Myocarditis in Cynomolgus Monkeys Following Treatment with Immune Checkpoint Inhibitors

CTLA4 (CTL-associated protein 4) and PD1 (programmed cell death 1)/programmed death ligand 1 (PDL1) are two clinically proven immune checkpoints for cancer immunotherapy. CTLA4 blocking antibody ipilimumab was the first approved (in 2011) checkpoint inhibitor for the treatment of metastatic melanoma. Several blocking antibodies targeting PD1 (nivolumab and pembrolizumab) or PDL1 (atezolizumab, durvalumab, and avelumab) have been approved subsequently for a variety of cancer indications. Coinhibitory receptors CTLA4 and PD1 are important for immunologic homeostasis, which helps maintain peripheral tolerance to self-antigens. They serve as immune checkpoints that effector T cells must pass to maintain/sustain their full functionality. Disruption of these immune checkpoints can result in breaking of immune tolerance and immune-related adverse events in patients.

CTLA4 is a cell surface coinhibitor molecule that is expressed nearly exclusively on CD4⁺ and CD8⁺ T cells and is critical for the function of regulatory T cells (Tregs). CTLA4 inhibits both early and late T-cell responses through multiple mechanisms. PD1 plays an important role in shaping the initial magnitude of the T-cell response, in fine tuning T-cell differentiation and effector T-cell fate and in the development of immunologic memory (1–3). PD1 has two natural ligands, PDL1 and PDL2 (4, 5). Mice genetically deficient in PD1 develop accelerated autoimmunity (6–9).

Clinical use of immune checkpoint inhibitors (ICI) is associated with a series of specific adverse events caused by the induced activation of immune response, collectively known as immune-related adverse events (irAE). CTLA4 inhibitors and PD1/PDL1 inhibitors have been associated with similar irAE profile, although the toxicities observed with CTLA4 inhibitors are generally more severe (10). These irAEs have been reported to affect many organs/tissues, such as skin, gastrointestinal tract, endocrine system, lung, liver, pancreas, kidney, neurologic system, kidney, joints, hematologic system, the skeletal muscle, and heart (11, 12). Although ICI-related cardiac toxicity is rare, it can be life-threatening; therefore, it warrants increased awareness and investigation to allow insights into potential pathogenesis, diagnostic strategy, and choice of drug combinations (13–19).

No significant irAEs as seen in humans are reported in laboratory animal species, including nonhuman primates, when PD1, PDL1, and CTLA4 inhibitors are administered alone or in combination. Here we report a monkey model in which we observed widespread inflammation in multiple organs including heart, liver, kidney, large intestine, adrenal medulla, and salivary gland following the administration of ipilimumab and nivolumab. Strikingly, myocarditis was found in monkeys with morphology, cardiac biomarker changes, and immune cell infiltrates similar to human ICI-related myocarditis. Blood immunophenotyping, IHC, and tissue transcriptomic analysis suggested increased T-cell activation and migration, increased phagocytosis, and antigen presentation in heart. These data suggest the organ toxicities in this monkey model are primarily mediated by T-cell immunity, consistent with the pharmacology of ipilimumab and nivolumab.

A 4-week study was conducted in female Chinese cynomolgus monkeys to evaluate the toxicity of coadministered ipilimumab and nivolumab. The study was conducted in an AAALAC accredited facility (Covance, Madison, WI), in compliance with the Animal Welfare Act, the Guide for the Care and Use of Laboratory Animals, and the Office of Laboratory Animal Welfare. The procedures used in this study have been reviewed and approved by the Institutional Animal Care and Use Committee. Two control monkeys (Group 1, P0001-0002) were dosed intravenously with saline once weekly (days 1, 8, 15, and 22) for a total of 4 doses. Five monkeys (group 2, P0101-0105) were dosed intravenously with clinical batches of nivolumab 20 mg/kg plus ipilimumab 15 mg/kg once weekly (days 1, 8, 15, and 22). Nivolumab and ipilimumab were purchased from Blue Door Pharma. The animals were evaluated for changes in clinical signs daily and body weights were recorded weekly. All animals were humanely euthanized by sodium pentobarbital anesthesia followed by exsanguination on day 29.

Serum samples were collected 5 minutes, 6, 24, 48, 96, and 168 hours following the first and last dose administrations and 5 minutes and 168 hours after the second and third dose administrations. Serum concentrations of ipilimumab and nivolumab were determined using a multiplexed ligand-binding assay. Briefly, biotinylated recombinant human CTLA4 is coupled to U-PLEX linker 1, and biotinylated recombinant human PD1 is coupled to U-PLEX linker 10. The U-PLEX linkers then self-assemble onto unique spots on a U-PLEX plate. Ipilimumab is captured by the recombinant human CTLA4, and nivolumab is captured by the recombinant human PD1. Bound ipilimumab and nivolumab are detected with a ruthenylated goat anti-human IgG. Plates are read on the MSD SECTOR Imager 6000, and final detection is conducted by using the ruthenium-labeled goat anti-human IgG antibody and Tripropylamine (TPA) to produce an electrochemiluminescent signal within the MSD instrument that is representative of the amount of bound ipilimumab and nivolumab. Sample concentrations are determined by interpolation from a standard curve that is fit using a 4-parameter logistic equation (note: weighing formula for standard curve is 1/y⁁2). The standard points in 100% cynomolgus monkey serum range from 2,560 ng/mL to 10.0 ng/mL, and the range of quantitation in 100% serum is 1,280 ng/mL to 40.0 ng/mL. Five samples ipilimumab and nivolumab at 1,280, 640, 160, 80, and 40 ng/mL serve as quality control samples. Toxicokinetics following the first and fourth doses (day 1 and day 22) were calculated using noncompartmental analysis using Phoenix WinNonLin v7.0.

