Novel RANKL DE‐loop mutants antagonize RANK‐mediated osteoclastogenesis

Bone is a dynamic tissue that is maintained by continuous renewal. An imbalance in bone resorption and bone formation can lead to a range of disorders, such as osteoporosis. The receptor activator of NF‐κB (RANK)–RANK‐ligand (RANKL) pathway plays a major role in bone remodeling. Here, we investigated the effect of mutations at position I248 in the DE‐loop of murine RANKL on the interaction of RANKL with RANK, and subsequent activation of osteoclastogenesis. Two single mutants, RANKL I248Y and I248K, were found to maintain binding and have the ability to reduce wild‐type RANKL‐induced osteoclastogenesis. The generation of RANK‐antagonists is a promising strategy for the exploration of new therapeutics against osteoporosis.


Introduction
Bone is a dynamic tissue that has multiple functions, including support of muscles, protection of vital organs, and fostering hematopoietic marrow [1,2]. Under normal conditions, bone homeostasis is achieved by continuous renewal, mediated by two processes: bone resorption by osteoclasts and bone formation by osteoblasts [3]. An imbalance in bone resorption and formation can cause bone-associated diseases: when the activity of osteoclasts exceeds that of osteoblasts, osteoporosis will be developed. In contrast, osteopetrosis will occur when osteoblastic activity exceeds [4].
Osteoclastogenesis is regulated by a cytokine system, consisting of three major proteins: receptor activator of nuclear factor-jB (RANK), RANK-ligand (RANKL), and osteoprotegerin (OPG) [5]. The binding between RANKL on the surface of osteoblast cells and the RANK on osteoclast precursor cells will trigger activation of several transcription factors and enzymes that induce osteoclast maturation and bone resorption. OPG is a soluble decoy receptor, acting as a natural antagonist of RANKL [6]. The RANK-RANKL pathway plays an important role in both physiological and pathological bone development. Therefore, this signaling pathway is a promising target in bone-related diseases [3].
Receptor activator of nuclear factor-jB-ligand is a member of the tumor necrosis factor (TNF)-superfamily, a group of cytokines involved in cell proliferation and cell death [7]. The TNF-superfamily consists of 19 multimeric ligands interacting with cognate receptor molecules; most of them require trimerization to initiate their signaling cascade [8]. Ligands belonging to the TNF-superfamily are mostly type II transmembrane glycoproteins, containing a C-terminal, receptor-interacting ectodomain, a transmembrane domain, and an N-terminal intracellular tail [4]; the extracellular domain can be either cleaved off by the proteolytic activity of metalloproteases or produced by alternative splicing [9,10].
Three-dimensional structures of TNF-a, TRAIL, and RANKL alone and in complex with their respective receptors have revealed very similar overall structures that comprise unique conserved elements involved in receptor binding [11][12][13][14][15][16][17][18]. One of these elements is the so-called DE-loop. There is a conserved tyrosine in the DE-loop of all TNF-ligands, with the exception of an arginine in CD40L and an isoleucine in RANKL. Interestingly, variants of human TNF-a with mutated DE-loop residues including the semiconserved residue Y87 were shown to be receptor antagonists: some mutants selectively bound to human TNF-receptor 1 (TNF-R1) and not TNF-receptor 2 (TNF-R2), while inhibiting wild-type TNF-a-mediated effects via TNF-R 1 [13]. Comparing the 3D structure of the antagonistic TNF-R1-selective mutant with that of wild-type TNF-a, it was concluded that the Y87H mutation changed the binding mode of the DE-loop to TNF-R1 from a hydrophobic to an electrostatic interaction [13]. In murine RANKL, mutation of the equivalent residue I248 to aspartate resulted in an eightfold lower ED 50 [12]. In contrast, another 3D structure has led to the suggestion that the DE-loop may not be critical for mRANKL-mRANK binding [11,19].
Here we have investigated the effect of mutations of the I248 residue in the mRANKL DE-loop on the binding to mRANK and on subsequent osteoclastogenesis. Our results show that RANKL mutants I248Y and I248K have the ability to reduce RANKL-induced osteoclastogenesis. Thus, mutagenesis at position I248 of the DE-loop region can form the basis of a general strategy to create RANK-antagonists for the RANKL-RANK pathway, paving the way to novel osteoporosis treatments.

