RANKL employs distinct binding modes to engage RANK and the OPG decoy receptor

SUMMARY

Osteoprotegerin (OPG) and receptor activator of nuclear factor kappa B (RANK) are members of the TNFR superfamily that regulate osteoclast formation and function by competing for RANK ligand (RANKL). RANKL promotes osteoclast development through RANK activation, while OPG inhibits this process by sequestering RANKL. For comparison, we solved crystal structures of RANKL with RANK, and RANKL with OPG. Complementary biochemical and functional studies reveal that the monomeric cytokine-binding region of OPG binds RANKL with ~500 fold higher affinity than RANK, and inhibits RANKL-stimulated osteoclastogenesis ~150 times more effectively, in part because the binding cleft of RANKL makes unique contacts with OPG. Several side chains as well as the C-D and D-E loops of RANKL occupy different orientations when bound to OPG versus RANK. High affinity OPG binding requires a 90s-loop Phe residue that is mutated in juvenile Paget’s disease. These results suggest cytokine plasticity may help to fine tune specific TNF-family cytokine/receptor pair selectivity.

INTRODUCTION

Normal skeletal mass reflects a balance between bone-forming osteoblasts and bone-resorbing osteoclasts (Leibbrandt and Penninger, 2009; Seeman, 2009; Zaidi, 2007). When the activity of osteoclasts substantially supersedes that of osteoblasts, patients develop osteoporosis, a condition characterized by reduced bone mineral density. In contrast, osteopetrosis, a condition of extremely dense bone, is the product of failed osteoclast formation or function.

The osteoclast is a polykaryon of hematopoietic origin whose differentiation from monocyte/macrophage precursors uniquely requires oligomerization and activation of the cell-surface receptor RANK by the TNF-like cytokine RANKL(Boyce and Xing, 2008; Kim et al., 2000; Kong et al., 1999; Lacey et al., 1998; Leibbrandt and Penninger, 2008; Teitelbaum, 2007; Yasuda et al., 1998). In fact, RANKL can be thought of as both an osteoclast differentiation and activation factor (Lacey et al., 1998). RANKL, in conjunction with M-CSF, is sufficient to prompt bone marrow macrophage differentiation into bone resorbing osteoclasts in vitro. Importantly, RANKL and its receptor RANK are required for osteoclastogenesis in vivo. Both RANKL (Kong et al., 1999) and RANK-deficient mice lack osteoclasts (Dougall et al., 1999; Li et al., 2000). The discovery that the later stages of osteoclast differentiation are blocked in mice that over express the decoy receptor, OPG (Simonet et al., 1997), established that RANK and OPG reciprocally regulate bone resorption. The clinical relevance of this relationship is underscored by the fact that many forms of osteoporosis are characterized by an increase in the ratio of circulating RANKL/OPG (Jabbar et al., 2011; Wasilewska et al., 2010; Xu et al., 2011). Manipulation of this ratio forms the basis of present anti-osteoporosis therapy. Hence, an understanding of the different mechanisms by which RANK and OPG recognize RANKL could prove useful in designing better therapeutics.

The RANKL gene was identified by expression cloning using OPG and RANK as probes (Anderson et al., 1997; Lacey et al., 1998; Wong et al., 1997; Yasuda et al., 1998). In addition to osteoclastogenesis, RANKL mediates lymph node formation, establishment of the thymic microenvironment, T-cell growth and dendritic-cell function. The cytokine also governs expansion and activity of mammary glands during pregnancy and lactation, and regulation of female basal body temperature (Anderson et al., 1997; Bachmann et al., 1999; Dougall et al., 1999; Fata et al., 2000; Hanada et al., 2009; Kong et al., 1999).

RANKL is a ~35 kDa type II transmembrane protein with a short N-terminal intracellular tail and a C-terminal extracellular region that contains a connecting stalk and receptor-binding domain (Figure 1A). Membrane-bound RANKL activates RANK to generate osteoclasts through cell-cell contact. In states of inflammatory osteolysis, RANKL is also cleaved to release a soluble, biologically-active product (Schlondorff et al., 2001). Like most TNF-family cytokines (Bodmer et al., 2002), RANKL forms a homotrimer in solution. Hydrophobic interactions at the core of the trimer drive monomer assembly around a 3-fold axis of symmetry. The individual monomers are composed entirely of β strands and loops connected in a “jelly-roll” fold, with the two β sheets of each monomer stacking as a sandwich (Lam et al., 2001). The sequence similarity among TNF-like cytokines is largely confined to the 25% - 30% of internal residues responsible for trimer stability. As each trimer assembles, loops at the edges of apposed monomers form the sides of the receptor-binding clefts, the shape of which determines receptor selectivity. For RANKL, the three identical receptor-binding clefts are spaced equally around the outside of the cytokine.

Figure 1. Schematic of the OPG, RANK, and RANKL proteins.

(A) RANKL domain architecture. The extracellular receptor-binding domain is encoded as the C-terminal region. (B) RANK domain architecture. In RANK, the CRDs are followed by a transmembrane domain (TM) and a cytoplasmic tail containing TRAF-binding motifs. (C) OPG domain architecture. In OPG, the CRDs are followed by two death domain related regions (DD) and a highly charged basic region. (D) OPG/RANKL and RANK/RANKL complexes in the context of cell membrane insertion.

RANK, the cell surface signaling receptor, is expressed as a ~67 kDa type I transmembrane protein consisting of four extracellular CRD’s linked to a long C-terminal intracellular region (Figure 1B). Like most members of the TNF-receptor superfamily, the CRDs of RANK consist of pseudo-repeats, each approximately 40 residues in length, containing between one and three disulfide bridges. These CRDs can be further characterized as consisting of pairs of structurally-conserved modules distinguished by fold type and number of disulfide bonds (Naismith and Sprang, 1998). Binding of RANKL to the CRDs of RANK stimulates receptor trimerization (Figure 1D). Indeed, the 3-fold symmetry enforced by the cytokine/receptor complex appears to be the stoichiometry of most TNF/TNFR superfamily signaling interactions (Bodmer et al., 2002; Hehlgans and Pfeffer, 2005; Locksley et al., 2001). The RANK protein lacks intrinsic enzymatic activity, but upon trimerization recruits signaling adaptors and other TNFR–associated factors that, in turn, activate osteoclastogenic signaling networks (Armstrong et al., 2002; Wong et al., 1998).

