A Genetic Mouse Model Recapitulates Immune Checkpoint Inhibitor–Associated Myocarditis and Supports a Mechanism-Based Therapeutic Intervention

Immune checkpoint inhibitors (ICI), including combination therapies such as ipilimumab plus nivolumab, have shown remarkable efficacy across multiple tumor types (1–5) but are associated with immune-related adverse events (irAE) including myocarditis (6–10). ICI-associated myocarditis is characterized by myocardial T-cell and macrophage infiltration and significant ECG disturbances, with preliminary data suggesting a female predominance, clinically distinguishing it from other forms of myocarditis (7, 11, 12). ICI-associated myocarditis is also often fulminant and fatal, despite corticosteroid therapy. Therefore, mechanism-based treatment strategies are needed to mitigate serious irAEs such as myocarditis. One major obstacle to improving our understanding of irAEs is the lack of preclinical animal models that recapitulate the clinical course. Herein, we report a murine model that phenotypically recapitulates the clinical pathology of ICI-associated myocarditis and provides mechanistic support for the inhibition of T-cell costimulation by CTLA4–Ig as a treatment for ICI-associated myocarditis.

To better define the risk factors for the development of ICI-associated myocarditis, we interrogated VigiBase, the World Health Organization's (WHO) global international pharmacovigilance database of individual case safety reports from January 2008 through June 2019. In comparing myocarditis rates in patients who received monotherapy versus combination ICI therapy, myocarditis was reported at higher frequency for patients treated with combination therapy compared with CTLA4 or PD-1/PD-L1 therapy alone [84/6,858 (1.22%) vs. 329/61,192 (0.54%), reporting odds ratio, 2.29; 95% confidence interval, 1.79–2.91, with n/N being the number of myocarditis case reports over the total number of case reports]. Therefore, consistent with previous analyses, the major risk factor associated with development of myocarditis is the combination of CTLA4 and PD-1 inhibition (7, 13).

We assessed whether ICI-associated myocarditis could be recapitulated by pharmacologic treatment of a mouse model of autoimmunity. MRL-Fas^lpr mice spontaneously develop lupus-like autoimmunity, and loss of PD-L1 in this model leads to autoimmune myocarditis, albeit with a disease pathology driven by antibody deposition that is not observed in patients with ICI-associated myocarditis (14, 15). MRL-Fas^lpr mice were treated with vehicle, anti-CTLA4 or anti–PD-1 monotherapy, or combination therapy twice weekly for 8 weeks. Treated mice did not display obvious clinical signs; however, histologic and electron microscopic examination revealed subtle immune infiltration of the myocardium and vasculature, as well as signs of endothelial cell damage and sarcomere disarray following combination therapy (Supplementary Fig. S1A–S1F). These findings not only are consistent with increased risk of myocarditis due to combination ICI therapy, but also underscore the absence of preclinical models that recapitulate the disease course of myocarditis associated with ICI therapy, which has been described elsewhere and is notably mechanistically distinct from myocarditis due to other causes (7, 13).

CTLA4 and PD-1 attenuate T-cell activation, but do so through distinct cellular and molecular mechanisms (16–18). Thus, the observed increase in myocarditis rates in combination therapy could either be due to additive effects of two distinct pathologies resulting from manipulation of these biological pathways, or through functional interaction between CTLA4 and PD-1 which exacerbates the development of this irAE. To distinguish these possibilities, we crossed transgenic mouse strains harboring loss-of-function alleles of Ctla4 (encoding CTLA4) and Pdcd1 (encoding PD-1) to understand how compound loss of CTLA4 and PD-1 would affect T-cell function and the manifestation of autoimmune pathologies (see Methods; refs. 19, 20). Consistent with prior findings, mice with homozygous loss of Ctla4 succumbed to fatal lymphoproliferative disease with complete penetrance (Fig. 1A; refs. 19, 21, 22). Approximately 50% of Ctla4^+/− Pdcd1^−/− mice died by 3 months of age, whereas littermate controls, including Ctla4^+/+ Pdcd1^−/− mice, remained healthy. This dramatic effect of single copy loss of Ctla4 in the absence of PD-1 is surprising, given that no haploinsufficiency phenotypes have been previously reported in Ctla4^± mice with homozygous wild-type Pdcd1 alleles at either the organismal or cellular level (23), suggesting that negative regulation by PD-1 is sufficient to buffer against perturbations in T-cell signaling caused by heterozygous loss of Ctla4. To confirm the mortality observed in Ctla4^+/− Pdcd1^−/− mice, we used a second breeding scheme that yields only Ctla4^+/− Pdcd1^−/− and Ctla4^+/+ Pdcd1^−/− littermate mice (Supplementary Fig. S2A; see Methods), which yielded similar findings with mortality of approximately 50% of Ctla4^+/− Pdcd1^−/− mice (Fig. 1B). The incomplete penetrance of mortality raised the possibility that subtle genetic heterogeneity, environmental factors (e.g., microbiome), or stochastic processes influence disease progression. Genetic heterogeneity (e.g., segregating MHC alleles) could not explain the incomplete penetrance of mortality in Ctla4^+/− Pdcd1^−/− mice, as all tested mice were 97% to 100% C57BL/6J and no associations between any segregating alleles and clinical signs were observed (Supplementary Fig. S2B). Similar findings of premature death in Ctla4^+/− Pdcd1^−/− mice were observed at a second institution, suggesting that environmental factors are not a primary driver of the phenotype (Supplementary Fig. S2C). The frequency of mortality was partially sex-dependent, with female mice dying at higher frequency than male mice, consistent with the characteristics of ICI-associated myocarditis in patients (refs. 11, 12; Fig. 1C). Mortality of Ctla4^+/− Pdcd1^−/− mice was often preceded by overt, nonspecific clinical signs (e.g., reduced weight gain, reduced movement, dyspnea) beginning as early as 1 month of age. Consistent with Ctla4 haploinsufficiency in the absence of PD-1, assessment of CTLA4 protein expression revealed reduced expression levels of total CTLA4 protein from stimulated T cells derived from lymph nodes from Ctla4^+/− Pdcd1^−/− mice (Supplementary Fig. S2D–S2F) and from those isolated from the thymus (Supplementary Fig. S3A–S3C).

