CXCR6+ T-cells drive Immune Checkpoint Inhibitor Myocarditis

Abstract

Background:

Myocarditis is a severe complication of immune checkpoint inhibitors (ICIs). The major risk factor for ICI-myocarditis is the use of combination ICI-treatment especially when relatlimab, a novel anti-LAG-3 antibody, is combined with anti-PD-1 therapy. While pathogenic T-cells are necessary for ICI-myocarditis, the specific signaling and T-cell populations that drive cardiac infiltration have not been fully elucidated, especially in setting of anti-LAG-3/PD-1 treatment.

Methods:

We utilized VigiBase, an international pharmacovigilance database, to assess the risk of myocarditis with anti-LAG-3 compared to other ICI treatment regimens. We identified a mouse model of LAG-3/PD-1 associated ICI-myocarditis through genetic deletion of immune checkpoints LAG-3 and PD-1 (Lag3−/−, Pdcd1−/− mice), and performed rigorous cardiac phenotyping using histology, flow cytometry, electrocardiography, single-cell RNA-sequencing, and antibody-induced cellular depletion.

Results:

We found an increased risk of myocarditis with anti-LAG-3 combination treatment clinically, confirming early clinical trial data. Lag3−/−, Pdcd1−/− mice were found to develop severe cardiac inflammation by histology with increased cardiac macrophages and clonal T-cells which was associated with the development of spontaneous arrhythmias leading to premature death by 6-8 weeks. We identified CXCR6 as a key marker of activated cardiac T-cells in this model, along with analogous signals in other preclinical models and patient data. CXCR6 marked a heterogenous group of cardiac T-cells, including distinct clusters of Gzmk, Gzmb, Cd4, and actively dividing T-cells. CXCL16, the sole known ligand for CXCR6, was similarly upregulated in the cardiac macrophage population. Treatment with anti-CXCR6 antibody prevented premature lethality, attenuated arrhythmias, and reduced the histologic severity of myocarditis – demonstrating that CXCR6+ T-cells are necessary for disease pathogenesis.

Conclusions:

Our findings suggest that ICI-myocarditis is driven by an expansion of CXCR6+ T-cells and identifies CXCR6 as a putative therapeutic target for this highly morbid condition.

Introduction:

Immune checkpoint inhibitors (ICIs), antibodies that target inhibitory cell-surface proteins on T cells such as Cytotoxic T-lymphocyte-associated protein 4(CTLA-4), Programmed cell death protein 1 (PD-1), and Lymphocyte-activation gene 3 (LAG-3), have dramatically changed the treatment of many cancers. Indeed, anti-PD-1 therapy has become a first line therapy for several cancers that were previously associated with poor survival including metastatic melanoma¹. However, ICIs (especially in combination) can lead to untoward immunological responses in healthy tissues, termed immune related adverse events (irAEs)². ICI-myocarditis is a rare (<1-2% of ICI-treated patients) but highly morbid adverse event, with a mortality rate of up to 46%³. The current mainstay of ICI-myocarditis treatment is corticosteroids which result in non-specific immunosuppression but have limited benefit and numerous adverse side effects⁴. As such, there is a critical unmet clinical need to identify novel therapies for this disease.

LAG-3 is a newly targeted T-cell immune checkpoint. Combination of relatlimab (anti-LAG-3) with nivolumab (anti-PD-1) has been FDA-approved for metastatic melanoma in 2022⁵. In the seminal clinical trial that led to the approval of relatlimab, in addition to the benefit in melanoma there was a nearly three-fold increased risk of myocarditis when anti-LAG-3 was added to anti-PD-1 treatment⁵. This significant increase in clinical myocarditis cases necessitates further preclinical investigation of LAG3-associated ICI-myocarditis, especially given that there are currently more than 15 different anti-LAG3 antibodies being tested in over 100 clinical trials for various cancer types and stages^6,7.

Growing data from our group and others have shown that T-cells drive ICI-myocarditis. Using a preclinical model of CTLA-4/PD-1 ICI-myocarditis, our group and others have demonstrated that an expansion of clonal, CD8+ T-cells, in part driven by autoreactivity against alpha-myosin heavy chain (a-MHC), is necessary for the development of ICI myocarditis.^8–11. Other investigations have found expansion of terminal effector memory CD8+ T-cells (TEMRA) expanded in the blood of patients with ICI-myocarditis¹². More recently, the chemokine signaling axis between CXCL9/CXCL10+ macrophages and CXCR3+ T-cells was identified to recruit peripheral T-cells to the heart in ICI myocarditis^13,14. Indeed, macrophage depletion attenuated myocarditis in multiple pre-clinical models, suggesting that specific groups of cardiac macrophages play an important supporting role in the pathogenesis of ICI-myocarditis^13,14. Outside of these data, the granular cardiac T-cell subsets and signals that drive cellular recruitment, expansion, and localization within the cardiac environment in ICI myocarditis remain poorly understood.

Here, we provide pharmacovigilance data confirming an increased risk for myocarditis with anti-LAG3 compounds. We perform deep cardiac phenotyping in mice lacking LAG-3 and PD-1 (Lag3−/−, Pdcd1−/−) which spontaneously develop severe cardiac inflammation with pathology reminiscent of ICI myocarditis in patients. Through cardiac immune scRNA-seq and flow cytometry we uncover cardiac enrichment of CXCR6+ T-cells and validated these findings with complementary preclinical models and patient data. We then demonstrate that the CXCR6 ligand, CXCL16, is upregulated in cardiac macrophages. Treatment with anti-CXCR6 prevented pre-mature lethality, attenuated arrhythmias, and reduced the histologic severity of myocarditis. These data highlight the necessity of CXCR6+ T-cells for myocarditis pathogenesis and give biologic plausibility to the use of anti-CXCR6 treatment in ICI-myocarditis.

