Abstract

Immune-checkpoint blockade has revolutionized cancer treatment. However, most patients do not respond to single-agent therapy. Combining checkpoint inhibitors with other immune-stimulating agents increases both efficacy and toxicity due to systemic T-cell activation. Protease-activatable antibody prodrugs, known as Probody therapeutics (Pb-Tx), localize antibody activity by attenuating capacity to bind antigen until protease activation in the tumor microenvironment. Herein, we show that systemic administration of anti–programmed cell death ligand 1 (anti–PD-L1) and anti–programmed cell death protein 1 (anti–PD-1) Pb-Tx to tumor-bearing mice elicited antitumor activity similar to that of traditional PD-1/PD-L1–targeted antibodies. Pb-Tx exhibited reduced systemic activity and an improved nonclinical safety profile, with markedly reduced target occupancy on peripheral T cells and reduced incidence of early-onset autoimmune diabetes in nonobese diabetic mice. Our results confirm that localized PD-1/PD-L1 inhibition by Pb-Tx can elicit robust antitumor immunity and minimize systemic immune-mediated toxicity. These data provide further preclinical rationale to support the ongoing development of the anti–PD-L1 Pb-Tx CX-072, which is currently in clinical trials.

Introduction

Immune checkpoints encompass multiple inhibitory and stimulatory mechanisms that are essential for maintaining self-tolerance and fine-tuning the duration and amplitude of immune responses. In patients with cancer, recombinant monoclonal antibodies (mAb) targeting inhibitory checkpoint molecules such as cytotoxic T-lymphocyte-associated antigen 4 (CTLA-4) or programmed cell death protein 1 (PD-1) and its ligand (PD-L1) can be effective strategies for reversing immune escape and elicit durable responses in a variety of indications (1, 2). A significant drawback to the administration of the mAb checkpoint inhibitors is the induction of systemic immune-related adverse events, which can be severe and life-threatening (3, 4). In addition to general inflammatory symptoms, patients treated with PD-1/PD-L1–targeting biologics can potentially develop life-threatening, organ-specific autoimmune reactions involving the gastrointestinal, hepatic, respiratory, endocrine, or central nervous systems (5–7). Although PD-1 expression is restricted to activated immune cells, its ligand PD-L1 can be found at low levels on a broad range of hematopoietic and nonhematopoietic cells such as those of the vascular endothelium, lung, liver, and muscle, indicating a potential role in maintaining the quiescence of T cells specific for self-antigens (8, 9). For some patients, the immune-related adverse events arising because of checkpoint inhibitor mAb therapy can necessitate dose reductions, delays, and discontinuations (10), which can be a significant impediment to successful treatment. In patients receiving multiple mAb checkpoint inhibitors, toxicities that limit therapy are not only more frequent but also more likely to be life-threatening and may require permanent discontinuation of checkpoint inhibitors (11). These clinical observations highlight the need for a new class of checkpoint inhibitors with reduced systemic toxicities.

In an attempt to decrease the toxicities of systemic mAb checkpoint inhibitor therapy, prior studies have examined the effectiveness of localized checkpoint inhibition (12, 13). These studies demonstrate that the localized delivery of checkpoint inhibitors can produce a therapeutic effect but do not establish whether this approach can also prevent systemic toxicity. It is reported that the therapeutic effect of checkpoint inhibition may require systemic activation of the immune system (14); therefore, it is not clear if the therapeutic and toxic effects can be separated with a local delivery strategy. We sought to test the hypothesis that localized PD-1/PD-L1 pathway inhibition can lead to effective antitumor immunity with minimal systemic enhancement of T-cell–mediated activity, including autoimmunity. To accomplish this, we generated constructs of Probody therapeutics (Pb-Tx) targeting the PD-1/PD-L1 pathway. Pb-Tx are a new class of proteolytically activated antibody prodrugs designed to minimize peripheral binding and promote tumor-specific activity (15, 16). The prodomain, an integral feature of Probody therapeutic technology, consists of a masking peptide that inhibits antigen binding connected to the antibody by a protease-cleavable linker (17). In healthy tissue where protease activity is tightly regulated, the prodomain remains predominantly intact, preventing target binding (18). However, there is considerable evidence for upregulation of specific proteases in the tumor microenvironment compared with normal noncancerous tissue (19–22). For example, LeBeau and colleagues document active matriptase, a serine protease that can also activate the Pb-Tx presented in our study, in human colon cancer samples by using an active-site mAb, whereas normal colon samples show no evidence of this protease activity (19). Therefore, in the tumor microenvironment, which has elevated protease activity, the prodomain is designed to be cleaved and the masking peptide removed so that the Pb-Tx can engage the target, eliciting its intended therapeutic effect (17, 23).

We conducted a series of studies to investigate whether tumor-localized checkpoint inhibition with Pb-Tx can induce tumor growth inhibition and minimize peripheral immune-mediated toxicities. We show that PD-1/PD-L1–targeting Pb-Tx elicited therapeutic responses that were similar to results elicited with parental unmasked antibody and have demonstrated attenuated systemic target engagement and toxicity. These findings may allow for the development of checkpoint inhibitors with a wider therapeutic window and improved clinical utility.

Materials and MethodsCell lines, antibodies, and recombinant proteins

Expi293F cells were purchased in 2013 from Life Technologies (catalog #A14527), subcultured, and cryopreserved per the vendor's recommendations and used for expression of recombinant proteins between passages 5 and 30. Human SAS cells were purchased in 2013 from the Japanese Collection of Research Bioresources (catalog #JCRB0260). MC38 cells were obtained in 2013 from Dr. Walter Storkus at the University of Pittsburgh, Pittsburgh, Pennsylvania. All cell lines were tested for a variety of viral and bacterial infectious agents, including Mycoplasma, by PCR prior to use (Idexx Laboratories IMPACT III PCR test). Human SAS cells and MC38 cells were maintained at fewer than 10 passages from the passage obtained from the vendor or source. Genotypic analysis for reauthentication purposes was not performed on the cell lines. Cells were cultured in RPMI-1640 (Thermo Fisher, catalog #72400047) or DMEM (Thermo Fisher, catalog #10564011) supplemented with 10% (v/v) fetal bovine serum (FBS; Thermo Fisher, catalog #10062147) as recommended by the vendors and maintained in a humidified atmosphere of 5% CO2 at 37°C.

