Abnormal Collagen Fibrils in Cartilage of Matrilin-1/Matrilin-3-deficient Mice*

Matrilins are oligomeric extracellular matrix adaptor proteins mediating interactions between collagen fibrils and other matrix constituents. All four matrilins are expressed in cartilage and mutations in the human gene encoding matrilin-3 (MATN3) are associated with different forms of chondrodysplasia. Surprisingly, however, Matn3-null as well as Matn1- and Matn2-null mice do not show an overt skeletal phenotype, suggesting a dominant negative pathomechanism for the human disorders and redundancy/compensation among the family members in the knock-out situation. Here, we show that mice lacking both matrilin-1 and matrilin-3 develop an apparently normal skeleton, but exhibit biochemical and ultrastructural abnormalities of the knee joint cartilage. At the protein level, an altered SDS-PAGE band pattern and a clear up-regulation of the homotrimeric form of matrilin-4 were evident in newborn Matn1/Matn3 and Matn1 knock-out mice, but not in Matn3-null mice. The ultrastructure of the cartilage matrix after conventional chemical fixation was grossly normal; however, electron microscopy of high pressure frozen and freeze-substituted samples, revealed two consistent observations: 1) moderately increased collagen fibril diameters throughout the epiphysis and the growth plate in both single and double mutants; and 2) increased collagen volume density in Matn1-/-/Matn3-/- and Matn3-/- mice. Taken together, our results demonstrate that matrilin-1 and matrilin-3 modulate collagen fibrillogenesis in cartilage and provide evidence that biochemical compensation might exist between matrilins.

Matrilins are oligomeric extracellular matrix adaptor proteins mediating interactions between collagen fibrils and other matrix constituents. All four matrilins are expressed in cartilage and mutations in the human gene encoding matrilin-3 (MATN3) are associated with different forms of chondrodysplasia. Surprisingly, however, Matn3-null as well as Matn1-and Matn2-null mice do not show an overt skeletal phenotype, suggesting a dominant negative pathomechanism for the human disorders and redundancy/compensation among the family members in the knock-out situation. Here, we show that mice lacking both matrilin-1 and matrilin-3 develop an apparently normal skeleton, but exhibit biochemical and ultrastructural abnormalities of the knee joint cartilage. At the protein level, an altered SDS-PAGE band pattern and a clear up-regulation of the homotrimeric form of matrilin-4 were evident in newborn Matn1/Matn3 and Matn1 knock-out mice, but not in Matn3null mice. The ultrastructure of the cartilage matrix after conventional chemical fixation was grossly normal; however, electron microscopy of high pressure frozen and freeze-substituted samples, revealed two consistent observations: 1) moderately increased collagen fibril diameters throughout the epiphysis and the growth plate in both single and double mutants; and 2) increased collagen volume density in Matn1 ؊/؊ /Matn3 ؊/؊ and Matn3 ؊/؊ mice. Taken together, our results demonstrate that matrilin-1 and matrilin-3 modulate collagen fibrillogenesis in cartilage and provide evidence that biochemical compensation might exist between matrilins.
Hyaline cartilage is a unique tissue, which contains relatively few cells (chondrocytes) embedded in an abundant extracellu-lar matrix (ECM). 3 The cartilage ECM assembles from collagens, proteoglycans, and non-collagenous glycoproteins that interact with each other and are organized into molecular networks. Heterotypic collagen fibrils, containing collagens II, IX, and XI, and the proteoglycan aggrecan, which forms large aggregates with hyaluronic acid, constitute the major structural components of the matrix providing tissue tensile strength and resistance against compressive forces, respectively. In addition to fibril forming collagens, articular cartilage contains beaded microfilament-forming collagen VI, while hypertrophic chondrocytes secrete collagen X. Other cartilage-specific ECM proteins play a role in matrix assembly by regulating collagen fibrillogenesis via binding to the surface of collagen fibers or by mediating interactions between the matrix and the chondrocytes.
The matrilin family of ECM proteins is comprised of four structurally related members (1,2). They contain one (matrilin-3) or two von Willebrand factor A (VWA) domain(s) (matrilin-1, -2, and -4), at least one epidermal growth factor (EGF)like motif, and a C-terminal coiled-coil domain, which allows the formation of homo-and hetero-oligomers of variable stoichiometry. In mouse, matrilin-1 and matrilin-3 are expressed almost exclusively in cartilaginous tissues (3,4), whereas matrilin-2 and matrilin-4 display a more widespread distribution pattern (5,6). In embryonic long bones, matrilin-1, -3, and -4 are abundant in epiphyseal as well as in growth plate cartilage, while matrilin-2 is prominent only in the outermost layer of the joint surface. In the adult knee, all matrilins are present in the growth plate and in the peripheral areas of the articular cartilage. In addition, matrilins, at least partially, co-localize in sternal, vertebral, and cranial cartilages (7).
