed to assess to which extent these processes underlie the skeletal abnormalities in mmp14a/b KO fish.
Taken together, our in vitro and in vivo studies highlight the significance of impaired proteolytic MMP14 activity underlying the WS skeletal phenotype. This observation has therapeutic implications. Attempts have been made in the past to treat WS and related disorders with bisphosphonates, with limited success (45,46). Bisphosphonates exert their therapeutic effect mostly by inhibiting osteoclasts (47). Whereas our observations as well as previously published work do support a role for osteoclasts in the WS bone phenotype, the major problem seems to be defective collagen remodeling. Thus, we would propose that any treatment of WS should seek to address this fundamental issue. Our zebrafish model could be used to develop and test such treatments, which might also be able to address osteoporosis and other conditions with decreased bone density.

Materials and Methods
Cell culture  Table S1). Likewise, the Sabe R92C mutant vector was created by mutating Codon 92 from CGA into TGC with fully overlapping primers. The 3HA-EGFP control vector was generated by deletion of the MMP14 coding sequence from the WT vector by inverse-PCR with fully overlapping primers binding to 24 bp of each linker sequence. Four expression vectors encoding either WT or mutant mmp14a or mmp14b with a C-terminal HA tag were generated. cDNA was obtained from 2-month-old WT and mmp14a D/D ; mmp14b D/D fish, respectively, as described below and, respectively, WT or mutant mmp14a/b coding sequence PCR amplified with the primers listed in Supplementary Material, Table S2. The forward primers started with zebrafish-optimized kozak sequence 5 0 -CCAACC-3 0 , the reverse primer ended with the HA coding sequence. The amplicon was gel-extracted and cloned into the pCR2.1 TOPO TA vector (Thermo; 450641) according to the manufacturer's protocol. The kozak-mmp14a/b-HA insert was subsequently cloned into the pCS2þ backbone.

Western blotting
For endogenous MMP14 and analysis of double-tagged MMP14, cells were seeded at a density of 100 000 cells per well of a sixwell plate. Cells were either transfected with pQCXIB vectors as described above or left non-transfected. Cells were harvested using trypsin-EDTA and whole cell lysate (WCL) was obtained using NP-40 lysis buffer (150 mM NaCl, 1% NP-40, 250 mM Tris pH 7.3) supplemented with protease and phosphatase inhibitors (Roche). Protein concentration was determined by Bradford protein assay (Sigma). An appropriate volume of Laemmli sample buffer was added to equal amounts of total protein lysate followed by boiling for 5 min. Samples were subjected to SDS-PAGE electrophoresis before transfer to an Amersham Hybond TM -PVDF membrane (GE Healthcare Life Sciences, Buckinghamshire, UK). Membranes were blocked in 5% (w/v) milk (Marvel dried skimmed milk) in 0.1% (v/v) Tween-20 (Melford; P1362) in TBS 1Â (Sigma; T5912-1L). After overnight incubation at 4 C in primary antibody (see Supplementary Material, Table S3) appropriately diluted in 0.5% BSA in TBS block buffer, membranes were incubated with secondary IRDye V R 680RD donkey anti-rabbit (LI-COR Biosciences, Lincoln, Nebraska, USA; 926-68073) and Alexa fluor 680 donkey antimouse (Thermo; A10038) antibody appropriately diluted in block buffer. Blots were imaged on the Odyssey V R Fc Imaging System (LI-COR).
For analysis of the effect of mmp14a/b frameshift mutations on putative protein products, 400 000 MRC-5V1 cells were seeded in 6 cm dishes. Cells were transfected 24 h post-seeding with 8 lg pCS2þ vector or left non-transfected using Invitrogen TM Lipofectamine 3000 (Thermo; L3000008) according to the manufacturer's protocol. Media was refreshed 6 h posttransfection. Cells were harvested 24 h post-transfection using trypsin-EDTA and WCL obtained using Triton V R X-100 lysis buffer [50 mM Tris-HCl pH 7.5, 150 mM NaCl, 1% (v/v) Triton V R X-100, 10% (v/v) glycerol, 0.1% (w/v) deoxycholate, 25 mM b-glycerophosphate] supplemented with protease inhibitors (Roche) and phosphatase inhibitors (10 mM NaF and 1 mM Na 3 VO 4 ). Protein concentration was determined by Bradford protein assay. Samples were mixed with Laemmli sample buffer and boiled as described above. Samples were subjected to SDS-PAGE electrophoresis before transfer to a PVDF membrane (Merck;

Zymography
Gelatin zymography was performed as previously described (49

Microcomputed tomography
Three-month-old fish were euthanized by hypothermic shock and fixed in 4% PFA and dehydrated through graded water into ethanol. Average sized fish were fixed per genotype, except for mmp14a D/D ; mmp14b D/D for which larger individuals were selected to better match the size of other genotypes. Microcomputed tomography images were acquired using an Inveon CT (Siemens AG, Berlin, Germany) at 55 kVp/110 mA. The exposure time per projection was 2500 ms and a binning factor of 2 was used, resulting in a reconstructed pixel size of 35 lm. Planar images were acquired from 181 projections over 360 of rotation in step-and-shoot mode. The images were reconstructed using a Feldkamp cone-beam algorithm. Threedimensional renders of the skeleton were made with AMIRA software (FEI, Mé rignac Cedex, France) with constant window settings. Images were exported as TIFF files and extracorporeal skeletal elements of previously imaged fish were manually removed with Pixelmator software. Raw data were viewed with AMIDE-bin 1.0.5 software (Andreas Loening), individual virtual sections exported as TIFF files and angles measured in Fiji software. Differences in mean bone density and angle of kyphosis between genotypes were tested for significance by two-sided Student's t-test.

Histology
Three-month-old fish were euthanized and fixed as described above. After trimming, the fish were placed into cassettes and processed with Sakura VIP Tissue Processor. Fish were dehydrated through graded ethanol into xylene (Chemtech Trading) before paraffin (Leica; 39601006) infiltration. After processing, tissues were embedded into paraffin blocks and sectioned midsagittally with a rotary microtome into 5 lm thick sections. The slides with sections were dried and placed into an incubator (60 C, 15 min). The sections were deparaffinized and rehydrated through graded ethanol into water. Rehydrated sections were subjected to hematoxylin solution (Richard-Allan Scientific TM ; 7231), Bluing (Richard-Allan ScientificTM; 7301), Clarifier (Richard-Allan Scientific TM ; 7442) and eosin-phloxine B Solution (AMPL). A separate batch of 5 lm thick sections of all the samples were stained with Weigert's hematoxylin, followed by Picrosirius red 0.1% (w/v) staining and rinsing with two changes of acidified (0.5%) water. Stained sections were dehydrated through graded ethanol into xylene and a cover slip added. Slides were imaged with an automated slide scanner equipped with Zeiss AxioImager Z.

Supplementary Material
Supplementary Material is available at HMG online.