BMPR1A is necessary for chondrogenesis and osteogenesis, whereas BMPR1B prevents hypertrophic differentiation

Bone morphogenetic proteins (BMPs) play a role in a multitude of processes during embryonic development, including skeletal development. Most skeletal elements are formed by endochondral ossification, which is initiated by mesenchymal cell condensation. Under the influence of several BMPs, these cells differentiate into chondrocytes that produce cartilaginous tissue, and further differentiate into hypertrophic chondrocytes. Afterward, vascularization takes place concomitantly with replacement of cartilage into bone, driven by osteoblasts and osteoclasts (Katagiri and Watabe, 2016; Salazar et al., 2016). Several conditional deletion studies have shown that BMP2 is essential for osteogenesis and bone formation during this process (Bandyopadhyay et al., 2006; McBride et al., 2014; Yang et al., 2013). Similarly, during bone fracture healing – where a similar mechanism takes place – conditional deletion of Bmp2 in mesenchymal progenitors or osteoprogenitors prevents fracture healing (Mi et al., 2013; Tsuji et al., 2006). In vitro, BMP2 provokes an induction of alkaline phosphatase (ALP) activity, osteocalcin expression and matrix mineralization in pluripotent mesenchymal progenitor cells (Cheng et al., 2003), and also stimulates chondrogenesis or adipogenesis (Date et al., 2004). Finally, BMP2 has been shown to promote bone repair in animal models (Kleinschmidt et al., 2013; Wulsten et al., 2011) and in a clinical setting (Ronga et al., 2013).

Growth differentiation factor 5 (GDF5; also named BMP14 or CDMP1) is also expressed in the developing limb at sites where joint cavitation occurs, and null mutation of Gdf5 in mice disrupts the formation of 30% of the synovial joints (Storm and Kingsley, 1996). Several gene expression or deletion studies strongly indicate that GDF5 (together with GDF6 and GDF7) provides a specialized function for joint morphogenesis (Settle et al., 2003; Storm and Kingsley, 1999). In vitro, GDF5 promotes chondrogenic differentiation and chondrocyte hypertrophy (Coleman et al., 2013; Erlacher et al., 1998). Subcutaneous implantation of collagen or collagen/hyaluronate loaded with GDF5 in the rat results in cartilage formation followed by bone formation (Erlacher et al., 1998; Spiro et al., 2000). In bone defect models, GDF5 caused delayed tissue mineralization compared to BMP2 and the presence of cartilage tissue in the defect (Kleinschmidt et al., 2013; Wulsten et al., 2011). Finally, intra-articular injections of GDF5 stimulated cartilage repair in a rat osteoarthritis model (Parrish et al., 2017). In conclusion, both BMP2 and GDF5 promote cartilage and bone formation, but BMP2 more strongly promotes osteogenesis and bone formation than does GDF5.

BMP2 and GDF5 elicit their effects through two types of serine/threonine kinase transmembrane receptors; type I and type II receptors. There are three type II and seven type I receptors that interact with BMPs, and the association of a type I with a type II receptor is required for the formation of an active signaling complex. BMP2 and GDF5 are both known to interact with all three type II receptors (BMPR2, ACVR1 and ACVR2) together with the type I receptors BMPR1A or BMPR1B. However, BMP2 and GDF5 preferentially interact with BMPR1A and BMPR1B, respectively (Heinecke et al., 2009; Nishitoh et al., 1996). In accordance with the role of BMPs in skeletal development, the receptors Bmpr1a and Bmpr1b are expressed in the developing limbs (Baur et al., 2000; Dewulf et al., 1995; Zou et al., 1997). Several studies have aimed to decipher the role of BMPR1A and BMPR1B in skeletal development. In vivo, the use of loss-of-function or gain-of-function mutations revealed some redundancy in the role of these two BMP type I receptors (Kobayashi et al., 2005; Yoon et al., 2005). In vitro, contradictory results have been published. For instance, it has been shown that constitutively active forms of BMPR1A and BMPR1B activate chondrogenesis in ADTC5 cells (Fujii et al., 1999), whereas forced expression of BMPR1A, but not BMPR1B, stimulates osteogenesis and chondrogenesis in C3H10T1/2 cells (Kaps et al., 2004).

