Immune Responses in Checkpoint Myocarditis Across Heart, Blood, and Tumor

Abstract

Immune checkpoint inhibitors (ICIs) are widely used anti-cancer therapies¹ that can cause morbid and potentially fatal immune-related adverse events (irAEs) such as immune-related myocarditis (irMyocarditis)^2–5. The pathogenesis of irMyocarditis and its relationship to anti-tumor immunity remain poorly understood. We sought to define immune responses in heart, tumor, and blood in patients with irMyocarditis by leveraging single-cell RNA-sequencing (scRNA-seq) coupled with T cell receptor (TCR) sequencing, microscopy, and proteomics analysis of 28 irMyocarditis patients and 41 controls. Analysis of 84,576 cardiac cells by scRNA-seq combined with multiplexed microscopy demonstrated increased frequencies and colocalization of cytotoxic T cells, conventional dendritic cells (cDCs), and inflammatory fibroblasts in irMyocarditis heart tissue. Analysis of 366,066 blood cells revealed decreased frequencies of plasmacytoid dendritic cells, cDCs, and B lineage cells but an increased frequency of other mononuclear phagocytes in irMyocarditis. Fifty-two heart-expanded TCR clones from eight patients did not recognize the putative cardiac autoantigens α-myosin, troponin I, or troponin T. Additionally, TCRs enriched in heart tissue were largely non-overlapping with those enriched in paired tumor tissue. The presence of heart-expanded TCRs in a cycling blood CD8T cell population associated with fatal irMyocarditis case status. Collectively, these findings highlight critical biology driving irMyocarditis and nominate putative biomarkers.

Introduction

Immune checkpoint inhibitors (ICIs) improve cancer outcomes for a wide range of tumor types¹ but can cause potentially dangerous immune-related adverse events (irAEs) ². ICI-related myocarditis (irMyocarditis) occurs in 0.3–1.7% of ICI recipients^3–6 and carries a mortality rate of 20–50%, the highest of any irAE and approximately 10-fold higher than myocarditis from other causes^7–9.

The molecular underpinnings of irMyocarditis remain poorly understood, but irMyocarditis is characterized by an infiltration of both CD4⁺ and CD8⁺ T cells as well as inflammatory macrophages^4,5,10–13. Sharing of T-cell clones from paired myocardium and tumor of an irMyocarditis patient suggest the possibility of a common antigen driving T cells responses in these tissues¹⁰. Bulk RNA sequencing analysis of irMyocarditis heart tissue demonstrated upregulation of many interferon-stimulated genes (ISGs)¹⁴, nominating a potential pathway contributing to pathogenesis. Furthermore, increases in circulating CD8⁺ T cell populations expressing high levels of cytotoxic genes and inflammatory chemokines have been associated with irMyocarditis¹⁵. Studies using Pdcd1^–/–Ctla4^+/– mice, a genetic model of irMyocarditis, demonstrated that CD8⁺ T cells were necessary to elicit myocarditis, and a fraction of these T cells recognize the cardiac protein α-myosin^16,17. T-cell clones specific for α-myosin have been found in the blood of healthy controls and the blood and hearts of irMyocarditis patients¹⁶, but their contribution to the clinical pathogenesis remains undefined. In this study, we used heart, blood, and tumor samples from human irMyocarditis patients to investigate the cellular and transcriptional mediators underlying irMyocarditis pathogenesis.

Results

Study design and sample collection

Heart, blood, and tumor specimens were collected from cancer patients at Massachusetts General Hospital (MGH) with irMyocarditis and from ICI-treated controls without irMyocarditis (Fig 1a-c; Supplementary Table 1; Methods). Heart tissue for scRNA-seq was collected during endomyocardial biopsy procedures on patients with suspected irMyocarditis (n=15; 13 irMyocarditis, 2 controls) or during early post-mortem autopsies (n=4; 3 irMyocarditis, 1 control)(Fig 1a-b, Supplementary Tables 1-2). Twelve irMyocarditis samples were collected prior to therapeutic corticosteroid initiation (“pre-steroid”). irMyocarditis patients in the scRNA-seq cohort had a median age of 72 years, were 80% male, and represented eight primary tumor types. All had received a PD-1/PD-L1 inhibitor, of which 7 (47%) were treated with dual PD-1/CTLA-4 inhibition. Matched heart, tumor, and tumor-adjacent normal tissues were collected, when possible, from irMyocarditis (n=7) and ICI-treated control (n=3) autopsy cases for immunohistochemistry, multiplexed microscopy, and bulk TCR-β sequencing experiments (Fig 1a, 1c, Supplementary Table 1).

Fig 1. Enrichment of multiple intracardiac and blood cell populations in irMyocarditis.

a, Summary of patient cohort. Control blood and serum are not shown (see Supplementary Table 1). b, Top: bar plots indicate the number of cells from each heart sample. Middle: cell lineage composition from each sample. Bottom: sample contribution to different analyses. “A” and “B” distinguish samples from donors with two myocardial samples. c, Summary of sample utilization. d,h, UMAP embedding displaying indicated cell populations from (d) heart tissue and (h) PBMC scRNA-seq datasets. “h-” and “b-” distinguish populations recovered from heart tissue and blood, respectively. e,i, Abundance analysis comparing intramyocardial cell population frequencies from (e) pre-steroid irMyocarditis heart samples (n=12, red) versus controls (n=8, blue) and (i) pre-steroid irMyocarditis PBMC samples (n=17, red) versus ICI-treated controls (n=28, blue). Left: dots represent logistic regression odds-ratios. Error bars represent 95% confidence intervals. Unadjusted P values are shown (likelihood ratio test). Bold font and colored bars indicate FDR<0.1. Right: box plots show median (line) and IQR, with whiskers no more than 1.5x IQR. Each dot represents one patient. f, j, Select GSEA results for (f) cardiac and (j) PBMC cells populations, color-coded by NES of the gene set in irMyocarditis cases versus controls. g, Hematoxylin and eosin staining of cardiac tissues obtained from two patients at autopsy showing intracardiac metastases (left column) and inflammation remote from metastatic foci (right column). k, Circulating frequency of MNPs (y-axis) versus serum troponin T (x-axis) from pre-steroid irMyocarditis PBMC samples (n=16); p=0.002, FDR<0.1, linear regression. Abbreviations: Allo, allograft rejection; AP, antigen presentation; CAMS, cell adhesion molecules; DNA, DNA synthesis; GEX, gene expression; FBC, feature bar code; H&E, hematoxylin and eosin; IFNG, interferon-γ signaling; ISH-IF, RNA in situ hybridization-immunofluorescence; HM, Hallmark; K, Kyoto Encyclopedia of Genes and Genomes; NES, normalized enrichment score; Viral Myo, viral myocarditis.

Peripheral blood mononuclear cells (PBMCs) were collected from 30 ICI-treated control cancer patients without irAEs and 25 irMyocarditis patients (13 with paired heart scRNA-seq data) (Fig 1a, 1c; Supplementary Tables 1, 3-4). Control and irMyocarditis cohort samples were collected at similar times after ICI initiation (control median 56 days, range 14–497; irMyocarditis median 62 days, range 17–252). For the irMyocarditis patient cohort, data was generated from 55 PBMC specimens across the following timepoints: prior to ICI treatment (“pre-ICI”; n=7); after ICI initiation but before irMyocarditis onset (“on-ICI”; n=4); prior to corticosteroids (“pre-steroid”; n=17); and after corticosteroids and in some cases other immunosuppressive medications (“post-steroid”; n=27). Pre-steroid serum was available for 16 irMyocarditis patients and 10 controls (Supplementary Table 1).

Heart cell populations and gene programs

ScRNA-seq analysis of our heart samples yielded transcriptomes from 33,145 single cells. These data were combined with publicly available control heart scRNA-seq data from donors without cancer and not receiving ICI therapy (n=6; 51,431 cells)(Fig 1c)¹⁸ to create an integrated heart dataset of 84,576 single cells. Throughout the text, immune cell subsets found in heart tissue and blood are denoted by subset names that begin with “h-” for heart and “b-” for blood; subsets are named according to their lineage and distinguishing genes. Low-resolution clustering of this heart data yielded 10 cell lineages: T and NK cells (h-T/NK), conventional dendritic cells (h-cDCs), plasmacytoid DCs (h-pDCs), other mononuclear phagocytes (h-MNPs), B cells and plasmablasts (h-B and plasma), endothelial cells, mural cells, fibroblasts, cardiomyocytes, and neural cells (Fig 1d; Extended Data Fig 1a; Supplementary Table 5-6). Samples from our dataset and the public heart atlas¹⁸ contributed to each lineage (Extended Data Fig 1b). Cellular composition of the heart was broadly consistent across different tumor histologies (Extended Data Fig 1c).

Cell subset abundance and gene expression analyses compared pre-steroid irMyocarditis heart samples (n=12) to a control cohort comprised of samples from ICI-treated donors without irMyocarditis (n=2) and heart atlas samples (n=6)¹⁸ (Fig 1b). h-T/NK cells were significantly enriched in irMyocarditis samples (false discovery rate [FDR]=0.007) (Fig 1e, Supplementary Table 7), and trends towards increased h-MNPs, h-cDCs, and fibroblasts were observed. Gene set enrichment analysis (GSEA) performed at the cell lineage level showed significant activation (FDR<0.1) of programs associated with DNA replication and cell adhesion in both the h-T/NK and fibroblast populations (Fig 1f, Supplementary Table 8). Gene sets associated with antigen processing, allograft rejection, cell adhesion, viral myocarditis, and interferon responses were upregulated in multiple non-immune cell subsets as well as h-MNPs. These results implicate both immune and non-immune cells in the pathophysiology of irMyocarditis.

Intracardiac metastases in irMyocarditis

Two irMyocarditis patients and one control patient (SIC_182; melanoma) had evidence of viable tumor in the myocardium at autopsy. Inflammation was seen both associated with and remote from tumor deposits in both irMyocarditis cases by hematoxylin and eosin (H&E) staining: SIC_136 (melanoma) had a solitary cardiac lesion, while SIC_232 (renal cell carcinoma) had diffuse tumor deposits (Fig 1g). A third irMyocarditis patient (SIC_171; melanoma) had received prior targeted (non-ICI) therapy and had evidence of melanin-laden cardiac macrophages on endomyocardial biopsy, suggesting prior intracardiac melanoma (data not shown). Notably, viable tumor cells were not detected in the scRNA-seq data. To assess whether samples with intracardiac metastases affected our findings, we repeated the cell population abundance and gene expression analyses after excluding the scRNA-seq data from patients with evidence of cardiac metastases (SIC_171, SIC_232, and SIC_182). The results were highly concordant with those of the full cohort (Extended Data Fig 1d-e; Supplementary Table 7, 9-10).

Circulating protein analysis

To assess changes in circulating proteins in irMyocarditis, we compared serum concentrations of 71 proteins in pre-steroid irMyocarditis samples (n=16) and ICI-treated controls (n=10). Levels of T-cell activating proteins IL-12p40 (FDR=0.076), IL-15 (FDR=0.076), and IL-27 (FDR=0.076) were increased, as well as CXCL13 (FDR=0.076), which has been associated with irAEs and anti-tumor responses after ICI therapy (Extended Data Fig 1f-g, Supplementary Table 11)^19,20. Trends towards increased CXCL9, CXCL10, IFN-γ, IL-2, IL-18, and TNF-α levels were observed. Excluding a donor with evidence of prior cardiac metastases (SIC_171) did not substantially affect the irMyocarditis versus control comparisons (Extended Data Fig 1h, Supplementary Table 11).

Blood cell populations and gene programs

Cellular indexing of transcriptomes and epitopes sequencing (CITE-Seq) was combined with TCR-seq to profile PBMCs from irMyocarditis patients (n=25) and ICI-treated controls (n=30) (Supplementary Table 1, 12). 366,066 cells passed quality control (QC) filters, and unsupervised clustering using only transcriptional data identified all expected blood immune cell lineages (Fig 1h, Extended Data Fig 1i; Supplementary Tables 6, 13). To assess the impact of corticosteroids, pre-steroid samples (n=17) were compared to post-steroid samples (n=19). b-pDCs (FDR=1×10⁻⁴), b-MNP^{FCGR3A,CDKN1C} (FDR=0.002), and b-DC2^CD1C,CLEC10A (FDR=0.018) subsets were less abundant in post-steroid samples (Extended Data Fig 1j, Supplementary Table 7). Differential gene expression (DGE) analysis and GSEA across blood cell lineages demonstrated marked transcriptional changes (4,613 significantly differentially expressed genes [DEGs]) and increased immune cell activation in pre-steroid samples (Extended Data Fig 1k-l, Supplementary Table 8-9). Fewer genes (312 significant DEGs across all lineages) and distinct gene sets correlated with time on steroids (Extended Data Fig 1m-o). For example, MNP expression of gene sets associated with adipogenesis, cholesterol homeostasis, and glycolysis were inversely associated with time on steroids, consistent with immunosuppression by steroids through altered MNP metabolism (Supplementary Table 8)²¹. Collectively, these data support the use of solely pre-steroid samples to define cell subset abundance and gene expression changes in irMyocarditis.