Mouse IgG2a clone C1.18.4 (catalog #BE0085, RRID:AB_1107771) was purchased from Bio X Cell. Human IgG4 isotype control was purchased from BioLegend (catalog #403701). All antibodies and Probody molecules targeted to PD-1 and PD-L1 were expressed at CytomX in Expi293F cells with the ExpiFectamine 293 Transfection Kit (Thermo Fisher catalog #A14524) per manufacturer's recommendation, purified to >95% purity and >95% monomer by protein A affinity chromatography (MabSelect SuRe, Cytiva, catalog #17543803) and size exclusion chromatography (Sepax SRT-10-SEC300) on an AKTA Pure system (Cytiva) using vendor's recommended protocols. All recombinant proteins were formulated in phosphate-buffered saline (PBS), pH 7.0 or 7.2, sterile filtered (0.2 micron), and confirmed to have an endotoxin level of <2 endotoxin unit (EU)/mg.

Antibody discovery

A PD-L1–blocking antibody was identified from a fully human single-chain variable fragment (scFv) phage-display library under contract by Creative Biolabs, using recombinant His-tagged human or mouse PD-L1 extracellular domain (Sino Biological, catalogs #10084-H02H and #50010-M08H, respectively). Five unique scFv clones that bound human PD-L1 (hPD-L1) extracellular domain were isolated and retested for specificity. Of these five, one cross-reactive clone was selected for affinity maturation and in silico assessment of immunogenicity. The light and heavy variable domains from the affinity-matured scFv were PCR-amplified and cloned into expression vectors based on pCDNA3 (Invitrogen) containing light (mouse and human kappa) and heavy (mouse IgG2a and human IgG1) constant regions, respectively. The affinity-matured lead clone (C5H9v2) was transiently expressed and purified as described above and confirmed to bind hPD-L1-Fc (R&D Systems, catalog #156-B7) and mouse PD-L1-Fc (R&D Systems, catalog #1019-B7) proteins by direct-binding enzyme-linked immunosorbent assay (ELISA), described below, with comparable affinity. C5H9v2 mAb was tested for specificity using the direct-binding ELISA described below and did not bind to several other related Fc-tagged proteins (both human and mouse proteins) including CTLA-4 (Sino Biological, catalogs #11159-H02H6 and #50503-M02H), PD-1 (Sino Biological, catalogs #10377-H02H and #50124-M02H), B7–1 (Sino Biological, catalogs #10698-H02H and #50446-M02H), CD137 (Sino Biological, catalogs #10041-H02H and #50811-M02H), OX40 (Sino Biological, catalogs #10481-H02H and #50808-M02H), and GITR (R&D Systems, catalogs #689-GR-100 and #524-GR-050; Supplementary Fig. S1). Amino acid sequences of murine C5H9v2 (muC5H9v2) mouse IgG2a (mIgG2a) are shown below.

muC5H9v2 mIgG2a heavy-chain amino acid sequence

EVQLLESGGGLVQPGGSLRLSCAASGFTFSSYAMSWVRQAPGKGLEWVSSIWRNGIVTVYADSVKGRFTISRDNSKNTLYLQMNSLRAEDTAVYYCAKWSAAFDYWGQGTLVTVSSAKTTAPSVYPLAPVCGDTTGSSVTLGCLVKGYFPEPVTLTWNSGSLSSGVHTFPAVLQSDLYTLSSSVTVTSSTWPSQSITCNVAHPASSTKVDKKIEPRGPTIKPCPPCKCPAPNLLGGPSVFIFPPKIKDVLMISLSPIVTCVVVDVSEDDPDVQISWFVNNVEVHTAQTQTHREDYNSTLRVVSALPIQHQDWMSGKEFKCKVNNKDLPAPIERTISKPKGSVRAPQVYVLPPPEEEMTKKQVTLTCMVTDFMPEDIYVEWTNNGKTELNYKNTEPVLDSDGSYFMYSKLRVEKKNWVERNSYSCSVVHEGLHNHHTTKSFSRTPGK

muC5H9v2 mIgG2a light-chain amino acid sequence

DIQMTQSPSSLSASVGDRVTITCRASQSISSYLNWYQQKPGKAPKLLIYAASSLQSGVPSRFSGSGSGTDFTLTISSLQPEDFATYYCQQDNGYPSTFGGGTKVEIKRADAAPTVSIFPPSSEQLTSGGASVVCFLNNFYPKDINVKWKIDGSERQNGVLNSWTDQDSKDSTYSMSSTLTLTKDEYERHNSYTCEATHKTSTSPIVKSFNRNEC

A humanized IgG4 anti-human PD-1 (anti–hPD-1; clone A1.5) was originally identified as a mouse hybridoma as described below. Six NZBWF1/J female mice (The Jackson Laboratory, catalog #100008) were immunized in the right flank with recombinant human PD-1 (Sino Biological, catalog #ABIN2181605) on days 0, 7, and 21. Serum was taken from immunized mice on day 28 and binding to HEK293-hPD-1 (cells transfected with an expression vector encoding human PD-1 (Origene, catalog # SC117011)) was measured. All six mice showed positive binding. Splenocytes were isolated from mice 1, 3, and 6 and fused with SP0 mouse B cells; likewise splenocytes were isolated from mice 2, 4, and 5 and fused with SP0 mouse B cells, giving rise to pools of hybridomas, m136 and m245. The hybridoma pools, m136 and m245, arising from the two fusions were plated for single clones, and clonal culture supernatants were assayed for antibodies capable of binding to HEK293-hPD-1, and not to untransfected HEK293, cells. Hybridoma clones expressing PD-1–specific antibodies were chosen for further analysis. Clone A1.5 mAb was tested for specificity using the direct-binding ELISA described below and did not bind to several other related Fc-tagged proteins (both human and mouse proteins) including CTLA-4 (Sino Biological, catalogs #11159-H02H6 and #50503-M02H), PD-1 (Sino Biological, catalogs #10377-H02H and #50124-M02H), B7–1 (Sino Biological, catalogs #10698-H02H and #50446-M02H), CD137 (Sino Biological, catalogs #10041-H02H and #50811-M02H), OX40 (Sino Biological, catalogs #10481-H02H and #50808-M02H), and GITR (R&D Systems, catalogs #689-GR-100 and #524-GR-050; Supplementary Fig. S2).

