Metalloprotease inhibitor TIMP proteins control FGF-2 bioavailability and regulate skeletal growth

Saw et al. show via the combinatorial deletion of Timp family members in mice that metalloprotease regulation of FGF-2 is a crucial event in the chondrocyte maturation program, underlying the growth plate development and bone elongation responsible for attaining proper body stature.


Introduction
Metalloproteases are present across all kingdoms of living organisms and have expanded widely during eukaryotic evolution, comprising the largest class of protease genes in humans (Gomis-Rüth, 2003;Quesada et al., 2009). Tissue inhibitors of metalloproteases (TIMPs) are well known to control the turnover of matrix proteins in connective tissue (Sterchi et al., 2008). The TIMP gene family has been diversified with phylogenic development. Flies have a single TIMP gene with conserved metalloprotease inhibitor function, loss of which causes blistered wings and death by digestive tract lysis (Godenschwege et al., 2000;Page-McCaw et al., 2003). The five putative TIMPs in zebrafish have not been well studied (Wyatt et al., 2009). Mammals possess four TIMPs, which inhibit most metzincins, a subfamily of 89 secreted and cell surface-bound metalloproteases (Sterchi et al., 2008). Beyond matrix turnover, the TIMP-metalloprotease axis controls major signaling pathways through ectodomain shedding, and deregulation of this axis has invariably been seen in human cancers and diseases (Murphy et al., 2008;Aiken and Khokha, 2010;Kessenbrock et al., 2010;Jackson et al., 2017). Overlapping enzyme inhibitory specificity among TIMP proteins undermines the ability of single TIMP knockout mice to reveal their critical biology. Therefore, concurrent deletion of the entire TIMP gene family is a prerequisite for understanding the function of this gene family in mammals.
Postnatal bone growth employs isometric scaling (Pietak et al., 2013;Stern et al., 2015), and skeletal proportionality varies in primates, where it has evolved to provide optimal biomechanical efficiency for species-specific adaptation to their habitat. In humans, the combined length of femur plus tibia is ∼50% of total stature, which is biomechanically efficient for a striding bipedal gait (Bogin and Varela-Silva, 2010). Body stature relies on height attained during the growth of long bones. Endochondral ossification is the sole process of bone elongation and is accomplished through the replacement of cartilage by bone matrix underneath the growth plate. Mesenchymal cells in the growth plate differentiate into chondrocytes, sequentially generating proliferating and hypertrophic zones, with the hypertrophic chondrocytes ultimately undergoing apoptosis to leave behind a cartilage matrix for subsequent ossification. Long bones continue to grow until growth plate closure with sexual maturity postpuberty. This defined chondrocyte program of proliferation, differentiation, and maturation is tightly regulated by local factors. The morphogen Indian hedgehog (IHH) and its interplay with parathyroid hormone-related protein are important to balance proliferating and hypertrophic chondrocytes (Kobayashi et al., 2002;Mizuhashi et al., 2018;Newton et al., 2019). Signaling from fibroblast growth factor receptor 3 (FGFR3) antagonizes Ihh expression in chondrocytes and leads to alteration in growth plate activity (Minina et al., 2002;Tang et al., 2016). Extracellular matrix proteins are also important regulators of growth plate activity (Aiken and Khokha, 2010). Aggrecan is a signatory proteoglycan of chondrocyte matrix and mutations in this molecule underlie various chondrodysplasias in humans and other animals. Metalloproteases are essential enzymes for aggrecan turnover, and endogenous TIMPs keep the protease activity in check. TIMPs are expressed in most organs including bone and cartilage, and their deregulation has been reported in cartilage and bone pathologies (Gendron et al., 2003;Nuttall et al., 2004;Aiken and Khokha, 2010;Chen et al., 2019).
