Structure-based development of a receptor activator of nuclear factor-κB ligand (RANKL) inhibitor peptide and molecular basis for osteopetrosis

Abstract

The receptor activator of nuclear factor-κB (RANK) and its ligand RANKL, which belong to the tumor necrosis factor (TNF) receptor-ligand family, mediate osteoclastogenesis. The crystal structure of the RANKL ectodomain (eRANKL) in complex with the RANK ectodomain (eRANK) combined with biochemical assays of RANK mutants indicated that three RANK loops (Loop1, Loop2, and Loop3) bind to the interface of a trimeric eRANKL. Loop3 is particularly notable in that it is structurally distinctive from other TNF-family receptors and forms extensive contacts with RANKL. The disulfide bond (C125-C127) at the tip of Loop3 is important for determining the unique topology of Loop3, and docking E126 close to RANKL, which was supported by the inability of C127A or E126A mutants of RANK to bind to RANKL. Inhibitory activity of RANK mutants, which contain loops of osteoprotegerin (OPG), a soluble decoy receptor to RANKL, confirmed that OPG shares the similar binding mode with RANK and OPG. Loop3 plays a key role in RANKL binding. Peptide inhibitors designed to mimic Loop3 blocked the RANKL-induced differentiation of osteoclast precursors, suggesting that they could be developed as therapeutic agents for the treatment of osteoporosis and bone-related diseases. Furthermore, some of the RANK mutations associated with autosomal recessive osteopetrosis (ARO) resulted in reduced RANKL-binding activity and failure to induce osteoclastogenesis. These results, together with structural interpretation of eRANK-eRANKL interaction, provided molecular understanding for pathogenesis of ARO.

Bone is a dynamic organ that is maintained by a balance between bone resorption by osteoclasts and bone formation by osteoblasts. The interaction between receptor activator of nuclear factor-κB ligand (RANKL) on osteoblast/stromal cells and the RANK receptor on osteoclast precursors results in the maturation of osteoclasts and subsequent bone resorption (1–4). Osteoprotegerin (OPG) functions as a soluble decoy receptor to RANKL and competes with RANK for RANKL binding. Accordingly, OPG has been shown to be an effective inhibitor of maturation and activation of osteoclasts in vitro and in vivo (5, 6). The ratio between RANKL and OPG elegantly regulates the orientation of bone metabolism to either bone formation or resorption; therefore, dysregulation of this ratio causes an imbalance between bone formation and resorption and results in bone diseases such as osteoporosis, rheumatoid arthritis, and osteolytic bone metastasis (7–10). For the same reasons, mutations in RANK, OPG, or RANKL are associated with genetic skeletal abnormalities such as autosomal recessive osteopetrosis (ARO) (11, 12). Because of the critical roles of RANKL/OPG/RANK proteins in bone metabolism, their interaction and RANK signaling are considered promising targets for the control of bone metabolic diseases (7). Consequently, RANK-Fc, Fc-OPG, and anti-RANKL antibodies have been developed as therapeutics for osteoporosis (13–19). Alternatively, peptide mimics of OPG (OP3-4 peptide) (20, 21) and the tumor necrosis factor (TNF) receptor (WP9QY peptide) (22) were also developed and showed inhibitory activity against the RANKL-induced osteoclastogenesis.

The RANKL-RANK complex belongs to the TNF ligand–receptor superfamily, whose members share a similar binding mode despite low sequence homology: The receptors bind to a groove at the junction of monomers in the trimeric ligand that is formed by edge-to-face packing of monomeric subunits (23–27). However, the key structural features in the binding interface that control the biological specificity of a particular ligand–receptor pair have not been defined. For example, the binding mode between RANKL and RANK is not yet clearly understood, although the crystal structure of RANKL was extensively characterized (28, 29).

We sought to identify structural determinants that govern the specific ligand–receptor recognition of RANKL-RANK and, thus, to provide a molecular foundation for further investigation of bone-related diseases and development of previously undescribed pharmaceuticals. In this study, based on crystal structure of the ectodomain of mouse RANKL (eRANKL) complexed with the ectodomain of RANK (eRANK) at 2.5-Å resolution and the biochemical and functional characterization of eRANK mutants, we identified the key structural determinants governing the recognition specificity of eRANK and developed potential inhibitors of RANK-RANKL interaction through structure-based approaches. Furthermore we were able to explain the molecular basis for mutations associated with ARO.