A commercially available hamster antibody specific for mouse PD-1 (clone J43; Bio X Cell, catalog #BP0033-2) was converted into a mouse IgG2a isotype harboring the D265A mutation that minimized Fc effector activity. Briefly, DNA fragments encoding the variable light and the variable heavy domains of J43 were synthetically produced (GeneArt, Thermo Fisher) and cloned into an expression vector (based on pCDNA3, Invitrogen) containing the constant mouse kappa light and constant mouse mIgG2a domains, respectively.

Engineering of Probody molecules

A previous description for methods regarding the design and engineering of Probody therapeutic constructs has been published (17). Briefly, PD-1 and PD-L1 antibody and Fab fragments (made by digesting antibodies with Immobilized Papain, Thermo Fisher, catalog #20341, per the vendor's recommended protocol) were used to select masking peptides from bacterial surface–displayed peptide libraries. The bacterial library cells were subjected to repeated rounds of magnetically activated cell sorting with Protein A Dynabeads (Thermo Fisher, catalog #10008D) after incubation with unlabeled PD-1 or PD-L1 antibody and fluorescently activated cell sorting (FACSAria II, BD) after incubation with Alexa Fluor 488 labeled (Thermo Fisher, catalog #20000) PD-1 or PD-L1 antibody or Fab fragments to enrich for potential masking peptides. Protease-cleavable substrates were identified as previously described by screening cellular libraries of peptide substrates and optimized based on the aberrant expression of tumor-associated proteases (17). Several individual clones (96–192) from the final selected masking-peptide population were sequenced via Sanger sequencing (ELIM Biopharmaceuticals) and a set of 15 to 20 peptides, chosen based on their frequency in the sequenced set and sequence diversity, were assessed for masking activity. The chosen peptides alongside a cleavable linker were recombinantly fused to the N-terminus of the parental light chain to generate masked light chains by cloning synthetic DNA fragments between the signal peptide and the mature light chain in the light-chain plasmid. The proteins encoded by the antibody heavy chain and the set of masked light chains were expressed and purified as described above. The binding activity of the parental unmasked antibody and its masked derivatives was determined by the direct-binding ELISA described below. The presence of a functionally masking prodomain decreases the binding of activatable antibodies and manifests as an increase in the observed ELISA half-maximal effective concentration (EC50) for the activatable antibody. The ratio of the observed EC50 of the activatable antibody to that of the original antibody is used as a measure of the masking activity of the prodomain and its component masking peptide. Amino acid sequences of CX-630 Pb-Tx are shown below.

CX-630 Pb-Tx heavy-chain amino acid sequence

EVQLLESGGGLVQPGGSLRLSCAASGFTFSSYAMSWVRQAPGKGLEWVSSIWRNGIVTVYADSVKGRFTISRDNSKNTLYLQMNSLRAEDTAVYYCAKWSAAFDYWGQGTLVTVSSAKTTAPSVYPLAPVCGDTTGSSVTLGCLVKGYFPEPVTLTWNSGSLSSGVHTFPAVLQSDLYTLSSSVTVTSSTWPSQSITCNVAHPASSTKVDKKIEPRGPTIKPCPPCKCPAPNLLGGPSVFIFPPKIKDVLMISLSPIVTCVVVDVSEDDPDVQISWFVNNVEVHTAQTQTHREDYNSTLRVVSALPIQHQDWMSGKEFKCKVNNKDLPAPIERTISKPKGSVRAPQVYVLPPPEEEMTKKQVTLTCMVTDFMPEDIYVEWTNNGKTELNYKNTEPVLDSDGSYFMYSKLRVEKKNWVERNSYSCSVVHEGLHNHHTTKSFSRTPGK

CX-630 Pb-Tx light-chain amino acid sequence

QGQSGSIFCHMGVVVPQCANYGGGSSGGSGGSGGTSTSGR-SANPRGGGSDIQMTQSPSSLSASVGDRVTITCRASQSISSYLNWYQQKPGKAPKLLIYAASSLQSGVPSRFSGSGSGTDFTLTISSLQPEDFATYYCQQDNGYPSTFGGGTKVEIKRADAAPTVSIFPPSSEQLTSGGASVVCFLNNFYPKDINVKWKIDGSERQNGVLNSWTDQDSKDSTYSMSSTLTLTKDEYERHNSYTCEATHKTSTSPIVKSFNRNEC

Protease activation of Pb-Tx molecules

Pb-Tx constructs were activated by cleavage with either urokinase-type plasminogen activator (uPA) or matriptase (MT-SP1) for in vitro binding and blocking assays. Test agents were incubated with 10% volume of uPA (R&D Systems, catalog 1310-SE) or MT-SP1 (catalog 3946-SEB) overnight (∼16 hours) at 37°C. The concentration of protease used in both instances was >10 times that required to ensure complete cleavage of the substrate. Protease-digested samples were purified using MabSelect SuRe LX beads (GE Lifesciences, catalog #17-5474-01), buffer exchanged into PBS pH 7.2 by dialysis, and analyzed by sodium dodecyl sulfate polyacrylamide gel electrophoresis and high-performance liquid chromatography to verify that only cleaved Pb-Tx molecules are present.

