Autopsies were performed and tissues were collected as previously described (6) in the National Cancer Institute’s Laboratory of Pathology at the National Institutes of Health Clinical Center following consent of the legal next of kin.

RNAscope was performed as previously described at the NIH (6) except 3,3’-diaminobenzidine (DAB) was used as a chromogen.

RNAlater (Invitrogen)-preserved autopsy tissues were homogenized and processed as previously described (6). Total RNA was extracted using RNeasy Mini, RNeasy Fibrous Tissue Mini, and RNeasy Lipid Tissue Mini Kits (Qiagen). The QX200 AutoDG Droplet Digital PCR (ddPCR) System (Bio-Rad) was used to detect SARS-CoV-2 RNA using the SARS-CoV-2 ddPCR Kit (Bio-Rad) (6). For samples to be considered positive for SARS-CoV-2 nucleocapsid 1 (N1) or 2 (N2) genes, the manufacturer’s limit of detection of >/= 0.1 copies/µl final ddPCR reaction and 2 positive droplets per well was required.

We performed immunohistochemical stains on formalin-fixed paraffin-embedded three micron tissue sections using an automated immunostainer Benchmark Ultra (Roche) with an Ultra-View Universal DAB brown detection kit (alongside an Ultra-View Universal AP Red detection kit for double staining). These antibodies were used for the following targets: CD3 (clone 2GV6, predilute Roche Cat # 790-4341), CD4 (clone SP35, predilute Roche Cat# 790-4423), CD8 (clone SP57, predilute Roche Cat#790-4460), granzyme B (clone GrB-7+D170, dilution 1:100, Millipore Cat# MAB 3070), perforin (clone KM585 PI-8) (dilution 1:200, Leica Cat# NCL-L-perforin) detection with Optiview kit (Roche), CD68 (clone KP1, predilute Roche Cat#790-2931), CD163 (clone MRQ-26, predilute Roche Cat#760-4437).

cDNA preparations were processed for viral capture high-throughput sequencing (VirCapSeq-VERT (7)) with an iScript cDNA synthesis kit (Bio-Rad). Library construction used Kapa HyperPlus kits (Roche) with enzymatic fragmentation to 300 bp average size. Individually dual-barcoded libraries were pooled and 150 cycle single-pass reads generated (Ilumina NextSeq 550). Per barcode, 7 – 10 million raw reads were obtained. Fastq files were adapter trimmed (Cutadapt v 3.0) (8) and quality reports generated (FastQC v 0.11.5) (9). Reads were quality filtered and end trimmed with PRINSEQ (v 0.20.3) (10). Host reads were removed by mapping against a human reference database using Bowtie2 (11). The remaining reads were subjected to homology search using NCBI Megablast against the viral GenBank nucleotide database. The blast reports were annotated with taxonomic information to assess read frequencies for viral families, genera, species and individual GenBank sequence accession numbers. Human endogenous retroviruses served as positive technical controls.

Genomic DNA was extracted from formalin-fixed paraffin-embedded (FFPE) tissue using a QIAamp DNA FFPE kit. Complementarity determining regions (CDR)3 of T-cell receptor (TCR)β chains present in the human tissue samples were sequenced in a high-throughput manner using the immunoSEQ assay (12) after amplification of the extracted DNA in a bias-controlled multiplex PCR. The resulting CDR3 sequences were collapsed and filtered to quantify the absolute abundance and frequency of each unique TCRβ CDR3 region with Adaptive Biotechnologies’ pipeline (13). We used CD-HIT (14) to test whether the most frequently observed CDR3 sequences derived from the patient samples clustered with the SARS-CoV-2-specific TCR sequences from ImmuneCODE (13). We generated the clonality and repertoire overlap figure using Immunarch (15).

Typing was performed as previously described (16).

A previously healthy 26-year-old man presented to the hospital with complaints of chest pressure and shortness of breath for 1 week and nausea, vomiting, and chills for 3 days. He was afebrile, had a heart rate of 80 beats per minute, and was normotensive with an unremarkable physical examination. SARS-CoV-2 PCR testing of a nasopharyngeal (NP) swab specimen was negative. Basic metabolic panel and complete blood count values were within normal range although troponin I and C-reactive protein were elevated to 21.2 ng/ml (ref: 0-0.045) and 9.84 mg/L (ref: 0-3), respectively, indicative of myocardial injury and inflammation. Electrocardiogram demonstrated inferolateral ST changes concerning for myocardial ischemia ( Figure S1A ). Chest X-Ray showed no abnormalities ( Figure S2 ). The patient was admitted to the medical ward for non-ST segment elevation myocardial infarction with plans for coronary angiogram in the morning.

The following morning, a transthoracic echocardiogram demonstrated an ejection fraction (EF) of 60-65%, without wall motion or valvular abnormalities. Coronary angiogram demonstrated mild non-obstructive coronary artery disease ( Figure S3 ) and left ventriculogram showed intact function without regional wall motion abnormalities or mitral regurgitation. A repeat NP swab for SARS-CoV-2 was negative. Post-cardiac catheterization he experienced ongoing chest pain with rising troponin I levels (43.7 ng/ml), and his electrocardiogram was unchanged. Colchicine and ibuprofen were initiated for presumptive myopericarditis.