A full necropsy was conducted on all monkeys. Microscopic examination of formalin-fixed, paraffin-embedded, hematoxylin and eosin–stained sections was performed on a standard comprehensive list of tissues. Tissues were examined by a board-certified veterinary pathologist and findings were recorded on a subjective scale as follows: minimal, an inconspicuous change; mild, a noticeable but not prominent change; moderate, a prominent change; marked, a dominant but not overwhelming change; and severe, an overwhelming change. A second board-certified veterinary pathologist reviewed the findings and the results reported herein represent the consensus opinion of the two pathologists.

For IHC, unstained tissue sections were deparaffinized in xylene and rehydrated with graded alcohols to distilled water. All tissue sections were pretreated for antigen retrieval by heating slides in either Borg Decloaker (Tris-based formulation, pH 9.5—Biocare Medical) or AR Citra solution pH6 (BioGenex) using a Retriever 2000 pressure cooker (Electron Microscopy Sciences). Endogenous peroxidase activity was inactivated with Peroxidazed 1 (Biocare Medical) for 10 minutes. Nonspecific protein interactions were blocked for 10 minutes with Background Punisher (Biocare Medical). The primary anti-CD3 (clone SP162, Spring Bioscience, 1:200), anti-CD4 (clone EP204, CellMarque, 1:100), anti-CD8α (clone D8A8Y, Cell Signaling Technology, 1:50), anti-CD68 (clone D4B9C, Cell Signaling Technology, 1:1,000), and anti-PD-L1 (clone SP142, Abcam, 1:50) antibodies were applied for 1 hour at room temperature and detected using MACH2 Rabbit HRP polymer (Biocare Medical) for 30 minutes at room temperature. Anti-FoxP3 (clone 236A/E7, Biocare Medical, Predilute), anti-PD-1 (clone NAT105, Abcam, 1:75), and anti-CD20 (clone L26, Biocare Medical, 1:200) were applied for 1 hour at room temperature and detected using MACH2 Mouse HRP polymer (Biocare Medical) for 30 minutes at room temperature. All immunoreactivity was developed using Betazoid DAB (Biocare Medical) for 5 minutes at room temperature. Immunostained sections were briefly counterstained with Tacha's Auto Hematoxylin (Biocare Medical), washed in tap water, dehydrated in graded alcohols, cleared in xylene and coverslipped with Permount mounting medium (Fisher Chemical). Isotype control rabbit or mouse IgGs were run on matching tissue sections using similar protocols.

All image analysis steps were carried out in Visiopharm software (Visiopharm). The quantification was performed in the ventricles, as there was non-specific DAB signal present throughout the atria and the surrounding heart tissue. For all the markers except PDL1, a custom algorithm was developed for each marker to detect positive and negative cells in each animal. Specifically, the algorithm made use of size, shape, and intensity of the DAB reaction product to detect infiltrating immune cells. Then for each marker, percent positive cells was calculated using the formula: 100 × (# of positive cells)/(# of positive cells + # of negative cells). For PDL1, the immunoreactivity of the anti-PDL1 primary antibody was such that it resulted in a patchy, noncontinuous membrane labeling pattern. Subsequently, it was not feasible to implement an automated algorithm to detect PDL1-positive cells. Therefore, we used a different endpoint, that is, DAB stain area percentage, to quantify the relative abundance of PDL1 in the IHC images. Specifically, for each animal, the algorithm calculated the total area occupied by the DAB reaction product and the total area occupied by hematoxylin stain in the ventricles. The DAB stain area percentage was then calculated by using the formula: 100 × (total area of DAB reaction product)/(total area of DAB reaction product + total area of hematoxylin stain). All graphs were plotted in GraphPad Prism (GraphPad Software).

