Denosumab, a Fully Human Monoclonal Antibody to RANKL, Inhibits Bone Resorption and Increases BMD in Knock-In Mice That Express Chimeric (Murine/Human) RANKL

Authors


  • Parts of this manuscript were presented at the 27th Annual Meeting of the American Society for Bone and Mineral Research, Nashville, TN, September 23–27, 2005.

  • All authors are regular full-time employees who may own stock and/or stock options in Amgen.

Abstract

RANKL is a TNF family member that mediates osteoclast formation, activation, and survival by activating RANK. The proresorptive effects of RANKL are prevented by binding to its soluble inhibitor osteoprotegerin (OPG). Recombinant human OPG-Fc recognizes RANKL from multiple species and reduced bone resorption and increased bone volume, density, and strength in a number of rodent models of bone disease. The clinical development of OPG-Fc was discontinued in favor of denosumab, a fully human monoclonal antibody that specifically inhibits primate RANKL. Direct binding assays showed that denosumab bound to human RANKL but not to murine RANKL, human TRAIL, or other human TNF family members. Denosumab did not suppress bone resorption in normal mice or rats but did prevent the resorptive response in mice challenged with a human RANKL fragment encoded primarily by the fifth exon of the RANKL gene. To create mice that were responsive to denosumab, knock-in technology was used to replace exon 5 from murine RANKL with its human ortholog. The resulting “huRANKL” mice exclusively express chimeric (human/murine) RANKL that was measurable with a human RANKL assay and that maintained bone resorption at slightly reduced levels versus wildtype controls. In young huRANKL mice, denosumab and OPG-Fc each reduced trabecular osteoclast surfaces by 95% and increased bone density and volume. In adult huRANKL mice, denosumab reduced bone resorption, increased cortical and cancellous bone mass, and improved trabecular microarchitecture. These huRANKL mice have potential utility for characterizing the activity of denosumab in a variety of murine bone disease models.

INTRODUCTION

RANKL is an essential mediator of osteoclast formation, function, and survival.math image The catabolic effects of RANKL are inhibited by osteoprotegerin (OPG), a soluble decoy receptor that binds RANKL and thereby prevents activation of its receptor RANK, which is found on osteoclasts and osteoclast precursors.math image Human OPG constructs have been valuable research tools because they potently inhibit bone resorption in a variety of species including rats,math image pigs,math image monkeys,math image and humans.math image,math image The clinical development of OPG-Fc was discontinued in favor of denosumab (AMG 162), a fully human monoclonal antibody that binds and inhibits RANKL. Denosumab reduced biochemical markers of bone resorption in postmenopausal women with low bone massmath image and in patients with multiple myeloma or with bone metastases.math image Denosumab has a longer circulating half-life than OPG-Fc,math image which might account for its ability to suppress bone resorption in postmenopausal subjects for up to 6 mo after a single subcutaneous injection.math image This prolonged inhibition of bone resorption might also be related to the potential for denosumab to inhibit osteoclast formation, because time would be required to replenish mature bone-resorbing osteoclasts after the drug is cleared from circulation. The ability of OPG to inhibit osteoclastogenesis is well established,math image and we have examined the effects of denosumab and of OPG-Fc in osteoclastogenesis assays.

OPG has been shown to bind the cytotoxic ligand TRAIL with reasonably high affinity.math image,math image Whereas there is currently little evidence that OPG and TRAIL influence each other's functions in vivo,math image,math image a more selective RANKL inhibitor could have advantages in TRAIL-dependent systems that have yet to be modeled and tested. The binding potential of denosumab for human TRAIL, and for other TNF family members including CD40L was therefore examined in direct binding assays. The inability of denosumab to inhibit bone resorption in normal mice or rats suggests that denosumab fails to bind RANKL in those species. Human OPG was recently reported to bind directly to murine RANKL,math image which would explain the ability of human OPG to reduce bone resorptionmath image and to increase BMDmath image and bone strength in mice.math image The affinity of denosumab and of human OPG-Fc for human versus murine RANKL was therefore examined in direct binding assays.

Apart from humans, the cynomolgus (cyno) monkey is the only other species reported to respond to denosumab. Denosumab was shown to suppress bone resorption in gonad-intact cynos, leading to increases in bone volume, density, and strength.math image The ovariectomized (OVX) cyno is the only preclinical bone disease model used to date to examine the effects of denosumab, and preliminary results were recently reported. Denosumab markedly reduced bone turnover in association with increased bone volume, density, and strength in these OVX cynos.math image,math image There is obvious utility in developing a small animal model for evaluating denosumab in other preclinical models of bone disease. The strategic foundation for such an effort was based on the ability of human RANKL to bind murine RANKmath image and to stimulate bone resorption in mice.math image These results suggested that bone resorption would occur in mice expressing an extracellular fragment of human RANKL in a manner that could be inhibited by denosumab. We therefore used knock-in technologymath image to replace the coding regions of the fifth exon of murine RANKL with its human ortholog. Mice that are homozygous for the resulting knock-in construct exclusively express a chimeric (murine/human) RANKL gene product, which remained under the control of normal endogenous regulatory elements. We used these huRANKL mice to examine the primary pharmacodynamics of denosumab, and we describe for the first time the effects of denosumab on bone histomorphometry endpoints and on cancellous and cortical bone density and architecture in mice.

MATERIALS AND METHODS

Protein reagents

All protein reagents were produced recombinantly at Amgen using Chinese hamster ovary (CHO) cells, unless otherwise noted. Human OPG-Fc is a recombinant fusion protein consisting of OPG (residues 22–194) fused to the Fc portion of human IgG1 to confer a longer circulating half-life.math image,math image Denosumab is a fully human monoclonal IgG2 antibody against human RANKL that was produced using transgenic Xenomouse technology.math image,math image CHO cells were used to express an active fragment of human recombinant RANKL (amino acids 143–317), which includes the complete TNF domain encoded by exon 5 of the human RANKL gene.math image Murine RANKL (158–316) was also expressed in bacteria and purified from the soluble fraction as described previously.math image These RANKL proteins were used in studies described in Fig. 1. Bacterially expressed recombinant human TRAIL comprised amino acids 95–281 and contained an N-terminal FLAG tag. Anti-FLAG M2 monoclonal antibody was purchased from Sigma-Aldrich. The relative affinity and potency of the chimeric (Mu/Hu) form RANKL expressed by huRANKL mice was studied by cloning and expressing FLAG-tagged versions of the ectodomains of human (159–317), murine (158–316), and chimeric RANKL (158–317) in bacteria. Chimeric RANKL comprised murine sequence 158–176 and human sequence 178–317, as described in Fig. 2. These FLAG-tagged RANKL proteins were used in experiments described in Fig. 3. Murine RANK-Fc was purchased from R&D Systems.