Demultiplexing pooled PBMC scRNA-seq data allowed use of 28 out of 30 controls for comparisons with pre-steroid irMyocarditis PBMC samples (n=17). B-pDCs (FDR=0.002), b-cDCs (FDR=0.012), b-B/plasma cells (FDR=0.012), and b-CD4T cells (FDR=0.090) were all less frequent in irMyocarditis, while b-MNPs were more frequent (FDR=0.026) (Fig 1i, Supplementary Table 7). Across blood lineages, GSEA showed enrichment of gene sets associated with interferon responses, allograft rejection, antigen presentation, cell adhesion, and viral myocarditis in irMyocarditis samples, broadly mirroring heart tissue findings (Fig 1f, 1j; Supplementary Table 8). b-MNP frequency in blood correlated directly with serum troponin T (FDR=0.008), a widely used clinical biomarker of myocardial injury that is associated with poor outcomes in irMyocarditis (Fig 1k, Supplementary Table 14)²².

Subclustering of b-CD8T/NK (n=12 subsets; 135,712 cells), b-CD4T (n=7; 45,022 cells), myeloid (comprised of b-MNP, b-cDC, and b-pDC; n=12; 149,101 cells), and b-B/ plasmablast (n=6; 21,171 cells) lineages were performed to investigate abundance changes of each sub-population in irMyocarditis samples relative to controls (Extended Data Fig 2a-l). No B or T cell subsets were more abundant in irMyocarditis, but significant decreases were observed in the circulating frequency of five b-CD8T/NK subsets, three b-CD4T subsets, and five b-B/plasmablast subsets (Extended Data Fig 2c, 2f, 2i, 2l; Supplementary Table 7). The frequency of two b-MNP subsets were increased in irMyocarditis: cluster 5 (b-MNP^FCGR3A,C1QB; FDR=0.082), and cluster 7 (b-MNP^CD14,IFI44L; FDR=0.047). Marked changes were noted in the b-cDC populations: cluster 8 (b-DC3^CD14,CLEC10A) was more abundant (FDR=0.014), while cluster 11 (b-DC1^CLEC9A,IDO1; FDR=4.0×10⁻⁵), and cluster 6 (b-DC2^CD1C,CLEC10A; FDR=2.6×10⁻⁴) were less abundant. None of these cell subsets had a circulating frequency associated with troponin T (Supplementary Table 14).

DGE analysis and GSEA performed on these lineages and subsets showed that T cells, and particularly b-CD8T subsets, had increased expression of genes related to antigen recognition and activation (PDCD1, TNFRSF9, IFNG, TNF) and decreased expression of transcription factors involved in cell stemness (TCF7) and blood recirculation (KLF2) (Extended Data Fig 2m-n, Supplementary Table 8-9). Subsets b-MNP^CD14,CSF3R and b-MNP^CD14,DUSP6 showed increased expression of CXCL10, which has been implicated in irMyocarditis pathogenesis¹², and IL27, whose protein product is increased in irMyocarditis patient serum (Extended Data Fig 1f-g). GSEA also showed associations with pathways related to interferon responses, allograft rejection, viral myocarditis, cell adhesion, and antigen presentation, as well as a broad decrease in TGF-β signaling.

To provide orthogonal validation of our scRNA-seq blood subset abundance findings, we analyzed CITE-Seq surface protein data, which is analogous to highly multiplexed flow cytometry²³. Sequential gates were used to define B cells, CD8T, CD4T, NK, pDC, DC2, and MNPs, and these assignments closely approximated those from the RNA-based clustering (Extended Data Fig 3a-c). Abundance analyses of pre-steroid irMyocarditis samples (n=17) and ICI-treated controls (n=28) mirrored our scRNA-seq findings: pDCs (FDR=0.003), CD8T (FDR=0.030), CD4T (FDR=0.041), NK (FDR=0.041), cDC2s (FDR=0.041), and B cells (FDR=0.041) were all less abundant in irMyocarditis patient blood samples, while MNPs were more abundant (FDR=0.003) (Extended Data Fig 3d, Supplementary Table 15). When CD8T and NK cell populations were combined (“CD8T/NK”) to mirror our scRNA-seq-defined lineages, this population was also less abundant in irMyocarditis (FDR=0.027). Collectively, these results nominated novel peripheral biomarkers of irMyocarditis onset.

Intracardiac T-cell phenotypes

We next subclustered the 9,134 intracardiac h-T/NK cells and defined six distinct subsets: four h-CD8T, one h-CD4T (cluster 3, h-CD4T^IL7R,LTB), and one h-NK cell subset (cluster 1, h-NK^KLRF1,FCER1G) (Fig 2a, Extended Data Fig 4a-b, Supplementary Tables 5-6). All h-CD8T cell subsets broadly expressed markers of cytotoxicity (GZMK, CCL5) and cell adhesion (ITGB7). Cluster 2 (h-CD8T^CD27,LAG3) and cluster 6 (h-CD8T^cycling) expressed a spectrum of activation and exhaustion markers (CTLA4, PDCD1, TOX), as well as the chemokine receptor CXCR3. Cluster 4 (h-CD8T^CCL5,NKG7) also included a small group of cycling cells with lower expression of the exhaustion markers LAG3 and TOX. Finally, cluster 5 (h-CD8T^KLRG1,CX3CR1) expressed markers of short-lived effector cells (KLRG1, CX3CR1)²⁴. No clear helper T cell populations could be defined within h-CD4T^IL7R,LTB using canonical markers (Extended Data Fig 4c), so these cells were retained as a single cluster for downstream analyses.

Fig 2. Cardiac T cell phenotypes in irMyocarditis.

a, UMAP embedding displaying seven subsets yielded from 9,134 T/NK cells in the heart tissue scRNA-seq dataset. b, Abundance analysis comparing intramyocardial frequencies of cell populations from pre-steroid irMyocarditis heart samples (n=12, red) versus controls (n=8, blue). Left: dots represent logistic regression odds-ratios. Error bars represent 95% confidence intervals. Unadjusted two-sided P values for likelihood ratio test are shown. Bold font and colored bars indicate FDR<0.1. Right: box plots show median (line) and IQR, with whiskers no more than 1.5x IQR. Each dot represents one patient. c, Volcano plot of T/NK lineage genes modeled by serum troponin T; select genes highlighted. Colored points represent FDR<0.1 (two-sided Wald test). d, Select pathways are shown from GSEA of T/NK cell DEGs modeled by serum troponin T (adjusted empirical P value). Abbreviations: NES, normalized enrichment score.

h-CD8T^CCL5,NKG7 (FDR=0.002) and h-CD4T^IL7R,LTB (FDR=0.073) were more abundant in irMyocarditis tissue, while h-CD8T^cycling and h-CD8T^CD27,LAG3 displayed trends towards greater abundance (Fig 2b, Extended Data Fig 4d-e, Supplementary Table 7). h-CD8T^cycling abundance correlated with serum troponin T (FDR=0.091), though this association was driven by few donors (Extended Data Fig 4f, Supplementary Table 14). None of the five other clinical measures of cardiac function or inflammation evaluated were similarly correlated with the abundance of any cell subset enriched in irMyocarditis (Supplementary Table 14).

DGE analysis was conducted for the intracardiac T cell lineage (“All T” cells) and for each of the defined subsets. Significant upregulation of antigen presentation (CD74), immunoreceptor signaling (LCP2, TOX), and interferon-stimulated genes (GBP5, IFITM1) were seen across multiple subsets (Extended Data Fig 4g-h, Supplementary Table 9). Other gene expression changes were cell subset-specific: cluster 6 (h-CD8T^cycling) showed upregulation of multiple immune checkpoints (CTLA4, LAG3), while cluster 2 (h-CD8T^CD27,LAG3) upregulated ITGA4, which encodes a therapeutically targetable adhesion molecule²⁵. GSEA similarly highlighted enrichment of gene sets involved in cell adhesion, allograft rejection, interferon gamma response, TCR signaling, and DNA replication at the lineage or subset level; viral myocarditis gene sets were notably not upregulated.

To help nominate genes associated with irMyocarditis pathogenesis, we performed gene expression modeling by serum troponin T. The gene expression changes most associated with troponin T were within the h-T/NK lineage (Extended Data Fig 4i, Supplementary Table 9). Positively correlated genes included those associated with cell cycle (MKI67, TOP2A), antigen presentation (HLA-DRA, HLA-DRB1), and activation/exhaustion (LAG3) (Fig 2c). Genes inversely correlated with troponin T included the receptors CX3CR1 and S1PR5. GSEA indicated that pathways associated with cell cycle and MTORC1 signaling were positively correlated with serum troponin T, while TNF-α signaling was inversely correlated (Fig 2d).

T-cell clones in heart, tumor, and blood

We sought to assess the impact of immunosuppression on irMyocarditis heart TCR repertoire using a pre-steroid biopsy specimen (SIC_264_A) and a second specimen obtained at autopsy seven days later (SIC_264_B) from the same patient after the administration of methylprednisolone and abatacept (Fig 1b). The top 13 most expanded TCR clones were found at both timepoints (Extended Data Fig 4j-k). Notably, pre-steroid expanded clones predominantly mapped to CD8T cells expressing cycling markers (e.g., STMN1), while post-steroid expanded TCR-β clones chiefly mapped to the h-CD8T^CD27,LAG3 subset, which lacks markers of cell division (Extended Data Fig 4a-b). Hence, immunosuppression appeared to alter the transcriptional profile of cardiac T cells but not the predominant TCR clones. Therefore, tissue samples collected both before and after steroid exposure were included in our subsequent TCR analyses.

Next, TCR-β sequencing was used to explore the relationship of T-cell clones in irMyocarditis heart tissue and paired tumor (Fig 3a). IrMyocarditis autopsy cases without diffuse myocardial metastases (n=6) were classified as “active” (n=2; SIC_17, SIC_264), “borderline”; (n=1; SIC_136), or “healing” myocarditis (n=3; SIC_3, SIC_175, SIC_266). IrMyocarditis tissues yielded more TCR-β sequences per total nucleated cells than controls (p=0.002) (Extended Data Fig 5a). The two active irMyocarditis cases had the lowest diversity (measured by Hill’s diversity index across all diversity orders), and each had single TCR-β sequence comprising >40% of their repertoire (Extended Data Fig 5b-d). Additionally, the TCR-β repertoire was more polyclonal in irMyocarditis patients with cardiac metastases (SIC_136 and SIC_232) than patients with active irMyocarditis and no cardiac metastases (Extended Data Fig 5e).

Fig 3. T-cell receptors enriched in irMyocarditis are largely distinct from those in tumor, distinguish fatal cases, and do not recognize putative cardiac autoantigens.

a, Schematic of bulk TCR-β sequencing showing macroscopic dissection of paired irMyocarditis, tumor, and tumor-adjacent normal parenchyma (“control tissue”). b,c, Each TCR-β clone enriched in heart or tumor relative to controls (FDR<0.05, two-sided Fisher’s exact test) is plotted according to its proportion (among all TCR-β clones in the respective tissue) in heart (x-axis) and tumor tissue (y-axis), normalized by its proportion in control tissue. Individual clones are colored by (b) tissue(s) of enrichment and (c) donor. d, The percent of all TCR-β clones shared between heart and blood cells found within each blood cell subset was calculated on a per-donor basis and stratified by fatal (n=4, orange) and non-fatal (n=9, grey) cases. Box plots display the median (line) and IQR, with whiskers no more than 1.5x IQR. Each dot represents one patient. b-CD8^cycling is highlighted. P values from two-sided t-tests of subsets meeting an FDR<0.1 are shown. e, UMAP embedding displaying CD8T/NK blood subsets with (top row, left) b-CD8^cycling (cluster 10) circled, and (top row, right and bottom row) feature plots indicating RNA transcript expression (logCPM) or protein expression (CLR) across CD8T/NK cells. f, Schematic illustrating the antigen screening assay. Created with BioRender.com. g, Graph indicates the background-subtracted percent of TCR-transduced (mTRBC⁺) T cells expressing CD137 after exposure to the indicated peptide pools (x-axis). mTRBC⁺CD8⁺ or mTRBC⁺CD8^- results are reported (Extended Data Figure 6a). Red dots represent a positive control TCR that recognizes the RINATLETK peptide¹⁶; blue dots represent control TCRs from unrelated patients; grey dots represent 52 heart-expanded TCRs. Abbreviations: APC, antigen presenting cell; CEF, cytomegalovirus, Epstein-Barr virus, and influenza virus; Ova, ovalbumin; TnI, troponin I; TnT, troponin T.

TCR-β repertoires from irMyocarditis hearts were then compared to those from autologous tumors in patients with available normal “control” tissue adjacent to tumors (n=4). The inclusion of such control tissue was intended to account for “bystander” T cells²⁶ and resident T-cell populations in tumor-infiltrated parenchyma²⁷. TCR-β clones enriched in heart relative to control tissue were recovered from all irMyocarditis patients (1–9 clones per donor; n=19 total), while clones enriched in tumor tissue relative to control were found in three of these patients (0–17 clones per donor; n=28 total)(Fig 3b-c, Extended Data Fig 5f-g, Supplementary Table 16). Across all irMyocarditis patients, five clones were enriched in both heart and tumor. However, in each irMyocarditis patient, the most enriched heart clone was not enriched in tumor. Grouping of lymphocyte interactions by paratope hotspots (GLIPH) analysis found that most expanded TCR-βs were not part of any similarity group, but the expanded motifs (n=14) tended to be distinctly enriched in either irMyocarditis or tumor (Extended Data Fig 5h-j). While shared epitope specificity among the repertoires of heart and tumor cannot be entirely excluded, these results suggest that enriched TCR-β clones and motifs in each tissue are largely distinct.

