Biophysical and Immunological Characterization and In Vivo Pharmacokinetics and Toxicology in Nonhuman Primates of the Anti-PD-1 Antibody Pembrolizumab

This article is featured in Highlights of This Issue, p. 1219

The major immune checkpoint receptor programmed cell death 1 (PD-1) is expressed on T cells following their activation, especially those that have been chronically stimulated due to viral infection or those in the tumor microenvironment (1, 2, 3). PD-1 is also expressed on NK cells in chronic infection and malignancy and on subsets of B cells and macrophages. PD-1 is a member of the immunoglobulin superfamily (4), and binds to PD-L1 and PD-L2 (5). Expression of PD-L1 increases following cell exposure to IFNγ. PD-L1 binding to PD-1 on T cells leads to recruitment of SHP-1 phosphatase to the immune synapse (6–9). The interaction of PD-1–associated SHP-1 with CD28 prevents CD28 phosphorylation and abrogates CD28-mediated costimulation. Tumor cells from a range of tissues show PD-L1 expression (10–19). PD-L1 expression in the tumor microenvironment leads to suppression of a T-cell response against the tumor (20).

Here we describe preclinical characterization of pembrolizumab (MK-3475, Keytruda®, SCH 900475), a humanized IgG4 (S228P) antibody that has been approved for treatment of cancer. Pembrolizumab is a potent inhibitor of PD-1 activity, achieved through blockade of both PD-L1 and PD-L2 by binding to a widespread structural epitope on PD-1.

SP2/0 cells and CHO cells expressing PD-1 were tested monthly for mycoplasma using a PCR kit. BC4-49, JY-PD-L1, and JY cells were negative for mycoplasma (last tested FEB-2018). Cell line authentication was not conducted. Peripheral blood mononuclear cells (PBMC) were collected under the Merck Research Laboratories Volunteer Donor program (MVD; Western Institutional Review Board Protocol 20121234) or clinical trial KEYNOTE-001 cohort A and with written informed consent.

Wild-type BALB/c mice (Charles River Laboratories) were immunized with human PD-1 DNA and final immunization with Chinese hamster ovary cells expressing human PD-1 [CHO-hPD1 cells; Merck Sharp & Dohme Corp., a subsidiary of Merck & Co., Inc. (MSD)]. Spleen cells were isolated and subjected to positive and negative panning procedures. Clones secreting functional antibodies were identified using B-cell culture, where B cells were immortalized by electrofusion to SP2/0 (ATCC) and tested in protein- and cell-based binding and ligand blockade assays, and for the ability to modulate IL2 secretion. Medical Research Council Technology (Cambridge, UK) humanized the selected antibody using complementarity-determining region grafting technology, resulting in the antibody known as pembrolizumab.

Binding to human and cynomolgus monkey PD-1 was assessed by solid-phase interferometry and surface plasmon resonance; binding to human PD-1 was also assessed by a kinetic exclusion assay. For the solid phase interferometry format using Octet (FortéBio by Molecular Devices), recombinant human PD-1-Fc protein (MSD) or cynomolgus monkey PD-1-Fc protein (MSD) was coupled to amine-reactive biosensors per the manufacturer instructions. Association was observed by placing the sensor in wells containing 10 to 80 nmol/L antibody and monitoring for 15 minutes. Dissociation was measured after transfer into PBS and monitoring signal for 2 hours. Data sets were fit globally to a pseudo-first-order kinetics association and first-order exponential decay.

For surface plasmon resonance (Biacore; GE Healthcare Life Sciences), pembrolizumab was captured with goat antihuman IgGκ-specific antibody immobilized on a CM5 Series S sensor chip using a Biacore T100 instrument. Human or cynomolgus monkey PD-1-Fc (Sino Biologicals) was injected at various concentrations for 3 minutes, followed by a 2-hour dissociation. The on-rate (k_on) and off-rate (k_off) were empirically determined; the equilibrium dissociation constant (K_D) was determined from the k_off and k_on (K_D = k_off/k_on).

For the kinetic exclusion assay, a KinExA instrument (Sapidyne) was used to evaluate the solution K_D for human PD-1-Fc. Serially diluted PD-1-Fc was incubated with one of two constant concentrations of pembrolizumab, one near and one much higher than the expected K_D. After reaching equilibrium, the amount of free pembrolizumab was measured to calculate the K_D using the KinExA Pro N-curve multicurve analysis software.