OPG, the naturally occurring decoy receptor for RANKL, is synthesized as a ~55 kDa monomer that self-associates to form a disulfide-linked homodimer prior to secretion (Figure 1D). Like RANK, the N-terminal half of OPG consists of four CRDs that are necessary and sufficient to inhibit osteoclast formation (Yamaguchi et al., 1998) (Figure 1C). The decoy receptor’s C-terminal half contains a dimerization cassette, comprised of two regions with similarity to cytoplasmic death domains, juxtaposed to a heparin-binding basic motif and having cysteine as the penultimate residue. Dimer formation enhances the receptor’s avidity for RANKL, which is likely central in the receptor’s ability to inhibit osteoclast formation in vivo (Schneeweis et al., 2005).

OPG is secreted primarily by osteoblasts and marrow stromal cells. By sequestering RANKL, OPG inhibits the RANKL/RANK interaction, blunting the maturation and bone degrading capacity of osteoclasts. Although human mutations in OPG are rare, loss of function severely affects bone growth. About 50 individuals worldwide have been identified with juvenile Paget’s disease, an autosomal recessively inherited osteopathy characterized by accelerated bone remodeling, low bone mineral density, fractures, and progressive skeletal deformity. The disease displays considerable phenotypic variation, the severity of which correlates with specific mutations in the OPG gene. The most affected individuals carry large homozygous deletions of OPG, or missense mutations in cysteine residues predicted to cause major disruption of the RANKL binding domain. Less affected individuals carry point mutations in the CRDs thought to alter RANKL binding (Chong et al., 2003). The physiologic role of OPG is not limited to the inhibition of bone resorption. OPG also binds to and inactivates TRAIL (TNF-related Apoptosis Inducing Ligand) (Emery et al., 1998), a member of the TNF family that promotes immune cancer surveillance. TRAIL also binds decoy receptors 1 (DcR1) and 2 (DcR2) that fail to induce apoptosis due to a lack of functional death domains.

The modular nature of TNF-receptor cysteine-rich domains permits determination of accurate sequence alignments even in the absence of significant sequence conservation. Still, structural modeling of TNF receptors has proven difficult. Further, without structural data, predicting the binding selectivity of specific TNF receptors is problematic due to uncertainties in the positions and orientations of successive modules, as well as the conformations of divergent loops. This is particularly relevant for the RANKL system. The inherent complexity (RANKL binding both OPG and RANK, and TRAIL binding OPG, DR4, DR5, DcR1, and DcR2) raises basic questions about binding modes and selectivity that can only be answered at the molecular level.

The capacity of OPG to dampen osteolysis makes it, and related molecules, candidate anti-osteoporosis therapeutic agents. With this in mind, we determined crystallographic structures for RANK and OPG in association with RANKL. The two TNF receptors compete for the same binding cleft, but for different biological purposes; RANK as a signaling receptor and OPG as a decoy receptor. This exercise provides structural insight into the determinants that support the decoy function; information that may prove important for the design of improved anti-osteoporosis drugs.

RESULTS

Structure determinations

To compare the interactions of OPG and RANK with RANKL, we prepared receptor/cytokine complexes for structural analysis. Both pairs formed crystals in space group P6₃ (Table I). Although the packing was similar, the unit cell of the OPG/RANKL crystal was smaller than that of its RANK/RANKL counterpart. Native diffraction data for the OPG/RANKL complex were collected to 2.70 Å resolution. Phasing was accomplished by a combination of molecular replacement based on our structural model of the cytokine (Lam et al., 2001) and MAD using selenomethionine labeled RANKL (Table S1). Structural refinement yielded a final model with an R_work of 20.2% and R_free of 24.0% and with root mean square deviations (RMSDs) from ideal values of 0.003 Å for bond lengths and 0.650° for bond angles (Table I). In the OPG/RANKL crystal structure, the first 8 N-terminal residues of OPG are disordered, as is the last half of CRD4. The crystallographic asymmetric unit consists of one molecule of OPG (residues 9 – 141) and one monomer of RANKL (residues 162 – 315). A 3:3 complex (or hetero-hexamer) becomes apparent upon application of the 3-fold crystallographic symmetry (Figure 2). Each OPG binding cleft incorporates two neighboring RANKL monomers, one contributing the A′-A″ loop side and the other the D-E loop side of the cleft (Figure S1).

Table I.

Data collection and refinement statistics

Data Set	OPG-RANKL	RANK-RANKL
Space Group	P63	P63
Unit Cell dimensions (Å)	a = b =109.9, c= 78.9	a = b = 120.6, c = 94.3
X-ray Source	ALS 4.2.2	Rigaku RUH3 CuKα
Wavelength (Å)	0.9951	1.5418
Resolution range (Å)	30.40–2.70	20.0–2.70
(outer shell) (Å)	(2.79–2.70)	(2.87–2.70)
Observations/Unique	42626/14915	107644/21055
Completeness (%)	99.3 (100.0)	98.1 (92.6)
Reflections Rsym (%)	8.4 (56.3)	8.0 (53.7)
<I/σI >	12.4 (2.1)	13.9 (2.2)
<Redundancy>	2.9	5.9
Refinement Statisticsa
Protein Atoms/Solvent	2280/192	2476/76
Rwork (%)	20.2 (31.8)	18.8 (30.8)
Rfree (%)	24.0 (35.4)	21.6 (34.3)
Ramachandran plot, most favored/additional (%)	95.0/5.0	96.2/3.8
RMSD bond length (Å)/angles (°)	0.003/0.650	0.002/0.576
RMSD Dihedral (°)	10.08	9.5

Advanced Light Source Beamline 4.2.2.

^aValues as defined in CNS.