Results
In silico mutagenesis of mRANKL demonstrates importance of position I248 Several 3D structures of mRANKL have been described previously [4,11,12,20,21]. The 3D structure of mRANK in complex with mRANKL (Protein Data Bank accession code 4GIQ) was used for in silico mutagenesis at position I248 in order to estimate differences in DDG i binding to RANK between RANKL mutants and wild-type RANKL (WT RANKL) (Fig. 1A,B). A similar strategy has allowed us in the past to successfully generate variants of hTRAIL with desired properties, such as enhancement of affinity and change of receptor selectivity [22][23][24]. As shown in Fig. 1B, the RANK residues surrounding I248 of RANKL are hydrophilic, including D49, T50, N52, E53, and E54, suggesting that the surface interaction between the DE-loop of RANKL and RANK is predominantly hydrophilic. This is different from the binding between TNF and TNF-R1, where Y87 of TNF is buried in a hydrophobic 'pocket' of the TNF-R1. The importance of RANKL residue I248 for the interaction with RANK was investigated by in silico mutagenesis. The effects of variation at position 248 of RANKL on the calculated DDG i are shown in Fig. 1C. Overall, our data indicate that substitution by other amino acids results in a strong effect on binding of RANKL to RANK, either positive or negative, suggesting that this residue is indeed important for binding. Given the overall estimated effects on binding energies between RANKL and RANK, we decided to focus on and construct a subset of RANKL mutants. I248Y was chosen because of the conserved tyrosine at the same position in the DE-loop of TNF-ligands and its calculated large difference in DDG i (Fig. 1C); I248K because this mutation also has a predicted large effect on the DDG i , plus it is homologous to the arginine found at this position in TNF-ligand family member CD40L; and finally I248D was chosen as it was previously reported to have an eightfold lower ED 50 compared to WT RANKL [12]. As the interaction between the DE-loop of RANKL and RANK is mainly hydrophilic, all the hydrophobic mutations including I248W were not taken into consideration.

RANKL mutations I248Y and I248K show increased binding affinity to RANK
After expression, purification, and characterization of WT RANKL and mutants, we determined their affinities to RANK by surface plasmon resonance (SPR). Using a CM4 sensor chip and a low density of RANK-Fc [< 60 response units (RU)], we were able to measure the interaction between one trimeric RANKL molecule binding to one RANK-Fc monomer [8]. Indeed, the sensorgrams for WT RANKL, I248Y, I248K, and I248D all show a~1 : 1 binding ratio (varying between 0.9 and 1.2) (Fig. 2). Hence, the data could be fitted with a 1 : 1 Langmuir fitting model and the kinetic parameters were calculated [8]. As shown in Table 1, I248Y and I248K showed approximately three-to fourfold higher affinities to the target receptor in comparison to WT RANKL. This was mainly caused by an increase in the association rate constant (k a ). The dissociation rate constants (k d ) of I248Y and I248K were comparable to WT RANKL. Mutant I248D showed a lower affinity to RANK-Fc compared to WT RANKL, which is mainly due to a much higher dissociation rate. Taken together, these results are consistent with the previous in silico predictions and they further confirm that the I248 residue indeed plays an important role in RANKL-RANK binding.
RANKL mutants compete for binding of WT RANKL to RANK and form high-order complexes with their receptor To further assess whether the mutants compete with WT RANKL for binding, we performed a competitive ELISA assay, in which wells coated with RANK-Fc were incubated with both WT RANKL and RANKL variants. Our competitive ELISA results (Fig. 3) showed that increasing concentrations of both I248K and I248Y can efficiently compete with WT RANKL for RANK-Fc binding at physiological concentrations. However, RANKL I248D could only partially compete with WT RANKL at concentrations over 1000 nM, which is in accordance with the lowered affinity of RANKL I248D to RANK-Fc compared to WT RANKL. Given that signaling of several TNF cytokines has been shown to require receptor oligomerization, we also tested whether mutants I248Y, I248K, and I248D could form multimeric complexes with RANK using high densities of RANK-Fc in SPR. As shown in Fig. 4, using the number of RU of bound RANK-Fc, we could estimate the ratio of complex formation (the number of trimeric RANKL units bound to monomeric RANK-Fc) using equation 1 (see Experimental Procedures). A binding ratio of 1 : 3.07 AE 0.01 was found for WT RANKL. For RANKL mutants I248Y, I248K, and I248D, mean binding ratios were 1 : 3.03 AE 0.04, 1 : 2.99 AE 0.02, and 1 : 3.12 AE 0.01, respectively. These results imply that the mutants, as WT RANKL, form a trimer-trimer complex with the RANK receptor.

