Recombinant Proteins and Inhibitors

Supplementary Methods:

Recombinant Proteins and Inhibitors

Recombinant ADAMTS-4 and ADAMTS-5 (262-624) were generated as described [1]. ADAMTS-5 (262-484) was expressed as ADAMTS-5 (1-484) with C-terminal Strep-Tag in a Chinese hamster ovary (CHO) cell line and purified from the conditioned medium via Strep-Tactin resin. ADAMTS-5 (1-567) and ADAMTS-5 (1-930) were expressed with C-terminal affinity tags in CHO cell lines; the prodomains were naturally processed during expression to generate ADAMTS-5 (262-567) and ADAMTS-5 (262-930). Due to endogenous proteolytic processing of the C-terminal affinity tags, conventional chromatography schemes were developed for purification. ForADAMTS-5 (262-567) the conditioned medium was concentrated and filtered prior to ammonium sulfate precipitation (20/55% saturations); the solubilized ammonium sulfate pellet was chromatographed on Source 15 Phenyl followed by sizing on Superdex 200. ADAMTS-5 (262-930) was captured from conditioned medium by Source 15 Q followed by Heparin Sepharose. The Heparin Sepharose eluate was precipitated with 45% ammonium sulfate saturation followed by sizing on Superdex 200. Appropriate N-terminal sequence was confirmed by Edman sequencing. Activity of purified ADAMTS-4 and ADAMTS-5 proteins was confirmed using a synthetic aggrecan IGD peptide substrate assay [1]. ADAMTS-5 and ADAMTS-4 selective mAbs (Table 1) and isotype matched control mAbs were expressed and purified from culture media via affinity chromatography from hybridomas (murine forms) or stably transfected Chinese hamster ovary (CHO) cell lines (humanized forms). Recombinant MMPs, aggrecan IGD (R&D Systems) and purified bovine aggrecan (Sigma Chemical) were purchased. GSK571949 is a broad spectrum small molecule metalloprotease inhibitor [1] used as an assay control in some experiments.

Immunization, mAb selection and humanization

SJL mice (Charles River), immunized with a combination of purified recombinant human ADAMTS-4 and ADAMTS-5, were routinely monitored for specific serum immunoreactivity and neutralization of specific protease activity. Splenocytes and lymph node-derived cells from mice with desirable profiles were progressed to fusion with P3XBcl-2.13 mouse myeloma cells and cultured underhypoxanthine-aminopterin-thymidine(HAT)selection. Resulting hybridomas were screened for binding specificity and inhibition of aggrecanase activity (independently against apurified bovine aggrecan and synthetic human peptide [QTVTWPDMELPLPRNITEGEARGSVILTVKPIFEVSPSPLK–biotin])substrate to identify hybridomas of interest and cloned by limiting dilution. Humanization of murine antibodies was performed essentially as described [2] using a best-fit CDR grafting technique. Briefly, mouse CDRs were grafted onto suitable human acceptor frameworks which showed high sequence homology with the original framework region of each mAb. GSK-proprietary databases of human germline sequences were searched to identify best match frameworks suitable for grafting of the CDRs. Key residues in the mouse framework were identified which might be important in maintaining the correct conformation of the CDRs. These were considered as possible “back mutations” which might be necessary to retain the binding affinity and were incorporated into the various designs for the humanized molecules. Humanized variants were selected by retained/increased binding affinity/selectivity and functional potency

ARGS Neoeptiope detection

Assay plates were coated with a monoclonal anti-aggrecan capture antibody (Invitrogen, Cat# AHP0022), washed and experimental samples were added to the wells followed by 2hr incubation. Plates were washed, and incubated with a biotin (DELFIA) or SulfaTag (electrochemiluminescent) labeled ARGS neoepitope specific detection antibody [3] (OA-1, 50ng/well) followed by another wash. Captured ARGS neoepitope was detected using streptavidin−europium (500pg/well) for 30 min and enhancement solution (200μL/well) for 5 min followed by europium signal acquisition (320 nm excitation, 615 nm emission) on an Envision plate reader (DELFIA) or directly in a MSD SECTOR Imager 6000 (electrochemiluminescent). In the case of serum analysis ARGS concentrations were calculated using a standard curve of ADAMTS-5-treated human aggrecan IGD as described [4].