For blood immunophenotyping, hematology data was collected to provide total lymphocyte counts. Approximately 1 mL of K2EDTA anticoagulated blood were collected predose on day 1 (considered baseline), and on days 8, 15, 22, and 29. An aliquot of each blood sample was used to determine lymphocyte counts, and the remainder was used for immunophenotyping analysis. For tissue immunophenotyping, a spleen sample and one mesenteric lymph node from each animal were collected at necropsy (day 29) and the spleen sample was weighed. Single-cell suspensions were prepared for spleen and lymph node samples. Approximately 100 μL blood or approximately 10⁶ spleen or lymph node cells in 100 μL suspension were aliquoted into each tube containing cocktails of antibodies with different fluorescent labels. After incubation on ice for 30 minutes, for blood and spleen cells, the red blood cells were lysed with 1–2 mL of FACSlysing solution supplied by Becton Dickinson. Samples not stained for Ki67 or FoxP3 were washed two times, reconstituted in 0.3 mL of Stain Buffer (BD Biosciences), and kept refrigerated until flow cytometry analysis. Samples stained for intracellular FoxP3 and Ki67 were treated with 0.5 mL FoxP3 Buffers A and C (BD Biosciences), incubated on ice for 20 minutes, washed two times with BD Perm Wash and then incubated with the Ki-67 and FoxP3 intracellular staining cocktail for approximately 30 minutes at room temperature. Samples were washed with BD Perm Wash and reconstituted in 0.3 mL of Stain Buffer (BD Biosciences) and kept refrigerated until flow cytometry analysis. The percent of each lymphocyte subset out of total lymphocytes from blood, as well as spleen and lymph node suspensions was determined by flow cytometry. These percentages from blood lymphocyte subsets were used to determine the absolute counts for each lymphocyte subset based on the absolute lymphocyte counts determined from hematology data (reported as cells/μL). The spleen cells were counted and the number of cells per spleen was calculated on the basis of the weight of the whole spleen and the weight of the partial spleen used for immunophenotyping.

Hematology and clinical chemistry parameters were evaluated using a Siemens Advia 2120 hematology and Advia 1800 chemistry analyzers, respectively. Blood samples were collected without anticoagulant during acclimation (baseline) and then at the following time points: day 1 hours 0 (predosing), 0.05, 6, 24, 48, 96, and 168; day 8 hours 0 (predosing), 0.05, and 168; day 15 hours 0 (predosing), 0.05, and 168; day 22 hours 0 (predosing), 0.05, 6, 24, 48, 96, and 168, and day 29. Serum samples were then submitted for measurement of cytokines (IL2, IL4, IL6, IL10, IFNγ, and TNFα), c-reactive protein, cardiac troponin I (cTnI), and n-terminal pro-b natriuretic peptide (NT-proBNP). cTnI results acquired from timepoints within 24 hours following dosing restraint were not interpreted due to likelihood of restraint stress (20). Cytokines were simultaneously quantified in cynomolgous monkey serum samples using the Millipore Milliplex MAP Non-Human Primate Cytokine Magnetic Bead Panel reagent kit (EMD Millipore Corporation) according to manufacturer's instructions. C-reactive protein was assayed with Siemens CardioPhase high-sensitivity c-reactive protein (HSCRP) reagents using a Siemens Advia 1800 chemistry analyzer following the manufacturer's instructions (Siemens Healthcare Diagnostics). cTnI was assayed using the Siemens Troponin I-Ultra assay using the Siemens Advia Centaur XP immunoassay system (Siemens Healthcare Diagnostics) according to manufacturer's instructions. NT-pro-BNP was assayed using the Biomedica BNP fragment EIA kit (Biomedica).

Two 10-μm curls, from formalin-fixed paraffin embedded heart tissue blocks from five control female cynomolgus monkeys dosed intravenously with saline or five female cynomolgus monkeys coadministered nivolumab and ipilimumab as described above were used to isolate total RNA. Three of the five control group animals were from a separate study, using the same vehicle and dosing regimen and schedule, and at similar body weights and ages. Total RNA was extracted from each sample using the Qiagen RNeasy FFPE kit per protocol with the single alteration of overnight incubation in Buffer PKD supplemented with 20 mmol/L 2-amino 5-methylphenyl phosphonic acid at 55°C for 18 hours (21). The extracted RNA samples were quantified using a Nanodrop 8000 instrument and qualified with the Agilent TapeStation RNA ScreenTape assay. One microgram of total RNA from each sample was then used as input into the Illumina TruSeq stranded RNA library prep kit using RiboZero for rRNA depletion prior to cDNA library generation. Perkin Elmer's SciClone automated liquid handling system was used to fabricate sequencing libraries per manufacturer protocol. The prepared Illumina RNASeq libraries were individually quantified by Qubit using the DNA broad range assay kit and qualified by Agilent TapeStation using the DNA 1000 ScreenTape assay kit. All libraries passing quality control, defined as the presence of a smear of double stranded DNA between 200 and 400 base pairs (bp) on the TapeStation DNA 1000 trace, were pooled at a concentration of 30 nmol/L in 50 μL, packaged and shipped out to GeneWiz, Inc. for sequencing on a HiSeq instrument (Illumina) using standard Illumina paired end 2 × 150 bp sequencing chemistry.