Here we characterize cartilage in genetically engineered mouse models (GEMMs) lacking all four TIMP genes. Complete TIMP deficiency produces severe skeletal defects due to an aberrant chondrocyte maturation program during the major period of bone growth spanning birth to puberty. We identify a TIMP-dependent regulation of FGF-2 and IHH signaling in the growth plate. Using compound GEMMs, we rescue the bone defects by incorporation of aggrecan knock-in mutations resistant to either matrix metalloproteinase (MMP) or ADAMTS cleavage, illustrating the requirement of TIMP-regulated metalloprotease activity for correct bone proportionality. The phenotypic manifestations in the TIMPless mice provide fundamental insights into the molecular drivers of mammalian skeletal growth and stature, as well as expose the functional redundancy in the metalloprotease inhibitor gene family.
Adult QT mice have short stature compared with WT or QT3 +/− littermates and exhibit a vivid skeletal phenotype (Fig. 2 A). Micro-CT imaging of the thoracic girdle exposed gross defects in bone architecture: a curved spine and bulging ribs at the costovertebral/costotransverse joints. Additionally, the sternum was shorter and displayed bright bands of calcification at the margins (Figs. 2 B and S1), with a small proportion (∼10%) of mice presenting pectus excavatum (Fig. 2 C), a sunken sternum, which is the most common congenital defect of the chest wall in humans (Tocchioni et al., 2013). Long bone defects were also prominent in adult QT mice, with their joints showing abnormal morphology and disappearance of the epiphysis such that only its remnants are visible by 7 wk of age (Fig. 2,C and D). Axial and appendicular bone segments, specifically the length of sternum, thoracic spine, femur, and tibia in WT, QT3 +/− , and QT cohorts, were measured (Fig. 2 E). While both femur and tibia were smaller, a decrease in the femur/tibia ratio indicated a more extensively shortened femur in the QT mouse, whereas the ratio of sternum/thoracic spine were similar to controls (Fig. 2, F and G). These data demonstrate a role for the TIMP family in normal postnatal development of mammalian skeleton and isometric scaling of long bones.
Growth plate closure in quadruple TIMP-deficient mice The epiphyseal growth plate is the developmental region responsible for bone elongation. The growth plate undergoes progressive narrowing and closure concomitant with pubertal growth in humans, although it does not fuse in mice (Börjesson et al., 2012;Staines et al., 2018). We examined the appendicular and axial skeleton ( Fig. 3 and 4). We observed cartilage overabundance as well as bone bridges across the ossification centers of long bones, resulting in aberrant growth plate closure in quadruple TIMP-deficient mice (Fig. 3 A). Chondrocyte columnar organization in this region was completely disrupted as early as 4 wk, with only small bone-encased islands of cartilage remaining at 10 wk ( Fig. 3 B), like those seen in a fusing human pubertal growth plate. QT3 +/− mice showed a less severe phenotype ( Fig. 3 B), although the growth plates of axial bones, sternum, and spine were also abnormal. Histology revealed an excess of sternal cartilage, which extended beyond the bone margins in 4-wk-old QT mice (Fig. 4,A and B). Similar changes occurred in the cervical and thoracic vertebrae (Fig. 4,C and D). Histomorphometry confirmed cartilage hyperplasia (1.7-fold) at the expense of highly compacted bone marrow cavities in both sternum and spine (Fig. 4,E and F). Overall, the absence of TIMP disrupted the postnatal chondrocyte program in both the axial and appendicular skeleton.
We then tracked skeletal development from the late embryonic to the prepubertal stage. TIMP-deficient groups showed a small reduction in rib cage and tibial length at embryonic day 17.5 (E17.5; Fig. S2, A-C), although long bone growth plates were indistinguishable between these groups at postnatal days 2 and 7 (Fig. S3, A and B). By postnatal day 16, we observed a marked reduction in proliferative and hypertrophic chondrocyte zones (Fig. 3 C) and delayed development of secondary ossification centers in QT, pointing to a defective chondrocyte maturation program. Chronological analysis of long bones and their growth plates demonstrates that the major deformities in TIMPdeficient bones arise postnatally, although the possibility of a subtle embryonic phenotype remains. A similar change in growth plate has previously been reported in mice that harbor an FGFR3 activating mutation, with concomitant lowering of Ihh expression (Naski et al., 1998;Wang et al., 1999).