Substrate selectivity

GSK2394002 was assessed for the ability to inhibit ADAMTS-5-mediated proteolysis of two substrates (aggrecan and brevican) using an in vitro biochemical assay format. Incubation of purified human ADAMTS-5 [100nM final] with varying concentrations of GSK2394002 [ranging from 15.6 – 1000nM final] was performed in individual tubes for 30 minutes to allow for binding. Following pre-incubation, 2ug of recombinant human aggrecan IGD [1.4uM final] or brevican [1.02uM final] substrate (both from R&D Systems) was added and incubated for 1 hour. Separate tubes containing: ADAMTS-5 alone, aggrecan IGD alone, aggrecan IGD + ADAMTS-5, GSK2394002 alone, GSK2394002 + ADAMTS-5, brevican alone and brevican + ADAMTS-5 were run in parallel as controls. All components were diluted in assay buffer (10mM HEPES, 150mM NaCl2, 1uM ZnCl2, 1mM CaCl2, 0.05% NP-40 at pH 7.4) and reactions were performed at 37oC in 25uL total volume. The reactions were stopped with addition of 2uL of 270uM EDTA [20uM final]. Differential substrate proteolysis in each sample was resolved by electrophoresis on NuPAGE Novex 4-12% Bis-Tris gradient gels (Invitrogen) with MOPS running buffer. Gels were stained with SimplyBLUE total protein stain (Invitrogen) and images captured using a Licor Odyssey imaging system.

Crystallography

Mouse anti-human ADAMTS-5 mAbs7B4.1E11 and 12F4.1H7 were expressed in mouse hybridoma cultures and purified. Fab fragments were generated by papain cleavage using an ImmunoPure Fab Preparation Kit (Pierce) and used without additional purification. Fabs were concentrated and crystallized using the hanging drop vapor diffusion method. Structures were solved by X-ray diffraction and submitted toProtein Data Bank (PDB);7B4.1E11 Fabat 2.6Åresolution (PDB: 4X80) and12F4.1H7 Fabat 2.3Åresolution (PDB:4X8J). The ADAMTS proteins used for subsequent modeling were publically available and obtained directly from the PDB site: ADAMTS-4catalytic + disintegrin (apo form)2.8Å resolution(PBD: 3B2Z) and ADAMTS-5 catalytic + disintegrin with inhibitor bound2.6Å resolution (PDB: 2RJQ).

Structure modeling

Computational mAb Structural Humanization

In order to generate humanized structures of the ADAMTS-5 Fabs (GSK2394000 and GSK2394002) and themurine ADAMTS-4 Fab7E8.1E3, the protein sequences (amino acid sequences listed below)inMolecular Operating Environment (MOE) antibody modeler [5,6]. The MOE modeler employs a knowledge-based approach that utilizes a database of available antibody structures (available in PSILO database;Chemical Computing Group, Montreal). In Fab homology modeling, appropriate templates for the framework region and CDR loops were first found by searching against the antibody structure database. Then, the respective loop templates were grafted onto the designated light chain and heavy chain frameworks, followed by energy minimization in the transition area between CDRs and frameworks in order to relax strained geometries using AMBER12EHT force field. In order to gain confidence in the modeled GSK2394000 and GSK2394002 Fab structures each was superimposed with its solved parental murine crystal structure (7B4.1E11 and 12F4.1H7, respectively) as shown in Supplementary Figure 3. Superimposed structures with a root mean square deviation (RMSD) less than 1.5Åfrom the murine crystal structure were considered a good model. RMSD is the square root of the mean of the square of the distances between the matched atoms in the two structures: RMSD = SQRT[{SUM(dii)2}/N], where diiis the distance between the ithatom of structure 1 and the ith atom of structure 2 and N is the number of atoms matched in each structure. The RMSD is 0 for identical structures, and its value increases as the two structures become more distinct. RMSD values are considered as reliable indicators of variability when applied to conformations of the same protein. RMSD comparisons for the superimposed structures: 7B4(416 atoms) = 0.363Å and 12F4 (346 atoms) = 1.249Å. One exception was noted in the 12F4.1H7 crystal structure where poor resolution for a portion of CDRH2 (spanning amino acids 53-68) led to an inability to accurately compare this region. Without a matched parental crystal structure the 7E8.1E3 Fab structure relied solely on computational modeling based on the variable region sequence (below) and comparison to the available antibody structure database in PSILO.