We next examined if TCR clones enriched in irMyocarditis heart tissue could also be detected among circulating CD4T and CD8T cells. Across all patients, significant TCR-β sequence sharing was found between the heart and five blood CD8T subsets but no CD4T subsets (Extended Data Fig 6a-b). On a per-patient basis, there was significantly more TCR-β sharing between expanded cardiac T cells and the b-CD8^cycling subset (CD8/NK cluster 10) in fatal versus non-fatal irMyocarditis (FDR=0.090) (Fig 3d, Extended Data Fig 6c-d, Supplementary Table 17). B-CD8^cycling expresses CXCR3, cycling genes (MKI67, STMN1), and CD45RO surface protein, indicating that these are antigen-exposed and dividing cells (Fig 3e). b-CD8^cycling also showed increased expression of PDCD1, TNFRSF9, and TNF in irMyocarditis samples versus controls (Extended Data Fig 2n), suggesting increased activation and cytotoxicity. In contrast, the b-CD8^CCL5,GNLY subset (CD8/NK cluster 3) had greater heart/blood TCR-β sharing in non-fatal samples (FDR=0.090). In a fatal irMyocarditis case (SIC_264), heart CD8T cells sharing TCR-β sequences with blood T cells were found in heart clusters expressing CXCR3, which can support cardiac recruitment of T cells²⁸, but not CX3CR1 (Extended Data Fig 6e-f), expression of which in cardiac T cells was inversely correlated with serum troponin T levels (Fig 2c). Collectively these data suggest that sharing of TCR-β between the heart and b-CD8^cycling cells may serve as a biomarker of disease severity.

TCR autoantigen screening

We then investigated whether heart-expanded TCRs were specific for the previously described cardiac autoantigens α-myosin¹⁶, troponin I, and troponin T^29–31. We assayed 52 heart-expanded TCRs identified from eight irMyocarditis patients; a TCR recognizing a peptide from the α-myosin protein (RINATLETK) served as a positive control¹⁶, and TCRs from unrelated individuals as negative controls (Fig 3f, Supplementary Table 18). While the positive control TCR recognized RINATLETK as a purified peptide and as part of a peptide pool (α-myosin Pool 4), none of the expanded TCRs derived from irMyocarditis donors in our study demonstrated recognition of any screened peptides (Fig 3g, Extended Data Fig 6g-h). These results suggest that the most expanded intracardiac T cell clones in our irMyocarditis cohort do not recognize α-myosin, troponin I, or troponin T.

Intracardiac cDCs, pDCs, and MNPs

We then subclustered the 9,824 MNP and dendritic cells, yielding five distinct MNP subsets, cDCs, and pDCs (Fig 4a, Extended Data Fig 7a-b, Supplementary Tables 5-6). Cluster 2 (h-MNP^LYVE1,C1QA) resemble cardiac resident macrophages^32,33. Cluster 3 (h-MNP^{FCGR3A,LILRB2}) and cluster 4 (h-MNP^S100A12,VCAN) express FCN1, which has been associated with monocyte-like signatures³⁴ and distinguishes them from cluster 1 (h-MNP^{S100A8-low,C1QA-low}). Cluster 5 (h-MNP^TREM2,APOC1) expressed GPNMB and FABP5, the latter of which is associated with lipid-rich macrophages^35,36. There were too few h-cDCs^CLEC9A,CD1C to further subcluster these cells into cDC1 (CLEC9A) and cDC2 (CD1C). h-MNP^{S100A8-low,C1QA-low}, h-MNP^{FCGR3A,LILRB2}, and h-cDCs^CLEC9A,CD1C demonstrated trends towards increased abundance in irMyocarditis tissue (Extended Data Fig 7c-e, Supplementary Table 7) but only h-cDCs^CLEC9A,CD1C frequency was correlated with serum troponin T (FDR=0.089) (Fig 4b, Supplementary Table 14).

Fig 4. Conventional dendritic cells (cDCs) are enriched in irMyocarditis heart tissue and associate with disease severity.

a, UMAP embedding displaying eight subsets yielded from the 9,824 MNP cells in the heart tissue scRNA-seq dataset. b, The intracardiac frequency of h-cDCs^CLEC9A,CD1C (y-axis) was plotted against serum troponin T levels (x-axis) for irMyocarditis heart samples (n=12); p=0.013, FDR<0.1, by linear regression. c-e, In situ visualization of cDCs (stained with pooled CLEC9A and CD1c antibodies) in tissue and colocalization with CD8T cells (CD8A⁺) cells. Representative image of irMyocarditis heart tissue stained with (c) H&E or by (d, e) RNA ISH-IF. The white box in d identifies a field of interest shown at higher magnification in e, highlighting cDCs (white arrows). f-g, Representative section of control heart tissue from an ICI-treated patient stained with (f) H&E and by (g) RNA ISH-IF. h, Box plot showing the intracardiac density of cDCs in irMyocarditis hearts (n=7, red) versus controls (n=3, blue); p=0.008, FDR<0.1, one-sided Mann-Whitney-U test. Median (line) and IQR are displayed, with whiskers encompassing no more than 1.5x IQR. Each dot represents one patient. Abbreviations: H&E, hematoxylin and eosin; IF/RNA-ISH, immunofluorescence/RNA in situ hybridization.

We sought to validate the presence of cDCs in irMyocarditis heart tissue and assess their proximity to CD8 T cells using microscopy. After detecting CD1c⁺ cells by immunohistochemistry (Extended Data Fig 7f), we performed multiplexed RNA-in situ hybridization-immunofluorescence (ISH-IF) on ICI-treated control (n=3) and irMyocarditis (n=7) heart tissue from autopsy samples (Fig 4c-h). Across whole slides, cDCs were increased in irMyocarditis (FDR=0.033) (Supplementary Table 20). Areas of inflammation in irMyocarditis tissue sections had 1.7-fold greater density of cDCs (FDR=0.033) and CD8T cells (FDR=0.033) than non-inflamed areas (Extended Data Fig 7g-h). cDCs were in closer proximity to CD8A-expressing cells in irMyocarditis cases (median distance 748.2 μm) than control samples (median distance 4686 μm) (Extended Data Fig 7i). Collectively, these data suggest that cDCs are enriched in irMyocarditis hearts, particularly in regions of active inflammation.

Critical genes and gene sets were upregulated across multiple heart MNP subsets in irMyocarditis (Extended Data Fig 7j; Supplementary Tables 8-9). DEGs include those related to antigen presentation (HLA-C, HLA-DQB2, PSMB9), CXCR3 signaling (CXCL9, CXCL10; Extended Data Fig 7k), cytokine signaling (IL15, TNF, STAT1), and ISGs (GBP5, IFITM1). GSEA indicated activation of transcriptional programs related to antigen processing, cell adhesion, interferon responses, allograft rejection, and viral myocarditis.

Stromal cells as inflammatory mediators

We next subclustered the 65,409 non-immune cells, yielding 17 distinct subsets: eight endothelial (EC), three pericyte, one fibroblast, one myofibroblast, one smooth muscle, one endocardial, one neural, and one cardiomyocyte population (Extended Data Fig 8a-e, Supplementary Tables 5-6). Three subsets were less abundant in cases, while fibroblasts trended towards enrichment in irMyocarditis (Extended Data Fig 8c, f-g; Supplementary Table 7). DGE analysis demonstrated marked transcriptional changes, as 654 unique genes were upregulated in at least one stromal cell subset (Extended Data Fig 8h, Supplementary Table 9). Antigen presentation (CD74, HLA-DPB1), chemokine (CXCL9, CXCL10, CXCL11), and cytokine (MDK, FLT3LG) transcripts were all increased in several cellular subsets. These results implicate non-immune cells in the orchestration of cardiac immune responses in irMyocarditis^37,38.

Subclustering of the 6,701 cells in the fibroblast lineage yielded a myofibroblast subset (cluster 2; Myofibroblast^ACTA2,ID4) and five fibroblast subsets (Fig 5a-b; Supplementary Tables 5-6). Two subsets were distinguished by the expression of EGR1: Fibroblast^DCN,EGR1low (cluster 1) and Fibroblast^{GPC3,EGR1high} (cluster 3). Fibroblast^{PCOLCE2,IGFBP6} (cluster 4) expresses genes involved in extracellular matrix remodeling¹⁸. Fibroblast^POSTN,F2R (cluster 5) expresses TGFβ-responsive genes implicated in fibrosis^18,39. Fibroblast^{CXCL9,HLA-DRA} (cluster 6) expressed proinflammatory chemokines (CXCL9, CXCL10, CXCL11, CXCL16, CCL5, CCL19), MHC-II machinery (HLA-DRA), and ISGs (GBP4)(Extended Data Fig 9a). Fibroblast^DCN,EGR1low and Fibroblast^{CXCL9,HLA-DRA} trended towards enrichment in irMyocarditis tissue (Extended Data Fig 9b, Supplementary Table 7), but only Fibroblast^{CXCl9,HLA-DRA} abundance was positively correlated with serum troponin T (FDR=0.089) (Fig 5c, Supplementary Table 14). DEGs associated with irMyocarditis in the fibroblast subsets include those involved in antigen presentation (HLA-DRA, PSMB9), cytokine signaling (CXCL9, MDK), and ISGs (GBP1, GBP4) (Fig 5d, Supplementary Table 9). GSEA showed similar patterns, but additionally highlighted pathways associated with both dilated and hypertrophic cardiomyopathy (Supplementary Table 8). Genes and GSEA pathways associated with viral myocarditis, interferon responses, and cytokine signaling were positively correlated with increasing serum troponin T (Fig 5e, Extended Data Fig 9c, Supplementary Table 8-9).

Fig 5. Inflammatory fibroblast enrichment in irMyocarditis.

a, UMAP embedding displaying seven subsets from 6,701 fibroblasts/myofibroblasts in the heart tissue scRNA-seq dataset. b, Top marker genes for each fibroblast subset. Dot size represents the percent of cells in subset with non-zero expression of indicated gene. Color indicates scaled expression. c, Intracardiac frequency of Fibroblasts^{CXCL9,HLA-DRA} (y-axis) plotted against serum troponin T levels (x-axis); n=12, p=0.008, FDR<0.1 (linear regression). d, Left: heatmap showing selected DEGs between irMyocarditis cases and controls across biological categories. Color scale indicates log2FC. Black dots indicate FDR<0.1 (Wald test). Right: select GSEA results, color-coded by NES of the gene set in irMyocarditis cases. “All Fibroblast” represents a pseudo-bulk analysis of the five fibroblast subsets and exclude myofibroblasts. Bolded genes in heatmap indicate leading edge genes for displayed GSEA pathways. e, Select GSEA pathways modeled by serum troponin T (adjusted empirical P value). f-l, In situ visualization of CXCL9/10/11⁺ fibroblasts. f-g, Representative control heart tissue stained with (f) H&E or (g) RNA ISH-IF. h-l, Representative irMyocarditis heart tissue stained with (h) H&E or by (i-l) RNA ISH-IF. White box in i is shown at higher magnification in j-l, highlighting co-staining of (j) DAPI, CLEC9A/CD1c, and COL1A1/A2; (k) DAPI, CLEC9A/CD1c, and CXCL9/10/11; and (l) all stains. m, Box plot of CXCL9/10/11⁺ fibroblasts in irMyocarditis heart (n=7, red) versus control tissue (n=3, blue), where each dot represents one patient; p=0.044, FDR<0.1, one-sided Mann-Whitney-U test. IrMyocarditis samples are denoted “Active” (dark red) or “Healing” (light red). Median (line) and IQR are displayed, with whiskers encompassing no more than 1.5x IQR. Abbreviations: Allo, allograft rejection; CAM, cell adhesion molecules; DCM, dilated cardiomyopathy; DNA, DNA synthesis; H, Hallmark; HCM, hypertrophic cardiomyopathy; H&E, hematoxylin and eosin; IFNG, interferon-γ signaling; IF/RNA-ISH, immunofluorescence/RNA in situ hybridization; K, Kyoto Encyclopedia of Genes and Genomes; Viral Myo, viral myocarditis.

ISH-IF validated the presence of cells expressing both fibroblast markers (COL1A1/COL1A2) and inflammatory cytokines (CXCL9/CXCL10/CXCL11) in proximity of CD8A-expressing cells and cDCs (Fig 5f-l). These inflammatory fibroblasts were observed in irMyocarditis hearts while absent from ICI-treated controls (FDR=0.074)(Fig 5m, Supplementary Table 20). Notably, these inflammatory fibroblasts were detected only in active but not in two cases of healing irMyocarditis (SIC_3, SIC_175). While cells similar to Fibroblast^{CXCL9,HLA-DRA} have not been described in prior human scRNA-seq studies of ischemic or dilated cardiomyopathy³⁹, they share features of inflammatory fibroblasts found in active human autoimmune disorders involving other organs and in the context of cellular rejection of murine heart allografts^40,41. These results suggest a coordinated immune response between immune and stromal cell subsets similar to other cardiac conditions^37,38 but not previously described for irMyocarditis.

Discussion

In this study, we present the first in-depth analysis of paired heart, blood, and tumor in irMyocarditis patients using systems immunology approaches. Our key findings, summarized in Extended Data Fig 9d, reveal novel circulating markers of irMyocarditis, including decreased frequencies of blood pDCs, cDCs, and B/plasma cells, and an increased frequency of blood MNPs. The correlations between serum troponin T levels and intracardiac frequencies of h-cDCs^CLEC9A,CD1C, h-CD8T^cycling, and Fibroblast^{CXCL9,HLA-DRA} nominate these cells as potentially pathogenic populations. Heart-expanded T-cell clones disproportionately shared TCR sequences with circulating b-CD8^cycling cells in fatal cases and are largely distinct from T-cell clones enriched in tumor tissue. Collectively, our results support a pathophysiologic model of irMyocarditis in which auto-reactive cytotoxic CD8T cells and antigen presenting cells are both recruited to and retained in the heart by stromal and immune-cell signaling networks.