Results

Overall Structure of the eRANK-eRANKL Complex.

The complex, with approximate dimensions of 60 Å × 70 Å × 100 Å, contains three eRANK molecules with four full cysteine-rich domains (CRDs) inserted into three crevices on the subunit interfaces of a trimeric eRANKL (Fig. 1A and Fig. S1A). Binding of eRANK induces local conformational changes in eRANKL near the AA″, CD, and EF loops, thereby disordering the N terminus of strand D, and the C terminus of strand E (Fig. S1B). eRANK contains four CRDs (Fig. 1B and Figs. S2 and S3) and shows some structural features distinct from other canonical receptors of the TNF family (23–27). Each CRD typically has six conserved Cys residues that form three disulfide pairs, but the disulfide bond between the third and fifth Cys residues is missing in CRD2, CRD3, and CRD4 of RANK (Fig. S3). Because disulfide bonds are essential for the overall fold of CRDs, the absence of a disulfide bond is likely to cause conformational diversity in eRANK. In fact, the structure of CRD3 of eRANK was the most divergent when compared with other TNF receptors (Fig. S4). However, CRD2 shows little structural heterogeneity (Fig. S4), probably because the hydrogen bonds of His90-Asn106, His90-Arg111, and Lys91-Arg111 functionally replace the missing disulfide bond (Fig. S3). Another feature of CRD3 is the presence of a noncanonical disulfide bond, Cys125-Cys127 (Fig. S3), which appears to play a role in RANKL binding and recognition, whereas other disulfide bonds mainly stabilize the overall fold.

Fig. 1.

Overall structure of eRANK-eRANKL complex. (A) Ribbon diagram of eRANK-eRANKL complex. Three eRANKs (pale cyan, pale green, and light blue) bind to a trimeric eRANKL (light orange, green cyan, and lime). The residues involved in ligand–receptor interaction, depicted as ball-and-stick models, are red (Loop1), green (Loop2), and blue (Loop3). N and C termini and β-strands of eRANKL are labeled. (B) Overall structure of eRANK in ribbon diagram. The four CRDs are pink, forest, marine, and magenta, and conserved disulfide bonds are in yellow ball-and-stick models. The disulfide bond between Cys125 and Cys127 is in a red ball-and-stick model. (C) Surface presentations of the binding interfaces of eRANK monomer (Left) and eRANKL dimer (Right). The eRANK residues in binding loops are red (Loop1), green, (Loop2), and blue (Loop3). Counterpart residues in RANKL are in the same color scheme. (D) Stereo presentation of the binding mode of Loop3 to RANKL. eRANKL and eRANK are green and pink, respectively. The residues involved in ligand–receptor binding are drawn as ball-and-stick models and labeled. Blue dashed lines represent hydrogen or ionic bonds, and important strands or loops are labeled. Cys125-Cys127 and waters are depicted as yellow and red ball-and-stick models, respectively.

Receptor-Ligand Interface.

In the eRANK-eRANKL complex, Loop1 and Loop2 in CRD2 and Loop3 in CRD3 fit into pockets of RANKL with perfect geometric and electrostatic complementarity, accounting for the recognition specificity (Fig. 1C and Fig. S5A). And 7680 Å² of the solvent-accessible area that is 15.4% of the total surface area of uncomplexed molecules is buried in the complex. The wide electrostatic network formed by charged residues appeared to be pivotal for specific interaction (Fig. S5A). The detailed atomic interactions between each loop and RANKL are described in Table S1 and Fig. S5.

Among the three RANKL-binding loops, Loop3 consisting of residues 119–130, provides the largest binding surface area (Fig. 1D). Glu126 in Loop3 is particularly notable due to its extensive interaction with RANKL. Glu126 not only interacts with Lys180 of RANKL, but forms bidentate water-mediated hydrogen bonds with Lys256, Asp301, and Asp303 on one side, and Asn253 and Gln291 on the other. It is also in van der Waals contact with Tyr240. It is unique that RANK has a noncanonical disulfide bridge (Cys125-Cys127) at the tip of Loop3, which most likely determines the shape of the loop and helps to orient Glu126 for extensive interactions with RANKL (Fig. 1D). The C127A replacement abolished eRANKL binding activity (Fig. S6), indicating the importance of this noncanonical disulfide bond for RANKL binding.