FastQ files were quality controlled and analyzed using version 11 of CLC Genomics Workbench (Qiagen). Read depth for samples averaged 103 million paired-end reads, with mean Pfred scores ranging from 38.8 to 39.2 ensuring base call accuracy greater than 99.9%. Short sequence reads were assembled using the RNASeq analysis workflow, and were mapped and annotated using the Macaca_fascicularis_5.0 reference genome as a template (https://www.ncbi.nlm.nih.gov/genome/annotation_euk/Macaca_fascicularis/101/). For mapping purposes, only alignments with a length fraction of 0.8 and a similarity fraction of 0.8 were considered, allowing for two mismatches and three insertions and deletions per read.

The resulting read count matrix with rows of genes, columns of samples, and intersection of raw read counts were analyzed for differential expression using DESeq2 (22). The five treated samples (ipilimumab and nivolumab) were compared with five vehicle control samples using DESeq methodology including the estimation of size factors, estimation of dispersion, negative binomial GLM fitting, Wald statistics, and Benjamini–Hochberg correction to arrive at log₂ fold change values, P values, and adjusted P values (P_adj). A significance threshold was established on the basis of an adjusted P_adj of 0.05.

Orthologs from cynomolgus monkey (Macaca fascicularis) to human were established using Ensembl 93 and Ensembl's ortholog tables, extracted via “biomaRt” (23, 24). The CLC Genomics Workbench and the NCBI Macaca_fascicularis_5.0 reference genome resulted in 27,616 mapped gene entities corresponding to 27,230 unique Entrez Gene IDs. Where possible, ensembl gene ids were mapped to entrez gene ids and the ortholog mappings resulted in 16,784 cynomologous monkey to human unique matches.

The raw read count matrix was converted to a matrix of transcript per million (TPM) values and subsequently log₁₀ transformed. The full matrix of log₁₀-TPM values were subsetted on the basis of the identified significant genes, based on the DESeq2 analysis. Using the R “pheatmap” package, the log₁₀-TPM hierarchically clustered heatmap was generated using Pearson correlation, Euclidean distance, and complete agglomeration method. Additional annotation was added including a column-wise sidebar for treated or vehicle and a row-wise sidebar to identify genes with human orthologs.

A total of 82 myocarditis genes were identified using the Ingenuity Knowledge Base (SI reference library). Of the 82 genes, 63 were successfully mapped to human entrez gene ids and their cynomolgus monkey orthologs. Using only the human ortholog–mapped DESeq2-analyzed dataset (n = 16,784 genes), the subset of 63 genes were evaluated for enrichment using a hypergeometric test. The complete set of myocarditis genes and their respective log₂ fold change values are displayed as a lollipop plot with the significant genes (P_adj ≤ 0.05) identified by red color.

Animals administered ipilimumab and nivolumab displayed occasional episodes of reduced food consumption and mild fecal alterations (nonformed and/or liquid feces). Low food consumption was most prevalent on the day after the first and second dose administration. The fecal alterations were seen as early as day 8, continued intermittently through day 24, and were associated with mononuclear cell infiltration in the gastrointestinal mucosa. Pharmacologically relevant plasma exposure of ipilimumab and nivolumab was detected over the course of the study (Supplementary Table S1; Supplementary Fig. S1).

Inflammation, characterized by minimal to marked mononuclear cell infiltration, was present in numerous organs/tissues in animals administered ipilimumab/nivolumab (summarized in Table 1), as compared with the minimal mononuclear infiltrates observed in occasional tissues in vehicle control animals. The tissues with the higher severity infiltrates (moderate or marked) in dosed monkeys were the heart, kidney, large intestine, salivary gland, and vagina. The mononuclear infiltration (Fig. 1) was composed primarily of lymphocytes and macrophages, often occurred with a perivascular orientation, and was accompanied by minimal to mild parenchymal degeneration/necrosis in the heart, kidney, and liver. Consistent with the widespread inflammation, on days 8 and/or 29 there were mild increases in circulating white blood cells, lymphocytes, monocytes, globulin and CRP, and mild decreases in albumin. Increases in circulating markers of specific organ damage (e.g., transaminases, blood urea nitrogen), exclusive of heart (discussed below), were not present. The distribution and severity of inflammatory cell infiltrates, accompanied by tissue damage in some organs and circulating indicators of inflammation (increased white blood cell counts, CRP, fibrinogen, and globulin), demonstrated development of immune-related findings in monkeys dosed with the combination of ipilimumab/nivolumab.

Because ICI-induced myocarditis is a life-threatening, emerging adverse event in patients, we further characterized the heart findings. All five ipilimumab/nivolumab–dosed monkeys had mononuclear cell infiltration within the heart as did both control monkeys (Supplementary Fig. S2). However, the severity in both controls was minimal, while in ipilimumab/nivolumab–dosed monkeys it was minimal in one, mild in two, and moderate in two (Supplementary Fig. S2). Furthermore, cardiomyocyte degeneration was present in two animals (P0102 and P0103) administered ipilimumab/nivolumab and associated with increased cTnI (up to 2.67× baseline). In addition, one monkey (P0104) with moderate mononuclear cell infiltration had increased NT pro-BNP (up to 5.06× baseline; Table 2).