RANKL mutants antagonize the biological activity of WT RANKL
We next tested the effect of RANKL mutants I248D, I248K, and I248Y on osteoclastogenesis by assessing their ability to induce multinucleation (three or more nuclei per cell) in osteoclast precursor murine RAW 264.7 cells and by testing their capability to inhibit WT RANKL (Fig. 5A, and Tables S1 and S2). Microscope photographs of osteoclasts from different treatments with RANKL variants (50 ngÁmL À1 ) only, WT RANKL (50 ngÁmL À1 ), or a combination of both are shown in Fig. 5A. Single treatment with WT RANKL caused osteoclast formation to a large extent and induced formation of giant multinucleated cells (two or more clusters of multinuclei per cell) which can cover large areas of the wells. RANKL variants themselves were also capable of inducing osteoclast formation, although the number of osteoclasts formed was lower than that induced by WT RANKL. Competition of either RANKL I248Y or I248K with 50 ngÁmL À1 WT RANKL led to a reduction in both number and size of osteoclasts, while the competitive effect of RANKL I248D was less than that of I248Y and I248K. Importantly, single treatment with each of the three mutants induced less osteoclastogenesis compared to treatment with WT RANKL at all concentrations measured ( Fig. 5B-D). The high-affinity mutants I248Y and I248K reached a maximum level at approximately 50 ngÁmL À1 , which is 70 and 80% of that of WT RANKL, respectively (Fig. 5B,C). In contrast, the low-affinity mutant I248D only reached up to 40% of the osteoclastogenic effect of WT RANKL even at the highest concentration of 150 ngÁmL À1 (Fig. 5D). Cotreatment of RAW 264.7 cells with WT RANKL and high-affinity mutants I248Y or I248K led to a reduction in the level of osteoclastogenesis by 30-40% already at low concentration of mutant RANKL, with  Table 1. Binding kinetics of the RANKL variants and WT RANKL to murine RANK-Fc as measured by surface plasmon resonance. no further reduction at higher concentrations (Fig. 5B, C). In contrast, the low-affinity mutant I248D only showed a reduction of the osteoclastic potential of WT RANKL at high concentrations (> 50 ngÁmL À1 ) (Fig. 5D).

