Affinity targeting of therapeutic proteins to the bone surface—local delivery of sclerostin–neutralizing antibody enhances efficacy

Abstract

Currently available biotherapeutics for the treatment of osteoporosis lack explicit mechanisms for bone localization, potentially limiting efficacy and inducing off-target toxicities. While various strategies have been explored for targeting the bone surface, critical aspects remain poorly understood, including the optimal affinity ligand, the role of binding avidity and circulation time, and, most importantly, whether or not this strategy can enhance the functional activity of clinically relevant protein therapeutics. To investigate, we generated fluorescent proteins (eg, mCherry) with site-specifically attached small molecule (bisphosphonate) or peptide (deca-aspartate, D₁₀) affinity ligands. While both affinity ligands successfully anchored fluorescent protein to the bone surface, quantitative radiotracing revealed only modest femoral and vertebral accumulation and suggested a need for enhanced circulation time. To achieve this, we fused mCherry to the Fc fragment of human IgG₁ and attached D₁₀ peptides to each C-terminus. The mCherry-Fc-D₁₀ demonstrated an ~80-fold increase in plasma exposure and marked increases in femoral and vertebral accumulation (13.6% ± 1.4% and 11.4% ± 1.3% of the injected dose/g [%ID/g] at 24 h, respectively). To determine if bone surface targeting could enhance the efficacy of a clinically relevant therapeutic, we generated a bone-targeted sclerostin-neutralizing antibody, anti-sclerostin-D₁₀. The targeted antibody demonstrated marked increases in bone accumulation and retention (20.9 ± 2.5% and 19.5 ± 2.5% ID/g in femur and vertebrae at 7 days) and enhanced effects in a murine model of ovariectomy-induced bone loss (bone volume/total volume, connectivity density, and structure model index all increased [P < .001] vs untargeted anti-sclerostin). Collectively, our results indicate the importance of both bone affinity and circulation time in achieving robust targeting of therapeutic proteins to the bone surface and suggest that this approach may enable lower doses and/or longer dosing intervals without reduction in biotherapeutic efficacy. Future studies will be needed to determine the translational potential of this strategy and its potential impact on off-site toxicities.

Introduction

Affinity targeting of proteins to the bone surface has been reported for over 2 decades.^1,2 The earliest examples involved chemical conjugation of bisphosphonates (BPs)—small molecules known both for their affinity to hydroxyapatite (HA) and therapeutic use in osteoporosis^3,4—to random amines on the protein surface.^1,5 This approach has proven effective in conferring bone affinity to both model proteins, such as albumin and lysozyme, and potential biotherapeutics like osteoprotegerin.^6,7 While studies have repeatedly demonstrated an enhanced accumulation in skeletal tissues, it has been unclear to what extent targeted proteins retain functional activity when anchored to the bone surface.^7,8 Moreover, there has been limited evidence to support bone surface targeting as a means of enhancing the therapeutic activity or index of biologics in the setting of bone disease.

One notable exception is the targeted enzyme replacement therapy, asfotase alfa (sold under the brand name, Strensiq).⁹ Developed for treatment of hypophosphatasia (HPP), a rare inherited metabolic disorder of defective skeletal mineralization, the drug is a bone surface targeted form of the genetically deficient enzyme, tissue-nonspecific alkaline phosphatase (TNSALP).¹⁰ Instead of BPs, asfotase alfa relies on anionic polyaspartate peptides to induce skeletal accumulation.¹¹ These chains of repeated acidic amino acids had been previously identified in endogenous bone binding proteins, such as osteopontin,¹² and have been shown to target fluorescent tracers,¹³ radionuclides,¹⁴ small molecules like estradiol¹⁵ and quinolones,¹⁶ peptides,¹⁷ and even nanocarriers¹⁸ to the bone surface in animals. In asfotase alfa, deca-aspartate sequences are fused to each arm of the Fc region of human IgG₁, which is in turn fused to the C-terminus of TNSALP. This 2-chain protein has been shown to accumulate on bone and reverse skeletal and dental abnormalities in a murine model of HPP.¹⁹ The drug subsequently proved effective in clinical trials in children with life-threatening disease, producing both functional benefits and prolonged survival, and was granted orphan drug designation by the FDA in 2008.⁹

While the success of asfotase alfa offers the first clear support for bone surface targeting of therapeutic proteins, it also raises questions regarding the key features necessary for optimal protein delivery. Apart from the choice of affinity ligand, the role of binding avidity and protein characteristics, such as size and circulation time, have not been rigorously studied. Indeed, it is unclear what role the various features of asfotase alfa (eg, 2-chain structure, Fc fragment, and multiple peptide affinity ligands) play in its bone delivery or ultimate therapeutic activity, as they have not been independently tested. To address these knowledge gaps, we develop here methods for the study of BP and deca-aspartate (D₁₀) affinity ligands using site-specific modification to enable their direct comparison and fluorescent proteins to facilitate the evaluation of functional activity on the bone surface. We test the roles of avidity, circulation time, and protein size via Fc fusion, leading to a strategy of bone surface targeting of therapeutic antibodies. Ultimately, we show that a targeted, murine-specific version of the anti-sclerostin antibody, Romosozumab, has enhanced therapeutic activity in an ovariectomy-induced bone loss model.²⁰ Together, our findings have both translational implications for existing bone therapies and significance for our basic understanding of affinity targeting of therapeutic proteins to the bone surface.

Materials and methods

Materials

All chemical reagents were purchased from Sigma Aldrich unless specifically mentioned in the below methods. Reagents were used as received unless otherwise specified. Bovine bone chips were obtained from Immunodiagnostic Systems (CAT# DT1BON10096). The HA disks were obtained from Biosurface Technologies. Expression vector pTT5 and mammalian cell line (HEK293-6E) were licensed from the National Research Council of Canada (NRC).