To confirm the role of each RANKL-binding loop, key residues that are involved in RANKL binding through ionic and hydrogen bonds (D85 in Loop1, K97 in Loop2, and E126 in Loop3) were replaced with Ala. Isothermal titration calorimetry (ITC) showed that the wild-type eRANK has strong affinity for eRANKL with K_d = 230 nM; however, no interaction was observed for E126A eRANK (Fig. 2A). E126A eRANK was not able to suppress eRANKL-induced osteoclast differentiation, whereas wild-type eRANK completely inhibited osteoclastogenesis (Fig. 2B). Loop2 contributes only 15% of the total buried surface area, but appears to exert a substantial effect on RANKL binding, because the K97A mutation completely abrogated eRANKL binding (Fig. 2A) and osteoclastogenesis inhibition (Fig. 2B). D85, a key residue in Loop1, did not appear to be as important as K97 in Loop2 or E126 in Loop3, because D85A eRANK retained marginal binding affinity (K_d = 24.2 μM) and inhibitory effects (Fig. 2).

Fig. 2.

Functional significance of residues involved in RANKL binding. (A) ITC analyses of eRANKL binding to the wild-type (eRANK-WT) and mutant eRANKs (D85A, K97A, and E126A). The raw data (Upper) and the best fit (Lower) are shown, with the dissociation constants (K_d). (B) Inhibitory effects of wild-type and mutant eRANKs on the eRANKL-induced osteoclast differentiation were estimated by the number of TRAP-positive multinucleated cells. Concentrations of eRANKs are indicated.

Binding Mode of OPG.

OPG is known as a decoy receptor to RANKL, and its binding affinity for RANKL is comparable to that of RANK (5, 6). Therefore, we hypothesized that OPG and RANK might have similar binding modes and modeled the OPG-RANKL complex (SI Text) to verify this hypothesis. In the modeling, OPG Loop3 was shown to fit into RANKL pocket (Fig. S7), and it is possible that E116 in Loop3 substitutes for E126 of RANK (Fig. 3A). We also created chimeric RANKs, which contain OPG loops (Fig. S2). The substitution of Loop3 (eRANK-Loop3_OPG) or all three loops (eRANK-Loop123_OPG) did not result in substantial changes in the inhibitory activity of eRANK (Fig. 3B), suggesting that binding loops of RANK and OPG may function in similar modes, and be functionally exchangeable.

Fig. 3.

The modeling of OPG and inhibitory activities of chimeric RANKs. (A) Ribbon diagrams of the ligand-binding domains (CRD2 and ½ CRD3) of RANK, and the enlarged view of Loop3 (yellow) superposed onto DR5 (magenta) and OPG (orange). Cys125-Cys127 in RANK, and Glu residues in RANK (Glu126), DR5 (Glu151), and OPG (Glu116) are shown in ball-and-stick models and labeled. (B) Functional analyses of wild-type (eRANK-wt) and chimeric eRANKs (Loop3_OPG and Loop123_OPG). Their inhibitory effects on the eRANKL-induced osteoclast differentiation were monitored by counting multinucleated TRAP-expressing cells. Concentrations of eRANKs are indicated.

Peptide Inhibitors Derived from RANK Loops.

The crystal structure of the eRANK-eRANKL complex and the biochemical characterization of RANK mutants proposed that Loop2 and Loop3 are the main determinants of the receptor–ligand interaction. To confirm this notion, synthetic peptides harboring the key residues in Loop2 and Loop3 were designed (Table 1), and their inhibitory effects on the eRANKL-induced osteoclastogenesis were examined. A Cys or Tyr residue was introduced for cyclization, so that the peptides have overall shapes similar to the binding loops in the complex. For example, Val100, which was located most proximal to Cys93 in the crystal structure of the complex, was replaced with Cys in order for the formation of a disulfide linkage with Cys93 (L2-1 and L2-2). Similarly, Trp121 (L3-2) and His120 (L3-3) were replaced with Cys, in consideration of their proximity to Cys128. Ala95 was changed to Gly in L2-2 to endow conformational flexibility (Table 1). L3-1 was expected to maintain the β-loop because it contains all residues of Loop3, so was not cyclized. However, Cys128 in L3-1 was changed to Ser to prevent undesired disulfide linkage with Cys125 or Cys127 (Table 1). Of the peptide mimics, L3-3 showed dose-dependent inhibitory activity, whereas the effects of L2-1 and L2-2 did not appear to be significant, implying that Loop2 is not as important as Loop3 (Fig. 4A) or that the folding of Loop2-mimic peptides was not similar to the native structure of Loop2. L3-1 and L3-2 might have lost the native folding necessary for effective RANKL binding, because of a loose or distorted conformation. In contrast, the restraint imposed on L3-3 may stabilize the conformation so that it is similar to Loop3 in the eRANK-eRANKL complex.