^bFrom Cross-Validated Luzzati plot.

Figure 2. Comparison of the OPG/RANKL and RANK/RANKL complexes.

(A) Lateral view. The RANKL trimer is illustrated as a white surface model, while OPG (magenta) and RANK (green) are presented as ribbons. The approximate position of each CRD is detailed (left). Disulfide bridges (yellow balls and sticks) form the framework of each receptor. To compare the relative orientation of RANK and OPG in the binding cleft, the co-complex structures were aligned by superposition of the RANKLs. Circles indicate the approximate locations of various RANKL loops. (B) Bottom-up view. Three OPG receptors are shown captured in three equally spaced binding clefts around the surface of the RANKL trimer. A black triangle marks the threefold crystallographic axis of symmetry.

The RANK/RANKL complex was crystallized, diffraction data collected to 2.7 Å resolution, and the structure phased by molecular replacement using our RANKL model (Lam et al., 2001). The RANK fragment contained only the ligand-binding ectodomain (residues 5 – 168). Structural refinement yielded an R_work of 18.8% and R_free of 21.6% with an RMSD from ideal values of 0.002 Å for bond lengths and 0.576° for bond angles (Table I). Satisfactory density was observed for RANK, excluding a few residues at the N-terminus. Similar to the OPG/RANKL crystal, the RANK/RANKL crystal also contains one subunit of receptor and cytokine in each asymmetric unit.

RANK is more elongated than OPG

Only 34% of the residues at structurally equivalent positions in OPG and RANK are identical, falling to 26% if only non-cysteine positions are considered (Table S2). Despite the low sequence identity throughout the cytokine binding regions, OPG and RANK share a similar structural framework. Superposition of the individual CRDs reveals substantial fold similarity, with RMSDs of equivalent Cα varying from 0.9 to 1.5 Å (Table S3A).

A flexible articulation point separating CRD2 and CRD3 is a common feature of TNFR family members. This “hinge”, identifiable by the characteristic CXC motif (Mongkolsapaya et al., 1999), enables TNFRs to adopt distinct orientations in different solvent conditions. The hinge conformation differs in the OPG and RANK co-complexes, and as a result OPG appears to be more bowed than RANK when bound to RANKL. This is in part due to a two-residue deletion in the CRD2 region of OPG (Figure 3A). The shorter OPG CRD2 loop combines with sequence disparity at the CRD2/CRD3 interface to shift the relative position of CRD3. Accordingly, a twist of 17° and a swing of 32° pivoting about the CXC hinge would place the OPG CRD3 onto the RANK CRD3 (Figure 2A).

Figure 3. Comparison of OPG and RANK contact residues at the RANKL interface.

(A) Structure-based sequence alignment of cytokine-binding regions. The mouse OPG (mOPG) sequence numbering, as counted from the amino-terminus of the mature protein, is given above the alignment. The mouse RANK numbering is given under the alignment. Individual amino acids interacting with the cytokine (≤ 4 A) are boxed and colored according to contact distance using the scale shown in panel B. The shortest distance between any atom interface pair determines the color for the entire residue. The CXC hinge region linking CRDs 2 and 3 (OPG residues 84–86) is boxed in black with a gray background. Red lines connect disulfide-bonded cysteine residues. The positions of the “50s” and “90s” loops, implicated in controlling the ligand binding specificity of TNFR family members, are shown under the alignment. (B) Surface representation of OPG, RANK, and RANKL. Each receptor/cytokine complex has been opened like a book to reveal the contacting surfaces. Key interacting residues are delineated and are colored by distance using the scale shown. Important RANKL loop regions are circled.

Like other TNFR family members, OPG and RANK use CRD2 and CRD3 to bind cytokine, with CRD2 contributing approximately 60% and CRD3 about 40% of the surface area buried upon RANKL binding by either receptor (Table S3B). The specific residues contacting RANKL are highlighted in the structure-based sequence alignment (Figure 3A). The CRD module classifications are detailed above the sequence according to the method of Naismith (Naismith and Sprang, 1998) wherein A or B indicates fold type and X1 and X2 denote number of amino acids between folds. RANK buries ~12% more total surface area than does the OPG decoy receptor. Interestingly, OPG and RANK divide the buried surface area differently between the two sides of the binding cleft (Figures 3B). RANK buries more surface area against the A′-A″ loop side of the cleft than against the D-E loop side. In contrast, OPG uses the D-E loop side more, and the A′-A″ loop side less than does RANK.

Different conformations of RANKL engage OPG and RANK

While both receptors bind at the interface formed between two RANKL monomers, there are distinct differences in the residues of RANKL that are utilized by OPG versus RANK (Figure 3B). Additionally, the shape of the RANKL binding cleft recognized by OPG is different than that recognized by RANK. Despite this difference, the shape complementarity values calculated for each receptor / cytokine interface are similar (RANK, Sc = 0.656 and OPG, Sc = 0.675), suggesting that the receptors have evolved to fit their respective binding clefts equally well.

Although the conformation of RANKL in the RANK/RANKL complex is strikingly similar to that of the cytokine alone (Ito et al., 2002; Lam et al., 2001), the RANKL D-E loop residues 246–250 and C-D loop residues 225–234 are rearranged apparently to accommodate OPG binding (Figure 4A). The D-E loop is not well ordered in the RANK/RANKL structure. It adopts a conformation closely resembling the D-E loop seen in the unliganded cytokine and likewise displays high B-factor values, suggesting that this loop is mobile. In the OPG complex, the D-E loop adopts a well-defined conformation where it makes unique contacts with the first half of OPG CRD2.

Figure 4. Conformational effects of C-D and D-E loop rearrangements.