Animals

Animal studies were conducted following guidance for the care and use of laboratory animals as adopted by the NIH, approved by the Institutional Animal Care and Use Committee of the University of Michigan. The C57BL/6J male or female mice (Jackson Laboratory, 000664), aged 12–16 wk, were used for all animal experiments.

Bacterial protein production and purification

The amino acid sequences of the fluorescent proteins mCherry, eGFP, and mCardinal were obtained from the UniProt database.²¹ The N- and C-terminal sortags²² or D₁₀ were appended to the coding sequences. The cDNAs encoding each protein were cloned into the pRSET vector (Thermo Fisher, CAT# V35120) and were transformed to BL21 (DE3) competent E. coli (Theromfisher, CAT# 600003) for production. Bacteria were grown in Terrific Broth medium to early log phase, as judged by OD600, induced with IPTG, and then incubated for 24 h at 18°C. Bacterial cell pellets were lysed by sonication and proteins were purified using Ni-NTA agarose His tag affinity resin (Qiagen, CAT# 30210). The >95% purity was verified by SDS-PAGE gel electrophoresis and size exclusion-HPLC. All proteins were stored at −80°C in aliquots, until their use, to avoid repeated freeze–thaw cycles.

Mammalian protein production and purification

For the production of mCherry-Fc fusion proteins, the fluorescent protein was codon optimized for production in human cells using an online tool (www.idtdna.com/CodonOpt), fused to the Fc fragment of human IgG₁, and cloned into the pTT5 mammalian expression vector (NRC).²³ For the production of anti-(mouse)-sclerostin antibodies, the sequences of heavy and light chain variable regions²⁴ were cloned into the pTT5 vector with mouse kappa and IgG_2a constant regions. The non-functional control IgG is an antibody generated by rat immunization using phosphorylated tau (pS422), which has been shown to have no target in mice.^25,26 In each case, bone-targeted versions were created by fusing a sequence encoding the amino acids DIDDDDDDDDDD (the D₁₀ peptide used in asfotase alfa¹⁹) to the C-terminus of the Fc fragment. Protein production was performed by transfecting 25 mL of mammalian cell culture (2 × 10⁶ cells/mL) with appropriate plasmids. In each conical tube, 15 μg of DNA was mixed with 45 μL of 40 kDa polyethylenimine (1 mg/mL) and 3 mL of F17 media (Gibco CAT# A13835-01). The mixture was added to HEK293-6E cells²³ and was transferred back to a 5% CO₂ incubator @ 37°C; 750 μL of yeastolate (20% w/w) (Thermo Fisher, CAT# 292804) was added to each tube of transfected cells 24 h post-transfection. Cells were harvested for 5–7 days after transfection and were centrifuged at 3500 g for 40 min. Protein was purified from the cell supernatant via Protein A agarose (Thermo Fisher, CAT: 20334). Proteins were stored at −80°C in aliquots to avoid repeated freeze–thaw cycles.

Site-specific modification of fluorescent proteins

Sortase modification of N- and C-terminal fluorescent proteins was performed as previously described.^22,27 In brief, LPET-tagged proteins were reacted with GGG peptide-azide at a 1:5 ratio (5 mol. GGG peptide-azide per 1 mol. LPET-tagged protein) for C-terminal modification in TBS buffer with the presence of calcium-dependent sortase (1 μM) and Ca²⁺ (1 mM) overnight at room temperature. For N-terminal modification, GGG-proteins were reacted with LPET peptide-azide under the same conditions. Unmodified proteins were removed using Ni-NTA agarose and excess peptide-azide were removed using a 10 kDa filter (Amicon, CAT# UFC501096). Azide-modified fluorescent proteins were then reacted with a 5-fold excess of BP-DBCO at room temperature overnight to yield protein-BP conjugates. The BP-modified protein was then dialyzed in a 10 kDa MWCO cassette (Thermoscientific, CAT# 87729) to remove excess BP-DBCO.

Anti-sclerostin and anti-sclerostin-D₁₀ binding affinity ELISA assay

The 96-well plate was coated with 0.05 μg recombinant murine sclerostin (R&D systems, CAT# 1589-ST) per well in 100 μL PBS overnight at 4°C. Next day, the plate was washed 3 times with PBS + 0.05% Tween20 (PBS-T) and then blocked with 200 μL PBS + 3% BSA at RT for 1 h. Post-blocking, the plate was washed 3 times with PBS-T, followed by adding 100 μL of anti-sclerostin and anti-sclerostin-D₁₀ at various concentrations and incubated at RT or 1 h. Afterward, the plate was washed 3 times with PBS-T and incubated with anti-mouse-HRP (Jackson Immunoresearch, CAT# 115035003) at RT for 30 min followed by PBS-T wash. The plate was developed using TMB substrate (Thermo scientific, CAT: 34021) following the manufacture instructions, and the result was read by plate-reader under the absorbance of 450 nm.

In vitro bone targeting assays

For in vitro assessment of function, bone-targeted and untargeted fluorescent proteins were incubated with HA disks or bovine bone chips at 2 μM concentration for 1 h at room temperature, prior to washing 5 times with PBS. An EVOS FL auto imaging system was used to visualize the fluorescence signal. All experiments are repeated at n ≥ 3, with representative images shown.