The integrative role of matrilins in cartilage matrix assembly was suggested by earlier observations demonstrating that matrilin-1, formerly called cartilage matrix protein, binds covalently to aggrecan (8) and associates with collagen II-con-taining fibrils (9). More recent studies have revealed that matrilin-1, -3, and -4 form complexes with the small leucine-rich repeat proteoglycans (SRLPs) biglycan and decorin (10), bind to cartilage oligomeric matrix protein (COMP) (11), and that matrilin-3 interacts with collagen IX (12). These matrilin-containing molecular associates are in turn connected to collagen networks such as collagen VI microfibrils via SRLPs or collagen fibers via COMP and collagen IX (2,10).
To date, human skeletal disorders have been associated only with matrilin-1 and matrilin-3. Matrilin-3 mutations are linked to autosomal dominant forms of multiple epiphyseal dysplasia (MED) (13)(14)(15), to bilateral hereditary micro-epiphyseal dysplasia (BHMED) (16) and to an autosomal recessive form of spondylo-epi-metaphyseal dysplasia (SEMD) (17). These disease-causing mutations are predominantly confined to the ␤-sheet regions of the VWA domain of MATN3 and probably result in the retention and accumulation of the protein in the endoplasmic reticulum (18,19). In contrast, matrilin-1 was excluded as a candidate locus for various forms of heritable chondrodysplasias (20). In osteoarthritic cartilage, the expression of both matrilin-1 and -3 are elevated (21,22). A low occurrence linkage of MATN3 to hand osteoarthritis has been reported (23,24,25), whereas the association of the matrilin-1 gene with osteoarthritis was found in the Dutch but not in the British population (26,27).
Surprisingly, matrilin-1 (28)-, matrilin-2 (29)-, and matrilin-3 (30)-null mice generated previously in our laboratory show no obvious phenotype. However, an independent study reported ultrastructural abnormalities of the cartilage collagen fibrils in the maturation and hypertrophic zones of the growth plate of matrilin-1 knock-out mice (31). More recently, transient expansion of the hypertrophic zone between embryonic day (E) 16.5 and 17.5, higher incidence of osteoarthritis and increased bone mineral density (BMD) were reported in matrilin-3 knock-out mice (32). The lack of an overt phenotype in our matrilin-deficient mouse strains suggests functional redundancy within the matrilin family. To explore this possibility, we report here the establishment and analysis of matrilin-1/matrilin-3 double-deficient mice.
Whole Mount Skeletal Staining and X-ray Analysis-The gross morphology of the whole skeleton was analyzed by Alcian Blue/Alizarin Red staining of newborn, 2-and 4-week-old mice and by x-ray of 1-year-old mice as previously described (30). To determine the length of the long bones, the humerus, femur, and tibia were dissected from five control and five double mutant mice and measured longitudinally using a fine calibrated ruler.
Peripheral Quantitative Computed Tomography (pQCT)-Isolated femora were assessed by pQCT using the XCT Research M scanner with software 5.50 (Stratec Medizintechnik GmbH, Pforzheim, Germany). Machine calibration was performed using a hydroxyapatite standard cone phantom. Bones were inserted into a syringe filled with saline solution and oriented longitudinally with the anterior surface upwards in the scanner. After taking a low-resolution scout view of the entire bone, three transverse sections were recorded at the distal femoral metaphysis (15, 17.5, and 20% bone length measured from the distal joint line) and one section was taken at the midshaft (50% bone length). The slice thickness was 500 m with an in-plane pixel resolution of 70 m ϫ 70 m. Each slice was analyzed by contour mode 1, peel mode 20 (30%) and cortical mode 1 (710 mg/cm 3 ). At the femoral metaphysis the following parameters were measured (means of three slices): total cross sectional bone area (CSA, mm 2 ), total bone mineral density (BMD, mg/cm 3 ), total bone mineral content (BMC, mg), trabecular cross sectional area (Tb.CSA, mm 2 ), trabecular BMD (Tb.BMD, mg/cm 3 ), and trabecular BMC (Tb.BMC, mg). At the midshaft the cortical area (Ct.CSA, mm 2 ), the cortical BMD (Ct.BMD, mg/cm 3 ), the cortical BMC (Ct.BMC, mg), the cortical thickness (Ct.thickness, mm), the periosteal circumference (mm), and the endosteal circumference (mm) were evaluated.
Reproducibility of the pQCT measurements was determined from four repeated scans of nine femora using the ex vivo multiple slice scanning method with repositioning. The root-mean square (RMS) average CV% values (33) were 2.2% for Tb.BMD and 0.7% for Ct.BMD. Statistical analysis was done using the Student's t test analysis. A value of p Ͻ 0.05 was considered significant.