(A) Alignment of RANKL from the OPG/RANKL and RANK/RANKL complexes. Only two RANKL monomers are visualized from each complex. The RANKL strands are labeled A-H with the protein termini marked “n” and “c”. Portions of the RANKL structure with Cα RMSDs <2 SD from the mean are white, while those with ≥2 SD variance are colored (OPG/RANKL, magenta; RANK/RANKL, green). (B) Lateral view of OPG CRDs 2 and 3 shown against one half of the receptor binding cleft formed by the A′-A″ loop side RANKL monomer. A small inset contains the OPG/RANKL complex boxed to show the region selected for magnification. The RANKL C-D loop is orange. OPG is again magenta in color with yellow disulfides. Hydrogen bonds appear as thin cyan lines. The atoms involved in non-bonded contacts are enclosed in a blue transparent surface generated using Ligplot+ at the default values (Laskowski and Swindells, 2011). The β2-β3 strands comprise the CRD3 projection often referred to as the “90s loop”. The right panel shows a close up view of the CRD3 projection looking straight into the cleft with important residues labeled. (C) Lateral view of RANK CRDs 2 and 3 in the binding cleft of RANKL in the same orientation as in (B). RANK is shown in green. The atoms involved in non-bonded contacts are enclosed in a pink transparent surface. Again, the right panel shows a close up view of the CRD3 projection looking straight into the binding cleft.

The C-D loop rearrangement is more complicated. The OPG/RANKL interaction requires a cascade of side chain movements at the base of the cytokine receptor-binding cleft (compare Figures 4B and 4C). The rotamer adopted by OPG-bound RANKL^F269 leaves the phenyl ring turned toward the C-D loop. Nearby, the imidazole ring of RANKL^H224 is turned away from the receptor. Additionally, the side chain carboxylic acid group of RANKL^E225 is rotated approximately 110° away from the receptor, shifting the backbone of the C-D loop. The loop pivots at RANKL^E225 and RANKL^Y234. As a result of these rearrangements, the side chain of RANKL^Y234 must swing approximately 90° out of the binding site to release its end of the C-D loop. These alterations create a novel hydrophobic pocket at the base of the binding cleft that accepts the phenyl ring of OPG^F96. Hence, a tight fit of OPG with RANKL does not appear possible without significant shifts in the positions of side chains within the receptor-binding cleft and displacement of the C-D loop (see Figure S2).

The OPG/RANKL interface is more hydrophobic than that of RANK/RANKL

The contacts between OPG CRD2 and the D-E loop of RANKL are largely hydrophobic with additional CRD2 contacts forming an elongated hydrophobic patch buried deep in the binding cleft along the E strand (Figure S3A). The key OPG binding element is contained in the CRD3 module. The β2 and β3 strands of CRD3 form a tight loop capped by OPG^F96 that juts into the binding site (Figure 4B). This projection corresponds to the “90s” loop, a region that controls the ligand-binding specificity of several TNF receptor family members(Hymowitz et al., 1999). Virtually every residue of this β2-β3 projection interacts with RANKL (Table S4A). The phenyl group of OPG^F96 is involved in a pi-stacking interaction with the phenyl group of RANKL^F269 at the beginning of the F strand. This interaction is possible because the rotamer of RANKL^F269 creates a deep pocket that accepts the phenyl ring of OPG^F96. Consequently, the C-D loop residues of RANKL converge around OPG^F96 to form a large, contiguous cluster of hydrophobic interactions with the OPG β2-β3 projection.

The hydrophobic contacts between RANK and RANKL are more isolated than those of OPG and more distributed around the binding interface (Figure S3B). The CRD2 module of RANK contains fewer residues juxtaposed to the D-E side of the RANKL cleft than does OPG, and thus the overall number of contact sites, in this region, are less. RANK^E54 generates a salt bridge with RANKL^K247 in the D-E loop and RANK^D55 does so with RANKL^R283 in the F-G loop. On the other side of the cleft, the CRD2 module of RANK contacts the A′-A″ loop (Figure 4C). Again, whereas the OPG/RANKL interactions are primarily hydrophobic, those of RANK/RANKL are generally charge driven, including two salt bridges (RANK^D64 to RANKL^R222 and RANK^K67 to RANKL^D299) as well as a hydrogen bond (RANK^K56 to RANKL^G191).

The RANK CRD3 β2-β3 strand also forms a tight loop and like OPG, virtually every residue in the RANK β2-β3 projection interacts with RANKL. The structural position analogous to OPG^F96 is occupied in RANK by a unique disulfide bond (C⁹⁵–C⁹⁷). However, unlike OPG^F96, this disulfide fails to make strong hydrophobic contacts with RANKL. Instead, the RANK β2-β3 projection forms several hydrogen bonds, including RANK^C98 to RANKL^H224 and RANK^Y89 to RANKL^E225 in the C-D loop, as well as RANK^S93 to RANKL^K180 in the A-A′ loop. The interaction of RANK with the RANKL A-A′ loop also includes two salt bridges (RANK^D94 to RANKL^H179 and RANK^E96 to RANKL^K180), again highlighting the ionic nature of the RANK/RANKL interface (a complete list of interface contacts are given in Table S4A for OPG/RANKL and S4B for RANK/RANKL).

To determine the relative hydrophobicity of the OPG/RANKL and RANK/RANKL interfaces we utilized the protein interfaces, surfaces and assemblies (PISA) server (Krissinel and Henrick, 2007). PISA assigns a P-value by comparing randomly selected surface atoms to those in the contact area. A P-value greater than 0.5 indicates the contact surface is more hydrophilic than one randomly generated, whereas a P-value less than 0.5 indicates greater hydrophobicity.

The P-values of the A′-A″ and D-E loop sides of the RANK/RANKL interface are 0.701 and 0.925, respectively. These data indicate the surface of RANKL buried by RANK is hydrophilic. In contrast, the respective P-values of the OPG/RANKL interface are 0.211 and 0.367 for the A′-A″ and D-E loop sides. Thus, the surface of RANKL buried by OPG is, in contrast, hydrophobic. While this difference partly reflects the distinct footprints of RANK and OPG on RANKL, it is primarily due to a change in the chemical nature of the RANKL binding groove caused by the side-chain shifts and loop rearrangement.