In conclusion, the respective role of BMPR1A and BMPR1B in chondrogenesis and osteogenesis is not yet fully clear. In this study, we tackled this question from a different angle. Although genetic manipulations are extremely helpful, the results should be interpreted with caution. Silencing the expression of one protein can be compensated by the expression of other similar proteins (Rossi et al., 2015), while overexpression induces non-physiological protein expression levels and might generate results that do not accurately reflect the biology. As an alternative, we generated several mutants of GDF5 that differentially activate BMPR1A or BMPR1B. These mutants were tested together with wild-type GDF5 and BMP2 for their chondrogenic, hypertrophic and osteogenic properties in the multipotent mesenchymal precursor cell line C3H10T1/2, primary chondrocytes and the human osteosarcoma cell line Saos-2.

The binding affinities of BMP2, GDF5 and the three GDF5 mutants for BMPR2, BMPR1A and BMPR1B were measured by surface plasmon resonance (SPR) (Table 1). The affinity measurements of BMP2, GDF5, R399E and W417F with BMPR2 resulted in K_d values in the 2-digit nM range, while no binding of W417R could be detected. The K_d values obtained for BMPR1B were in the 2- to 3-digit pM range. GDF5 showed the strongest affinity for BMPR1B, followed by BMP2, R399E, W417F and W417R. Regarding BMPR1A, BMP2 showed the strongest affinity, followed by GDF5 and the GDF5 mutants, which displayed strongly reduced affinities for BMPR1A; the affinity of R399E and W417F for BMPR1A was 6 and 41 times lower compared to GDF5 and no binding of W417R to BMPR1A could be detected. The ratios of the K_d BMPR1A to K_d for BMPR1B were calculated to describe the binding preferences of the compounds to BMPR1A or BMPR1B. As previously described, compared to BMP2, GDF5 has an increased binding preference for BMPR1B over BMPR1A (Heinecke et al., 2009). Here, the GDF5 mutants R399E and W417F showed a strongly increased selectivity for BMPR1B in comparison to GDF5.

BMP2, GDF5 and the GDF5 mutants were further evaluated in a cellular assay to determine their ability to dimerize BMPR2 with BMPR1A or BMPR1B (Fig. 1). All compounds were found to be active in both assays. The corresponding EC₅₀ K_d values were calculated (Table 2). BMP2 was the most potent compound for the dimerization of BMPR2 with BMPR1A, followed by R399E, GDF5, W417F and W417R. Regarding the dimerization of BMPR2 with BMPR1B, R399E was the most potent compound, followed by BMP2, GDF5, W417R and W417F. Similar to the K_d values, the ratio of EC₅₀ values were calculated. The same selectivity profile was obtained; BMP2 was more selective for BMPR1A than BMPR1B and the selectivity was shifted toward BMPR1B with GDF5, R399E, W417F and W417R.

To ensure that the cells selected to study chondrogenesis and osteogenesis are relevant, we verified that they express the receptors we intend to study. The C3H10T1/2 cells, Saos-2 cells and porcine chondrocytes all express Bmpr1a/BMPR1A, Bmpr1b/BMPR1B and Bmpr2/BMPR2 (Fig. S1). In addition, because the proteins we use in this work derive from the human GDF5 sequence and were tested on human receptors for the affinity or dimerization measurements, we also evaluated the percentage identity between human, porcine or mouse BMPRs (see Table S1). For all sequences, a high identity was found (>96%) and variations within the sequence are limited to regions that are not involved into ligand binding (Kotzsch et al., 2009; Weber et al., 2007; Mace et al., 2006).

