To profile the virus-specific CD8 T cell response, we leveraged three previously described models of murine viral infection. In an attempt to maintain initial target cell tropism, the initial distribution of early infection events, and consistency with previous TCR repertoire studies in the context of LCMV infection, acute and chronic LCMV infections were elicited by a low (200 ffu) or high dose (2 x 10⁶) of an identical LCMV clone 13 intravenously (i.v.) as previously performed (4, 10, 16, 17). We additionally induced a latent viral infection via i.v. administration of MCMV-ie2-gp33 (2x10⁵ pfu) to obtain a stem-like T cell subset comparable to a previously described population of Tcf7-expressing cells that sustain the inflationary response during latent infection (5). An advantage of using these three viruses is that they all contain the gp_33-42 (GP33) viral peptide epitope, which enables isolation of endogenous virus-specific CD8+ T cells using MHC-tetramers. While the GP33 peptide is naturally encoded in the LCMV genome, it has been engineered into the ie2 gene locus of MCMV and gives rise to a population of virus-specific T cells termed inflationary T cells (5, 18). We therefore isolated GP33-specific CD8+ T cells from the spleen of mice at 28 dpi, separated into cohorts of either acute LCMV, chronic LCMV, or latent MCMV-ie2-gp33 infection. While we intended to include GP33-specific CD8+ T cells following peptide immunization with the GP33 peptide in CpG, we were unable to obtain sufficient numbers of GP33-specific cells 28 dpi and therefore excluded this group for future sequencing ( Supplementary Figure S1A ). The virus-specific CD8+ T cells were then processed for single-cell sequencing of their TCR repertoires and transcriptomes by following the 10X genomics workflow (5’ immune profiling with V(D)J and GEX protocol) ( Figure 1A and Supplementary Figure S1A ). Following single-cell sequencing and alignment to the murine reference transcriptome, we recovered GEX information from thousands of virus-specific CD8+ T cells from each mouse ( Figures 1B, C ), with the median number of genes per cell ranging from 866 to 1210 for all mice ( Figure 1D ), accompanied by comparable percentages of mitochondrial genes and sequencing reads for all samples ( Supplementary Figures S1B, S1C ).

Acute, chronic, and latent infections have been reported to have distinct phenotypes of virus-specific CD8+ T cells, corresponding to memory, exhaustion, and inflationary subsets (9, 19–22). Performing unsupervised clustering based on total gene expression [excluding genes relating to TCRs, i.e., V-, D-, J-, and constant region (C) genes] and subsequently visualizing the cells from all mice revealed infection-specific clustering ( Figures 1B, D, E ). Quantifying the proportion of cells in each cluster demonstrated that the transcriptional profiles were highly reproducible across biological replicates ( Figure 1E and Supplementary Figures S1D, S1E ), with CD8+ T cells found in distinct clusters for acute LCMV (clusters 0,1,3,4), chronic LCMV (5,6,7,8), and latent MCMV (0,1,2,6) infections. Further unbiased investigation into the most expressed genes per cluster revealed a plethora of genes previously reported in the context of viral infections, such as Il7r, Tcf7, Zeb2, Klrg1, Gzma, Gzmb, Gzmk, Vim, Lgals3, Tox, Lag3, Pdc1, Id3 ( Figure 1B and Supplementary Figures S2, S3, S4 ). In the case of chronic and latent infections, we observed one cluster (cluster 7) where CD8 T cells express Tcf7 and Slamf6, which is characteristic of previously described stem-like T cells capable of sustaining the effector population during chronic infection and during checkpoint blockage (2, 20, 23). Together, these patterns of gene expression suggested the presence of distinct memory (clusters 0, 3, 4), effector (clusters 1, 2, 5), exhausted (cluster 8), memory-like (cluster 7), proliferative (cluster 6) subsets, in addition to a small population of B cells present (cluster 9) in all samples and CD4+ cells, suggesting minor contamination ( Figure 1F ) and were therefore removed for the remainder of the gene expression analyses. We observed considerable overlap of cells from acute LCMV and MCMV-ie2-gp33 in the largest effector-like cluster (cluster 1) and subsequently questioned the extent that these cells could be differentiated by gene expression. Performing differential gene expression analysis between the cells of cluster 1 separated by infection type demonstrated that T cells isolated from MCMV-ie2-gp33 had significantly increased expression of granzymes (Gzmk, Gzma), activation genes (Ccl5, Nkg7, Klrg1), and MHC class I genes ( Supplementary Figure S6 and Supplementary Tables S7, S8 ), even though the cells occupied the same cluster ( Figure 1F ). When comparing the cells in cluster 1 arising from either chronic or latent infection, we observed a significant upregulation of genes involved in T cell exhaustion ( Supplementary Figure S6 and Supplementary Tables S7, S8 ). Finally, as we observed a population of stem-like Tcf7-expressing cells in both chronic LCMV and MCMV infection, we questioned whether these cells similarly maintained characteristics of exhaustion and T cell inflation, respectively. Differential gene expression between cells within this cluster demonstrated indeed demonstrated that T cells arising from chronic infection within this cluster maintained relatively higher expression levels of exhaustion-program genes such as Pdcd1, Tox, and Lag3, in addition to genes such as Bcl2, Tigit, and Id3 ( Supplementary Figure S6 ). This was in contrast to Tcf7+ cells from latent infection, which maintained relatively higher levels of Gzmm, Ifngr1, and Ccl5 ( Supplementary Figure S6 ).