Light Microscopic Analysis of the Skeleton-For histological analysis, specimens of various stages were fixed overnight in fresh 4% paraformaldehyde (PFA) in phosphate-buffered saline (PBS), pH 7.4, decalcified in 10% EDTA-PBS if needed, and processed for paraffin embedding. Sections (6 m) were routinely stained with Hematoxylin and Eosin (HE). For morphometric analysis of growth plate zones, HE-stained sections of the proximal tibia were investigated at various stages (E16.5, newborn, 2-and 4-week-old) as described previously (34,35). To analyze whether matrilin-1/matrilin-3 deficiency causes osteoarthritic (OA) changes of the articular cartilage, knee joints from 14-month-old male and female wild-type (n ϭ 6 and 6), Matn1 ϩ/Ϫ /Matn ϩ/Ϫ (n ϭ 6 and 3) and Matn1 Ϫ/Ϫ / Matn3 Ϫ/Ϫ mice (n ϭ 6 and 5) were dissected. PFA-fixed samples were decalcified in a solution containing 1.35 N hydrochloric acid and 0.003 M EDTA for 3 days at room temperature and processed for paraffin embedding. For each knee joint, 6-mthick serial sections (n ϭ ϳ300) were cut, and every 10th section was either stained with HE, or with Safranin O and Fast Green. To assess cartilage degeneration, sections were evaluated by three blinded observers according to criteria described in Table 1.
Chondrocyte apoptosis and proliferation were analyzed on tissue sections using the in situ cell death detection kit (Roche Applied Science) and BrdU (5-bromo-2Ј-deoxyuridine) incorporation assay (38), respectively.
For cryofixation and high pressure freezing, tissue slices from proximal tibiae of 2-and 7-day-old animals were cut using a razor blade. Slices were placed in a Petri dish with 1-hexadecene and a circular ophthalmic punch (Miltex, Inc., York, PA) was used to excise 1.7-mm diameter disks. Individual disks were placed between two thin aluminum plates and mounted in the specimen holder of an Empact highpressure freezer (Leica Microsystem, Vienna, Austria). Cryofixation was achieved using a pressure of 2100 bar and a surface-cooling rate of 10,000°C/s. Following cryoimmobilization, samples were immediately immersed and stored in liquid nitrogen. Freeze substitution was performed in a liquid nitrogen-cooled cryostat: disks were immersed in an anhydrous solution of acetone containing 2% (w/v) osmium tetroxide for 17 h at Ϫ90°C, and for 12 h at Ϫ60°C and Ϫ30°C. The temperature was then raised to 0°C and, after a period of 1 h, the pure anhydrous acetone was exchanged three times prior to stepwise embedding in Epon 812. Polymerization (using fresh resin) was carried out at 60°C for 5 days. Both conventionally fixed and freeze-substituted sections were stained with uranyl acetate and lead citrate, and examined by electron microscopy.
Stereology of Cryofixed Cartilage Samples-Ultra-thin sections from two blocks of each experimental group were examined. The original pictures, taken with a Hitachi 7100 (Tokyo, Japan) electron microscope, were magnified to photographic papers to a final magnification of ϫ120,000. For collagen fibril diameter measurements, the widest area along the longitudinal axis of well visible fibrils in the intercellular matrix was chosen. Only fibrils running either perpendicular or near-parallel to the section plane were selected for measurement throughout the whole photographic image. Fibril diameters were divided into 9 incremental group sizes using transparency circles (40). The difference in diameters between individual group circles was 0.5 mm (ranging from 0.5 mm to 4 mm) on the final magnification scale used. Each measured fibril was classified into a size group, to which the circle was just fitting (or was slightly smaller). Statistical differences between genotype groups were evaluated using the Mann-Whitney U test.
To measure collagen fibril volume density in the intercellular matrix compartment, a point-grid was used according to Gundersen et al. (41). Collagen fibril density was defined as the sum of the area occupied by the fibrils in the image divided by the total image area. Statistical analysis was performed using the one-way analysis of variance test, the level of significance set at p Ͻ 0.05. SAS statistical software (version 8.2) was employed. Post-hoc comparisons were made using Bonferroni corrections.