This indicates that single copy loss of Ctla4 in the absence of PD-1 results in lowered CTLA4 levels and raises the possibility that the early mortality of Ctla4^+/− Pdcd1^−/− mice arises due to reduced regulatory capacity of CTLA4 and defects in immunologic tolerance. Because Ctla4-null mice die due to systemic lymphoproliferative disease and Pdcd1-knockout mice develop mild autoimmunity in specific strains of mice, we explored the possibility that Ctla4^+/− Pdcd1^−/− mice develop systemic autoimmunity. No differences in morphology of lymphoid tissues between genotypes were detected in Ctla4^+/− Pdcd1^−/− mice (Supplementary Fig. S4A–S4F). Furthermore, in contrast to Ctla4-null mice, Ctla4^+/− Pdcd1^−/− mice do not die from uncontrolled lymphoproliferation or a major defect in central tolerance. There was no evidence of systemic organ immune infiltration or elevations in levels of serum cytokines or antibodies (Supplementary Fig. S5A and S5B). This indicates that Ctla4^+/− Pdcd1^−/− mice do not die due to cytokine storm or systemic autoimmunity. However, Ctla4^+/− Pdcd1^−/− mice were hypoglycemic and had elevations in serum troponin, a marker of cardiac damage, raising the possibility that cardiac involvement underlies the premature death of Ctla4^+/− Pdcd1^−/− mice (Supplementary Fig. S5C). In a small subset of mice where matched flow cytometry data were also available, elevated troponin was associated with increased cardiac CD45⁺ immune cell infiltration (Supplementary Fig. S5D). Comprehensive histologic analysis revealed prominent lymphocytic infiltration in the heart and pancreas of Ctla4^+/− Pdcd1^−/− mice (Supplementary Table S1). Interestingly, histopathologic analysis of other tissues, such as skeletal muscle (Supplementary Fig. S6A), colon (Supplementary Fig. S6B), and lung (Supplementary Fig. S6C), showed minor or no increased immune infiltration in Ctla4^+/− Pdcd1^−/− mice compared with Ctla4^+/+ Pdcd1^−/− mice (Supplementary Fig. S6D–S6F), which did not generally result in overt tissue death and was assessed as unlikely to be pathophysiologically relevant. These infiltrations were substantially less pronounced in comparison with the pathologies observed in the heart and pancreas. Together, these data suggested that Ctla4^+/− Pdcd1^−/− mice die due to a breach in peripheral tolerance manifesting in organ-specific autoimmunity.

Histologic examination of the heart tissue indicated dense cellular infiltrates with associated cardiomyocyte necrosis, consistent with myocarditis, in Ctla4^+/− Pdcd1^−/− mice. Although myocarditis was generally widespread among atrial and ventricular sites, the infiltrates were mostly epicardial and endocardial, consistent with fulminant myocarditis observed in ICI-treated patients (7). Lymphocytic infiltration into the myocardial tissue and myocyte destruction demonstrated patterns similar to those observed in human patients with ICI-induced myocarditis (Fig. 2A). Many of the affected Ctla4^+/− Pdcd1^−/− mice displayed evidence of end-stage cardiomyopathy including cardiomegaly, presence of atrial thrombus (Supplementary Fig. S7A), as well as liver congestion. Other cardiovascular pathologic observations included aortic dissection, vasculitis, pulmonary vasculitis, and pericarditis (Supplementary Fig. S7A). Interestingly, the other site of dramatic lymphocytic inflammation occurred in the exocrine pancreas, with sparing of the islet cells (Supplementary Fig. S7B). The intensity of the inflammation in Ctla4^+/− Pdcd1^−/− mice varied, consistent with the incomplete penetrance of mortality. These pathologic abnormalities were not observed in Ctla4^+/+ Pdcd1^−/− mice, underscoring the critical importance of Ctla4 in disease pathogenesis. More minor inflammation was observed in other organs of both Ctla4^+/− Pdcd1^−/− and Ctla4^+/+ Pdcd1^−/− mice, consistent with autoimmunity observed upon loss of PD-1 (20, 24, 25). Lymphocytic infiltration consisted of CD3⁺ CD8⁺ T cells, and CD4⁺ T cells to a lesser abundance in Ctla4^+/− Pdcd1^−/− mice, mirroring observations in patients (ref. 7; Fig. 2B and C). In agreement with phenotypic observations, histopathology scores of cardiac hematoxylin and eosin (H&E) sections in Ctla4^+/− Pdcd1^−/− mice demonstrated increased prevalence of myocarditis with high CD3⁺ lymphocytic infiltrate, high F4/80⁺ macrophage infiltrate, and low FOXP3⁺ regulatory T-cell abundance (Fig. 2D). Increased PD-L1 expression was also observed in cardiac tissue of Ctla4^+/− Pdcd1^−/− mice similar to prior clinical observations (Supplementary Fig. S7C and S7D; ref. 7). We further characterized cellular infiltrates in female 3- to 6-week-old Ctla4^+/− Pdcd1^−/− and Ctla4^+/+ Pdcd1^−/− mice by flow cytometry. Increased frequencies of total CD45⁺ immune cells, CD8⁺ T cells, and CD68⁺ macrophages were observed in Ctla4^+/− Pdcd1^−/− mice compared with littermate Ctla4^+/+ Pdcd1^−/− mice (Fig. 2E; Supplementary Figs. S8 and S9A). In contrast, no difference was observed in matched splenic populations (Supplementary Fig. S9B). Mice were highly heterogeneous in levels of infiltrate, consistent with the heterogeneity and incomplete penetrance observed at the phenotypic level. Given these observations, we then tested whether the disease pathogenesis of Ctla4^+/− Pdcd1^−/− mice is immune-mediated. Premature mortality appeared to be attenuated in Ctla4^+/− Pdcd1^−/− Rag1^−/− mice compared with littermate Ctla4^+/− Pdcd1^−/− Rag1 competent mice (Fig. 2F; Supplementary Fig. S9C). Although this comparison did not meet statistical significance (potentially due to the impact of Rag1^−/− immunodeficiency independent of myocarditis, evidenced by premature deaths observed in Ctla4^+/− Pdcd1^+/− Rag1^−/− mice), it is consistent with a role for adaptive immunity in disease pathogenesis.