Finally, we also evaluated the noggin (Nog) sensitivity of the GDF5 mutants. Nog is an inhibitor of BMPs that can be produced by chondrocytes or C3H10T1/2 cells (Kameda et al., 2000; Zehentner et al., 2002). It was previously shown that Noggin inhibits GDF5, and that specific mutations of GDF5 can suppress Noggin inhibition. (Seemann et al., 2009). We found cells that stimulated with GDF5, R399E and BMP2 were strongly sensitive to Nog treatment but those stimulated with W417F and W417R were less affected by Nog (Fig. S1). We also observed that GDF5 treatment caused increased Nog expression in C3H10T1/2 cells but treatment of R399E, W417R or W417F did not.

To summarize, the chosen cells express all three receptors, and porcine and murine BMPRs present a high identity to the human BMPRs. However, we do not know whether mouse or porcine BMPRs behave in a similar manner to the human receptor, which is a limitation of the present study. In addition, all mutants were found to cause different sensitivities to Nog to that of GDF5, and they did not stimulate Nog expression while GDF5 did.

C3H10T1/2 cells were cultured as pellets in a chondrogenic medium in the presence of 300 ng/ml of BMP2, GDF5, R399E, W417R or W417F. After 14, 21 or 28 days, samples were harvested for the analysis of chondrogenic and hypertrophic marker expression (Fig. 2).

In the absence of compound (control), none of the analyzed genes were modulated at day 14 in comparison to day 0. At later time points it was impossible to isolate RNA, indicating a low level of cellular activity. Similarly, no RNA could be isolated from the cells cultured with W417R (data not shown).

With GDF5 treatment, the expression of Acan, Col2a1 and Sox9 were increased, which indicates that GDF5 promotes chondrogenesis. Col10a1 was also increased in the presence of GDF5 but Runx2 and ALP activity were not. Similarly, R399E increased Acan and Col2a1 expression (albeit both to a lesser extent than GDF5) and Sox9 expression. In contrast to GDF5, in the presence of R399E the expression of Col10a1 remained low; Runx2 was also not modulated at early time points and ALP activity was not detectable. W417F had only a weak effect on all parameters. Finally, BMP2 increased Acan expression but not Col2a1, while all hypertrophy markers – Col10a1, Runx2 expression and ALP activity – were stimulated by BMP2. From these results it appears that both GDF5 and R399E are chondrogenic with a limited hypertrophic effect, while BMP2 showed weaker chondrogenic but a stronger hypertrophy effect.

The chondrogenic and hypertrophic activity of the different compounds was evaluated in primary chondrocytes. Cells were cultured in 3D for 28 days with 300 ng/ml GDF5, R399E, W417F, W417R or BMP2. Similar to the C3H10T1/2 cells, chondrogenic and hypertrophic markers were evaluated (Fig. 3).

GDF5, R399E and BMP2 all stimulated at least two chondrogenic markers, and all increased COL2A1 expression. In addition, GDF5 and BMP2 stimulated ACAN expression while R399E and BMP2 showed elevated SOX9 expression. W417F and W417R showed no significant effects on these parameters. BMP2 was the only compound to significantly increase COL10A1 expression. RUNX2 was slightly increased by all compounds (only reaching significance for R399E, but no significant difference between R399E and GDF5 was observed). Finally, ALP activity was increased by all compounds except W417R. GDF5 and BMP2 showed the greatest ALP activity increase, followed by R399E and W417F. These results indicate that GDF5, R399E and BMP2 exert a chondrogenic effect in primary chondrocytes. In addition, GDF5 and R399E stimulated hypertrophy but to a lesser extent than BMP2.

R399E exerted the same effects on primary chondrocytes as GDF5, except for ALP activity, which was significantly lower with R399E, and both W417F and W417R presented reduced chondrogenic and hypertrophic capabilities in comparison to GDF5.

