A novel chordin-like BMP inhibitor, CHL2, expressed preferentially in chondrocytes of developing cartilage and osteoarthritic joint cartilage

Bone morphogenetic proteins (BMPs) belong to the transforming growth factor(TGF) β super family. These molecules play important roles during many organogenic processes, even though they were originally identified as factors promoting the ectopic formation of cartilage and bone. The concentration of active BMP is controlled in part by inhibitors from three BMP binding protein families: short gastrulation/chordin, noggin and cerberus (reviewed by Balemans and Van Hul, 2002). Among these, chordin and noggin were the first proteins found to inhibit the activity of bound BMPs by preventing interactions with their BMP receptors(Piccolo et al., 1996; Zimmerman et al., 1996). Although noggin is encoded by a single gene in mammals, chordin belongs to a family of proteins that share a cysteine-rich pro-collagen repeat [or chordin-like cysteine-rich repeat (CR)], which is also found in various extracellular matrix proteins (reviewed by Garcia Abreu et al., 2002). Without exception, the homology between chordin family members lies within their CRs.

The chordin polypeptide contains four CRs, of which the first and the third(CR1 and CR3) are responsible for BMP binding(Larrain et al., 2000). Binding of chordin to BMP4 is specific and tight(Piccolo et al., 1996). Proteolysis by Tolloid (or BMP1), which liberates CR1 and CR4 from chordin, is required to release bound BMP4 (Piccolo et al., 1997; Scott et al.,1999). The importance of CR for BMP interactions has been strengthened by the recent finding that connective tissue growth factor functions as a BMP-binding inhibitor, and that its single CR domain is essential for this activity (Abreu et al.,2002).

We previously described a small chordin-like secreted protein, CHL1 (for chordin-like 1, re-designated from CHL), a novel BMP-binding inhibitor with three CRs (Nakayama et al.,2001). CHL1 was isolated originally from mouse bone marrow stromal cells. Interestingly, CHL1 expression correlates with the stem/progenitor-support activities of over 19 stromal cell lines established from the aorta-gonadsmesonephros region, the site at which definitive hematopoietic stem cells first arise during embryogenesis(Oostendorp et al., 2002). However, CHL1 mRNA is also detected in various mesenchymal derivatives associated with (1) the dermatome, limb bud and chondrocyte precursors of the skeleton during embryogenesis, and (2) digestive tract connective tissues,kidney tubules and marrow stromal cells in adults. In addition, CHL1is expressed in olfactory bulb and cerebellum, suggesting a wider array of physiological functions. Two other groups have independently isolated CHL1,naming it neuralin-1 and ventroptin(Coffinier et al., 2001; Sakuta et al., 2001) and demonstrating its ability to correctly specify retinotectal projections along the dorsoventral retinal axis during development.

We provide evidence that CHL2, a novel chordin family member with structural homology to CHL1, is a BMP-binding inhibitor whose expression is uniquely restricted to the superficial layers of developing joint cartilage,in contrast to that of other family members. Potential downregulation of cartilage matrix accumulation and/or cartilage mineralization by CHL2 is suggested by in vitro observations using cartilage particles derived from embryonic stem (ES), cell-derived mesodermal cells(Nakayama et al., 2003) and with marrow-derived mesenchymal stem/progenitor cells (MSCs). CHL2 is also induced in osteoarthritic joint cartilage, implying a potential role during cartilage regeneration in the adult.

Enzymes for the polymerase chain reaction (PCR) and cDNA library constructions; recombinant human proteins (and corresponding antibodies for western blot detection) for BMP4, BMP5, BMP6, activin A, TGFβ2, and BMP receptor 1B-Fc fusion protein (BMPR1B-Fc); human platelet-derived growth factor (PDGF)-BB, human TGFβ3, and mouse noggin-Fc fusion protein(noggin-Fc); monoclonal antibodies against collagens type II (COL2, clone 2B1.5) and type X (COL10, clone X53); other staining reagents, and all culture vessels were obtained as described previously(Nakayama et al., 2003; Nakayama et al., 2001). Mouse chordin (mCHD-His); human BMP2, BMP7 and TGFβ1 (and mouse monoclonal antibodies against them); mouse monoclonal anti-human TGFβ3; mouse growth and differentiation factor 5 (GDF5); and affinity-purified goat polyclonal anti-mouse GDF5 were from R&D Systems (Minneapolis, MN). Goat polyclonal anti-human IgG1-Fc fragment (IgG-Fc) was obtained from Sigma. Human kidney epithelial cell lines 293 and 293T were maintained as described previously(Nakayama et al., 2001). The human MSCs (hMSCs) were obtained from BioWhittaker (Walkersville, MD). The E14 mouse ES cells were grown, and embryoid body (EB) cells were harvested,treated and sorted as described previously(Nakayama et al., 2003).

