key: cord-0326845-ulvcdsmv
authors: Mantri, Madhav; Hinchman, Meleana M.; McKellar, David W.; Wang, Michael F. Z.; Cross, Shaun T.; Parker, John S. L.; De Vlaminck, Iwijn
title: Spatiotemporal transcriptomics reveals pathogenesis of viral myocarditis
date: 2022-05-11
journal: bioRxiv
DOI: 10.1101/2021.12.07.471659
sha: 33c38380a179c2c837965323ec86484700f5c898
doc_id: 326845
cord_uid: ulvcdsmv

A significant fraction of sudden death in children and young adults is due to myocarditis, an inflammatory disease of the heart, most often caused by viral infection. Here we used integrated single-cell and spatial transcriptomics to create a high-resolution, spatially resolved map of reovirus-induced myocarditis in neonatal murine hearts. We assayed hearts collected at three timepoints after reovirus infection and studied the temporal, spatial, and cellular heterogeneity of host-virus interactions. We further assayed the intestine, the primary site of reovirus infection to establish a full chronology of molecular events that ultimately lead to myocarditis. We implemented targeted enrichment of viral transcripts to establish the cellular targets of the virus in the intestine and the heart. Our data give insight into the cell-type specificity of innate immune responses, and into the transcriptional states of inflamed cardiac cells that recruit circulating immune cells, including cytotoxic T cells which induce pyroptosis in the myocarditic tissue. Analyses of spatially restricted gene expression in myocarditic regions and the border zone around those regions identified immune-mediated cell-type specific injury and stress responses. Overall, we observe a dynamic and complex network of cellular phenotypes and cell-cell interactions associated with viral myocarditis.