RNA Isolation, Northern Blot Analysis, RT-PCR and in Situ Hybridization-Total RNA was isolated from newborn limb chondrocytes using the Qiagen RNeasy kit according to the instructions of the manufacturer. For Northern blot analysis, 20 g of total RNA was size-fractionated on 1% agarose-2.2 M formaldehyde gel and transferred to a Hybond XL membrane (Amersham Biosciences). Filters were consecutively hybridized with 32 P-labeled cDNA probes specific for matrilin-1, matrilin-2, matrilin-3, matrilin-4, and glyceraldehyde phosphodehydrogenase (Gapdh). For semi-quantitative RT-PCR, 2 g of total RNA was transcribed into cDNA using the SuperScript TM III RNaseH Ϫ Reverse Transcriptase (Invitrogen). 2 l of the reverse-transcribed material were amplified by PCR as described earlier (29) using primers specific for the coding sequences of matrilin-4 (forward, 5Ј-ACCCTCCCGCTCCAGGTTA-3Ј and reverse, 5Ј-CCACA-GTGTCACAGCCTGCA-3Ј) and Gapdh (forward, 5Ј-TCG-TGGATCTGACGTGCCGCCTG-3Ј and reverse, 5Ј-CACC-ACCCTGTTGCTGTAGCCG-TAT-3Ј) as positive control. PCR reactions were performed three times and relative mRNA levels were determined using the Bio-Profile computer-assisted imaging system (Vilbert Lourmat). Radioactive in situ hybridization was performed as described (34) (PAGE), and Immunoblotting-Knee joints and sterna were dissected, weighed, and frozen at Ϫ80°C. On the day of extraction, the specimens were thawed and cut into 1-mm 3 pieces. Ten volumes (ml/g of wet tissue) of chilled buffer I (0.15 M NaCl, 50 mM Tris-HCl, pH 7.4) (TBS) were added, and the tissue extracted for 7-10 h at 4°C with continuous stirring. The extracts were clarified by centrifugation, and the supernatants stored at Ϫ20°C. The pellets were re-extracted in an identical manner with buffer II (1 M NaCl, 10 mM EDTA, 50 mM Tris, pH 7.4) and the remaining insoluble material was extracted with buffer III (4 M GuHCl, 10 mM EDTA, 50 mM Tris, pH 7.4). All extraction buffers con- tained the protease inhibitors 2 mM phenylmethylsulfonylfluoride and 2 mM N-ethylmaleimide. Aliquots (100 l) of the extracts obtained with buffers I, II, and III were precipitated with 1 ml of 96% ethanol overnight at 4°C. The precipitates were washed with a mixture of 9 vol of 96% ethanol and 1 vol of TBS for 2 h at 4°C with gentle agitation. After centrifugation, the pellets were air-dried and suspended in 150 l of water, and the same volume of double concentrated non-reducing SDS-PAGE sample buffer was added. Aliquots were applied to 4 -15% SDS-polyacrylamide gels and SDS-PAGE performed as described by Laemmli (42). For immunoblots, the proteins were transferred to nitrocellulose and incubated with the appropriate affinity-purified rabbit or chicken antibodies (see immunohistochemistry) diluted in TBS. Bound antibodies were detected either by luminescence using peroxidase-conjugated swine anti-rabbit immunoglobulin G (Dako), 3-aminopthalhydrazide (1.25 mM), p-coumaric acid (225 M), and 0.01% H 2 O 2 or by infrared fluorescence using Alexa Fluor 680-labeled anti-rabbit (Molecular Probes), and IRDye TM 800 conjugated affinity purified anti-chicken immunoglobulin G (Rockland). The fluorographs were developed with the ODYSSEY infrared imaging system (LI-COR Biosciences).

Analysis of the Skeleton at Macroscopic and Light Microscopic
Levels-Matrilin-1/matrilin-3 double-deficient mice were obtained with the normal Mendelian ratio, showed no obvious abnormalities, were fertile, and had a normal life span. Neither whole mount skeletal staining of newborn, 2-, and 4-week-old mice (Fig. 1, A-E, and not shown) nor x-ray analysis of 1-yearold mice (Fig. 1F) revealed skeletal abnormalities in double mutants. pQCT measurements of distal femoral metaphysis and femoral diaphysis of male and female mice at 4 months of age indicated no differences in bone parameters between control (Matn1 ϩ/Ϫ /Matn3 ϩ/Ϫ ) and double-deficient animals ( Table 2).
It was recently reported that Matn3 Ϫ/Ϫ mice display a transient increase of the hypertrophic zone of the growth plate between E16.5 and E17.5 (32). We therefore examined long bones in control, Matn1 Ϫ/Ϫ , Matn3 Ϫ/Ϫ , and Matn1 Ϫ/Ϫ / Matn3 Ϫ/Ϫ mice at E16.5. HE-stained serial sections showed comparable appearance of the proximal tibia in each genotype group ( Fig. 2A and not shown) and morphometric analysis did not confirm a statistically significant expansion of the hypertrophic zone in matrilin mutants (Fig. 2B) The relative ratio of the hypertrophic zone to the proximal tibial length was nearly identical between the genotype groups (26.7, 25.5, 24.1, 25.6, and 25.8%, respectively). Organization of the postnatal tibial growth plate (Fig. 2, C and D), the expression of chondrocyte differentiation markers (Fig. 2E), cell proliferation and survival (not shown) were also indistinguishable between control and double mutant samples. In addition, no histological abnormalities were detected in the development of ribs, vertebral bodies and intervertebral discs (not shown).