OPG/RANKL is more stable than RANK/RANKL

We determined single-site affinities for RANKL binding to monomeric CRD-spanning regions from OPG and RANK using surface plasmon resonance (SPR). In this circumstance, RANKL (cleaved and purified to remove the GST-tag) was coupled to a CM5 sensor chip using NHS/EDC chemistry. Monomeric constructs spanning the cytokine-binding regions of RANK, OPG, or OPG variants were flowed over the chip surface. The binding appeared both specific and saturable. By Scatchard analysis, an equilibrium affinity of K_{D, equilibrium} = 2.09 μM was obtained for the interaction between monomeric RANK and RANKL (Figure 5B). The monomeric CRDs of OPG bound with significantly higher affinity than RANK (for OPG, K_{D, equilibrium} = 4.24 nM). Similar single site binding affinities were obtained using biolayer interferometry on an octet red system (Table S5). SPR kinetic analysis indicates that OPG exhibits an approximately 30 fold faster on-rate and an approximately 20 fold slower off-rate than RANK (Figure 5A). The higher affinity and, in particular, the longer off-rate of the OPG fragment are consistent with the function of the intact protein as a decoy receptor.

Figure 5. Comparison of the affinity for RANKL of monomeric cytokine-binding regions from OPG, OPG^F96A,OPG^F96L, and RANK.

(A) Binding curves for the interaction of monomeric truncated receptor (CRDs 1 through 4) with RANKL obtained from SPR data using a Biacore T100. Soluble RANKL was coupled directly to a CM5 chip and various concentrations of soluble OPG, OPG^F96A, OPG^F96L, and RANK receptor CRDs were injected through the flow cell at the concentrations listed on the sensorgrams. A summary table of the affinity constants determined by kinetic analysis is given under each sensorgram, k_a(1/Ms), k_d(1/s), and K_{D kinetic}(M). The values in each table represent the average of three independent experiments and the standard deviation. The RANK CRD curves were fit simultaneously to determine k_d/k_a using a 1:1 (Langmuir) binding model. The OPG CRD data were fit using a two state reaction (binding with conformational change) model. The fits are included on each set of sensorgrams as a thinner set of curves. A cartoon of the experimental setup is given below each set of binding curves. (B) Saturation curves and Scatchard plots for the experiments in (A) at equilibrium. Scatchard analysis of the cytokine-receptor fragment interaction yields the affinity at steady state, K_{D equilibrium}(M). The values reported for each are from the representative experiment shown with the standard deviation determined from three individual experiments.

High-affinity RANKL binding requires OPG ^F96

The equivalent position of OPG^F96 in humans (OPG ^F117) is mutated to leucine in a subset of families with juvenile Paget’s disease (Chong et al., 2003), a disorder of accelerated bone resorption due to enhanced osteoclastogenesis. This phenotype is consistent with OPG dysfunction (Bucay et al., 1998; Mizuno et al., 1998). FoldX analysis (Schymkowitz et al., 2005) of the interface suggested F96 as the most energetically important residue of OPG for RANKL binding. Based on the disease association, and the apparent importance of OPG^F96 for RANKL binding, we constructed an OPG^F96A mutant to determine whether complete removal of the hydrophobic, pi-stacking head group at this position would yield a dramatic decrease in affinity. Indeed, this single substitution diminished the affinity for RANKL by approximately 60-fold (Figure 5). The dissociation rate of the OPG^F96A CRDs from RANKL was essentially the same as the wild-type OPG CRDs. However the on-rate of the OPG^F96A CRDs was approximately 30 fold slower than wild-type (Figure 5A). Clearly the aromatic ring of OPG^F96 is important for rapid high-affinity binding. Interestingly, the juvenile Paget’s disease mutation, OPG^F96L, decreased binding of the CRD-spanning fragment to RANKL even further, by approximately 2900-fold compared to wild-type. The disease mutant CRDs displayed about a 270 fold slower on-rate and a 5 fold faster off-rate. Indeed, the juvenile Paget’s disease CRDs bound RANKL with approximately a 6-fold lower single-site affinity than did monomeric RANK.

To complement our binding analysis, we tested the ability of the monomeric CRD-spanning fragments from wild-type OPG, OPG^F96A, OPG^F96L, and RANK to inhibit osteoclastogenesis in vitro. Bone marrow macrophages were cultured with osteoclastogenic amounts of M-CSF and RANKL in the presence of increasing concentrations of soluble monomeric RANK, OPG, OPG^F96A, or OPG^F96L CRD-spanning fragments. Binding of soluble receptor fragments to RANKL is expected to block its association with RANK on the cell surface and thereby prevent precursor cells from differentiating into mature osteoclasts. After five days, the cells were stained for tartrate-resistant acid phosphatase (TRAP) activity and osteoclasts counted (Figure 6). No osteoclasts developed in the absence of RANKL. Although all the receptor CRD regions were ultimately inhibitory, the wild-type OPG CRDs were nearly 20-fold more effective than OPG^F96A, 200 fold more than OPG^F96L, and 150-fold more than RANK. Thus, the pathological mutation, OPG^F96L, renders the decoy receptor incapable of blocking RANK/RANKL signaling, explaining the osteolytic phenotype in patients with juvenile Paget’s disease.

Figure 6. Effect of soluble monomeric OPG, OPG^F96A, OPG^F96L, or RANK receptor fragments on osteoclast formation.

Bone marrow macrophages were cultured in M-CSF +/− RANKL with increasing concentrations of OPG, OPG^F96A, OPG^F96L, or RANK truncated receptor regions (CRDs 1 through 4) as soluble inhibitors. Also included was a set of control wells without any monomeric soluble receptor fragment (CONT). On day 5, the cells were stained for TRAP activity and osteoclasts were counted. Data represent the mean ± the standard deviation of 4 well replicates for each point and are representative of 3 individual experiments.