Table 1.

Peptide design for the development of RANKL inhibitors

The original loop sequences		Peptide sequences
Loop2	93- CDAGKALV-100	L2-1	YC DAGKAL CY
		L2-2	YC DGGKAL CY
Loop3	118-GYHWNSDCECCRRN-131	L3-1	GYHWNSDCECSRRN
		L3-2	YC NSDCEC CY RR
		L3-3	YC WNSDCEC CY RR
OPG	113-LEIEFCLKHR-122	OP3-4	YC EIEF CY LIR

Residues involved in RANKL interaction are highlighted and underlined. Cys and Tyr residues introduced for cyclization of peptides are also highlighted.

Fig. 4.

Inhibitory activities of synthetic peptides and ARO-associated eRANK mutants. (A) Osteoclast precursors were treated with 10 ng/ml eRANKL, and varying concentrations (10 to 50 μM) of peptides (L2-1, L2-2, L3-1, L3-2, and L3-3), and their differentiation was monitored by TRAP assay. (B) Inhibitory activities of L3-3 and OP3-4 were compared by TRAP assay under the same conditions as in Fig. 4A over a wide concentration range (1 to 200 μM). (C) Concentration-dependent inhibitory activities of wild-type and mutant eRANKs (G54R, K171G, and C176R) were compared by TRAP assay. eRANKL at the concentration of 100 ng/mL was used for cell differentiation.

The inhibitory activity of L3-3 was characterized by the cell differentiation assay in a wide concentration range. OP3-4, a well-known OPG-mimic inhibitor of the osteoclastogenesis (20, 21), served as a positive control. L3-3 effectively inhibited the eRANKL-induced osteoclastogenesis, even more effectively than OP3-4 at high concentrations (Fig. 4B). This was further confirmed by ITC analysis of L3-3 and OP3-4 (Fig. S6B), which showed that L3-3 has 16-fold higher binding affinity to eRANKL (K_d = 17.3 μM) than OP3-4 (K_d = 284.4 μM). This result demonstrated that L3-3 is a potent inhibitor of RANKL and could be developed as a potential therapeutic agent for the treatment of bone-related diseases.

Molecular Basis for the Pathogenesis of ARO.

The proposed binding modes of eRANK and eRANKL provided a molecular basis for the pathogenesis of ARO, a human inherited bone disease in which the differentiation of osteoclasts is impaired by mutations in RANKL or RANK (11, 12). Three mutations of RANKL are associated with ARO: the deletion of β-strand A and half of the AA″ loop; the substitution mutation M199K; and the frameshift causing the deletion of β-strands G, H, and F (11). Deletion of the AA″ loop could abolish the interaction with RANK, because this loop forms an extensive interaction network with Loop1 and Loop3 of eRANK (Fig. 1D, Fig. S5B, and Table S1). The frameshift mutation and the resultant loss of strand F, which is important for the trimerization of RANKL, could significantly affect the conformation and binding activity. Of seven mutations identified in human RANK, four (G53R, R129C, R170G, and C175R) are located in the extracellular domain (12). In the mouse eRANK, Arg130, the equivalent of Arg129 of human RANK, forms ionic and hydrogen bonds with Glu225 and Asn266 of RANKL (Table S1). Therefore, the R129C substitution is expected to impair the affinity for RANKL and, accordingly, the ability to induce the differentiation of osteoclasts.