We next performed differential gene expression and calculated most up and down-regulated genes to determine if infection conditions would further separate transcriptional phenotypes ( Figure 2A ). Genes characteristic of T cell exhaustion were upregulated in the chronic LCMV infection (e.g., Pdcd1, Tox, Lag3), whereas genes associated with memory formation and inflationary phenotypes were upregulated in the acute LCMV (e.g., Il7r) and latent MCMV (e.g., Klrg1) infections ( Figure 2A , Supplementary Tables S1–S3 ). As many of these genes have been previously described in the context of viral infection, we investigated whether the expression of additional genes commonly used to differentiate populations of CD8+ T cells could further differentiate infection types. Genes such as Cd8a and Cd3e were expressed ubiquitously across all cells ( Supplementary Figures S2–S4 ), whereas exhaustion markers such as Pdcd1, Tim3, Lag3, and Ctla4 were preferentially localized to the cells arising from chronic infection ( Figure 2B and Supplementary Figures S2–S4 ). We additionally observed a population of cells from chronically infected mice coexpressing Pdcd1 and Tcf7 ( Figure 2B and Supplementary Figures S2–S4 ), which has been described previously as “stem-like”, “memory-like” or “progenitor-exhausted”, and serves to sustain the effector and exhausted population in the context of chronic infection (2, 23, 24). Expression of Klrg1 was similarly localized to cells arising from the MCMV-ie2-gp33 infection, consistent with the known effector-memory phenotype of inflationary CD8+ T cells (5) ( Figure 2B and Supplementary Figures S2–S4 ). These findings were consistent when performing gene ontology (GO) and gene set enrichment (GSEA) analyses using the 100 most upregulated genes from our previous differential gene analysis ( Supplementary Figures S6–S8 and Supplementary Tables S4–S6 ).

To better annotate our observed CD8 T cell states, we utilized ProjectTILS (25), a recently develop T cell reference atlas to interpret T cell states. Quantifying the fraction of cells residing in each CD8-specific ProjectTILS-defined subset again highlighted the robust phenotypic differences of T cells arising from the three infection types ( Figures 2C, D and Supplementary Figures S9, S10, S11 ). Chronic LCMV infection resulted in a clear increase in the exhausted subset, whereas acute and latent infections were largely dominated by short-lived effector cells (SLECs) ( Figure 2C ). When using the generalized T cell reference subsets containing both CD4 and CD8 T cells, however, we observed an increase in naïve-like and early effector-like T cells following acute infection relative to the increased proportion of effector memory cells found following MCMV infection ( Supplementary Figures S10, S11 ). Together, this further highlighted distinct transcriptional phenotypes arising from acute, chronic and latent infections.

After observing the transcriptionally diverse gene expression profiles following the different infection types, we determined if TCR repertoires showed infection-specific features. By restricting our analysis to cells containing only one TRA and one TRB sequence, we obtained information ranging from hundreds to thousands of cells from each mouse ( Figure 3A and Supplementary Figure S12A ). Quantifying the number of unique clones [defined by unique complementarity determining region 3 beta (CDRb3) + CDR3 alpha (CDRa3) nucleotide sequence] revealed hundreds of clones for each mouse ( Figure 3B and Supplementary Figure S12B ), indicating both a polyclonal GP33-specific repertoire in all three infection conditions and the presence of clonal expansion. Next, we visualized the percentage of the repertoire comprised by each clone, which demonstrated that mice which had received acute LCMV infection (and therefore had already cleared the virus 28 dpi) had a higher fraction of clones supported by only a single cell barcode ( Figure 3A ). Quantifying the Shannon evenness, a commonly used entropy metric that provides a global view of clonal frequencies (26–28), further confirmed the notion that acute LCMV infection resulted in relatively less clonal expansion than the other two infection types, where antigen is still present ( Supplementary Figure S12C ). A closer examination of the 30 most expanded clones from each repertoire ( Figure 3A ) revealed that in many cases individual clones were represented by hundreds of cells, particularly in chronically and latently infected mice, and, in the case of a single latently infected mouse, more than a thousand cells ( Figure 3B ).