Does exchange of RANKL subunits play a role in inhibition of WT RANKL-induced osteoclastogenesis?
In our studies, we showed inhibition of WT RANKLinduced osteoclastogenesis in the murine RAW 264.7 cell line by RANKL variants I248K and I248Y (Fig. 5). Although they themselves can also cause osteoclast formation, the numbers of osteoclasts are lower than those of WT RANKL (50 ngÁmL À1 ). The increased affinities of these mutants can explain the results of the inhibition experiments: both mutants showed higher affinities to RANK-Fc than WT RANKL (Fig. 2), which can make them compete efficiently with WT RANKL to occupy RANK and induce less osteoclastogenesis. An alternative explanation may be that the subunits of RANKL are interchanged, allowing formation of heterotrimeric RANKL molecules that are no longer able to activate RANK. To confirm or refute this hypothesis, we mixed purified sfGFP-WT RANKL with WT RANKL at equimolar ratio and incubated this for either 24 h at 37°C or 44 h at 4°C. The protein samples were loaded onto a size exclusion column while the absorbance was followed at 280 and 475/488 nm (Fig. 6). The trimeric fusion protein sfGFP-WT RANKL with its higher molecular weight eluted at approximately 12 mL showing absorbance at both wavelengths, whereas WT RANKL eluted at approximately 14.8 mL, showing only absorbance at 280 nm. Mixing of the two trimeric proteins did not result in shifting peaks, demonstrating that no exchange occurred during the incubation (Fig. 6).

Discussion
Bone homeostasis is achieved by continuously renewing the tissue [6]. Loss of bone remodeling homeostasis can lead to a range of disorders, including osteoporosis [6], bone lesions associated with rheumatoid arthritis [25], Paget's disease [26], and malignancy-induced bone diseases [27]. Due to the important role of the RANK-RANKL pathway in the bone remodeling process [5], interfering with their interaction is a promising strategy for therapeutic intervention. Initially, OPG-Fc was developed as a RANKL scavenger [28]; as OPG cross reacts with TRAIL, recently OPG variants lacking TRAIL binding were developed, showing significantly reduced osteoclastogenesis [29]. RANKL-targeted peptides were also developed and confirmed to prevent bone loss and inflammation in rheumatoid arthritis in vivo [19,30]. Finally, denosumab, an FDA-approved RANKL-specific antibody, was developed and is currently used in the treatment of osteoporosis and bone metastases [19]. Denosumab inhibits osteoclastogenesis by scavenging RANKL thus inhibiting receptor interaction. In contrast, we focused on novel RANKL variants, which will be of interest as these are blocking the receptor directly. Three-dimensional structures of TNF-TNF-R1 and TRAIL-DR5 complexes show a conserved tyrosine residue in the DE-loop of ligands, which induces a hydrophobic interaction with loop 1 of the receptors [15,31]. Mukai et al. [32] created a series of receptorselective TNF-a mutants with full bioactivity by using the phage display technique; the results showed that all active TNF-R1-selective mutants retained Y87, suggesting that Y87 is an essential residue for receptor binding. Furthermore, TNF-a mutation Y87H was generated as a part of a TNF-R1-selective antagonist, and Y87H was found to change the binding mode of the DE-loop from a hydrophobic to an electrostatic interaction [13]. In silico calculation of the effects of mutations of RANKL residue I248, which is homologous to TNF-a Y87, on the interaction with its receptor RANK, suggested that mutagenesis of the I248 residue of RANKL has strong effect on binding of RANKL to RANK.