Mouse placentas were isolated at E18. A signal-trap cDNA library and a regular full-length cDNA library were constructed as described previously(Nakayama et al., 2001). From 400 trap-positive clones sequenced, a cDNA fragment encoding the NH₂-terminal sequence of a putative secreted protein with significant homology to Xenopus chordin was identified (designated mouse CHL2 or mCHL2) by a BLAST search (Accerlys, San Diego, CA). Using this partial cDNA as a probe, the corresponding full-length cDNA (approximately 1.8 kb) was isolated, and designated as pSPORTmCHL2.

A human placenta library was constructed with size-selected (>1.5 kb)oligo(dT)-primed cDNAs in the pSPORT1 vector (Gibco). A full-length human cDNA clone (hCHL2, 1.5 kb in length) was isolated using the mouse cDNA probe, and designated pSPORThCHL2. Rat CHL2 (rCHL2) cDNA was cloned by PCR from a rat fetal liver cDNA library (Stratagene) using the following primers: sense,5′-TCCTCTCATCCTCACCTTAG-3′ (based on mCHL2-5′UTR), and antisense, 5′-GAGGGTAATGCGACTTCTTT-3′ (based on mCHL2-3′UTR). A 1.2 kb fragment was amplified using Advantage-HF2 enzyme(Clontech), cloned into pTOPO2.1 (Invitrogen), and designated pTOPOrCHL2.

To prepare mCHL2, the mCHL2 open reading frame (ORF) was mutated by PCR to replace the stop codon with a SalI site, inserted into a pFLAG-CMV5a expression vector (Sigma) to attach the FLAG sequence in-frame to mCHL2 at its COOH terminus (mCHL2-FLAG), and designated pFLAGmCHL2. Expression was checked by transient transfection of 293T cells, followed by direct western blot analysis of conditioned media, using the anti-FLAG monoclonal antibody M2 as described previously(Nakayama et al., 2001).

A large-scale, transient transfection-based expression was performed as described with 293T cells bearing pFLAGmCHL2(Nakayama et al., 2001),yielding approximately 10 μg/ml of mCHL2-FLAG. The protein was purified by affinity chromatography using anti-FLAG M2 affinity gel (Sigma) under high-salt conditions, as described by Piccolo et al.(Piccolo et al., 1997), after which positive fractions were subjected to hydroxyapatite column chromatography (equilibrated with 10 mM phosphate, with gradient elution from 10 mM to 400 mM phosphate) at pH 6.9. Purity was confirmed by SDS-polyacrylamide gel electrophoresis followed by Coomassie Blue staining. Approximately 5 mg of >95% pure mCHL2-FLAG protein were obtained from 2.5 l of conditioned medium.

A rabbit polyclonal antibody for mCHL2 (αCHL2-COOH) was raised to the peptide NH₂-CPEDEAEDDHSEVISTR-COOH, and affinity purified against the corresponding peptide (Harlow and Lane, 1988).

Immunoprecipitations to demonstrate direct interactions between BMPs,TGFβs and activin A were performed as described previously(Nakayama et al., 2001) except that only one condition was used: 200 ng mCHL2-FLAG were mixed with 100 ng of BMP/GDF/activin/TGFβ in 1 ml binding buffer, followed by 12 μg/ml ofαCHL2-COOH. The BMP/GDF/activin/TGFβ immunocomplex was precipitated with 20 μl protein A agarose beads (Santa Cruz), fractionated on an SDS-polyacrylamide gel (NuPAGE, Invitrogen), blotted, and visualized with the corresponding antibody as described previously(Nakayama et al., 2001), or with 1 μg/ml of anti-BMP2, anti-BMP7, anti-TGFβ1 or anti-TGFβ3 or 0.3 μg/ml of anti-GDF5. Each blot then was treated with 4.4 μg/ml M2 to confirm the precipitation of mCHL2-FLAG. The inhibitory effect of mCHL2-FLAG(0.1-1 μg, in 1 ml binding buffer) on BMP4 binding (at 100 ng/ml) to the BMPR1B ectodomain (BMPR1B-Fc at 1 μg/ml) was performed as described by Nakayama et al. (Nakayama et al.,2001), except that BMP4 visualization on blots was followed by CHL2 and BMPR1B-Fc detection using 4.4 μg/ml M2 and 2.2 μg/ml anti-IgGFc, respectively.