To further characterize clinical cardiac disease, we performed extensive cardiovascular and physiologic examination in all Ctla4^+/− Pdcd1^−/− and littermate Ctla4^+/+ Pdcd1^−/− mice. Considerable heterogeneity was observed in echocardiographic parameters of Ctla4^+/− Pdcd1^−/− mice compared with Ctla4^+/+ Pdcd1^−/− mice such that no statistically significant differences were observed. However, stratification of mice based on immune infiltrates revealed an association between increased immune infiltration and cardiomegaly as manifested by increased heart-weight to body-weight ratio and ventricular wall thickening on echocardiography without systolic cardiac dysfunction (Fig. 3A and B). Detailed ECG examination revealed significant arrhythmogenic disturbances, including sinus node dysfunction, sinus arrest, and atrioventricular conduction block, and severe bradycardia in Ctla4^+/− Pdcd1^−/− mice, but not Ctla4^+/+ Pdcd1^−/− mice (Fig. 3C–E), consistent with the arrhythmogenic nature of ICI myocarditis in patients (ref. 7; Fig. 3F). The pattern of myocarditis and ECG instability observed in these mice is consistent with clinical observations from patients with ICI-associated myocarditis (7, 26, 27).

The development of a robust mouse model of ICI-associated myocarditis enables the testing of therapeutic regimens. Our finding that the development of myocarditis is gene dosage–dependent raises the hypothesis that restoration of CTLA4 and/or PD-1 signaling would be sufficient to prevent disease progression. To test this hypothesis, 3-week-old Ctla4^+/− Pdcd1^−/− mice were treated with vehicle or abatacept (recombinant CTLA4–Ig) to block T-cell costimulation by binding to B7 ligands, prior to the development of any overt clinical signs. Abatacept treatment significantly reduced mortality of Ctla4^+/− Pdcd1^−/− mice (Fig. 4A). At a cellular level, abatacept did not decrease myocardial immune infiltrates at 2 weeks; however, after 14 weeks, immune infiltrates in the heart returned to near-baseline levels (Fig. 4B). In an effort to identify biomarkers associated with disease progression or response, we assessed heart weight–body weight ratios. No statistically significant difference was observed between genotypes; however, decreased cardiomegaly was observed following abatacept treatment (Fig. 4C). These observations provide preclinical rationale for further investigation of CTLA4–Ig as a treatment for autoimmune irAEs associated with ICI therapy. Clinically, we have previously published a case report of a corticosteroid-refractory ICI myocarditis successfully treated with abatacept (28) and provide clinical details on two additional similarly responding patients with ICI-associated myocarditis in which abatacept led to amelioration of the clinical features of myocarditis (Supplementary Table S2 and Supplementary Fig. S10). Thus, preliminary evidence exists that response to CTLA4–Ig in this model seems to replicate that observed in clinical patients, though more investigation is clearly needed. Nonetheless, these data further support the mechanistic basis for the ability of CTLA4–Ig intervention to attenuate autoimmunity caused by either anti–PD-1 monotherapy or combination anti-CTLA4 plus anti–PD-1 treatment.

Here, we identify gene dosage–dependent genetic and functional interaction between the T cell–negative costimulatory genes Ctla4 and Pdcd1, which manifests as fatal myocarditis in Ctla4^+/− Pdcd1^−/− mice, providing a robust genetic mouse model for myocarditis and other cardiovascular toxicities associated with CTLA4 and PD-1 ICI therapy. This model addresses a key limitation in the field and enables mechanistic investigation of ICI-induced irAEs. Importantly, this mouse strain models myocarditis associated with monotherapy and combination ICI therapy, which are both fundamentally driven by aberrant T-cell activation. In patients, most T cells are antigen experienced, whereas most T cells in laboratory mice are naïve, which explains why PD-1 blockade can lead to myocarditis in humans, but additional perturbations are required in mice. Specifically, naïve T cells require CD28 costimulation to activate (which will not normally be provided for self-antigens), whereas in contrast, the activity of antigen-experienced cells can be enhanced by PD-1 blockade alone. Dual PD-1 and CTLA4 blockade enhances both of these processes, providing a mechanistic rationale for the association with a higher frequency of myocarditis. Fundamentally, these data support a threshold model of T-cell activation in which CTLA4 and PD-1 regulatory signals functionally integrate to provide a critical buffering system to restrain T-cell activation. Gene dosages of Ctla4 and Pdcd1 together define the functional threshold of T-cell self-recognition. These findings are consistent with the understanding that CTLA4 and PD-1 attenuate T-cell activation through distinct cellular and molecular mechanisms (16–18), provide mechanistic insights into how ICI-associated myocarditis arises, and present a preclinical model to test therapeutic interventions to mitigate irAEs.

Based on the functional interaction between CTLA4 and PD-1, intervention with CTLA4–Ig, which blocks T-cell costimulation, may be an effective therapeutic intervention for myocarditis associated with monotherapy and combination ICI therapy—a provocative possibility that is supported by preliminary clinical reports and warrants future prospective clinical testing. However, the effect of CTLA4–Ig treatment on tumor growth must be fully understood, particularly given that CD28 signaling is required for T-cell priming and for responses to checkpoint blockade (16, 17, 29). It is tempting to speculate that CTLA4–Ig could be administered after the initial priming of tumor-reactive T cells, allowing antitumor responses to be maintained; however, this possibility requires investigation.

Critical open questions remain, including how ICI efficacy and irAEs are related, how irAEs may be mitigated, and in what contexts ICI therapy is warranted. Our results reveal a functional interaction between CTLA4 and PD-1, with particular relevance to cardiovascular homeostasis. Although mice heterozygous for Ctla4 appear normal, complete loss of CTLA4 leads to unrestrained fatal lymphoproliferation; however, the inflammation is not specific to the cardiovascular system in these models. On the other hand, the phenotype of Pdcd1^−/− mice is dependent on genetic background. Although PD-1 loss in a C57BL/6 strain does not result in significant cardiovascular toxicity, BALB/c mice deficient for PD-1 display autoantibody-mediated cardiomyopathy upon aging (24, 25, 30). However, this model does not recapitulate ICI myocarditis in humans clinically or pathologically. On the other hand, the disease etiology of Ctla4^+/− Pdcd1^−/− mice closely recapitulates ICI-associated myocarditis. The sex imbalance seen in our mice is consistent with emerging data from patients where female sex is associated with increased risk of irAEs and myocarditis specifically (11, 12). These early observations in ICI myocarditis are in contrast with general myocarditis which appears to occur with higher frequency in men (31).