C3H10T1/2 cells were cultured in monolayer in an osteogenic medium for 28 days with 300 ng/ml GDF5, R399E, W417F, W417R or BMP2 (Fig. 4A). Both GDF5 and BMP2 increased three of the six osteogenic markers tested; Spp1 and Bglap expression were elevated in the presence of GDF5, Bglap expression and ALP activity were increased by BMP2 and both showed positive Alizarin Red staining after 28 days. R399E, W417F and W417R showed no osteogenic activity.

The culture of Saos-2 cells in the presence of the different compounds was conducted to further evaluate the osteogenic capacity of the compounds. Only BMP2 increased ALP levels. In addition, the strongest Alizarin Red staining was obtained in the presence of BMP2 (Fig. 4B).

GDF5, BMP2 and the GDF5 mutants were evaluated for their binding affinities to BMPR2, BMPR1A and BMPR1B (by SPR) and their ability to induce dimerization of BMPR2 with BMPR1A or BMPR1B in a cellular dimerization assay. The resulting affinities (K_d values), potencies (EC₅₀ values) and calculated ratios (K_d for BMPR1A to K_d for BMPR1B; EC₅₀ for BMPR1a–BMPR2 to EC50 for BMPR1b–BMPR2) enabled comparison of the compounds for their selectivity for BMPR1B or BMPR1A. As expected, both the ratios for the K_d and EC₅₀ values showed that BMP2 binds preferentially to BMPR1A over BMPR1B, followed by GDF5, R399E, W417F and W417R. Of note, the three GDF5 mutants showed a strongly reduced binding affinity for BMPR1A in comparison to GDF5, while the affinity for BMPR1B was less affected. When comparing the obtained K_d and EC₅₀ values for BMPR1A or BMPR1B, there were some discrepancies. For instance, R399E had a lower affinity for BMPR1A but was more potent at inducing dimerization of BMPR1A–BMPR2 than GDF5. In addition, no binding of W417R with either BMPR2 or BMPR1A could be detected by SPR, while W417R was found to induce dimerization of these two receptors in the dimerization assay. These differences can be explained by the fact that SPR evaluates binding to one receptor subtype only and does not take into account that the oligomerization of type I and type II receptor is required for signaling. In contrast, the dimerization assay might better encompass the complexity of a ternary receptor–ligand complex (Heinecke et al., 2009). Consequently, we have chosen to use the results from the dimerization assay to subsequently interpret the roles of BMR1A and BMPR1B in chondrogenic, hypertrophic and osteogenic differentiation.

Chondrogenesis and hypertrophic differentiation of the multipotent mesenchymal precursor cells, C3H10T1/2, was stimulated by GDF5. BMP2 stimulated chondrogenesis in C3H10T1/2 cells to a lesser extent than GDF5, and increased Acan and Sox9 expression but not Col2a1 expression. However, the hypertrophic capacity of BMP2 was stronger than GDF5; it increased Col10a1 more strongly than GDF5 and stimulated Runx2 expression and ALP activity while GDF5 did not. In mature chondrocytes, both GDF5 and BMP2 had a similar impact on chondrogenic markers, but only BMP2 stimulated COL10A1 expression. Finally, GDF5 and BMP2 stimulated osteogenesis in C3H10T1/2 cells, but only BMP2 stimulated ALP activity and increased Alizarin Red staining in the osteoblast-like Saos-2 cells. These results confirm that BMP2 and GDF5 can both promote chondrogenesis and osteogenesis, but that GDF5 appears to favor chondrogenesis while BMP2 favors osteogenesis. This is in accordance with the results from other studies (Kleinschmidt et al., 2013; Wulsten et al., 2011).