Figure Figure 1.

Specificity of denosumab binding and its effects on osteoclastogenesis. (A) Different concentrations of denosumab or human OPG-Fc were added to plates coated with human or murine RANKL, and their binding was monitored by colorimetric detection of HRP-conjugated goat anti-human IgG. (B) Different concentrations of denosumab or human OPG-Fc were added to plates coated with human flag-tagged TRAIL (95–281). Binding was monitored as described above. (C) RAW murine macrophages were stimulated to form osteoclasts by the addition of human RANKL and murine M-CSF. TRACP-5b was measured after 5 days in the presence of varying concentrations of human OPG-Fc or denosumab. *Significantly different from 0 nM control, p < 0.05. (D) Effects of denosumab or OPG-Fc on hypercalcemia in normal mice challenged twice daily with subcutaneous injections of human RANKL (0.5 mg/kg) or PBS (n = 6/group). Single subcutaneous injections of denosumab or human OPG-Fc were delivered after the first injection of RANKL, and blood ionized calcium was monitored daily for 4 days. Doses listed are in mg/kg. *Significantly different from RANKL + PBS, p < 0.05; #significantly different from RANKL + either dose of denosumab, p < 0.05 by ANOVA.

Figure Figure 2.

Creation and characterization of huRANKL knock-in mice. (A) Schematic representation of the genomic organization of the murine RANKL locus. Introns are shown as thin black lines. Brown boxes represent the coding region of mouse RANKL exons, red boxes represent the coding region of human RANKL exon 5, and gray boxes represent 3′ and 5′ untranslated regions of the mouse RANKL gene. PGK-neo cassette was placed in the same orientation immediately after the 3′ untranslated region of the gene. LA, long arm; SA short arm. (B) Amino acid alignment of human and murine RANKL. Amino acid residues highlighted in red represent in the knock-in where residues of murine exon 5 were replaced with human exon 5. The predicted amino acid sequence of the full-length chimeric mouse/human RANKL molecule is shown in uppercase letters, with lowercase letters denoting differences between the species. The arrow denotes the start site for recombinant RANKL molecules that retain full biological activity.math image This region contains 10 β-strands that are labeled as B−I and underlined. The protease cleavage site for the shedding of soluble RANKL follows residue 138 in mouse.math image

Figure Figure 3.

Binding and potency of chimeric RANKL. (A) Binding of murine, chimeric (Mu/Hu), or human RANKL to immobilized murine RANK-Fc. (B) Potency of murine, chimeric (Mu/Hu), and human RANKL in a murine osteoclastogenesis culture system. Mu/Hu RANKL exhibited potency (EC50 = 4 ng/ml) that was intermediate to that of human (EC50 = 2.4 ng/ml) and murine (EC50 = 12 ng/ml) RANKL. Data represent means ± SE of triplicate cultures. (C) Potency of murine, chimeric (Mu/Hu), or human RANKL in stimulating bone resorption in normal mice that were challenged twice daily with either vehicle (PBS) or varying doses of RANKL for 3.5 days (n = 6/group). Serum was obtained from blood collected 3 h after the final morning injection of RANKL and analyzed for total serum calcium (an index of bone resorption). Data represent means ± SE. *Significantly different from vehicle control by ANOVA, p < 0.05. (D) Binding of denosumab to immobilized murine, chimeric (Mu/Hu), or human RANKL.

Binding properties of denosumab and human OPG-Fc for RANKL and TRAIL

To examine RANKL binding, Costar EIA/RIA 96-well plates were coated with 75 μl/well of recombinant human RANKL (amino acids 143–317) or murine RANKL (158–316) at 3 μg/ml in PBS. After overnight incubation at 4°C, RANKL solutions were removed, and plates were blocked with 5% chicken serum (Gibco) in PBST (PBS plus 0.05% Tween 20) and incubated at room temperature for 3 h with agitation. Plates were washed with 1× KP wash solution (Kirkegaard-Perry Laboratories) in distilled water. Human OPG-Fc or denosumab was serially diluted in PBST and added to the RANKL-coated plates. Plates were incubated for 7 h at room temperature with agitation and washed with 1× KP wash solution. Goat anti-human IgG (Fc)-horseradish peroxidase (HRP) (Pierce) was diluted 1:3000 in 5% chicken serum in PBST (C-PBST) and added to the wells. Plates were incubated for 1 h at room temperature with agitation and washed with 1× KP. Undiluted ABTS substrate was added and the plate was incubated at room temperature. Color development was stopped after 4 min by addition of 1% SDS and measured at 405 nm.

To examine TRAIL binding, Costar EIA/RIA plates were coated with anti-FLAG M2 monoclonal antibody (3 μg/ml) in PBS. After overnight incubation at 4°C, the antibody was removed and plates were blocked and washed as described above. FLAG-human TRAIL (2 μg/ml in C-PBST) was added to each well and incubated for 1–2 h, and plates were washed with 1× KP wash solution. Serial dilutions of human OPG-Fc or denosumab in C-PBST were added to wells, and plates were agitated for 2 h at room temperature. After three washes with 1× KP solution, a 1:5000 dilution of the anti-human IgG (Fc)-HRP conjugate was added to each well, and plates were agitated for 1 h. After three final washes, ABTS substrate was added, and color development was monitored as described above.

Binding affinity of denosumab and OPG-Fc for human RANKL

Binding affinities for denosumab and human OPG-Fc were assessed through solution equilibrium binding analysis using a KinExA 3000 system (Sapidyne Instruments). Briefly, Reacti-Gel 6× beads were precoated with 20 μg/ml of human RANKL at 4°C overnight, blocked with 1 mg/ml BSA for 2 h, and washed three times in PBS. Denosumab (50 pM) or human OPG-Fc (10 pM) were incubated with various concentrations of soluble human RANKL (0–5 nM) at room temperature for >6 h to allow for equilibrium binding before being passed through the RANKL-coated beads. The binding of free denosumab or OPG-Fc to the beads was quantified by fluorescently labeled (cyanine Cy5 dye) goat anti-human antibody.