Inhibition of BMP4 by mCHL2 was assessed in Xenopus embryos. The EcoRI-NotI fragment of pSPORTmCHL2 was cloned into the EcoRI-NotI sites of pCS2+(Rupp et al., 1994), and the resulting plasmid was linearized with NotI. Capped mRNAs were synthesized with SP6 polymerase, quantified, diluted and injected into two ventral blastomeres as described previously(Nakayama et al., 2001).

Promyoblast C2C12 cells were maintained and differentiated according to the method of Kirsch et al. (Kirsch et al.,2000a). Briefly, cells were plated at 3×10⁴cells/well in a 96-well plate, and after 1 day, stimulated to differentiate for 72 hours in 120 μl of Dulbecco's Modified Eagle's medium with 2% calf serum (Gibco) in the presence or absence of BMP and/or CHL2. Cells were washed and lysed. Alkaline phosphatase (AP) activity was measured with p-nitrophenyl phosphate (Sigma), with specific activity calculated as the amount of p-nitrophenol produced in 30 minutes at 37°C,normalized to total protein content, as determined with BCA reagent(Pierce).

Northern blotting was performed against the entire open reading frame (ORF)of CHL2 on human and mouse multiple tissue RNA blots (Clontech) using the ³²P-labeled XbaI-SalI fragment from pSPORThCHL2 as a probe (Sambrook et al.,1989).

In situ hybridization (ISH) was performed on 5 μm paraffin sections of zinc formalin-fixed, formic acid-decalcified tissue according to the method of Wilcox (Wilcox, 1993),including a high stringency wash with 0.1× SSC at 55°C. Slides hybridized with ³³P-labeled RNA probes were exposed to NTB2 emulsion (Kodak) for 3 weeks, developed and counterstained with Hematoxylin and Eosin. Expression was subjectively evaluated under dark-field microscopy.

For mCHL2, the 1234-bp SalI-NotI fragment and the 653 bp HindIII fragment were deleted from pSPORTmCHL2 to obtain pSPmCHL2-COOH and pSPmCHL2-NH₂, respectively. Both plasmids were linearized with EcoRI to generate non-overlapping probes. An hCHL2-bearing plasmid, pSPhCHL2NH2, was made by deleting the 811 bp BamHI fragment from pSPORThCHL2, leaving the 746 bp hCHL2 NH₂-terminal ORF. The plasmid was linearized with SalI. RNA probes for hCHL2 and mCHL2 were synthesized with SP6 RNA polymerase. For rCHL2, an approximately 1.2 kb fragment of rCHL2 cDNA derived from pTOPOrCHL2 was cloned into the pBluescript vector (pBSrCHL2). The plasmid was linearized with XbaI and a labeled RNA probe synthesized with T7 RNA polymerase.

CHL2 expression was assessed by ISH in mouse embryonic/adult tissues, and in structurally normal and arthritic joints of adult humans (knee cartilage) and rats (hind paws). Human diseased cartilage samples from rheumatoid arthritis (RA) or osteoarthritis (OA)/degenerative joint disease(DJD) patients were provided by Cooperative Human Tissue Network, and normal control samples were from Zoion Diagnostics (New York, NY), with pre-approval of the Institutional Review Board (758WIRB and CP1098/01, respectively). Collagen-induced arthritis (CIA) was induced, as described(Trentham et al., 1977), in anesthetized female Lewis rats (7-8 weeks old, 150-170 g) by intradermal injection of porcine COL2 (Chrondex, Seattle, WA) emulsified 1:1 in incomplete Freund's adjuvant (Difco, Detroit, MI) at 10 different sites over the back (50μg COL2/100 μl/injection). Disease developed between 10 and 12 days after injection, as determined by caliper measurements (Cole Parmer, Vernon Hills, IL) of ankle width and ambulatory difficulties. Paws were harvested for ISH 7 days after CIA onset. These experiments were conducted in accordance with federal animal care guidelines and were pre-approved by the Amgen Institutional Animal Care and Use Committee.