Immunohistochemical, Biochemical, and RNA Analyses-To determine whether matrilin deficiency leads to alterations in the expression of other family members or of known binding partners, control (wild-type or Matn1 ϩ/Ϫ /Matn3 ϩ/Ϫ ), Matn1 Ϫ/Ϫ , Matn3 Ϫ/Ϫ , and Matn1 Ϫ/Ϫ /Matn3 Ϫ/Ϫ cartilage samples were investigated using various methods. First, immu-  JULY 27, 2007 • VOLUME 282 • NUMBER 30 nohistochemical staining confirmed the complete lack of matrilin-1 and matrilin-3 in newborn limb cartilage of double mutants and showed no obvious alteration in the deposition of either matrilin-2 and matrilin-4 ( Fig. 4A) or the interaction partners collagen II, collagen IX, collagen VI, aggrecan, COMP, decorin (Fig. 4B), and biglycan (not shown) compared with wild-type. Second, Northern blot analysis of total RNA isolated from newborn limb cartilage revealed that the expression of matrilin-2 and matrilin-4 mRNAs was similar in the different genotype groups (Fig. 5A). Furthermore, we quantified the matrilin-4 mRNA level using semiquantitative RT-PCR (Fig. 5, B and C) and observed no statistically significant differences between wildtype, double heterozygous, single and double knock-out mice. Finally, proteins were sequentially extracted from sternal and knee cartilage at newborn and 4.5week-old stages with neutral salt (fraction I), high salt with 10 mM EDTA (fraction II), and 4 M GuHCl (fraction III). The reproducibility of the extraction was checked on Coomassie Blue-stained SDS-polyacrylamide gels. The amounts of matrilin-1, -2, -3, and -4, COMP, collagen II, biglycan, and decorin were examined by immunoblotting after nonreducing SDS-PAGE (Fig. 6, A and B and not shown). Knee cartilage from newborn (Fig. 6A) and 4.5-week-old (Fig. 6B) double knock-out mice contained matrilin-2 and matrilin-4. The amounts of matrilin-2 oligomers in mutant cartilages were similar to those of the control at both stages, whereas the homotrimeric form of matrilin-4 gave significantly stronger signals in fractions II and III of cartilages extracted from knees of newborn mice (Fig.  6A). This result was confirmed in three independent extraction experiments (Fig. 6C). Sternal cartilages showed no increase in matrilin-4 either in newborn or in 4.5-week old mice (not shown). Interestingly, the up-regulation of matrilin-4 homotrimers was seen also in mice lacking only matrilin-1, whereas it was not detected in the matrilin-3 knock-out mice (Fig. 6A). Furthermore, the matrilin-4 band pattern was altered in the matrilin-1 and the matrilin-1/matrilin-3-deficient newborn mice (Fig. 6A). On a higher resolution gel, two bands between the trimeric and dimeric forms of matrilin-4 were missing in these mice (Fig. 6C). By double fluorescence staining of immunoblots specific for matrilin-1 and matrilin-4 we showed that these bands are positive only for matrilin-4, suggesting that they do not represent hetero-oligomers of matrilin-1 and -4 (Fig. 6D). In addition to the changes in matrilin-4 expression, alteration of the matrilin-1 banding pattern could be detected on Western blots at both newborn and 4.5-weekold stages (Fig. 6, A and B). In control mice, two minor bands were found above the main matrilin-1-containing band. These minor bands were missing in the matrilin-3-null mice suggesting that they represent hetero-oligomers formed of matrilin-1 and -3. In contrast, immunoblot analysis of the matrilin interaction partners COMP, biglycan, decorin, and collagen II (Fig.  6, A and B) revealed no apparent differences between control and mutant cartilage.

Matrilin-1 and Matrilin-3 Modulate Collagen Fibril Diameter
Ultrastructural Analysis-Ultrastructure of the cartilaginous matrix in control, Matn1 Ϫ/Ϫ , Matn3 Ϫ/Ϫ , and Matn1 Ϫ/Ϫ / Matn3 Ϫ/Ϫ mice was analyzed by electron microscopy on conventionally fixed and on cryofixed and freeze-substituted sections. Upon glutaraldehyde fixation of epiphyseal and growth plate samples, the cartilage matrix appeared as a bulk of electron-dense proteoglycan granules embedded into a meshwork of collagen fibrils (Fig. 7). In general, no obvious differences in chondrocyte morphology and in the structural organization of the collagen fibrillar network were detected in the epiphyseal cartilage (not shown) or in the proliferative (not shown) and maturation (Fig. 7) zones of the growth plate between control (Matn1 ϩ/ϩ /Matn3 ϩ/ϩ or Matn1 ϩ/Ϫ / Matn3 ϩ/Ϫ ) and double mutants at 3 days (Fig. 7, A, B, D and  E, F, H) or 4 weeks of age (not shown). Similarly, no patho-logically relevant ultrastructural alterations of the various matrix compartments were evident in single matrilin-1 or single matrilin-3 knock-out mice (Fig. 7, C and G and not  shown).