Our findings differ from published results that implicated shared tumor antigens¹⁰ and α-myosin-reactive T cells in irMyocarditis pathology¹⁶. Interpatient and intertumoral heterogeneity as well as differences in methodology may partially contribute to this discrepancy. Our TCR-β repertoire analyses sought to account for potential “bystander” TCR clones²⁶ by normalizing heart and tumor repertoires to those found in normal tissues, which was not previously performed¹⁰. Additionally, our in vitro antigen screening assay included only the most expanded TCRs and did not fully recapitulate in vivo antigen processing. Thus, while we cannot exclude the possibility that lower abundance heart T-cell clones or those that recognize tumor antigens may contribute to irMyocarditis, our findings suggest that the most expanded T-cell clones in irMyocarditis patients are directed against as-yet undetermined antigens and are distinct from those driving anti-tumor immunity. An important unanswered question is whether TCR specificity is deterministic towards irMyocarditis incidence or severity.

Our results help to contextualize irMyocarditis within the emerging literature describing non-cardiac irAEs. IrAEs in barrier organs, such as the colon, may be driven by both resident and recruited immune cells^42,43, but irAEs in sterile organs such as the heart and joints are likely mediated by recruited immune cells⁴⁴. To our knowledge, our results are the first to demonstrate the presence and enrichment of fibroblasts expressing CXCR3 ligands (CXCL9/10/11) in an irAE-affected organ and implicate this signaling axis in the pathogenesis of irMyocarditis^12,45. These fibroblasts co-localized with cDCs, and while no tertiary lymphoid structures were observed, our data suggest that non-immune cells and cDCs may play a substantial role in irAE pathophysiology. CXCR3 signaling has been implicated in various irAEs^{14,42,44,46,47}; disrupting this pathway has demonstrated benefits in models of heart failure, type 1 diabetes, and myocarditis^28,48,49, but such disruption may also blunt anti-tumor responses⁵⁰. Deciphering irAE biology across organ systems may also help to explain why certain immunosuppressive agents are helpful in some irAEs but harmful in others^51,52.

Certain analyses were limited by the availability of matched blood, heart, and tumor specimens. ICI-treated heart controls lacking any pathology remain challenging to obtain, and our heart scRNA-seq analyses included six patients not exposed to ICIs from a published heart atlas¹⁸. Though the anatomic locations and tissue processing protocols were similar, technical variation may have contributed to observed changes in cell population abundance and gene expression changes. Additionally, irMyocarditis is often characterized by regions of inflammation interspersed within normal-appearing myocardium^10,11,53,54; hence, sampling variation may have contributed to some of the heterogeneity seen across irMyocarditis samples analyzed.

Additional research will be needed to identify drivers of irMyocarditis and optimal strategies to diagnose, risk-stratify, and therapeutically manage affected patients while preserving anti-tumor responses. These efforts will ultimately help to maximize oncologic benefits of ICIs while minimizing harm from irAEs.

Methods

Study design, patient identification, and sample selection – heart specimens for scRNA-seq

Heart tissue was collected at our institution through endomyocardial biopsy or autopsy of patients receiving ICI agents for the treatment of cancer. Informed consent was obtained from all patients or their appropriate representatives. This consent included consent to publish indirect patient identifiers, such as age, sex and patient-identified race. All research protocols were approved by the Dana-Farber/Harvard Cancer Center Institutional Review Boards (#11–181 and 13–416). Endomyocardial biopsies were collected in the catheterization laboratory as part of the clinical evaluation for suspected irMyocarditis, and a single tissue fragment (~1–2 mm³) primarily derived from the right side of the interventricular septum was collected for research. One patient (SIC_264) had a biopsy of the left ventricle for clinical reasons. Autopsy samples were obtained by an autopsy pathologist from the right ventricular free wall. Autopsy samples were obtained rapidly (within 6 hours) following the death of patients on ICI agents with or without clinically suspected irMyocarditis. ICI-treated donors were classified as having irMyocarditis based on one of the following: a histopathological diagnosis based on review of cardiac tissue by cardiac pathologists as part of routine clinical care (n=24) or a diagnostic cardiac MRI (based on Lake Louise criteria⁵⁵;n=4). Control heart samples were obtained through two sources: (i) two samples (SIC_182 and SIC_333) were obtained at our institution from cancer patients who received ICIs and underwent endomyocardial biopsies and/or autopsy with a clinical suspicion for irMyocarditis, but they ultimately did not have histopathological features of irMyocarditis. (ii) Six additional controls were obtained from hearts that were offered but not selected for transplantation and were used to generate a scRNA-seq heart atlas, as described¹⁸; none of these patients were known to be on ICI therapy. To account for differences in cell proportions or gene expression that occurs in different regions of the heart^18,56, we only used samples from this control cohort that were obtained from the right ventricle septum or right ventricular free wall to match the anatomic location of samples collected at our institution. The heart atlas also included samples that were enriched for CD45⁺ cells, which we included in our clustering and select downstream analyses, as outlined in Fig 1b.

Study design, patient identification, and sample selection – heart and tumor specimens for bulk TCR-β sequencing

Samples for bulk T-cell receptor β chain (TCR-β) sequencing were identified from the autopsies of cases enrolled in our scRNA-seq cohort as well as additional patients who consented to our tissue banking protocols and were found to have histological evidence of irMyocarditis at time of autopsy. Control heart tissues for bulk TCR-β sequencing were identified from patients who were receiving an ICI, did not have evidence of irMyocarditis at autopsy, and for whom archival tissue was available through our biobanking protocol. irMyocarditis cases without diffuse myocardial metastases patients (n=6) were classified by a cardiac pathologist as “active myocarditis” (n=2; SIC_17 and SIC_264), lymphocytic infiltrate without associated myocyte injury suggestive of irMyocarditis (“borderline”; n=1; SIC_136, which also had a single macroscopic cardiac metastasis in another tissue block), or “healing myocarditis” (n=3; SIC_3, SIC_175, and SIC_266), in accordance with the Dallas Criteria⁵⁷.

Study design, patient identification, and sample selection – blood and serum specimens

Paired blood and serum were sought prior to the initiation of corticosteroids for the treatment of irMyocarditis in our tissue scRNA-seq cohort as well as additional patients who were diagnosed with irMyocarditis but did not have tissue data. Where available, blood samples were collected or accessed from collaborating biobanking efforts at the following clinically relevant timepoints: (i) prior to the start of ICI (“pre ICI” timepoints); after the start of ICI but without clinical irMyocarditis (“on ICI”); (ii) at the time of clinical irMyocarditis diagnosis but prior to steroid initiation (“pre-steroid”); (iii) after steroid initiation or subsequent therapies (collectively, “post-steroid” samples). Control serum samples were identified from biobanked specimens from patients receiving an ICI who presented to MGH with concern for an irAE but ultimately were to have alternative causes for their symptoms and did not develop irMyocarditis or any other irAE.

Additional control PBMC samples were identified through the MGH Melanoma Biobank (DFCI/HCC Protocol #11–181). Controls received at least two doses of an ICI regimen that contained a PD-1 inhibitor and had at least eight weeks of follow-up before next therapy or death. Patients were included if they had no documented irAEs on that line of therapy. >300 clinical charts were reviewed. Samples collected within the first 100 days after starting therapy were prioritized to match the general timing if irMyocarditis onset, which is typically in the first three months of ICI therapy⁵³.

Clinical covariates

Clinical data were obtained retrospectively from electronic medical records, including patient demographics, cancer type, prior cancer therapies received, specific ICI agents used, method of irMyocarditis diagnosis, myocardial biopsy pathology grade, troponin T measurements, routine laboratory tests (NT-proBNP, erythrocyte sedimentation rate [ESR], C-reactive protein [CRP]), echocardiographic parameters (left ventricular ejection fraction), electrocardiogram data (QRS duration), and concomitant irAEs. Peak troponin T was defined as the highest value during the index admission. All clinical metadata can be found in Supplementary Table 1.

Preparation of tissue samples for scRNA-seq

Tissue samples obtained by biopsy or autopsy were immediately placed in ice-cold HypoThermosol solution (BioLife Solutions, Bothell, WA) and kept on ice during transfer to the research facility. Tissue was then washed twice with cold phosphate buffered saline prior to being dissociated with the human tumor dissociation kit according to the manufacturer’s instructions (Miltenyi Biotic, Bergisch Gladbach, Germany), with modification such that calcium chloride was added to the enzymatic cocktail to a final concentration of 1.25 mM. Biopsies were cut into ≤ 1 mm pieces with standard laboratory tissue dissection scissors. Tubes containing tissue fragments in the enzymatic cocktail were placed in a heated shaker at 37° C with shaking at 750 RPM for 25 minutes with the machine placed on its side to prevent tissue fragments from settling. Following incubation, the reaction was quenched through the addition of 100 μl human serum. The mixture was further dissociated through manual trituration followed by filtration through 70 μm mesh. Following centrifugation at 350 x g for 12 minutes, the supernatant was removed, and RBC lysis was performed for two minutes (ACK lysing buffer, Lonza, Basel, Switzerland). Following a wash step, cells were resuspended in phenol-free RPMI with 2% (v/v) human AB serum.

Due to excessive debris, the following myocardial samples obtained from autopsy were sorted by FACS for live (DAPI), non-red blood cells (CD235a) following tissue dissociation. These include the following samples: SIC 176, SIC 182_B, SIC 232, and SIC 264_B (B samples refer to the second of two samples when collected from the same patient). Tissue was prepared for cell sorting by first undergoing dissociation as described above. Following the resuspension of cells in sort buffer (phenol-free RPMI with 2% [v/v] human AB serum), Fc receptors were blocked (Human TruStain FcX, Biolegend 422302), after which cells were incubated with CD235a-PE-Cy5 (Biolegend 306606) for 30 minutes. Following a wash, cells were resuspended in sort buffer containing DAPI. Flow cytometric sorting was performed on a Sony MA900 Cell Sorter (Cell Sorter Software v 3.3.0) to collect live, singlet, CD235a^- cells (Extended Data Fig 10). Sorted cells were centrifuged and resuspended in sort buffer prior to loading in the 10x Chromium chip.

PBMC CD45⁺ enrichment, cell hashing, and CITE-Seq staining

Cryopreserved PMBC samples were used for CITE-Seq data generation (Supplementary Tables 1)²³. Cells were thawed at 37°C, diluted with a 10x volume of RPMI with 10% heat-inactivated human AB serum (Sigma), and centrifuged at 300 x g for 7 minutes. Cells were resuspended in CITE-seq buffer (i.e., RPMI with 2.5% [v/v] human AB serum and 2mM EDTA) and added to 96-well plates. Dead cells were removed with an Annexin-V-conjugated bead kit (Stemcell 17899) and red blood cells were removed with a glycophorin A-based antibody kit (Stemcell 01738); modifications were made to manufacturer’s protocols for each to accommodate a sample volume of 150 uL. Cells were quantified with an automated cell counter (Bio-Rad TC20), after which 2.5 × 10⁵ cells were resuspended in CITE-Seq buffer containing TruStain FcX blocker (Biolegend 422302) and MojoSort CD45 Nanobeads (Biolegend 480030). In batches where hashtags were used for demultiplexing, hashtag antibodies (Biolegend) were added to samples followed by a 30-minute incubation on ice and then washed three times with CITE-Seq buffer using a magnet to retain CD45⁺ cells. For each washing step, 1.4ml of CITE-Seq buffer was added to the sample, cells were resuspended followed by centrifugation at 300g for 7min at 4C. Supernatant was then removed before starting the next washing step. For samples where genetic demultiplexing was used, this hashtag staining step was omitted (Supplementary Tables 3-4). Live cells were counted with trypan blue, and 6–8 samples (each sample with 60,000 cells) were pooled together at equal concentrations. Pooled samples were filtered with 40 μM strainers, centrifuged, and resuspended in CITE-Seq buffer with TotalSeq-C antibody cocktail (Biolegend; Supplementary Table 12). Cells were incubated on ice for 30 minutes, followed by three washes with CITE-seq buffer and a final wash in the same buffer without EDTA (RPMI with 2.5% [v/v] human AB serum). Cells were resuspended in this buffer without EDTA, filtered a second time, and counted.

scRNA-seq data generation

For heart samples, single cell suspensions containing up to 12,000 cells (or all available cells when total was <12,000) were loaded per channel on 10x Chromium chips. For some samples, two channels on the 10x Chromium chip were loaded to maximize cell recovery; downstream data from these multiple wells were later combined and considered as a single sample. For hashed PBMC samples, samples were diluted to a concentration of 1,200 cells/μl, and 50,000 cells were loaded per channel into the 10x Chromium chip. According to the manufacturer’s instructions, cell/bead emulsions were generated and transferred to PCR tube strips, after which cDNA libraries were generated. Libraries from heart samples were generated using Chromium Single Cell 5’ V1 kits (10x Genomics PN-1000006) except for one sample (SIC_317), which was generated with a 5’ V2 kit (10x Genomics PN-1000263) (Supplementary Table 2). Hashed PBMC single cell libraries were generated with the Chromium Single Cell 5’ kit (V1.1, 10x Genomics PN-1000020) together with the 5’ Feature Barcode library kit (10x Genomics PN-1000080) (Supplementary Tables 3-4). TCR-enriched cDNA libraries were generated with the Chromium Single Cell V(D)J Enrichment kit (10x Genomics PN-1000005). Library quality was assessed with an Agilent 2100 Bioanalyzer.