Fab Antibody Amino Acid Sequence(CDRs)

GSK2394000 Heavy chain

EVQLVESGGGLVQPGGSLRLSCAASGFTFSDAWMDWVRQAPGKGLEWVAEIRNKANNHARHYAESVKGRFTISRDNAKNSLYLQMNSLRAEDTAVYYCARTYYYGSSYGYCDVWGQGTLVTVSSASTKGPSVFPLAPSSKSTSGGTAALGCLVKDYFPEPVTVSWNSGALTSGVHTFPAVLQSSGLYSLSSVVTVPSSSLGTQTYICNVNHKPSNTKVDKKVEPKSCDKTH

GSK2394000 Light chain

DIQMTQSPSSLSASVGDRVTITCRTSENIYSYLAWYQQKPGKAPKLLIYNAKTLAEGVPSRFSGSGSGTDFTLTISSLQPEDFATYYCQHHYGTPWTFGQGTKLEIKRTVAAPSVFIFPPSDEQLKSGTASVVCLLNNFYPREAKVQWKVDNALQSGNSQESVTEQDSKDSTYSLSSTLTLSKADYEKHKVYACEVTHQGLSSPVTKSFNRGEC

GSK2394002 Heavy chain

EVQLLESGGGLVQPGGSLRLSCAASGFTFSDAWMDWVRQAPGKGLEWVAEIRHKANDHAIFYDESVKGRFTISRDDSKNTVYLQMNSLRAEDTAVYYCTSPFAYWGQGTLVTVSSASTKGPSVFPLAPSSKSTSGGTAALGCLVKDYFPEPVTVSWNSGALTSGVHTFPAVLQSSGLYSLSSVVTVPSSSLGTQTYICNVNHKPSNTKVDKKVEPKSCDKTH

GSK2394002 Light chain

DIQMTQSPSSLSASVGDRVTITCKASQSVGTTIVWYQQKPGKAPKLLIYSASNRHTGVPSRFSGSGSGTDFTLTISSLQPEDFATYYCQQYTSYPFTFGQGTKLEIKRTVAAPSVFIFPPSDEQLKSGTASVVCLLNNFYPREAKVQWKVDNALQSGNSQESVTEQDSKDSTYSLSSTLTLSKADYEKHKVYACEVTHQGLSSPVTKSFNRGEC

7E8.1E3 Heavy Chain QVQLQQSGAELVRPGASVTLSCKASGYTFTDYEIHWVKQTPVHGLEWIGPIDPETGNTAYNQKFKGKAIMTVDKSSSTAYMELRSLTSEDSAVYYCTREGLRGHWYFDVWGAGTTVTVSSAKTTPPSDYPLVPGSAAQTNSMVTLGCLVKGYFPEPVTVTWNSGSLSSGVHTFPAVLQSDLYTLSSSVTVPSSTWPSQTVTCNVAHPASSTKVDKKIVPR

7E8.1E3 Light Chain

QIVLIQSPAIMSAFPGERVTMTCSASLSVTYMYWYQQRPGSSPRLLISDTSNLASGVPVRFSGSGSGTSYSLTISRVEAEDAATYYCQQWSYYPVTFGAGTKLELKRADAAPTVSIFPPSSEQLTSGGASVVCFLNNFYPKDINVKWKIDGSERQNGVLNSWTDQDSKDSTYSMSSTLTLTKDEYERHNSYTCEATHKTSTSPIVKSFNRNE

Structure best-fit modeling

High resolution structures of antibody Fab fragments and antibody-antigen complexes are useful for analyzing the binding interface and to make rational choices for affinity maturation. Using a Rosetta computational docking program [5] the structural complex is modeled to predict and characterize a binding interface between two proteins. ADAMTS5-5 catalytic and disintegrin domain structure (PDB: 2RJQ), with small molecule inhibitor removed,was independently docked with the modeled humanized versions of the 7B4.1E11(GSK2394000)and 12F4.1H7(GSK2394002) structures as described above. ADAMTS-4 catalytic and disintegrin domain (PDB: 3B2Z) was similarly docked with the modeled 7E8.1E3Fab. The first stage of the algorithm employs a rigid-body Monte Carlo search, translating and rotating the antigen around the surface of the Fab, using residue-scale interaction potentials. An alignment score directs the antigen toward the antibody CDR loops. After the low-resolution search, explicit side chains are added to the protein backbones using a backbone-dependent rotamer packing algorithm. A Monte Carlo-plus-minimization scheme then efficiently samples a set of local minima in a small region of docking conformation space by simultaneously optimizing the side-chain conformations and the rigid-body position resulting in identification of the top 200 best-fits. Ultimately, the top 10binding predictions were overlaid andFab orientation clustering was used to assess confidence in a single preferential binding site and docking orientation. Models demonstrating a high degree of consistency for the top 10 fits (within 2.5Å resolution) in the clustering around a specific region of the protease met the confidence threshold and the top ranked binding prediction wasprogressed to bond energy analysis at the protease/antibody interface, while those that did not (ADAMTS-4 with ADAMTS-5 mAbs, for example, which demonstrated a wide distribution across the protease surface) were deemed inefficient binders, consistent with the observed binding selectivity profile for these mAbs (manuscript Table 1 and Figure 1).