Immunohistochemically, myocardial infiltrates contained a mixture of cells including T lymphocytes (CD3⁺), macrophages (CD68⁺), and B cells (CD20⁺; Fig. 2). T Lymphocytes were the most prominent cell type, with CD4⁺ T cells tending to be more prominent than CD8⁺ T cells. Regulatory T cells (FoxP3⁺) composed a minor portion of the lymphocyte infiltrate. Infiltrating cells were also immunohistochemically positive for PD1 and PDL1 (Fig. 2). Rare faint cardiomyocyte immunoreactivity was present for PDL1. Digital analysis of IHC images in the ventricle areas from the 2 control monkeys and 5 ipilimumab/nivolumab–dosed monkeys showed substantially greater numbers of CD4⁺ and CD8⁺ T cells, Tregs, B cells, and macrophages in the ipilimumab/nivolumab–dosed animal heart tissues in comparison with the control animal heart tissues, although large variations were present between individual animals (Supplementary Fig. S3).

To monitor the immunopharmacology of ipilimumab and nivolumab in the dosed monkeys, we conducted immunophenotyping of lymphocytes in the blood, spleen, and mesenteric lymph nodes. Following intravenous administration of ipilimumab/nivolumab, there were increases in the absolute numbers of both CD4⁺ and CD8⁺ T cells (Fig. 3A and B). These correlated with increased proliferation of T cells, shown as increased percentages of CD4⁺ and CD8⁺ T cells with a proliferation phenotype (Ki67⁺; Fig. 3C and D). Similarly, the regulatory T cells (Tregs) were also increased (Fig. 3H). The increases in CD4⁺ T cells were more substantial than the increases in CD8⁺ T cells and were observed at all timepoints (in at least 2 animals per time point), and were more pronounced on days 15 (5 animals, 2.67×–3.86× baseline) and 22 (4 animals, 2.25×–4.15×). There was also a trend of increases in the percentages of activated (CD69⁺) CD4⁺ and CD8⁺ T cells in the blood (Fig. 3E and F); and a trend of increases in the percentages of central memory CD8⁺ T cells (Tcm), defined as CD28⁺CD95⁺ CD8⁺ T cells (Fig. 3G).

The increases in total T cells, CD4⁺ T cells, CD8⁺ T cells, and Tregs were also found in the spleen (Fig. 3I). Similar to blood, these increases were attributable to T-cell proliferation (Ki67⁺). The increases in splenic T cells correlated with microscopic findings of increased lymphocytes in the red pulp and increased size of the follicles and marginal zones, and contributed to increased splenic weights (1.75–1.96 times higher mean absolute and relative splenic weights). In the mesenteric lymph nodes, there were increases in the percentages of NK cells, CD4⁺ central memory T cells, and CD8⁺ central memory and effector memory T cells (Fig. 3I). These changes correlated with microscopic findings of mildly increased paracortical lymphocytes. Increased paracortical lymphocytes were also present in the inguinal and mandibular lymph nodes of animals administered nivolumab and ipilimumab.

Sporadic minimal to moderate increases in IL4, IL6, IL10, IFNγ, TNFα in three monkeys (P0101-0103) were consistent with increased activation and proliferation of T cells in these monkeys (Fig. 3J–L).

To explore the immune mechanisms of ICI-induced myocarditis in our monkey model, RNA-sequencing (RNAseq) analysis was performed. Heart tissues from monkeys from this study and from three other control monkeys from a separate study using the same vehicle and dosing regimen were sequenced to an average read depth of 103 million paired-end reads; which, when aligned to the NCBI Macaca_fascicularis_5.0 genome assembly, resulted in an average of 27% exonic reads, 28% intronic reads, and 48% intergenic reads across samples. All samples passed quality control analysis.

A total of 1,114 genes (of 27,230) were determined to be differentially expressed in the hearts of monkeys receiving coadministration of ipilimumab/nivolumab using a statistical threshold of P_adj ≤ 0.05 (Supplementary Table S2). A representative heatmap of these differentially expressed genes was generated using gene-wise median log₁₀-TPM values subtracted from each sample such that each row was zero centered on the median of the control responses (Fig. 4A). A majority of differentially expressed genes were elevated (807) in the heart of treated monkeys, while a smaller number was reduced (307). Animals P0103 and P0104 showed the most significant increases in these 807 upregulated gene clusters, which correlated with the highest grade of inflammatory cell infiltrates in the heart. For pathway analysis purposes, we then identified human orthologs of the differentially expressed monkey genes. Of the 1,114 monkey genes in this dataset, 891 were matched to human genes (Fig. 4A; Supplementary Table S2).

This data set of 891 differentially expressed genes was further analyzed using Ingenuity Pathway Analysis (IPA) software (Qiagen Redwood City, www.qiagen.com/ingenuity). A log₁₀ P value cutoff of 4.9 was used to define the top 50 canonical pathways enriched in this dataset (Supplementary Table S3). The top ten pathways are associated with immune response, are highly statistically significant (−log₁₀ P > 8) and are predicted to be activated (z-score > 2; ref. 25). Strikingly, each of the top 4 enriched pathways are involved in Th cell differentiation (Th1 and Th2 pathways), and costimulation of proliferation and cytokine production (iCOS-iCOSL signaling in T helper cells and CD28 signaling in T helper cells).