All heart sample gene expression libraries were sequenced on an Illumina Nextseq 500/550 instrument using the high output v2.5 75 cycles kit with the following sequencing parameters: read 1=26; read 2=56; index 1=8; index 2=0. TCR-enriched libraries from heart tissue were sequenced on an Illumina Nextseq 500/550 instrument using the high output v2.5 150 cycle kit with the following sequencing parameters: read 1=30, read 2=130, index 1=8, index 2=0. All hashed PBMC CITE-seq, feature barcode (i.e., ADT and HTO libraries), and TCR libraries were sequenced on an Illumina Novaseq instrument using the S4 300 cycles flow with the following sequencing parameters: read 1=26 ; read 2=91 ; index 1=8; index 2=0.

scRNA-seq read alignment and quantification

Raw sequencing data was pre-processed with CellRanger (v3.0.2, 10x Genomics) to demultiplex FASTQ reads, align reads to the human reference genome (GRCh38, v3.0.0 from 10x Genomics), and count unique molecular identifiers (UMI) to produce a cell x gene count matrix⁵⁸. All count matrices were then aggregated with Pegasus (v1.1.0, Python) using the aggregate_matrices function⁵⁹. Low-quality droplets were filtered out of the matrix prior to proceeding with downstream analyses using the percent of mitochondrial UMIs and number of unique genes detected as filters (heart tissue=< 20% mitochondrial UMIs, > 300 unique genes; PBMCs=< 15% mitochondrial UMIs, > 300 unique genes). The percent of mitochondrial UMI was computed using 13 mitochondrial genes (MT-ND6, MT-CO2, MT-CYB, MT-ND2, MT-ND5, MT-CO1, MT-ND3, MT-ND4, MT-ND1, MT-ATP6, MT-CO3, MT-ND4L, MT-ATP8) using the qc_metrics function in Pegasus. The counts for each remaining cell in the matrix were then log-normalized by computing the log1p(counts per 100,000), which we refer to in the text and Figs as logCPM. The detailed quality control statistics for these datasets are compiled in Supplementary Tables 2 and 4.

Demultiplexing of PBMC data

For the PBMC data from patients who developed irMyocarditis, cells underwent cell hashing with TotalSeq-C hashing antibodies (see scRNA-seq data generation) and were demultiplexed using DemuxEM⁶⁰. For the PBMC data from our control cohort of patients that lacked any irAEs, cells were pooled prior to loading the 10x Chromium instrument, and an aliquot of cells from each donor was lysed in Buffer TCL + 1% beta-mercaptoethanol (BME) at a concentration of 1,000 cells/uL. 25 uL of lysed cells per sample was then used to generate low-input bulk RNAseq using the SmartSeq2 protocol⁶¹ to enable genetic demultiplexing. Bulk RNAseq data was sequenced on a NextSeq 550 with the parameters: read 1=38; read 2=38; index 1=8; index 2=8. In short, VCF files were generated from the bulk RNAseq data for each donor using CellSNP-lite (v1.2.3)⁶². Next, the pooled scRNA-seq data was demultiplexed with Souporcell (v2.5)⁶³ and the resulting VCF files for each predicted donor were matched to the VCF files from our bulk RNAseq data using Vireo (v0.5.8, Python)⁶⁴. All donors from the cohort were able to be demultiplexed with confidence with the exception of two donors (donor_730 and donor_327) due to poor recovery of variants from their bulk RNAseq data. Therefore, these donors were included in our clustering solution but were not included in any donor-driven analyses (i.e., differential gene expression and differential abundance).

Basic clustering

All cells from all samples were utilized for clustering. First, 2,000 highly variable genes were selected using the highly_variable_features function in Pegasus and used as input for principal component analysis. To account for technical variability between donors, the resulting principal component scores were aligned using the Harmony algorithm⁶⁵. The resulting principal components were used as input for Leiden clustering and Uniform Manifold Approximation and Projection (UMAP) algorithm (spread=1, min-dist=0.5).

Canonical markers were used to distinguish broadly defined cell lineages: T and NK cells when assessed together (CD3D and KLRB1); CD8T and NK cells (CD3D, CD8A, KLRF1) and CD4T cells (CD3D, CD4, IL7R) when assessed separately; cDCs (HRA-DRA, CD1c, CLEC9A); pDCs (LILRA4, IL3RA); other MNPs (LYZ, CD14, CD68); B cells and plasmablasts (CD79A, MZB1); endothelial cells (VWF, CA4); mural cells (RGS5, KCNJ8); fibroblasts (DCN, PDGFRA); cardiomyocytes (TNNI3, MB); and neural cells (PLP1, NRXN1). The number of principal components used for each blood clustering (CD4T=16; CD8T/NK=20 ; MNP=24; B cells=16; all cells=30) and tissue clustering (T/NK=24; MNP=35; non-immune=50) was decided via molecular cross-validation⁶⁶.

Marker gene identification and cell annotation

The marker genes defining each distinct cell subset from our global and lineage-specific subclustering analyses were determined by applying two complementary methods. First, we calculated the area under the receiver operating characteristic (AUROC) curve for the logCPM values of each gene as a predictor of cluster membership using the de_analysis function in Pegasus. Genes with an AUROC ≥ 0.75 were considered marker genes for a particular cell subset. Second, we created a pseudobulk count matrix⁶⁷ by summing the UMI counts across cells for each unique cluster/sample combination, creating a matrix of n genes x (n samples*n clusters). We performed “one-versus-all” (OVA) differential expression (DE) analyses for each cell subset using the Limma package (v3.54.0, R)⁶⁸. For each subset, we used an input model gene ~ in_clust, where in_clust is a factor with two levels indicating if the sample was in or not in the subset being tested. A moderated t-test was used to calculate P values and compute a false discovery rate (FDR) using the Bejamini-Hochberg method. We identified marker genes that were significantly associated with a particular subset as having an FDR<0.05 and a log2 fold change > 0. The AUROC and OVA pseudobulk marker genes for all cell subsets can be found in Supplementary Tables 5 and 13. Marker genes for each cell subset were interrogated and investigated in the context of other published immune profiles to guide our cell subset annotations.

Abundance analysis

To identify the association between cell subset abundance and either patient group (irMyocarditis versus control) or timepoint (pre-corticosteroid versus post-corticosteroid), we used a mixed-effects association logistic regression model similar to that described by Fonseka et al.⁶⁹. We used the glmer function from the lme4 package (v1.1–31, R) to fit a logistic regression model for each cell subset. Each subset was modeled independently with a “full” model as follows:

$$cluster~1+comparison+(1\mid id)$$

where cluster is a binary indicator set to 1 when a cell belongs to the given cell subset or 0 otherwise, comparison is a factor with 2 levels which represented either the patient groups (irMyocarditis or control) or the timepoints (pre-corticosteroid or post-corticosteroid), and id is a factor indicating the donor. The notation (1|id) indicates that id is a random intercept. To determine significant associations, a “null” model of cluster ~ 1 + (1|id) was fit and a likelihood ratio test was used to compare the full and null models. A false discovery rate was calculated using the Benjamini-Hochberg approach and clusters with an FDR<10% were considered significant.

Since we were underpowered to investigate differences between fatal and non-fatal irMyocarditis in our scRNA-seq data, we sought to identify associations between cell subset frequencies and clinically relevant cardiac measurements, including serum troponin T levels, NTproBNP, LV ejection fraction, erythrocyte sedimentation rate (ESR), C-Reactive Protein (CRP) and QRS duration. In addition, we aimed to understand how the length of time between corticosteroid initiation and post-corticosteroid sample collection altered cellular abundances.

For associations with clinical metrics, we limited our analysis to include only cell subsets that were differentially abundant in irMyocarditis cases versus controls (unadjusted P value < 0.1) given we do not be expect subsets that are unchanged in the presence of irMyocarditis to correlate with clinical metrics of cardiac injury. Similarly, for the modeling of gene expression by time between corticosteroid initiation and sample collection, we limited our analysis to only include cell subsets that were differentially abundant between pre- and post-corticosteroid samples (unadjusted P value < 0.1). For every metric, we performed linear regression on each cell subset individually, using the model log(abundance + 1) ~ log(variable) where variable is the value of the clinical metric and abundance is the relative proportion of the subset in a given sample. All clinical metrics represented measurements closest to time of tissue sample collection with cutoffs of no greater than ±2 days for troponin T measurements and no greater than ±7 days for all other measurements. For the analysis of time between corticosteroid initiation and sample collection, variable was the number of days.

To accurately measure the abundance of each cellular component of heart tissue, we used only unenriched samples (“native” fractions) and excluded the CD45⁺-enriched samples from the published heart atlas¹⁸. In the case of SIC_182, a control patient who underwent a biopsy (sample “SIC_182_A”) and then an autopsy (“SIC_182_B”) but did not have irMyocarditis and never received immunosuppression, data from this patient’s two samples were combined and treated as a single timepoint.

Differential gene expression analysis

Comparisons between irMyocarditis and control heart tissue was limited to samples collected prior to corticosteroids administration (“pre-steroid”) and included both native and CD45⁺ enriched samples from the heart cell atlas to account for the paucity of immune cells in uninflamed hearts²⁹ (Fig 1b). To capture differences in PBMCs associated with irMyocarditis or control, we compared pre-steroid PBMC samples from irMyocarditis cases to control samples from patients who were treated with a checkpoint inhibitor but showed no symptoms of any irAEs. To understand the effects of corticosteroids on our PBMC dataset, we compared pre-steroid irMyocarditis samples to samples collected after corticosteroid administration (“post-steroid”). These differential gene expression analyses (irMyocarditis case versus control, pre-steroid versus post-steroid) were performed on pseudobulk count matrices using the DESeq2 package (v. 1.38.2, R). The input model was gene ~ comparison, where comparison was a binary variable indicating either if the sample came from an irMyocarditis patient or control or indicating the timepoint for the corresponding sample (pre-steroid, post-steroid).

To identify gene expression associations with disease severity, we used DESeq2 with the input model gene ~ log(troponin) where troponin is an individual’s serum troponin T level measured closest to the time of tissue sample collection (no greater than ±2 days).

To identify genes associated with the length of time between corticosteroid initiation and post-corticosteroid sample collection, we used DESeq2 with the input model gene ~ log(change in post-steroid troponin) + log(time from steroid start to post-steroid sample collection) where change in post-steroid troponin was the fold-change between a donor’s peak serum troponin T level and their serum troponin T level at the time of post-steroid sample collection. Time from steroid start to post-steroid sample collect was the number of days between corticosteroid initiation and post-corticosteroid sample collection. This model allowed us to find changes associated with the amount of time on corticosteroids while controlling for potential improvement in irMyocarditis.

Significant differentially expressed genes (DEGs) were identified using a Wald test (FDR < 0.1) (Supplementary Table 9). Select DEGs were visualized with the ComplexHeatamp package (v.2.14.0, R).

Gene set enrichment analysis

GSEA was performed using the fgsea function from the fgsea package (v1.18.0, R) with 10,000 permutations to test for independence. For both comparisons of irMyocarditis cases versus controls as well associations with serum troponin T levels, the input gene rankings for each cell subset were based on decreasing Wald statistic values from DE analysis where the gene with the highest Wald statistic (i.e., associated with irMyocarditis or associated with higher troponin) was ranked first and the gene with the lowest Wald statistic (i.e., associated with control or associated with lower troponin) was ranked last. The input gene sets tested were derived from the KEGG and HALLMARK pathways database and are compiled in Supplementary Table 8.

Secreted factors analysis

Peripheral blood was collected from irMyocarditis and control patients in serum separator tubes. Following centrifugation for 10 minutes at 890g, serum was collected and stored at −80° C. Aliquots of undiluted serum were analyzed for the presence of 71 secreted factors through multiplex immunoassay (Human Cytokine Array/Chemokine Array 71-Plex Panel [catalog #HD71], Eve Technologies, Calgary, AB, Canada). Log +1 transformed values were compared using a two-sided t-test.

CITE-Seq analysis

CITE-Seq protein feature data was used to validate changes in cell population abundance between in irMyocarditis cases (n=17) and controls (n=28) that were observed using gene expression (GEX) scRNA-seq data. Cells with recovered CITE-Seq information were first limited to non-doublet cells present in the blood GEX dataset through barcode matching. Cells were further filtered for those with at least a sum of 100 UMI counts across all CITE-Seq features. This data was then normalized using a centered log-ratio transformation and gated by protein expression value. Eleven manually curated protein features representing key canonical immune cell lineage markers were used to define the following gating strategies that clearly delineated positive and negative populations.

pDCs: CD3^-CD19^-CD56^-HLA-DR⁺CD123⁺

CD8T: CD19^-CD3⁺CD8⁺

CD4T: CD19^-CD3⁺CD4⁺

NK: CD19^-CD3^-CD14^-CD56⁺

cDC2s: CD3^-CD19^-CD56^-CD123^-CD11c⁺CD1c⁺

B cells: CD3^-CD19⁺

MNPs: CD3^-CD19^-CD56^-CD123^-CD1c^-CD94^-CD11c⁺

As with the GEX abundance analysis, a mixed-effects logistic regression model of cluster ~ 1 + comparison + (1|id) using the glmer function from lme4 in R was used.