Binding to human and cynomolgus monkey PD-1 and ability to block binding of PD-L1 and PD-L2 was assessed using fluorometric microvolume assay technology (FMAT). CHO-hPD-1 (MSD) or CHO-cynomolgus monkey PD-1 (CHO-cynoPD-1; MSD) and PD-L1-Fc or PD-L2-Fc fusion proteins (R&D Systems) were used. Cells were premixed with Alexa Fluor 647-labeled PD-L1-Fc or PD-L2-Fc, then mixed with unlabeled antibody at varying concentrations in FMAT plates. After reaching equilibrium, the plates were read using the FMAT system (Applied Biosystems). Total number of events detected and mean fluorescence per cell were used to calculate the IC₅₀.

Binding to CD28, CTLA-4, and ICOS used an electrochemiluminescence (ECL) immunoassay (see Supplementary Data for details). Pembrolizumab was evaluated for binding to mouse PD-1 using Octet and Biacore as described above.

A gene construct encoding recombinant soluble PD-1 (sPD-1) encompassing amino acids Leu25 through Ser157 of the human PD-1 sequence with a single substitution (Cys93Ser) was expressed in Escherichia coli. Aliquots of guanidine hydrochloride-denatured sPD-1 were diluted into refolding buffer containing L-arginine. After concentration, monomeric sPD-1 species were isolated by size-exclusion chromatography, followed by dialysis, and ion-exchange chromatography. For nuclear magnetic resonance (NMR) experiments, ²H, ¹³C, ¹⁵N isotopic-labeled sPD-1 was produced as described using E. coli grown in isotope-enriched media.

A Fab based on the sequence of nivolumab was expressed in 293-F cells (Life Technologies) after transfection with DNA plasmids encoding the heavy and the light Fab chains at a 1:1 ratio. Harvested supernatant was purified using Protein G (GE Healthcare) chromatography and dialyzed into 50 mmol/L HEPES, 100 mmol/L sodium chloride, pH 7.5.

Pembrolizumab Fab was expressed in E. coli and purified using Protein A chromatography (GE Healthcare). Purified Fab was formulated in PBS, 98% purity by analytical size-exclusion chromatography and >90% purity by reverse-phase chromatography.

NMR studies of PD-1-antibody interactions rely on the assessment of chemical shift changes to ¹⁵N-labeled sPD-1 protein due to unlabeled Fab binding. Standard triple resonance techniques were first used to obtain sequence-specific assignments of ¹³C, ¹⁵N-labeled sPD-1 protein, which were supplemented from published assignments of a related construct. Next, a ¹⁵N heteronuclear single quantum coherence (HSQC) data set was recorded of the ¹⁵N-sPD-1/Fab complex, where only sPD-1 is ¹⁵N labeled. The identification of sPD-1 residues affected by Fab binding is achieved by comparing ¹H and ¹⁵N sPD-1 peak positions in the free and Fab complex data sets. Peak shift comparisons are made from well-resolved regions of the ¹⁵N-HSQC spectrum. Mapping the NMR chemical shift perturbation data to the 3D structure of PD-1 was helpful to localize the interaction site and to set a threshold for affected residues. Data were recorded on a 500 MHz Bruker DRX500 NMR spectrometer with a high-sensitivity 5 mm cryoprobe. Peak shift analyses were made using NMRFAM-SPARKY (21) using an overlay mode to facilitate comparisons.

Heparinized whole blood was obtained under written informed consent from 8 patients with prostate cancer, from 8 patients with stage IV melanoma, and from healthy volunteers; and the staphylococcal enterotoxin B (SEB) stimulation assay was performed as described previously (22).

Heparinized blood was collected from cynomolgus monkeys in the pharmacokinetic study (described below) predose (days −7 and 0) and on days 1, 7, 28, 56, and 84 postdose. Heparinized whole blood was also collected from monkeys in the 1-month toxicology study (described below), twice pretest, 24 hours after dosing on days 0 and 28, and on days 56, 98, 126, and 154 from animals assigned to the recovery period. Blood was diluted 1:10 with RPMI culture medium and preincubated for 30 to 60 minutes with pembrolizumab (at 25 to 0.00025 mg/mL) or medium, after which 1 μg/mL SEB was added and mixtures cultured for 3 days. Supernatant IL2 was measured with an ELISA Kit (eBioscience).

Cryopreserved PBMCs from consenting donors revaccinated <1 year prior with tetanus toxoid (TT; Sanquin Blood Bank Nijmegen) were thawed and plated in DMEM-F12 media with 10% FBS and antibiotics as described previously (22). Cells were preincubated with antibody at concentrations up to 25 μg/mL for 30 to 60 minutes prior to addition of 1 μg/mL TT. After 7 days, supernatants were analyzed for IFNγ by ELISA (eBioscience).