Human MSCs were cultured and differentiated as described(Mackay et al., 1998). Briefly, the pellet culture was performed in serum-free chondrogenesis medium supplemented with 10 ng/ml TGFβ3, with or without 2 μg/ml mCHL2-FLAG,1 μg/ml noggin-Fc, or 1 μg/ml IgGFc. On days 21-28, cartilage-like particles were formalin-fixed, paraffin-embedded, sectioned centrally and stained with Toluidine Blue to detect sulfated glycosaminoglycans(Nakayama et al., 2003; Sheehan and Hrapchak, 1987). Three sections from different regions were examined to confirm staining reproducibility.

To analyze gene expression, two to five particles were harvested at designated times and disrupted immediately in guanidine isothiocyanate solution (RNeasy kit, Qiagen). Total RNA was purified using the manufacturer's protocol, including DNase I treatment. Reverse transcription (RT) and PCR were performed as described previously(Nakayama et al., 1998),except that the PCR used 30 cycles, an annealing temperature of 62°C and one primer set per gene. Primers for aggrecan, cartilage oligomeric matrix protein (COMP), COL1, COL2, COL10, SOX9, CHL1, CHL2 and glyceraldehyde-3-phosphate dehydrogenase (GAPD) are shown in Table 1.

Mineralizing cartilage particles were produced as described by Nakayama et al. (Nakayama et al., 2003). Briefly, 3-4×10⁵ FACS-purified EB cells were pellet-cultured in serum-free chondrogenesis medium with 10 ng/ml TGFβ3 and 50 ng/ml PDGF-BB. On day 10, TGFβ and PDGF were replaced by 50 ng/ml BMP4 to generate a hyaline cartilage particle. On day 15, cultures were adjusted to the hypertrophic differentiation medium without T3 (Sigma), containing 3μg/ml mCHL2-FLAG, 2 μg/ml noggin-Fc or 50 ng/ml BMP6 for 3 days; 10 nM T3 was added to induce hypertrophic differentiation on day 18. On days 24-26,each particle was stained with Toluidine Blue as described above(Sheehan and Hrapchak, 1987). Additional serial sections were immunostained with 2B1.5 for COL2 and X53 for COL10, and stained to reveal mineral deposition (von Kossa). Immunostained sections were counterstained with Gill 2 Hematoxylin, and von Kossa sections with Nuclear Fast Red (Sheehan and Hrapchak, 1987).

Random nucleotide sequencing of a mouse placenta library enriched for genes of secreted and membrane-bound proteins revealed a cDNA encoding a protein precursor consisting of a potential signal peptide followed by three chordin-like cysteine-rich repeat (CRs) homologous to those of chordin and CHL(Fig. 1A,C). A GAP search(Accerlys) revealed amino acid sequence identities with mCHL(l3) and chordin of 41% and 33%, respectively. The predicted precursor size of 426 amino acids resembled mCHL(l3) (448 amino acids) more than mouse chordin (999 amino acids). Because this novel cDNA was clearly homologous to CHL, we designated it CHL2 and renamed CHL as CHL1. Human and rat CHL2 cDNAs were cloned also. The human gene encoded a 429 amino acid precursor with 73% amino acid homology to mCHL2, while the rat variant was 95% identical to mCHL2(Fig. 1B).

A genomic DNA fragment containing a region spanning at least the 5′-half of mCHL2 was cloned from a 129SV genomic BAC library(Genome Systems). Using this clone, mCHL2 was localized to chromosome 7, mapping to a position 66% of the distance from the heterochromatic-euchromatic boundary to the telomere. A search of public genome databases confirmed that mCHL2 is on 7E1 and hCHL2 is on 11q13. Therefore, unlike CHL1, CHL2 is an autosomal gene. Interestingly, the exon-intron junctions of CHL2 were nearly identical to those of CHL1 (not shown), suggesting that both genes may originate from a common ancestor.

Chordin is known to dorsalize the gastrulating Xenopus embryo by inhibiting BMP4 activity, so the impact of CHL2 on Xenopus embryo development was examined (Table 2). Injection of 1 pg mCHL2 RNA per blastomere induced trunk duplication in 74% of embryos, compared with 0% for uninjected controls and embryos given EF1α mRNA. As a positive control, injection of 10 pg mCHL1(s2) RNA yielded an axis duplication rate of 80%(Nakayama et al., 2001). These results indicated that mCHL2 actively antagonized an endogenous ventralizing factor (presumably BMP4). The improved efficacy afforded by a 10-fold lower CHL2 dose suggested that it might be a more stable and/or potent BMP inhibitor than CHL1.

mCHL2-FLAG protein was purified to near homogeneity(Fig. 2A). Unlike mCHL1(s2)-FLAG, the overall yield was higher, and degradation products were not detected.