INTRODUCTION 27 Viral infection is the most common cause of myocarditis 1,2 . The resulting inflammatory 28 cardiomyopathy can lead to arrhythmias, dilated cardiomyopathy, and death 1,3,4 . In humans, 29 viral myocarditis is challenging to study because of the low sensitivity of available diagnostic 30 testing, the acute onset of the disease, the focal nature of the disease, and the extreme 31 heterogeneity of immune-virus interactions in complex cardiac tissues 4-6 . In mice, mammalian 32 orthoreovirus offers a flexible model system 7 . After oral inoculation, the Type 1 Lang (T1L) 33 reovirus strain initially infects the gastrointestinal tract. Within days the infection then spreads 34 to secondary sites in the body, including the heart, leading to myocarditis in up to 50% of 35 infections 7-9 . Yet, even in this mouse model, the molecular pathogenesis of viral myocarditis is 36 difficult to study because of the complex network of cardiac and immune cell types involved 37 and the cellular, spatial, and temporal heterogeneity of the disease 2,10 . Consequently, neither 38 the cell types that are responsible for the innate immune response, nor the cell types that are within infected cardiac tissues is needed to address these knowledge gaps. 47 Here we used integrated single-cell and spatially resolved RNA-sequencing (RNA-seq) to 48 study the cellular and spatial heterogeneity of myocarditic processes in the hearts of reovirus- 49 infected neonatal mice at multiple time points after infection. We also applied these 50 technologies to study the innate response to reovirus infection in the intestine. In addition, we 67 To elucidate the pathogenesis of reovirus-induced myocarditis, we analyzed heart tissues 68 collected from neonatal mice infected orally with either the T1L strain of reovirus or a mock 69 control (Methods, Fig. 1A) . We generated scRNA-seq data for 31,684 cells from infected damage identified by H&E staining (Fig. 1C, Supp Fig. 2A-B) . The combination of scRNA-seq 80 and spatial transcriptomics allowed us to resolve and visualize cell types and gene expression 81 in a spatial context (Supp Fig. 2C ). Because the virus first infects the gastrointestinal tract 82 before it spreads to other body sites including the heart, we also performed scRNA-seq and 83 spatial transcriptomics on ileum. We obtained 7,695 single-cell transcriptomes and 8,027 84 spatial spot transcriptomes for ileum from mock and infected samples at 1 and 4 dpi (Fig. 1D , 85 Supp Fig. 3A-D) . 86 To faithfully identify reovirus transcripts in the ileum and heart, which are not polyadenylated, 87 we performed hybridization-based enrichment of viral fragments captured in the scRNA-seq 88 libraries (Methods, Supp Fig. 4A-C) . In the ileum, we captured a total of 13,100 unique viral 89 transcripts, with viral load decreasing from 1 dpi to 4 dpi. At 1 dpi, entero-endocrine cells had 90 the highest fraction of infected cells followed by enterocytes and goblet cells, all of which are 91 present in the gut epithelium. Lymphatic endothelial cells were infected at 4 dpi, suggesting 92 that the virus reaches the bloodstream via lymphatic drainage to allow transmission of the virus 93 to secondary sites in the body, including the heart, as shown before 12 (Supp Fig. 4D, Fig 1E) . 94 We captured 2,762 unique viral transcripts from 392 cells in the T1L-infected hearts. The viral 95 load first increased from 4 dpi to 7 dpi and then decreased from 7 dpi to 10 dpi, consistent with 96 viral titer assays performed on whole hearts 9,13 (Fig. 1E, Supp Fig. 4E ). Endocardial and 97 endothelial cells were the most frequently infected cell types at 4 dpi, suggesting that 98 endocardial cells lining the ventricular lumen and endothelial cells lining the cardiac 99 vasculature are among the first cells to be infected (Fig. 1E) . We detected an increased 100 infection in endothelial cells from 4 dpi to 7 dpi, consistent with viral titer assays performed on 101 whole hearts 9,13 (Fig. 1E, Supp Fig. 4E) . We further detected viral transcripts in neutrophils, 102 dendritic cells, and T cells in the 7 dpi heart (Fig. 1E, Supp Fig. 4E ). This observation 103 suggests that antigen-presenting cells and immune cells may contribute to the spread of 104 infection to other organs in the body. The role of infected dendritic cells in bringing more 105 reovirus to the cardiac tissue during systemic infection has been discussed previously 8 . 