TIMP1 is critical for sustaining normal growth plate
We investigated which of the four TIMPs are important for normal growth plate structure by evaluating the length of long bones and growth plate closure phenotypes of individual TIMP knockouts as well as several combinatorial knockouts (  (Fig. 3 D). Interestingly, the only two T2 −/− T3 −/− mice that serendipitously survived in utero lethality exhibited only a minor distortion in the growth plate without bone bridge development (Fig. 3 D) and no alteration in bone length (Fig. 3 E). Partial or complete growth plate aberration in specific genotypes is summarized in Fig. 3 F. These data collectively show that TIMP1 along with TIMP3 is crucial for growth plate integrity.

Differential rescue of long bone proportionality by MMP-and ADAMTS-resistant aggrecans
Aggrecan is a core proteoglycan of the cartilage matrix and is critical for its load-bearing function. Mutations in aggrecan result in severely disrupted growth plates and skeletal deformities in humans, chickens, and mice (Kimata et al., 1981;Li et al., 1993;Warman et al., 2011). Aggrecan loss in articular cartilage is predominantly a proteolytic process mediated by MMP and ADAMTS proteases and has been extensively studied in arthritis (Roughley and Mort, 2014). Multiple metalloprotease cleavage sites are located in the interglobular domain (IGD) and the chondroitin sulfate-rich region of aggrecan ( Fig. 5 A); proteolysis in the IGD releases the entire glycosaminoglycan-containing portion, with concomitant loss of mechanical properties, whereas proteolysis in the chondroitin sulfate rich region is part of normal aging and does not appear to affect biomechanics (Ilic et al., 1998). We used a genetic approach to test the contribution of aggrecan cleavage to the skeletal abnormalities seen in TIMPdeficient mice by crossing in Chloe or Jaffa knock-in mutations that block either MMP (Chloe) or ADAMTS (Jaffa) cleavage sites in the IGD of aggrecan (Little et al., 2005(Little et al., , 2007, as modeled in Fig. 5 A. Specifically, amino acid sequence mutation 342 FFG to 342 GRT eliminates the MMP-cleavage site at N 341 /F 342 in Chloe; and 374 ALGS to 374 NVYS eliminates the ADAMTS cleavage site at E 373 /A 374 in Jaffa. Chloe and Jaffa mice were crossed with QT3 +/− mice to produce C-QT3 +/− and J-QT3 +/− cohorts. We found that incorporation of the Chloe mutation rescued the shortening of both the tibia and femur of QT3 +/− mice, demonstrating that TIMP regulation of MMP activity at this site is necessary for normal postnatal growth plate development. Surprisingly, the Jaffa mutation rescued the tibia but not the femur length (Fig. 5, B and C), showing a differential requirement for regulation of ADAMTS activity in these two long bones. Consistent with the report that the distal femur and proximal tibia growth plates are the main contributors to bone elongation (Serrat et al., 2007), C-QT3 +/− growth plates displayed normal histology at both these sites, while the J-QT3 +/− femur growth (C) Embryos were genotyped and observed versus expected ratios compared using the χ 2 test. Gray shading distinguishes postnatal time points. Enumeration of embryos (E13.5, E15.7, and E17.5) and born offspring showed that most T2 −/− T3 −/− die by E17.5, in contrast to QT, which were observed at >75% of the expected number. Therefore, the loss of TIMP2 and TIMP3 is detrimental at late gestation, while further additive loss of both TIMP1 and TIMP4 rescues lethality. It is conceivable that networks causing lethality in the T2 −/− T3 −/− scenario are either bypassed or opposed by the new milieu generated by complete TIMP loss. Numbers examined at E13.

Exacerbation of sternum defect by MMP-resistant aggrecan
We also evaluated the outcome of blocking MMP-and ADAMTSmediated aggrecan cleavage in axial bones, in the context of TIMP loss. The length of the thoracic vertebral column and sternum were measured in all cohorts (Fig. 5, B and C). Axial bone length did not further change with the incorporation of ADAMTS-resistant aggrecan in the J-QT3 +/− cohort, whereas MMP-resistant aggrecan (C-QT3 +/− ) had the unexpected and dramatic effect of further shortening the sternum. The exceptionally short sternum observed in C-QT3 +/− mice, compared with J-QT3 +/− and QT3 +/− mice, demonstrates the requirement for MMP-mediated cleavage of aggrecan for normal sternum growth. Histologically, C-QT3 +/− mice also displayed thick fibrocartilaginous pads between the sternebrae (Fig. S4). The uneven rescue of axial bone segment shortening with MMPresistant aggrecan is highlighted by the sternum:spine ratio ( Fig. 5 C). These data demonstrate that postnatal growth of skeletal segments requires natural metalloprotease inhibitor regulation of MMP and ADAMTS processing of aggrecan.

Expression profiling reveals IHH and FGF-2 deregulation in chondrocytes lacking TIMPs
To seek the mechanism underlying the disrupted chondrocyte maturation program, we macrodissected sternal cartilage excluding the xiphoid process for expression profiling (  Table S1).
Functional enrichment analysis using g:Profiler pointed to the smoothened and FGF pathways along with alterations in metabolism, embryonic development, inflammatory response program, and others ( Fig. 6 D). Additionally, KEGG (Kyoto Encyclopedia of Genes and Genomes) pathway enrichment using Enrichr also pointed to IHH as the top down-regulated pathway in TIMP-deficient cartilage ( Fig. S5 A). IHH is a major regulator of chondrocyte maturation (St-Jacques et al., 1999). It binds to the receptor Ptch1, relieving inhibition of Smo to initiate signaling, an interaction promoted by the coreceptors Cdo, Boc, and Gas1 and hindered by Gpc3 and Hhip (Briscoe and Thérond, 2013). Quantitative PCR verified the down-regulation of these IHH pathway components in QT cartilage (Fig. 6, E and G), whereby the extent of decrease in the canonical IHH target genes, Gli1, Hhip, and Ptch1, correlated with phenotype severity. The enrichment map also pointed to FGF-2 signaling up-regulation in QT samples ( Fig. 6 D). Chondrocytes express multiple genes in response to FGF-2 stimulation such as Inhba, Mmp3, Mmp19, Pdpn, Tnfaip6, Tnfsf11, and Timp1 (Chong et al., 2013); several of these genes were increased in TIMP-deficient sternal cartilage ( Fig. 6 F). We then assessed the proximal tibia head to determine whether TIMP-deficient long bones had mechanistic alterations similar to the sternal cartilage. Ihh down-regulation and Fgf2 up-regulation was confirmed in the growth plate cartilage containing region of the QT3 +/− long bones (Fig. 7, A-D). Specifically, Mmp3 and Mmp19 were up-regulated and Ihh and Acan were down-regulated as the FGF-2 target genes. Furthermore, several markers of chondrocyte maturation (resting, collagen II; proliferating, aggrecan; hypertrophic, IHH and collagen X) were found to be altered, consistent with the reduction of hypertrophic chondrocyte zone in TIMP-deficient long bones (Figs. 3 C and 7 D). Increased FGF-2 signaling prompted us to look for additional skeletal changes in the quadruple TIMP-deficient mice. We noted shortening of the nasal and frontal bones, kyphosis, and misaligned closure of the upper and lower incisors (Fig. 7, E and F), phenotypes similar to those reported in mice harboring an FGFR3 activating mutation (Wang et al., 1999). Collectively, TIMPless chondrocytes have both decreased IHH and increased FGF-2 growth factor signaling in both sternum and long bone growth plates.