Purified mCHL2-FLAG co-immunoprecipitated BMP2, BMP4, BMP5, BMP6, BMP7 and GDF5 (Fig. 2B). Like chordin and CHL1, mCHL2-FLAG did not bind activin A, TGFβ1 or TGFβ3. However, unlike CHL1, no interaction between mCHL2-FLAG and TGFβ2 was found. The control protein noggin-Fc exhibited comparable qualitative binding specificity (not shown), suggesting that CHL2 may be a pan-BMP-binding protein like noggin and chordin.

Chordin and CHL1 inhibit BMP activity by blocking their interactions with their receptors. Therefore, we determined whether CHL2 had a similar function by mixing an Fc-fusion protein, incorporating the extracellular domain of BMP receptor 1B (BMPR1B-Fc), with BMP4 and mCHL2-FLAG, followed by precipitation of the BMPR1B-Fc complex with protein A beads. As shown in Fig. 2C, BMP4 co-precipitated specifically with BMPR1B-Fc, but not IgG-Fc, in the absence of mCHL2-FLAG. However, the signal for co-precipitated BMP4 weakened appreciably as increasing amounts of mCHL2-FLAG were added (particularly at 0.3 μg/ml or higher). Co-precipitation of mCHL2-FLAG was not detected. These results suggest that CHL2 acts like chordin and CHL1 to prevent BMP4 interacting with its receptor.

Next, we demonstrated that the recombinant mCHL2 inhibited BMP activity by quantifying BMP-dependent AP induction in C2C12 cells. A serially diluted BMP inhibitor [chordin (mCHD-His), mCHL2-FLAG, or noggin-Fc] was mixed with BMP2,BMP4, BMP6 or BMP7 at a concentration corresponding to the EC50 for AP induction in C2C12 cells and then cultured for 3 days. Cell-bound AP activity was then measured (Fig. 3, Table 3). CHL2 inhibited AP induction by all four BMPs. Noggin-Fc and mCHL2-FLAG reproducibly elicited similar, dose-dependent inhibitions of BMP4, with complete suppression occurring at concentrations of 1-3 μg/ml (20-60 nM). The mCHD-His activity was weakest for BMP4; it was approximately fivefold less potent than mCHL2-FLAG. Partially purified mCHL1(s2)-FLAG and mCHD-FLAG(Nakayama et al., 2001)inhibited BMP4 with a potency indistinguishable from that of mCHD-His (not shown). In contrast, both noggin-Fc and mCHD-His displayed weaker inhibitory activities than mCHL2-FLAG toward BMP6 and BMP7. In particular, noggin-Fc was approximately sevenfold less potent on BMP6 than mCHL2-FLAG. Thus, CHL2 inhibits BMP2, BMP4, BMP6 and BMP7 as well as or better than noggin and chordin.

In mouse embryos, ISH for CHL2 mRNA revealed expression restricted to the surface chondrocytes of developing joint cartilage(Fig. 4) and to the connective tissue of reproductive organs (Fig. 5). In the adult mouse, CHL2 was weakly expressed in cartilage of the femoral head and patella(Fig. 4C,D) and articular facets of vertebrae (Fig. 4E,F); the latter location had a relatively stronger signal. CHL2 was also expressed weakly in the annulus fibrosus of intervertebral discs in adults (not shown). In general, CHL2expression in adult cartilage was weaker than that of embryonic cartilage. CHL2 in normal cartilage was confined to articular chondrocytes,especially in the superficial zone. Expression was always observed on both sides of a joint. CHL2 transcripts were never detected in growth plate cartilage or bone during development or adulthood. In rats, CHL2 also showed low to moderate expression in sternal cartilage during embryogenesis(not shown) and in joint cartilages of adult paws(Fig. 6). Thus, among skeletal compartments, CHL2 seems to be expressed preferentially in the superficial zone chondrocytes of developing articular cartilage.