We have focused on three mutations on position 248: I248Y, I248K, and I248D. Detailed analysis of the interaction of RANKL proteins with RANK using SPR showed an increased affinity for mutants I248K and I248Y and a decreased affinity for I248D ( Fig. 2 and Table 1). In accordance, competitive ELISA confirmed that RANKL I248Y and I248K can efficiently compete with WT RANKL for binding to RANK, whereas I248D cannot (Fig. 3). In line with these results, structural analysis of the impact of these mutations shows that the positively charged lysine on position 248 may interact with the negatively charged E54 residue on the RANK surface. Mutation I248Y could also result in an extra hydrogen bond and strengthened van der Waals forces, also reinforcing the binding of RANKL and RANK. The lower affinity for RANK that we measured for I248D can be explained by an electrostatic repulsion with E54 residue in RANK.
High-affinity mutants RANKL I248Y and I248K are still able to induce osteoclastogenesis but to a 20-30% lower maximum than WT RANKL, whereas low-affinity mutant I248D does not even reach 40% of this maximum activation level even at 150 ngÁmL À1 (Fig. 5). Interestingly, the high-affinity mutants were able to reduce WT RANKL-mediated osteoclastogenesis already at low concentrations, whereas I248D shows a dose-dependent reduction of the WT RANKL signal up to the concentration of 150 ngÁmL À1 (Fig. 5). All three mutants showed higher association rate constants (Table 1) allowing them to bind faster to the receptor than WT RANKL. Although high-affinity mutants I248K and I248Y have similar dissociation rate constants as WT RANKL, lowaffinity mutant I248D has a higher dissociation rate constant ( Table 1). The latter one leads to a less stable complex with RANK, which can explain why this mutant in itself is not very biologically active and why it reduces WT RANKL activity only at high concentrations (Fig. 5D). A similar explanation was formulated for antagonistic TNF-a mutation Y87H [13]. Our RANKL mutants can reduce WT RANKL activity by competing in binding to RANK, resulting in a reduction in biological activity to the level of the mutants themselves. What could be an explanation for this lower biological activity? In a molecular dynamics study on binding between TNF-ligand family member TRAIL and its death receptor DR5, Wassenaar et al. [33] proposed that through binding to TRAIL the conformational freedom of extracellular binding domains of DR5 was strongly restricted; this might also influence the stability of the intracellular death complex. Furthermore, Neumann et al. [34] showed that the signaling capabilities of the death receptors DR4 and DR5 do not only depend on binding of TRAIL but the transmembrane domains together with their adjacent stalk regions also control the signaling strength. Likewise, the transmembrane domain and adjacent extracellular stalk regions of RANK could play a similar role. RANKL may also play a role in controlling the conformational freedom of the extracellular domains of RANK. Mutants RANKL I248Y and I248K may slightly change the conformation of the RANKL-RANK complex, thus influencing the conformational freedom of RANK extracellular domains and subsequently the orientation of intracellular domains, which further influences the recruitment of downstream effector proteins.
In conclusion, our data show that the DE-loop is of importance for osteoclastogenesis and can be explored for the creation of new ligands with inhibitory properties. Given the importance of the DE-loop for RANK-RANKL signaling, the newly created mutants I248Y and I248K can form a starting point for novel therapeutic proteins in new osteoporosis treatments.