On the other hand, M199 of human RANKL, and G53, R170, and C175 of human RANK, which correspond to mouse M198, G54, K171, and C176, do not seem to directly affect the formation of the complex, because they are not on the RANK-RANKL interface. To understand how these mutations cause ARO, recombinant proteins were expressed in a bacterial expression system and characterized. The M198K mutant was completely insoluble under the expression conditions of wild-type eRANKL, suggesting that this mutation might result in instability or misfolding of the protein, although we cannot rule out the possibility of the soluble expression in animal cells. In the osteoclastogenic analysis, the G54R, K171G, and C176R mutants all showed significant decreases in inhibitory activity (Fig. 4C). Consistent with this observation, eRANKL binding was not detected for G54R and K171G by ITC under current experimental conditions (Fig. S6A). These results suggest that mutant proteins may still retain very weak binding affinity to RANK and thus affect in vitro osteoclastogeneis only at the higher concentration. The low solubility of the C176R mutant prevented ITC measurement, but considering the close correlation between the binding affinity and osteoclastogenesis, C176R seems to have reduced RANKL-binding activity. Conformational changes caused by these mutations were proposed to diminish the RANKL-binding affinity, consequently impairing bone resorption. This was confirmed by CD analyses of G54R and K171G, which showed that their spectra did not overlap with that of wild type and indicated that the secondary structures were altered (Fig. S6C). Also, the C176R substitution is likely to change the conformation of RANK substantially, because Cys176 is a key residue to determine the folding of CRD4 by the formation of the disulfide bond.

Discussion

RANK has the overall fold common to all TNF-family receptors (Fig. S4), but it also has distinctive conformational features and ligand-binding mode that confer the binding specificity for RANKL. The most striking difference between RANK and other TNF-family receptors is the presence of an additional disulfide bond (Cys125-Cys127) in Loop3 that restrains the conformation and determines its unique topology (Figs. 1 and 3A). In eDR5, the loop corresponding to the RANK Loop3 has a loose conformation because the conformational restraint is alleviated (Fig. S7). RANKL has a smaller and deeper binding pocket, whereas that of TNF-related apoptosis-inducing ligand (TRAIL) is larger and shallower (Fig. S7). The overlapping of eRANK onto the eDR5-eTRAIL complex, or eDR5 onto the eRANK-eRANKL complex, shows severe steric clashes in the Loop3 binding region. Therefore, it is thought that the structural heterogeneity in Loop3 is key to the recognition specificity for eRANKL or eTRAIL, even though the central ligand-binding domains of RANK and DR5 are very well conserved in structural aspects (Fig. 3A and Fig. S7).

By analogy, the structural comparison of Loop3 explains the dual specificity of OPG for both RANKL and TRAIL. OPG Loop3 is two and four residues shorter than those in RANK and DR5, respectively (Fig. S2). Therefore, it has more compact and shorter shape. Furthermore, E116, located on the tip of OPG Loop3, overlaps well with E126 in RANK and is proximal to E151 in DR5. Considering the structural flexibility of the loop, it is expected that OPG Loop3 can fit into pockets of both RANKL and TRAIL without steric hindrance, and the major interactions between Loop3 and the binding pocket are preserved.

Loop1 of RANK seems to be less important than those in other TNF-family receptors. In the TNFR1-TNFβ or DR5-TRAIL complex, the hydrophobic interaction between Loop1 of the receptor and the DE loop of the ligand is the main force for binding, with the conserved tyrosine residue in the DE loop penetrating the hydrophobic groove formed by hydrophobic residues in Loop1 (23–25). However, Ile248 of eRANKL, which is equivalent to the Tyr residue in TNFR or DR5, is thought to be only marginal for receptor binding, because the I248D mutant maintained binding activity to RANK (28). This was corroborated in the structures presented here, because the electron densities of the Ile248 residues in the DE loop (residues 245–250) were very poor in both free (29) and ligand-bound forms (current study).

Recombinant proteins that inhibit RANK-RANKL interaction, such as RANK-Fc, Fc-OPG, and anti-RANKL antibodies, have been developed as therapeutic agents for osteoporosis (13–19). Denosumab, a humanized monoclonal antibody against RANKL, is currently in phase III trials for osteoporosis treatment (16), and in phase II trials for the treatment of rheumatoid arthritis and osteolytic bone metastases (18, 19). AMGN-0007, a recombinant Fc-OPG, is in phase I clinical trials for the treatment of osteolytic bone metastasis (14, 15). However, the use of large macromolecules for therapeutic intervention can be hindered by drawbacks including low stability, poor bioavailability, high cost, and difficulties in administration. Therefore, small peptides or peptidomimetics with high specificity to RANKL can be an alternative to overcome disadvantages of using macromolecules. For this purpose, small peptides, WP9QY and OP3-4, were developed and their inhibitory activity for bone resorption was confirmed in vivo (20–22). However, further improvement of their efficacy is necessary for the application for therapeutic purposes.