Methods:

Animals

Pdcd1 null and Lag3 null mice were purchased from the Jackson Laboratory (Strain 026644 and 028276). Lines were crossed to generate Lag3/Pdcd1 null animals (referred to as Lag3−/−,Pdcd1−/−). Mice were housed at the UCSF Cardiovascular Research Institute Barrier, an animal facility accredited by the Associated for Assessment and Accreditation of Laboratory Animal Care International. Animals were maintained in a controlled environment with a 12-hour light-dark cycle with access to water and a standard chow diet at all times. All experiments were performed in accordance with an IACUC protocol (AN207188). The study adheres to the ARRIVE guidelines for reporting animal research. Both male and female mice were studied in equal number. Mice were euthanized via deep isoflurane anesthesia followed by cervical dislocation. Mouse survival was assessed via Kaplan-Meier Survival curves with log rank statistical analysis. For Ctla4+/−,Pdcd1−/− mice, animal studies were performed in compliance with guidelines set forth by the National Institutes of Health Office of Laboratory Animal Welfare and approved by the Washington University institutional animal care and use committee. Ctla4+/−, Pdcd1−/− female mice were maintained and used as previously described⁹.

Generation of single cell cardiac suspension for scRNA-TCR Seq

Mice were sacrificed and hearts were harvested and washed thoroughly with PBS. Hearts were finely minced and enzymatically digested in mixture of 250U/mL Collagenase 3 (Worthington, LS004182), 125U/mL DNAse I (Worthington, LS002138), 170U/mL Hyaluronidase (Sigma, H3506) in RPMI for 50 minutes on GentleMACS system. Red blood cells were lysed using ACK lysing buffer (KD Medical/MediaTech, NC0274127). Cells were stained with PerCP/Cyanine5.5 anti-mouse CD45 (BioLegend, 157612, clone QA17A26) for 30 minutes at 4C. Following staining and washing with PBS, cells were resuspended in PBS with DAPI (1:10,000). Live CD45+ immune cells were sorted by FACS on PerCP/Cy5.5 positive, DAPI negative events. Given the paucity of cardiac lymphocytes at homeostasis, the wild-type control sample consisted of pooled cardiac immune infiltrates from 6 mice. The myocarditis sample consisted of inflamed hearts from 4 Lag3−/−, Pdcd1−/− mice. All mice were approximately 6 weeks of age. Lag3−/−, Pdcd1−/− mice were multiplexed with TotalSeq-C Hashtag antibodies (TotalSeq-C0301, TotalSeq-C0302, TotalSeq-C0303, TotalSeq-C0304).

scRNA–TCR-seq

Each sample (targeting 5,000–15,000 cells per sample) was processed for single-cell 5′ RNA and TCR sequencing utilizing the 10x genomics Chromium Next GEM 5’ Gene Expression Reagents v2. Libraries were prepared following the manufacturer’s protocol with the assistance of Gladstone Genomics Core. The libraries were sequenced using NovaSeq X and analysis was completed using 10x Genomics Cell Ranger software. We identified 13,741 cells from pooled Lag3−/−, Pdcd1−/− mice (1,126 median genes per cell; 39,289 reads per cell) and 4,014 cells from pooled WT mice (536 median genes per cell; 106,046 reads per cell). Data were analyzed in R using the filtered h5 gene matrices in the Seurat. In brief, samples were subset to include cells with >200 but <2,500 unique transcripts to exclude probable non-cellular RNA reads and doublets, respectively. Cells with >5% mitochondrial transcripts were also excluded. The DoubletFinder R-package was used to detect and remove doublets from analysis. Seven clusters were identified using a resolution of 0.1. UMAP was used for dimensionality reduction with 15 nearest neighbors and minimum distance of 0.5. For differential gene expression analysis of “Hyperexpanded” clones were compared to all “Activated T-cells”. The FindMarkers Seurat function was used with minimum percent of 25% and minimum log2fold change of 0.25. Gene Ontology Pathway enrichment was performed using ClusterProfiler R-package on marker genes of Isg15 monocytes and cardiac macrophages (markers were defined using FindAllMarkers Seurat function with abundance of >50% and log2fold change of >0.5).

Histology

Organs were dissected from mice and fixed in 10% formalin for 48h and then transferred to 70% ethanol. Samples were grossed, embedded, sectioned and H&E stained at AML Laboratories. The severity of the cardiac inflammation was scored by an independent, blinded pathologist. Lesional area for anti-CXCR6, anti-CXCR3, and Isotype depletion was calculated by a blinded pathologist by tracing the perimeter of the inflammatory infiltrate and recording the area. This area was then normalized to the total area of ventricular cardiac tissue present on the slide. All slides were sectioned in the transverse short axis through the mid left ventricle to minimize variation between mice.

Chemistries

Blood was obtained upon sacrifice of animals and spun at 10,000g for 10 mins, from which plasma was isolated. Analysis of chemistries was performed at University of Michigan ULAM Pathology Core. Cardiac troponin I analysis was quantified by ELISA (Life Diagnostics, CTNI-1-US).

Antibody Depletion

At 21 days of age, Lag3−/−, Pdcd1−/− mice were randomly assigned to anti-CXCR6 antibody (courtesy of Edelweiss Immune), anti-CXCR3 antibody (BioXCell, BE0249), or rat IgG1 isotype control (BioXCell, BE0088) administered at a dose of 250ug i.p. twice weekly in 100uL. Experiments were concluded when mice reached 70 days of age. Weekly EKGs were performed until time of death or experimental endpoint. Organs were harvested for histology. To detect an anticipated mortality difference of 100% (for control) to 40% (for an intervention that reduces mortality) with an α of 0.05 and 80% power, a sample size of 8 mice per group was used.

Anti-CXCR6 treatment of Ctla4+/−, Pdcd1−/− mice

Ctla4+/−,Pdcd1−/− female mice were randomly assigned to receive vehicle control or anti-CXCR6 treatment at 21 days of age. Mice were treated with anti-CXCR6 or vehicle twice weekly. After 3 weeks of treatment mice were sacrificed and hearts were enzymatically digested for flow cytometry.