To preserve the chondrocytes and their surrounding matrix in a more natural state we performed ultrastructural analysis on high pressure frozen, freeze-substituted, and Epon-embedded cartilage samples at postnatal (P) days 2 and 7. At low magnification, the morphology of control (Matn1 ϩ/Ϫ /Matn3 ϩ/Ϫ ) and double mutant chondrocytes was similar in all zones of the growth plate and the epiphyseal cartilage (not shown). At higher magnification a fine, cross-striated collagenous network was identifiable in the matrix. This network was random and stretched from the cell surface through all compartments. Initial analysis at P2 showed increased fibrillar density and higher proportion of thicker fibrils in the matrix of double mutant mice (n ϭ 2) compared with controls (not shown). These abnormalities were more obvious at P7 (Fig. 8, A and B) and thick fibrils were also observed in Matn1 Ϫ/Ϫ and Matn3 Ϫ/Ϫ mice (Fig. 8, C and D). Morphometric analysis of the interterritorial matrix in the epiphyseal and growth plate regions at P7 revealed that collagen fibrils had an average diameter of 11.16 nm, 11.69 nm, 12.73 nm, and 13.30 nm in controls, double mutants, single matrilin-3, and single matrilin-1 mutants, respectively (Fig. 9). Statistical analysis indicated that the mean values are significantly different between each of the experimental groups (p Ͻ 0.05). In all groups more than 99.5% of the fibrils ranged between 2.45 nm and 26.96 nm, however the distribution profile of collagen fibrils mildly shifted toward to large diameters in matrilin mutants (Fig. 9). In controls, 15.3% of the fibrils had a diameter larger than 12.26 nm, whereas the proportion of thick fibrils was 21.4, 31.0, and 33.0% in Matn1 Ϫ/Ϫ / Matn3 Ϫ/Ϫ , Matn3 Ϫ/Ϫ and Matn1 Ϫ/Ϫ mice, respectively (Fig.  9). In addition, collagen volume density significantly increased in double matrilin-1/-3 and single matrilin-3 mutants (0.390 Ϯ 0.074, p Ͻ 0.0001; and 0.261 Ϯ 0.036, p Ͻ 0.05), but not in matrilin-1 mutants (0.233 Ϯ 0.067), compared with double heterozygous control mice (0.214 Ϯ 0.065) (Fig. 10).

DISCUSSION
We have generated mice deficient in matrilin-1 and matrilin-3, two members of the matrilin subfamily of ECM molecules, which are highly expressed in cartilage, to elucidate their role in skeletogenesis and to investigate redundancy/compensation between matrilins. Previous results from our laboratories indicated that single Matn1, Matn2, and Matn3 knock-out mice develop without obvious abnormalities of the skeletal system (28 -30). Here, we report that Matn1/Matn3 double knock-out mice have an apparently normal skeleton at the macroscopic and light microscopic levels, but display biochemical and ultrastructural abnormalities of the cartilage. Some of these abnormalities were also evident in Matn1-and/or Matn3-null mice leading us to re-evaluate the consequence of single matrilin deficiency for cartilage matrix organization.
Previous biochemical analysis of single matrilin knock-out mice revealed no phenotype (28 -30). Here we observed an increased amount of matrilin-4 homotrimers in cartilage

Matrilin-1 and Matrilin-3 Modulate Collagen Fibril Diameter
extracts of newborn double knock-out mice compared with control mice. Interestingly, the analysis of the matrilin-1 knock-out mice revealed the same effect. When the matrilin-1 knockouts were originally characterized (28) antibodies against matrilin-4 were not yet available. As Northern blot and RT-PCR analyses indicated no up-regulation at the transcriptional level, either the translation of the protein is increased or the turnover is decreased. Alternatively, the extractability of matrilin-4 could be increased. Considering the role of the matrilin family as adapter proteins, it is possible that the loss of one member leads to a less stable supramolecular assembly. Further evidence for a close association of matrilin-1 and -4 is provided by the loss of two particular oligomers in both matrilin-1-and matrilin-1/-3-deficient animals but not in Matn3-null mice. Although in vitro experiments with recombinant coiled coil domains have demonstrated the propensity of matrilin-1 and matrilin-4 for hetero-oligomer formation (43), we could by double fluorescence staining exclude that these bands indeed represent such hetero-oligomers. Based on the electrophoretic mobility, the molecular masses of the bands lie between that of a full-length matrilin-4 trimer and that of a proteolytically cleaved dimer (5), which could be because of shorter splice variants (44) or N-terminally processed forms. The reason for the loss of these forms of matrilin-4 in matrilin-1-deficient mice remains unclear. We speculate that the lack of matrilin-1 may, as a compensational mechanism, alter the extracellular processing of matrilin-4 favoring the formation of homotrimers. In contrast, the extractability of COMP, collagen II, biglycan, and decorin, the binding partners of matrilins, was not affected.