Effects of denosumab in osteoclastogenesis assays

Osteoclastogenesis was monitored in cultures of murine RAW 264.7 macrophages that served as osteoclast precursors, as previously described.math image Briefly, 5 × 103 RAW cells/ml were cultured at 37°C with murine M-CSF (30 ng/ml; R&D Systems) and human RANKL (143–317) (30 ng/ml) for 5 days. Triplicate cultures were treated every 3 days with fresh media containing various concentrations of human OPG-Fc, denosumab, or vehicle (α-MEM + 10% FBS). Osteoclast formation was measured by quantifying the presence of TRACTP (optical density, 405 nm) using a solution assay (Sigma), as previously described.math image

Effects of denosumab or human OPG-Fc on hypercalcemia induced by human RANKL

Young (4 wk old) male BDF1 mice were injected with human RANKL (143–317; 0.5 mg/kg) or with PBS (n = 6/group) twice daily (morning and afternoon). Immediately after the first RANKL challenge, mice were treated with a single subcutaneous injection of either human OPG-Fc (3 mg/kg) or denosumab (0.3–10 mg/kg). Blood ionized calcium was measured exactly 3 h after each morning injection of RANKL, using a calcium/pH analyzer (Model 634; Chiron Diagnostics).

Generation and characterization of human RANKL knock-in mice

Based on the genomic structure of mouse RANKL,math image exon 5 was identified as containing ∼90% of the minimal extracellular receptor binding portion of the molecule needed to fully promote osteoclast formation.math image This exon comprises all but the first of the 10 β strands that form the homotrimeric ligand and which are conserved among TNF family members.math image Fugure 3 describes the strategy for replacing the coding region of exon 5 of the murine RANKL gene with its human homolog, along with the primary amino acid sequence of the resulting mouse/human RANKL chimeric molecule. A murine BAC clone containing mouse RANKL genomic DNA was identified by PCR screening using primer sets directed toward mouse exons 3, 4, and 5. The human RANKL knock-in construct was assembled by directed insertion of three PCR products described below into the pAMGENKO3 targeting vector using engineered restriction sites. A 1.4-kb PCR product (short arm) of the mouse intron following exon 5 and a 4.9-kb PCR product (long arm) of the intron between mouse exon 4 and exon 5 were inserted into the vector, along with a 1.9-kb sequential PCR product that contained a 0.25-kb piece of the mouse intron between exon 4 and exon 5 (introduced BstZ17I), a 0.4-kb piece of the coding region of exon 5 from human RANKL cDNA and a 1.24-kb piece of the 3′ untranslated region of mouse exon 5. Plasmid DNA containing the human RANKL knock-in construct was transfected into GS-1 embryonic stem cells (129SvJ) (Genome Systems). Incorporation of the knock-in gene by ES cells was confirmed by Southern blot analysis (data not shown), and ES cells containing the knock-in gene were injected into 2.5-day C57Bl/6 (Taconic) blastocysts. The line was expanded by multiple back-crosses, breeding male chimeras with female black Swiss/129 mice. Mice that were homozygous for the knock-in were identified by PCR-based genotyping.

Histopathologic and phenotypic assessments were conducted using male and female huRANKL mice (13–16 wk old) and wildtype (WT) littermates (n = 8–10/sex/genotype). Body weight was recorded at necropsy. The heart, liver, kidneys, spleen, thymus, coronary artery, and mesenteric lymph nodes were harvested. The liver, spleen, heart, kidney, and thymus were weighed. All harvested organs were fixed in zinc formalin, sectioned, stained with H&E, and analyzed by a board-certified veterinary pathologist. Blood drawn at necropsy was analyzed for complete blood counts (Advia 120 Analyzer; Bayer Corp.), clinical chemistry analyses (BUN, AST, ALT, TBil, Alb, Ca, Phos; Olympus AU Automated Analyzer), and serum TRACP-5b, RANKL, and OPG. Murine TRACP-5b was measured by ELISA (IDS). Human and murine RANKL and murine OPG were measured using Luminex antibody-immobilized microbead kits (Linco Research). Additional standard curves were generated to assess the ability of the Luminex polyclonal anti-human RANKL capture antibody to recognize human (158–317) versus chimeric (158–317) recombinant RANKL standards (Amgen). The skeletal phenotype of huRANKL mice was evaluated with 8- to 12-wk-old male huRANKL mice and WT littermates (n = 8/group). Serum was obtained from retro-orbital blood drawn from anesthetized mice, and TRACP-5b was measured as described above. The proximal tibial metaphysis was harvested, decalcified, and processed for histomorphometry analyses as described below. The left femur was harvested and analyzed by μCT as described below.

Binding and potency of murine, chimeric, and human RANKL

Binding affinity for murine RANK

To assess the relative binding affinity of chimeric RANKL for murine RANK, 96-well EIA plates were coated overnight at 4°C with 1 μg/ml murine RANK-Fc, washed, and blocked. Serial dilutions of FLAG-tagged forms of RANKL (murine, chimeric, and human) were added and incubated for 1–2 h at room temperature, 37°C or 42°C. After washing, the level of RANKL binding was assessed after incubation with an HRP-conjugated anti-FLAG M2 monoclonal antibody, washing, and colorimetric detection using TMB as a substrate. Absorbance was monitored at 450 nm (SpectraMax 340). Data are expressed as percent of maximum binding for each protein.

Binding affinity for denosumab

Polystyrene plates (384-well) were precoated overnight at 4°C with 2 μg/ml of murine, human, or chimeric RANKL. Plates were washed with 1× KP solution and blocked with 1% BSA in 1× PBS for 1 h at room temperature. After washing, varying concentrations of denosumab were added for 1 h at room temperature, and wells were washed. HRP-conjugated anti-human IgG (Abcam) was added for 1 h at room temperature to detect denosumab. Tetramethylbenzidine (TMB) substrate (KPL) was added and the colorimetric reaction was stopped with 2 M sulfuric acid. Absorbance was read at 450 nm with a SpectraMax M5 plate reader (Molecular Devices).