From a perspective of basic T-cell biology, the precise molecular and cellular mechanism by which CTLA4 and PD-1 functionally interact remains unclear. These data further support a threshold model of T-cell costimulation, in which loss of PD-1 and CTLA4 leads to a decreased threshold for T-cell activation and enables aberrant activation of self-reactive T cells. The manifestation of severe autoimmunity in specific tissues in this model and individual patients receiving ICIs may reflect one or more potential mechanisms, including the predisposition of particular antigens to autoimmune recognition in the absence of negative costimulation, functional differences between tissue-specific regulatory T-cell (Treg) populations (32), or environmental factors.

Our findings also raise the possibility that subtle changes in gene dosage of CTLA4 and PDCD1 may predispose some patients to the development of cardiovascular irAEs. Heterozygous germline loss of function alleles of CTLA4 in humans leads to immune dysregulation with highly variable clinical presentation and incomplete penetrance (33–35). Our preclinical data provide biological rationale and support further investigation of CTLA4–Ig as a potential therapy for ICI-induced autoimmune irAEs. Our results are further supported by preliminary patient data whereby treatment with abatacept attenuated cases of steroid-refractory ICI-associated myocarditis. Although there was no evidence of tumor growth due to abatacept treatment in the small sample of patients that we had, longer follow-up and more patient cases are needed. Further data are needed to understand the effects of abatacept on tumor growth in preclinical and clinical settings. These clinical data are provocative but preliminary, but they support the pivotal importance of manipulating T-cell costimulation for the treatment of ICI myocarditis and the need for prospective interventional trials with abatacept for ICI-associated myocarditis, where both cardiovascular endpoints as well as cancer endpoints would be assessed.

Ctla4^tm1All mice (19) were bred to Pdcd1-knockout mice (Pdcd1^tm1.1Shr; ref. 20), which were purchased from The Jackson Laboratory (021157). Pdcd1-knockout mice were backcrossed once to C57BL/6J, prior to breeding crosses with Ctla4 mice. We first used a heterozygous intercross-breeding scheme using Ctla4- and Pdcd1-knockout transgenic mice (Supplementary Fig. S1A) to allow for the generation of all possible permutations of mutant alleles from a single cross and for estimation of the observed recombination frequency between Ctla4 and Pdcd1. This first breeding scheme specifically utilized F1 mice derived from the cross of Ctla4^+/− (which are wild-type for Pdcd1) and Pdcd1^−/− mice (which are wild-type for Ctla4). This is important, because this breeding scheme ensures that mutant alleles of Ctla4 and Pdcd1 in F1 mice are in trans, and thus the recombination frequency can be calculated. We observed a recombination frequency of 17.03 cM between the murine Ctla4 and Pdcd1 loci, consistent with the 16.89 cM distance estimate by the Mouse Phenome Database (RRID:SCR_003212; ref. 36). The genetic distance between Ctla4 and Pdcd1 was calculated by assessing the number of recombination and total events in this cross.

To verify findings from the first breeding scheme, a second related breeding scheme was utilized (Supplementary Fig. S1A). Importantly, this breeding approach utilized different genotypes; the mutant alleles could be either in cis or in trans and would generate Ctla4^+/− Pdcd1^−/− (experimental) and Ctla4^+/+ Pdcd1^−/− (control littermates) in a 1:1 ratio. This allows for the generation of many more Ctla4^+/− Pdcd1^−/− mice than in the initial breeding approach. Specifically, male Ctla4^+/− Pdcd1^−/− and female Ctla4^+/+ Pdcd1^−/− were bred. Female Ctla4^+/+ Pdcd1^−/− were used to eliminate the possibility that the autoimmunity or other phenotypes observed in Ctla4^−/+ Pdcd1^−/− might affect fetal-maternal tolerance or the ability to produce viable litters.

For the generation of survival curves, events were defined as either death (i.e., mice found dead) or identification of mice by veterinary staff as requiring euthanasia (e.g., due to lethargy, moribund, dyspnea). For mice identified as requiring euthanasia, the date of death was defined as the day the mouse was identified by veterinary staff. Animal phenotypes associated with mortality were identified and reported by veterinary staff. Mice utilized for breeding or other analyses (e.g., histologic examination) were included in survival analyses in addition to initial cohorts primarily dedicated to assessing survival. Mice were censored from survival analyses at the time they were utilized for alternative purposes or analyses. Survival analyses of Ctla4^+/− Pdcd1^−/− Rag1^−/− and littermate Rag1 competent mice were described as above. Notably as a caveat of these specific analyses, because Ctla4^+/− Pdcd1^−/− Rag1^−/− mice are the only immunodeficient genotype within the groups expected to develop fatal myocarditis (i.e., Ctla4^+/− Pdcd1^−/− groups), premature deaths in this genotype can be attributed to either causes associated solely with immunodeficiency (e.g., infection) or incomplete rescue of autoimmune phenotypes (e.g., myocarditis) by loss of adaptive immunity. Analysis of female mice in which the premature death is more penetrant provided further support for the former possibility (Supplementary Fig. S9C).

All mice were housed at The University of Texas MD Anderson Cancer Center South Campus Vivarium, an Association for Assessment and Accreditation of Laboratory Animal Care International (AAALAC)–accredited, specific pathogen-free (SPF) animal facility, or at the Vanderbilt University Medical Center, also an AAALAC-accredited, SPF animal facility. Given that the vivaria are SPF-free and not germ-free, it is presumed that untested infectious agents may be present and it is possible that they could modulate the manifestation of disease. All experiments were performed in accordance with The University of Texas MD Anderson Cancer Center Institutional Animal Care and Use Committee (IACUC) guidelines or the Vanderbilt University Medical Center IACUC guidelines.