106 To validate these observations, we performed histology, multiplexed RNA fluorescence in-situ 107 hybridization (FISH), and immunofluorescence assays on tissue sections from myocarditic 108 hearts and controls (multiple infected mice litters, Supp Fig. 5A-5B, Methods) . We used RNA-109 FISH to visualize expression of genes specific to cardiomyocytes, fibroblasts, endothelial cells, 110 macrophages, neutrophils, and T cells (Supp Fig. 5C & 5E, Methods) . On consecutive tissue 111 sections, we labeled reovirus antigen using immunofluorescence to identify reovirus infected 112 cells (Supp Fig. 5C) . These experiments revealed infection foci and immune infiltration in 113 myocarditic regions. We found macrophages and neutrophils in the border zones and outside 114 the myocarditic regions at 7 dpi (Supp Fig. 5C, 5D) . In contrast, most T cells were found inside 115 the myocarditic regions at 7 dpi (Supp Fig. 5D ). Endothelial cells that were positive for the To detect early transcriptional differences in the cardiac tissue after infection, we performed 137 Differential Gene Expression Analysis (DGEA, mock vs infected hearts at 4 dpi, Methods). 138 This analysis revealed a significant upregulation of 226 genes in the infected heart (two-sided 139 Wilcoxon test, log fold-change > 1.0 and p-value < 0.01), including genes related to the 140 interferon-β pathway, interferon signaling, and innate immune responses (Supp Fig 6A-6B (Fig. 2B) . In response to infection, an increase in IR was observed for all cardiac cell types, 147 but the greatest increase in IR was observed for endothelial cells (Fig. 2B) . These data 148 suggest that endothelial cells lining the cardiac vasculature are important initiators of the host 149 defense to viral infection. Comparison of IR scores using the spatial transcriptomic data 150 showed increased IR scores in the infected hearts at 4 and 7 dpi with the highest scores found 151 in myocarditic regions (Fig. 2C) . Given our observation that endothelial cells within the heart 152 had the highest IR score in the absence of infection, we asked if this observation was unique 153 to heart tissue or was a more general phenomenon. To this end, we used the Tabula Muris 154 scRNA-seq mouse atlas 14 and estimated the IR of ~16,000 cells of five major cell types 155 (epithelial cells, fibroblasts, endothelial cells, smooth muscle cells, and mesenchymal cells) 156 across 10 different organs and tissues. This analysis revealed that endothelial cells 157 consistently had the highest IR score across all tissues in mice (Fig. 2D) . These results 158 indicate that endothelial cells lining the vasculature have a higher basal expression of innate 159 response genes within most tissues, which may prime these cells to respond to viral 160 dissemination within the blood and lymphatics. 161 To investigate the cell-type-specific IR in the ileum, the primary site of reovirus infection, we 162 performed DGEA on reovirus-infected and mock-infected ileal cells at 1 dpi and found a 163 significant upregulation of 438 genes (two-sided Wilcoxon test, log fold-change > 1.0 and p-164 value < 0.01), related to the interferon-beta pathway, interferon signaling, and innate immune 165 responses in reovirus-infected ileal cells (Supp Fig. 6C-6D) . We computed an IR score using 166 this module of 438 genes and observed higher basal IR scores in enterocytes and entero-167 endocrine cells as compared to other ileal cell types (Fig. 2E) . Enterocytes further showed the 168 highest increase in IR score after infection, followed by entero-endocrine, endothelial, and 169 lymphatic cells (Fig. 2E) . Comparison of IR scores for spatial transcriptomic data further 170 supported our analysis of the scRNA-seq data, showing increased IR scores in the infected 171 ileum at 1 and 4 dpi with the highest scores evident within intestinal mucosa and villi (Fig. 2F) . 172 The intestinal epithelial cells must tolerate commensal microorganisms present in the lumen of 173 the gut and yet still be responsive to invasive pathogens. Our data suggest that to achieve this, test, log fold-change > 1.0 and p-value < 0.01) in the reovirus-infected heart as compared to the mock-infected 184 heart at 4 dpi. C) Infection response score (defined above) across spots in spatial transcriptomics data. D)