Cytomegalovirus recall assay

Peripheral blood mononuclear cells (PBMC) from a cytomegalovirus (CMV)-positive donor (HemaCare) were plated at 2.5 × 105 cells per well in the presence of 4 μg/mL CMV antigen (Astarte Biologics, catalog #1004) and titrations of C5H9v2 (hPD-L1 mAb), CX-072 Pb-Tx, A1.5 mAb (hPD-1 mAb), CX-188 Pb-Tx or hIgG4 isotype control in a total volume of 200 μL DMEM (Thermo Fisher, catalog #10564011) supplemented with 10% (v/v) FBS (Thermo Fisher, catalog #10062147). Cells were incubated at 37°C in 5% CO2. After 4 days, IFNγ levels were detected and quantified from the supernatant using an IFNγ ELISA kit (Life Technologies, catalog #KHC4021). The ELISA assay was performed according to the manufacturer's instructions. Absorbance was read at a wavelength of 450 nm using a microplate reader (BioTek ELx405). Data were graphed in Prism software version 6 (GraphPad Software). EC50 values were determined using nonlinear regression 4-PL analysis.

ELISAsDirect-binding ELISA

Assessments of direct binding to recombinant PD-L1 derived from various species were performed using a sandwich ELISA in which immobilized recombinant proteins were incubated with titrations of the respective antibodies or Pb-Tx (± proteolytic preactivation with uPA or MT-SP1 protease). High-binding 96-well plates (Nunc MaxiSorp, Thermo Fisher) were coated overnight at 4°C with 50 μL of 1 μg/mL recombinant hPD-L1-Fc (R&D Systems, catalog #156-B7), cynomolgus monkey PD-L1-Fc (Sino Biological, catalog #90251-C02H), and rat PD-L1-Fc (Sino Biological, catalog #80450-R02H); mouse PD-L1-Fc (R&D Systems, catalog #1019-B7) was coated on plates at 0.5 μg/mL. Plates were washed six times with 300 μL in PBS, 0.05% Tween-20 (wash buffer), then blocked with PBS, 0.05% Tween, 1% bovine serum albumin (BSA blocking buffer) for 1 hour at room temperature. Plates were washed as before and incubated in 50 μL of an 11-point, three-fold serial dilution for antibodies targeting PD-L1 or PD-1 (C5H9v2 or J43v2, respectively) or Probody molecules (CX-072 or CX-284) accordingly. Activated Pb-Tx constructs were generated by incubating with uPA or MT-SP1 diluted in blocking buffer, for 1 hour at room temperature. Plates were washed as before, and 100 μL of anti-human IgG–horseradish peroxidase (HRP) conjugate, Fab-specific (Sigma-Aldrich, catalog #A0293-1ML; 1:3,000 dilution), or anti-mouse IgG–HRP conjugate (Sigma-Aldrich, catalog #A2554-1ML; 1:5,000 dilution) was added for 30 minutes at room temperature. Plates were washed as before and developed with 100 μL of chromogenic substrate 1-Step Ultra TMB-ELISA Substrate solution (Thermo Fisher, catalog #34029), and the reaction was stopped with 100 μL of 1N HCl. Absorbance was read at a wavelength of 450 nm using a microplate reader (BioTek). Data were graphed using Prism software. The EC50 values were determined using nonlinear regression and a 1-site–specific binding analysis.

Competition ELISA

The competition ELISA was performed as described above for the direct-binding ELISA with the following modifications. High-binding 96-well plates (Nunc MaxiSorp, Thermo Fisher) were coated with 50 μL of recombinant hPD-L1-Fc or recombinant mouse PD-1-Fc (R&D Systems, catalog #10-PD21). The associated human PD-1-Fc (Chimerigen Laboratories, catalog #CHI-HF-210PD1), human B7-1-Fc (Chimerigen Laboratories, catalog #CHI-HF-210CD80), or mouse PD-L1-Fc (R&D Systems, catalog #1019-B7) ligands were biotinylated using EZ-Link Sulfo-NHS-LC-Biotin (Thermo Fisher, catalog #21327) according to the manufacturer's instructions and dialyzed overnight with PBS buffer to remove unreacted Sulfo-NHS-LC-Biotin. Biotinylated PD-1-Fc (4 nmol/L) or PD-L1-Fc fusion (0.6 nmol/L) proteins were added to the wells and incubated for 1 hour at room temperature in the presence or absence of an 11-point, 3-fold serial dilution of respective blocking mAb, Pb-Tx, or protease-activated Pb-Tx. Plates were washed as described in the binding ELISA. Streptavidin–HRP conjugate (Thermo Scientific, catalog #21126) was added to the wells at a 1:5,000 dilution and incubated for 30 minutes at room temperature. Plates were washed as previously described and developed with 100 μL of chromogenic substrate 1-Step Ultra TMB-ELISA Substrate solution (Thermo Fisher, catalog #34029). The reaction was stopped with 100 μL of 1N HCl. Absorbance was read at a wavelength of 450 nm using a microplate reader (BioTek ELx405). Data were graphed in Prism software.

The ability of mouse PD-L1–specific antibody/Pb-Tx to block mouse PD-1/PD-L1 interactions was determined as described above with the exception that mouse PD-L1 was coated on plates at 0.5 μg/mL. Biotinylated mouse PD-1 was used at 2 nmol/L.