Genomic DNA was isolated using DirectPCR Lysis Reagent (Viagen Biotech, 101-T) supplemented with Proteinase K. PCR-based genotyping using crude lysate was performed for Ctla4- and Pdcd1-knockout mice. Ctla4^tm1All mice were genotyped as previously described (19). The expected band sizes for the Ctla4 wild-type and mutant alleles are approximately 75 and 150 bp, respectively. Pdcd1-knockout mice were genotyped as previously described (20). The expected band sizes for the Pdcd1 wild-type and mutant alleles are 418 and 350 bp, respectively. Ctla4 primers: 5′ AAACAACCCCAAGCTAACTGCGACAAGG 3′, 5′ CCAGAACCATGCCCGGATTCTGACTTC 3′, 5′ CCAAGTGCCCAGCGGGGCTGCTAAA 3′. Pdcd1 primers: PD1 KO common 24743 (5′ CACTATCCCACTGACCCTTCA), PD1 KO WT rev (5′ AGAAGGTGAGGGACCTCCAG), and PD1 KO Mut rev (5′ CACAGGGTAGGCATGTAGCA).

Crude genomic DNA lysate was submitted to The University of Texas MD Anderson Cancer Center Laboratory Animal Genetics Services core facility for SNP typing using a 100-marker panel. Ctla4^−/+ Pdcd1^−/− mice derived from both the first and second (validation) breeding schemes were analyzed. For the purpose of determining whether genetic variants associate with autoimmunity, “unaffected” (defined as no observable clinical symptoms within the 6 months of age) and “affected” (defined as mice that died or were identified as requiring euthanasia) Ctla4^−/+ Pdcd1^−/− mice were compared. 129 (the original derivation strain) and B6 SNPs were assessed to test for association with phenotypic outcomes.

Animal necropsies were performed by personnel in The University of Texas MD Anderson Cancer Center veterinary medical histology laboratory, in the Allison laboratory, or at Vanderbilt University Medical Center. Automated serum chemistry analysis using a Cobas Integra 400Plus (Roche Diagnostics) was performed on a blood sample collected at euthanasia. Formalin-fixed tissues were processed routinely into paraffin blocks, sectioned at 5 μm, and stained with H&E. Additional sections were used for IHC staining of particular tissues of interest, using an antibody directed against CD3 (ab16669, Abcam), CD4 (25229, Cell Signaling Technologies), CD8a (98941, Cell Signaling Technologies), F4/80 (ab100790, Abcam), or FOXP3 (12653, Cell Signaling Technologies), followed by secondary reagents for chromogenic detection (Bond Polymer Refine Detection system, DS9800, Leica). Stained sections were examined using a Leica DM2500 microscope with Leica DFC495 camera and Leica Application Suite v4.12 software.

Formalin-fixed tissues were routinely processed, cut at 4 to 5 μm, and stained with H&E by standard protocols in Vanderbilt University Medical Center's Translational Pathology Shared Resource (TPSR) core laboratory. To further characterize the mononuclear cardiac infiltrates detected by light microscopy, a panel of IHC markers was employed. IHC staining was performed in the TPSR using standard, validated protocols for chromogenic IHC. All steps besides dehydration, clearing, and coverslipping were performed on the Leica Bond-Max IHC autostainer (Leica Biosystems Inc.). Slides were deparaffinized. Antigen retrieval was performed using EDTA (CD markers) or proteinase K (F4/80). Slides were incubated with primary antibodies as indicated below. Secondary antibody labeling was performed for all markers except CD3 by incubating in rabbit anti-rat antibody (BA-4001, Vector Laboratories, Inc.) for 15 minutes at a 1:650 dilution. Immunolabeling by rabbit antibody was visualized using the Bond polymer refine detection system (#DS9800, Leica Biosystems, Inc.). Slides were then dehydrated, cleared, and coverslipped. For primary antibodies, anti-CD45R (BD Biosciences, RA3–6B2) was used at 1:20,000 dilution as a B-cell marker, anti-CD3 (Abcam, Ab16669) was used at 1:250 dilution as a pan–T-cell marker, anti-CD4 (eBioscience, 14–9766–82) was used at 1:1,000 dilution as a Th cell marker, anti-CD8 (eBioscience, 14–0808–82) was used at 1:1,000 dilution as Tc cell marker, and anti-F4/80 (Novus Biologicals, NB600–404) was used at 1:900 dilution as macrophage marker.

Microscopic analysis and semiquantitative scoring of morphologic lesions were conducted by two board-certified veterinary anatomic pathologists, one of whom was blinded to the mouse genotypes/treatments. Semiquantitative scoring was performed to characterize the magnitude of myocardial infiltrates in six distinct cardiac regions: ventricular endocardium, ventricular myocardium and perivascular adventitia, ventricular epicardium, atria, aorta, and pericardium. The sum of scores assigned for all regions was calculated for each mouse for a composite score, and group means of the composite scores are presented. Scores were performed as follows: 0, within normal limits; 1, minimal (<10% affected); 2, mild (10%–20% affected); 3, moderate (20%–50% affected); and 4, marked (>50% affected).

Image analysis was conducted by a board-certified veterinary anatomic pathologist under masked conditions. Heart sections from the same mice evaluated by H&E and immunophenotyping were analyzed. Slide scanning was performed on the Pannoramic 250 Flash III digital scanner (3DHISTECH Ltd.). Quantification of immunolabeled cells in the hearts was performed in QuPath version 0.2.0-m4, an open source digital pathology platform (37), using manual region of interest delineation of the atria and ventricles and a single-threshold–positive cell detection feature. Data were collected per heart section, and then group-wide means of percentage of positive CD3, CD4, and CD8 cells were calculated. Parameters were set up as follows: detection image: hematoxylin OD, pixel size: 0.5 μm, nucleus—background radius: 8 μm, nucleus—median filter radius: 0 μm, nucleus—sigma: 1.2 μm, nucleus—min area: 10 μm², nucleus—max area: 400 μm², intensity—threshold: 0.1, max background intensity: 2, intensity: split by shape, cell expansion: 4 μm, cell parameter: include cell nucleus, boundaries: smooth boundaries, measurements: make measurements, score compartment: nucleus—DAB OD mean, threshold 1+: 0.2.