The human CD4⁺, HLA-DR alloantigen-specific T-cell clone BC4-49 was generated from PBMCs of a consenting donor (Stanford Blood Center) after priming with JY cells (gift from Gilbert M. Lenoir) and cloning by limited dilution. JY-PD-L1 clone 6, a high PD-L1 cell surface-expressing puromycin-resistant clone, was generated by transfection of wild-type JY cells with a lentivirus vector pLX-301 encoding human PD-L1 (Millipore Sigma). For the activity assay, BC4-49 T cells were harvested on day 7 after antigen restimulation, washed with PBS, 2 mmol/L EDTA and resuspended in Yssel's medium (Gemini Bioproducts). The BC4-49 T cell suspension was added at 2 × 10⁴ cells in 50 μL per well into wells containing 100 μL of serially titrated antibody. The antibody/T-cell mixture was preincubated for 1 hour. Irradiated JY and JY-PD-L1 cells were washed twice with PBS, 2 mmol/L EDTA, resuspended in Yssel's medium, and filtered. JY or JY-PD-L1 cell suspension (1 × 10⁴ cells per well) was added to the preincubated antibody/T-cell mixtures, using a BC4-49 T cell to JY cell ratio of 2:1. After 3 days, supernatants were analyzed for IFNγ using the Human IFNγ Quantikine ELISA Kit (R&D Systems). See Supplementary Data for additional details.

Fresh non–small cell lung cancer (NSCLC) tumor tissues were collected and shipped overnight in AQIX media (AQIX) by collection providers (Folio Biosciences, Bio-Options, and Boston Biosource). Single cells were dissociated from tumors by fine cutting with a scalpel, followed by a 30-minute incubation at 37°C in DMEM with 100 mg/mL collagenase type I, and 10,000 U/mL DNase I. Digested samples were triturated, filtered through a 70 μm strainer, and washed in DMEM complete medium. A total 0.1 × 10⁶ tumor cells per well were stimulated with 10 ng/mL soluble anti-CD3 antibody (BioLegend, clone OKT3) in the presence of pembrolizumab or isotype control antibody. On day 7 supernatants were assayed for IFNγ using a human IFNγ Kit (Meso Scale Discovery).

PBMC cytokine release assays were conducted with plate-bound antibody using blood from eight healthy human donors (MVD). An in vitro whole blood assay using blood from eight healthy human donors (MVD) was also conducted with pembrolizumab in solution. Controls included known stimulatory antibodies anti-CD3 (OKT3 analog; eBioscience) and anti-CD28 (MSD TGN1412 analog), and trastuzumab. Induction of IFNγ, IL10, IL12 p70, IL1β, IL2, IL6, IL8, and TNFα was assessed by multiplex immunoassays (Meso Scale Discovery). See Supplementary Data for additional details.

Activated human PBMCs were used as effector cells and activated human CD4+ T cells from the same donor (MVD) were used as target cells for the antibody-dependent cell-mediated cytotoxicity (ADCC) assays. The ADCC assays were conducted according to the method used to assess nivolumab (23). Activated CD4⁺ T cells were used as target cells, and human serum complement as effector molecules in the complement-dependent cytotoxicity (CDC) assays. The same activated human CD4⁺ T cells were used as target cells in both the ADCC and CDC assays. Saturation of PD-1 on activated target human CD4⁺ T cells by pembrolizumab was confirmed by flow cytometry. See Supplementary Data for additional details.

The effect of treatment with pembrolizumab on overall T-cell responses to viral infection was assessed ex vivo for clinical trial KEYNOTE-001 cohort A patients with solid tumors using an IFNγ ELISPOT assay (Cellular Technology Ltd.). KEYNOTE-001 was conducted in accordance with recognized ethical guidelines and approved by an institutional review board. PBMCs from 17 patients were collected before and after treatment with pembrolizumab (days 3, 8, 15, and 22; and cycle 2, day 1 preinfusion time points) under written informed consent and stimulated with a pool of peptides of Epstein–Barr virus, cytomegalovirus, and influenza virus restricted to MHC class I molecules (CEF-32; Cellular Technology Ltd.; Panatecs). Samples were analyzed using an IFNγ ELISPOT assay. See Supplementary Data for additional details.