Perturbed biological networks associated with the differential gene expression response to ipilimumab/nivolumab combination treatment were also identified using IPA software. We identified 25 networks in which, at minimum, 20 dysregulated genes are represented within this dataset (Supplementary Table S4). Many of the highest scoring enriched networks are associated with cell development and immune response. Of note, networks identified include the Infectious Diseases, Cell-To-Cell Signaling and Interaction, Hematological System Development and Function (Fig. 4B) and Lymphoid Tissue Structure and Development, Tissue Morphology, Antigen Presentation (Fig. 4C) networks, which support the canonical pathways discussed previously.

In addition to the pathway and network analysis, the IPA Regulator Effects Tool (25) was employed. This tool aims to predict activation or deactivation of upstream regulators based upon the magnitude and significance of the differential gene expression and information collected in the Ingenuity Knowledge Base (extracted from published data). In addition, upstream regulator effects and empirical expression data are used to predict downstream biological processes or diseases that occur using Ingenuity Knowledge Base. In identifying regulator effects, a custom analysis was performed. Specifically, only upstream regulators with a −log₁₀ P > 10, and a z-score > |3| were considered. In addition, the maximum number of regulators considered was set to 4. For downstream disease and functions the same statistical thresholds were applied, yet the functions per network were limited to 1. This analysis yielded 25 regulator effects with consistency scores greater than 7 (Supplementary Table S5), all of which are associated with immune cell function (e.g., recruitment/migration, chemotaxis, homeostasis, interaction). The top two scoring regulator effects are associated with lymphocyte migration (Fig. 4D) and phagocytosis of cells (Fig. 4E).

To evaluate whether gene expression changes in the heart of ipilimumab/nivolumab dosed monkeys are associated with clinical myocarditis or preclinical disease models, we queried the Ingenuity Knowledge Base. A total of 82 genes were identified on the basis of the published literature, of which 63 were detected within our dataset (Supplementary Table S6). The fold change in expression and P_adj value for these genes are represented in Fig. 4F. In total, 22 of 63 (35%) genes reported to be dysregulated during myocarditis were found to be differentially expressed (enrichment P = 5.4e⁻¹⁴). These include chemokine receptors CCR2, CCR5, and CXCR3, MHC molecules HLA-DRA, HLA-DMB, HLA-DPB1, T-cell activation, and Th1 differentiation–associated genes or markers PRKCQ (PKC theta), TNFSF8 (CD153), CD40 ligand, TBX21 (encoding t-bet), STAT4, which responds to Th1-driving cytokine IL12, and CD8 effector molecules PRF1 (encoding perforin), and TNF. This observation suggests that the molecular changes are in high concordance with previously published data.

ICI-related myocarditis has drawn significant attention recently due to its high fatality rate, the absence of good understanding of its pathogenesis, and a lack of early diagnosis and effective monitoring capabilities. Its importance is also emphasized by a recent ICI-associated myocarditis workshop that drew multidisciplinary experts from academia, industry, and regulatory agencies (17). During the workshop, the lack of animal models of nonclinical myocarditis was recognized.

CTLA4 inhibitor and PD1/PDL1 inhibitors share the same common irAEs, with slightly higher frequency and severity of adverse events seen in CTLA4 inhibitor–treated patients (12). The combination of these ICIs generally results in increased risk and severity of irAEs (26), including myocarditis (27). In fact, emerging evidence suggests that combination ICI therapy is the main risk factor for ICI-mediated myocarditis (16). Therefore, we used a ipilimumab/nivolumab combination to induce irAEs in monkeys. Four weekly doses of ipilimumab at 15 mg/kg/dose and nivolumab at 20 mg/kg/dose resulted in significant systemic inflammation and multiorgan irAEs, including myocarditis.

In this study, doses higher than the clinical ones were used to increase the chance of eliciting irAEs. The exposures of ipilimumab and nivolumab were significantly higher (AUC: approximately 3.7× and 8.5×, respectively) than the exposures with the clinical ipilimumab/nivolumab combination regimen of 3 mg/kg and 1 mg/kg (refs. 28, 29; Supplementary Table S1). All dosed monkeys had increased proliferation of lymphocytes and mononuclear cell infiltration in multiple organs, which is consistent with the high percentages of severe irAEs (68.7% of ≥ grade 3) in patients receiving ipilimumab/nivolumab combination (26).