Single-cell suspensions were obtained from hearts of 3- to 6-week-old female wild-type, Ctla4^+/+ Pdcd1^−/− or Ctla4^+/− Pdcd1^−/− mice by mincing followed by enzymatic digestion with 125 U/mL DNase I (Worthington; cat no. LS002138) and 250 U/mL Collagenase 3 (Worthington; cat no. LS004182). Dissociated hearts were filtered through a 70 μm filter. Red blood cells were lysed using ACK lysing buffer (KD Medical/MediaTech; cat no. NC0274127). Cells were washed with PBS and stained for flow cytometry. Cells were first stained with Zombie Violet fixable viability dye for 20 minutes at room temperature, followed by Fc block [TruStain FcX (anti-mouse CD16/32) Antibody; BioLegend; cat no. 101320] for 5 minutes at 4°C. Cells were washed and stained with surface stain for 20 minutes at 4°C, as indicated below. Cells were then fixed with either eBioscience Fix/Perm (Thermo Fisher Scientific; cat no. 88–8824–00) kit for cytoplasmic antigens (i.e., CD68) or True-Nuclear Transcription Factor Buffer Set (BioLegend; cat no. 424401) for nuclear antigens (i.e., FoxP3). Samples were run on an Attune NxT Acoustic Focusing Cytometer. Analysis was performed in FlowJo, using two different panels of antibodies on the same sample. Gating was first done on forward scatter (FSC) and side scatter to exclude debris. Doublets were excluded by gating on FSC area versus FSC height. Given massive cell death in the cardiomyocytes, percentages were calculated out of total cardiac single cells. Next, viable (zombie violet negative) immune cells (CD45⁺) were gated and used for all downstream immune cell gates. T cells were defined as CD3⁺ and gated into CD3⁺CD4⁺CD8⁻ or CD3⁺CD8⁺CD4⁻. Tregs were defined as CD3⁺CD4⁺CD8⁻FoxP3⁺. Macrophages were defined as CD68⁺. Natural killer/NK-T cells were defined as CD68-NK1.1⁺. B cells were defined as CD68⁻CD19⁺. CD68⁻CD11c⁺ cells are also enumerated, which are primarily comprised of dendritic cells. Zombie Violet (cat no. 423114), PerCP/Cy5.5 anti-mouse CD45 (clone 30-F11; cat no. 103132), Alexa Flour 488 anti-mouse CD3 (clone 17A2; cat no. 100210), APC anti-mouse CD4 (clone GK1.5; cat no. 100412), PE/Cy7 anti-mouse CD8a (clone 53–6.7; cat no. 100722), APC/Cy7 anti-mouse CD19 (clone 6D5; cat no. 115530), Alexa Flour 488 anti-mouse CD11c (clone N418; cat no. 117311), PE/Cy7 anti-mouse IA/IE (clone M5/114.15.2; cat no. 107630), PE anti-mouse CD68 (clone FA-11; cat no. 137013), APC anti-mouse NK1.1 (clone PK136; cat no. 108710), and APC/Cy7 anti-mouse CD11b (clone M1/70; cat no. 101225) were purchased from BioLegend. PE anti-mouse FoxP3 (clone FJK-16s; cat no. 101225) was purchased from Thermo Fisher Scientific.

Single-cell suspensions from lymph nodes were prepared by mashing pooled inguinal, axillary, and brachial lymph nodes through a 70 μm filter using the back of a plastic syringe into RPMI-1640 supplemented with 10% FBS and 1% penicillin/streptomycin. A 96-well flat-bottom plate was coated 200 μL per well of 1 μg/mL anti-CD3e and 2 μg/mL anti-CD28 in PBS overnight at 4°C the previous night. Cells were then stained with CellTrace Violet Proliferation kit as per the manufacturer's protocol (Invitrogen, C34557). Triplicates of each sample were plated 10⁶ cells/mL per well in 200 μL of RPMI-1640 supplemented with 10% FBS, sodium pyruvate, 0.1% b-ME, and penicillin/streptomycin and incubated at 37°C for 46 hours. Cells were then transferred to a U-bottom 96-well plate, washed twice with FACS buffer, and incubated with 2% of each bovine, murine, rat, hamster, and rabbit serum PBS with 25 mg/mL 2.4G2 antibody at 4°C for 10 minutes prior to surface staining with an antibody cocktail at 4°C for 30 minutes in a 50 μL volume. Cells were washed twice with FACS buffer, fixed, and permeabilized using the FoxP3 fix and permeabilization kit according to the manufacturer's protocol (eBioscience). Cells were subsequently stained with an intracellular antibody cocktail at room temperature for 30 minutes. Cells were then washed twice with Foxp3 permeabilization buffer, then twice with FACS buffer, and analyzed on a LSRII (BD). PE mean fluorescence intensity was then quantified to assess CTLA4 protein levels in T cells from mice.

LIVE/DEAD Fixable Blue Dead Cell Stain (cat no. L23105) was purchased from Thermo Fisher Scientific. BV786 Hamster Anti-Mouse CD3e (clone 145–2C11; cat no. 564379) was purchased from BD. Brilliant Violet 605 anti-mouse TCR β chain antibody (clone H57–597; cat no. 109241), Brilliant Violet 650 anti-mouse CD19 antibody (clone 6D5; cat no. 115541), PE anti-mouse CD152 antibody (clone UC10–4B9; cat no. 106306), and PE Armenian Hamster IgG Isotype Ctrl antibody (clone HTK888; cat no. 400908) were purchased from BioLegend. FITC Anti-mouse CD4 Antibody (clone RM4.5; cat no. 11–0042–82), APC Anti-mouse CD8a Antibody (clone 53–6.7; cat no. 17–0081–82), and Alexa Fluor 700 Anti-mouse CD45.2 Antibody (clone 104; cat no. 56–0454–82) were purchased from ebio.

BV786 Hamster Anti-Mouse CD3e (clone 145–2C11; cat no. 564379), Brilliant Violet 605 anti-mouse TCR β chain antibody (clone H57–597; cat no. 109241), PE anti-mouse CD152 (CTLA4) Antibody (clone UC10–4B9; cat no. 106306), and PE Armenian Hamster IgG Isotype CTLA4 Ctrl antibody (clone HTK888; cat no. 400908) were purchased from BioLegend. FITC Anti-mouse CD4 Antibody (clone RM4.5; cat no. 11–0042–82) and APC Anti-mouse CD8a Antibody (clone 53–6.7; cat no. 17–0081–82) were purchased from ebio.