Osteoclastogenesis

The osteoclastogenic potency of chimeric (murine/human) RANKL was evaluated in an assay relying on bone marrow cells from C3H/HeN mice (Charles River Laboratories) as osteoclast precursors. Nonadherent bone marrow cells were obtained as previously describedmath image and cultured (in triplicate) in the presence of MEM supplemented with 10% heat-inactivated FCS, murine colony-stimulating factor (CSF)-1 (30 ng/ml; R&D Systems), and varying concentrations of murine, human, or chimeric RANKL for 5 days. By day 5, large multinucleated osteoclast-like cells were evident in RANKL-treated cultures by light microscopy. Cultures were gently rinsed, and adherent cells were stained for TRACP activity using a solution assay (Sigma) as previously described.math image

In vivo potency in normal mice

Female BDF1 mice (8–9 wk of age, n = 6/group) were challenged twice daily (morning and evening; subcutaneously) with either vehicle (PBS) or varying doses of human, murine, or chimeric RANKL for 3.5 days. Serum was prepared from retro-orbital blood collected exactly 3 h after the final morning injection of RANKL. Total serum calcium was measured with an Olympus AU Automated Analyzer.

Pharmacokinetics and pharmacodynamics of denosumab in huRANKL mice

Female HuRANKL mice (4–5 mo old, n = 5–6/group) were injected once (subcutaneously) with vehicle (PBS) or with denosumab (0.2, 1.0, or 5.0 mg/kg). Serum was obtained from blood drawn 1, 4, and 7 days later for measurement of the serum TRACP-5b (IDS). Serum denosumab concentrations were assessed by coating 384-well polystyrene plates with 2 μg/ml human RANKL (143–317) by overnight incubation at 4°C. Standard curves were generated using denosumab diluted in PBS (9.8 ng/ml to 10 μg/ml). Wells were blocked for 1 h at room temperature in a solution of 1% BSA in 1× PBS, washed, and followed by the addition of standards, study samples (serially diluted in PBS), and blanks for 1 h at room temperature. Wells were washed, and an HRP-conjugated monoclonal anti-human IgG detection antibody (Abcam) was added to the plate for 1 h at room temperature. After a final wash, a TMB-peroxidase substrate was added. The colorimetric reaction was stopped with 2 M sulfuric acid, and absorbance was measured at 450 nm using a SpectraMax M5 plate reader (Molecular Devices). Four-parameter curve-fitting software (Softmax Pro) was used to convert optical density values to denosumab concentrations. Pretreatment serum was used to determine nonspecific binding, and the sensitivity of the assay was ∼10 ng/ml.

Skeletal effects of denosumab or OPG in young wildtype and huRANKL mice

Young male and female homozygous huRANKL mice (6–8 wk old, n = 3–4/group) were treated with vehicle (PBS), denosumab, or human OPG-Fc (both at 5 mg/kg, twice weekly, subcutaneously) for 3 wk. Male and female WT littermate controls (n = 3–4/group) were treated with PBS or denosumab (5 mg/kg, twice weekly). The right tibia was harvested and decalcified for histomorphometry as previously described.math image Histomorphometry parameters were assessed on trichrome-stained sections of the proximal tibial metaphysis using Osteomeasure (Osteometrics) bone analysis software. Analyses were performed in the trabecular region adjacent to the growth plate, avoiding primary spongiosa. Four fields were analyzed per section at ×20 magnification. Immunohistochemistry was used to visually assess the skeletal disposition of denosumab (a fully human IgG2) in huRANKL mice. Proximal tibia sections from denosumab- and PBS-treated huRANKL mice were deparaffinized, blocked with CAS Block (Zymed Laboratories), and incubated with rabbit anti-human IgG (Jackson Laboratory) at 1:4000 for 1 h. This IgG was detected by biotinylated goat anti-rabbit IgG (Vector Laboratories) and followed with HRP ABC Complex (Vector Laboratories). The reaction was visualized with DAB substrate (DAKO). All sections were counterstained with hematoxylin.

Skeletal effects of denosumab in adult huRANKL mice

Adult female homozygous huRANKL mice (10 mo old) were injected subcutaneously with PBS or denosumab at 10 mg/kg once weekly for 3 wk (n = 6/group). WT control mice (n = 6) were injected with PBS once weekly for 3 wk. Mice were injected intraperitoneally with calcein (25 mg/kg) and demeclocycline (25 mg/kg) on days −9 and −8 and days −3 and −2 (respectively), relative to necropsy. Blood was collected before death to assess serum TRACP-5b (murine ELISA; IDS), and the right tibia was harvested for histomorphometry. Static bone histomorphometry parameters were assessed with trichrome-stained undecalcified methylmethacrylate sections of the proximal tibia, whereas dynamic parameters were assessed with unstained sections using Osteomeasure (Osteometrics) bone analysis software. Histomorphometry analyses were performed in the trabecular region as described above. The left femur was harvested for μCT analysis using an eXplore Locus SP MicroCT System (GE Healthcare). Bones were scanned at 0.5° rotations for 200° (80 kVp, 80 μA), calibrated using a hydroxyapatite density phantom and reconstructed to yield images with a voxel size of 18.2 μm. The central 10% of the femur diaphysis was analyzed for cortical volumetric BMD (vBMD), periosteal perimeter, endocortical perimeter, and cortical area. Trabecular analyses at the distal femur metaphysis were determined for an endocortical-contoured region comprising 10% of femur length, 0.2 mm proximal to the growth plate. Image rendering and analyses were generated using a threshold of 320 mg/ml for trabecular bone and 640 mg/ml for cortical bone (MicroView v2.1). All animal protocols were reviewed and approved by Amgen's Institutional Animal Care and Use Committee.

Statistical analyses

All results were expressed as the mean ± SE. A one-way ANOVA followed by Dunnett's comparison was used to determine significant differences between drug-treated or vehicle-treated cultures or animals, using p < 0.05 to indicate significance. Serum analysis of RANKL and TRACP-5b in huRANKL knock-in mice versus WT littermate controls used Student's t-test. All statistical analyses were done using JMP software v5.1 (SAS Institute, Cary, NC, USA).

RESULTS

Denosumab binding specificity and affinity for human RANKL

An enzyme immunoassay showed that denosumab bound to a human RANKL fragment (143–317) that included the complete TNF-like domain encoded by exon 5 of the human RANKL gene (Fig. 1A). Human OPG-Fc showed similar binding to human RANKL and to murine RANKL. Denosumab showed no binding to murine RANKL (Fig. 1A). Denosumab also failed to bind immobilized TRAIL under conditions wherein human OPG-Fc bound TRAIL with an EC50 of ∼1 ng/ml (Fig. 1B). In similar enzyme immunoassays, denosumab failed to bind to other TNF family members including TNF-α, TNF-β, and CD40L (data not shown). The binding affinities of denosumab and human OPG-Fc for human RANKL were assessed by solution equilibrium binding analysis, which showed dissociation equilibrium constants (KD) of 3 pM for denosumab and 0.3 pM for OPG-Fc.