Earlier studies demonstrating the association of matrilin-1 with collagen II-containing cartilage fibrils (9) and with the major cartilage proteoglycan aggrecan (8) lead to the hypothesis that matrilin-1, and maybe also the structurally similar other family members, play a role in collagen fibrillogenesis and matrix assembly. This putative function is controversially discussed in the literature. Whereas conventional electron microscopic analyses of Matn1-and Matn3-null mice generated in our laboratory revealed no significant abnormalities in the ultrastructure of the cartilaginous matrix (28,30), an independent study reported mild alterations in collagen fibril organization in newborn tibiae of Matn1 knock-out mice (31). Those included bundling and about 30% increase in mean fibril diameter exclusively in the maturation (postmitotic) and hypertrophic zones of the growth plate. However, this spatially restricted phenotype is contradictory to the expression pattern of matrilin-1 in mouse: Matn1 is transcribed by epiphyseal and proliferative growth plate chondrocytes (3), and the protein is deposited throughout the growth plate (3,7).
These discrepancies prompted us to perform a careful ultrastructural analysis of cartilage isolated from control, single and double knock-out mice. Electron microscopy of glutaraldehyde fixed samples revealed normal organization of the collagen network with no apparent changes of fibril diameters in the epiphyseal and growth plate cartilages (including the maturation zone) independent of the genotype. Using high-pressure freezing followed by freeze substitution, which preserves the carti-   lage ultrastructure in a more native state (45), we were, however, able to detect mild abnormalities of the collagen network architecture in single and double matrilin mutants, which were not obvious in chemically fixed samples. These include 1) a mild but significant increase in large diameter fibrils in Matn1 Ϫ/Ϫ , Matn3 Ϫ/Ϫ , and Matn1 Ϫ/Ϫ /Matn3 Ϫ/Ϫ mice; and 2) an increase in collagen volume density in Matn3 Ϫ/Ϫ and Matn1 Ϫ/Ϫ / Matn3 Ϫ/Ϫ animals. Because these abnormalities were evident in both the epiphyseal and growth plate matrices, we propose that a biological function of matrilin-1 and matrilin-3 is to modulate collagen fibrillogenesis and collagen network organization in cartilage extracellular matrix.
The mechanism behind the ultrastructural phenotypes is still unclear. One possibility, that matrilins bind directly to collagen II has been proposed for matrilin-1 (9) and another, that they attach to the surface via an interaction with collagen IX was recently demonstrated for matrilin-3 (12). Collagen IX, a heterotrimer of ␣1(IX)␣2(IX)␣3-(IX) polypeptide chains, is covalently associated with the surface of collagen II/XI fibrils and mutations in any of the three genes encoding the polypeptides result in MED (46). Mice with targeted inactivation of the Col9a1 gene lack the functional collagen IX protein (47) and develop early onset osteoarthritis without overt morphological alterations in collagen fibrils of the knee cartilage (48). However, collagen IX-deficient fibrils isolated from costal cartilage have strongly reduced matrilin-3 content and have a slightly increased average fibril diameter compared with control fibrils (12). Alternatively, matrilins could bind to the fibrils via adaptor proteins such as COMP (11,12), decorin, and biglycan (10). COMP binds to collagen I/II (49) and collagen IX (50,51), and it has been implicated in collagen fibrillogenesis (46). Although COMP-null mice display normal skeletal development (52), COMP mutations in human are associated with pseudoachondrodysplasia and some types of MED (46). The highly homologous class I small leucine-rich repeat proteoglycans (SLRP) biglycan and decorin are also present in cartilage (53) as well as in other tissues such as bone, tendon, and skin. Mice deficient in these proteoglycans exhibit a variety of connective tissue disorders primarily associated with abnormal collagen fibrillogenesis (54). In Swarm rat chondrosarcoma tissue, decorin, and biglycan in complex with matrilin-1, -3, and -4 connect collagen VI-containing microfibrils to the collagen II network or to the aggrecan core protein (10). Independent of the manner in which matrilins are attached, the present study demonstrates that the lack of one or more family members in the cartilage matrix disturbs collagen fibril organization. Single matrilin-1 or matrilin-3 deficiency similarly increases fibrillar diameters, implicating that both matrilins are involved in the fine control of lateral fibril growth. In double knock-out animals the collagen volume density is almost doubled, whereas the means and distribution profiles of collagen fibril diameters show less pronounced changes compared with controls. It is tempting to speculate that single and double matrilin-deficiency differently alter the molecular assemblies controlling the structure of the collagen network.  The absence of matrilin-1 or matrilin-3 may predominantly modulate fibril diameters, while the lack of both matrilins induces changes important for interfibrillar spacing. The physiological consequences of these alterations are not clear, but they could influence the mechanical properties of the cartilage. Collagen fibrils are the major determinants of tissue strength. Changes in the diameter profile and spacing of collagen fibrils in matrilin-deficient mice may therefore lead to alterations in cartilage tensile strength and elasticity.