Flow Cytometry

Lag3−/−, Pdcd1−/− samples were run on an Attune NxT Acoustic Focusing cytometer (Life Technologies). Data were collected using Attune NxT software v.3.2.1. Ctla4+/−,Pdcd1 samples were collected on a BD FACSMelody or Cytek Aurora. Analysis was performed in FlowJo v.10.10. The gating strategy consisted of forward scatter and side scatter to exclude debris, FSC area versus FSC height to exclude doublets, and Zombie Violet Viability to exclude dead cells (VWR, 10761-308). Gating strategy can be found in Figure S1. The following antibodies were used for Lag3−/−, Pdcd1−/− samples: CD45-PerCP/Cy5.5 (BioLegend, 103132, clone 30-F11; dilution 1:400); CD3-AF488 (BioLegend, 100210, clone 17A2; dilution 1:200); CD4-APC (BioLegend, 100412, clone GK1.5; dilution 1:100); CD8a-PE/Cy7 (BioLegend, 100722, clone 53-6.7, dilution 1:400), CD11b-APC/Cy7 (BioLegend, 101225, clone M1/70, dilution 1:400), and F4/80-PE-eFluor610 (Fischer Scientific, 50-112-9765, clone BM8, 1:100). The following antibodies were used for Ctla4+/−,Pdcd1−/− samples: CD45-BV510 (Biolegend, 103137, clone 30F-11, dilution 1:200), CD3-FITC (eBioscience, 4338511, clone 145-2C11, dilution 1:200), CD4-PE/Cy7(Biolegend, 100422, clone GK1.5, dilution 1:200), CD8-APC/Cy7 (Biolegend, 100714, clone 53-6.7, dilution 1:200), CD44-APC (Biolegend, 103012, clone IM7, dilution 1:200), and CD62L-PE (Biolegend, 10440, clone MEL-14, dilution 1:200).

Electrocardiography

Mice were anesthetized at 2% isoflurane in accordance with IACUC protocol and one-lead electrocardiogram (lead pins inserted in right arm and left leg) was performed using AD Instruments Power Lab-C and continuous ECG tracing was recorded for 30 seconds and analyzed with LabChart Pro 8. Arrhythmias were defined as high grade atrioventricular block or ventricular arrhythmias including premature ventricular contractions and ventricular escape rhythms.

Echocardiography

Mice were anesthetized at 2% isoflurane and echocardiography was performed using Vevo 3100 Ultrasound Machine. Parasternal long axis images of the left ventricle were analyzed and analyzed using Vevo software.

Chemotaxis Assay

Lag3−/−, Pdcd1−/− mice were harvested and cardiac immune cells and PBMCs were isolated as described above. Both PBMCs and cardiac immune cells were stained with DAPI, CD45-PerCP/Cy5.5 (BioLegend, 103132, clone 30-F11; dilution 1:400), and CD3-AF488 (BioLegend, 100210, clone 17A2; dilution 1:200). Cells were then sorted via FACS for DAPI negative/PerCP-Cy5.5 positive/AF488 positive cells. A total of 20,000 blood T-cells and 10,000 cardiac T-cells were plated in replicate on QCM Chemotaxis Cell Migration Assay, 24-well (5 μm), colorimetric (Millipore Sigma, ECM506) with 10% Fetal Bovine Serum (FBS), RPMI only and RPMI + 0.15ug/mL of recombinant mouse CXCL16 (Peprotech, 250-28) in bottom chamber and placed for 4h in 37C incubator. Per kit instructions, colorimetric analysis was performed on bottom chamber at 560nm.

Pharmacovigilance Analysis

We queried VigiBase, the international World Health Organization pharmacovigilance database encompassing over 32 million reports from over 130 countries since 1967 to January 1st, 2024 (http://www.vigiaccess.org/)^16,17. We searched for the following ICI: anti-CTLA4: ipilimumab, tremelimumab; anti-PD1: nivolumab, pembrolizumab, cemiplimab, dostarlimab, retifanlimab, toripalimab, tislelizumab; anti-LAG-3: relatlimab and specifically, myocarditis within 24 immune related adverse events (irAE) established to cover all organs potentially affected by ICI-induced autoimmunity using the Medical Dictionary for Regulatory Activities dictionary¹⁷. The combination of preferred term levels used to identify these irAE were recently published and made available¹⁷. The rate of each specific irAE in Vigibase over all adverse drug reactions reported in ICI-treated cases including anti-LAG-3 combined with anti-PD-1 to those reported in cases on anti-PD-1 without anti-LAG-3 using multivariate logistic regression adjusted on cancer types, age, sex, period of reporting and concomitant use of anti-CTLA-4. This study was registered on clinicaltrials.gov under NCT05934214.

Statistical Analysis:

Mouse survival was assessed via Kaplan-Meier Survival curves with log rank statistical analysis. Significance of quantification results was tested by Welch’s t test, Student’s t test or Mann-Whitney test using Prism 10.0 (GraphPad Software, San Diego, CA). P value <0.05 was considered as significant. Overlaid bar graphs show mean and error bars show standard error measurement.

Results:

Anti-LAG-3/anti-PD-1 combination therapy is associated with an increased risk of ICI-myocarditis.

Using the international pharmacovigilance database, VigiBase, (accessed in January 2024) we found 130,113 reports of adverse drug reactions associated with anti-PD-1 of which 1,702 were ICI-myocarditis. Among these 130,113 cases, 365 (1.9%) received concomitant anti-LAG-3 therapy and 19,923 (15.3%) received anti-CTLA-4 therapy. Anti-PD-1 therapy was the only ICI used in 109,869 cases (84.4%) and combined with anti-CTLA-4 in 19,879 cases (15.3%). Notably, ICI-myocarditis (and other ICI-induced myotoxicities) was more frequently reported in cases treated with anti-LAG-3 (28/365, 7.7%) than in cases treated with anti-PD-1 without anti-LAG-3 (1,674/129,748, 1.3%; after adjustment for sex, age, cancer type, combination with anti-CTLA-4: adjusted OR=4.0, 95%CI=2.6-5.8, p=8x10⁻¹², Figure S2)^4,18,19. In a multivariate analysis, anti-LAG-3 combination therapy portended a higher risk of myocarditis than anti-CTLA-4 combination therapy (OR=1.7, 95%CI=1.5-1.9, p=1x10⁻¹⁹).