Denosumab suppression of osteoclastogenesis

Denosumab and human OPG-Fc inhibited the ability of human RANKL (143–317) to stimulate the formation of osteoclasts derived from murine RAW 264.7 cells, with IC50 values of 1.64 and 1.15 nM, respectively (Fig. 1C). Denosumab had no inhibitory effects on osteoclastogenesis in a fully murine co-culture system, wherein nonadherent bone marrow cells were stimulated to form osteoclasts by murine RANKL that was provided by ST-2 stromal cells (data not shown).

Inhibitory effects of denosumab in normal mice treated with human RANKL

Twice-daily injections of a soluble human RANKL fragment (143–317) caused a significant increase in blood ionized calcium from day 2 through day 4 (p < 0.05 versus PBS control; Fig. 1D). Single injections of OPG-Fc or denosumab caused significant reductions in human RANKL-mediated hypercalcemia from days 2 to 4 (p < 0.05 versus PBS control). Blood ionized calcium on days 2–4 was significantly lower in RANKL-challenged mice treated with denosumab (3 and 10 mg/kg) compared with those treated with OPG-Fc (3 mg/kg; p < 0.05).

Generation and characterization of human RANKL knock-in mice

Having shown denosumab efficacy in mice rendered hypercalcemic by a human RANKL (143–317) fragment, we inserted coding sequences for the majority of this RANKL domain (all of exon 5) into the mouse genome by targeted homologous recombination (Fig. 2). Simultaneous deletion of the endogenous mouse RANKL exon 5 was achieved by transfecting the chimeric targeting construct into mouse embryonic stem cells. PCR-based genotyping from tail DNA of resulting homozygous huRANKL mice showed the expected substitution of their murine exon 5 with a human exon 5 (data not shown). Phenotypic analysis of 13- to 16-wk-old animals indicated that homozygous huRANKL mice were slightly smaller than WT controls, with average body weights of 39.8 ± 1.1 (SE) g for male WT mice versus 34.7 ± 1.2 g for male huRANKL mice (p < 0.05) and 32.7 ± 1.0 g for female WT mice versus 26.3 ± 0.6 for female huRANKL mice (p < 0.05). Hematology parameters (including hemoglobin levels and numbers of lymphocytes, monocytes, neutrophils, and platelets) were normal in huRANKL mice, as were blood clinical chemistry parameters (blood urea nitrogen [BUN], aspartate amino transferase [AST], alamine amino transferase [ALT], total bilirubin [TBil], alkaline phosphatase [ALP], calcium [Ca], and phosphate [Phos]; data not shown). Relative weights of major organs (liver, heart, kidneys) were similar in WT and huRANKL mice (data not shown). There was also no difference in the relative weights of the spleen or thymus (data not shown). Histopathologic assessment of H&E-stained sections from each of these organs and of H&E-stained sections from bone, aorta, mesenteric lymph nodes, and Peyer's patches showed no differences between WT and huRANKL mice (data not shown). HuRANKL mice exhibited normal nursing behavior and normal tooth eruption.

The 199 amino acid soluble form of RANKL (found in serum) would be ∼80% human in huRANKL mice. We therefore tested a commercially available human RANKL assay (Luminex; Linco Research) for its ability to measure serum RANKL in huRANKL mice. The polyclonal anti-RANKL capture antibody included with this human assay kit exhibited an average cross-reactivity of 100% between human (158–317) and chimeric (158–317) RANKL standards (two experiments; data not shown), indicating that this assay is suitable for measuring RANKL in huRANKL mice. Serum RANKL in male and female huRANKL mice was 198.3 ± 24.2 (SE) pg/ml, whereas this human assay was unable to detect any RANKL in serum from WT mice and showed very weak cross-reactivity (∼5%) with murine RANKL (157–316) (data not shown). Serum RANKL in WT mice (measured with a murine RANKL assay) was 258.6 ± 11.8 pg/ml. Serum OPG was similar in huRANKL mice (557.7 ± 24.6 pg/ml) versus WT littermates (621.3 ± 46.1 pg/ml), whereas serum TRACP-5b was significantly lower in these 13- to 16-wk-old huRANKL mice (9.93 ± 1.32 U/liter) compared with their WT controls (17.63 ± 2.73 U/liter); p < 0.05 by ANOVA). These data represent combined values from equal numbers of males and females, and there were no sex-related differences in any of these variables (data not shown).

The skeletal phenotype of slightly younger huRANKL mice (8–12 wk old) and WT littermates was assessed by biochemical, histomorphometric, and densitometric methods. These huRANKL mice exhibited a trend toward lower serum TRACP-5b (−32%, p < 0.070) and had significantly lower osteoclast and osteoblast surface compared with WT controls (Table 1). No significant differences were observed between these huRANKL mice and WT controls for bone volume or osteoblast surface at the proximal tibial metaphysis. No significant differences were observed in μCT-derived parameters for the distal femur (vBMD and bone volume fraction) or the femur diaphysis (cortical area, endocortical perimeter, or periosteal perimeter; Table 1).

Affinity and potency of murine, chimeric, and human RANKL

Receptor binding of chimeric (murine/human) RANKL was assessed by its ability to bind murine RANK-Fc. Chimeric RANKL exhibited dose-dependent binding to RANK-Fc, and this binding was similar to that shown for murine or human RANKL to RANK-Fc (Fig. 3A). These analyses were conducted at room temperature, and similar findings were observed at 37°C and 42°C (data not shown). Murine, chimeric, and human RANKL also appeared to bind OPG equally well, because binding competition studies showed that soluble murine or human OPG-Fc competed equally well for RANKL binding to immobilized RANK-Fc (data not shown).

The in vitro potency of these RANKL constructs was evaluated in a murine osteoclastogenesis assay, and chimeric RANKL exhibited potency (EC50 = 4 ng/ml) that was intermediate to that of human (EC50 = 2.4 ng/ml) and murine (EC50 = 12 ng/ml) RANKL (Fig. 3B). The in vivo potency of these constructs was evaluated by their injection into normal BDF1 mice and monitoring of serum calcium. In this system the chimeric RANKL showed slightly reduced potency relative to human or murine RANKL (Fig. 3C). The binding of chimeric versus human RANKL to denosumab was similar (Fig. 3D).