Cell-binding assay

Binding of hPD-L1 mAb (C5H9v2), CX-072 Pb-Tx, and preactivated CX-072 Pb-Tx to PD-L1–positive cells of human (SAS human cell line), mouse (MC38 mouse cell line), rat (Sprague Dawley rat peripheral blood CD4+ T lymphocytes, BioIVT, catalog #RATWBLIHPO-M), and cynomolgus monkey (peripheral blood CD4+ T lymphocytes, BioIVT, catalog #CYNWBLIHP-M) origin was evaluated by flow cytometry. Adherent cells were detached from flasks by rinsing once with PBS followed by Versene (Gibco, catalog #15040–066) at 37°C, washed with PBS, 2% FBS (staining buffer), and plated between 100,000 and 200,000 cells per well in a 96-well U-bottom plate and then pelleted at 200 × g for 3 minutes. Cells were incubated for 1 hour at 4°C with 50 μL of 8-point, 4-fold serial dilution of hPD-L1 mAb (C5H9v2), hPD-L1 Pb-Tx, or activated hPD-L1 Pb-Tx, serially diluted in staining buffer. After the incubation, the cells were washed three times with 200 μL of staining buffer and pelleted by centrifugation as described above. Anti-human IgG-AF647 (Jackson ImmunoResearch Laboratories, catalog #709-605-149) was added to cells at a 1:400 dilution for 30 minutes at 4°C. Cells were diluted with 200 μL of staining buffer and washed 3 times as previously outlined. Staining buffer (200 μL) was added to cells and cells analyzed on a flow cytometer (MACSQuant Analyzer 10, Miltenyi Biotec). Live cells were gated from forward/side scatter plots and the median fluorescence intensity (MFI) of AF647 was measured. Representative dot plots and gating are provided in Supplementary Fig. S3. Data were analyzed using FlowJo software (FlowJo, LLC) and graphed in Prism software (GraphPad Software). EC50 values were determined using three-parameter nonlinear regression analysis.

MC38 mouse tumor model

All in vivo work was performed in accordance with the Guide for the Care and Use of Laboratory Animals under a protocol approved by the Institutional Animal Care and Use Committee (protocols AP103; MC38 model and AP010; nonobese diabetic mouse model). CytomX Therapeutics, Inc. is an AAALAC accredited institution. Female C57BL/6 mice, obtained from Charles River Laboratories (RRID:IMSR_CRL:027), at 7 weeks of age were each inoculated with 1.5 × 106 MC38 cells subcutaneously into the right hind flank in a volume of 0.1 mL serum-free RPMI-1640 medium (Life Technologies). Tumor volumes and body weights were recorded twice weekly. Tumor dimensions were determined by caliper measurements, and tumor volume was calculated using the formula (a × b2)/2, where a is the longest and b is the shortest diameter. Tumor volumes were used to calculate percent tumor growth inhibition, expressed as [1 − (T/C)] × 100, where T and C are the mean tumor volumes of the treated and control groups, respectively, on a given day. When mean tumor size reached 80 mm3, mice were randomized into treatment groups. All test and control agents, including muC5H9v2, CX-603 Pb-Tx, CX-284 Pb-Tx, J43v2 mAb, and anti–CTLA-4 mAb (Bio X Cell, catalog #BP0164), were blinded and were administered by intraperitoneal injection beginning on day 1. The timing and dose differed in different experiments and can be found in figure legends for individual experiments. All mice were examined daily for health status and were euthanized by CO2 inhalation when tumors reached 2,000 mm3, body weight loss exceeded 20%, another humane endpoint was met, or at study termination.

Western capillary electrophoresis

Tumor, spleen, and liver were harvested 4 days after dose from mice withMC38 tumor dosed with 5 mg/kg of either CX-630 or the parental antibody and frozen. Frozen tissues (15–35 mg) were cut on the ThermalTray slim low-profile platform over dry ice (BioCision, catalog #BCS-252). Immediately after cutting, each sample was placed in a cryostat (−20°C). Immediately after taking each sample out from the cryostat, 100 μL of Pierce IP Lysis Buffer (Thermo Fisher, catalog #87788) with 1X Halt Protease Inhibitor Cocktail and 0.5 M ethylenediaminetetraacetic acid (EDTA; Thermo Fisher, catalog #78438) was added. Protein was then extracted using the Barocycler 2320EXT (Pressure BioScience) for 40 cycles of 30 seconds at 35,000 psi and 10 seconds at ambient pressure at 20°C. The samples were centrifuged at 18,500 × g for 15 minutes at 4°C, and the supernatant was collected. Samples were frozen at −80°C until use. Activated and intact anti–PD-L1 Pb-Tx was detected using an anti-idiotype primary antibody (clone 17G1, CytomX Therapeutics) and anti-mouse secondary antibody (ProteinSimple, catalog #042-205). Capillary electrophoresis was performed on the Sally Sue instrument (ProteinSimple) following standard instrument protocol. Probody concentrations in tumor/tissue were obtained from Compass software peak area data in SoftMax Pro software where Basic Endpoint protocol was used to back-calculate against active and intact anti–PD-L1 Pb-Tx by 4-parameter logistic standard curves.

CD8 immunohistochemistry

C57BL/6 mice bearing MC38 tumors (average volume of 80 mm3) were treated with muC5H9v2 mAb, CX-630 Pb-Tx, or isotype control (5 mg/kg) on days 0, 3, and 7. On day 8, tumors were excised, fixed in formalin, and embedded in paraffin blocks. CD8 immunohistochemical analysis was performed by Cureline, Inc. Briefly, 5 μm sections were cut, deparaffinized, and rehydrated followed by antigen retrieval using an EDTA buffer. Sections were washed and incubated in a blocking solution. Primary staining was performed with a 200-fold dilution of rabbit anti-CD8 antibody (Abcam, catalog #ab217344) or isotype control (Abcam, catalog #ab172730) in staining buffer (PBS with 0.5% Tx-100 and 1% FBS) for 1 hour at room temperature. After several washes, tissue sections were stained with an anti-rabbit HRP-conjugated secondary antibody for 20 minutes at room temperature (Dako, catalog #P0448). After a final wash step, the slides were incubated with a 3,3′-diaminobenzidine chromogen for 10 minutes at room temperature. After rinsing, slides were counterstained with hematoxylin, dehydrated, cleared, and mounted. Brightfield whole-tissue scans where performed on the slides using the Vectra Polaris Imaging System (Akoya Biosciences) at a 20× magnification. Snapshots images were captured digitally from the scan at 10× magnification using Phenochart software version 1.0.