Quantitative radiotracing of protein biodistribution

Proteins were directly radioiodinated with [¹²⁵I]NaI (Perkin Elmer) using Pierce iodination reagent (Thermo Fisher, CAT# 28601) and were purified using Zeba desalting columns. Radiochemical purity was assessed via thin layer chromatography (TLC) performed using aluminum TLC silica gel 60 F254 plates (Millipore Sigma, CAT #105554) and a 75%:25% mixture of methanol and 1 M sodium acetate (pH 6.8) as a mobile phase. For radiotracing experiments, a tracer dose (1 or 2 μg) of ¹²⁵I-labeled protein was added to the appropriate mass of non-radioactive protein to give the desired dose (eg, 2 or 5 mg/kg). Doses were administered intravenously via retro-orbital injection. Each animal was subjected to 3 blood draws at designated time points for measuring blood pharmacokinetics (PK). The first blood draw was from the retroorbital plexus contralateral to the injection site, while the second blood draw was performed on the same side as the initial injection. The final blood draw was from the inferior vena cava at the time of euthanasia. Animals were transcardially perfused with 15 mL of PBS to flush the residual blood content from organs. Radioactivity of blood and organs were measured via gamma counter (PerkinElmer, 2470 Automatic Gamma Counter).

Fluorescence imaging of ex vivo bone tissue

To determine the localization and retention of bone-targeted proteins in vivo, mice were administered with fluorescent proteins intravenously via retro-orbital injection. Animals were transcardially perfused at the appropriate time point and bones were harvested for fixation in 4% PFA for 48 h. Post-fixation, bones were directly embedded in OCT (Thermo Fisher, CAT# 23730571) for cryosectioning. Sections were collected on Kawamoto film and were imaged using confocal microscopy. Nikon Eclipse C2 and Zeiss LSM 780 inverted confocal microscopes were used to visualize the fluorescence signal.

Ovariectomy-induced bone loss model

The 13-wk-old healthy female C57BL/6J mice were received from Jackson Laboratories (000664). Surgical ovariectomy (OVX) was performed as previously described.²⁸ Briefly, mice were acclimated for 1 wk. A 0.25-cm incision was made bilaterally to locate the ovaries. In “sham control” mice, the ovary and fat pad were immediately returned to the peritoneal cavity. In “OVX” mice that received the full surgery, the exposed ovary and the oviduct were carefully removed using sterile scissors, and hemostasis was achieved. The uterus and remaining oviduct were placed back into the abdominal cavity and sutured. Post-operative carprofen was administered at 24 h (5 mg/kg subcutaneous injection) and the animals were monitored daily for the first 10 days and then twice weekly until euthanasia. Bone-targeted and untargeted sclerostin antibodies and targeted control (non-functional) antibodies (5 mg/kg) were administered weekly via retro-orbital injection, alternating left and right administration, starting at 7 days post-surgery. After 4 wk, mice were euthanized by CO₂ asphyxiation and bilateral pneumothorax. Long bones and vertebrae were harvested at the endpoint, fixed with 4% paraformaldehyde at 4°C for 48 h, and then kept in PBS at 4°C for subsequent analysis.

Micro-CT analysis

Micro-CT analysis was performed following published guidelines.²⁹ Briefly, samples were placed in a 19-mm diameter specimen holder and were scanned over the entire length of the tibia using a micro-CT system (μCT100 Scanco Medical, Bassersdorf, Switzerland) with voxel size 10 μm, 70 kVp, 114 μA, 0.5 mm AL filter, and integration time of 500 ms. For femur analysis: A 1.0-mm region of trabecular compartment was analyzed immediately below the growth plate using a fixed global threshold of 18%; and a 0.3-mm region of cortical compartment at the midpoint was analyzed using a fixed global threshold of 28% (280 on a grayscale of 0–1000). For vertebral analysis: the L4 vertebrae was identified; the trabecular compartment was analyzed using a fixed global threshold of 18%. Trabecular bone volume fraction (bone volume/total volume [BV/TV]), connectivity density (Conn dens), structure model index (SMI), trabecular thickness (Tb.Th), trabecular no. (Tb.N), trabecular separation (Tb.Sp), cortical bone volume fraction (BV/TV), cortical porosity, cortical thickness, tissue mineral density (TMD), and sub-periosteal area and sub-endosteal area were analyzed using an evaluation software from the manufacture.

Statistical analysis

Statistical analyses were done using 1-way ANOVA among experimental groups and this was followed by a Tukey test. All experiments were done with at least 3 biological replicates or more per group. The results are expressed as the mean ± SEM. Statistical significance is considered to be P < .05. For OVX experiments, a sample size calculation was performed using data from a pilot study and G^*Power 3 software.³⁰ In analyzing femoral micro-CT data, the left- and right-sided measurements from the same mouse were not considered as independent data points, so these values were averaged prior to statistical comparisons.

Results

Site-specific attachment of bone-targeting affinity ligands to fluorescent proteins

To enable rigorous comparison of affinity ligands, we developed methods for attachment of either BP or D₁₀ peptides to equivalent sites on fluorescent model proteins. While D₁₀ was genetically fused to the N- or C-terminus, BPs were site-specifically conjugated following the scheme shown in Supplementary Figure S1. Briefly, a FITC-label and azide functional group were attached via the bacterial enzyme, sortase.^22,27 Azide-modified fluorescent protein (eg, mCherry) was then reacted with BP-DBCO, resulting in a final product with a single, covalently attached BP at either terminus.