In a single-dose PK study, female cynomolgus monkeys (Macaca fascicularis) received intravenous 0.3, 3, or 30 mg/kg pembrolizumab, and serum concentrations were determined by ELISA. For the toxicokinetic analyses included in the repeat-dose toxicity studies, serum pembrolizumab concentrations were determined using an ECL immunoassay. In both the ELISA and ECL assays, serum pembrolizumab concentrations were determined by ligand capture using human PD-1-Fc and antihuman kappa chain antibody for detection. For anti-pembrolizumab antibody detection (ADA), a bridging ECL method was used with biotinylated pembrolizumab as capture, ruthenylated pembrolizumab as detection, and affinity purified rabbit polyclonal anti-pembrolizumab as a positive control. To assess target engagement, an ex vivo whole blood assay was used to determine potentiation of IL2 secretion in PBMCs after SEB stimulation (described above).

Safety of pembrolizumab was evaluated in 1- and 6-month repeat-dose toxicity studies with a 4-month recovery period in cynomolgus monkeys. In the 1-month study, animals (6/sex/group) received 5 weekly intravenous doses of vehicle or pembrolizumab at 6, 40, or 200 mg/kg for total of five doses. Four animals/sex/group were euthanized during Week 5 whereas the remaining two animals/sex/group were euthanized during Week 23, after a 4-month treatment-free period. In this study, an assessment of target engagement was evaluated using the ex vivo IL2 SEB bioactivity assay (described above). In the 6-month study, animals (5/sex/group) received intravenous doses of vehicle or pembrolizumab at 6, 40, or 200 mg/kg once every other week for a total of 12 doses. Three animals/sex/group were designated for necropsy at the end of the 6-month dosing phase, and the remaining animals underwent necropsy following the 4-month treatment-free period. In both studies, in-life examinations included monitoring of food consumption, body weight, and clinical observations. Clinical pathology parameters (hematology, coagulation, serum chemistry, and urinalysis) were assessed periodically during the dosing phase on all animals. In addition, ophthalmic and electrocardiogram evaluations were performed at the end of dosing periods on all animals. Postmortem examinations included organ weights, and macroscopic and microscopic evaluations of all major tissues.

Procedures involving the care and use of animals were reviewed and approved by the Institutional Animal Care and Use Committee at Research Laboratories of Merck & Co., Inc. and adhere to the US Public Health Service Policy on Humane Care and Use of Laboratory Animals. Toxicology studies were conducted in compliance with Good Laboratory Practice at the Safety Evaluation Center, Schering Plough Research Institute (part of Merck & Co., Inc.), Merck & Co., Inc., and Charles River Laboratories.

Statistical analyses for data presented in Table 1, Fig. 2 all panels, and Fig. 3 panel A were performed using Prism software (GraphPad). Arithmetic means are reported in Table 2; PK analyses were performed using WinNonlin. Analyses for Fig. 3 panel B were performed using NONMEM software.

Following immunization of mice and short-term B-cell culture (24), B-cell clones were identified producing antibodies that bind and block human PD-1. After immortalization and further characterization, clone hPD1.09A was humanized by CDR grafting. Pembrolizumab, the final humanized antibody construct of hPD-1.09A, was generated on a human IgG4/kappa isotype with the stabilizing Adair mutation (25) S228P sequence alteration, where serine 228 (EU index) was converted to proline.

Pembrolizumab bound human PD-1 with an affinity ranging from 1.1 to 130 pmol/L in studies using solid-phase interferometry, surface plasmon resonance, or kinetic exclusion (Table 1). The K_D for binding of pembrolizumab to human PD-1 resulting from the Octet experiment was 29 pmol/L. The Biacore and KinExA studies involved the same 16 batches. Most batches had a K_D value assessed by Biacore of about 50 pmol/L, whereas the K_D values obtained by KinExA range from 0.4 to 1.9 pmol/L.

Pembrolizumab bound to cynomolgus monkey PD-1 with a similar K_D of 118 to 470 pmol/L. Pembrolizumab inhibited the binding of human PD-L1 and PD-L2 to human PD-1 and the binding of cynomolgus monkey PD-L1 and PD-L2 to cynomolgus monkey PD-1 (Table 1). Pembrolizumab did not bind to mouse PD-1, or to immunoglobulin superfamily members human CD28, human CTLA-4, or human ICOS (see Supplementary Fig. S1).

NMR was used to further characterize the PD-1/pembrolizumab Fab interactions (Fig. 1A–D). The hPD-1/hPD-L1 complex x-ray crystal structure (PDB code 4zqk; ref. 26) was used as a point of reference because this is the exact interaction that is being disrupted by antibody binding. The structure identifies residues in the C′CFG β strands of PD-1 that interact with PD-L1. In addition, the CC′ loop undergoes rearrangement upon PD-L1 binding. In all, 21 PD-1 residues are involved at the complex interface that spans 1970 Ų. A flexible C′D loop is observed in the hPD-1/hPD-L1 structure that is disordered in the sPD-1 structure.