Pharmacokinetics and pharmacodynamics of denosumab in huRANKL mice

Single subcutaneous injections of denosumab resulted in dose-dependent increases in serum drug levels, which were measurable for at least 7 days (Fig. 4A). The highest dose of denosumab (5 mg/kg) resulted in significant suppression of serum TRACP-5b as early as 24 h after administration (Fig. 4B), and TRACP-5b remained suppressed on days 4 and 7 days at levels 72–76% below those of vehicle-treated controls (p < 0.05). A denosumab dose of 1 mg/kg reduced serum TRACP-5b by 64–65% on days 4 and 7 (p < 0.05 versus Veh control). The lowest dose tested (0.2 mg/kg) significantly suppressed TRACP-5b on day 4 (p < 0.05), followed by recovery of serum TRACP-5b by day 7, coincident with a significant reduction in serum denosumab concentration (Figs. 4A and 4B). These analyses indicated that denosumab doses of 1–5 mg/kg, delivered once or twice weekly, would be expected to cause strong suppression of bone resorption in huRANKL mice. Higher doses (5 or 10 mg/kg, once or twice weekly) were used in subsequent studies with the goal or overwhelming any possible immune responses that might be raised against the fully human denosumab protein in repeated dose settings.

Figure Figure 4.

Pharmacokinetics and pharmacodynamics of denosumab in huRANKL mice. Female huRANKL mice (n = 5–6/group) were given single subcutaneous injections of vehicle (PBS) or denosumab at 0.2–5.0 mg/kg. Serum was obtained from blood drawn before treatment (baseline, BL) and on days 1, 4, and 7 after treatment. (A) Serum levels of denosumab were measured by an ELISA wherein serum samples were added to plates precoated with human RANKL, and denosumab was detected with a monoclonal anti-human IgG. (B) Bone resorption was assessed by measuring serum TRACP-5b. *Significantly different from Veh control, p < 0.05 by ANOVA.

Effects of denosumab or OPG in young wildtype or huRANKL mice

Denosumab or OPG treatment of huRANKL mice was associated with >95% reduction in OcS/BS and a 2-fold increase in BV/TV (p < 0.05 versus PBS controls; Fig. 5A and 5C). Denosumab treatment of WT mice was not associated with any significant changes in any measured histomorphometry parameter. Immunohistochemistry was performed to provide a visual assessment of the skeletal disposition of denosumab in huRANKL mice. An anti-human IgG antibody showed no staining in bone sections from PBS-treated huRANKL mice. In denosumab-treated huRANKL mice, strong staining was apparent within blood vessels found in the medullary cavity and within blood vessels penetrating the cortical bone (Fig. 5D). There was no apparent staining of bone matrix or bone surfaces, suggesting that denosumab is primarily a circulating soluble protein.

Figure Figure 5.

Effects of denosumab on bone histomorphometry in young (6–8 wk old) huRANKL mice. HuRANKL mice (n = 3–4/group) were treated with vehicle (PBS), denosumab, or human OPG-Fc (5 mg/kg, SC, twice per week) for 3 wk. WT littermate controls were treated with PBS or denosumab. Histomorphometry was performed on decalcified sections of the proximal tibial metaphysis. (A) Osteoclast surface (OcS/BS) was significantly lower in huRANKL mice treated with denosumab or OPG-Fc, whereas denosumab failed to reduce OcS/BS in WT controls. (B) Osteoblast surface (ObS/BS) was statistically similar in all groups. (C) Trabecular bone volume (BV/TV) was significantly greater in huRANKL mice treated with denosumab or OPG-Fc, whereas denosumab failed to increase BV/TV in WT controls. For A-C, * represents significant differences from huRANKL mice treated with PBS, p < 0.05. (D) Immunohistochemical assessment of denosumab in the proximal tibia of a huRANKL mouse. Sections were incubated with rabbit anti-human IgG. (Left panel) Negligible specific staining was observed in bones from huRANKL mice treated with PBS. *Note for reference the large unstained blood vessel in the marrow of the diaphysis. (Right panel) In a denosumab-treated huRANKL mouse, strong staining was observed within a major blood vessel in the marrow diaphysis, and within a vessel penetrating the cortex (circle).

Effects of denosumab in adult huRANKL knock-in mice

The effects of denosumab were evaluated in more skeletally mature adult (10 mo old) female huRANKL mice. PBS-treated WT female littermates were included to assess the skeletal phenotype of huRANKL mice in adulthood. At this age, huRANKL mice (36.0 ± 2.5 g) had body weights similar to their WT littermates (37.5 ± 3.5 g). Denosumab treatment of huRANKL mice resulted in >95% reductions in serum TRACP-5b at all time points (Fig. 6A; p < 0.05 versus Veh). Denosumab treatment of huRANKL mice led to significant reductions in serum osteocalcin on days 14 and 21 of treatment (Fig. 6B). Histomorphometry of the proximal tibia at week 3 showed no statistically significant differences in static or dynamic parameters between vehicle-treated huRANKL mice and their WT vehicle-treated littermates (Figs. 6C and 6H), although trabecular bone volume showed a trend toward higher values in huRANKL mice. Denosumab treatment of huRANKL mice was associated with markedly lower osteoclast surface, significantly lower osteoblast surface, and significantly greater trabecular bone volume (all p < 0.05 compared with vehicle-treated huRANKL mice; Figs. 6C and 6E). Denosumab also resulted in significant reductions in trabecular mineralizing surface, mineral apposition rate, and bone formation rate as a reflection of normal coupling between bone resorption and formation on these remodeling surfaces (Figs. 6F and 6H).

Figure Figure 6.

Effects of denosumab in adult (10 mo old) huRANKL mice (n = 6/group). (A) Serum TRACP-5b was significantly reduced by denosumab at all time points during treatment. (B) Serum osteocalcin was significantly reduced by denosumab on days 14 and 21. (C-H) Histomorphometry of the proximal tibial metaphysis after 3 wk of treatment showed significant denosumab-related reductions in osteoclast surface, osteoblast surface, mineralizing surface, mineral apposition rate, and bone formation rate (BFR/BS) and significantly greater trabecular bone volume. *Significantly different from vehicle-treated huRANKL mice by ANOVA (p < 0.05).