Northern analyses revealed little mCHL2 mRNA on the mouse embryo blot, in contrast to the abundant expression of mouse chordin and mouse CHL1(Nakayama et al., 2001). On the mouse adult tissue blot, a faint 2 kb transcript was detected in liver and kidney extracts, and a trace amount of a >2.4 kb transcript was detected in skeletal muscle and testis, although ISH failed to localize a signal in these tissues (not shown). The adult mouse blot did not include uterus. On the human adult tissue blot, a 2 kb CHL2 transcript was detected readily in uterus and weakly in colon, heart, prostate, stomach and skeletal muscle(Fig. 5A); only the uterus and colon had expression patterns consistent with the ISH signals in mouse tissues(Fig. 5B). Traces of hCHL2 mRNA were found in placenta, testis, small intestine, trachea and bone marrow. Expression was not detected in either human liver or kidney. Thus, low levels of CHL2 expression in soft tissues may differ between humans and mice.

In mouse reproductive organs, ISH detected CHL2-positive connective tissues such as ligaments of the ovary and oviduct in females, and of testis,epididymis and certain male accessory sex glands in males. Expression was high in uterine myometrium (Fig. 5Ba), as was that of CHL1 (not shown). CHL2 was also present in maternally derived placental tissues (not shown). A trace CHL2 signal was found on colonic serosa (Fig. 5Bb); in contrast, CHL1-positive cells lie between the colonic submucosa and muscularis (Nakayama et al.,2001). Interestingly, CHL1 but not CHL2 was expressed in stomach and small intestine(Nakayama et al., 2001). In rat tissues, CHL2 occurred at low to moderate levels in cervical muscles and discrete regions of the placenta (not shown).

Degenerating cartilage from human arthritis patients and rats with CIA were assessed by ISH (Fig. 6). In two relatively normal specimens from knees of adult humans, CHL2 mRNA was expressed in a few chondrocytes in the superficial zone as well as in the middle zone (Fig. 6A). In 19 OA cases, expression was limited to chondrocytes in the middle zone, where numerous well-labeled cells were observed(Fig. 6B,C); positive cells were not found in the superficial zone in any OA sample. Interestingly, 50-90%of such CHL2-expressing chondrocytes existed in clusters of 2-3 cells. Unlike OA, two RA specimens exhibited weak expression in both the superficial and middle zones (Fig. 6D). As with humans, scattered chondrocytes in normal articular cartilage of rats expressed CHL2(Fig. 6E), while CIA paw joints had similar (Fig. 6F) or fewer labeled chondrocytes relative to controls.

In summary, CHL2 was expressed in normal and diseased cartilage in humans and rats. It was expressed most strongly in human OA patients, although the signal had shifted to the middle zone. Interestingly, CHL2expression levels and patterns were not significantly altered, relative to normal cartilage, in the rat CIA model and human patients with RA.

We further addressed the relevance of CHL2 to cartilage formation using MSC, which can differentiate into chondrocytes in vitro(Mackay et al., 1998) in association with upregulated BMP transcription(Roh et al., 2001). As shown by RT-PCR in Fig. 7A, hCHL2 mRNA, but not COL2 mRNA, was expressed by undifferentiated MSCs. Transcripts for SOX9, COL1 and COL10 (not shown) as well as CHL1, aggrecan and COMP were also detected. The CHL2 signal fell as chondrogenesis progressed; those for SOX9 and COL10 (not shown) as well as CHL1, aggrecan and COMP(Fig. 7A) maintained a similar level throughout the culture period. In contrast, COL2 mRNA was tightly regulated, with production induced by day 7 in culture conditions favoring chondrogenesis but absent under osteogenic conditions (not shown)(Jaiswal et al., 1997).

Next, mCHL2-FLAG protein was added at various concentrations to the chondrogenic pellet culture of MSC. mCHL2-FLAG (2 μg/ml) significantly inhibited cartilage nodule formation (Fig. 7B, Table 4),yielding definitive cartilage nodules in less than 10% of particles. Inhibition was absent at 0.1-0.2 μg/ml, while complete inhibition occurred at 3-10 μg/ml. Noggin-Fc at 0.1 and 1 μg/ml provided similar,dose-dependent inhibition.

These results implicate CHL2 as a negative regulator of cartilage formation/regeneration from immature mesenchymal cells, by preventing or reducing the rate of matrix accumulation.