Mouse plasma IgG2a capture ELISA

The plasma concentrations of muC5H9v2 mAb, CX630 Pb-Tx, or mouse IgG2a isotype control were measured by a capture ELISA using immobilized anti-mouse IgG2a(a), a haplotype-specific antibody to detect the test agents in a C57BL/6 background. Briefly, high-binding 96-well plates (Nunc MaxiSorp, Thermo Fisher) were coated overnight at 4°C with 60 μL of 4 μg/mL anti-mouse IgG2a(a) (clone 8.3, BD Biosciences). Plates were washed with wash buffer (PBS, 0.05% Tween-20), then blocked with a PBS/1% BSA solution for 1 hour at room temperature. Samples were diluted 1:20, 1:200, or 1:2,000 in blocking buffer. Plates were washed as before and incubated in 60 μL of the test agents: an 11-point, 2-fold serial dilution (concentration range, 0.488–500 ng/mL) performed in replicates of muC5H9v2 mAb or CX630 Pb-Tx diluted in blocking buffer, for 1 hour at room temperature. Plates were washed as before and 60 μL of anti-mouse IgG2a–biotin conjugate (clone 8.3, BD Biosciences, catalog #553502) in blocking buffer was added for 1.5 hours at room temperature. Plates were washed as before and 60 μL of streptavidin–HRP diluted 1:5,000 (Pierce, catalog #21140) in block buffer was added for 30 minutes at room temperature. Plates were washed as before and developed with 100 μL of chromogenic substrate 1-Step Ultra TMB-ELISA Substrate Solution (Thermo Fisher, catalog #34029). The reaction was stopped with 100 μL of 1N HCl. Absorbance was read at a wavelength of 450 nm using a microplate reader (BioTek ELx405). Data were graphed in Prism software, and concentrations were interpolated using 4-parameter logistic nonlinear regression analysis.

Target occupancy assay

Female C57BL/6 mice with established subcutaneous MC38 tumors 60 to 120 mm3 in size were given a single intraperitoneal injection of muC5H9v2 mAb, CX-630 Pb-Tx, or mouse IgG2a control antibody at 5, 3, 1, or 0.3 mg/kg. Four days after dosing, terminal bleeds were taken with a minimum of 200 μL (+ EDTA) for flow cytometry analysis and 100 μL (lithium heparin) for processing to plasma and stored at −80°C for exposure assessment in single aliquots. Blood was taken from the remaining untreated control tumor-bearing mouse to serve as a negative control and for compensation control for flow cytometry analysis. The amount of muC5H9v2 or CX-072 bound to the surface of CD4+ T cells was assessed by flow cytometry. A 50 μL sample of whole blood from each animal was incubated with 5 μL of mouse FcR block (Miltenyi Biotec; catalog #130-092-575) in duplicate wells for 10 minutes at 4°C. The following dilutions were prepared in 1:10 mouse FcR block in Stain Buffer BSA (BD Biosciences, catalog #554657) and incubated for 1 hour: For the ex vivo samples, 55 μL of PBS was added to one set of wells. For the saturated samples, 55 μL of 40 μg/mL mouse PD-L1 antibody (C5H9v2) was added to one set of duplicate wells. For background control, 55 μL of 40 μg/mL of mouse IgG2a isotype control was added to the wells of untreated animals. Red blood cells were lysed using 1,100 μL of RBC Lysis Solution (Miltenyi Biotec; catalog #130-094-183) for 10 minutes. Cells were pelleted at 300 × g for 5 minutes. Cells were washed once with RBC lysis buffer followed by two washes with 200 μL of stain buffer. Anti-mouse H+L IgG-biotin (Jackson ImmunoResearch Laboratories, catalog #715-065-151, RRID:AB_2340785) at 1:500 dilution was added to the wells for 30 minutes. Cells were washed twice with 200 μL of stain buffer and incubated with a cocktail of anti-mouse CD4-Pacific Blue and anti-mouse CD45-APC-Cy7, and 1:500 dilution of streptavidin-phycoerythrin (streptavidin-PE; Invitrogen catalog #12-4317-87) for 30 minutes. Cells were washed twice with 200 μL of stain buffer before 7-aminoactinomycin D (7-AAD, BD Bioscience catalog #559925) was added to the wells to detect dead cells. Cells were analyzed on a flow cytometer (MACSQuant, Miltenyi Biotec). Live lymphocytes were gated from forward/side scatter plots, negative 7-AAD, and positive CD45 gates. The MFI of PE was measured in the CD4+ cells and graphed in Prism software. Background MFI was subtracted from all samples. Percent target occupancy was determined using the following equation: (ex vivo MFI of PE)/(saturated MFI of PE) × 100.

Nonobese diabetic diabetes model

Female nonobese diabetic (NOD)/ShiLtJ mice (The Jackson Laboratory, stock #001976) were maintained with free access to standard food and water. Nine-week-old female mice confirmed to be prediabetic (negative for glucosuria) were randomized and dosed with muC5H9v2 mAb, CX-630 Pb-Tx, or mouse IgG2a (BIW) by intraperitoneal injection. In the combination study, 5-week-old prediabetic female NOD mice were injected intraperitoneally with muC5H9v2 mAb, CX-630 Pb-Tx, or mIgG2a at 10 mg/kg on days 0, 4, and 7 in combination with either anti–CTLA-4 clone 9D9 (Bio X Cell, catalog #BP0164) or mIgG2b (Bio X Cell, catalog #BP0086) at 10 mg/kg. Body weights and urinary glucose levels were monitored daily. Urinary glucose was monitored daily using a glucose test strip (Diastix; Bayer Corporation). Blood obtained by means of a tail nick was tested for glucose with an Alpha TRAK 2 test strip inserted into a glucometer (Abbott Animal Care, now owned by Zoetis). Any animal with detectable glucose in urine (≥100 mg/dL) was immediately tested for blood glucose level. Animals with blood glucose levels ≥250 mg/dL for 2 consecutive days were designated as diabetic and euthanized by CO2 inhalation.