Table Table 1.. Skeletal Phenotypic Assessment of 8- to 12-wk-old Male huRANKL Knock-in Mice and WT Male Littermates
original image

μCT analysis indicated that the femurs of PBS-treated huRANKL mice had significantly greater trabecular number in the distal metaphysis and greater cortical area at the midshaft compared with PBS-treated WT controls (Fig. 7). Whereas there were no statistically significant differences in periosteal or endocortical perimeters among the three groups, periosteal perimeter was 7.3% greater and endocortical perimeter was 5.3% lower in PBS-treated huRANKL mice versus WT controls (p = 0.31 and 0.74, respectively, by ANOVA; data not shown). Total femur length was also similar in PBS-treated WT mice (mean ± SE = 15.5 ± 0.3 mm), PBS-treated huRANKL mice (15.5 ± 0.3 mm), and denosumab-treated huRANKL mice (15.9 ± 0.1 mm; all p > 0.05). Denosumab treatment was associated with significantly greater cortical vBMD, trabecular vBMD, trabecular bone volume fraction, and trabecular thickness (p < 0.05 versus vehicle-treated huRANKL controls; Figs. 7A and 7F).

Figure Figure 7.

μCT assessment of femurs from adult (10 mo old) female huRANKL mice treated with denosumab (10 mg/kg) once per week for 3 wk (n = 6/group). (A-D) Trabecular bone within the distal femoral metaphysis was analyzed by μCT to assess volumetric BMD (vBMD), trabecular bone volume fraction (BVF), trabecular thickness, and trabecular number. (E and F) Cortical bone in the femur diaphysis was analyzed by μCT to assess cortical area and vBMD. *Significantly different from vehicle-treated huRANKL mice by ANOVA (p < 0.05).

DISCUSSION

RANKL is a key mediator of bone resorption, and the inhibition of RANKL by its soluble decoy receptor OPG results in suppression of bone resorption and increases in cortical and trabecular bone volume, density, and strength in mice, rats, and nonhuman primates.math image Recombinant human OPG constructs have also been shown to suppress bone resorption in postmenopausal women with low bone massmath image and in patients with multiple myeloma or breast cancer.math image The clinical development of OPG was discontinued in favor of denosumab, a fully human monoclonal antibody against human RANKL. Denosumab and OPG are thought to function in a similar manner by binding RANKL and preventing it from activating its receptor (RANK), which is found on osteoclasts and osteoclast precursors.math image,math image These data provide the first direct evidence that, like OPG, denosumab directly binds human RANKL and potently inhibits osteoclastogenesis driven by human RANKL. Unlike OPG, denosumab failed to bind murine RANKL and failed to increase BMD in normal mice, or in normal rats (data not shown). Human and nonhuman primate RANKL are the only proteins to which denosumab has shown binding, and denosumab did not exhibit binding to other members of the TNF family including human TRAIL, TNF-α, TNF-β, or CD40L.

Its lack of biological activity in mice or rats limited the evaluation of denosumab in the majority of preclinical bone disease models. OPG constructs have been shown to reduce bone resorption and improve bone mass and/or strength in mouse or rat models of ovariectomy, bone metastasis, humoral hypercalcemia, rheumatoid arthritis, skeletal unloading, inflammatory bowel disease, viral infection, vascular calcification, and other disease models.math image However, the ovariectomized cynomolgus monkey is the only preclinical disease model wherein the safety and efficacy of denosumab have been examined.math image The potential to develop a denosumab-responsive mouse model was suggested by the ability of a human RANKL fragment to cause hypercalcemia in normal mice, which indicated that human RANKL can activate murine RANK in vivo. Denosumab fully inhibited the catabolic effects of this RANKL fragment, which included all 10 β strands thought to be involved in RANKL trimerization and its ability to bind and activate RANK.math image,math image These results provided a strategy to engineer mice that would maintain their ability to resorb bone in a manner that would be inhibited by denosumab. Knock-in technology was used to create mice that exclusively express a chimeric form of RANKL, wherein the coding region of mouse exon 5 was replaced with the corresponding coding region of human exon 5. This exon encodes 90% of the minimal extracellular receptor binding portion of the molecule,math image and the few remaining residues within the active TNF domains of RANKL are well conserved between mice and humans.

There are several potential advantages with this knock-in methodology compared with transgenic approaches, including the maintenance of normal murine regulatory regions controlling the expression of this chimeric RANKL gene. Untranslated regions involved in transcriptional regulation of the knock-in gene also remained fully murine. HuRANKL mice might therefore be as suitable as normal mice for examining disease-related changes in RANKL. The stalk region of murine RANKL, which contains proteolytic cleavage sites involved in the shedding of RANKL from the cell surface,math image is also fully murine in huRANKL mice, as are the cytoplasmic and transmembrane domains. It is therefore unlikely that these mice would have alterations in the relative abundance of soluble versus surface-bound RANKL. HuRANKL mice seemed to have slightly lower average serum RANKL levels than WT controls, but it is difficult to directly compare their absolute levels because of the use of different antibodies and standards for their measurement.

Male and female huRANKL knock-in mice were smaller than their WT littermate controls at 13–16 wk of age, whereas at 10 mo of age, the average body weight of huRANKL mice and WT littermates was similar. The only other phenotypic difference observed to date with huRANKL mice is modestly higher bone mass and inconsistent evidence of reduced parameters of bone remodeling such as TRACP-5b, OcS/BS, and ObS/BS. Trabecular bone volume, measured histomorphometrically, was not significantly different from WT controls in young (8–12 wk old; n = 8/group) or adult (10 mo old) huRANKL mice (n = 6/group). However, a trend toward increased bone volume was noted in the 10-mo-old huRANKL mice. Certain μCT endpoints, including trabecular number and cortical area, were significantly greater in 10-mo-old huRANKL animals compared with their WT littermates. In general, the magnitude of skeletal differences between huRANKL mice and WT controls were within the range of natural variations that exist between different strains of inbred mice.math image For example, the largest difference observed in trabecular bone volume between huRANKL and WT mice was 2.4-fold (at the distal femur of 10-mo-old mice), whereas trabecular bone volume in the proximal tibia varied by 4.2-fold in the widely used C57BL/6J and BALB/c mouse strains.math image Those two strains also exhibited a 33% difference in femur diaphysis cortical area, whereas that difference was 21% in adult huRANKL versus WT mice.