Experimental procedures
In silico calculations of the interaction energy of RANK in complexes with RANKL I248 mutants Simulations were performed using the available 3D structure of murine RANKL in complex with RANK (Protein Data Bank accession code 4GIQ) [4]. All calculations and predictions were performed using BIOVIA Discovery studio 4.5. Briefly, the structures of RANKL and the I248 mutants were refined using the CHARMm force field. For the minimization of the receptor-ligand complex, the Smart Minimizer was used with maximum steps of 2000 and RMS gradient of 0.1, and Generalized Born with molecular volume (GBMW) was used as implicit solvation model. Predicted differences in RANK-binding energy (DDG i ) of the RANKL I248 mutants compared to WT RANKL were determined through interaction energy calculation. The change in energy was calculated in kcalÁmol À1 and applies to a single binding interface. A negative DDG i value indicates an increase in receptor binding energy, whereas a positive DDG i value indicates a decrease in receptor binding energy.
Site-directed mutagenesis, production, and purification of the RANKL variants cDNA encoding murine soluble RANKL (aa 160-316) was introduced in a pET15b expression plasmid (Novagen, Darmstadt, Germany) using the restriction sites NcoI and BamHI (Fermentas, St. Leon, Lithuania). Mutations at position 248 were introduced using megaprimers [35]. The PCR product was digested using DpnI (Fermentas) to remove template DNA and used for transformation of Escherichia coli DH5a cells. Plasmid DNA was purified from an overnight culture (Nucleospin, Macherey-Nagel, Du¨ren, Germany). After confirming the introduction of mutations by DNA sequencing, the plasmids were used for transformation of E. coli BL21(DE3). Transformants selected using ampicillin (Duchefa, Haarlem, the Netherlands) were grown overnight at 37°C in 1 x LB medium (Tryptone, NaCl, and yeast extract all from Duchefa) containing 100 lgÁmL À1 ampicillin. This culture was used to inoculate a larger culture (1 : 100 dilution), which was grown for approximately 2 h to OD 600 0.5, after which protein production was initiated by adding 1 mM IPTG (Duchefa), and continued for 16 h at 20°C.
After harvesting cells by centrifugation and resuspending at 3 mLÁg À1 wet cells in buffer A containing 50 mM MES pH 5.8 (Duchefa), 10% v/v glycerol (Duchefa), and 2 mM dithiothreitol (Sigma, Saint Louis, MO, USA), the cells were lysed by sonication. The RANKL proteins were mainly soluble. Supernatants were firstly loaded on a 5 mL cation exchange column (SP column, GE Healthcare, Uppsala, Sweden) using buffer A. Buffer B, similar to A but containing 2 M NaCl (Duchefa), was used to elute the protein. Elution of the fractions containing RANKL occurred with 300-400 mM NaCl. The presence of proteins was confirmed by western blotting using an anti-mRANKL antibody (R&D Systems, Minneapolis, MN, USA) and corresponding anti-goat-horse radish peroxidase (HRP) antibody (Biosource, Waltham, UK).
The obtained protein-containing fractions were pooled and loaded onto a HiLoad Superdex75 16/60 size exclusion chromatography column (GE Healthcare), using buffer A. The retention volumes of the RANKL proteins corresponded to the hydrodynamic radius of a trimeric complex. The proteins were checked by SDS/PAGE to determine the purity which was estimated to be more than 90% pure. Concentrations were determined with the Coomassie Bradford protein assay kit (Pierce, Rockford, IL, USA) using bovine serum albumin (BSA, ThermoScientific, Rockford, IL, USA) as a standard. The final yield of the variants was approximately 2 mg purified protein per liter culture. When calculating molarity, we used the molecular weight of trimeric RANKL proteins.
Superfolder green fluorescent protein was amplified from pQE30_sfGFP [36] by PCR, digested with NcoI/NcoI site and ligated into the vector pET15b_WT RANKL to encode sfGFP fused to the N terminus of WT RANKL. sfGFP-WT RANKL was produced in BL21(DE3) cells similar to our variants. For the purification of sfGFP-WT RANKL, cleared cell lysate was firstly loaded onto a 5 mL anion exchange column (Q column, GE Healthcare), using a buffer of 20 mM Tris pH 8.5, 2 mM dithiothreitol, and 10% v/v glycerol. Buffer B, similar to A but containing 2 M NaCl, was used to elute the protein. Elution of the fractions containing sfGFP-WT RANKL occurred at 300-500 mM NaCl. The obtained protein-containing fractions were pooled and loaded onto a 5 mL cation exchange column, followed by a HiLoad Superdex200 16/60 size exclusion chromatography column (GE Healthcare), and the purification steps were similar with that of WT RANKL.