Infection response score calculated for five common cell types across 13 tissues from the tabula-muris mouse 186 atlas data. E) Infection response score for ileal cell types in scRNA-seq data across mock-infected and reovirus-187 infected ileum at two distinct stages. The infection response score represents the gene module score for a panel 188 of 438 genes significantly upregulated (two-sided Wilcoxon test, log fold-change > 1.0 and p-value < 0.01) in the 189 reovirus-infected ileum at 1 dpi as compared to the mock-infected ileum. F) Infection response score for spatial 190 transcriptomics data from mock-infected and reovirus-infected ileum at two distinct stages. 192 To explore the heterogeneity of endothelial cell phenotypes in more detail, we reclustered all 193 9,786 cardiac endothelial cells in the scRNA-seq data. We observed four distinct phenotypes: hearts at 4 and 10 dpi, and iv) inflamed endothelial cells from the heart at 7 dpi, with both 198 inflamed endothelial cell clusters expressing Isg15, Iigp1, and Ly6a ( Fig. 3A-B) . DGEA across 199 endothelial subclusters revealed that the inflamed 7 dpi endothelial cells overexpressed 200 chemokines Cxcl9 and Cxcl10, which are generally involved in immunoregulatory and 201 inflammatory processes, but more specifically in the recruitment of T cells and NK T cells 16 202 ( Fig. 3B-3C, Supp Fig. 7A ). In line with this observation, T cells in the 7 dpi hearts expressed (Supp Fig. 9D) . Ctss was expressed mainly in monocytes (Supp Fig. 9D) . Ctss 311 encodes a protease used for degradation of antigenic proteins to peptides for presentation by 312 MHC class II molecules. Increased formation of immunoproteasomes in susceptible mice has 313 been shown to affect the generation of antigenic peptides and subsequent T cell activity in viral 314 myocarditis 25,26 . GO term analysis of genes upregulated in the border zone revealed 315 enrichment of terms related to the response to tumor necrosis factor, response to interleukin-1, 316 and NIK/ NF-κB signaling (Supp Fig. 9E ). 317 To further understand the effect of immune cell infiltration on the cell type composition 318 surrounding the myocarditic regions, we assessed cell type proportions as a function of 319 distance from myocarditic regions in the tissue. We quantified the cell type proportions in 320 myocarditic regions, the border zones, and the rest of the ventricular tissue, and found that the 321 fraction of Cxcl9-high endothelial cells, Ccl2+ fibroblasts, T cells, dendritic cells, and NK cells 322 was increased in the myocarditic regions, and the fraction of cardiomyocytes was reduced in 323 myocarditic regions (Fig. 4C, Supp Fig. 2C) . 324 To understand the phenotype of Ccl2+ fibroblasts enriched in myocarditic region and border 325 zone, we reclustered 9,192 fibroblast cells from the scRNA-seq dataset and identified a distinct 326 cluster of inflamed Ccl2+ fibroblasts from the infected heart at 7 dpi (Supp. Fig. 9F, 9G) . The 327 Ccl2+ fibroblasts expressed high levels of MHC class 1 (H2-D1 and H2-K1) , adhesion marker 328 genes Vcam1 and Icam1, and other genes such as Serpina3g, C3, and Ms4a4d (Supp Fig.   329   9H, 9I) . Moreover, these cells also expressed Casp1 and Casp4, suggesting that fibroblasts 330 also undergo pyroptosis in response to cytotoxic T cells (Supp Fig. 9H) . 331 To investigate the effect of inflammation on cardiomyocytes in myocarditic hearts, we 332 reclustered 502 cardiomyocytes from the scRNA-seq dataset and identified three distinct 333 phenotypes: i) ventricular myocytes expressing Myl2, Myl3, and Mb derived from mock and 334 infected hearts at 4 and 10 dpi, ii) atrial myocytes expressing markers Myl4, Myl7, and Nppa 335 derived from mock and infected hearts at 4 and 10 dpi, and iii) inflamed myocytes from the 336 infected heart at 7 dpi expressing innate immunity genes Isg15, Igtp, and Iigp1 27 (Fig. 4D-E) . 337 Inflamed myocytes from the infected heart at 7 dpi had a distinct phenotype when compared to 338 the myocytes from hearts at 4 and 10 dpi, which clustered with myocyte cells from mock-339 infected hearts (Fig. 4E) . To find transcriptional signatures for myocytes present in the border 340 zone, we selected genes that were both enriched in cardiomyocytes in the scRNA-seq data 341 and upregulated in the border zone. This analysis revealed that cardiomyocytes in the border 342 zone expressed Gm4841, Gm12185, Mt1, Mt2, Ankrd1, and Nppb (Fig. 4F, Supp Fig. 9J) . 10 dpi. We generated a total of 16,771 single-cell transcriptomes and integrated the data with 376 the data from the WT virus. We did not observe sample-specific clusters after data integration, 377 suggesting minimal experimental batch effects (Fig. 5A, Supp Fig. 10A) . We performed viral 378 transcript enrichment and compared the mean viral transcripts in WT-and mutant-infected 379 cells. We found similar levels of mean viral transcripts for WT and K287T viruses at 4 dpi but a 380 60-fold lower viral load for K287T at 7 dpi, consistent with viral titer assays 9 (Supp Fig. 10B (Fig. 5B) . 385 We analyzed the cell type composition of inflamed Cxcl9-high endothelial cells and immune 386 cells detected in K287T-and WT-infected hearts. We observed fewer Cxcl9-high endothelial 387 cells and immune cells including cytotoxic T cells, infiltrating the heart at 7 dpi compared to 388 WT-infected heart (Fig. 5C & 5F) . These differences correlate with the reduced levels of 389 inflammation associated with the K287T mutant (Supp Fig. 11A & 11B) . To validate these 390 observations, we performed RNA FISH and immunofluorescence staining on K287T-infected 391 hearts and compared them to mock-infected and reovirus WT-infected hearts (Supp Fig. 11A) . 392 Immunostaining for reoviral antigen in tissue sections confirmed both a significantly reduced 393 area with viral replication (two-sided Mann-Whitney test, p-value < 0.05) and significantly lower 394 viral antigen within those areas (two-sided Mann-Whitney test, p-value < 0.05), consistent with 395 the scRNA-seq analysis and viral titer assays (Supp. Fig. 11C, Fig. 5E ). We observed a 396 reduction in infiltration of T cells in K287T-infected hearts as compared to WT-infected hearts 397 at 7 dpi (Fig. 5D, 5F ). The fraction of total cytotoxic immune cells (Prf1+) was significantly 398 reduced in K287T-infected hearts as compared to WT-infected hearts (two-sided Mann-399 Whitney test, p-value < 0.05, Supp. Fig. 11D, Fig 5F) . These findings support the reduced 400 immune-mediated cytotoxicity seen in K287T-infected hearts. This was further supported by a 401 significant reduction in Casp1 protein expression in K287T-infected hearts as compared to 402 WT-infected hearts (two-sided Mann-Whitney test, p-value < 0.05, Supp. Fig. 11C, Fig. 5G ). 403 Our results show that cardiac endothelial cells mount a potent and robust innate immune Viral myocarditis has been recognized as a cause of heart failure for more than 50 years, but it 430 is still a challenging disease to study, diagnose, and treat 31 . Here, we used integrated spatial 431 and single-cell RNA-seq to dissect the temporal, spatial, and cellular heterogeneity of reovirus-432 induced acute myocarditis in a neonatal mouse model. We assayed ileum and heart tissues at 433 multiple time points after infection. We investigated the cell types that are infected, and the 434 cellular and spatial heterogeneity of innate and adaptive immune responses. We generated a 435 total of thirteen scRNA-seq and eight spatial transcriptomics datasets, spanning two organs, Sample preparation for single-cell transcriptomics of cardiac tissue. 507 We sacrificed mock-infected and reovirus-infected C57BL/6J mice on day 4, day 7, and day 10 508 post-infection and collected cardiac tissues for single cell transcriptomics. Single heart tissue 509 from respective stages (one heart per stage) were isolated aseptically, washed with ice-cold segmentation mask were files and morphological opening was performed to remove noise. 782 The segmentation was enhanced using watershed algorithm followed by a morphological 