In patients, histopathologic diagnosis of myocarditis requires the presence of inflammatory cell infiltrates in association with myocyte degeneration/necrosis, without evidence of an ischemic event (30). In our study, 2 of 5 ipilimumab/nivolumab dosed monkeys had inflammatory infiltrates combined with cardiomyocyte degeneration/necrosis (without evidence of an ischemic event), thus meeting the human histopathologic diagnostic criteria for myocarditis. These two monkeys had minimally increased cTnI, a serum marker indicative of cardiomyocyte necrosis, which is frequently increased in clinical cases of myocarditis. In addition, a third ipilimumab/nivolumab–dosed monkey had a moderate mononuclear infiltrate with increased NT pro-BNP, a biomarker change also seen in patients with ICI-related myocarditis (18). Thus, at least 3 of 5 monkeys developed ipilimumab and nivolumab–related myocarditis. Morphologically, the myocarditis present in ipilimumab/nivolumab dosed cynomolgus monkeys was similar to ICI-induced myocarditis in patients (14, 15, 31–33). In ICI-induced myocarditis case reports, heart findings often also include fibrosis. The lack of fibrosis in our monkey study is likely related to the relatively short duration of the study. In patients, the infiltrates have been characterized as containing CD3⁺, CD4⁺, and CD8⁺ T lymphocytes, CD68⁺ macrophages, and FoxP3⁺ T regulatory cells. Infiltrating cells have been reported to express PD1 and PDL1, and cardiomyocytes have been reported to express PDL1. Expression of all these markers was present in the inflamed areas of ipilimumab/nivolumab–dosed monkey hearts.

The high incidence of heart findings in our monkeys compared with patients may result from several different factors. Monkeys may be more susceptible to immune-related inflammation in the heart, as suggested by the observation that inflammatory cell infiltration, mostly of low grade, is common in the heart tissues of cynomolgus monkeys (31, 32, 34, 35). Indeed, these low-grade infiltrates were present in both control monkeys in our study, and it is possible that blocking of the immune checkpoints CTLA4 and PD1 exacerbated the immune cell infiltration into the heart. The higher doses and exposures in our monkeys compared with those of patients may have contributed to the higher incidence of heart findings. Heart tissue sampling differences may have also been a factor, because a much larger fraction of heart area was evaluated in the monkeys than is typically examined on human endocardial biopsies.

Previously, Selby and colleagues reported gastrointestinal inflammation in cynomolgus monkeys coadministered four weekly doses of ipilimumab and nivolumab (36). Three of 5 animals in the high-dose group (10 mg/kg ipilimumab plus 50 mg/kg nivolumab) had gastrointestinal pathology findings. No irAEs in nonlymphoid tissues or organs were reported, and neither proliferation nor activation of lymphocytes was noted. In our model, the dose of ipilimumab is higher but the dose of nivolumab is lower than what were used in Selby and colleagues' study. This suggests that the inhibition of CTLA4 checkpoint mechanisms by ipilimumab at a 10 mg/kg dose in monkeys may be incomplete, and an increase in dose to 15 mg/kg may provide sufficient exposure and pharmacology. This may also suggest that CTLA4 inhibition contributes more than PD1 inhibition in the development of irAEs including myocarditis in this monkey model, which is consistent with the higher frequency and greater severity of irAEs observed in patients receiving ipilimumab monotherapy than those receiving nivolumab monotherapy (37). Unlike Selby and colleagues' study, our model was characterized by substantial lymphoproliferation accompanied by T-cell activation and widespread infiltration of mononuclear cells into various tissues (Figs. 1 and 3; Table 1); this distinction potentially explains the multiorgan irAEs, including myocarditis, observed in our model. Infiltration of T cells and macrophages into heart tissues was evidenced by IHC analysis. Unlike other laboratory animal species, nonhuman primates used in research come from diverse backgrounds, which may partly explain differences seen in lymphoproliferation and widespread inflammatory cell infiltration in our study.

It seems probable that both PD1 inhibition and CTLA4 inhibition contribute to the development of multiorgan irAEs and myocarditis in monkeys, because single pathway inhibition rarely results in any irAEs, myocarditis, or other organ toxicities. This combinational effect can potentially be explained by the nonoverlapping mechanisms of PD1 and CTLA4 pathways. CTLA4 harnesses multiple mechanisms to control T-cell activation and enforce peripheral tolerance. CTLA4 deficiency in mice caused rapid development of lymphoproliferative disease with multiorgan lymphocytic infiltration and tissue destruction, with severe myocarditis and pancreatitis and death within 4 weeks of birth (38, 39). PD1 signaling restrains autoreactive T cells that enter the heart, maintaining them in an anergic state (40). PD1- or PDL1-deficient autoimmune-prone MRL mice developed lymphocytic myocarditis with massive infiltration of CD4⁺ and CD8⁺ T cells (6, 41). The role of PD1 in maintaining immune tolerance in the heart was also demonstrated in myocarditis mouse models. Mice receiving PD1-deficient CD8⁺ T cells developed myocarditis, and PD1-deficient mice developed enhanced myocarditis in a CD4⁺ T-cell–mediated experimental autoimmune myocarditis model (42). Therefore, it is plausible that combining an ICI with an oncology drug with cardiotoxic potential can result in additive or synergistic myocarditis risk and a careful consideration may be needed in susceptible patients.