Kinetic analysis of complex formation between trimeric RANKL and monomeric RANK-Fc
Binding between RANKL and RANK-Fc was determined using a Biacore 3000 system (GE Healthcare). A CM4 chip was directly coated with 70 lgÁmL À1 protein A (Sigma), making use of the amine coupling kit (GE Healthcare) according to the manufacturer's instructions. The amount of immobilized protein A was typically around 1000 RU. Next, murine RANK-Fc (R&D Systems) was captured at a low density, never exceeding 60 RU. The flow rate used for immobilization was 50 lLÁmin À1 . RANKL was injected at a flow rate of 50 lLÁmin À1 in concentrations between 0.01 nM and 20.48 nM for 3 min. Dissociation of the formed complex was followed up for 1000 s. HBS-N buffer (GE Healthcare) with 0.005% v/v surfactant P20 (GE Healthcare, Uppsala, Sweden) was used as a running buffer. The chip surface was regenerated between cycles with two injections of 10 mM glycine (Duchefa) pH 1.5 for 30 s each. The responses were corrected for buffer effect and binding to the protein A chip surface [8]. At low receptor densities, the data could be fitted to a 1 : 1 Langmuir model using the BIAEVALUATION software version 4.2 (GE Healthcare) [8].

Determination of trimer-trimer complexes using SPR
To confirm whether a trimer-trimer complex could be formed between RANKL and RANK, an SPR experiment was set up as previously described [8]. RANK-Fc was captured at high density, > 1000 RU, on the surface of a CM5 chip by immobilized protein A. RANKL was injected at a high concentration of 5000 nM with a flow rate of 10 lLÁmin À1 . After each cycle, the chip surface was regenerated as described above. With equation 1, the ratio of the complex formation, n, can be calculated by measuring the maximal response (R max ) that can be reached by formation of the complex AB n between trimeric RANKL (A; with molecular weight MW A = 52.7 kDa) and monomeric RANK-Fc (B; with molecular weight MW B = 60 kDa) captured at density RU captured [8].
Competitive ELISA at concentrations in the range of 0-8000 nM, mixed with 10 nM sfGFP-WT RANKL. After washing with TBST buffer, bound sfGFP-WT RANKL was detected by incubation with a murine anti-GFP antibody (Sigma) and a secondary anti-mouse HRP-conjugated antibody (Millipore, Billerica, MA, USA). The signal was quantified using the One-step Turbo TMB reagent (ThermoScientific) and absorbance was measured at 450 nm. The percentage of sfGFP-WT RANKL bound was measured relative to the binding of 10 nM sfGFP-WT RANKL alone.

Inhibition of osteoclastogenesis
Cells from the murine leukemic monocyte macrophage cell line RAW 264.7 were cultured as described before [37]. Cells were subcultured every third or fourth day using a 1 : 10 dilution in fresh Dulbecco's modified Eagle's medium (Life Technologies, Carlsbad, CA, USA) containing 10% v/v fetal calf serum (FCS, Life Technologies) and 2 mM penicillin/streptomycin (Life Technologies). Cells were seeded at a density of 1000 cells per well in a 96-well plate (Greiner). At day 3, the medium was changed to Alpha-MEM (Invitrogen, Carlsbad, CA, USA) containing 2 mM penicillin/streptomycin and 10% v/v FCS (Invitrogen); different concentrations of RANKL were added, ranging from 3.1 to 150 ngÁmL À1 . Inhibition of WT RANKL-induced osteoclastogenesis by RANKL mutants was assessed via treatment of RAW 264.7 cells with 50 ngÁmL À1 WT RANKL and a concentration range (3.1-150 ngÁmL À1 ) of a determined RANKL mutant. At day 5, the medium was exchanged, and both WT RANKL and RANKL mutants were added using the same concentrations as described for day 3. At day 7, osteoclastogenesis was determined using the tartrate resistance acid phosphatase (TRAP) assay (Sigma) according to the manufacturer's instructions. The number of osteoclasts in the wells containing 50 ngÁmL À1 WT RANKL was set to 100%, and the number of osteoclasts in all other wells was calculated relative to this reference well. Multinucleated (three or more nuclei) TRAP-positive cells were considered as osteoclasts. Osteoclasts with two or more clusters of multinuclei were considered as giant osteoclasts.

Determination of exchange of RANKL subunits
The exchange of RANKL subunits was determined using size exclusion chromatography. WT RANKL and sfGFP-WT RANKL at concentrations of 12 lM were incubated

Supporting information
Additional Supporting Information may be found online in the supporting information tab for this article: Table S1. Absolute numbers of normal and giant osteoclasts per RANKL treatment (experiment 1). Table S2. Absolute numbers of normal and giant osteoclasts per RANKL treatment (experiment 2).