Balanced Salt Solution and then blood was carefully removed by perfusing the hearts with 702 fresh HBSS through the apex. Fresh tissues were immediately embedded in Optimal Cutting 703

Compound (OCT) media and frozen in liquid nitrogen cooled isopentane, cut into 10 µm 704 sections using a Thermo Scientific Microm 550 cryostat, and mounted on -20°C cooled 705 histological glass slides which were then stored at -80°C until used

RNA fluorescence in-situ hybridization (FISH) split probe design and Signal 707 Amplification using Hybridization Chain Reaction HCR-V3

Two-step hybridization strategy with split probe design and Hybridization Chain Reaction 709 (HCR)-V3 47 was used to label up to three transcripts in a single tissue section. Probes were 710 designed using NCBI primer-blast which uses primer3 for designing internal hybridization oligo 711 and BLASTn to check for binding specificity. We designed 20-21 bp primer pairs for an 712 amplicon length of 40-42 bp

-10 sets of reverse complemented 714 forward primers and reverse primers were then concatenated to flanking initiator sequence for 715

Split probes for each gene target, mixed and diluted in nuclease-free water to 717 create a split probe pool stock solution at 10µM total probe concentration for every target

Hairpin pairs labeled with three different fluorophores namely Alexa-488, Alexa-546

Supp Data 4) were used for HCR V3

RNA fluorescence in-situ hybridization (FISH) experiments

Slides with tissue sections were then brought to room temperature until the OCT melts and 722 were then immediately fixed in 4% paraformaldehyde for 12 minutes at room temperature

Post fixation, the sections were washed for 5 mins in 1x PBS twice, incubated for 1 hour in 724 70% ethanol for tissue permeabilization, washed again for 5 mins in 1x PBS, and then used for 725 primary hybridization. Hybridization Buffer (HB) mix was prepared with 2x SSC

01% SDS, 1uM of probe pool mix 727 per target for the hybridization reaction. 20 µl of HB mix (with probes) per section was then put 728 on each slide to cover the tissue section, covered with parafilm, and incubated overnight at 729 37°C inside a humidifying chamber for primary hybridization. After primary hybridization, 730 parafilm was removed and slides were washed in Hybridization

EDTA) for 20-30 minutes at 48°C. Amplification Buffer 732 (AB) mix was prepared with 2x SSC, 5x of Denhart Solution

After fixation, the slides were then rehydrated to Milli-Q water for 2 minutes and then 753 washed twice in 1X PBS. Samples then underwent an antigen retrieval step via incubation in 754 1X citrate buffer for 10-15 minutes at 95°C

756 blocked for one hour at room temperature in blocking buffer (1% bovine serum albumin and 757 10% normal donkey serum in PBS. 20ul of primary antibodies diluted in antibody solution (1% 758 bovine serum albumin in PBS) were then added on to the slides, covered with parafilm, then 759 incubated in a humidifying chamber overnight at 4°C. Primary antibodies used were rabbit anti-760 reovirus VM1:VM6 polyclonal sera (1:30000), rat anti-Casp1 monoclonal antibody (1:200, #14-761 9832-82, Invitrogen)

Lastly, samples were washed thrice in PBS for 10 minutes with shaking, counter 766 stained with DAPI, and mounted in Prolong antifade mounting media. Images were acquired 767 on a Zeiss Axio Observer Z1 Microscope using a Hamamatsu ORCA Fusion Gen III Scientific 768 CMOS camera. Immunostaining images were shading corrected, stitched, rotated, 769 thresholded