DISCUSSION

Our comparison of OPG and RANK, in complex with RANKL, reveals structural features that dictate the biological properties of each. Although they differ in primary sequence and number of disulfide bonds, the cytokine-binding domains of OPG and RANK are similar in fold. However, despite the topological resemblance, the receptors adopt different orientations within the RANKL binding groove. Both contact RANKL via their CRDs 2 and 3. Yet, RANK is more elongated than OPG, in part because the decoy receptor bends sharply around the hinge region joining CRDs 2 and 3. It is possible that this bend is required for the cytokine-binding regions of OPG to reach the dimerization domains, so as to occupy two RANKL binding sites simultaneously.

The OPG/RANKL and RANK/RANKL interactions diverge in other ways. RANK distributes its contacts evenly around the RANKL binding site. In contrast, the contacts made by the first half of OPG CRD2 are limited to part of the E-strand and D-E loop region of RANKL, avoiding almost completely the A′-A″ loop region of the binding cleft. Further, the majority of OPG CRD3 contacts concentrate around the deeply buried OPG^F96, which we have shown is required for rapid high affinity binding. Even when RANK and OPG interact with the same RANKL residues, they often do so differently. For example, RANKL^H224 extends out of the binding site to form a hydrophobic interaction with OPG^L69. In the analogous region, RANKL^H224 forms a hydrogen bond with the backbone carbonyl oxygen of RANK^C98. Regardless, there are some shared contacts between the signaling and decoy receptors, for example RANKL^Q236 contacts backbone atoms in CRD3 of both OPG and RANK. RANKL^Q236 makes non-bonded contacts with OPG^F96 and a hydrogen bond to the backbond oxygen of RANK^C95. It is worth noting that the structurally equivalent residue in TRAIL, Q205, makes contact with DR5, and alanine substitution of Q205 decreases the apoptotic activity of TRAIL by approximately 690-fold (Hymowitz et al., 2000). This analysis underscores the importance of structural information to guide residue-specific targeting of the interface for therapeutic purposes.

It is still unclear what role avidity plays in the competition for RANKL. In vivo, intact full-length OPG likely binds as a dimer, forming a bivalent interaction with RANKL at the cell surface, either to a single trimer or possibly by cross-linking adjacent trimers. In addition, OPG may recognize the RANKL cytokine shed in solution. The extra avidity afforded by OPG dimerization likely contributes to its role as a decoy receptor (Schneeweis et al., 2005). In contrast, RANK is constrained to move in the plane of the cell membrane where oligomerization of three RANK receptors, by RANKL, is likely required for signal induction. The ability of bivalent OPG to block assembly of the trivalent RANK/RANKL complex probably depends on its higher intrinsic affinity, which appears from our monovalent binding data to be driven by both a faster association-rate and a slower dissociation-rate. Our findings, that monomeric OPG CRDs alone are roughly 150-fold more potent at inhibiting osteoclastogenesis than their RANK counterparts, bolster this concept.

OPG interacts with another TNF family cytokine, TRAIL(Wiley et al., 1995). TRAIL induces tumor apoptosis upon binding to DR4 and DR5, two cell surface TNFR-family members whose cytoplasmic regions contain death domains(Baetu and Hiscott, 2002; Degli-Esposti, 1999). Although the relative affinity may be weak(Truneh et al., 2000), TRAIL sequestration has been argued to be a potentially negative consequence of therapeutic treatment with OPG(Reid and Holen, 2009; Zauli et al., 2009). Our OPG/RANKL structure, by providing atomic-level detail of the decoy receptor complex, may aid in the engineering of OPG variants incapable of TRAIL binding that could mitigate these concerns.

The only other secreted decoy TNF receptor in humans is DcR3. By sequence and structure, OPG appears more similar to DcR3 than to any other TNFR family member. Both receptors contain four CRDs, with the CRD3s two residues shorter than many other TNF-family receptors including RANK. DcR3 neutralizes three TNF-like ligands: TL1A, FasL, and LIGHT. The co-complex structure of DcR3/TL1A revealed that the CRD3 of DcR3 contributes little to TL1A binding specificity (Zhan et al., 2011). These authors suggest, by analogy, that the shorter OPG CRD3 might contribute less to RANKL binding than the longer RANK CRD3. However, our data establish that the CRD3s of RANK and OPG account for approximately the same fraction of the surface area buried upon RANKL binding (40% by both receptors). Further, our mutagenesis experiments demonstrate that the CRD3 of OPG contains a major determinant for RANKL binding (residue F96). Whether the CRD3 of DcR3 functions in a more conventional manner when binding FasL or LIGHT is unknown.

Two groups have reported X-ray determination of the RANK/RANKL complex (Liu et al., 2010; Ta et al., 2010) (see Table S6 for a comparison of the RMSD’s between domains and the CXC hinge angles observed in each structure). Our RANK/RANKL structure most closely resembles 3ME2, and both models include a sodium ion at the CRD3/CRD4 domain junction, which has been proposed to stabilize the interface(Liu et al., 2010). The evidence for the sodium ion in our model is based on several factors; the proximity of the ion to four main chain oxygen atoms, the octahedral configuration of the coordination sphere, the expected distance between the ion and its coordinating atoms, the good match between the B-factors of the ion and its coordinating atoms, and the presence of sodium in the crystallization conditions.

Based on an alignment of modules that make up the CRD2 and CRD3 domains of RANK and OPG, Liu et al. predicted a model for OPG/RANKL binding. Our efforts, however, provide unforeseen details of the RANKL/OPG interaction. We discovered a large shift of the OPG CRD2 away from the RANKL A′-A″ loop, disagreeing with modeling predicting that OPG^Y61 and RANKL^Y187 comprise a ring-stacking interaction. Our structural analysis reveals these residues to be separated by more than 5 Å. Despite the lack of contact with the RANKL A′-A″ loop, CRD2 engagement of the RANKL D-E loop represents a major binding element for OPG. Our structure suggests that the OPG CRD2 could be engineered to include RANKL A′-A″ loop contacts and thereby enhance binding affinity.