The systemic hyperproliferation of T cells observed in our monkey model (Fig. 3) and in CTLA4^−/− mouse models (38, 39) suggests polyclonal T-cell expansion and subsequent infiltration into various tissues. The infiltrating cells, of course, may include self-reactive T-cell clones that directly cause tissue damage. Trafficking of T cells to tissues requires the expression of chemokine receptors on T cells and the expression of the corresponding chemokines in tissues. One of the interesting findings from gene expression analysis is that the expression of multiple chemokine receptors was increased in the ipilimumab and nivolumab–dosed monkey heart tissues, including the CXCR3–CXCL9/CXCL10 and CCR5/CCL5 chemokine axes molecules (Supplementary Table S2). The increased expression of CXCR3, CCR2, and CCR5 were also found to be strongly associated with myocarditis through Ingenuity Knowledge Base analysis (Fig. 4F). The increased expression of CXCR3 chemokine axis was also reported previously in the ICI-associated myocarditis heart tissues by others (15, 43). Interestingly, CXCR3–CXCL9/CXCL10 and CCR5/CCL5 chemokine axes have also been shown to be particularly important in T-cell homing to tumors (44).

The presence of activated T cells was supported by the increased CD69⁺ T-cell populations (Fig. 3) and IPA analysis that suggested Th1 and Th2 activation, iCOS–ICOSL signaling, T-cell receptor signaling, and CD28 signaling (Supplementary Table S3). There was evidence supporting increased CD8⁺ T-cell activities, such as increased expression of key CD8⁺ T-cell functional mediators including Granzyme K, A, and B, Fas ligand, and perforin (Supplementary Table S2). IPA analysis identified Th1 pathway being the most significantly upregulated pathway by ipilimumab/nivolumab combination (Supplementary Table S3), and the hallmark Th1 transcription factor t-bet (ref. 45; encoded by TBX21) gene expression was upregulated and identified as one of the genes associated with myocarditis in our model (Fig. 4F). It is highly likely that there was constant antigen presentation and activation of T cells in the heart tissues from the ICI-treated monkeys. Transcriptomic data suggest that 38% of genes (14/38) associated with the antigen presentation pathway are upregulated (Supplementary Table S3) in treated monkeys. In addition, the regulatory network analysis suggested increased phagocytosis and activation of APC (Fig. 4B and E).

Currently, there are neither standard diagnostic criteria nor consistent biomarker monitoring guidelines for ICI-related myocarditis (17). Increases in circulating biomarkers cTnI and/or T and BNP or NT-proBNP are considered to be among the most useful noninvasive diagnostic tests and have additionally been evaluated as promising prediction risk factors (13, 16). In this study, increases in NT-proBNP or cTnI were observed at multiple time points as early as 15 days after the first dose in the two animals (P0102 and P0103) with myocardial degeneration and a third animal (P0104) with moderate mononuclear cell infiltration (Table 2), providing additional support that cTnI and NT-proBNP may serve as valuable biomarkers for ICI-related myocarditis. Interestingly, there were small increases in some cytokines for animals P0102 and P0103.

In conclusion, we developed a monkey model of ICI-induced multiorgan irAEs including myocarditis. Investigative studies revealed immune-mediated mechanisms suggestive of polyclonal expansion of T cells and potential evidence of antigen-specific cellular immune responses. This monkey model can be used to further elucidate the pathogenesis of myocarditis and various irAEs. In addition, it can potentially be used to benchmark immunomodulators and to evaluate combination therapies before entering clinical studies.

F. Barletta, A.T. Hooper, and P. Sapra hold ownership interest (including patents) in Pfizer Inc. No potential conflicts of interest were disclosed by the other authors.

Conception and design: C. Ji, A. Vitsky, A.T. Hooper, P. Sapra, N.K. Khan, M. Finkelstein, M. Guffroy, B.S. Buetow

Development of methodology: C. Ji, M.D. Roy, F. Barletta, B.S. Buetow

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): C. Ji, J. Golas, S.W. Kumpf, F. Barletta, W. Meier

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C. Ji, M.D. Roy, A. Vitsky, S. Ram, M. Martin, F. Barletta, W. Meier, N.K. Khan, M. Guffroy, B.S. Buetow

Writing, review, and/or revision of the manuscript: C. Ji, M.D. Roy, J. Golas, A. Vitsky, S. Ram, M. Martin, F. Barletta, W. Meier, A.T. Hooper, N.K. Khan, M. Finkelstein, M. Guffroy, B.S. Buetow

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M.D. Roy, J. Golas, M. Martin, B.S. Buetow

Study supervision: C. Ji, P. Sapra

We would like to thank Ellen Evans for her critical review of this manuscript and the following other Pfizer colleagues for their technical support: Daniel Weaver (RNA sequencing); Carol Donovan, Amy Shen (immunophenotyping); Thomas Cummings (in vivo study monitoring); and Alyson McGuinty, Melba Dokmonovich, Sandra Summers (biomarker data); Alyssa Myers and Beth Leary (pharmacokinetic analysis). We also want to acknowledge the following individuals from Covance Laboratories for their support of the in vivo study: Jacqueline Miller and Niraj Tripathi.

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.