W417R was as potent as GDF5 in the BMPR2–BMPR1B dimerization assay but possessed the lowest potency for inducing BMPR2–BMPR1A dimerization (16 times lower than GDF5). It was also the only mutant that had no effect on C3H10T1/2 chondrogenesis or osteogenesis, and no impact on the expression of chondrogenic, hypertrophy and osteogenic markers in primary chondrocytes or Saos-2 cells, respectively. This reduced activity cannot be due to an effect mediated by Nog as W417R was found to be poorly inhibited by Noggin (Fig. S1B). These results indicate that BMPR1A, but not BMPR1B, signaling is necessary for chondrogenesis and osteogenesis. Similarly, Kaps et al. (2004) observed that, in C3H10T1/2 cells, the forced expression of a dominant negative Bmpr1a strongly affected the initiation of chondrogenesis and osteogenesis, while the expression of a dominant negative Bmpr1b did not. In addition, Jing et al. (2013, 2017) observed that a conditional Bmpr1a knockout in the cartilage of mice resulted in impaired postnatal chondrogenesis, reduced cartilage matrix in the growth plate and the arrest of long bone growth.

R399E showed a similar profile to GDF5 on chondrogenesis in C3H10T1/2 cells and the expression of chondrogenic markers in primary chondrocytes, but had little or no effect on hypertrophy marker expression. In contrast, W417F and W417R had little/no effect on both chondrogenesis and hypertrophy in C3H10T1/2 cells and primary chondrocytes. To better understand how selectivity for BMPR1A or BMPR1B influences the chondrogenic and hypertrophic capacities of the compounds, Col2a1/COL2A1 (as a chondrogenesis marker) and Col10a1/COL10A1 (as a hypertrophy marker) expression were plotted against the EC₅₀ values obtained in the dimerization assay. No correlation between the EC₅₀ values for BMPR2–BMPR1A or BMPR2–BMPR1B dimerization and Col2a1/COL2A1 or Col10a1/COL10A1 expression in C3H10T1/2 cells or chondrocytes was observed (data not shown). The ratio of the EC₅₀ values (EC50_{BMPR1a–BMPR2}:EC50_{BMPR1b–BMPR2}) were then plotted against the ratio of Col2a1/COL2A1 to Col10a1/COL10A1 (Fig. 5). The latter was used as an indicator of the intensity of chondrogenesis versus hypertrophy stimulation. A correlation was observed between the EC₅₀ ratio and the ratio of Col2a1:Col10a1 or COL2A1:COL10A1 expression. This indicates that, in contrast to BMPR1A, stronger BMPR1B signaling favors chondrogenesis over hypertrophy. As a result, R399E, for example, induced chondrogenesis similarly to GDF5 but exerted a low hypertrophic effect.

We observed that GDF5 and the GDF5 mutants are differently inhibited by Noggin, and that the GDF5 mutant did not influence Nog expression in C3H10T1/2 cells while GDF5 did. In addition, others have demonstrated that BMP2 is inhibited by Nog and increases Nog expression similarly to GDF5 (Seemann et al., 2009). These differences can influence the activity of the compounds, but we postulate that Nog would similarly inhibit chondrogenesis, and hypertrophy and would not affect the Col2a1:Col10a1 or COL2A1:COL10A1 ratios.

In this study, instead of using genetic manipulation to elucidate the role of BMPR1A and BMPR1B in chondrogenesis and osteogenesis, we used a different approach. We engineered GDF5 mutants that bind to these receptors with different affinities and evaluated their effects on the chondrogenesis and osteogenesis of C3H10T1/2 cells, the expression of chondrocyte and hypertrophy markers in primary chondrocytes and the expression of osteogenic markers in Saos-2 cells. Our results confirm that BMPR1A is necessary for the initiation of chondrogenesis and osteogenesis. On the other hand, BMPR1B does not seem to play a role in osteogenesis; we propose that BMPR1B acts as an inhibitor of chondrocyte hypertrophy and that the relative strength of BMPR1A versus BMPR1B signaling determines cellular fate. This role of BMPR1B is in accordance with its expression pattern in cartilage and areas prefiguring cartilage during skeletal development (Yi et al., 2000; Zou et al., 1997). Therefore, we suggest that BMPR1B signaling might prevent hypertrophy and stabilize the cartilage surface during joint morphogenesis. Bmpr1b is also highly expressed during early chondrogenesis in mesenchymal cells before and during condensation of progenitor cells (Yi et al., 2000; Yoon et al., 2005) and might play a different role at this stage. Indeed, it currently appears that both BMPR1A and BMPR1B, as well as several BMPs (Salazar et al., 2016), are involved at this stage explaining the redundancy between BMPR1A and BMPR1B (Yoon et al., 2005) observed during early chondrogenesis.