Despite the mild ultrastructural abnormalities, double and single matrilin-deficient mice (28,30) have a normal cartilaginous and bony skeleton without any sign of pathological malformations suggesting that matrilin-1 and matrilin-3 are dispensable for skeletal development. However, we cannot exclude the possibility that matri-lin1/3 and a properly organized cartilage matrix are important in stress situations, such as bone fracture healing. Unfixed bone fractures are repaired by the formation of the cartilaginous callus tissue which is, subsequently, remodeled into bone recapitulating the sequence of endochondral ossification. Mice, lacking the matrilin interaction partner collagen IX display an altered matrilin-3 deposition (12) accompanied by delayed cartilaginous maturation and ossification of the callus (55). Whether or not matrilins play a direct role in the healing of fractured bones remains to be elucidated.
Interestingly, three distinct abnormalities were recently reported in Matn3 Ϫ/Ϫ mice generated by van der Weyden et al. (32): 1) a transient expansion of the hypertrophic zone of the growth plate at E16.5-E17.5 because of premature chondrocyte differentiation; 2) an increased whole body and knee joint BMD at 18 weeks of age; and 3) a higher predisposition for severe knee osteoarthritis in 1-year-old mutant mice. Contrary to these observations, none of these defects are visible in Matn1 Ϫ/Ϫ /Matn3 Ϫ/Ϫ mice. Histo-morphometry at E16.5 revealed comparable lengths of the hypertrophic zone in control, Matn1 Ϫ/Ϫ , Matn3 Ϫ/Ϫ , and Matn1 Ϫ/Ϫ /Matn3 Ϫ/Ϫ mice; pQCT measurement of isolated femora obtained from 16-week-old control and double matrilin-null mice showed no statistically significant difference in bone parameters including cortical and trabecular BMD; and histopathological grading of the articular cartilage from 14-month-old knee joints did not confirm a prevalence of moderate-to-severe OA in double-deficient mice compared with controls. The reasons for the discrepancies between our data and the results of van der Weyden et al. (32) are not clear, but some of them could partially be explained by genetic background and/or gender differences. For example, we found that bone parameters and the incidence of moderate-to-severe osteoarthritis are affected by the sex ( Table 2 and Fig. 3E) in both control and double mutant mice. Female mice consistently show lower BMD and milder erosion of the articular cartilage than male mice. Because van der Weyden et al. did not discriminate between males and females we speculate that an unequal male/female ratio could influence their results. Clearly, large scale studies with mutant mice on defined genetic and sexual background are necessary to clarify

Matrilin-1 and Matrilin-3 Modulate Collagen Fibril Diameter
the impact of matrilins on osteoarthritis and bone mineral density.
Furthermore, our results suggest that human chondrodysplasias associated with mutations in the MATN3 gene such as MED and SEMD are not caused by the absence of matrilin-3 in the cartilage matrix. We have recently reported that two MED/ SEMD causing point mutations in MATN3 lead to the almost complete retention of the mutant proteins in the rough endoplasm reticulum of transfected primary articular chondrocytes (19). Similarly, MED mutations in MATN3 cause misfolding of the VWA domain and intracellular accumulation of the mutant matrilin-3 in CHO-B2 cells (18). In addition, light and electron microscopic analyses of cartilage from an MED patient with a MATN3 mutation revealed intracellular staining of matrilin-3 in chondrocytes and a dilated rER (18). These findings implicate that the intracellular retention of the mutant matrilin-3 results in chondrocyte dysfunctions such as reduced viability or the general perturbance of the secretory pathway leading to the co-retention of other cartilage matrix components (e.g. COMP or collagen IX) together with the mutant matrilin-3.
In summary, the present study demonstrates that both matrilin-1 and matrilin-3 act as modulators of collagen fibrillogenesis in cartilage and suggests that the biological function of these matrilins is similar. Furthermore, the mild ultrastructural phenotype and the temporary up-regulation of matrilin-4 homotrimers in matrilin-1/matrilin-3 double knock-out mice indicate that matrilin-4, and possibly also matrilin-2, may, at least partially, compensate for the lacking matrilins in hyaline cartilage. Alternatively, functional redundancy may exist among matrilins and their interaction partners such as COMP, decorin and biglycan. Nevertheless, a more precisely defined role of matrilins in cartilaginous matrices can only be identified by the generation of mice lacking all of the four matrilins in cartilage.