We next evaluated the binding of BP-conjugated mCherry to HA and bovine bone chips in vitro. The mCherry decorated with surface BP in a non-site-specific manner (similar to prior reports for albumin and lysozyme)^1,6 was used as a positive control. As shown in Figure 1A, both C-terminal and non-site-specific conjugation of BP resulted in binding of functional fluorescent protein to HA and bone chips. No fluorescent signal was detectable in the absence of BP modification. We repeated this approach with 2 other fluorescent proteins, eGFP and mCardinal, and observed similar results (Figure 1B). Similarly, bone targeting and surface function were consistent whether the BP affinity ligand was attached to the N- or C-terminus of the fluorescent proteins (Figure 1C). Finally, we compared the in vitro binding of equimolar concentrations of BP-conjugated and D₁₀-fused mCherry (Figure 1D). While the fusion of D₁₀ to the N terminus produced similar binding and surface fluorescence as covalent attachment of BP, no bone surface targeting was observed when D₁₀ peptide was fused to the C-terminus of mCherry.

Figure 1.

Binding of bone surface targeted fluorescent proteins in vitro. (A) The mCherry with a single BP attached to the C-terminus (BP-mCherry)-bound HA (top panel) and bovine bone chips (bottom panel) similarly to mCherry decorated with BP via non-selective attachment to primary amines. (B) Bone binding of BP-GFP and BP-mCardinal vs BP-mCherry. (C) Comparison of bone binding by mCherry with N- and C-terminal BP. (D) Comparison of bone binding by mCherry modified at the N- and C-terminus with BP vs D₁₀.

Comparison of bone surface targeting of BP- and D₁₀-mCherry in vivo

We next sought to quantitatively compare the bone surface targeting of BP- and D₁₀-modified mCherry following intravenous injection in vivo, with untargeted mCherry as a control. To ensure that the 3 proteins were equivalent except for the affinity ligand, both D₁₀-mCherry and untargeted mCherry were synthesized with a sortag and modified with FITC-labeled, azide-containing peptide (ie, analogous to BP-mCherry, but without subsequent attachment of BP-DBCO). Each protein was radioiodinated and administered at an equimolar 2 mg/kg dose. Animals were transcardially perfused to clear residual blood from the organs. As shown in Figure 2A, the 3 proteins cleared the blood quickly via renal filtration, with <5% of the injected dose (ID) in the blood and most of the radioactive signal in the kidney and/or urine at 4 h post-injection (Figure 2B). Detailed results are provided in Supplementary Table S1. In spite of rapid clearance, both BP- and D₁₀-modified mCherry demonstrated selective accumulation in skeletal tissues at 4 h post-injection—femur (0.75 ± 0.14, vs 0.60 ± 0.06 vs 0.31 ± 0.06%ID/g for BP-mCherry, D₁₀-mCherry, and unmodified mCherry, respectively, P = .0016) and vertebrae (0.61 ± 0.13, 0.49 ± 0.06, 0.26 ± 0.07%ID/g, P = .0052) (Figure 2C). To confirm that this quantitative increase reflected the anchoring of protein to the bone surface, we sectioned calcified tissue and performed confocal imaging. As shown in Figure 2D, both BP- and D₁₀-modified mCherry were visualized on the surface of trabecular bone, confirming both localization and functional activity in vivo. Interestingly, all experimental samples demonstrated red fluorescent signal in the bone marrow, presumed to be auto-fluorescence, as it was also present in PBS-injected mice (Supplementary Figure S2). Thankfully, this auto-fluorescent signal was easily distinguished from surface-anchored mCherry, for which co-localization of the red fluorescent protein and FITC-labeled peptide was observed.

Figure 2.

Biodistribution of bone surface targeted mCherry in vivo. Quantitative radiotracing of BP-mCherry, D₁₀-mCherry, and unmodified mCherry at a 2 mg/kg dose. (A) Blood pharmacokinetics. (B) Uptake by vital organs at 4 h. (C) Biodistribution to femur and lumbar vertebrae at 4 h; n = 3–5 male mice per group. One-way ANOVA was employed for statistical analysis. Data presented as mean ± SEM. ns= non-significant, ^**  P < .01. (D) Confocal fluorescence imaging of femurs of female mice injected with 2 mg/kg BP-mCherry vs D₁₀-mCherry at 4 h post-injection. (E) Time course of BP-mCherry on trabecular bone with peak fluorescent signal seen at 4 h post-injection. White arrowheads indicate areas of isolated green fluorescent signal, which suggest inactivation or catabolism of surface-anchored mCherry. All scale bars = 200 μm.

Dose and time dependence of bone surface targeting in vivo

Our next goal was to determine the effect of dose on the accumulation of targeted protein in skeletal tissues. Given the relatively fast renal clearance of the proteins and the large surface area of mineralized bone in the body, we expected the PK to be largely independent of dose—that is, bone accumulation would not saturate. Indeed, injection of a 5 mg/kg vs 2 mg/kg dose of BP-mCherry showed proportionate increases in blood, bone, and organ concentrations, resulting in nearly identical biodistribution when expressed as a percent of the ID (Supplementary Figure S3, Supplementary Table S2). We next performed fluorescence imaging of bone tissue from animals injected with 5 mg/kg BP-mCherry (Figure 2E). Consistent with the radiotracing results, the higher dose produced a brighter fluorescent signal, facilitating the study of the time dependence of bone surface targeting. As shown, the amount of fluorescent protein anchored to the bone surface gradually increased over the first few hours, reaching an apparent peak at approximately 4 h post-injection. By 24 h, the fluorescent signal was visibly diminished due to either clearance of intact protein, loss of activity, or catabolism. Interestingly, we observed areas of isolated green fluorescent signal at the 24-h time point, suggesting that surface-anchored mCherry may have been inactivated or catabolized at those sites, leaving only the surface-anchored FITC intact (white arrowheads, Figure 2E).