The structural interpretation and functional studies of eRANK mutants indicated that the disulfide-restrained Loop3 is the major determinant for specific RANK-RANKL recognition. This enabled us to design previously undescribed inhibitor peptides that mimic the structure and interaction of Loop3. The cyclic peptide L3-3 strongly bound to eRANKL and blocked RANKL-induced osteoclast differentiation more efficiently than OP3-4 under our experimental conditions (Fig. 4B and Fig. S6B). An OPG-RANKL model and cell differentiation assay using chimeric RANKs containing OPG loops suggested similarity in the binding mode of RANK and OPG. Accordingly, it was expected that Loop3 in OPG is also involved in RANKL binding in the analogous way, which was also supported by the fact that OP3-4, a peptide comprising essential binding residues of Loop3, was proven as a potent RANKL inhibitor in vitro and in animal models (20, 21). The enhanced inhibitory activity of L3-3 might be due to additional binding residues (Table 1). Further modification of L3-3 in parallel comparison with OP3-4 will allow the development of more potent peptides or peptide-derived chemical agents that can be applied for therapeutic purposes. In this aspect, current study provides a framework for rational design of RANKL inhibitors.

Structure analysis of RANK-RANKL combined with mutant studies enabled us to explain the molecular basis for the inherited bone disease, ARO. The crystal structure revealed that some of the mutations found in ARO patients are on the RANK-RANKL interface and are likely to directly interfere with the binding. Current study proved that some ARO-associated mutations of RANK cause defective folding and affect the binding to RANKL, although they are not on the binding interface.

Materials and Methods

Structure Determination.

The crystal structure of the eRANK-eRANKL complex was determined by molecular replacement using the crystal structure of RANKL as a search probe (PDB ID code 1IQA; SI Text). The final model refined at 2.5-Å resolution with R and R_free values of 22.6% and 24.9%, respectively, comprises residues 35–198 of eRANK and residues 161–315 of eRANKL. Poly-Ala peptides were modeled at residues 35–46 in eRANK and residues 245–250 in eRANKL due to weak electron densities in these regions. The refinement statistics are summarized in Table S2.

Site-Directed and Loop Replacement Mutagenesis.

Point mutations of eRANK were introduced using a Quick Change mutagenesis kit (Stratagene). Chimeric RANK constructs were created using PCR in which Loop1 (residues 73–89), Loop2 (residues 94–100), or Loop3 (residues 119–130) was replaced by corresponding regions of OPG, Loop1 (residues 67–82), Loop2 (residues 88–94), and Loop3 (residues 111–120). Mutant and chimeric proteins were expressed and purified as described for the wild-type protein. The elution profile of Superdex 200 gel filtration chromatography and the size distribution profile determined by dynamic light scattering were used to verify proper folding and assembly of mutant proteins in comparison to the wild type.

Osteoclast Differentiation Assay.

The biological activities of eRANKL, eRANK, and inhibitory peptides were estimated using an osteoclast differentiation assay. Mouse osteoclast precursors were extracted and purified from the tibia and femur of 5- to 8-wk-old male mice. Osteoclast precursors were cultured in 96-well plates (2 × 10⁴ cells/well) in α-MEM supplemented with 10% FBS and 30 ng/mL macrophage colony stimulating factor. Varying concentrations of eRANKL and eRANK were added, and the cells were incubated at 37 °C in a humidified atmosphere containing 5% CO₂. Differentiation of osteoclast precursors was measured in vitro by quantitation of tartrate-resistant acid phosphatase (TRAP) activity. TRAP staining assay was performed using Naphthol AS phosphate substrate in 100 mM sodium acetate (pH 5.2), and the reaction product was quantified by spectroscopic measurement at 405 nm. TRAP-positive multinucleated cells were counted under the microscope, and each experiment was repeated three times for statistical analysis.