The x-ray crystal structure of PD-1/pembrolizumab Fab (PDB code 5ggs; ref. 27) revealed a widespread structural epitope consisting of atoms from PD-1 BC loop residues T59-F63; C strand residues V64, N66, and Y68; C´ strand residues Q75-K78 and A81; C′D loop residues P83 and D85-G90; F strand residue I126; FG loop residues L128-A132; G strand residue I134. The regions highlighted in blue in Fig. 1 include 27 residues and cover 2081 Ų corresponding to 30% of the protein surface.

NMR binding studies were used to detect specific PD-1 residues affected by pembrolizumab Fab binding (Fig. 1) and were mapped to the existing structure (PDB code 5ggs). In these studies, the ¹⁵N-edited HSQC data from ¹⁵N-labeled sPD-1 was compared with data collected from a complex of ¹⁵N-labeled PD-1 bound to unlabeled pembrolizumab Fab. These experiments collected data only from PD-1 residues; changes that occurred upon Fab binding can be identified for each residue. From NMR spectroscopy, residues that experienced the largest changes were: PD-1 BC loop residues N58, S62; C strand residues N66 and Y68, C´ strand residues D77, K78, and L79; C′D loop residues F82, D85, R86, Q88, G90, Q91, D92, F95; FG loop residues A129, A132, Q133; G strand residues I134 and S137. Other regions that were affected included D strand residue V97; F strand residues L122, G124, and I126 that were likely affected by the disruption of the FG loop. The regions highlighted in green (Fig. 1) include 24 residues that cover 1337 Ų, corresponding to 19% of the protein surface. Rearrangement of the C′D loop and F strand residues L122-C123-G124 due to PD-L1 binding to sPD-1 was previously noted (28).

NMR studies were also used to determine the specific PD-1 residues affected by binding of a nivolumab Fab (Fig. 1E–G). In these studies, the ¹⁵N-edited HSQC data from ¹⁵N-labeled sPD-1 was compared with data collected from a complex of ¹⁵N-labeled PD-1 bound to unlabeled nivolumab Fab. Relatively few peaks were shifted in the complex spectra consistent with a focused epitope. The 13 residues that were most affected were N loop residues S27 and D29, BC loop residues N58 and S60 and FG loop residues A129, A132, and Q133. Other affected residues were not localized in loops; they included T36, S54, N64, H107, and Y121. When mapped to the PDB code 5wt9 crystal structure, the N-terminal BC and FG loops were spatially clustered together and, as a result, can be identified as the Fab binding location. The regions cover 711 Ų, corresponding to 9% of the protein surface.

Single residue resolution of the NMR HSQC data allowed for the differentiation of nivolumab from pembrolizumab binding to PD-1 even without knowledge of sequence-specific assignments. Pembrolizumab and nivolumab demonstrated distinct binding interaction sites on PD-1 (Fig. 1G).

Pembrolizumab enhanced SEB-induced production of IL2 (29) by 3- to 4-fold over SEB alone in an ex vivo assay on PBMCs from healthy human donors (22) as well as from patients with advanced metastatic melanoma (22) or prostate cancer (22), indicating that blood cells derived from healthy donors and patients with cancer contained T cells sensitive to modulation by pembrolizumab. Pembrolizumab activity was also confirmed in SEB-stimulated PBMC cultures from cynomolgus monkeys with similar enhancement of IL2 production (22).

To assess the influence of pembrolizumab on antigen-specific memory T-cell responses, a human T-cell recall assay using TT was conducted using PBMCs from volunteers who had received a TT booster vaccination within the previous year (22). Production of IFNγ was enhanced by PD-1 blockade using pembrolizumab in a dose-dependent manner (Fig. 2A), evidence that PD-1 blockade relieved PD-1/PD-L1/PD-L2 immune suppression and resulted in enhanced antigen-specific T cell IFNγ production.

Pembrolizumab ability to enhance IFNγ production in a T-cell clone co-culture assay was also assessed. T-cell clone BC4-49 expresses PD-1 and is susceptible to checkpoint inhibition, including inhibition of proliferation and IFNγ production when binding either PD-L1 or PD-L2 in the context of allo-antigen recognition. Using a JY transfected clone expressing high levels of PD-L1, coculture of T-cell clone BC4-49 with JY-PD-L1 cells resulted in low levels of IFNγ production. This was reversed by pembrolizumab in a dose-dependent manner (Fig. 2B).