Prolonging circulation time markedly increases bone surface accumulation

The short plasma half-life of BP- and D₁₀-targeted mCherry (Figure 2A), together with fluorescence imaging indicating gradual accumulation on the bone surface (Figure 2E), suggested that prolongation of blood PK might significantly enhance bone surface accumulation. To further investigate, we fused mCherry to the Fc-fragment of human IgG₁, which extends plasma half-life due to interaction with the murine neonatal Fc receptor.³¹ The D₁₀ affinity ligands were fused to the C-terminus of each chain (Figure 3A). The D₁₀ was chosen over BP in this case due to greater ease of synthesis and purification as well as the similarity of the resulting design to that of asfotase alfa.¹⁹ The D₁₀-mCherry (without Fc) and mCherry-Fc (ie, without D₁₀ affinity ligands) were used as controls.

Figure 3.

The Fc-fusion greatly enhances bone accumulation of targeted mCherry. (A) Schematic depicting the structure of two-chain mCherry-Fc fusion protein with and without C-terminal D₁₀ affinity ligands. (B) The D₁₀-mCherry and mCherry-Fc-D₁₀, but not untargeted mCherry-Fc, bind to bone chips in vitro. (C) The Fc-containing proteins demonstrate markedly prolonged circulation time at 5 mg/kg dose, with a small, but significant, reduction in blood pharmacokinetics for bone-targeted vs untargeted mCherry-Fc (^*  P < .05 for all time points). (D) Biodistribution to the femur and lumbar vertebrae at 4 and 24 h post-injection. (E) Bone:blood ratio. (F) Uptake by vital organs at 24 h post-injection, n = 3–5 mice for each condition. Note (D)–(F) share the same figure legend. One-way ANOVA was employed for statistical analysis. Data are presented as mean ± SEM, ^***  P < .001, ^****  P < .0001. (G) Confocal fluorescence microscopy at 4 and 24 h shows the accumulation of mCherry-fc-D₁₀, but not mCherry-Fc, on the trabecular surface, while the latter is seen predominantly in the marrow (scale bar = 200 μm). Male mice were used in the biodistribution experiment (C–F). Female mice were used in confocal fluorescence imaging experiments (G).

As shown in Figure 3B, D₁₀-mCherry and mCherry-Fc-D₁₀ demonstrated similar function in vitro, with roughly equal binding to bovine bone chips at equimolar concentrations. Addition of the Fc fragment greatly prolonged plasma half-life, with mCherry-Fc and mCherry-Fc-D₁₀ demonstrating 108- and 77-fold increases in the area under the curve of the blood concentration vs time curve, respectively (732 ± 50 vs 1024 ± 45 vs 12.23 ± 1.7 vs 9.5 ± 0.9 for mCherry-Fc-D₁₀, mCherry-Fc, D₁₀-mCherry, and unmodified mCherry, P < .001) (Figure 3C,  Supplementary Table S3). A relatively small, but significant, difference in blood PK was observed for bone-targeted vs untargeted mCherry-Fc (P < .05), with faster clearance presumed to be the result of bone accumulation (ie, target-mediated disposition). Correspondingly, mCherry-Fc-D₁₀ showed a marked increase in bone accumulation, with 6.73 ± 1.89 and 5.05 ± 0.87%ID/g on femur and vertebrae at 4 h post-injection, respectively (as compared to 2.12 ± 0.29 and 1.23 ± 0.25%ID/g for mCherry-Fc, P < .001 for each). Given the prolonged plasma half-life, we hypothesized that bone accumulation of mCherry-Fc-D₁₀ would continue past 4 h. Indeed, femoral and vertebral uptake increased to 13.59 ± 1.6 and 11.42 ± 1.27%ID/g at 24 h, with bone:blood ratios of 213.5 ± 60.4 and 179.1 ± 47.1 (Figure 3D and E,  Supplementary Table S3). Apart from bone, the organ biodistribution of mCherry-Fc-D₁₀ and mCherry-Fc were similar (Figure 3F,  Supplementary Table S3). Interestingly, mCherry-Fc demonstrated significantly greater accumulation in bone tissue than untargeted mCherry or even D₁₀-mCherry. While this was initially thought to be the residual blood content that could have not been completely flushed from the bone tissue on transcardial perfusion, it increased from 4 to 24 h despite a decrease in blood concentration (Figure 3D and E,  Supplementary Table S3). Fluorescence imaging indicated that the majority of the protein taken up in bone is localized to the bone marrow, where numerous cell types express Fc receptors.³² As shown in Figure 3G, only mCherry-Fc-D₁₀ was observed on the mineralized surface of the bone.