Tissue/bulk TCR data generation

Four autopsy cases were identified with available matched irMyocarditis, tumor, and histologically normal tissue adjacent to tumor. This normal adjacent tissue was used as “control” tissue. Formalin-fixed, paraffin-embedded (FFPE) slides were stained for hematoxylin and eosin (H&E) and annotated by a cardiac pathologist to indicate regions of myocardial inflammation, tumor, and normal parenchyma. Marked regions of interest were manually macroscopically dissected using a scalpel to scrape tissue from serial unstained slides into 1.5 ml Eppendorf tubes pre-filled with 1 ml of xylene. Genomic DNA was extracted using the AllPrep DNA/RNA FFPE Kit instructions (Qiagen). T-cell receptor β chain (TCR-β) sequencing of the CDR3 region was performed (immunoSEQ ® human T-cell receptor beta [hsTCRB], Adaptive Biotechnologies)⁷⁰. We also profiled heart tissue for three cancer patients receiving an ICI who consented to our collection protocol, underwent an autopsy, and did not have evidence of irMyocarditis (“heart controls”) (Supplementary Table 1).

Comparison of irMyocarditis and tumor TCR-β repertoires

For comparisons between heart and tumor TCR repertoires, the frequencies, and not the absolute counts, of TCR clones were compared to account for differential recovery of TCR-β sequences from various tissues⁷¹. Expanded TCR-β sequences were defined as those accounting for >0.5% of all TCR-β sequences recovered from the tissue of interest (heart or tumor). These expanded sequences were then filtered to exclude bystander TCR-β sequences by performing a Fisher’s exact test on each, comparing the proportion of the TCR-β sequence in adjacent normal tissue to that of the tissue of interest (heart or tumor). Sequences with an FDR <5% were considered enriched in their given tissue site (heart or tumor).

TCR-β sequence diversity

Diversity curves that measured Hill’s diversity metric across diversity orders 0–4 were created using the package alakazam (v1.0.2, R) with the alphaDiversity function⁷². Hill’s diversity metric was only calculated on samples with ≥100 total TCR-β sequences.

Examination of shared TCR-β clones in heart and blood

Heart and PBMC single-cell data was combined with bulk TCR-β sequencing data yielded from heart tissue from “active” (SIC_17, SIC_264) or “borderline” irMyocarditis (SIC_136). For bulk TCR-β-seq data, expanded TCR-β sequences were defined as those that represented > 0.5% of the bulk repertoire and were found to have a read count ≥ 2. For scTCR-β-seq data, expanded TCR-β sequences were those that represented > 0.5% of the scTCR-βseq repertoire and were found in ≥ 2 cells. For patients with both bulk TCR-β-seq and scTCR-β-seq data, these data were analyzed together. Expanded heart TCR-β clones were then compared to TCR-β clones recovered from PBMC scRNA-seq data. First, PBMC cell subsets were investigated for their association with the presence of expanded TCR-β heart clones by fitting a logistic regression model using the glmer function from the lme4 package (v1.1–31, R). Each subset was modeled independently with the full model cluster ~ 1 + myo_clone, where myo_clone is a binary variable where the value was 1 if the cell had an expanded irMyocarditis TCR-β and 0 if it did not. An alternative null model cluster ~ 1 was then fit and a likelihood ratio test was used to compare the full and null models. Cell subsets with an FDR < 5% were considered to be enriched for expanded irMyocarditis TCR-β sequences. Additionally, we calculated the proportion of the repertoire that was overlapping with expanded heart TCR-β in each cellular subset for each donor and compared the proportions for each cluster. Log + 1 transformed proportions for each cluster were compared between fatal and non-fatal cases using a two-sided T test in R.

GLIPH

To discern TCR motifs, GLIPH2 was run on the TCR-βs derived from each donor individually^73,74 using the built-in naïve CD8 TCR repertoire as a reference. The resulting GLIPH groups were then limited to those that had ≥ 3 unique TCR-β sequences. The TCR-βs from each GLIPH group were used to calculate the proportion they represented in each tissue repertoire (irMyocarditis, tumor, control). GLIPH groups where the proportion of the member TCR-βs represented >0.5% of the irMyocarditis and/or tumor tissue repertoires were considered expanded GLIPH groups.

TCR Identification and Antigen Screening

52 TCRs were selected for antigen screening based on if they met any of the following criteria: (a) containing a TCR-β that was enriched in the heart compared to adjacent normal control tissue from our bulk TCR-βseq analysis (n=7); (b) enriched in heart and tumor when compared to adjacent normal control tissue (n=2); (c) expanded in the heart and found in the b-CD8^cycling subset of cells from our scRNA-seq PBMC data (n=7); (d) expanded in the heart and found in our blood CD4 subclustering (n=4); or comprising >1% of the captured bulk or scTCRseq repertoire from the heart (n=32) (Supplementary Table 18).

V-D-J rearrangement of TCR-α and β chains were identified from scRNA-seq or bulk TCR-β seq. For the latter TCRs, TCR-β chains were tracked in matched single-cell TCRseq in patient-matched blood and tissue data, and paired α chains were deducted through analysis of single-cell TCR-seq. In cases where a single TCR-β had multiple corresponding TCR-α chains, both were tested separately. Classification of the 52 clonotypes were based on CD4 or CD8 based on scRNA-seq based clustering of the cells that transcribed that TCR; based on such distinction, the downstream analysis was performed on CD4⁺ or CD8⁺ transduced T cells, respectively. TCRs were reconstructed and expressed in T cells from healthy donors, as described previously⁷⁵. An additional TCR that was reported to recognize a 9 aa peptide sequence in the α-myosin protein (RINATLETK) when displayed on an HLA-A*03:01 tetramer was also reconstructed for use as a “positive control” TCR (“Myo-TCR”)¹⁶. Background reactivity was assessed on T cells transduced with irrelevant TCRs isolated from unrelated patients⁷⁶; untransduced T cells were tested in parallel as additional negative controls.

The full-length TCRα and TCRβ chains, separated by Furin SGSG P2A linker, were synthesized in the TCRβ/TCRα orientation (Integrated DNA Technologies) and cloned into a lentiviral vector under the control of the pEF1a promoter using Gibson assembly (New England Biolabs Inc). For generation of TCRs, full-length TCRα V-J regions were fused to optimized mouse TCRα constant chain, and the TCRβ V-D-J regions to optimized mouse TCRβ constant chain to allow preferential pairing of the introduced TCR chains, enhanced TCR surface expression and functionality^77,78.

Donor T cells enriched from PBMCs using Pan-T cell selection kit (Miltenyi) were activated with antiCD3/CD28 dynabeads (ThermoFisher Scientific) in the presence of 5ng/mL of IL-7 and IL-15 (Peprotech). Activated cells were transduced with a lentiviral vector encoding the TCRB-TCRA chains for 3 subsequent days. Briefly, lentiviral particles were generated by transient transfection of the lentiviral packaging Lenti-X 293T cells (Takahara) with the TCR encoding plasmids and packaging plasmids (VSVg and PSPAX2)⁷⁵ using Transit LT-1 (Mirus). Lentiviral supernatant was harvested on days 1–2-3 after transfection and used to transduce activated T cells. On day 2, transduction efficiency was increased by performing spinoculation of the virus at 2,000 rpm, 37°C for 2 hours in presence of polybrene (8ug/mL, Sigma), and cells were cultured on viral supernatant for 3 days. 6 days after activation, the beads were removed from culture, and cells were expanded in media enriched with IL-7 and IL-15. Transduction efficiency was determined by flow cytometric analysis using the anti-mTCRB antibody (PE, clone H57–597, eBioscience, catalog # 12–5961-82, dilution 1:50). Transduced T cells were used at 14 days post-transduction for TCR reactivity tests.

We utilized screening through upregulation of CD137 expression to determine the reactivity of reconstructed TCRs⁷⁵. In brief, TCR-transduced T cell lines were washed, resuspended in PBS at 1 × 10⁶ cells per ml and labelled with a combination of three dyes (Cell Trace CFSE, Far Red or Violet Proliferation Kits, Life Technologies). Up to 4 dilutions of Cell-trace Violet, 3 dilutions of Cell-trace Far Red or CFSE were created and then mixed, resulting in up to 36 color combinations. After incubation at 37 °C for 20 min, T cells were washed twice, resuspended in complete medium and divided in pools. As internal controls, each pool contained a population of mock-transduced lymphocytes, a population of T cells transduced with an irrelevant TCR, and a population of T cells expressing the Myo-TCR¹⁶. 2.5×10⁵ TCR-transduced T cells were put in contact with an equal number of patient-derived antigen presenting cells pulsed with selected peptide pools described below.

Epstein–Barr virus-immortalized lymphoblastoid B cell lines (EBV-LCLs) served as antigen presenting cells, which were generated for five patients (SIC_48, SIC_171, SIC_232, SIC_258, SIC_264). Additionally, for three patients (SIC_17, SIC_175, SIC_317) for whom autologous EBV-LCL lines were not available, we screened 17 CD8 TCRs using single-HLA matching EBV-LCLs or monoallelic HLA class I lines (Abelin et al, Immunity 2017) covering 6 out of 6 (SIC_17) or 5 out of 6 (SIC_175, SIC_317) patients’ HLA class I alleles. Antigen presenting cells were pulsed with the following controls: DMSO (vehicle), or 10⁷ pg/mL of RINATLETK peptide (positive control), Ova Class I peptide (SIINFEKL; negative control), Ova class I peptide (ISQAVHAAHAEINEAGR; negative control), CEF peptides (JPT, catalogue #PM-CEF-S-1; negative control). The test conditions comprised of custom crude peptide pools designed to cover the full-length α-myosin (encoded by MYH6), troponin-I (encoded by TNNI3), and troponin-T (encoded by TNNT2) proteins. Peptide sequences were up to 20 aa long and tiled to achieve a 5 aa overlap, similar to previously reports identifying α-myosin-reactive TCRs¹⁶. 6–8 test crude peptides were combined into a pool (Genscript; Supplementary Table 19).

After an overnight incubation, TCR reactivity was measured trough detection of CD137 surface expression (PE, clone 4B4–1, Biolegend, catalog # 309804, dilution 1:50) on CD8⁺ (BV785, clone RPA-T8, Biolegend, catalog # 301046, dilution 1:100) or CD8^- (CD4⁺) TCR- transduced (mTCRB⁺, PE-Cy7, clone H57–597, eBioscience, catalog # 25–5961-82, dilution 1:50) T cells using a Fortessa flow cytometer (BD Biosciences). Data were analyzed using Flowjo software (v10.8.2, BD Biosciences). For CD8 TCRs, CD137 upregulation was evaluated from the CD8⁺ gate (Extended Data Fig 6g), and for CD4 TCRs, CD137 upregulation was evaluated from a CD8^- gate. Background CD137 reactivity, measured in the presence of target cells pulsed with DMSO, was subtracted. Data from the “positive control” Myo-TCR represent the results of that TCR tested against EBV-LCLs from a donor who has the HLA-A*03:01 allele.

IHC and H&E Staining

FFPE tissue sections from seven hearts from irMyocarditis patients and three hearts from ICI-treated control patients without irMyocarditis were stained using a Leica Bond RX automated stainer. Anti-CD1c-OTI2F4 mAb (Abcam ab156708) were labeled with DAB chromogen (Leica Bond Polymer Refine Detection DS9800) using EDTA based pH 9 epitope retrieval condition and 20 ug/ml antibody concentration. FFPE tissue sections from irMyocarditis cases and heart controls were manually stained for H&E.

Image Acquisition of IHC and H&E Images

Whole slide images of stained slides were acquired at 40x (0.13um/pixel) resolution using a MoticEasyScan Infinity digital pathology scanner.

ISH-IF Tissue Staining

FFPE tissue sections from seven irMyocarditis cases and three heart controls were stained using a Leica Bond RX automated stainer. The staining panel was developed using RNAscope ISH probes (Advanced Cell Diagnostics) with Opal fluorophores (Akoya Biosciences). The panel consisted of 6 RNA probes (Hs-COL1A-pool, Hs-CXCL9, Hs-CXCL10, Hs-CXCL11, Hs-FLT3LG, Hs-CD8A) and two antibodies targeting dendritic cells, Anti-CD1c-OTI2F4 (Abcam ab156708) and Anti-CLEC9A [EPR22324] (Abcam ab223188). Hs-COL1A-pool consisted of probes for COL1A1 and COL1A2. The three CXC chemokine probes were pooled into a single fluorophore channel. The two antibodies were also pooled into a single fluorophore channel. Tyramide Signal Amplification (TSA) was used to boost fluorophore signal (Opal 690, Opal 520, Opal 620, Opal 480, and Opal 780). Opal fluorophore concentrations were optimized to balance signal intensity across all channels. DAPI was used as a nuclear counterstain. FLT3LG was expressed diffusely, and we were unable to discern a clear signal for downstream analysis.

ISH-IF Image Acquisition

Whole slide images were acquired on an Akoya PhenoImager HT multi-spectral slide scanner at 20x (0.5um/pixel) resolution. Exposures were set for each image individually to avoid pixel saturation or underexposure. Akoya inForm software was used to spectrally separate signals from each fluorophore and to mitigate the effect of native tissue autofluorescence. The spectral library used to unmix the raw images was created using the synthetic Opal spectra and autofluorescence spectra from unstained human heart tissue. Unmixed tiles produced by inForm were stitched to reproduce a whole-slide pyramidal TIF using Indica Labs HALO software.