We next determined the extent of clonal convergence in the GP33-specific repertoire by quantifying the number of identical TCRs found in each mouse, which revealed minimal overlap detected regardless of infection condition ( Figure 3C ). We observed no prominent relationship between clonal expansion and clonal overlap, as performing a similar analysis restricted to expanded clones (clones supported by two or more cells), including the 10 most expanded clones per mouse, did not reveal any substantial overlap ( Supplementary Figures S12D, E ). We subsequently questioned if signs of clonal convergence could be detected by focusing the analysis on clones with similar, but non-identical TCR sequences. We therefore constructed sequence similarity networks based on the edit distance of the CDRb3 and CDRa3 sequences. Despite investigating a range of edit distance thresholds, we visually failed to observe any infection-specific clustering ( Figure 3D and Supplementary Figure S13A ). Formally quantifying the number of edges shared either within the same or across different infection groups supported our visual observation that infection-specific overlap was not present ( Supplementary Figure S13B ).

After observing the low degree of sequence similarity across all mice, we determined if distinct patterns of germline gene usage could be observed in any of the infection conditions, as previous repertoire sequencing experiments investigating virus-specific T cells in the context of LCMV or MCMV infection have demonstrated preferential germline use (5, 10). Leveraging the ability of our single-cell immune repertoire profiling, we could investigate the variable (V) gene usage for both TRB and TRA chains, in addition to quantifying how often certain pairings occurred. Quantifying and visualizing the number of cells using a given germline pairing revealed certain V genes dominated the repertoire across multiple mice in different infection conditions, such as TRBV13-1, TRBV19, and TRBV29 ( Figure 4A and Supplementary Figure S14A ). Calculating pairwise correlation coefficients for germline gene usage between all mice demonstrated that TRBV gene usage loosely clustered repertoires by infection type ( Supplementary Figure S14B ), although this effect was not observed when both TRBV and TRAV genes were included into the calculation ( Supplementary Figure S14C ). We lastly questioned whether including repertoires lacking specificity to GP33 could provide contrast for how similar the 6 repertoires were following acute, chronic, and latent infection. We included two naïve datasets from pooled PBMCs from naïve C57BL/6 mice (10x genomics) and CNS-resident CD4 and CD8 T cells from naïve C57BL/6J mice (12). Additionally, GP_66-77-specific CD4 T cells from C57BL/6 mice following acute LCMV infection (15) and T follicular helper cells from NP17-OVA immunized C57BL/6J mice (29) were included as another source of polyclonal T cell repertoires. Including these publicly available data demonstrated a clear separation between the repertoires of GP33-specific T cells and those T cells arising from other contexts ( Figure 4B and Supplementary Figures S15, S16 ), further supporting the notion that the GP33-specific repertoire selects for distinct germline gene usage irrespective of the infection condition.