Female MRL/MpJ-Fas/lpr mice were purchased from The Jackson Laboratory at 3 to 4 weeks. After 1 week of acclimatization, mice were treated with anti–CTLA4, anti–PD-1 antibody (Thermo Fisher Scientific, both 200 μg), anti–PD-1 (Thermo Fisher Scientific, 200 μg), both, or vehicle. Immunofluorescence staining of cardiac tissues was performed at Vanderbilt University using anti-CD3 (Novus) and anti-CD68 (Novus) antibodies, as described above.

Mice were injected i.p. with either 100 μL PBS with 200 μg CTLA4 Ig (Abatacept, purchased from Vanderbilt pharmacy) or vehicle, 3 times a week, for up to 80 days, starting at 21 days of age. Although abatacept is a human CTLA-4-Ig fusion, it also functions in mouse models due to the high similarity between human and mouse CTLA4 (38).

ECG records were created in the different genetic models to assess presence of arrhythmogenic disturbances. Once mice were anesthetized with 3% isoflurane inhalation, needle electrodes were placed s.c. into all four limbs. Isoflurane was then lowered to 1% to maintain stable sedation, and ECG recording was started. ECG traces were recorded for 4 to 5 minutes. ECG analysis was conducted in LabChart 8 by two observers blinded to the genotype and then compared for precision. Conduction disturbances were quantified as the number of mice in each group as percentage. ECG records were reviewed and ventricular ectopic beats identified based on standard criteria (i.e., wide QRS complex, AV dissociation).

Before sacrifice, in vivo cardiac dimensions were assessed with M-mode and two-dimensional transthoracic echocardiography (Vevo 2100; Fujifilm VisualSonics). Mice were not anesthetized during echocardiography. Parasternal long axis views were used to record M-mode tracings at left ventricular midpapillary level to measure left ventricular dimensions (interventricular septum, left ventricular posterior wall, left ventricular internal diameter, and % ejection fraction).

Serum was collected from Ctla4^+/− Pdcd1^−/− and control littermate mice (including both Ctla4^+/+ Pdcd1^−/− mice and mice competent for both CTLA4 and PD-1 such as Ctla4^+/− Pdcd1^+/− mice) from both breeding schemes described above. Briefly, blood was collected by terminal cardiac puncture after deep anesthesia was established during CO₂ euthanasia. Blood was allowed to coagulate at room temperature, centrifuged at 2,000 × g for 10 minutes, supernatant serum collected, snap frozen in liquid nitrogen, and stored at −80°C for later analysis. Serum levels of antibodies cytokine and chemokine were assessed using the Cytokine & Chemokine 36-plex Mouse ProcartPlex luminex assay (Thermo Fisher Scientific) per the manufacturer's protocol. All samples were analyzed in parallel in a single batch for each respective analysis. Serum samples were diluted 1:10,000 for analysis of serum antibody levels. Cardiac TnI concentration in serum was measured by a mouse-specific ELISA (Cat# 2010–1-HSP, Life Diagnostics, Inc.). Standard and sample volumes were reduced to 25 μL from the manufacturer's recommended 100 μL per well to conserve limited sample volume.

mRNA was purified with TRIzol (Invitrogen) and RNeasy column (Qiagen). Total RNA (0.5 μg) was reverse transcribed (StrataScript First Strand cDNA Synthesis Kit; Stratagene) and analyzed by real-time PCR using RT2 Profiler PCR Arrays (SABiosciences) and Mx3005 thermocycler (Stratagene). The following primers were used for the cd274 (PD-L1) mRNA: 5′-TGCGGACTACAAGCGAATCACG-3′, 3′-CTCAGCTTCTGGATAACCCTCG-5′; 5′-GCTCCAAAGGACTTGTACGTG-3′, 3′-GCTCCAAAGGACTTGTACGTG-5′.

Paraffin-embedded tissue sections were immunostained for PD-L1 using a murine PD-L1 antibody (Cell Signaling Technology) at 1:75 dilution.

Hearts were fix-perfused with ice-cold buffer containing 2.5% paraformaldehyde, 2.5% glutaraldehyde, and 0.1 mol/L sodium cacodylate (pH 7.4; No. 15949; Electron Microscopy Sciences). Heart tissues from the endocardial aspect of the left ventricular wall were embedded for transmission electron microscopy (Tecnai G2 Spirit BioTWIN; FEI Co.) with a cooled charge-coupled device camera (XR41C; Advanced Microscopy Techniques) and Image Caption Engine acquisition software (Advanced Microscopy Techniques).

PD-1 inhibitors considered for disproportionality analysis in VigiBase were nivolumab, pembrolizumab, and cemiplimab. PD-L1 inhibitors considered for this analysis were atezolizumab, durvalumab, and avelumab, and CTLA4 inhibitor was ipilimumab. Methods utilized were as previously described (13, 39). Myocarditis (n = 413) was identified as reported (autoimmune myocarditis and/or myocarditis; preferred term levels of Medical Dictionary for Regulatory Activities) in patients who received combination ICI targeting CTLA4 and PD-1/PD-L1 (e.g., ipilimumab and nivolumab) compared with single CTLA4 or PD-1/PD-L1 therapy. With VigiBase (N = 16,337,777 reports since January 2008), reporting odds ratio allows comparisons across toxicities and drug groups.

Statistical analyses were performed in Prism 7.0 or 8.0 (GraphPad Software), unless otherwise noted.

Publicly available data were obtained from VigiBase (https://www.who-umc.org/vigibase/vigibase/). All other data are available upon request from the authors after appropriate approvals.

All evaluations of clinical data on adverse drug reactions from multisource databases and real-world signal detection were performed under ethics guidelines established by and with the approval of the Vanderbilt University Medical Center Human Research Protections Program and Institutional Review Board (IRB # 181337), who determined that the work did not meet the definition of human subject research. No identifiable private information was obtained by the research team.