Image Analysis

Quantitative image analysis was performed using the HALO image analysis platform (v3.6.4134.137, Indica Labs). Images were manually annotated for regions to analyze. For all samples, the whole image was analyzed excluding manually-annotated areas of debris, tissue folds and large vessels. For the irMyocarditis cases, additional targeted analysis was performed on regions of myocarditis and non-myocarditis tissue that were annotated based on pathologist review of serial section H&E slides.

Cell segmentation and phenotyping was performed within annotated tissue regions using the HALO FISH-IF (v2.2.5) and AI (v3.6.4134) modules. Nuclear detection and cell segmentation were performed using the default AI nuclear segmentation algorithm based on the DAPI channel. Cells were phenotyped as CD1c/CLEC9A⁺ based on signal intensity of the antibody channel within the nuclear and cytoplasmic compartments. RNA-ISH positivity was based on signal intensity and dot size within a cell, with brighter and larger dots having more weight. Cell phenotypes were also defined using multiple marker criteria (COL1A1/2⁺CXCL9/10/11⁺). The positivity thresholds were manually validated for each sample for cell phenotyping quality. Summary tables containing cell phenotype information for each region of analysis (whole slide, inflamed regions, non-inflamed regions) were exported for further analysis across all samples. HALO Nearest Neighbor Spatial Analysis was performed on the phenotyped cells to determine the average distance from each CD1c/CLEC9A⁺ to the nearest CD8A⁺ cell. For comparisons of cell proportions between irMyocarditis versus Control, a one-sided Mann-Whitney U test was used given our data did not have a normal distribution (likely to the low number of samples). For our paired analysis between inflamed and non-inflamed tissue, we used a one-sided paired t-test.

Extended Data

Extended Data Fig 1. Cell lineages in heart and blood defined by scRNA-seq.

a, Dot plot showing top marker genes for each lineage in the heart. Dot size represents the percent of cells in the lineage with non-zero expression of a given gene. Color indicates scaled expression across lineages. b, Stacked bar plots showing the composition of major cell lineages, colored by data source (“MGH” refers to data generated in this study; “Sanger’’ refers to public heart atlas data)²⁹. c, Stacked bar plot showing the lineage composition per patient, grouped by the histology of the primary malignancy. d, Differential abundance of cell subsets in scRNA-seq data from the heart, comparing irMyocarditis cases to controls. Each point represents a cell subset. The x-axis represents the odds ratio (OR) in all cases and controls, and the y-axis represents the OR in an analysis that excluded cases (SIC_171 and SIC_232) and a control (SIC_182) with evidence of cardiac metastases. e, Differentially expressed genes in scRNA-seq data from the heart, comparing irMyocarditis cases to controls. Each point represents a gene contrasted in each cluster. The x-axis represents the log2 fold change (log2FC) in all cases and controls, and the y-axis represents the log2FC in an analysis that excluded samples with evidence of cardiac metastases. f, A volcano plot depicting 71 cytokines and chemokines measured in serum from irMyocarditis and control patients, where each point represents a protein. The x-axis represents log2FC between irMyocarditis (n=16) and control samples (n=10), and the y-axis represents –log10(P value) of the comparison. Proteins are colored based on enrichment in irMyocarditis cases (red, right side) and controls (blue, left side). Select proteins with an unadjusted p<0.05 are labeled, and those with FDR<0.1 are in bold (two-sided t-test). g, Box plots showing serum concentrations of selected cytokines and chemokines in irMyocarditis cases (n=16, red) and controls (n=10, blue) (two-sided t-test). Data points from a patient with evidence of regressed cardiac metastases (SIC_171) are highlighted in yellow. h, Circulating protein analysis, comparing irMyocarditis cases to controls where each point represents a circulating protein. The x-axis represents the t-statistic in all cases and controls, and the y-axis represents the t-statistic in an analysis that excluded the sample with evidence of cardiac metastases (SIC_171). i, Dot plot showing top marker genes for each lineage in the blood. Dot size represents the percent of cells in the lineage with non-zero expression of a given gene. Color indicates scaled expression across lineages. j, Abundance analysis comparing PBMC samples from irMyocarditis cases prior to the initiation of corticosteroids (“pre-steroid,” n=17, light green) to samples from irMyocarditis patients shortly after the initiation of high-dose corticosteroids (“post-steroid,” n=19, dark green). Left: dots represent logistic regression odds-ratios. Error bars represent 95% confidence intervals. Unadjusted two-sided likelihood-ratio test P values for each subset are shown. Color and bolded P value indicates FDR<0.1. Right: individual-level data showing the frequency of the indicated lineages; each dot represents a patient. k, the number of DEGs for each cell lineage when comparing pre-steroid (right, light green) to post-steroid samples (left, dark green). l, GSEA of select gene sets by heart cell lineages, color-coded by NES of differential gene expression of irMyocarditis pre-steroid samples compared to post-steroid samples. m, Left: timelines of blood collections for irMyocarditis patients included in the pre- and post-steroid analysis (n=24). The length of the lines represents the relative time interval between the pre- and post-steroid samples and the time of steroid initiation. Right: peak troponin (ng/L) and percentage troponin reduction from peak at the time of post-steroid blood collection are shown. Fatal cases are shown in red, and cases are ordered by fatal status followed by interval between steroid initiation and post-steroid sample collection. Full patient metadata is available in Supplementary Table 1. n, The number of DEGs positively (red, right) and inversely (blue, left) correlated with time from steroid initiation in post-steroid samples are shown. o, GSEA of select gene sets when modeled by time from steroid initiation. For g and j, box plots display the median (line) and IQR, with whiskers no more than 1.5x IQR, and each dot represents one patient Related to Fig 1. Abbreviations: Allo, allograft rejection; AP, antigen presentation; CAM, cell adhesion molecules; DNA, DNA synthesis; IFNG, interferon-γ signaling; KEGG, Kyoto Encyclopedia of Genes and Genomes NES, normalized enrichment score; Viral Myo, viral myocarditis.

Extended Data Fig 2. PBMC cell subsets and transcriptional changes associated with irMyocarditis onset.

a, b, c, 135,712 circulating CD8T and NK cells are (a) displayed on a UMAP embedding with (b) a dot plot to show top marker genes for each subset and (c) case versus control abundance analysis. d,e,f, 45,022 circulating CD4T cells are (d) displayed on a UMAP embedding with (e) a dot plot of top marker genes and (f) case versus control abundance analysis. g, h, i, 149,101 circulating cDC, pDC, and MNP cells are (g) displayed on a UMAP embedding with (h) a dot plot of top marker genes and (i) case versus control abundance analysis. j, k, l, 21,171 circulating B and plasmablast cells (j) displayed on a UMAP embedding with (k) a dot plot showing selected marker genes for each subset and (l) case versus control abundance analysis. m, The number of DEGs (FDR<0.1, Wald test) for each cell lineage when comparing pre-steroid irMyocarditis samples (n=17, red, right) to ICI-treated controls (n=28, blue, left). n, Heatmap showing selected differentially expressed genes (irMyocarditis versus control) and gene sets across each PBMC cell subset and lineage, grouped by biological themes. Color scale indicates log2FC difference between irMyocarditis cases and controls for DEG analysis and NES for GSEA. Black dots indicate FDR<0.1 (Wald test). In a, d, g, and j, UMAP pseudocoloring corresponds to the cell subsets labeled on the right. In b, e, h, and k, dot plots show top marker genes for each cell subset. Dot size represents the percent of cells in the subset with non-zero expression of a given gene. Color indicates scaled expression across subsets. For c, f, i, and l, cell subset differential abundance analysis comparing pre-steroid samples from irMyocarditis cases at the pre-steroid timepoint (n=17, red) to ICI-treated controls (n=28, blue). Left panels: dots represent logistic regression odds-ratios. Error bars represent 95% confidence intervals. Unadjusted two-sided likelihood-ratio test P values for each subset are shown. Color and bolded P value indicates FDR<0.1. Right panels: box plots display the median (line) and IQR, with whiskers no more than 1.5x IQR. Each dot represents one patient. Abbreviations: Allo, allograft rejection; AP, antigen presentation; CAM, cell adhesion molecules; CC, co-inhibition or co-stimulation; CS; cytokine signaling; GG, glycolysis and gluconeogenesis; IFNA, interferon-α signaling; IFNG, interferon-γ signaling; ISG; interferon-stimulated genes; KEGG, Kyoto Encyclopedia of Genes and Genomes; NES, normalized enrichment score; TF, transcription factors; TGFB; TGF-β signaling; Viral Myo, viral myocarditis. Related to Fig 1.

Extended Data Fig 3. Blood CITE-Seq surface protein analysis.

a, b, Scatter plots showing transformed counts of the listed proteins in our CITE-Seq data. Boxes and arrows indicate the gating strategy used to define the populations labeled in red and analyzed in panels c and d. The same data are shown pseudocolored by the lineage assigned to those cells by (a) scRNA-seq and (b) as density plots. c, Bar plots showing the percent of the cells in each indicated CITE-Seq-defined lineage that were assigned to each scRNA-seq-defined blood lineage (rows). d, Abundance analysis comparing the frequencies of CITE-seq defined populations from pre-steroid irMyocarditis cases (n=17, red) to ICI-treated controls (n=28, blue). Left: dots represent logistic regression odds-ratios. Error bars represent 95% confidence intervals. Unadjusted two-sided likelihood-ratio test P values for each population is shown. Color and bolded P value indicates FDR<0.1. Right: box plots display the median (line) and IQR, with whiskers no more than 1.5x IQR. Each dot represents one patient. Composition is reported as the percent of all PBMCs from a patient in each subset. The last row, “CD8/NK” is the sum of the “CD8T” and “NK” gates, included to mirror the scRNA-seq lineage level clustering solution. Related to Fig 1.

Extended Data Fig 4. Lymphoid cells in irMyocarditis tissue.

a, Selected marker genes for each T/NK cell subset. Dot size represents the percent of cells in the subset with non-zero expression of a given gene. Color indicates scaled expression. b, Feature plots using color to indicate gene expression (logCPM) levels of the indicated genes projected onto the T/NK UMAP embedding. Cell numbers and percentages represent gene expression across all heart T and NK cells. c, Feature plots using color to indicate gene expression (logCPM) levels of the indicated genes projected onto the heart h-CD4T^IL7R,LTB subset. Cell numbers and percentages represent gene expression in this subset only. d, UMAP embedding of cell density displaying the relative proportion of cells from irMyocarditis cases (n=4,686) and controls (n=4,448). e, Stacked bar plots showing the per-subset cellular composition per donor of each pre-steroid or unenriched control sample. Red patient labels denote samples from patients with fatal irMyocarditis. f, Plot showing the intracardiac frequency of h-CD8T^cycling cells (y-axis) versus serum troponin T level (x-axis) for irMyocarditis samples (n = 12). Linear regression p=0.018, FDR<0.1. g, Feature plots using color to indicate gene expression (logCPM) levels of the indicated genes expressed by control (left) or irMyocarditis samples (right), projected onto the heart T/NK UMAP embedding. Cell numbers and percentages represent gene expression across T/NK cells in control or irMyocarditis samples. h, Top: heatmap showing selected DEGs across T/NK subsets from irMyocarditis versus control heart samples, grouped by biological categories. Color scale indicates log2FC. Black dots indicate FDR<0.1 (Wald test). Bottom: select GSEA results, color-coded by NES of the gene set in irMyocarditis cases versus controls. “All T” row depicts pseudo-bulk DGE analysis of pooled T-cell subsets (subsets 2–6) that excluded NK cells. Bolded genes in the heatmap indicate leading edge genes in ≥1 GSEA gene set in ≥4 subsets. i, Bar graph showing the number of DEGs per lineage when gene expression was modeled against serum troponin T level, showing positively correlated (right, red) and negatively correlated (left, blue) genes. Only lineages with at least one troponin T-associated DEG are displayed. j, k, The 33 expanded TCR-β sequences (> 0.5% of TCR-β repertoire) from patient SIC_264 in T/NK UMAP space, color coded by whether the cell was found prior to (“pre-steroid”, red) or after (“post-steroid”, blue) administration of corticosteroids and second-line immunosuppression. Data shown by (j) TCR-β clone and (k) in aggregate of all clones. Abbreviations: Allo, allograft rejection; AP, antigen presentation; CAM, cell adhesion molecules; CC, co-inhibition or co-stimulation; DNA, DNA synthesis; HM, Hallmark; IFNG, interferon-γ signaling; IS, immunoreceptor signaling; ISG, interferon-stimulated genes; K, Kyoto Encyclopedia of Genes and Genomes; MA, motility and adhesion; NES, normalized enrichment score; TCR Sig, TCR signaling; TF, transcription factors; Viral Myo, viral myocarditis. Related to Fig 2.