Saw et al.
Journal of Cell Biology ( 3 H-217; high-affinity binding to the active site of most MMPs) showed a heightened signal in the skeleton of QT mice (Fig. S5 B). Likewise, ex vivo imaging of organs from mice injected with MMPSense750 (fluorescent pan-MMP beacon upon cleavage) highlighted skeleton, with an intense signal in the growth plate. The long bones, sternum, and spine of TIMP-deficient mice displayed elevated MMP activity (Figs. 8 A and S5 C).
To address the causal relationship between increased metalloprotease activity and down-regulation of the IHH pathway in TIMP-deficient cartilage, we set up primary chondrocyte cultures derived from sternum and ribs of neonatal pups (Fig. 8 B). Alcian blue staining indicated the production of matrix component by chondrocytes in vitro, attesting to their functional capacity ( Fig. S5 D), and we verified that TIMPs are expressed in WT chondrocytes (Fig. S5 E). These cultures were treated with two metalloprotease inhibitors, either a broad-spectrum inhibitor (BB94) or one with specificity for adamalysins (TAPI-1; Fig. 8, C-F). We found that WT chondrocytes responded to BB94 by significantly elevating Ihh expression (Fig. 8 C) and by partially up-regulating Hhip expression (Fig. 8 E), whereas the IHH pathway inhibitor cyclopamine (a control) blocked the induction of Hhip (Fig. 8 E). TIMP-deficient chondrocytes responded to both metalloprotease inhibitors with a far greater induction of Ihh and Hhip expression (Fig. 8, D and F). These data indicate that metalloproteases normally act to negatively regulate IHH signaling in chondrocytes, and blocking the hypertrophic zone (HZ) and proliferating zone (PZ). RZ, resting zone. (D) H&E-stained knee joints of different TIMP knockout combinations. Upper panels display knee joints of individual TIMP knockouts. Middle panels depict different levels of growth plate closure (*) of multiple-TIMP knockouts. Safranin O-stained tibiae of 8-wk-old WT and multiple TIMP knockouts (lower panels) exhibit growth plate closure (*; 4×/0.5-NA objective). (E) Faxitron x-ray image of femur, demonstrating the lengths of WT and different multiple TIMP knockouts (8-wk-old). (F) Summary of TIMP knockout combinations that result in growth plate closure or have an intact growth plate. FGF-2 signaling in chondrocytes is known to suppresses Ihh expression, and both are implicated in various bone pathologies (Su et al., 2014). For instance, transgenic FGF-2 overexpression results in dwarfism in mice (Coffin et al., 1995), and activating FGFR3 mutations lead to achondroplasia and shortened long bones with disorganized chondrocyte columns in growth plate (Chen et al., 1999;Lee et al., 2017). We therefore probed the effect of metalloprotease inhibitors on Ihh expression in the presence or absence of recombinant FGF-2 in chondrocyte cultures (Fig. 8 G). FGF-2 treatment lowered Ihh expression whereas pan-FGFR inhibitor AZD4547 induced Ihh indicating the presence of endogenous FGF-2 signaling in this culture system. Importantly, combining BB94 with FGF-2 did not further alter Ihh expression suggesting that metalloprotease activity does not influence the bioactivity of exogenously added FGF-2.