Statistical analysis

Statistical analysis was performed using GraphPad Prism software version 6 (GraphPad Software). All the data points represent the mean ± SD. Differences in tumor size between groups were assessed by one-way analysis of variance using the Dunnett correction for multiple comparisons. A P value of <0.05 was considered statistically significant. Kaplan–Meier survival plots were generated with Prism and were analyzed using the survival analysis tool, which performs a log-rank test.

ResultsCharacterization of PD-L1–neutralizing antibody

A phage-displayed scFv library was used to identify clones that bound specifically to PD-L1. Positive clones were characterized for their affinity and binding specificity. From this pool, a clone was selected for further development (C5H9v2; Supplementary Table S1). The variable light and heavy domains of the lead antibody were cloned into expression vectors for hIgG4 (S228P; C5H9v2) and mIgG2a muC5H9v2. The equilibrium binding affinity of C5H9v2 hIgG4 mAb to recombinant human, monkey, rat, and mouse PD-L1 fusion protein was determined by ELISA; nanomolar affinity was observed (Fig. 1A; ref. 24). Similar subnanomolar EC50 values were obtained for on-cell binding to PD-L1–expressing human, monkey, rat, and mouse cells by flow cytometry (Supplementary Table S2). The capacity of this antibody to inhibit the interaction of PD-L1 with its two known receptors, PD-1 and B7-1 (25, 26), was also evaluated by ELISA. The addition of single-digit nanomolar concentrations of C5H9v2 mAb abrogated all binding of PD-1 [half-maximal inhibitory concentration (IC50) = 0.65 nmol/L] or B7-1-Fc fusions (IC50 = 0.46 nmol/L) to plate-bound PD-L1 (Fig. 1B and C).

Figure 1.

C5H9v2 mAb is a high-affinity blocker of PD-L1 and can potentiate T-cell responses. A, The capacity of C5H9v2 hIgG4 mAb to bind and block immobilized PD-L1 was evaluated in an ELISA format. Serial dilutions of C5H9v2 mAb were incubated with recombinant plate-bound PD-L1-Fc fusion protein of human, mouse, rat, or cynomolgus monkey origin. Bound antibody was detected using an anti-human HRP-conjugated secondary antibody and is shown in relation to the concentration. Data shown are representative of 3 independent experiments. Serial dilutions of C5H9v2 mAb were added to wells containing biotinylated hPD-1 (B) or hB7-1 (CD80; C) protein and plate-bound PD-L1-Fc to measure blocking capacity. The binding of hPD-1 or hB7-1 to PD-L1 was detected using a streptavidin–HRP conjugate. Data shown are representative of 3 independent experiments. D, Blockade of PD-L1 by C5H9v2 mAb enhanced IFNγ production in a CMV recall assay. CMV-positive hPBMCs were plated at 350,000 cells/well in media containing serial dilutions of C5H9v2 mAb or isotype control and stimulated with 4 μg/mL CMV lysate for 4 days, at which point IFNγ was measured in the supernatants by ELISA. Data shown are representative response from one donor (n = 3 donors tested) of 3 independent experiments. All data points represent the mean ± SD. Abs, absorbance; Cyno, cynomolgus monkey; hB7-1, human B7-1; hPBMC, human peripheral blood mononuclear cell.

Next, the ability of C5H9v2 mAb to enhance T-cell activation was evaluated by an antigenic recall assay. This assay measures IFNγ secretion by CMV-specific T cells derived from human PBMCs in response to CMV peptide–pulsed antigen-presenting cells. Blockade of the PD-1/PD-L1 interaction enhances IFNγ secretion by CMV-specific memory T cells when reexposed to CMV antigens (27). Human PBMCs from CMV-positive donors were stimulated with lysate from CMV-infected cells for 4 days in the presence of increasing concentrations of antibody. An increase of IFNγ production was observed when C5H9v2 mAb was present (EC50 = 0.35 nmol/L) compared with an isotype-matched control antibody, consistent with an enhanced antigen-specific T-cell response (Fig. 1D). Collectively, this data set validates C5H9v2 as a high-affinity neutralizing antibody to PD-L1 suitable for further development as a Pb-Tx.

Engineering of PD-1/PD-L1–targeting Pb-Tx

C5H9v2 IgG and antigen-binding fragment (Fab) constructs were used to identify potential masking peptides from a diverse bacterial peptide-display library, as described previously (28). Peptides conferring the desired masking properties were paired with protease-cleavable linker substrates to create unique prodomains as illustrated in Supplementary Fig. S4. The degree of functional attenuation achieved by a given prodomain can be expressed as a ratio of the antigen-binding EC50 values of the Pb-Tx and the nonmasked parental antibody. A Pb-Tx composed of C5H9v2 with a prodomain conferring the desired degree of masking and protease cleavability was designated CX-072 and selected for further development. This Pb-Tx is currently being studied in a phase I/II clinical trial (NCT03013491).

To further characterize the blockade of PD-1/PD-L1 with Pb-Tx in mouse models, we simultaneously generated muC5H9v2, a surrogate chimeric antibody composed of a mouse IgG2a constant region and the variable domains of the C5H9v2 mAb. Given the differences in observed affinity of C5H9v2 mAb to mouse PD-L1 and hPD-L1, masks for muC5H9v2 were generated such that the binding affinity of the surrogate Pb-Tx construct for mouse PD-L1 (Fig. 2) was comparable with that of CX-072 to hPD-L1 (Supplementary Table S1). In addition, the protease-sensitive substrate linker was optimized to match cleavability by mouse and human proteases. For the purposes of clarity, the Pb-Tx construct engineered for mouse studies was designated CX-630. All mAbs and Pb-Tx used in this report are described in Table 1.

Figure 2.