LAG-3/PD-1 KO mice die prematurely with associated cardiac inflammation and arrhythmias

Guided by initial studies performed in the context of anti-tumor immunity, we crossed Lag3−/− and Pdcd1−/− mice to generate Lag3−/−, Pdcd1−/− mice on a C57BL/6 background^20,21. Consistent with previous studies, Lag3−/−, Pdcd1−/− mice demonstrated pre-mature lethality, with most mice dying by 8-10 weeks of age compared to Pdcd1−/− mice which survived (Figure 1A). Given that this model results in global disruption of T-cell signaling with loss of both Pdcd1 and Lag3, we carefully investigated the cause of death in Lag3−/−, Pdcd1−/− mice by harvesting organs of 5-6-week old Lag3−/−, Pdcd1−/− mice and Pdcd1−/− mice including heart, lungs, stomach, kidney, liver and pancreas. While Pdcd1−/− mice had evidence of modest autoimmunity, Lag3−/−, Pdcd1−/− mice experienced severe cardiac inflammation, though varying levels of inflammation were noted in other organs (Figure 1B). There were no significant differences in survival or other cardiac phenotype between male and female Lag3−/−, Pdcd1−/− mice (Figure S3). When scored by a pathologist blinded to mouse genotypes, differences in inflammation of non-cardiac organs appeared unchanged between Lag3−/−, Pdcd1−/− and Pdcd1−/− mice, with exception to a non-significant increase in inflammation within the pancreas and lungs (Figure 1C). To further assess for evidence of multi-organ injury in the Lag3−/−,Pdcd1−/− mice, we performed plasma chemistry analyses in wildtype (WT), Pdcd1−/−, and Lag3−/−, Pdcd1−/− mice. We found a mild increase in alanine aminotransferase (ALT) and creatinine (Cr) in the Lag3−/−, Pdcd1−/− compared to WT mice, but notably not compared to Pdcd1−/− mice. In contrast, Lag3−/−, Pdcd1−/− mice had a robust increase in plasma troponin I levels, indicating extensive cardiac injury with normal levels in both WT and Pdcd1−/− mice (Figure 1D).

Figure 1:

Lag3−/−, Pdcd1−/− mice exhibit pre-mature lethality with cardiac-enriched inflammation and arrhythmias. A) Survival curves of Lag3−/−, Pdcd1−/− mice (n=14, 7 males/7 females) as compared to Pdcd1−/− (n=30, 15 males/15 females) mice, log-rank test. B) Representative Histological Images of Lag3−/−, Pdcd1−/− and Pdcd1−/− mice C) Histological grading of Lag3−/−, Pdcd1−/− (n=10, 5 males/5 females) and Pdcd1−/− mice (n=6, 3 males/3 females), Welch’s t-test. D) Plasma chemistry analysis of Lag3−/−, Pdcd1−/− (troponin: n=16, 8 males/8 females; all others: n=12, 6 males/6 females), Pdcd1−/− (troponin: n=14, 7 males/7 females, all others: n= 7, 4 males/3 females), and WT mice (troponin: n=4, 2 males/2 females, all others: n= 7, 4 males/3 females), Welch’s t-test. E) Representative arrhythmias seen on EKG of Lag3−/−, Pdcd1−/− mice. F) Arrhythmia burden (AV block, ventricular escape, PVCs) over time in Lag3−/−, Pdcd1−/− and Pdcd1−/− mice.

Given that the phenotype in Lag3−/−, Pdcd1−/− mice appeared to be largely cardiac injury, we performed detailed cardiac phenotyping. Electrocardiography (ECG) showed significant rhythm disturbances including atrioventricular (AV) block and ventricular arrhythmias in the Lag3−/−, Pdcd1−/− mice. Weekly ECG measurements demonstrated an increasing prevalence of arrhythmias prior to death (Figure 1E) as compared to Pdcd1−/− mice which remained in normal sinus rhythm (Figure 1F). To determine left ventricular systolic function and volume, we performed echocardiography on the Lag3−/−, Pdcd1−/− mice. Similar to other preclinical models of ICI-myocarditis and human data, we did not observe changes in left ventricular ejection fraction (LVEF) or left ventricular end diastolic volume (LVEDV) (Figure S4)^8,22.

LAG-3/PD-1 KO mice have robust expansion of activated clonal cardiac T-cells

Given the severe cardiac phenotype in Lag3−/−, Pdcd1−/− mice, we next sought to define the immune cell populations within the heart. Flow cytometry on cardiac isolates revealed both increased T-cell count along with T-cell enrichment within cardiac immune population of Lag3−/−, Pdcd1−/− mice (Figure 2A). Specifically, we observed an increased relative frequency of CD8+ T-cells, without significant changes in relative abundance of CD4+ T-cells (Figure S5).

Figure 2:

Lag3−/−, Pdcd1−/− mice have dynamic changes in cardiac immune populations with robust expansion of infiltrating T-cells. A) Flow cytometry for relative frequency of CD3+ T-cells of 6 week old Lag3−/−, Pdcd1−/− mice (n=14, 7 males/7 females), Pdcd1−/− mice (n=8, 4 males/4 females) and WT (n=3, 2 males/1 female) and T-cell count of Lag3−/−, Pdcd1−/− mice (n=5, 3 males/2 females) and WT (n=4, 2 males/2 females) associated controls, Welch’s t-test. B) UMAP of cardiac CD45+ sorted scRNA-seq, split by genotype (Lag3−/−, Pdcd1−/− mice n=4, 2 males/2 females; WT mice n=6, 3 males/3 females). C) Heatmap of markers defining each annotated cluster. D) Feature Plot representing activated T-cells markers. E) UMAP of cardiac T-cell clones in WT and Lag3−/−, Pdcd1−/− mice. F) Distribution of clone size in WT and Lag3−/−, Pdcd1−/− mice including sole (non-clonal T-cells).