To further understand the role of BMPR1A and BMPR1B it would be of interest to evaluate how their expression level might change during chondrogenesis and osteogenesis and how BMPs with different receptor affinities influence their expression. We could not observe any influence of any of the tested compound on Bmpr1a and Bmpr2 during chondrogenesis and osteogenesis in C3H10T1/2 cells (data not shown). In our experimental setting Bmpr1b expression was too low to be detected.

GDF5 was previously shown to promote cartilage repair in an osteoarthritis rat model and was proposed as a possible therapy for osteoarthritis (Parrish et al., 2017). However, in the same study the unwanted formation of osteophytes was observed in GDF5-treated animals, which could be the result of the hypertrophic and osteogenic properties of GDF5. Interestingly, our results demonstrate that R399E can promote chondrogenesis of mesenchymal cells, stimulate cartilage matrix molecule expression in primary chondrocytes and simultaneously be less hypertrophic and osteogenic than GDF5. Consequently, R399E might be a better therapeutic candidate than GDF5 for cartilage repair.

In conclusion, we characterized the chondrogenic, hypertrophic and osteogenic profile of BMP2, GDF5 and three GDF5 mutants that have different affinities for BMPR1A and BMPR1B in vitro. We conclude that BMPR1A signaling is necessary for chondrogenesis and osteogenesis, and that BMPR1B does not play a role in osteogenesis. In addition, we postulate that stronger BMPR1B signaling compared to BMPR1A signaling could prevent chondrocyte hypertrophy and might act as a cartilage stabilizer during joint morphogenesis. Finally, because of its chondrogenic properties and low hypertrophic and osteogenic activities, we suggest that R399E is a promising growth factor for cartilage repair.

DNA coding for the mature parts of human GDF5 proteins was been isolated from human ROB-C26 osteoprogenitor cells (Yamaguchi and Kahn, 1991) via RT-PCR and subsequently ligated into prokaryotic plasmid vectors. Single mutations were introduced in residues in the binding site through which GDF5 binds to BMPR1A and BMPR1B (Schreuder et al., 2005) via site-directed mutagenesis and the proteins expressed in E. coli, isolated from inclusion bodies, renatured and purified, as described elsewhere (Ploeger and Wagner, 2017, United State Patent US9,718,697B2). The amino acid substitution consisted of replacing a hydrophobic amino acid with a hydrophilic/polar amino acid and vice versa, or replacing an amino acid with a different steric demand or a different charge. The resulting mutants (about 20) were profiled in several cell lines for their bioactivity and tested for their binding affinities to BMPR1A and BMPR1B (Ploeger and Wagner, 2017, United State Patent US9,718,867B2). The selection of the mutants used in this study was based on their increased preference to bind BMPR1B over BMPR1A compared to GDF5. The selected mutants are R399E (also named M1673 in other studies), W417F and W417R where the number refers to the position of the residue in the GDF5 wild-type sequence (accession number AAH32495) and the first and last letter correspond to the original amino acid and the mutated amino acid, respectively.

BMP2 was from R&D Systems (cat. no. 355-BM-500/CF, Minneapolis, MN, USA). All proteins were resuspended in 10 mM HCl.