Higher bone mass in huRANKL mice could be related to less bone resorption, as suggested by evidence of reduced serum TRACP-5b and OcS/BS. The extracellular domain of chimeric RANKL expressed by these mice showed modestly reduced potency for stimulating bone resorption in normal mice compared with murine RANKL. This difference was not associated with evidence of reduced affinity of chimeric RANKL for murine RANK or with reduced potency in a murine osteoclastogenesis study. It remains possible that the chimeric RANKL produced by huRANKL mice has reduced stability or greater clearance in vivo, which could in theory contribute to reduced bone resorption in huRANKL mice.

HuRANKL knock-in mice were used to obtain the first direct comparisons of the pharmacologic effects of OPG-Fc versus denosumab. Denosumab and OPG-Fc caused similar increments in trabecular bone volume in huRANKL mice, and both RANKL inhibitors resulted in >95% reduction in trabecular osteoclast surfaces. These results suggest that TRAIL binding did not contribute to the effects of OPG-Fc on bone. The relative potency of denosumab versus OPG-Fc could not be adequately assessed in this study because dose regimens were designed to maximally suppress bone resorption. A dose-response study in normal mice suggested that denosumab was more potent than OPG-Fc at inhibiting the resorptive response to the twice-daily injection of human RANKL. This greater potency could be related to the longer circulating half-life of denosumab relative to human OPG-Fc, which was shown in cynomolgus monkeysmath image and which was confirmed in huRANKL mice (data not shown). A longer half-life could have maintained therapeutic levels of denosumab for a longer duration, thereby neutralizing a greater proportion of RANKL injected into those mice. Denosumab had a 10-fold lower affinity than did OPG-Fc for human RANKL, whereas in vitro osteoclastogenesis assays showed similar potency between denosumab and OPG-Fc. Together, these results suggest that circulating half-life is a more important attribute than drug-ligand affinity or in vitro potency in predicting the efficacy of these two agents.

Pharmacokinetic and pharmacodynamic analyses of denosumab in huRANKL mice showed dose-dependent suppression of bone resorption using a dose range similar to that which was tested clinically in postmenopausal women.math image In those subjects and in huRANKL mice, the subcutaneous injection of denosumab at ∼1 mg/kg resulted in strong suppression of bone resorption and sustained serum levels of drug. Higher doses (5–10 mg/kg, once or twice weekly) are recommended when treating huRANKL mice with denosumab for >2 wk, because of the potential for immune responses to develop against this fully human protein after multiple injections. These high doses have been shown to maintain potent inhibition of bone resorption in huRANKL mice for several weeks without evidence of toxicity.math image Doses of denosumab <1 mg/kg resulted in more transient suppression of bone resorption and more rapid clearance of the drug in humansmath image and in huRANKL mice, highlighting the reversible nature of RANKL inhibition with denosumab. We also determined that denosumab binding to chimeric and human RANKL was similar, which further supports the use of huRANKL mice as a test system for evaluating the effects of denosumab.

Insights into possible mechanisms of reversibility with denosumab were provided by immunohistochemical analysis of denosumab disposition within bone. Denosumab was primarily localized to blood vessels within the medullary cavity and cortex. The primarily vascular localization of denosumab in huRANKL mice was also consistent with similar immunolocalization studies performed in bone sections from normal rats treated with human OPG-Fc.math image In that study, the clearance of OPG-Fc from peripheral bloodmath image was well coordinated with its clearance from bone.math image These analyses provided no evidence for denosumab localization to bone matrix or to cells that lined bone surfaces. Whereas this technique has only modest sensitivity at the level of light microscopy, it is reasonable to assume that the majority of denosumab remained within the vasculature. The soluble circulating nature of denosumab has potential implications for the suppression of osteoclast precursors that are found in human peripheral blood and in bone marrowmath image and that can differentiate into mature osteoclasts on stimulation by soluble RANKL.math image The ability of OPG to suppress osteoclast formation is well established,math image,math image and similar inhibition was shown with denosumab. The rapid suppression of bone resorption by denosumab (within 24 h in humansmath image and in huRANKL mice) suggests that denosumab is also capable of inhibiting mature osteoclasts.

In addition to marked reductions in bone resorption parameters, denosumab treatment of huRANKL mice was associated with significant reductions in osteoblast surface and in bone formation parameters within trabecular bone. Bone turnover in this skeletal compartment is dominated by bone remodeling, wherein resorption and formation are coupled in a manner that is maintained with RANKL inhibition.math image The suppression of bone formation parameters in cynomolgus monkeys treated with denosumab was strongly correlated with improvements in bone strength parameters,math image suggesting a lack of safety concerns for the suppression of bone formation under those conditions. Bone formation is an important aspect of the early phase of fracture repair, but fortunately this type of bone formation (endochondral) is not coupled with bone resorption and is therefore not impacted by antiresorptive therapy. Potent antiresorptive therapy, whether by bisphosphonates or by RANKL inhibitors, was shown to have neutral effects on callus formation and positive effects on callus strength despite delays in callus remodeling.math image,math image

In summary, OPG and denosumab inhibit osteoclastogenesis by their binding and inhibition of human RANKL. Unlike human OPG-Fc, denosumab did not bind human TRAIL or murine RANKL. Knock-in mice were therefore made to express a chimeric (murine/human) form of RANKL. This form of RANKL showed no alterations in its binding to denosumab or to murine RANK and maintained bone resorption in a manner that was fully inhibited by denosumab. HuRANKL mice represent the only current small-animal model system for studying the effects of denosumab, and histomorphometry showed that denosumab caused marked reductions in osteoclast numbers, more modest reductions in osteoblast numbers, and a resulting increase in bone volume. HuRANKL mice also provided the first evidence that denosumab can improve trabecular microarchitecture and cortical volumetric BMD without incorporation into bone matrix. Regarding their potential utility in preclinical bone disease models, preliminary data indicate that huRANKL mice lose bone mass after ovariectomymath image and after glucocorticoid administrationmath image in a manner that was inhibited by denosumab.

Acknowledgements

The authors are grateful for the contributions of Cathy Christiansen, Laura Martin, Jhun Viray, and staff from the Laboratory Animal Resource Department at Amgen for animal husbandry and for assistance from the staff of the Department of Protein Sciences in expression and purification of recombinant OPG, RANKL, and TRAIL. We also thank Diane D'Agostin for ES cell work, Betsy Daugherty for genotyping, and Zhaopo Geng for histomorphometry support. This study was supported by Amgen, Thousand Oaks, CA, USA.

Ancillary