S.C. Wei reports personal fees from The University of Texas MD Anderson Cancer Center Odyssey Program during the conduct of the study; personal fees and other from Spotlight Therapeutics; and non-financial support from American Association of Immunologists outside the submitted work; in addition, S.C. Wei has a patent for a genetic mouse model of autoimmune adverse events and immune checkpoint blockade therapy (PCT/US2019/050551) pending to Board of Regents, The University of Texas System. D.B. Johnson reports other from Array Biopharma, grants and other from BMS, other from Catalyst, other from Iovance, other from Jansen, other from Merck, other from Novartis, and other from Oncosec outside the submitted work. E. Whitley reports grants from NIH during the conduct of the study. L. Lehmann has served on the advisory board for Daiichi Sankyio, Senaca, and Servier and received speakers' honoraria from Novartis and MSD. P.-Y. Courand reports non-financial support from Servier, personal fees and non-financial support from AMGEN, personal fees from BMS, personal fees from Janssen, and personal fees from Takeda outside the submitted work. L.E. Himmel reports grants from NIH and grants from NIH during the conduct of the study. P. Sharma reports personal fees and other from Jounce Therapeutics, other from BioNTech, other from Adaptive Biotechnologies, personal fees and other from Codiak, personal fees and other from Forty-Seven, personal fees and other from Dragonfly, personal fees and other from Gympse, personal fees and other from Phenomics, personal fees and other from Lava, personal fees and other from Oncolytics, personal fees and other from Lytix, personal fees and other from Achelois, other from Constellation, personal fees and other from BioAtla, personal fees and other from Polaris, personal fees and other from Marker, personal fees and other from Hummingbird, personal fees and other from ImaginAb, and personal fees and other from Infinity outside the submitted work. J.-E. Salem reports personal fees from BMS during the conduct of the study; in addition, J.-E. Salem has a patent for Abatacept for the treatment of immune-related adverse events issued. J.M. Balko reports grants from Genentech/Roche, grants from Bristol Myers Squibb, grants from Incyte, and personal fees from Novartis outside the submitted work; in addition, J.M. Balko has a patent for Immunotherapy biomarkers issued. J.J. Moslehi reports personal fees from BMS, personal fees from Immunocore, personal fees from Boston Biomedical, personal fees from Deciphera, personal fees from Janssen, personal fees from AstraZeneca, personal fees from Myovant, personal fees from Triple Gene/Precigen, personal fees from Cytokinetics, personal fees from Takeda, and personal fees from GSK during the conduct of the study. J.P. Allison reports personal fees from Achelois, personal fees from Apricity Health, personal fees from BioAtla, personal fees from Codiak BioSciences, personal fees from Dragonfly Therapeutics, personal fees from Forty-Seven Inc., personal fees from Hummingbird, personal fees from ImaginAb, personal fees from Jounce Therapeutics, personal fees from Lava Therapeutics, personal fees from Lytix Biopharma, personal fees from Marker Therapeutics, personal fees from BioNTx, and personal fees from Polaris outside the submitted work; in addition, J.P. Allison has a patent for a genetic mouse model of immune checkpoint blockade induced immune-related adverse events pending to The University of Texas MD Anderson Cancer Center; and have received royalties from intellectual property licensed to BMS and Merck. No disclosures were reported by the other authors.

S.C. Wei: Conceptualization, resources, supervision, funding acquisition, writing-original draft. W.C. Meijers: Conceptualization, investigation. M.L. Axelrod: Investigation. N.-A.A.S. Anang: Investigation. E.M. Screever: Investigation. E.C. Wescott: Investigation. D.B. Johnson: Investigation. E. Whitley: Formal analysis, investigation, visualization, writing-review and editing. L. Lehmann: Investigation. P.-Y. Courand: Investigation. J.J. Mancuso: Project administration, writing-review and editing. L.E. Himmel: Investigation, writing-review and editing. B. Lebrun-Vignes: Investigation. M.J. Wleklinski: Investigation. B.C. Knollmann: Investigation.J. Srinivasan: Investigation. Y. Li: Investigation. O.T. Atolagbe: Investigation, visualization. X. Rao: Formal analysis. Y. Zhao: Formal analysis. J. Wang: Formal analysis, supervision. L.I.R. Ehrlich: Investigation. P. Sharma: Writing-review and editing. J.-E. Salem: Investigation. J.M. Balko: Formal analysis, supervision, funding acquisition, investigation, writing-review and editing. J.J. Moslehi: Conceptualization, formal analysis, supervision, writing-original draft, writing-review and editing. J.P. Allison: Conceptualization, resources, supervision, funding acquisition, investigation, writing-original draft, writing-review and editing.

We acknowledge the contributions of Mary Barber, Gosife D. Okoye, Yueli Zhang, and Elie Tannous for the early parts of this project as well as Drs. Maël Richard, Anissa Bourli, Brahim Harbaoui, Raphaele Nove-Josserand, Julien Peron, Tanguy Fenouil, and Pierre Lanteleme. This work was supported by a grant from Cancer Prevention and Research in Texas to J.P. Allison (R1203) and NCI Cancer Center Support Grant (CCSG) P30CA16672 which supports core facilities at MD Anderson Cancer Center including the Reverse Phase Proteomic array, Veterinary Pathology, Laboratory Animal Genetics, and Statistical Support cores that were utilized in the described work, as well as NCI/NIH Cancer Center Support Grant 5P30 CA68485–19 (to Vanderbilt) and the Vanderbilt Mouse Metabolic Phenotyping Center Grant 2 U24 DK059637–16. J.J. Moslehi was supported by NIH grants R56 HL141466 and R01 HL141466. D.B. Johnson and J.M. Balko were supported by NCI/NIH R01CA227481. Additional funding was provided by NIH T32GM007347 (M.L. Axelrod) and F30CA236157 (M.L. Axelrod). S.C. Wei was an MD Anderson Cancer Center Odyssey postdoctoral fellow and is currently an employee of Spotlight Therapeutics. J.P. Allison is a CPRIT Scholar in Cancer Research and a Director of the Parker Institute for Cancer Immunotherapy. The supplied data from VigiBase come from various sources. The likelihood of a causal relationship is not the same in all reports. The information does not represent the opinion of the WHO. We also thank Lin Zhong from the Vanderbilt University Cardiology Physiology Core for assistance with echocardiogram.