Viral myocarditis-diagnosis, treatment 825 options, and current controversies

Viral myocarditis from the perspective of the virus

Myocarditis and inflammatory cardiomyopathy: current evidence and future 830 directions

Viral myocarditis. A review

An overview of the immune mechanisms of viral myocarditis

Derivation and characterization of an 835 efficiently myocarditic reovirus variant

Mechanisms of reovirus bloodstream dissemination

The multi-functional reovirus σ 3 protein is a virulence factor that suppresses stress 839 granule formation and is associated with myocardial injury

Cardioimmunology: the immune system in cardiac homeostasis 841 and disease

Lymphocytes protect against and 843 are not required for reovirus-induced myocarditis

Interferon Responses Are Critical for Control of Systemic Reovirus Dissemination

Interferon Regulatory Factor 3 Attenuates Reovirus Myocarditis and 848 Contributes to Viral Clearance

Single-cell transcriptomics of 20 mouse organs creates a Tabula Muris

Single-Cell Transcriptome Atlas of Murine Endothelial Cells

CXCL9, CXCL10, CXCL11/CXCR3 axis for immune activation -a target for 854 novel cancer therapy

856 Infectious myocarditis: the role of the cardiac vasculature

An evolving new paradigm: endothelial cells 858 -conditional innate immune cells

Cell death 860 mechanisms induced by cytotoxic lymphocytes

Silencing of CCR2 in myocarditis

Mesenchymal Stromal Cells Modulate Monocytes Trafficking in Coxsackievirus 863 B3-Induced Myocarditis

Increased expression of extracellular matrix regulators TIMP1 and MMP1 in 865 deteriorating heart failure

The translation of non-canonical open reading frames controls mucosal 867 immunity

J/clusterin induction in myocarditis: A localized response gene to myocardial injury

Ongoing Coxsackievirus Myocarditis Is Associated with Increased Formation 872 and Activity of Myocardial Immunoproteasomes

Myocarditis elicits dendritic cell and monocyte infiltration in the heart 874 and self-antigen presentation by conventional type 2 dendritic cells

IFN-β) and IFN-β-Stimulated Gene Expression Are Cell Type Specific in the Cardiac Protective 878 Response

Metallothioneins 1 and 2 Modulate Inflammation and Support Remodeling in 880

Ischemic Cardiomyopathy in Mice

Induction of Ankrd1 in dilated cardiomyopathy correlates with the heart 882 failure progression

Spatiotemporal single-cell analysis reveals critical roles of mechano-sensing 884 genes at the border zone in remodeling after myocardial infarction

Spatiotemporal single-cell RNA sequencing of developing chicken hearts 889 identifies interplay between cellular differentiation and morphogenesis

A Spatiotemporal Organ-Wide Gene Expression and Cell Atlas of the Developing 892 Human Heart

Spatial multi-omic map of human myocardial infarction

Association of Cardiac Infection With SARS-CoV-2 in Confirmed COVID-19 896 Autopsy Cases

Cardiac SARS-CoV-2 infection is associated with pro-inflammatory 898 transcriptomic alterations within the heart

The innate immune response to coxsackievirus B3 predicts progression to 900 cardiovascular disease and heart failure in male mice

Testosterone and interleukin-1β increase cardiac remodeling during 902 coxsackievirus B3 myocarditis via serpin A 3n

Dissecting the Cellular 904 Landscape and Transcriptome Network in Viral Myocarditis by Single-Cell RNA Sequencing

Cytopathogenic effect in cardiac myocytes but not in cardiac fibroblasts 907 is correlated with reovirus-induced acute myocarditis

PKR's protective role in viral myocarditis

Cardiac Cell-specific Apoptotic and Cytokine Responses to Reovirus 911 Infection: Determinants of Myocarditic Phenotype

Large-scale single-cell gene expression data 913 analysis

scvi-tools: a library for deep probabilistic analysis of single-cell omics data

Gene set enrichment analysis: A knowledge-based approach for 917 interpreting genome-wide expression profiles

Third-generation in situ hybridization chain reaction: multiplexed, 920 quantitative, sensitive, versatile, robust

Gene Expression Omnibus: NCBI gene expression and hybridization array data 922 repository

Reovirus-induced-myocarditis

 797 We would like to thank Peter Schweitzer and the Cornell Genomics Center for help with single- 

The authors declare no conflicts. 823