The potential of pembrolizumab to induce IFNγ production in human primary tumors was evaluated in cultures of anti-CD3-stimulated primary tumor cell digests (30). In these experiments, NSCLC tumor specimens were treated with pembrolizumab or an isotype control antibody. A dose-dependent increase in IFNγ secretion with pembrolizumab treatment was observed (Fig. 2C).

In both the PBMC cytokine release assay with plate-bound antibodies and in solution in the whole blood assay, cytokine levels induced by pembrolizumab were low and comparable to those induced by trastuzumab, a therapeutic antibody with a safe clinical profile for cytokine release (see Supplementary Tables S1 and S2). PD-1 was saturated by pembrolizumab at 0.1 μg/mL on activated target human CD4⁺ T cells. PD-1 expression was comparable to the levels observed on the activated target human CD4⁺ T cells used in the CDC and ADCC assays. A concentration of pembrolizumab exceeding that needed for saturation of PD-1 was used. Pembrolizumab ADCC and CDC induction was comparable to the negative control, human IgG4 anti-PCSK9 antibody and mouse IgG2a isotype control antibody, whereas an anti-MHC class I positive control antibody induced robust cell killing (Supplementary Figs. S2 and S3).

The effect of treatment with pembrolizumab on overall T-cell response to viral infection was assessed ex vivo in the clinical trial KEYNOTE-001 on patients with solid tumors using a viral recall assay. PBMCs from 17 patients isolated prior to and after up to 4 weeks of treatment with pembrolizumab were stimulated with a pool of peptides of Epstein–Barr virus, cytomegalovirus, and influenza virus restricted to MHC class I molecules and assessed for IFNγ expression. There was no significant change in IFNγ production after pembrolizumab treatment using Wilcoxon-signed rank test (paired difference test), indicating no compromise of overall virus specific T-cell–mediated immune response (Fig. 2D).

After administering single intravenous doses of pembrolizumab at 0.3, 3, or 30 mg/kg to cynomolgus monkeys, a slightly more than dose-proportional exposure between 0.3 and 3.0 mg/kg was observed, and approximately dose-proportional exposure was observed between 3.0 and 30 mg/kg (Table 2). Clearance and terminal half-life (t_1/2) appeared to be dose-dependent, with the higher dose groups exhibiting slower clearance and longer t_1/2, suggesting a possible involvement of target mediated clearance at the low dose. See Supplementary Table S3 for more detail.

Pembrolizumab was well-tolerated in cynomolgus monkeys with a systemic exposure (AUC_{0-7 days}) of up to approximately 170,000 μg·day/mL with weekly dosing over the course of a 1-month study, and with a systemic exposure (AUC_{0-14 days}) of up to approximately 67,500 μg·day/mL with biweekly dosing over the course of a 6-month study (Tables 2 and 3).

In both the 1-month and the 6-month studies, no adverse findings were observed in any antemortem monitoring. At terminal evaluations (end of dosing and end of treatment-free period necropsy), there were no pembrolizumab-related organ weight, macroscopic, or histopathologic findings. Most histomorphologic changes observed in a few scattered organs/tissues generally consisted of small, focal, mononuclear cellular infiltration, composed primarily of lymphocytes and/or histiocytes, without parenchymal organ tissue disruption or degeneration, similar in character and magnitude to spontaneous changes commonly observed in cynomolgus monkeys (31, 32). On the basis of these results, the no-observed-adverse-effect-level (NOAEL) in both monkey studies was determined as the highest tested dose of 200 mg/kg.

In both the PK and 1-month toxicity study, monkeys' biological response to pembrolizumab was confirmed using ex vivo potentiation of IL2 secretion in peripheral blood cells following stimulation with SEB (Fig. 3A). Prior to dosing, ex vivo addition of pembrolizumab and SEB to blood potentiated IL2 production 2- to 3-fold. However, the potentiation of IL2 production was not observed in blood collected from pembrolizumab-dosed animals, indicating prior PD-1 engagement by pembrolizumab. Engagement was observed across all dose levels in both the PK and toxicology study immediately following dose administration on both the first and last day of dosing, and returned to baseline following antibody clearance in a dose-dependent manner (Fig. 3B). During the 4-month postdose period in the 1 month toxicology study, in the 6 and 40 mg/kg dose groups target engagement activity returned to prestudy levels around 1 and 2 months, respectively, whereas in the 200 mg/kg group engagement was still evident at 4 months, indicating full PD-1 engagement for the entire duration of the toxicology study.