A Biacore4000 surface plasmon resonance (SPR) instrument (GE Healthcare, Chicago, IL) was used. A total of 400 to 900 resonance units (RU) of the Fc-chimera of the ectodomain of the receptors (from R&D Systems; BMPR1A, cat. no. 2406-BR-100; BMPR1B, cat. no. 505-BR-100; BMPR2, cat. no. 811-BR-100) were immobilized on a Protein A-coated CM-5 sensorchip. GDF5, BMP2 or the GDF5 mutants were injected (ten steps of a 1-to-2 dilution series, the highest tested concentration was 200 nM) for 300 s at 30 µl/min. Sensorgrams were processed with the Biacore 4000 Evaluation Software version 1.1 (GE Healthcare). In detail, non-specific interactions of the test samples to the reference surfaces (Protein A), as well as buffer injections (double referenced), were subtracted from the recorded binding data. Processed data were analyzed using a simple 1:1 interaction model including a term for mass transport using numerical integration and nonlinear curve fitting to determine kinetic rate constants (k_on and k_off) and the dissociation constants (K_d). Each experiment was performed two times.

U2OS cells expressing BMPR2 together with BMPR1A (cat. no. 93-1006E3) or BMPR1B (cat. no. 93-1053C3) were acquired from Discoverx (Fermont, CA). This cell assay utilizes enzyme fragment complementation technology, where two fragments of the galactosidase are unified in a functional enzyme upon receptor dimerization and generate a luminescent signal. A dose–response curve was established for each ligand and tested according to the recommendation of the manufacturer. The EC₅₀ values were calculated with GraphPad Prism software (Version 7.0). Two experiments were performed (without duplicate for the BMPR2–BMPR1b dimerization assay and in duplicate for the BMPR2–BMPR1a dimerization assay) and the obtained EC₅₀ values were averaged.

C3H10T1/2 clone 8 cells were obtained from ATCC (cat. no. CCL226) and used at passage 8. To initiate chondrogenesis, 10⁶ cells/microtube were centrifuged (500 g for 5 min) to form a pellet, and these pellets were cultured in chondrogenic medium, composed of high-glucose DMEM (Gibco, Thermo Fisher Scientific, Waltham, MA, USA), 50 µg/ml ascorbate-2-phosphate (Sigma-Aldrich, St Louis, MO), 0.4 mM proline (Sigma-Aldrich), 1×10⁻⁷ M dexamethasone (Sigma-Aldrich), 0.1% (v/v) ITS (insulin, transferrin and sodium selenite supplement; Sigma-Aldrich) and 1% (v/v) 100× penicillin/streptomycin (Gibco).

To stimulate osteogenesis, cells were seeded at 100,000 cells/well in a 24-well plate and cultured for 2 days in high-glucose DMEM with 10% fetal bovine serum (FBS) and 1% penicillin/streptomycin. Osteogenesis was stimulated by further culturing the cells in osteogenic medium, composed of high-glucose DMEM, ascorbate-2-phosphate, 10 mM β-glycerol phosphate (Sigma-Aldrich), 1×10⁻⁸ M dexamethasone and 1% (v/v) 100× penicillin/streptomycin.

For both chondrogenesis and osteogenesis, cells were treated with 300 ng/ml GDF5, R399E, W417R, W417F or BMP2 or 12.5 µM HCl (negative control). Twice a week the medium was changed, and the compounds were freshly added. At day 0, 7, 14, 21 and 28, cells were harvested for gene expression, histology analysis (for chondrogenesis only) or Alizarin Red staining (for osteogenesis only). In addition, the medium was also harvested and analyzed for ALP. Two experiments were realized and for each experiment three (n=3) or five (n=5) cultures were run in parallel for chondrogenesis and osteogenesis, respectively. The results of one experiment are shown.