Computational protease/antibody interface bond energy prediction

The top ranked docking prediction for the ADAMTS-5 mAbs (GSK2394000 and GSK2394002) with ADAMTS-5 from above were used in an Alanine substitution scan for each predicted hydrogen bond site to determine the importance of a residue to the overall stability and affinity of the interaction. Alanine was used because of its side chain is non-bulky and chemically inert, but still has secondary structure preferences similar to most amino acids. In the model, stabilizing or destabilizing energy relative to the wild-type antibody was assessed computationally using the Rosetta program [5] and resulting change in relative hydrogen bond energy(Kcal/mol) as shown (Figure 2c). If the predicted change in bond energy is positive, the wild-type is contributing more favorably to the binding free-energy than the alanine mutant, in contrast, if the energy change is negative, the mutant is contributing more favorably. The magnitude indicates how important a particular energy estimator is to determining the relative free energy change of the mutant compared to the wild-type.

Human OA CartilageExplant Processing andAssessment of Treatment Efficacy

Cartilage was carved from bone, cut into uniform 3-mm diameter disks of similar thickness and cultured[DMEM with 10% FBS, Pennicillin (10IU/mL), Streptomycin (10g/mL) and L-glutamine (0.29mg/mL)] in 96-well plates. Tissue from a single donor was used in each plate. Following a 3-5 day culture equilibration, media was removed and treatments were addedwithout exogenous stimulation using a continuous ortransient(pulse-chase) treatment design. Each treatment was performed in replicates of seven/plate. Supernatants were harvested twice weekly throughout the 3-week unstimulated phase when half the well volume was removed (100L), stored frozen(-20°C) until analysis, and replaced with fresh media containingeither consistent treatment concentration (continuous design) or no treatment (pulse-chase design). Immediately following,in most experiments, explants were stimulated with exogenous human interleukin-1beta and oncostatin M [10ng/mL each]. Cartilage degradation in culture supernatants wasmonitoredusing ARGS neoepitope detection. Percent inhibition was calculated compared to isotypecontrol and small molecule inhibitor (GSK571949) treated samples

Animal use compliance

All studies were conducted in accordance with the GSK Policy on the Care, Welfare and Treatment of Laboratory Animals and were reviewed by Institutional Animal Care and Use Committees either at GSK orRush University.

Mechanical Allodynia

Male C57BL/6 mice (10 weeks of age) were pre-dosed with ADAMTS-5 (7B4.1E11) or isotype control mAbs [10mg/kg] 3 days prior to DMM surgery followed by weekly maintenance doses and direct effectson pain-related endpointswere assessed(supplementary methods). Throughout the 8-week study mice were tested using the up–down staircase method [6]. Mice were placed on a perforated metal floor (with 5-mm diameter holes placed 7 mm apart) within small Plexiglas cubicles, and a set of eight calibrated von Frey fibers (Stoelting Touch Test Sensory Evaluator Kit; ranging from ~0.04 g to ~2.0 g of force) were applied to the plantar surface of the hind paw until the fibers bowed, and then held for 3 s. The threshold force required to elicit withdrawal of the paw (median 50% withdrawal) was determined twice on each hind paw (and averaged) on each testing day, with sequential measurements separated by at least 5 min. Thresholds were assessed prior to surgery, and again 2, 4, and 8 weeks post-surgery.