To further phenotype the immune cell populations, scRNA-seq was performed on sorted cardiac CD45+ immune cells from Lag3−/−, Pdcd1−/− and WT mice, which allowed for the the characterization of immune cell populations and cluster cells into (1) dendritic cells, (2) dividing T-cells, (3) natural killer (NK) cells, (4) B-cells, (5) naïve T-cells, (6) macrophages/monocytes and (7) activated T-cells (Figure 2B). Marker genes were used to define both the immune cell type (e.g., T-cell) and their specific cell state (e.g., Mki67 expression within the dividing T-cell population). Within our Lag3−/−, Pdcd1−/− mice, we specifically observed upregulation of activated T-cells, dividing T-cells, and macrophages/monocytes. Markers of the activated T-cell cluster included Cd8a and canonical cytotoxicity genes such as Nkg7, both the chemokine Ccl5 and chemokine receptor Cxcr6, co-stimulatory genes (Icos), and markers of T-cell exhaustion (Ctla4, Tigit) (Figure 2C–D). scRNA-seq populations between all 4 Lag3−/−, Pdcd1−/− mice were comparable (Figure S6). Re-clustering of cardiac T-cells demonstrated markedly different cardiac T-cell subtypes between WT and Lag3−/−,Pdcd1−/− (Figure S7A–B). Upon further analysis of gene expression profiles, WT cardiac T-cells expressed markers of naïve/quiescent T-cells (Tcf7, Il7r, Lef1, Ccr7) in comparison to Lag3−/−,Pdcd1−/− which showed a shift toward proinflammatory CD4 (Ccl3, Ccl4, and Icos) and effector memory CD8 (marked by Prf1, Nkg7, and Ccl5) T-cell populations (Supplement Figure S7C–D).

We performed simultaneous TCR sequencing to determine clonality of cardiac T-cells within the Lag3−/−, Pdcd1−/− mice. Interestingly, 41% of Lag3−/−, Pdcd1−/− cardiac T-cells were clonal (defined as >2 cells with same TCR, Table S1). Further, there was a marked expansion of clones in the Lag3−/−, Pdcd1−/− as compared to WT mice (Figure 2E). Indeed, over 20% of cardiac T-cells in the Lag3−/−, Pdcd1−/− mice were noted to be either “Large” or “Hyperexpanded” clones with >10 T-cells per clone (Figure 2F).

Cardiac CXCR6+ T-cells dynamically upregulate in mouse and human ICI-myocarditis tissues

To better understand the role of recruitment and positioning of T-cells in the cardiac environment we focused on CXCR6, which was both a marker of activated cardiac T-cells and robustly expanded in the Lag3−/−, Pdcd1−/− mice compared to the WT controls (Figure 3A). To determine if Cxcr6 expression was also dynamically upregulated in other mouse models of ICI myocarditis, we analyzed Cxcr6 expression in published datasets of Pdcd1−/− and Ctla4+/−,Pdcd1−/− mice. While there was minimal Cxcr6 expression in WT hearts, there was an increase in Cxcr6 expression in the heart of both Pdcd1−/− and Ctla4+/−,Pdcd1−/− mice (Figure S8). These data suggest that CXCR6 T-cells are a common feature of ICI myocarditis models. While the role of CXCR3 is an established mechanism of recruitment in irAEs such as myocarditis – the role of CXCR6 T-cells in myocarditis has been understudied, though both CXCR3 and CXCR6 have been implicated in anti-tumor immunity^{13–15,23–29}. To confirm the cardiac CD45+ scRNA-seq findings and determine whether similar changes occur in other inflamed organs in the Lag3−/−, Pdcd1−/− mice – we performed flow cytometry on immune cells from the pancreas, liver, and blood. While there was a marked expansion of CXCR6+ T-cells in comparison to total T-cell population in the heart, there was no significant differences in other assayed tissues – suggesting that a reliance on CXCR6+ T-cells may be unique to the heart in this model (Figure 3B). To assess for human relevance, we probed a publicly accessible repository of cardiac scRNA-seq data on 15 ICI treated patients with myocarditis and 6 healthy human hearts^15,30. Though the patients included in this study were on a variety of different ICI-regimens, cardiac CXCR6 expression was robustly increased in ICI-myocarditis patient samples compared to controls (Figure 3C–D)¹⁵. These findings suggest that CXCR6+ T-cells are dynamically upregulated in myocarditis across different ICI therapies.

Figure 3.

CXCR6+ T-cells are upregulated in mouse models and human ICI-myocarditis. A) Feature Plot of Cxcr6 expression in WT and Lag3−/−, Pdcd1−/− mice. B) Flow cytometry of CXCR6+ T-cells in different inflamed tissues in Lag3−/−, Pdcd1−/− mice (blood: n=4 (2 males/2 females), pancreas: n=5 (3 males/2 females), liver: n=5 (3 males/2 females), heart: n=11 (5 males/6 females)) and WT controls (blood: n=4 (2 males/2 females), pancreas: n=4 (2 males/2 females), liver: n=4 (2 males/2 females), heart: n=7 (3 males/4 females)). Welch’s t-test . C) UMAP of publicly available human ICI-myocarditis (n=15) and healthy controls (n=6) cardiac scRNA-seq. UMAP obtained from PMID 39506125. D) Feature Plot of CXCR6 split by condition. E) Subclusters of Cxcr6 expressing T-cells defines 5 unique clusters. F) Heatmap of Cxcr6 T-cell subcluster markers. G) Volcano Plot of markers Hyperexpanded T-cell clones as compared to all cells in Activated T-cell cluster.

To characterize the different populations in the CXCR6+ T-cells in Lag3−/−, Pdcd1−/− mice – we performed downstream clustering on the Cxcr6 expressing T-cells, defined as Cxcr6 expression >1. We defined five groups of T-cells (Figure 3E–F) including Gzmk expressing T-cells, Gzmb expressing T-cells, Gm29617 expressing T-cells, Cd4 T-cells, and dividing T-cells (Pclaf, Birc5). The dichotomy of Gzmk and Gzmb CD8+ T-cells has been observed in other T-cell driven inflammatory settings – however the functional relevance of the two subsets is still an area of active investigation³¹. We observed that Cxcr6 expression overlapped with clonal T-cells graphically – and therefore sought to characterize the marker genes of hyperexpanded T-cell clones as compared to all cells within the activated T-cell cluster. Indeed, Cxcr6, along with Prf1 and Nkg7 were markers of hyperexpanded clonal T-cells (Figure 3G).