Chondrocytes from OA joints express markers of hypertrophy, such as COL10 and AP (Kirsch et al., 2000b; Von der Mark et al., 1992), so we addressed whether CHL2 induction in OA cartilage would affect differentiation and mineralization of hypertrophic chondrocytes. We demonstrated previously that mesodermal progenitor cells, purified from differentiating ES cells, can form hyaline cartilage particles in vitro that will undergo further mineralization(Nakayama et al., 2003). We isolated FLK1^– PDGFRα⁺ mesodermal cells,subjected them to pellet micromass culture, and induced cartilage matrix mineralization (verified by von Kossa staining) in the presence or absence of mCHL2-FLAG, noggin-Fc or BMP6 (Fig. 7C, Table 5). Addition of 3 μg/ml mCHL2-FLAG significantly reduced the von Kossa-positive matrix area in 75% of particles examined, of which half showed near-complete inhibition. In contrast, COL10 expression was reduced slightly, while no significant change was detected in COL2(Fig. 7C). Noggin-Fc at 2-3μg/ml provided similar, but somewhat weaker, inhibition. A positive in vitro role for BMP6 has been suggested in chondrocyte hypertrophic differentiation (Grimsrud et al.,1999). However, BMP6 at 50 ng/ml, which was sufficient to enlarge the particle size, did not affect the degree of mineralization.

Taken together, these results suggest that CHL2 induction in damaged joint cartilage may reduce the extent or speed of hypertrophic differentiation in articular chondrocytes.

We have demonstrated here that CHL2, the second chordin-like gene, encodes a protein that directly interacts with different BMPs, inhibiting their actions in vitro as well as in vivo in a manner similar to chordin, noggin and CHL1. Expression analysis suggests a possible role for CHL2 during formation and maintenance of articular cartilage and reproductive organs. We have also provided evidence that CHL2 might negatively regulate cartilage formation/regeneration in diseased joints.

Searches of human and mouse genome databases indicated that CHL2 is most homologous to CHL1. Injection of CHL2 RNA induced trunk duplication in early Xenopus embryos similar to those produced by chordin and CHL1(s2)RNAs (Table 2)(Nakayama et al., 2001). Recombinant mCHL2 protein interacted directly with five BMPs and one GDF(Fig. 2 and Nakayama et al., 2001) thereby inhibiting, in vitro, several BMP/GDF-dependent processes including,osteogenic differentiation of C2C12 mesenchymal progenitor cells by several BMPs (Fig. 3, Table 3), ATDC5 embryonal carcinoma cells by GDF5 (not shown) and BMP4-dependent lymphohematopoietic(CD34⁺CD31^hi and CD34⁺CD31^lo)progenitor cell development from ES cells (not shown)(Nakayama et al., 2000). Under our conditions, CHL2 provided 50% inhibition (IC50) by blocking a half to a third of available BMP dimers, suggesting that tight CHL2 binding to one BMP subunit might be sufficient for full inhibition. Furthermore, as with related factors (chordin, noggin, CHL1), CHL2 prevented BMP interactions with the BMP receptor (Fig. 2C), although CHL2 activity was two- to sevenfold more potent than chordin(Table 3) and CHL1 (not shown). Thus, CHL2 is structurally and functionally similar to chordin and CHL1.

Cartilages within hip and knee joints and at the costochondral junction were the major CHL2 expression sites during embryogenesis(Fig. 4). CHL2 mRNA was also expressed strongly in connective tissues anchoring reproductive organs(Fig. 5). CHL2 in developing joints was restricted to superficial zone chondrocytes; expression was substantially diminished in adult joint cartilage(Fig. 4). The CHL2-expressing areas did not overlap domains expressing chordin (non-chondrogenic mesenchyme of limb buds), CHL1 (condensing mesoderm, hypertrophic chondrocytes) and gremlin (non-chondrogenic regions of limb buds, including interdigital mesenchyme) during limb formation(Nakayama et al., 2001; Scott et al., 1999; Scott et al., 2000),suggesting that these four factors have divergent biological roles. However,as with CHL1 (Nakayama et al.,2001), CHL2 expression in developing cartilage overlapped with noggin expression (Brunet et al.,1998; Capdevila and Johnson,1998; Merino et al.,1998; Nifuji and Noda,1999; Pathi et al.,1999).