The OPG CRDs bind RANKL with an approximately 500-fold higher single-site affinity than the RANK CRDs, attributable to both a faster on-rate and a slower off-rate. Surprisingly, RANKL buries less surface area and makes fewer salt bridges and hydrogen bonds with OPG compared to RANK. The slower off-rate is probably a reflection of the more hydrophobic nature of the OPG binding cleft. The rearrangement of RANKL C-D loop residues, to form the OPG^F96 binding pocket, dictates partner predilection as evidenced by the decreased affinity of OPG^F96A and OPG^F96L for RANKL. Further, the key mouse and human residues involved in the RANKL C-D loop rearrangement are conserved in man. OPG^F96 corresponds to human OPG^F117, a residue mutated in a subset of patients with juvenile Paget’s disease, a disorder of accelerated bone resorption. Furthermore, we expressed recombinant wild-type OPG, OPG^F96A, and OPG^F96L proteins in mammalian cells. Thus, the human mutation OPG^F117L likely yields correctly folded protein that does not function efficiently as a decoy receptor. (For a list of known human disease causing mutations that occur at the RANK/RANKL or OPG/RANKL interface and structure-based predictions of their effects, see Table S7).

Our most novel observation is that conformational changes in the RANKL binding cleft permit preferential recognition of OPG. Previous studies established sequence diversity, amino acid insertions, receptor flexibility, and receptor placement as determinants of receptor/ligand selectivity in the TNF superfamily. The modified shape and chemical nature of RANKL’s binding groove, in response to OPG association, provides the first example wherein realignment of residues within the receptor binding site results in a shift to a higher-affinity conformation. Our crystallographic efforts demonstrate that plasticity of the receptor-binding cleft may be an additional feature regulating TNF/TNFR specificity.

EXPERIMENTAL PROCEDURES

Expression and purification of OPG, RANK, and RANKL

The CRDs of OPG (mouse strain C57BL/6) were expressed in High Five insect cells (Invitrogen Life Technologies) using a baculovirus expression system. A cDNA fragment encoding the N-terminal four CRDs immediately following the signal sequence of OPG (residues 22–197 of accession NP_032790) was inserted into a modified baculovirus shuttle vector such that the OPG sequence was located downstream of a bee-melittin-derived signal peptide followed by a thrombin protease cleavage site and a 6-His tag. Recombinant baculovirus was generated by cellfectin (Invitrogen Life Technologies) mediated co-transfection of the transfer plasmid with flashbac genomic DNA (Oxford Expression Technologies) into SF9 cells. Soluble OPG protein was recovered from the supernatants of infected High Five cells by Ni-NTA chromatography, cut with thrombin, and then purified by gel filtration chromatography. The purified protein was stored at 4°C for use in crystallization in buffer consisting of 25 mM HEPES (pH 7.5), 20 mM sodium chloride, and 0.01% sodium azide. The extracellular cytokine-binding domain of mouse RANK was also produced in High Five insect cells. The transfer vector encoded the endogenous signal peptide of RANK and all four CRDs (residues 1–198 of accession NP_033425) fused through a thrombin cleavage site to a BirA biotin-ligase recognition site and ending in a 6-His tag. Recombinant viral DNA was generated by site-specific transposition of the transfer plasmid onto a baculovirus bacmid in DH10bac E. coli cells (Invitrogen Life Technologies). DNA from the bacmid was transfected into SF9 cells to produce recombinant virus particles and the RANK protein recovered from infected cell supernatants by Ni-NTA chromatography. Alternately, 6HIS-tagged OPG protein (residues 22–191 of accession NP_032790) was recovered from supernatants of transiently transfected HEK293F cells (Invitrogen Life Technologies). A construct encoding the bee-meletin signal peptide, the OPG fragment and a 6His tag was inserted downstream of the CMV promoter in the IRES-GFP expresson vector pFM-1.2 (a gift of Dr. Filippo Mancia). The mammalian expressed OPG was used for binding studies because it more closely matched the RANK protein in length.

Recombinant mouse RANKL (residues 162–316 of NP_035743) was expressed as a soluble GST fusion protein in E. coli strain BL21-CodonPlus (DE3)-RIL cells (Agilent Technologies) as described previously (Lam et al., 2001). The GST-RANKL fusion protein was captured on glutathione sepharose and cleaved with prescission protease to release the RANKL ectodomain. The RANKL trimer was purified by gel filtration chromatography. Fractions containing mono-dispersed protein were pooled and kept at 4°C in sizing buffer consisting of 25 mM HEPES (pH 7.5), 150 mM sodium chloride, and 0.01% sodium azide. Selenomethionine labeling of RANKL was achieved by feedback inhibition of methionine biosynthesis prior to induction. In brief, a fresh colony containing the expression plasmid in BL21-CodonPlus (DE3)-RIL cells was seeded into 10 ml of LB containing 100 μg/ml carbenicillin at 37°C with shaking. When an OD₆₀₀ of 0.6 was reached, the 10 ml starter culture was used to inoculate 1000 ml of methionine deficient medium (Athena Enzyme Systems) at 30°C in a 2-liter Erlenmeyer flask (1:100 dilution). When the density again reached OD₆₀₀ of 0.6, 0.5 g feedback inhibition stock (made by combining the following amino acids: 0.1 g of lysine, 0.1 g threonine, 0.1 g phenylalanine, 0.05 g leucine, 0.05 g isoleucine, 0.05 g valine, and 0.05 g L-selenomethionine) was added to the culture. After 15 minutes, protein expression was induced by addition of 0.1 mM IPTG. The temperature was reduced to 25°C and the culture left for 16 hours before harvest.

Crystallization conditions

To characterize the OPG-RANKL binding interface, we purified the N-terminal ligand-binding fragment of OPG (CRDs 1 through 4) and mixed it with soluble RANKL trimer to form homogeneous complexes having a subunit ratio of 3:3. We recovered the complex containing fractions by size exclusion chromatography (Superdex 200) and concentrated the proteins to 15 mg/ml in 25 mM HEPES (pH 7.4) buffer containing 20 mM sodium chloride. Crystals of the OPG-RANKL complex grew in hanging drops at 20°C. Optimum crystal growth occurred in 100 mM sodium phosphate/citrate buffer (pH 4.1), 14% polyethylene glycol 8000 and 250 mM sodium chloride. The crystals were cryoprotected by a short soak in well solution containing 30% ethylene glycol. Similarly, RANK and RANKL were mixed at molar ratio of 4.5:3 and incubated at 4°C overnight before crystallization. Hexagonal-rod-shaped crystals formed in hanging drops at 20°C produced by mixing equal volumes of protein and well buffer containing 2.0 M sodium chloride, 100 mM EDTA, and 100 mM sodium acetate (pH 4.6). The crystals were cryoprotected using 25% xylitol in well solution. In both cases, the crystals were held in a stream of nitrogen gas (100°K) for data collection.