Porcine chondrocytes were isolated from the femoral head of an ∼1-year-old pig provided by a local slaughterhouse. The cartilage was firstly digested for 45 min with 0.25% collagenase (collagenase NBG4, Serva, Heidelberg, Germany) and subsequently overnight with 0.1% collagenase. The resulting cell suspension was filtered and washed, and 10⁶ cells/well were incubated in a low-binding 96-well plate in high-glucose DMEM with 10% FCS (Biochrom, Berlin, Germany), 0.4 mM proline, 50 µg/ml ascorbate-2-phosphate and 1% 100× penicillin/streptomycin. After 1 week, the resulting 3D constructs were transferred to a 24-well plate and cultured for an additional 4 weeks in the same medium supplemented with 300 ng/ml of GDF5, R399E, W417F, W417R or BMP2. The medium was changed twice a week. At the end of the culture, the cell constructs were homogenized in RLT buffer (from the RNeasy Mini Kit, Qiagen, Hilden, Germany) and processed for gene expression analysis. Three experiments were realized (only one with BMP2) and for each experiment three (n=3) cultures were run in parallel. The results of one experiment are shown.

Saos-2 cells (ATCC HTP-85) were seeded at 50,000 cells/well in a 24-well plate with a differentiation medium composed of high-glucose DMEM, 10% FBS, 50 µg/ml ascorbate-2-phosphate, 10 mM β-glycerol phosphate, 10⁻⁸ M dexamethasone and 1% 100× penicillin/streptomycin supplemented with 300 ng/ml of GDF5, R399E, W417F, W417R or BMP2. Alizarin Red staining was performed after 14 days of culture and the ALP concentration was measured in the medium at days 4, 7, 11 and 14. The medium was changed twice a week. Three experiments were realized and for each experiment three to six (n=3–6) cultures were run in parallel. The results of one experiment are shown.

Cells were stained directly in the culture plate. Cells were first washed with PBS and then fixed with 4% paraformaldehyde (w/v) for one hour at room temperature. The cells were washed twice with ddH₂O before being stained with 400 µl/well of 0.5% (w/v) Alizarin Red for 20 min at room temperature. Finally, cells were washed with ddH₂O until the water became clear and the cells were covered with PBS before being photographed.

For ALP measurement, 100 µl of medium was mixed with 100 µl p-nitrophenolphosphate (10 mg/ml). After 1 h of incubation, the absorbance was read at 405 nm using 490 nm as a reference wavelength and compared to that of a nitrophenol standard curve.

RLT buffer (from the RNeasy Mini Kit, Qiagen, Hilden, Germany) was added to the cells. For the pellets and chondrocyte 3D constructs, a proteinase K (Qiagen) digestion was performed for 10 min at 55°C before proceeding with the RNA isolation. RNA isolation was then performed with the RNeasy Mini Kit according to the recommendations of the manufacturer. mRNA concentration and quality were analyzed using an Agilent Bioanalyzer (Model 2100, Agilent, Santa Clara, CA) with an Agilent RNA 6000 Nano Chip G2938-80023.

Reverse transcription was performed using SuperScript III First-Strand Synthesis SuperMix (Invitrogen, Carlsbad, CA), followed by an RNase H treatment. Quantitative PCR was performed using the SYBR Green Jumpstart Taq Ready Mix (Sigma-Aldrich) with 200 nM of the reverse and forward primers (Eurofins Genomics, Ebersberg, Germany; see Table S2) and a Mx3000P Thermocycler (Agilent). Data were acquired and analyzed with the software MxPro v4.10 (Agilent).

Data was statistically analyzed with GraphPad Prism software (Version 7.0). One-way analysis of variance (ANOVA) was used with Dunnett's post-hoc test to correct for multiple comparisons. Statistical analyses were realized on replicates from parallel cultures.

We would like to thank Claudia Arras and Yvonne Wilhelm, who contributed to this work.

The peer review history is available online at https://jcs.biologists.org/lookup/doi/10.1242/jcs.246934.reviewer-comments.pdf