As CHL2 is a BMP-binding inhibitor, and BMPs regulate multiple steps during chondrogenesis, expression of CHL2 in superficial chondrocytes in developing joints suggests a role in joint specification. The ability of exogenous mCHL2 to inhibit chondrogenesis by hMSCs supports this hypothesis(Fig. 7A,B, Table 4). The surface of developing cartilage consists of proliferating mesenchymal cell layers that are differentiating into chondrocytes. By its location, CHL2 might act as an important boundary in joint formation. A possible role could be to prevent articular cartilage from becoming too massive, by keeping mesenchymal cells in the joint space from being recruited to the chondrocyte developmental pathway.

Alternatively, CHL2 could play more subtle roles. The superficial zone of articular cartilage is composed of flattened chondrocytes separated by tangential arrays of thin collagen fibrils, but no proteoglycan matrix. In contrast, the middle zone consists of rounded chondrocytes surrounded by a proteoglycan-rich matrix containing radial bundles of thick collagen fibrils. Osteogenic BMPs accumulate in the pericellular matrix of articular cartilage,with highest levels in the middle to deep zone(Anderson et al., 2000). Conversely, the osteogenic antagonist BMP3(Daluiski et al., 2001) is more highly expressed in the superficial zone. We failed to detect an interaction between CHL2 and BMP3 (not shown), suggesting that preferential expressions of CHL2 and BMP3 in the surface chondrocytes act to regulate a BMP gradient in normal articular cartilage.

CHL2 mRNA was never detected in the growth plate, where proliferation and hypertrophic differentiation of prehypertrophic chondrocytes normally occur,implying that CHL2 is not relevant to normal pathways of chondrocyte proliferation and maturation. However, the up-regulation of CHL2 transcripts specifically in middle zone cartilage of adult joints with OA(Fig. 6) prompted our speculation that CHL2 has a role in cartilage repair. We examined the associations between CHL2 and three principal phenotypes of OA cartilage: (1)reduced proteoglycan levels (which precede overt histological changes), (2)aberrant chondrocyte proliferation (resulting in clonal chondrocyte expansion), and (3) upregulation of molecules (e.g., COL1, COL3, COL10, and AP) found in hypertrophied or de-differentiated chondrocytes but not normal articular chondrocytes (Aigner et al.,1993; Kirsch et al.,2000b; Von der Mark et al.,1992). First, the proteoglycan content in CHL2-expressing regions of OA cartilage was not reduced, as detected in Toluidine Blue-stained sections by the retained metachromasia (not shown). Second, CHL2-expressing chondrocytes in OA cartilage were typically found as aggregates; however,middle to deep zone chondrocytes are normally arranged in a cylindrical fashion, so this association might reflect normal middle zone anatomy. In contrast, the weak but significant inhibition of cartilage mineralization by CHL2 (Fig. 7C, Table 5) suggested that in OA cartilages this molecule might delay and/or reduce the degree of chondrocyte hypertrophy, thereby ameliorating cartilage degeneration. Further support for this premise is that medium turbidity, which indicates mineral deposition and excess Ca²⁺ excretion, was delayed during hypertrophic differentiation culture of cartilage particles by CHL2 or noggin (not shown). However, we have not addressed whether CHL2 is involved in the de-differentiation of mature articular chondrocytes. Co-localization analyses between cells expressing the CHL2 mRNA and those expressing transcripts for COL1, COL10 or proliferating cell-nuclear antigen are underway to answer this question.

In conclusion, abundant evidence suggests that BMP functions are regulated by numerous extracellular BMP-binding proteins in developing joints. Our current data support this paradigm and add a new BMP inhibitor, CHL2, to this pathway. Our findings also provide the first evidence that a chordin-like BMP-binding inhibitor might be intimately involved in the pathogenesis of degenerative joint disease.

First, we wish to acknowledge the many scientists and technicians in the Amgen Genomics and Bioinformatics groups who constructed and analyzed the nucleotide sequence databases. We greatly appreciate the superb technical support provided by R. Haldankar, H. Yamane, and M. Haniu for mid-scale mCHL2 preparation and NH₂-terminal amino acid sequence determination, and by R. Manoukian for cell sorting. We are grateful to D. Chang, L. Daugherty,C. Baikalov and H. Yamane for preparing and purifying anti-mCHL2 peptide antibodies. Finally, we warmly thank W. Boyle and S. Simonet for support and encouragement. R.N. and T.Y. are supported by Amgen Japan, K.K. Cooperative Human Tissue Network is funded by the National Cancer Institute.