Bone surface targeting of therapeutic antibodies

Our next priority was to determine if affinity targeting to the bone surface would enhance the functional activity of a clinically relevant protein therapeutic (Figure 4A). As a first test, we chose a murine analog of the FDA-approved monoclonal antibody, Romosozumab. This antibody binds and inhibits murine sclerostin, a small protein expressed predominantly by osteocytes, which potently inhibits bone formation and stimulates bone resorption.³³ While bone surface targeting was expected to enhance local antibody concentration and scavenging of sclerostin, it was not clear if this would increase the therapeutic activity or if the sclerostin binding elsewhere (eg, the systemic circulation) might play a significant role in its functional effects.³⁴ To further investigate, we synthesized targeted (anti-sclerostin-D₁₀) and untargeted (anti-sclerostin) versions of the antibody, fusing D₁₀-affinity ligands to each C-terminus of the murine Fc fragment. After confirming the structure and purity of bone-targeted and untargeted antibodies (Supplementary Figure S4), we tested binding to recombinant mouse sclerostin and found equivalent affinities (Figure 4B). We next radiolabeled the antibodies and injected them intravenously at a dose of 5 mg/kg. Overall, the blood PK mirrored that of mCherry-Fc and mCherry-Fc-D₁₀, with both targeted and untargeted antibodies demonstrating prolonged circulation time, but the former cleared the blood slightly faster due to bone uptake (Figure 4C,  Supplementary Table S4). Indeed, anti-sclerostin-D₁₀ demonstrated femoral and vertebral accumulation of 14.47 ± 3.19 and 13.40 ± 2.96%ID/g at 24 h post-injection, roughly an order of magnitude higher than anti-sclerostin (2.41 ± 0.44 and 1.45 ± 0.21%ID/g for femur and vertebrae, P < .001 for each) (Figure 4D and E,  Supplementary Table S4). Notably, bone distribution of anti-sclerostin-D₁₀ continued to increase over the course of 1 wk, reaching 20.9 ± 2.5 and 19.5 ± 2.5%ID/g on femur and vertebrae at 7 days post-injection, respectively. By contrast, the bone content of anti-sclerostin decreased over this same time frame and the radioactive signal was no longer detectable 1 wk post-injection. We also tested the blood and bone PK and organ biodistribution of a bone surface targeted, but non-functional antibody, IgG-D₁₀, and we directly compared its distribution in adult male and female mice (Supplementary Figure S5). While blood PK and distribution to vital organs were closely matched across the 2 sexes, there were significant differences in both the timing and magnitude of bone accumulation—especially in vertebrae (Supplementary Figure S5C). Finally, we took advantage of the multichain IgG structure to directly evaluate the impact of binding avidity on bone surface targeting. Non-functional IgG (no affinity ligands) was compared to monovalent IgG-D₁₀ (ie, affinity ligand on just 1 of the 2 heavy chains) and bivalent IgG-D₁₀ (ie, affinity ligands on both heavy chains). As shown in Supplementary Figure S6, the bivalent molecule demonstrated significantly higher bone accumulation at 24 h post-injection, indicating avidity is a key determinant of bone surface targeting of IgGs.

Figure 4.

Bone surface targeting of sclerostin–neutralizing antibody. (A) Schematic showing sclerostin as an osteocyte-derived, paracrine factor which inhibits bone formation and stimulates bone resorption. To determine if bone surface targeting could enhance the therapeutic efficacy of a clinically relevant biotherapeutic, we generated D₁₀-targeted anti-sclerostin antibody (anti-sclerostin-D₁₀) and tested its bone accumulation and retention. (B) Comparative affinity of anti-sclerostin-D₁₀ and anti-sclerostin on plate-based ELISA. (C) Blood pharmacokinetics at 5 mg/kg, with anti-sclerostin-D₁₀ significantly lower at all time points (^***  P < .001). (D and E) Time course of biodistribution to femur and lumbar vertebrae shows accumulation and retention over 1 wk post-injection; n = 4 female mice each group and time point. One-way ANOVA was employed for statistical analysis. Data are presented as mean ± SEM, ^***  P < .001, ^****  P < .0001.

We next sought to compare the therapeutic efficacy of anti- sclerostin-D_10, anti- sclerostin, and IgG-D₁₀ in a murine model of ovariectomy-induced bone loss (OVX). Before testing therapeutics, a pilot study of the OVX model was performed to guide sample size calculation and ensure adequate statistical power. Measurement of BV/TV on micro-CT analysis²⁹ showed a mean effect size of −21.2% for OVX (5.4 ± 3%) vs sham (26.6 ± 3.5%). Based on these results, we determined that a sample size of n = 8 mice per group would be sufficient to detect a 35% decrease in bone loss with 80% power (Supplementary Figure S7). The experimental design for testing therapeutic efficacy is shown in Figure 5A. The 14-wk-old female C57BL/6J WT mice were subjected to a bilateral ovariectomy²⁸ and, starting 1 wk later, were given weekly 5 mg/kg doses of anti-sclerostin-D_10, anti-sclerostin, IgG-D₁₀, or vehicle control. Dosing was based on a prior rodent trial of anti-sclerostin in which the untargeted antibody was given twice weekly at 25 mg/kg.^20,24,35 Based on our pharmacokinetic results, we estimated that the lower dose and less frequent administration of anti-Scl-D₁₀ would give roughly similar bone exposure.

Figure 5.

The OVX model and micro-CT of femoral trabecular bone. (A) Schematic of murine ovariectomy-induced bone loss model, in which healthy female mice underwent bilateral ovariectomy at 14 wk of age. Starting 1 wk after the surgery, mice were given weekly injections of 5 mg/kg anti-sclerostin-D₁₀, anti-sclerostin, targeted but non-functional antibody (IgG-D₁₀), or vehicle. Mice were sacrificed after 4 wk and tissues harvested for micro-CT. (B) Three-dimensional rendering highlighting the ROI. (C) Micro-CT parameters for quantitative analysis: BV/TV = bone volume to total volume fraction. (D) Conn dens = connectivity density. (E) SMI = structural model index. Note (C)–(E) share the same figure legends; N = 8 mice (16 femurs, average of 2 femurs = 1 mouse) for each condition. One-way ANOVA was employed for statistical analysis. Data are presented as mean ± SEM, ^**  P < .01, ^***  P < .001, ^****  P < .0001. (F) Representative 3D rendering of femoral cross-section for each group.