CXCL16+ cardiac macrophages expand and position CXCR6+ T-cells in LAG-3/PD-1 myocarditis

As both cardiac T-cells and macrophages/monocytes were particularly expanded in the Lag3−/−, Pdcd1−/− mice, we then turned our attention to defining our macrophage/monocytes populations. Re-clustering the cardiac macrophage/monocytes population resulted in four unique clusters (Figure 4A). We and others have previously reported on the role of CXCL9/CXCL10 in the recruitment of CXCR3+ T-cells in ICI-myocarditis. Indeed, we were able to identify infiltrating monocytes that expressed both Cxcl10 and interferon inducible genes (Isg15 Monocytes). There was also a distinct cluster of cardiac macrophages expressing Cxcl16. These macrophages had robust expression of antigen presentation genes (H2-Aa, H2-Ab1, H2-Eb1) and complement (C1qa, C1qb, C1qc) (Figure 4B). Pathway enrichment between these two clusters demonstrated cytokine signaling upregulated in Isg15 monocytes and MHC Class II antigen presentation in cardiac macrophages (Figure S9). As mentioned earlier, CXCL16 is the only known ligand for CXCR6. While there were modest levels of Cxcl16 in WT cardiac macrophages, there was a notable increase in expression in the Lag3−/−, Pdcd1−/− mice (Figure 4C–D). To validate the gene expression findings at the protein level, flow cytometry was performed which similarly demonstrated an expansion of CXCL16+ macrophages (CD11b+, F4/80+) in Lag3−/−, Pdcd1−/− hearts (Figure 4E)

Figure 4.

CXCL16, the cognate ligand of CXCR6, is dynamically increased in cardiac macrophages. A) UMAP of macrophage/monocyte subclusters demonstrating distinct populations. B) Heatmap of the defining markers of each macrophage/monocyte subcluster. C) Feature Plot of Cxcl16 expression in WT and Lag3−/−, Pdcd1−/− mouse hearts. D) Violin Plot, split by cluster and sample, of Cxcl16 demonstrating increased Cxcl16 expression in cardiac macrophages. E) Flow Cytometry on relative CXCL16+ Macrophage population (gated on Live/CD45+/CD11b+/F4-80+) between Lag3−/−,Pdcd1−/− (n=3, 2 males/1 female) and WT (n=3, 2 males/1 female), Welch’s t-test. F) Chemotaxis Assay of sorted T-cells from Lag3−/−, Pdcd1−/− blood (n=3) and heart (n=3) to recombinant mouse CXCL16 (rmCXCL16) and positive (10% FBS) and negative (RPMI only) controls, student t-test.

To test the relevance of CXCL16 within the cardiac T-cell population, T-cells from Lag3−/−, Pdcd1−/− mice heart and blood were isolated and a CXCL16 chemoattraction assay was performed. While peripheral blood T-cells were chemoattracted to FBS as a positive control – they did not appear to respond to recombinant mouse CXCL16, in line with our observation of few CXCR6+ T-cells in the blood. In contrast, cardiac T-cells were significantly attracted to CXCL16 (Figure 4F). This suggests that CXCL16 production from cardiac macrophages likely plays a role in the positioning of cardiac T-cells – presumably through the CXCL16-CXCR6 axis.

CXCR6+ T-cells are necessary for LAG-3/PD-1 myocarditis

To this point, we demonstrated that LAG-3/PD-1 myocarditis is associated with cardiac-specific enrichment of CXCR6+ T-cells recruited, in part, through CXCL16 produced by cardiac macrophages. We hypothesized that these CXCR6+ T-cells were critical for induction of myocarditis. To test this hypothesis, we treated Lag3−/−, Pdcd1−/− mice with anti-CXCR6 antibody or isotype control from 3 to 10 weeks of age. Anti-CXCR6 treatment rescued the pre-mature lethality phenotype with 100% of the treated mice living until experimental endpoint while all of the isotype control mice died (Figure 5A). Histology revealed a marked reduction in inflammation within hearts of anti-CXCR6 treated mice (Figure 5B–C). This was associated with a substantial reduction in the incidence of arrhythmias with only one mouse in the anti-CXCR6 treated group developing an arrhythmia compared to 100% arrhythmia burden among isotype-treated controls (Figure 5D–E). Flow cytometry of anti-CXCR6 treated animals demonstrated a decrease in the total frequency of T-cells and CXCR6+ T-cells (Figure 5F). Notably, while there was a decrease in CXCR3+ CXCR6+ T-cells, there was no difference in relative frequency of CXCR3+ CXCR6- T-cells with anti-CXCR6 treatment (Figure S10). Taken together, anti-CXCR6 treatment effectively rescued the spontaneous myocarditis phenotype in the Lag3−/−, Pdcd1−/− model establishing the necessity of CXCR6+ T-cells in the propagation of LAG-3/PD-1 myocarditis.

Figure 5:

CXCR6+ T-cells are necessary for Lag3−/−, Pdcd1−/− myocarditis. A) Survival curves of Lag3−/−, Pdcd1−/− treated with 250ug twice weekly isotype control (n=8, 4 males/4 females) and 250ug twice weekly anti-CXCR6 (n=8, 4 males/4 females), log-rank test. B) Representative histology of isotype control and anti-CXCR6 treated mice (Scale Bars: 500um and 100um). C) Myocarditis Lesional area in isotype (n=8) and anti-CXCR6 (n=8) treated mice, Welch’s t-test D) Representative EKGs from isotype and anti-CXCR6 treated mice. E) Burden of arrhythmias based on weekly EKGs of isotype and anti-CXCR6 treated mice F) Flow Cytometry of Relative T-cell and CXCR6+ T-cell populations in untreated (n=6, 3 males/3 females) and anti-CXCR6 treated (n=6, 3 males/3 females) mice, Welch’s t-test.