Metalloprotease-resistant aggrecans alter FGF-2 localization and release
Perlecan is an FGF-2-binding heparan sulfate proteoglycan that localizes to the cell surface, and multiple MMPs are capable of releasing sequestered FGF-2 by cleaving perlecan (Whitelock et al., 1996;Tholozan et al., 2007). We reasoned that elevated FGF-2 bioavailability due to excess metalloprotease activity may be responsible for growth plate closure. FGF-2 and perlecan immunofluorescence staining to mark chondrocyte pericellular space in situ revealed colocalization of perlecan and FGF-2 on the chondrocyte cell surface in the growth plate of WT femoral and tibial distal heads (Fig. 9 A). Both chondrocyte organization and perlecan/FGF-2 colocalization were lost in the QT and QT3 +/− animals. The Jaffa and Chloe mutations restored chondrocyte organization and perlecan/FGF-2 colocalization in the tibia (Jaffa and Chloe) and femur (only Chloe; Fig. 9 A). Surprisingly, the Chloe mutation resulted in higher chondrocyte number and higher Perlecan/FGF-2 costaining, changes that were also reflected in the aggrecan staining pattern (Fig. 9 B). The differential rescue of chondrocyte organization by the Chloe and Jaffa mutations corresponded to the recovery in bone lengths in those mice (Fig. 5 C and Fig. 10, A and B).
The release of perlecan-sequestered FGF-2 from the pericellular space of chondrocytes is thought to be important for its bioactivity (Jonca et al., 1997;Gavrilovic, 2009). To determine whether FGF-2 release is affected in the TIMP-deficient or Chloe/Jaffa crosses, we set up the femoral distal head explant culture system (Stanton et al., 2011) and used ELISA to measure FGF-2 in the supernatants. The concentration of FGF-2 in the WT supernatant was lower in the presence of the metalloprotease inhibitor BB94 (Fig. 10 C). Although variable, the level of FGF-2 was higher in the supernatant of QT3 +/− than WT femur head explant cultures, consistent with the higher metalloprotease activity playing a role in FGF-2 release from the cell surface. J-QT3 +/− mice released far less FGF-2, consistent with there being less FGF-2 by immunofluorescence staining (Fig. 9  A), whereas Chloe mice released more FGF-2 into the media, again consistent with their higher FGF-2 staining. We noted that BB94 did not reduce the FGF-2 levels in the Chloe explant, possibly due to the presence of MMP-resistant aggrecan in this setting. Collectively, these data demonstrate that a TIMP loss promotes FGF-2 signaling, which in turn suppresses the IHH pathway, triggering aberrant cartilage homeostasis and growth plate integrity, ultimately resulting in abnormal postnatal bone development and body posture, as summarized in Fig. 10 D.

Discussion
Whole-body removal of the entire TIMP gene family has directly demonstrated their redundancy and identified their indispensable role in skeletal growth, stature, and lifespan. Adult quadruple TIMP-deficient mice have widespread chondrodysplasia throughout the appendicular and axial skeleton and epiphyseal growth plate closure of long bones, a phenomenon underlying growth retardation. Pectus excavatum is an extreme manifestation of the TIMP-deficient sternum and, to our knowledge, reported in only one other mouse GEMM model, of Ptpn11 loss in the mesenchymal lineage (Lapinski et al., 2013). Pectus deformities affect up to 0.3% of the human population in association with a range of connective tissue disorders including Marfan syndrome and related disorders (Tocchioni et al., 2013). Our combinatorial TIMP knockouts also reveal the exclusive requirement of TIMP2 plus TIMP3 for embryonic survival, although their deletions do not impose remarkable skeletal deformities in the rare surviving animal. Further deletion of Timp1 and Timp4 greatly improves embryonic survival, although these mice are runted and have a severely shortened lifespan. Surprisingly, a single Timp3 allele is sufficient for a near-normal lifespan, as shown by QT3 +/− mice, yet these mice display pervasive bone abnormalities. Overall, TIMP-deficient GEMMs offer new models to study pectus excavatum, bone elongation, and chondrocyte biology. Furthermore, these combinatorial GEMMs help reveal the redundancy of TIMP activity in select systems such as the chondrocytes in growth plate cartilage and the mammary stem and progenitor pools in the mammary epithelium (Jackson et al., 2015). (G) Quantitative PCR confirming IHH pathway down-regulation (n = 6/genotype). Expression of genes cdo, Boc, Ihh, Hhip, Ptch1, Gli1, Gli2, and Sfrp5. Expression of Ihh, Hhip, Ptch1, and Gli1 were further subdivided by the severity of QT sternal phenotype (n = 3/group). Mean values of each dataset are plotted in graphs with error bars representing SEM. Data were compared using unpaired t test: *, P < 0.05; **, P < 0.01.