Structure determination and refinement

The OPG-RANKL crystals belong to spacegroup P6₃ with unit cell dimensions a = b = 109.6 Å, and c = 78.7 Å. The asymmetric unit consists of one monomer of OPG and one monomer of RANKL. Diffraction data were collected to 2.70 Å resolution at the Advanced Light Source Synchrotron (beamline 4.2.2) using a CCD detector (Noir-1) and processed with HKL2000 (Otwinowski and Minor, 1997). A combination of molecular replacement using our previously published RANKL model (Lam et al., 2001) and MAD technique using labeled RANKL allowed phasing of the structure. The RANK-RANKL crystals also belong to spacegroup P6₃ but have slightly larger unit cell dimensions a and b = 120.6 Å, c = 94.3 Å. Complete data to 2.7 Å resolution was collected using a Rigaku rotating anode generator equipped for Cu-Kα radiation with Osmic confocal mirrors and an R-Axis IV detector. Data sets were processed with DENZO and SCALEPACK. The RANKL-RANK complex structure was determined initially by molecular replacement using the RANKL coordinates as a search probe (Lam et al., 2001). After locating RANKL, a putative RANK model was fit into the electron density map with COOT (Emsley and Cowtan, 2004). The theoretical RANK coordinates were generated using 3D-PSSM (Kelley et al., 2000) using a related TNF receptor (p55, PDB 1TNR) as the model (Banner et al., 1993). The structure was refined using CNS (Brunger, 2007; Brunger et al., 1998) and refmac (Murshudov et al., 1997). Repeated iteration between manual rebuilding and error minimization was applied as guided by R-free. The final refined structures of both the OPG-RANKL and RANK-RANKL complexes have good crystallographic R factors and stereochemistry. The refinement statistics are summarized in Table I. Figures were prepared using PyMOL. Buried surface area was calculated using the PISA server (Krissinel and Henrick, 2007). NCONT (CCP4 Program Suite) (Collaborative Computational Project, 1994) was used to determine ligand-receptor contacts within 4.0 Å. HBPLUS (McDonald and Thornton, 1994) was used to assess neighbor interactions and the geometry of hydrogen bonds. Shape complementarity was calculated using Sc (Lawrence and Colman, 1993).

Quantitative and kinetic binding measurements

The binding of soluble monomeric OPG and RANK CRD regions to RANKL was examined using a Biacore T100 surface plasmon resonance-based instrument. Between 400 and 1000 response units of murine RANKL (cleaved and purified to remove the GST-affinity tag) were immobilized to the dextran matrix of a CM5 sensor chip by amine coupling in 10 mM sodium phosphate buffer (pH5.2) at a flow rate of 5 μl/min. The flow cell was activated at 5 μl/min using a 1:1 mixture of 0.1 M N-hydoxysuccinimide and 0.1 M 3-(N,N-dimethylamino)-propyl-N-ethylcarbondimide. An adjacent flow cell was coupled with neutravidin to serve as a control for nonspecific binding. The coupling reaction was stopped by injection of 1 M ethanolamine (pH8.5) for 5 min. Varying concentrations of monovalent receptor fragment (the N-terminal cytokine-binding CRDs 1 through 4) of RANK, OPG, or its variants were injected over the chip at a flow rate of 20 μl/min. The binding buffer consisted of 10 mM HEPES (pH7.4), 150 mM NaCl, 3 mM EDTA, and 0.05% v/v Surfactant P20. Regeneration was accomplished by passing binding buffer over the sensor surface until complete dissociation had occurred. For RANK, the sensorgrams were fit to a simple 1:1 reaction model using BIA evaluation software (Biacore, GE Healthcare). For OPG and its variants, the data were fit using a two state reaction model (binding with conformational change), which is consistent with the required conformational changes observed for bound OPG. The data fit the two state reaction model better. For example, for the wild-type OPG CRD binding data shown in figure 5A, values of Chi² were 0.231 for the two state model, and 8.48 for the 1:1 binding model. In addition the curve-fit residuals were always less than 10% for the two state model, but ranged as high as 35% for the 1:1 state model.

Osteoclast formation and tartrate-resistant acid phosphatase staining

Primary bone marrow macrophage cells (BMMs) were obtained by flushing bone marrow cells from femurs and tibia of 6–8 week-old C57BL/6 mice. After red blood cell lysis, the cells were resuspended in α-MEM supplemented with 10% (v/v) heat-inactivated FBS and cultured in 100 ng/mL M-CSF at 37°C for 4 days. Floating cells and debris were removed by washing with PBS and then the adherent BMMs lifted with trypsin-EDTA. Cells were plated into 96-well tissue-culture-treated plates under conditions to support further growth of BMMs (20 ng/mL M-CSF) or osteoclasts (20 ng/mL M-CSF plus 100 ng/mL RANKL) but with varying concentrations of monomeric RANK or OPG CRDs 1 through 4. The medium was changed every 2 days. On day 5, the cells were fixed with room-temperature paraformaldehyde (4% in PBS) and stained for TRAP expression using a kit (Sigma-Aldrich). TRAP-positive osteoclasts (≥3 nuclei) were counted using a light microscope.

HIGHLIGHTS.

• Crystal structures of the cytokine-binding regions of OPG and RANK bound to RANKL.

• High affinity OPG engagement is associated with a distinct RANKL binding cleft.

• Structural basis of how the OPG^F96L mutation promotes juvenile Paget’s disease.