Given the growing body of literature suggesting a role for the CXCL9/10-CXCR3 axis in ICI-myocarditis, we investigated the differences between CXCR3+ T-cells relative to CXCR6+ T-cells. Flow cytometry of CXCR3+ T-cells in our Lag3−/−, Pdcd1−/− mice demonstrated an expansion within the T-cell population in the blood, pancreas, and heart (Figure S11A), which ultimately made up approximately one-third of cardiac T-cells. However, in contrast to the clonal CXCR6 T cells, CXCR3 was not a marker of hyperexpanded clones in the scRNA/TCR-Seq (Figure S11B). Interestingly, similar to anti-CXCR6, anti-CXCR3 treatment rescued Lag3−/−, Pdcd1−/− mice from pre-mature lethality (Figure S11C). However, while survival was prolonged, a larger portion of anti-CXCR3 treated mice (4/8) developed arrhythmias (Figure S11D). Histology from anti-CXCR3 treated mice showed improvement in myocarditis by lesional area as compared to isotype controls, and lesional area of myocarditis between anti-CXCR3 and anti-CXCR6 treated mice was not significantly different (p=0.11) (Figure S11E–F).

Given analogous signals of Cxcr6 expression in cardiac T-cells in Ctla4+/−, Pdcd1−/− mice, we hypothesized that treatment with anti-CXCR6 would similarly target cardiac T-cells in this model. To test this, Ctla4+/−,Pdcd1−/− mice were treated with anti-CXCR6 or vehicle for three weeks and assessed hearts by flow cytometry. We identified a significant, specific decrease in relative frequency of cardiac CD8+ T-cells compared to the total immune population, not seen in total T-cell or CD4+ T-cells (Figure S12A). Although there was a notable trend, there was a no significant difference in the number of cardiac CD8 T-cell by count in the Ctla4+/−,Pdcd1−/− model treated with anti-CXCR6 likely due to the established heterogeneity of this model (Figure S12B). Interestingly, effector memory T-cells defined as CD44+/CD64L- were significantly lower in anti-CXCR6 treated Ctla4+/−,Pdcd1−/− animals (Figure S12C). We have previously shown that CD8+ T-cells were necessary for myocarditis in this model⁸. These data suggest that anti-CXCR6 treatment decreases activated effector CD8+ T-cells in a preclinical ICI model outside of LAG3/PD1 myocarditis.

Discussion:

Herein, we have defined a new model of ICI-myocarditis to approximate combinatorial treatment of LAG-3 and PD-1 using genetic knockout mice. Lag3−/−, Pdcd1−/− mice manifest cardiac enriched inflammation and pre-mature lethality due to myocarditis. ScRNA-seq of the heart immune cells demonstrated a dynamic increase in Cxcr6 expressing T-cells that was not seen in other organs, which mirrored clinical ICI myocarditis data. CXCR6’s cognate ligand, Cxcl16, was upregulated on a specific group of cardiac macrophages. Most notably, depletion of CXCR6+ T-cells rescued the mice from myocarditis, suggesting that cardiac CXCR6+ T-cells are key facilitators of ICI-myocarditis.

With the emergence of ICIs (especially combinatorial therapy) as a common first line cancer regimen, we anticipate a growing burden of ICI-myocarditis³². As such, it is imperative to generate pre-clinical models of ICI-myocarditis to elucidate molecular mechanisms that can aid in the diagnosis and treatment of this highly morbid adverse event. Moreover, immunotherapy is a rapidly advancing field with new ICIs and combinations of anti-LAG-3/PD-1 are poised to become a mainstay of ICI therapy in the coming years due to its favorable irAE profile outside of ICI-myocarditis⁵. Therefore, having preclinical models that effectively recapitulate cardiotoxicity from anti-LAG-3/PD-1 therapy is paramount to delineating the disease pathogenesis. Our results from Lag3−/−, Pdcd1−/− mice highlight their utility as a model of ICI-myocarditis.

One of the major challenges for studying ICI-myocarditis in a pre-clinical model is the more subtle phenotypes seen in some pharmacological treatment models. In contrast, genetic deletion of LAG-3 and PD-1 results in 100% pre-mature lethality due to myocarditis characterized by activated, clonal T-cells within 10 weeks of age. As such, it permits careful characterization of immune cell subsets, their necessity, and their therapeutic relevance more efficiently and reliably. However, we also recognize the limitations posed by global genetic deletion models including unique mechanisms of injury that may be due to alternations in T-cell development in the absence of LAG-3 and PD-1 immune checkpoints. Therefore, it is critical to perform orthogonal validation of pre-clinical models in human tissue samples to determine if our pre-clinical model effectively recapitulates the clinical pathogenesis. For example, we do not observe concomitant myositis in the Lag3−/−, Pdcd1−/− mice which has commonly been reported in patients with ICI-myocarditis.

While our mouse model of ICI-myocarditis is focused on LAG-3 and PD-1, the finding of robust upregulation of CXCR6+ on cardiac T-cells appears to be a common thread among pre-clinical ICI-myocarditis models as well as patients with ICI-myocarditis as demonstrated by our data in the Ctla4+/−,Pdcd1−/− model and publicly accessed human data⁸. In particular, our results identify the cardiac enrichment of a heterogenous group of CXCR6+ T-cells in Lag3−/−, Pdcd1−/− mice. These CXCR6+ T-cell clusters included distinct expression of Granzyme K, Granzyme B, Helper CD4+ T-cells, and actively dividing T-cells, indicating that in the context of our preclinical model of ICI-myocarditis, CXCR6 is not an exclusive marker of a unique T-cell population.

The mechanisms that drive higher rates of myocarditis in anti-LAG-3/anti-PD-1 treated patients remain unclear but may relate to LAG-3 activity on CD4+ T-cells. LAG-3 is thought to be expressed canonically on CD4+ T-cells and similarly bind peptide-MHC class II complexes³³. However, further studies are needed to clearly elucidate the role of CD4+ T-cells in anti-LAG-3 associated myocarditis.