We next integrated TCR sequence with transcriptomes at single-cell resolution. It has been previously demonstrated that highly expanded clones upregulate effector molecules such as Nkg7, Ccl5, and granzymes (12, 13). Therefore, we first focused our analysis on the 30 most expanded clones for each infection by quantifying the fraction of cells present in each transcriptional cluster ( Figure 5A ). Cells arising from different infection conditions occupied distinct transcriptional states, thereby suggesting transcriptional heterogeneity within the majority of expanded clones ( Figure 5A ). Extending this analysis to all clones, regardless of clonal expansion, would reveal differences between highly and lowly expanded clones. This analysis showed that more expanded clones (thicker lines in Supplementary Figure S17A ) were predominantly connected to clusters 5 and 2 for chronic and latent infections, respectively, whereas lowly expanded clones (narrower lines) were often connected to clusters 7 and 0, respectively ( Supplementary Figure S17A ). Quantifying the cluster membership demonstrated a trend that those cells with higher expansion levels were located in cluster 1 for acute, 5 for chronic, 2 for latent infection ( Supplementary Figure S17B ), which characteristically expressed Zeb2, Tox, and Klrg1, respectively ( Figures 1B, F and Supplementary Figure S2 ). Conversely, lowly expanded clones were more often located in clusters 4 for acute and 7 for chronic and latent ( Figure 5B and Supplementary Figures S17B, C ), which were characterized by high expression of genes associated with memory phenotypes such as Id3, Sell, and Tcf1 ( Figures 1B, F and Supplementary Figure S2 ). We additionally observed that the stem-like cluster 7 (characterized by Tcf7 expression) was entirely absent in the highly expanded clones isolated from mice infected with MCMV-ie2-gp33. Integrating the ProjectTILS annotations into the clonal expansion data demonstrated similarly support a model in which clonally expanded clones were more likely to reside in certain subsets (e.g., SLEC, effector memory, and exhausted) relative to others (e.g., Pex, naïve-like T cells) ( Figure 5C and Supplementary Figure S11 ). Calculating the differentially expressed genes revealed that genes such as Nkg7, Lgals1, Pdcd1, and Ccr2 were significantly upregulated in expanded cells in at least one infection condition and demonstrated consistent trends in expression for all infection groups ( Figure 5D and Supplementary Tables S9–S11 ). Further inspection into the CDR3 motifs of different transcriptional clusters and the expanded and lowly expanded clones suggested the enrichment of certain residues were enriched in either highly-expanded clones and effector and exhausted clusters ( Supplementary Figure S20 ). This was exemplified when visualizing that the lowly-expanded clones from the latent MCMV infection and the Tcf1-expressing cluster had increased use of asparagine amino acid in residue nine relative to expanded clones from the latent MCMV infection, which had a preferential use of arginine and higher proportion of clones using glutamine in position 11 similar to expanded cells in effector cluster two ( Supplementary Figure S20 ). Taken together, our findings suggest that biophysical properties dictating clonal expansion influence the transcriptional phenotype towards more effector and exhausted phenotypes and away from memory and naïve-like ones.

All animal experiments were performed in accordance with institutional guidelines and Swiss federal regulations. Experiments were approved by the veterinary office of the canton of Zurich under animal experimentation licenses 115/2017 and ZH058/20. 6-8-week-old female C57BL/6 mice from Janvier were housed under specific-pathogen-free conditions in individually ventilated cages with bedding and nesting material for enrichment. Acute LCMV infections were infected intravenously (i.v.) with 200 focus forming units (ffu) of LCMV clone 13 in the tail vein. Chronic LCMV infections were performed i.v. with 2 x 10⁶ ffu LCMV clone 13. Latent infections were established by injecting 2x10⁵ pfu dose of MCMV-ie2-gp33 i.v., which was obtained from Dr. L. Cicin-Sain and contains a functional m157 gene as previously described (18). MCMV viral stocks were propagated on M2-10B4 cells and purified by ultracentrifugation using a 15% sucrose gradient. LCMV clone 13 was produced as previously described (4). Upon sacrifice with CO2 at 28 dpi, spleens were harvested and single-cell suspensions were prepared by mashing the tissue through a 70 uM cell strainer and rinsing with complete RPMI (RPMI-1640 supplemented with 10% fetal bovine serum, 2 mM L-glutamine, 1% penicillin-streptomycin, 1 mM sodium pyruvate, 50 nM beta-mercapthoethanol, 0.1 mM non-essential (glycine, L-alanine, L-asparagine, L-aspartic acid, L-glutamic acid, L-proline, L-serine) amino acids, 20 mM HEPES). The single cell suspension was then incubated with CD8-PE (clone 53-6.7, Biolegend), MHC class 1 tetramer for gp_33-41 conjugated to APC diluted in FACS buffer (PBS, 2 mmEDTA, 2% FCS) at room temperature for 30 minutes, as previously described (39), and LiveDead nearIR. Tetramer positive cells were isolated via flow cytometric sorting (FACSAria with FACSDiva software) and subsequently supplied as input for single-cell immune repertoire sequencing.