Extended Data Fig 5. Expanded T-cell receptor β (TCR-β) sequences in irMyocarditis tissue and tumor.

a, A box plot showing the relative proportion of cells that recovered a TCR-β CDR3 sequence from marked areas of irMyocarditis (n=6, red) and control (n=3, blue) tissue; p=0.002 by two-sided t-test (via Adaptive Biotechnologies). Median and IRQ are shown, with whiskers no more than 1.5x IQR. Each dot represents one patient. b, Expanded TCR-β CDR3 (left) and total unique TCR-β CDR3 sequences (right) recovered on a per patient basis from both scRNA-seq and bulk TCR-β sequencing. c, Expanded TCR-β CDR3 from bulk sequencing (top) and scRNA-seq data (bottom) from irMyocarditis patients with matched blood samples. Donors with “healing” irMyocarditis were excluded. d, e, Smoothed Hill’s diversity index curves at diversity orders 0–4 for the TCR-β repertoires of irMyocarditis tissues (d) without diffuse metastases, colored by histologic appearance at the time of autopsy, and (e) with an additional case with diffuse cardiac metastases (SIC_232) colored by histologic appearance at the time of autopsy and by the presence of cardiac metastases (SIC_232 and SIC_136). f, Within each tissue type in each patient (“control”, “tumor”, or “irMyocarditis”), the frequency of each TCR-β clone is plotted on a per-patient basis and labeled by the pathological designation of the macroscopically dissected regions (SIC_17: Active; SIC_136: Borderline; SIC_3: Healing; SIC_175: Healing). Each point represents a TCR-β clone. For the left and middle columns, the y-axis represents the proportion of a given TCR-β clone in the patient’s control tissue repertoire, and the x-axis represents the proportion of the TCR-β clone in their (left column) tumor TCR-β repertoire or (middle column) irMyocarditis TCR-β repertoire. Right column: proportion of TCR-β clone in the tumor is shown on the y-axis and the proportion of that clone in irMyocarditis is shown on the x-axis. Points are pseudocolored to represent a TCR-β clone that was expanded (> 0.5% of the heart or tumor repertoire) and enriched (Fisher’s exact test FDR < 5% compared to control) in heart (green), tumor (purple), both tissues (red), or neither tissue (grey). g, Each TCR-β clone enriched in heart or tumor relative to controls (FDR<0.05, Fisher’s exact test) is plotted according to its proportion (among all TCR-β clones in the respective tissue) in heart (x-axis) and tumor tissue (y-axis), normalized by its proportion in control tissue. Each plot shows the enriched TCR-β clones within each donor projected onto the aggregate data across all donors (Fig 3b-c). h, TCR-β CDR3 amino acid sequences for select GLIPH groups. Highlighted in red are the predicted enriched motifs. i, The same data as shown in g, with color indicating if the corresponding TCR-β was (red; “true”) or was not (grey; “false”) found in a GLIPH group. j, The frequency of each expanded GLIPH group in heart and tumor tissue was calculated and then normalized by dividing by the frequency of that same GLIPH group in control tissue. Normalized GLIPH group frequencies for heart (x-axis) and tumor (y-axis) are plotted. Color indicates the donor from which each GLIPH group was found, and the label shows the associated amino acid motif of the GLIPH group. Throughout the Figure, red patient labels denote cases of fatal irMyocarditis. Related to Fig 3.

Extended Data Fig 6. TCR clonotype sharing and antigen recognition screen.

a, Blood CD8T/NK UMAP highlighting circulating cells (in red) that express a TCR-β sequence found to be expanded in irMyocarditis hearts (combined scRNA-seq and bulk TCR-β sequencing data). b, A volcano plot showing the results of a logistic regression model investigating the likelihood of a cell in a given CD8T/NK cell subset from the blood containing a TCR-β CDR3 sequence that was expanded in heart tissue (n=13). Points represent odds ratios for each cell subset. Red points denote cell subsets with statistically significant sharing (FDR<0.05, two-sided likelihood-ratio test). Error bars represent 95% confidence intervals. c,d, Cells in blood for which the same TCR-β was expanded in paired irMyocarditis heart and blood sample from the same patient are shown in red projected on the (c) CD8T/NK blood UMAP embedding and (d) CD4T blood UMAP embedding. Each patient with paired heart and blood samples is shown. e, UMAP of heart T and NK cells highlighting cells that express expanded TCR-β CDR3 sequences that were found in CD8T/NK blood cells on a per-patient basis. f, Feature plots using color to indicate gene expression (logCPM) levels of the indicated genes projected onto the heart T/NK UMAP embedding. Cell numbers and percentages represent gene expression across all heart T and NK cells. g, The flow cytometry gating strategy for the CD137 expression assay is shown. Lymphocytes are analyzed as either CD8⁺ or CD8^- (to assess for TCRs isolated from CD8 or CD4T cells, respectively) and then gated on intensity of Far Red, CT-Violet, or CT-CFSE staining; each combination of stains represents a unique TCR. Each TCR is then gated on mTRBC⁺ to identify cells expressing the transduced TCR construct. Each row of histograms represents a condition of Epstein-Barr virus-immortalized lymphoblastoid cell lines (EBV-LCLs), which served as antigen presenting cells, pulsed with target peptides/peptide pools, and CD137 expression is shown. PMA/Iono serves as a positive control; “DMSO,” “Ova peptide,” and “CEF” represents negative controls; RINATLETK is an α-myosin peptide known to be recognized by the positive control TCR (“Myo-TCR”)¹⁶; α-myosin pool 4 contains a 20-aa peptide with the RINATLETK sequence; and all other pools are considered test peptide pools. Irrelevant TCRs are TCRs from an unrelated donor⁷⁶ and serve as negative controls along with untransduced (UT) T cells. h, A heatmap showing the background subtracted CD137 expression of each TCR and peptide/peptide pool combination. Each column represents a unique TCR, color coded by patient and then grouped by CD8 and CD4 TCRs. Each row represents peptides or peptide pools tested; these include the controls and pools covering the full length of the α-myosin, troponin I (TnI), and troponin T (TnT) proteins. α-myosin pool 11 could not be evaluated for TCR63 and is shown as empty space on the heatmap. In c, d, and e, red patient labels denote cases of fatal irMyocarditis. Abbreviations: CEF, Cytomegalovirus, Epstein-Barr virus, and influenza virus; EVB-LCLs, Epstein-Barr virus-immortalized lymphoblastoid cell lines; HLA, human leukocyte antigens; PMA/Iono, Phorbol 12-myristate 13-acetate/ionomycin; mTRBC, murine T-cell receptor β chain; TCR, T-cell receptor; TnI, troponin I; TnT, troponin T; UT, untransduced. Related to Fig 3.

Extended Data Fig 7. MNP populations in heart and blood.

a, Selected marker genes for each MNP subset. Dot size represents the percent of cells in the subset with non-zero expression of a given gene. Color indicates scaled expression. b, Feature plots using color to indicate gene expression (logCPM) levels of the indicated genes projected onto the heart MNP UMAP embedding. Cell numbers and percentages represent gene expression across all heart MNP cells. c, Embedding of cell density plot displaying the relative proportion of cells from irMyocarditis cases (n=3,606) and controls (n=6,218). d, Stacked bar chart depicting the relative contributions of cells in each MNP subset (colored coded on the right) on a per donor basis from each pre-steroid irMyocarditis or unenriched control sample. e, Abundance analysis comparing intramyocardial frequencies of cell populations from pre-steroid irMyocarditis cases (n=12, red) versus control (n=8, blue). Left: dots represent logistic regression odds-ratios. Error bars represent 95% confidence intervals. Unadjusted two-sided likelihood-ratio test P values for each subset are shown. Statistics for h-pDC^{LILRA4, IRF8} are not included due to extremely low recovery from this cell subset. Right: box plots display the median (line) and IQR, with whiskers no more than 1.5x IQR. Each dot represents one patient. f, Representative image following immunohistochemical staining of irMyocarditis heart tissue section (n=6) for CD1c (brown) and hematoxylin nuclear counterstaining (blue). g, h, Comparison of (g) cDCs (CLEC9A/CD1c⁺) and (h) CD8T cell density as measured by immunofluorescence staining in non-inflamed regions (left column) and inflamed regions (right column) of irMyocarditis heart sections (n=6); one-sided t-test P values are shown. i, Bar graph showing average distance from each cDC to the nearest CD8 T cell (per section), with each row representing an individual case (red) or control (blue); the control slide labeled “NC” (no cells) had neither CD8A+ nor CLEC9A/CD1c+ cells detected. j, Heatmap showing select DEGs and GSEA pathways for irMyocarditis versus control heart tissue grouped by biological themes. The “All MNP” row depicts DEG results from pooling MNP subsets 1–5. Color scale indicates log2FC between irMyocarditis cases and controls and NES for GSEA. Black dots indicate FDR < 0.1 (Wald test). Genes in bold were a part of the leading edges genes of the displayed GSEA pathways. k, UMAP Feature plots using color to indicate gene expression (logCPM) levels of CXCL10 expressed by control (top) or irMyocarditis samples (bottom). Cell numbers and percentages represent expression across control or irMyocarditis MNP cells. Abbreviations: Allo, allograft rejection; AP, antigen presentation; CAM, cell adhesion molecules; CC, co-inhibition or co-stimulation; CS, cytokine signaling; DNA, DNA synthesis; IFNG, interferon-γ signaling; INF, inflammasome; ISG, interferon-stimulated genes; KEGG, Kyoto Encyclopedia of Genes and Genomes; MA, motility and adhesion; NES, normalized enrichment score; TF, transcription factors; Viral Myo, viral myocarditis. Related to Fig 4.

Extended Data Fig 8. Non-immune populations in irMyocarditis heart tissue.

a, UMAP embedding of 65,409 non-immune cells isolated from the heart, segregated by endothelial cells (top) and non-endothelial cells (bottom), and colored by the 18 defined cell subsets labeled on the right. b, Dot plot showing top marker genes for each non-immune subset. Dot size represents the percent of cells in the subset with non-zero expression of a given gene. Color indicates scaled expression across subsets. c, Abundance analysis comparing intramyocardial frequencies of cell populations from pre-steroid irMyocarditis heart samples (n=12, red) versus controls (n=8, blue). Left: dots represent logistic regression odds-ratios. Error bars represent 95% confidence intervals. Unadjusted two-sided likelihood-ratio test P values for each subset are shown. Color and bolded P value indicates FDR<0.1 Right: box plots display the median (line) and IQR, with whiskers no more than 1.5x IQR. Each dot represents one patient. d, Feature plots using color to indicate marker gene expression (logCPM) levels of the indicated genes projected onto the heart non-immune UMAP embedding. Cell numbers and percentages represent gene expression across all heart non-immune cells. e, Feature plots showing (top panel) the number of genes expressed by each cell and (bottom panel) percent mitochondrial UMIs on the UMAP of non-immune cells derived from heart scRNA-seq data. f, Embedding of cell density plot displaying the relative proportion of cells from irMyocarditis cases (n=20,439) and controls (n=44,970). g, Stacked bar chart depicting the relative contributions of cells in each non-immune cell subset (color coded to the right) from each pre-corticosteroid irMyocarditis or unenriched control sample. h, Top: the number of DEGs for each cell subset when comparing irMyocarditis cases (up, red) to control (down, blue). Bottom: heatmap showing select DEGs for irMyocarditis versus control heart tissue grouped by biological themes. Color scale indicates log2FC difference between irMyocarditis cases and controls. Black dots indicate FDR < 0.1 (Wald test). Abbreviations: AP, antigen presentation; CC, co-inhibition or co-stimulation; CS, cytokine signaling; ISG, interferon-stimulated genes; MA, motility and adhesion; TF, transcription factors. Related to Fig 5.

Extended Data Fig 9.

a, Feature plots using color to indicate marker gene expression (logCPM) levels of the indicated genes projected onto the heart fibroblast UMAP embedding. Cell numbers and percentages represent gene expression across all heart fibroblast and myofibroblast cells. b, Abundance analysis comparing intramyocardial frequencies of cell populations from pre-steroid irMyocarditis heart samples (n=12, red) versus controls (n=8, blue). Left: dots represent logistic regression odds-ratios. Error bars represent 95% confidence intervals. Unadjusted two-sided likelihood-ratio test P values for each subset are shown. Right: box plots display the median (line) and IQR, with whiskers no more than 1.5x IQR. Each dot represents one patient. c, Volcano plot of the genes within the fibroblast lineage modeled by serum troponin T, with select genes highlighted. Colored points represent FDR < 0.1 (two-sided Wald test). d, Summary of key findings. Clusters whose abundances correlated with serum troponin T levels were considered correlates of disease severity. Abbreviations: FB, fibroblast. Related to Fig 5.

Extended Data Fig 10. Flow QC.

a, Pseudocolor plots showing the applied sequential gating strategy to sort live cells for downstream scRNA-seq profiling from a representative myocardial sample. Numbers indicate the percentage within the indicated gate. DAPI^-CD235a^- cells were collected for downstream analysis. Abbreviations: BSC-A, back scatter-area; FSC-A, forward scatter-area; FSC-H; forward scatter-height; DAPI, 4’,6-diamidino-2-phenylindole. Related to all Figures; Methods.

Data availability

scRNA-seq count matrices from this study and related data as well as TCR sequencing data is deposited in the GEO database (under accession #GSE228597), and raw human sequencing data is available in the controlled access repository dbGaP (https://www.ncbi.nlm.nih.gov/gap; accession phs003413.v1.p1). Public Heart Atlas data¹⁸ utilized as heart controls in our study are accessible via the Human Cell Atlas (HCA) Data Coordination Platform (DCP) with accession number: ERP123138 (https://www.ebi.ac.uk/ena/browser/view/ERP123138). Reads from scRNA-seq experiments were aligned to the human reference genome (GRCh38, v3.0.0 from 10x Genomics).

Code availability

Source code for data analysis is available on GitHub (https://github.com/villani-lab/myocarditis) and has been archived to Zenodo (https://zenodo.org/doi/10.5281/zenodo.11519192)⁷⁹. A full list of software packages and versions included in the analyses is included in Supplementary Table 21. A user-friendly portal⁸⁰ is available to browse the single-cell data generated in this manuscript at https://villani.mgh.harvard.edu/myocarditis.