As shown in Figure 5B–E and Supplementary Table S5, volumetric analysis of the femur trabecular compartment in the defined ROI demonstrated a greater therapeutic effect for anti-sclerostin-D₁₀ than anti-sclerostin, IgG-D₁₀, or vehicle-treated OVX mice. Specifically, animals treated with anti-sclerostin-D₁₀ had significantly greater BV/TV (11.9% ± 1.2%) than those in the other groups (7.5% ± 1.8% vs 4.5% ± 1.0% vs 4.2% ± 1.0% for anti-sclerostin, IgG-D₁₀, and vehicle, respectively, P < .0001 for anti-sclerostin-D₁₀ vs each) (Figure 5C). Anti-sclerostin-D₁₀ treatment was also associated with significant changes in 2 clinically relevant measures of bone morphology and structure—connectivity density (112.8 ± 24.5 vs 35.5 ± 15.2 vs 15.3 ± 7.3 vs 13.6 ± 5.9, P < .0001 for anti-sclerostin-D₁₀ vs each) and SMI (1.99 ± 0.18 vs 2.90 ± 0.28 vs 3.34 ± 0.17 vs 3.30 ± 0.27, P < .0001 for anti-sclerostin-D₁₀ vs each) (Figure 5D and E). For these latter metrics, OVX mice treated with anti-sclerostin-D₁₀ had values similar metrics to sham controls, indicating that the targeted treatment preserves well-organized and morphometrically normal bone mass and is protective against estrogen-induced bone loss. The observed differences in these quantitative metrics were reflected in the 3D reconstructions of the micro-CT scans (Figure 5F,  Supplementary Figure S8). Effects on TMD and trabecular morphology (Tb.N, Tb.Th, and Tb.Sp) are shown in Supplementary Figure S9 and Supplementary Table S5. Finally, the impact of OVX and various treatments on midshaft femoral cortical bone are shown in Supplementary Figure S10 and Supplementary Table S6. The OVX did not significantly affect cortical bone metrics, including cortical thickness, as compared to sham surgery—possibly due to the relatively short duration of our model. In spite of this, both anti-sclerostin and anti-sclerostin-D₁₀ enhanced cortical bone mass, mineral density, and thickness, although the 2 treatment groups were not significantly different.

Figure 6 shows the micro-CT analysis of lumbar vertebrae. Consistent with the effects seen on femoral bone, anti-sclerostin-D₁₀ significantly improved the metrics of trabecular bone mass, morphology, and structure in L4 vertebrae of OVX mice. The effects of anti-sclerostin-D₁₀ were superior to anti-sclerostin for BV/TV (25.8% ± 1.7% vs 21.3% ± 1.6%, P < .0001) and SMI (0.51 ± 0.19 vs 1.06 ± 0.19, P < .0001). By contrast, both treatments significantly improved Conn dens without a significant difference between the targeted and untargeted antibody (126.6 ± 23.34 vs 114 ± 12.27, P = .5307) (Figure 6B–D). Effects on other metrics are shown in Supplementary Figure S11 and Supplementary Table S7. As with the femur, 3D reconstructions of L4 vertebrae matched the quantitative findings from micro-CT analysis (Figure 6E).

Figure 6.

Micro-CT of L4 vertebral trabecular bone. (A) The ROI for L4 vertebrae. (B–D) Micro-CT parameters: BV/TV= bone volume to total volume fraction, Conn dens = connectivity density, and SMI = structure model index. Note (B)–(D) share the same figure legend; n = 8 mice for each condition. One-way ANOVA was employed for statistical analysis. Data are presented as mean ± SEM, ns = non-significant, ^**  P < .01, ^***  P < .001, ^****  P < .0001. (E) Representative, 3D-renderings of L4 vertebrae from each treatment group.

Discussion

As the use of biologics for skeletal disorders (eg, osteoporosis, inflammatory arthritis, and rare bone diseases) has become more common, interest in the delivery of therapeutic proteins to bone has grown despite limited insight into its requirements or ability to enhance the functional activity or efficacy.^2,36 The most popular approach to date—affinity targeting to the bone surface—is a unique pharmacologic strategy, which aims to create an extracellular depot of protein within the bone microenvironment, enhancing local concentration and/or reducing the dose necessary to achieve therapeutic thresholds. Targeting the mineralized surface of bone has a no. of challenges from the standpoint of drug delivery, including a lack of obvious protein targets, which in turn limits the development of antibody-based affinity ligands that form the basis of selective delivery to most tissues.³⁷ At a more fundamental level, there are questions regarding the validity of the depot strategy, which may increase the local concentration without guaranteeing access to cells or other key biological targets. Finally, the bone surface is constantly being remodeled, creating the possibility of phagocytosis and degradation of surface-anchored proteins, or alternatively encasement within newly formed bone.³⁸ With these challenges in mind, we sought to systematically and quantitatively characterize the determinants of protein targeting to the bone surface, while investigating its potential utility and applications.

Our results offer a no. of insights for those considering this approach—notably, the finding that a single bone affinity ligand is sufficient to drive the accumulation of functional protein on the bone surface, both in vitro and in vivo. While prior work has shown that affinity ligands—particularly BP—are capable of inducing bone uptake, the use of fluorescent proteins and controlled, site-specific bioconjugation clarifies the degree of modification required and the activity of proteins once in bone tissue. Apart from the degree of modification, our results also help inform the selection of sites for attachment of affinity ligands to protein cargo. Modification at either terminus seems adequate for BP conjugates, but the situation is less clear for a genetic fusion of peptide affinity ligands, with our results indicating that some degree of trial and error may be required. For example, we found that C-terminal fusion of D₁₀ was sufficient to confer bone binding to mCherry-Fc, but not mCherry alone. Whether or not this trial and error can be avoided via the use of flexible linkers or other related strategies will require additional study.