In the setting of recurrent chest pain, an electrocardiogram demonstrated persistent ST segment changes concerning for ischemia ( Figure S1B ). Repeat troponin I level was 83.5 ng/ml, and lactate was 3.0 mmol/L (ref: 0.7-2.0). He developed sustained ventricular tachycardia (VT) progressing to pulseless VT arrest. Return of spontaneous circulation was achieved. However, post-arrest point-of care echocardiogram showed global left ventricular hypokinesis with an estimated EF <10%. He was started on an epinephrine infusion and cannulated for veno-arterial extracorporeal membrane oxygenation (ECMO) by cardiothoracic surgery at bedside. He was transferred to a tertiary care ECMO facility, where his NP sample again tested negative for SARS-CoV-2. Shortly after arrival he suffered a pulseless electrical activity arrest with profound vasodilation and died.

Cardiac histopathological findings featured microscopic foci of contraction band myocyte necrosis ( Figure 1A ) and extensive lymphocytic infiltration reflecting severe myocarditis ( Figure 1B ). The lymphocytic infiltrate colocalized with dying myocytes and was more extensive in the right ventricle (RV) than the left ventricle (LV). Immunohistochemical studies were performed on a section from the RV. The vast majority of lymphocytes were T-cells ( Figure 1C ). Both CD4+ and CD8+ T-cells were present ( Figure 1D ). CD8+ cells were positive for perforin ( Figure 1E ), and granzyme B ( Figure 1F ). Only rare CD20+ B cells and no CD56+ natural killer cells were observed. Macrophages staining with CD68 ( Figure 1G ) and CD163 (marker of M2 tissue repair phenotype; Figure 1H ) outnumbered CD3+ T-cells.

Extracardiac findings included mild pneumonia with pulmonary edema and mild interstitial pneumonitis. Sinusoidal histiocytosis of pulmonary hilar lymph nodes was observed, and lymphoid depletion was noted in the spleen.

Assays for SARS-CoV-2 RNA in cardiac tissues by RNAscope and ddPCR were negative and below the limit of detection, respectively. However, we detected SARS-CoV-2 RNA in the mucosal epithelium of the patient’s appendix ( Figure S4 ) and duodenum by RNAscope, and in the ileum by ddPCR ( Table S1 ). We used VirCapSeq-VERT to test for the presence of other viruses in myocardial and pericardial tissues and found none ( Table S2 ).

We sequenced TCRs in cardiac tissues and found a high degree of TCR repertoire overlap across RV, LV, and intraventricular septum ( Figure 2A ). We observed three hyperexpanded clones across these tissues ( Table S3 ; Figure 2B ). The most dominant clonotype (clonotype1) comprised 27-59% of T-cells across tissues, while clonotypes 2 and 3 comprised 2.5-4.3% and 1.0-1.7% of T-cells, respectively ( Table S3 ).

An epitope is often recognized by many distinct TCRs with highly similar CDR3 sequences (17). Therefore, we applied a widely-used CD-HIT clustering approach (14) to assess whether the CDR3 sequences of clonotypes1-3 ( Table S4 ) were highly similar to the 149,845 unique TCRβ CDR3 sequences that have been verified to bind to SARS-CoV-2 epitopes and catalogued in ImmuneCODE (13). Clonotypes1-3 had nine, one, and two matches with >90% amino acid sequence identity to TCRs in ImmuneCODE, respectively ( Table 1 ). No CDR3 sequences from clonotypes with a frequency <1% in myocardial tissues had an ImmuneCODE hit.

The CDR3 sequences clustered with clonotype1 have been shown to bind 15 unique epitopes from three SARS-CoV-2 antigens (ORF1ab, spike, ORF7b) ( Table S5 ). CDR3 sequences clustered with clonotypes2 and 3 have been shown to bind 19 unique SARS-CoV-2 epitopes from ORF10, ORF1ab, and ORF7b. The Multiplex Identification of Antigen-Specific T-Cell Receptors Assay (MIRA), used to define the list of epitopes in ImmuneCODE (13), screens for a finite set of epitopes per unique TCR sequence. Given this, we queried The Immune Epitope Database (IEDB) and identified an additional 26 well-characterized SARS-CoV-2 epitopes that shared >90% amino acid sequence similarity with ImmuneCODE epitopes ( Table S6 ) (18).

Lastly, to identify possible cross-reactive human epitopes that could exacerbate the patient’s T-cell infiltrate, we queried whether there were human epitopes that matched the ImmuneCODE SARS-CoV-2 epitopes with >90% amino acid similarity. We found seven human epitope matches ( Table S7 ). Three of these epitopes originate from plectin, a cytoskeletal protein that maintains tissue integrity and has been previously associated with myocarditis and dilated cardiomyopathy (19).