We noted significant bradyarrhythmias in our Lag3−/−,Pdcd1−/− mice that included both atrioventricular block and ventricular escape rhythms. ICI-myocarditis is particularly arrhythmogenic – 67% of patients present with conduction disease, and 17% develop third-degree atrioventricular block³⁴. In our preclinical model, spontaneous arrhythmias proved to be a reliable way to track our myocarditis phenotype noninvasively. We find the particular presence of atrioventricular blocks intriguing. Notably, cardiac macrophage populations have been implicated in AV node signaling through the gap junction protein Connexin 43³⁵. While the Connexin 43 was not altered in the macrophages in our affected mice, future investigations into how T-cell enriched inflammation drive bradyarrhythmias may yield new insight the disease and therapeutic strategies.

Unlike other chemokine-chemokine receptor pathways with multiple ligand/receptors, the sole chemokine for CXCR6 is CXCL16. As such, we identified CXCL16 upregulation, at both the RNA and protein level, in cardiac macrophages in our preclinical model. CXCL16, unlike other CXCL chemokines can exist in a membrane-bound state and has previously been described to mediate the direct interactions between CXCR6+ T-cells and CXCL16+ myeloid cells in other microenvironments. Interestingly, CXCL16 appeared to segregate from CXCL10 expression in cardiac monocytes – where we and others have shown the CXCL9/10-CXCR3 axis as critical in the recruitment of peripheral T-cells to the heart in ICI-myocarditis^13,14. From the data generated in this paper, CXCR6+ T-cells represent a more clonal population of T-cells as compared to CXCR3+. Anti-CXCR3 treatment attenuated the myocarditis phenotype, though a portion of these mice still developed arrhythmias and mild myocarditis.

Importantly, our data demonstrated an important functional role for CXCR6+ T cells in ICI myocarditis. Treatment with anti-CXCR6 effectively rescued the myocarditis phenotype in several orthogonal ways (histology, arrhythmias and most critically survival). These data strongly indicate that CXCR6+ T-cells are critical drivers of our preclinical ICI-myocarditis model. This study provides the biologic plausibility of anti-CXCR6 treatment as a potential therapeutic for ICI-myocarditis. As CXCR6 has been implicated in anti-tumor immunity, further investigations of anti-CXCR6 in preclinical models that combine both ICI-myocarditis and tumor progression are necessary to delineate whether anti-CXCR6 treatment will attenuate both the cardiac inflammation without sacrificing anti-tumor immunity.

CXCR6 is best known as a marker of tissue resident memory T-cells, and while our preclinical model demonstrates fulminant, active inflammation – a population of these CXCR6+ T-cells may evolve into memory T-cells should these mice not succumb to their fatal disease. Recent investigations have demonstrated that existing tissue resident memory T-cells, in the setting of previous cardiac injury, play a role in ICI-myocarditis³⁶. Based on the lack of expansion of CXCR6+ T-cells in circulating T-cells, we suspect that CXCL9/10-CXCR3 axis may be involved in recruitment of peripheral T-cells with subsequent expression of CXCR6 in the heart for positioning within the cardiac environment and establishing tissue residency. However, these hypotheses require further experimentation. Lastly, another outstanding question is the necessity of CXCR6 in the pathogenic role of CXCR6+ T-cells. In the tumor microenvironment, CXCL16+ dendritic cells recruit CXCR6+ T-cells through the CXCL16-CXCR6 axis in order to trans-present IL-15 and induce pro-survival signaling²⁶. As such, further investigation into the loss of CXCR6 on cardiac T-cells and how this may affect cardiac T-cell retention, T-cell survival and overall cardiac phenotype is warranted.

In summary, we have characterized a mouse model of Lag3 and Pdcd1 deletion that effectively recapitulates ICI-myocarditis and serves as a robust tool to study the immunological drivers of the disease. Like other pre-clinical models and clinical data, we observed clonal, T-cell predominant cardiac infiltration with specific upregulation of the chemokine receptor CXCR6. The cognate ligand of CXCR6, CXCL16, is upregulated in a distinct population of cardiac macrophages. Loss of CXCR6+ T-cells through anti-CXCR6 treatment rescues the myocarditis phenotype. Overall, our study highlights cardiac CXCR6+ T-cells as key mediators in our pre-clinical model of ICI-myocarditis.

Clinical Perspective:

What is new?

• Anti-LAG-3/PD-1 combination therapy significantly increases myocarditis risk (compared to anti-PD-1 therapy alone).

• We have generated a novel pre-clinical model of ICI-myocarditis mimicking anti-LAG-3/PD-1 combination therapy.

• CXCR6+ T-cells are critical mediators and are necessary for development of myocarditis.

What are the clinical implications?

• CXCR6+ T-cells represent a central driver of ICI-myocarditis, suggesting a novel therapeutic target.

• Anti-CXCR6 therapy may provide a cardiac-specific pathway for targeted immunomodulation.

Non-standard Abbreviations and Acronyms

CTLA-4

Cytotoxic T-lymphocyte-associated protein 4

CXCL9/10

C-X-C motif chemokine ligands 9 and 10

CXCL16

C-X-C motif chemokine ligand 16

CXCR3

C-X-C motif chemokine receptor 3

CXCR6

C-X-C motif chemokine receptor 6

ICI

Immune Checkpoint Inhibitors

LAG-3

Lymphocyte-activation gene 3

PD-1

Programmed cell death protein 1

TCR

T-cell receptor

Data Availability

The single cell raw expression matrices and code that support the findings of the study are available on Zenodo (10.5281/zenodo.17050583). Published mouse cardiac scRNA-sequencing (RNA-seq) data from Pdcd1−/− and Ctla4+/−,Pdcd1−/− mice that were used to support the findings of this study are available on the Gene Expression Omnibus (GSE227437, GSE213486)^8,14. Published scRNA-seq data from patients with ICI-associated myocarditis are available in the Gene Expression Omnibus (GSE228597)¹⁵. For all previously existing datasets, the published analyzed and annotated data was used in this manuscript.