Saw et al. Journal of Cell Biology
The human growth plate undergoes progressive narrowing, ultimately disappearing with the cessation of growth upon sexual maturity, although the complex mechanisms governing this process are not completely understood (Staines et al., 2018). Recent studies emphasized a role of matrix remodeling in this process as active matrix fragments are shed, and endochondral bone formation is also influenced by epigenetic regulation of several metalloproteases (Carpio et al., 2016;Coghlan et al., 2017). Several MMPs (e.g., MMP-2, -3, -8, -9, -10, -12, and -13) and ADAMTSs (e.g., ADAMTS1, 4, and 5) enzymes are present in the growth plate, and null mutations in some MMPs (MMP-9, -13, and -14) affect growth plate morphology (Gack et al., 1995;Mattot et al., 1995;Hou et al., 2004;Little et al., 2005). Growth plates also express TIMPs in chondrocyte zones: Timp1 in all zones, Timp2 in proliferating and hypertrophic zones, and Timp3 in the hypertrophic zone (Takahashi et al., 2005). Axial and appendicular QT growth plates had intense metalloprotease activity. The unleashed metalloproteases resulted in closure of both femoral and tibial growth plates through bone bridges, although a lower ratio of femur:tibia length indicated a greater sensitivity of the femur growth plate to metalloprotease degradation.

Saw et al.
Journal of Cell Biology ADAMTS inhibitor (Murphy, 2011). Interestingly, pronounced bone phenotypes in TIMPless and combinatorial TIMP-deficient mice also point to the redundancy among TIMP functions. Appearance of weak or no phenotypes in single gene knockouts of most MMPs have often raised speculation about functional redundancy among these matrix remodeling enzymes, which has been confirmed through the complete deletion of TIMPs. MMP and ADAMTS members degrade a wide range of cartilage matrix components including aggrecan and their cleaved neopeptides can be detected in humans and rodents suggesting an active role of these metalloproteases in aggrecan turnover (Lark et al., 1997;van Meurs et al., 1999;Makihira et al., 2003). As reported, a baseline level of aggrecan turnover is mediated by MMPs, whereas ADAMTSs are known to initiate cartilage damage in arthritis. Previously, aggrecan turnover by MMPs was thought to be a dispensable mechanism in growth plate biology, as it was studied in WT background where growth plate disruption was not apparent (Little et al., 2005). When we compared aggrecan degradation by MMPs versus ADAMTSs in mice lacking most TIMPs, i.e., QT3 +/− , Jaffa (ADAMTS resistant) and Chloe (MMP resistant) aggrecan knock-in mutations produced distinct outcomes on cartilage development in appendicular and axial skeleton. MMP-resistant aggrecan rescued chondrocyte/FGF2 organization, growth plate integrity, and bone length in both femur and tibia of C-QT3 +/− mice, while only the tibial cartilage was rescued in the ADAMTS-resistant J-QT3 +/− mice. This demonstrates that TIMP regulation of MMP activity is necessary for normal femur and tibia growth plate, and that the femoral growth plate is clearly more susceptible to excess MMP activity than the tibial growth plate. The lack of rescue in the femur of J-QT3 +/− mice argues that at least some ADAMTS activity is necessary for normal femoral growth plate development. Similarly, exacerbation of sternum shortening in the C-QT3 +/− mouse points to MMP cleavage at aggrecan N 341 /F 342 as necessary for normal sternum chondrocyte development. Overall, these two mutations in aggrecan affect the axial skeleton differently, since MMP resistance exacerbates rather than rescues the QT3 +/− shortened sternum, and ADAMTS resistance had no effect. Therefore, simultaneous regulation of two major metalloprotease classes, MMP and ADAMTS, by the TIMP gene family is crucial to processes that determine optimal bone growth and isometry in the mouse.

Saw et al.
Journal of Cell Biology reduced bone mass (Xiao et al., 2004;Sobue et al., 2005). Mice with activating FGFR3 mutations have a smaller body size, kinky tails, dorsal kyphosis, a dome-shaped skull, shorter long bones encompassing disorganized chondrocyte columns in growth plates, shortening of the nasal and frontal bones of the skull, misaligned closure of the upper/lower incisors, and delayed formation of secondary ossification centers (Wang et al., 1999), traits seen in the quadruple TIMP knockouts. Further, the epiphyseal growth plates of mice with FGFR3 mutations have smaller proliferating and hypertrophic zones (Wang et al., 1999;Lee et al., 2017) similar to TIMPless mice. Humans with FGFR3 gain-of-function point mutations also present a variety of skeletal dysplasias (Lee et al., 2017). In a pharmacological model, dosing with an IHH pathway inhibitor led to a shortened femur and premature growth plate closure (Kimura et al., 2008), similar to our TIMP-deficient mice. Collectively, the strong similarities in phenotypes among these GEMMs strengthen FGF-2 and IHH as core cartilage development pathways dependent on the natural metalloprotease inhibitor activity. Given the complexity of the skeletal system, which also heavily relies on balanced osteoblast and osteoclast activity, we have recently reported low bone mass in TIMPdeficient long bones due to higher Rankl activity in osteoblasts, which is also downstream of FGFR3 signaling (Su et al., 2010;Wen et al., 2016;Chen et al., 2019).
In summary, this comprehensive study of combinatorial TIMP-deficient GEMMs uses global expression profiling, biochemical and in situ analyses, and rationalized aggrecan-resistant knock-in GEMMs to uncover metalloprotease regulation of FGF-2 as a crucial event in the chondrocyte maturation program, underlying correct growth plate development and bone elongation responsible for attaining proper body stature.