Dorsal Root Ganglion (DRG) Immunofluorescence

Eight weeks post-DMM mice from the pain study described above were anesthetized by ketamine and xylazine and perfused transcardially with PBS followed by 4% paraformaldehyde in PBS at the completion of the study. The spinal column was dissected and postfixed in 4% paraformaldehyde overnight followed by cryopreservation in 30% sucrose in PBS. Individual ipsilateral L3-L5 DRG were embedded with OCT (Tissue-Tek), snap frozen, and cut into 12 μm sections. DRG sections were stained with the primary antibody anti-F4/80 (Abcam ab6640) and isotype-specific secondary AlexaFluor-633 antibody (Invitrogen) [7]. All captured images were exported to Adobe Photoshop CS 5.1 (Adobe, San Jose, CA) and adjustments were made to the brightness and contrast to reflect true colors.

Supplementary Tables:

Supplementary Table 1: Demographics and circulating ARGS profile of cynomolgus monkeys tested

Supplementary Table 2: Statistical correlation analysis of various parameters in study monkeys

The relationships were analyzed using Pearson correlation method, correlation coefficient and p-value were to be reported and positive value of correlation coefficient is indicative of positive correlation.

Supplementary Figures:

Supplementary Figure 1: GSK2394002 inhibits ADAMTS-5-mediated proteolysis of multiple substrates. Titration of GSK2394002 in the presence of ADAMTS-5 and either aggrecan IGD (top) or brevican (bottom) demonstrate dose-dependent inhibition of proteolysis (lanes 7-13). Individual assay controls in each experiment (lanes 2-6) are shown for molecular weight comparison of individual components and were treated similarly. Molecular weight markers are shown (lane 1). Identities of each fragment based on respective control comparison are shown on the right.

Supplementary Figure 2: Pre-treatment ARGS neoepitope levels in individual human OA cartilage explant cultures correlate with response to GSK2394002 treatment. A heat map depiction shows ARGS neoepitope levels are elevated in approximately half of individual OA patient cartilage explants prior to treatment. Background samples were taken immediately following treatment addition on day 0 while the experimental phase (response to treatment) was assessed following a single 5 day treatment with 670nM of GSK2394002 and isotype control mAbs. Values shown for pre-treatment, background and experimental phase are average ARGS neoepitope levels from 7 biological replicates at each timepoint. Note: although 17 independent experiments were performed in this study, two utilized separate samples from a single patient, 33745 and 33745a, to assess consistency of response in cartilage from an individual patient.

Supplementary Figure 3: Superimposed alignments of modeled humanized and crystallized parental murine ADAMTS-5 Fab structures support accurate modeling. (a) Overlay of 7B4.1E11 Fab crystal structure (red) and computationally modeled humanized GSK2394000 Fab (green). (b) Overlay of 12F4.1H7 Fab crystal structure (red) and computationally modeled humanized GSK2394002 Fab (green).

Supplementary References

1.Deng H, O'Keefe H, Davie CP, Lind KE, Acharya RA, Franklin GJ, et al. Discovery of highly potent and selective small molecule ADAMTS-5 inhibitors that inhibit human cartilage degradation via encoded library technology (ELT). J Med Chem 2012; 55: 7061-7079.

2.Co MS, Queen C. Humanized antibodies for therapy. Nature 1991; 351: 501-502.

3.Pratta MA, Su JL, Leesnitzer MA, Struglics A, Larsson S, Lohmander LS, et al. Development and characterization of a highly specific and sensitive sandwich ELISA for detection of aggrecanase-generated aggrecan fragments. Osteoarthritis Cartilage 2006; 14: 702-713.

4.Germaschewski FG, Matheny, C., Larkin, J., Liu, F., Thomas, L., Saunders, J., Sully, K., Whitall, C., Boyle, Y., Peters, G., Graham, N.M. Quantitation of ARGS aggrecan fragments in synovial fluid, serum and urine from osteoarthritis patients. Osteoarthritis Cartilage 2014; 22: 690-700.

5.Lyskov S, Gray JJ. The RosettaDock server for local protein-protein docking. Nucleic Acids Res 2008; 36: W233-238.

6.Chaplan SR, Bach FW, Pogrel JW, Chung JM, Yaksh TL. Quantitative assessment of tactile allodynia in the rat paw. J Neurosci Methods 1994; 53: 55-63.

7.Miller RE, Tran PB, Das R, Ghoreishi-Haack N, Ren D, Miller RJ, et al. CCR2 chemokine receptor signaling mediates pain in experimental osteoarthritis. Proc Natl Acad Sci U S A; 109: 20602-20607.