Pembrolizumab is a humanized IgG4 (S228P) anti-PD-1 antibody that binds to human and cynomolgus monkey PD-1 with high affinity. Although there is no international standard for determining affinity, three different methods show that pembrolizumab binds human PD-1 with picomolar affinity. In addition, pembrolizumab potently blocks both PD-1 ligands PD-L1 and PD-L2 from binding to PD-1.

The binding interaction sites between PD-1 and either PD-L1 or PD-L2 are known (26–28, 33–37). Structural studies have shown that the pembrolizumab binding interaction site on PD-1 overlaps with the PD-L1 binding site (27, 36–38) and involves the C′D and FG loops of PD-1. Structural, biochemical, and biophysical methods have been used to better understand the interactions of PD-1 with pembrolizumab and another anti-PD-1 therapeutic, nivolumab (36, 38–40). The pembrolizumab binding interaction site with PD-1 is distinct from that of nivolumab (36, 38–40). High-resolution x-ray crystal structures and solution NMR studies reveal that pembrolizumab binding affects many of the same PD-1 residues that directly interact with PD-L1 and PD-L2, directly blocking the binding of the endogenous ligands (41). In contrast, PD-1-nivolumab interactions primarily affect flexible PD-1 N-terminal residues (39, 41). Deletion of the PD-1 N-terminus abolished nivolumab binding whereas pembrolizumab binding was unaffected.

Lepir and colleagues (38) hypothesized that the affinity strength of pembrolizumab is mainly due to the HCDR3 loop, which binds both the C′D and FG loops of PD-1. Na and colleagues (37) structurally superimposed the human PD-1/pembrolizumab Fab complex and their model of the human PD-1/human PD-L2 complex and concluded that the binding interaction sites overlap in regions similar to those observed for pembrolizumab and PD-L1.

The ability of pembrolizumab to block binding of both PD-L1 and PD-L2 to PD-1 is an important feature of its clinical mechanism of action. The signaling functions of PD-1/PD-L1 and PD-1/PD-L2 overlap partially but not completely (42–44). The importance of being able to block not just PD-L1 but also PD-L2 is highlighted by the fact that tumors and their microenvironments may express either PD-L1, PD-L2, or both (44, 45). A preclinical mouse study (43) suggests that although PD-1/PD-L1 suppression is the primary mechanism for cancer immune evasion, PD-L2 upregulation may be a compensatory mechanism once PD-L1 function is nullified.

Pembrolizumab demonstrated functional activity by reversing PD-1 immune-mediated suppression in human T-cell SEB stimulation, CD4⁺/HLA-DR alloantigen-specific T-cell clone co-culture, and primary tumor cell histoculture assays by inducing or enhancing production of IL2 and IFNγ, a marker of tumor-infiltrating lymphocytes (30). Although pembrolizumab allows stimulated T cells to become active, it does so in an antigen-specific manner as evidenced by the CD4⁺/HLA-DR alloantigen-specific T-cell clone, TT recall response, and viral antigen response results.

The pembrolizumab IgG4 framework heavy chain is constructed using an S228P mutation, which prevents Fab arm exchange with endogenous IgG4 molecules (46). Consistent with its IgG4 framework, pembrolizumab does not mediate effector functions, as assessed by ADCC and CDC. Nor does pembrolizumab directly cause cytokine release, as demonstrated using human PBMC and whole blood assays.

Pembrolizumab was well-tolerated up to 200 mg/kg either once weekly for 1 month or once every 2 weeks for 6 months in cynomolgus monkeys, where dose dependent and significant systemic exposure was demonstrated. Most histopathology observations in these studies were limited to incidental histomorphologic changes observed in scattered organs/tissues in individual monkeys that were similar in character to spontaneously occurring background findings in cynomolgus monkeys (31, 32). Although potential exacerbation of spontaneous tissue changes by the treatment with pembrolizumab cannot be ruled out, the findings were nonadverse to the study animals. The safety profile of pembrolizumab in monkeys appears to be similar to nivolumab although the study designs for the two antibodies were different in dose level, dose frequency, and duration of treatment. The nivolumab toxicology study (23) in cynomolgus monkeys with twice weekly intravenous administration of 10 or 50 mg/kg over a 3-month period (27 total doses) demonstrated the highest well-tolerated dose was 50 mg/kg. The only toxicology-related finding was a reversible decrease (by 28%) in serum triiodothyronine (T3) hormone in females treated with 50 mg/kg at week 13, whereas thyroxine (T4) and thyroid-stimulating hormone (TSH) levels were unchanged. The significance of this finding was not discussed and no other findings from the nivolumab toxicology study were reported (23).