Apart from these considerations, we found relatively few differences between the 2 affinity ligands. Other groups have identified meaningful distinctions between bone-targeting peptides and small molecules, but their efforts focused on the delivery of peptides rather than proteins and selective targeting to fracture sites rather than healthy bone.¹⁷ By contrast, our results indicate that both BP and D₁₀ affinity ligands induce modest accumulation of protein (<1% ID/g) on the bone surface. While it was a relatively small percentage, the amount increased proportionately with administered dose, at least within the range tested in our experiments. This implies the lack of saturation of the bone surface and suggests that therapeutic concentrations may be achievable given a high enough dose even without any optimization of pharmacokinetic parameters. Our results also indicate a surprising degree of retention at the bone surface, with a single dose of BP-mCherry producing over 24 h of detectable fluorescent signal. This finding was unexpected, given the relatively modest binding affinity of BP and anionic peptides on bone—orders of magnitude lower than most tissue-targeting antibodies³—and suggests that abundance of target, rather than strength of binding, may be responsible. In other words, proteins bearing these affinity ligands likely dissociate and repeatedly rebind the bone instead of being tightly anchored to the surface. This phenomenon, if confirmed, could have important implications for the ability of surface-targeted proteins to exert therapeutic effects in the bone microenvironment.

While the results with singly modified mCherry provide important insights for bone surface targeting, they also indicate that the majority of injected protein is cleared from the circulation without ever gaining access to its intended target. This is not atypical for proteins with affinity to antigens not accessible from the bloodstream, for which escape from the vasculature is necessary for affinity to affect PK.³⁹ In these situations, half-life extension via Fc fusion or albumin binding may enhance the effects of affinity targeting or further reduce its impact by decreasing penetration across the endothelial barrier and into the tissue parenchyma.⁴⁰ In this respect, our results are clear—bone affinity and circulation time are both critically important to effective bone surface targeting, and the combination of the Fc fragment and the D₁₀ affinity ligand seems particularly promising, increasing the biodistribution of mCherry to >10% ID/g in both femur and vertebrae. Our results with monoclonal antibodies are even more impressive, with as much as 50% of the ID (~25–30% ID/g, assuming equal distribution throughout the mouse skeleton) ending up on the bone surface. Importantly, this accumulation is gradual and dependent on sustained blood concentrations, so it may vary from antibody to antibody, depending on the amount of accumulation in other tissues. The IgG-D₁₀, for example, demonstrates greater bone accumulation than anti-sclerostin-D₁₀, as it has no other target in the mouse. These details aside, the degree of tissue-specific delivery achieved is almost unprecedented within the field of drug delivery.³⁷

Of course, the significance of these quantitative findings is ultimately dependent on the ability of surface-targeted proteins to exert therapeutic effects within the bone microenvironment. In this regard, our most significant finding may be the enhanced effects of anti-sclerostin-D₁₀ in the murine OVX model as compared to its untargeted counterpart. Critically, our experiments used both a lower dose (5 vs 25 mg/kg) and longer dosing interval (once vs twice weekly) than that tested in the original rodent trials.³⁵ While dose response was not reported in those studies, our results indicate that untargeted anti-sclerostin is only partially effective when dose and frequency are reduced. This is further supported by results from other groups,^41–44 and from testing of romosozumab in non-human primates, in which monthly doses of 3, 10, and 30 mg/kg were found to have increasing effects on trabecular bone volume and formation.⁴⁵ Taken together, these results strongly support the notion that bone surface targeting enhances the local activity of anti-sclerostin and enables equal therapeutic efficacy at lower dose and longer dosing interval, albeit in a prophylactic mode of treatment. Future experiments will be needed to assess the ability of anti-sclerostin-D₁₀ to restore bone mass in animals with established osteoporosis.

Given the numerous reported activities of anti-sclerostin,^46–48 it will also be interesting to see if surface targeting enhances its activity and efficacy in settings other than estrogen-deficiency-induced bone loss. Similarly, an important area for future research will be determining what other antibodies and range of therapeutics have enhanced efficacy when targeted to the bone surface. Beyond this, it will be important to determine the specificity of the targeting strategy for the bone surface—not only in young, healthy animals but also in those with bone diseases and comorbid conditions—especially those known to be associated with extraosseous calcification (eg, diabetes, atherosclerosis, and renal insufficiency).^49,50 Finally, future studies will be needed to confirm the translation of the targeting strategy to large animals and, ultimately, its ability to influence dosing and off-target toxicity in patients.

Author contributions

Boya Zhang (Data curation, Formal analysis, Investigation, Methodology, Software, Validation, Visualization, Writing—original draft, Writing—review & editing), W. Benton Swanson (Data curation, Formal analysis, Investigation, Methodology, Software, Validation, Visualization, Writing—original draft, Writing—review & editing), Margaret Durdan (Data curation, Formal analysis, Investigation, Methodology, Validation), Heather N. Livingston (Data curation, Investigation), Michaela Dodd (Data curation, Investigation), Sachith M. Vidanapathirana (Data curation, Investigation), Alec Desai (Data curation, Investigation), Lindsey Douglas (Data curation, Investigation), Yuji Mishina (Data curation, Formal analysis, Investigation, Methodology, Resources, Supervision, Writing—review & editing), Megan Weivoda (Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Resources, Supervision, Writing—review & editing), and Colin F. Greineder (Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Supervision, Writing—review & editing).

Funding

NHLBI (C.F.G., K08-130430), NIAMS (M.W., R01-AR077538), and the U-M MiMHC/NIAMS (P30-AR069620), NIDCR (F30 DE029359, W.B.S.).

Conflicts of interest

None declared.

Data availability

The data supporting this study’s findings are available in this article’s supplementary material.