Key resources
Key resources are listed in Table S2.

Mice
All mice used in this study are of pure C57BL/6 background. Mice were housed and cared for in accordance with the guidelines approved by the Canadian Council for Animal Care and the Animal Care Committee of the Princess Margaret Cancer Centre (Toronto, Canada). We used individual TIMP knockout mice that were previously generated using homologous recombination; T1 −/− mice have a stop codon within each reading frame of the third exon of Timp1 gene (Soloway et al., 1996); T2 −/− mice are devoid of the first exon of Timp2 and additional 59 genomic sequences (Wang et al., 2000); T3 −/− mice lack 6 kb from exons 2 and 3 of Timp3, including translation initiation sequence and Saw et al.
Journal of Cell Biology sequences encoding amino acids important for TIMP inhibitory activity (Leco et al., 2001); and T4 −/− mice have deletion of a 2.4-kb genomic fragment containing exons 1-3 of Timp4, including translation initiation codon (Koskivirta et al., 2010; Fig. 1 A). These mice were bred to produce different combinations of whole-body TIMP knockout GEMMs, and with Chloe and Jaffa knock-in mice (Little et al., 2005(Little et al., , 2007. In Jaffa knock-in mice, a mutation was inserted in exon 7 of aggrecan gene (Acan) to change the amino acid sequence from 374ALGS to 374NVYS, eliminating aggrecanase cleavage site 373E/374A, whereas in the Chloe knock-in strategy, exon 7 of Acan was mutated to change the amino acid sequence from 342FFG to 342GTR, disrupting the MMP cleavage site 341N/342F. These knock-in strains in C57BL/6 background were bred through several crosses of combinatorial TIMP-deficient mice to produce Chloe-QT3 +/− (C-QT3 +/− ) and Jaffa-QT3 +/− (J-QT3 +/− ) GEMMs.

Radiography
Formalin-fixed bones were imaged using a Faxitron MX-20 digital x-ray system with a 24kV, 4-s exposure time for 2D analysis. For whole-body micro-CT imaging, formalin-fixed mouse skeletons were placed in a GE Locus Ultra Micro-CT (GE Medical Systems) and subjected to a 16-s Anatomical Scan Protocol (total of 680 images) at 80 kV, 70 mA, using a 0.15-mm Cu Filter, to achieve ∼150-µm resolution. The same machine was also used for live-mouse imaging at acquisition parameters 80 kV, 50 mA; 16-s anatomical scan; 154-µm isotropic voxels (total of 680 slices) and 3D rendered using Siemens Inveon. For highresolution micro-CT imaging, fixed mouse legs were immobilized on 1.25% agarose. Specimens were scanned in 360°rotation using a Siemens Inveon Micro-CT high-resolution scanner (Siemens Medical Systems) with the x-ray source at 80 kVp and 0.5 μA. 3D micro-CT data were reconstructed at 13.5-µm resolution. Raw data processing was performed using ImageJ software (National Institutes of Health), and 3D isosurfaces were rendered using Microview software (GE Healthcare). Bone length was measured digitally with ImageJ.

Histomorphometry
To establish the relative proportion of tissues in the sternum and spine, H&E-stained sections were examined with a Merz eyepiece graticule (Merz and Schenk, 1970), and the cross-sectional areas of cartilage, bone matrix, and bone marrow were measured and expressed as percentages of total area examined.

Skeletal staining
The skin and soft tissues were removed from embryos or adult mice before fixation in 95% EtOH. Mice were stained for 2 d in 0.1 mg/ml alcian blue (Sigma-Aldrich) in an 80%:20% volumetric solution of EtOH:glacial acetic acid, and then rehydrated in sequential EtOH solutions of 70% (twice), 40%, and 15% and distilled H 2 O. Skeletons were cleared overnight in 1% KOH (Sigma-Aldrich) and stained for 2 d in 0.01 mg/ml alizarin red (Sigma-Aldrich) in 1% KOH. Stained specimens were photographed using an Infinity2 camera (Lumenera) and Olympus SZ2-1LST dissecting microscope.