PD-L1–targeting Pb-Tx exhibits protease-dependent blocking activity in vitro. A, Serial dilutions of muC5H9v2 mAb or Pb-Tx CX-630, either pretreated or not with recombinant human protease (uPA), were incubated with immobilized PD-L1-Fc. The binding affinity of the agents was measured using an anti-mouse HRP-conjugated secondary antibody and is shown in relation to the concentration. Data shown are representative of 3 independent experiments. B, The capacity of CX-630 to block ligand binding was impaired relative to the parental unmasked antibody, as indicated by a competition ELISA assay. Dilutions of muC5H9v2 mAb or CX-630, either pretreated or not with iPA, were mixed with biotinylated PD-1 in a well that was precoated with PD-L1-Fc. The amount of PD-1 bound to PD-L1 was measured using a streptavidin–HRP conjugate and is shown in relation to the amount of blocking antibody or Pb-Tx. Data shown are representative of 3 independent experiments. C, CMV-positive human PBMCs were plated at 250,000 cells/well and stimulated with 5 μg/mL CMV lysate in the presence of CX-072 Pb-Tx, C5H9v2 mAb, or isotype control for 4 days, at which point IFNγ was measured in the supernatants by ELISA. Data shown are representative response from one donor (n = 3 donors tested) of 3 independent experiments. All data points represent the mean ± SD. Abs, absorbance.

Table 1.

List of mAbs and Pb-Tx.

In the absence of proteases, CX-630 Pb-Tx bound to mouse PD-L1 with approximately 120-fold lower affinity compared with muC5H9v2 mAb [equilibrium dissociation constant (KD) of 47.23 vs. 0.40 nmol/L, respectively] and showed a corresponding reduced capacity to inhibit PD-1/PD-L1 interactions (IC50 of 94.9 nmol/L vs. 0.44 nmol/L, respectively; Fig. 2A and B). To evaluate whether affinity for antigen is restored after removal of the mask, CX-630 was activated overnight with the recombinant human protease uPA. In vitro, the binding and blocking properties of uPA-activated CX-630 were indistinguishable from those of the parental muC5H9v2 antibody (Fig. 2A and B). The physiologic significance for attenuated binding to PD-L1 was demonstrated in the CMV recall assay, where the Pb-Tx CX-072 was less potent in inducing IFNγ than the C5H9v2 hIgG4 mAb (Fig. 2C).

To further characterize localized blockade of the PD-1/PD-L1 pathway, Pb-Tx targeting PD-1 were developed using the anti-murine PD-1 J43 antibody (29). The binding affinity for the murine-specific antibody (J43v2) and that for its Pb-Tx derivative (CX-284) to mouse PD-1 were evaluated by ELISA. Although the J43v2 mAb bound with subnanomolar affinity (KD = 0.627 nmol/L), CX-284 exhibited a 20-fold reduction in its ability to bind to mouse PD-1 relative to J43v2 (KD = 11.54 nmol/L; Fig. 3A; ref. 30). Furthermore, the Pb-Tx molecule exhibited an attenuated capacity to block the interaction of PD-1 with PD-L1 (IC50 = 35.04 nmol/L) relative to the parental antibody J43v2 (IC50 = 2.17 nmol/L; Fig. 3B). In vitro activation of CX-284 by recombinant proteases restored full binding and blocking activity (IC50 = 2.77 nmol/L; Fig. 3A and B). To evaluate the impact of masking PD-1 on human cells, a CMV recall assay was performed using an antibody and Pb-Tx targeting the hPD-1 receptor (A1.5 mAb and CX-188 Pb-Tx). The ability of the Pb-Tx to enhance IFNγ production by CMV-positive human PBMCs was attenuated compared with the A1.5 antibody (EC50 = 1.67 nmol/L vs. 0.047 nmol/L; Fig. 3C).

Figure 3.

Pb-Tx targeting the PD-1 pathway exhibits attenuated, protease-dependent binding. A, CX-284 Pb-Tx binds mouse PD-1 with a lower affinity relative to the parental antibody, as indicated by direct ELISA. Serial dilutions of J43v2 mAb or Pb-Tx CX-284, either pretreated or not with recombinant human protease (uPA), were incubated with immobilized mouse PD-1. The binding affinity of the agents was detected using an anti-mouse HRP-conjugated secondary antibody. Data shown are representative of 3 independent experiments. B, The ability of CX-284 to block ligand binding is impaired relative to the parental antibody, as indicated by a competition ELISA assay. Biotinylated PD-L1 was added to wells coated with PD-1-Fc in the presence of serially diluted J43v2 mAb or CX-284 (± uPA preactivation). The amount of bound PD-L1 was measured using a streptavidin–HRP conjugate and is shown in relation to the amount of blocking antibody or Pb-Tx. Data shown are representative of 3 independent experiments. C, CMV-positive human PBMCs were plated at 250,000 cells/well and stimulated with 4 μg/mL CMV lysate in the presence or absence of the human PD-1–targeted antibody A1.5, the human PD-1–targeted Pb-Tx CX-188, or isotype control. After 4 days, the level of IFNγ was measured in the supernatants by ELISA. Data shown are representative response from one donor (n = 3 donors tested) of 3 independent experiments. All data points represent the mean ± SD. Abs, absorbance.

Targeted intratumoral blockade of the PD-1/PD-L1 pathway is an effective therapeutic strategy

Pb-Tx are engineered to be activated by proteases within the tumor microenvironment, allowing the activated Pb-Tx to bind locally while limiting peripheral engagement of the target antigen. Using this approach, we set out to determine whether localized intratumoral blockade of the PD-1/PD-L1 axis was sufficient for therapeutic activity. Single-agent efficacy studies were conducted in the MC38 mouse tumor model. Syngeneic C57BL/6 mice with established subcutaneous MC38 tumors were randomly assigned for treatment with muC5H9v2 mAb, CX-630 Pb-Tx, or mouse IgG2a isotype as a control. As illustrated in Fig. 4A, the PD-L1 Pb-Tx CX-630 elicited comparable therapeutic activity relative to the unmasked parental antibody, achieving maximal tumor growth inhibition of >60% (P ≤ 0.05 for each vs. isotype, one-way analysis of variance).