Pembrolizumab has demonstrated efficacy in multiple cancers and is currently approved for treatment of more than 15 different indications (47). Clinical studies continue to examine the effectiveness of pembrolizumab alone or in combination with other agents in multiple cancers.

All authors are current or former employees of Merck Sharp & Dohme Corp., a subsidiary of Merck & Co., Inc., and may own stock/stock options in Merck & Co., Inc. J. Dulos has a patent W02008156712A1, relevant to work. H. van Eenennaam has stock from Aduro, Inc., outside the submitted work. In addition, H. van Eenennaam has a patent WO/2008/156712 issued, relevant to work. Gregory J. Carven is a current employee of Scholar Rock and is a named inventor on patents and patent applications related to pembrolizumab.

Conception and design: L. Fayadat-Dilman, M. Hsieh, S. Li, S. Sadekova, M.A. McCoy, D. Herzyk, F.M. Poulet, J. Dulos, C.L. Andrews, R. de Waal Malefyt, J. Lee, H. van Eenennaam, A. Willingham, Y. Yu, M. Streuli, G.J. Carven, A. van Elsas

Development of methodology: L. Fayadat-Dilman, M. Hsieh, S. Li, M.A. McCoy, F.M. Poulet, J. Dulos, C.L. Andrews, G. Ayanoglu, C. Cullen, R. de Waal Malefyt, J. Lee, S. Sadekova, H. van Eenennaam, A. Willingham, Y. Yu, M. Streuli, G.J. Carven, A. van Elsas

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): C. Cullen, M.A. McCoy, F.M. Poulet, J. Dulos, S.P. Kang, G. Ayanoglu, J. Lee, S. Li, D. Malashock, G. Soder, Y. Yu, G.J. Carven

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): D. Herzyk, S.P. Kang, R.A Kastelein, S. Sadekova, A. van Elsas, B. Hutchins, G.C. Starling, M.A. McCoy, F.M. Poulet, J. Dulos, L. Liu, L. Fayadat-Dilman, C.L. Andrews, C. Cullen, R. de Waal Malefyt, S. Le Saux, S. Li, D. Malashock, H. van Eenennaam, A. Willingham, Y. Yu, M. Streuli, G.J. Carven

Writing, review, and/or revision of the manuscript: B. Hutchins, G.C. Starling, M.A. McCoy, D. Herzyk, F.M. Poulet, J. Dulos, L. Liu, S.P. Kang, L. Fayadat-Dilman, M. Hsieh, C.L. Andrews, G. Ayanoglu, C. Cullen, R. de Waal Malefyt, R.A. Kastelein, S. Le Saux, J. Lee, S. Li, D. Malashock, S. Sadekova, G. Soder, H. van Eenennaam, A. Willingham, Y. Yu, M. Streuli, G.J. Carven, A. van Elsas

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J. Dulos

Study supervision: J. Dulos, C. Cullen, R.A. Kastelein, S. Sadekova, H. van Eenennaam, M. Streuli, A. van Elsas

Other (data verification): B. Hutchins

Other (drafting of the first draft): B. Hutchins, G.C. Starling, M.A. McCoy, D. Herzyk, L. Liu, R. de Waal Malefyt

This study was funded by Merck Sharp & Dohme Corp., a subsidiary of Merck & Co., Inc. The authors wish to acknowledge the following individuals (all of Merck Sharp & Dohme Corp., a subsidiary of Merck & Co., Inc., Kenilworth, NJ USA): Karen Kolz, Yifan Song, Dong-Hun Lee for collection of data; Corey Strickland for structural biology input; Jayanthi Wolf for supervision of research and input on study design and data interpretation; Lauren Nardini for collection of data and bioanalytical development, validation and sample analysis; Shree Bhagwat, David Sanden, and Shawn Foti for technical help and collection of data; Michael Judo for collection of data for specificity; Saraswathi Naravula and Kasia Marullo for technical help and collection of data; Wiebe Olijve and Ton Adang for supervision of research; Amy Levonas for technical help, writing assistance, and administrative or logistic support; Sheila Erespe for submission assistance; Narendra Kishnani for PK/TK analysis; Cathelijne Kloks for collection of data, supervision of research, and development and validation of pharmacokinetic and immunogenicity assays for (pre)clinical studies of pembrolizumab.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.