Microarray
Flash-frozen sternums from 4-wk-old WT and QT mice were thawed in RNAlater-ice per the manufacturer's instructions (Ambion). Sternal cartilage pads were dissected and homogenized in 800 µl of Trizol (Invitrogen). RNA was extracted as described previously for cartilage (Ali and Alman, 2012) (Gentleman et al., 2004;Smyth, 2004) in R statistical environment (v 2.14.2); linear modeling was applied to identify genes altered in QT sternal cartilage relative to WT samples. Bayesian moderation of the standard error implemented was conducted on all model-based t tests (Smyth, 2004). In addition, multiple testing was corrected by a false discovery rate adjustment (Storey and Tibshirani, 2003). Significant differential expressed genes between WT and QT were identified based on a false discovery rate cutoff of <0.05. Pathway enrichment analysis was performed using the Enrichr gene list enrichment analysis tool (Chen et al., 2013). Further, gene ontology functional enrichment analysis was performed using g:Profiler, and an enrichment map was generated using cystoscope (v3.6.0) with a q value cutoff of 0.01. Enriched pathways were identified using AutoAnnotate (Kucera et al., 2016).

RT-PCR
Whole sternums and proximal tibia heads of mice were cleaned of connective tissue, snap frozen immediately following dissection, and kept at −80°C. Frozen bones were pulverized, and total RNA was extracted using TRIzol reagent (Invitrogen; 15596-026

Imaging
For radioactive imaging, a mixture of 1 H-217 and 3 H-217 was injected intravenously (150 µCi/ml PBS for total 0.7 mM; 10 mg/kg into 12-wk-old mice). Mice were anesthetized, and body temperature was maintained at 38°C for 1 h before sacrifice. Mice were immersed in dry ice/isopentane to prevent tissue redistribution, and whole-body sections (20 µm) were prepared at −20°C with a slicing microtome (Leica Microsystems). Sagittal sections were desiccated (24 h, RT) before radioimaging (β-imager TM 2000, Biospace Lab). For fluorescence imaging, 4-wk-old mice were injected with MMPSense750 Fast (intravenously, 0.08 nmol/g; PerkinElmer) and sacrificed 7 h later. Organs were extracted and imaged (Maestro system; Cambridge Research and Instrumentation). Multispectral image cubes containing multiple 10-nm bandwidth fluorescence emission signals were acquired at identical exposure times and spectrally unmixed using the fluorescence emission spectra from the injected compound.
Femoral distal head explant culture 3-wk-old mice were dissected to collect femurs, and after complete removal of muscles, the femoral distal head was carefully separated from the rest of long bone using forceps under sterile conditions. The femoral head was then transferred to a 48-well plate (as shown in Fig. 10 C) and cultured in 300 µl serum-free DMEM for 72 h (5% CO 2 , 37°C) with BB94 (10 µM) or vehicle control DMSO, as done elsewhere for femoral distal head explant culture (Stanton et al., 2011). The inhibitor was added every 24 h, and culture supernatant was collected for ELISA.
FGF-2 ELISA FGF-2 ELISA was performed using the commercially available kit EMFGF2 (Thermo Fisher Scientific) and following the manual. Culture supernatant and standard were added (100 µl) to the well of a precoated 96-well ELISA plate and incubated overnight at 4°C with gentle shaking. After washing with provided wash buffer, 100 µl of biotinylated antibody was added to the wells and incubated at RT for 1 h. Streptavidin-HRP antibody was added after washing and incubated for 45 min at RT. After washing, plates were developed using tetramethylbenzidine substrate and incubating at RT for 30 min. Absorbance was read at 450 nm using a POLARstar Omega microplate reader (BMG Labtech). FGF-2 concentration was estimated using the standard curve.

Statistical analysis
Prism software (GraphPad) was used for all analyses. Data distributions were assumed to be normal. Unpaired Student's t test, Wilcoxon matched-pair test, and one-way ANOVA tests were used for pairwise and multiple comparisons, respectively. Oneway ANOVA with Sidak's, Dunnett's, and Bonferroni's post hoc multiple comparison tests were performed for multiple comparison. The χ 2 test was used to determine whether offspring of a given genotype were observed in a Mendelian ratio. The χ 2 test for independence was used to compare mouse survival rates between genotypes. All graphs are plotted as mean with error bars representing SEM; *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001. Statistical analyses performed on microarray and mass spectrometry datasets are detailed in their respective sections.
Data and software availability Microarray data generated in this study was submitted to Gene Expression Omnibus under accession no. GSE60451.

Online supplemental material
In Fig. S1, rib head enlargement at both costovertebral and costotransverse joints of QT mice are shown. In Fig. S2, minor defects in axial and appendicular bone of QT embryos are displayed. In Fig. S3, growth plate organization and chondrocyte maturation zone are shown spanning days 2 to 28 in QT and WT, as well as tibia and femur lengths of individual TIMP knockouts. In Fig. S4, shortening of sternebrae and enlargement of cartilaginous joint of sternum are shown across several GEMMs. In Fig. S5, up-regulated and down-regulated pathways are summarized from microarray analyses of sternal cartilage, along with MMP activity in skeletal system and Timp gene expression in primary chondrocytes across experimental and control mouse cohorts. Table S1 lists genes significantly altered in TIMPless sternal cartilage, Table S2 lists key resources, and Table S3 lists primers used for SYBR Green analysis.