Single-cell immune repertoire sequencing was performed as according to the 10x Genomics Chromium Single Cell V(D)J Reagents Kit (CG000166 Rev A) as previously described (40). In brief, single cells for all six samples were simultaneously encapsulated with gel emulsion microdroplets (10x Genomics, 1000006) in droplets using 6 lanes of one Chromium Single Cell A Chip (10x Genomics, 1000009) with a target loading of 13,000 cells per reaction. cDNA amplification was performed using 14 cycles and subsequently split for downstream GEX and VDJ library preparation. GEX libraries were amplified using the Chromium Single Cell 5’ Library Kit (10x Genomics, 1000006). TCR libraries were amplified using the Chromium Single Cell V(D)J Enrichment Kit, Mouse T Cell (10x Genomics, 1000071). Final libraries were pooled and sequenced on the Illumina NovaSeq S1 using a concentration of 1.8 pM with 5% PhiX.

Paired-end sequencing files for GEX and VDJ libraries were aligned to the murine reference genome (mm10) and V(D)J germlines (GRCm38) using 10x Genomics cellranger (v4.0.0) count and vdj arguments, respectively. The filtered feature matrix directory was supplied as input to the automate_GEX function in the R package Platypus (v2.0.5) (13), which uses the transcriptome analysis workflow of the R package Seurat (41). Only those cells containing less than 20% of mitochondrial reads were retained in the analysis. Genes involved in the adaptive immune receptor (e.g., TRB, TRBV1-1), were removed from the count matrix to prevent clonal relationships from influencing transcriptional phenotypes. Gene expression was normalized using the “scale.data” argument in automate_GEX, which first performs log-normalization with a scaling factor of 10000 and then scales mean expression and variance to 0 and 1, respectively. 2000 variable features were selected using the “vst” selection method and used as input to principal component analysis (PCA) using the first 10 dimensions. Graph-based clustering using the Louvain modularity optimization and hierarchical clustering was performed using the functions FindNeighbors and FindClusters in Seurat using the first ten dimensions and a cluster resolution of 0.5. UMAP was similarly inferred using the first ten dimensions. The FindMarkers function from Seurat was used when calculating differentially expressed genes (both across groups or across clusters) with both the minimum log fold change and the minimum number of cells expressing each gene set to 0.25. Mitochondrial and ribosomal genes were removed when either visualizing DE genes or supplying the top DE genes as input to gene ontology and gene set enrichment analyses. Cluster 9, which contained B cells, was removed for further analysis. Gene ontology and gene set enrichment analysis was performed using the GEX_GOterm and GEX_GSEA functions in Platypus by supplying either the top N or bottom N genes as input. In the case of GEX_GSEA, the C7 immunological signatures gene set from the Broad institute was supplied as input to the function (42). The GEX_GSEA function uses the R package fgsea (43) to conduct gene set enrichment analysis and GEX_GOterm is based on the R package edgeR (44). Projection of cells onto a reference UMAP was done using the R package ProjecTILs (25). Each infection condition was projected individually onto the acute and chronic viral infection CD8 T cell atlas and the tumor-infiltrating T lymphocytes (TIL) atlas.

For TCR repertoire analysis, the output directory of 10x Genomics cellranger vdj function was supplied as input to the VDJ_analyze function in Platypus maintaining the default clonotyping strategy (CDRa3+CDRb3 nucleotide sequence) as performed by cellranger. Those clones not containing exactly one TRA and one TRB chain were removed from the analysis. Clonal frequency was determined by counting the number of distinct cell barcodes for each unique CDR3. Overlap matrices were calculated by first appending the CDRa3 and CDRb3 nucleotide sequences and then quantifying the exact matches across repertoires. Similarity networks were calculated based on the VDJ_network function in Platypus, which first calculates the edit distance separately for TRB and TRA CDR3s, and then draws edges between those clones with a distance below the specified threshold. Circos plots were created using the VDJ_circos function in Platypus with a label.threshold of 5. Those cells in clones supported by only one cell were considered lowly expanded clones, whereas those clones supported by two or more cells were considered expanded. Logo plots were created using the VDJ_logoplot_vector function in Platypus on unique CDRa3 and CDRb3 sequences.

Heatmaps displaying differential gene expression were produced using the DoHeatmap function in the R package Seurat (v4.0.1) (45). Gene enrichment analysis was performed using the GEX_GOterm function in Platypus, which is based on the analysis pipeline in edgeR (44). Enrichment plots were produced using the R package ggplot (46). Gene set enrichment analysis was performed using the GEX_GSEA function in Platypus (v3.1) under default parameters, which utilizes fgsea (v1.12), tibble (v2.1.3), and the C7 gene set from the molecular signatures database MSigDB (43, 47). Similarity networks were produced using the R package igraph (48). Circos plots were produced using the chordDiagram function of the R package Circlize (50). Logoplots were produced using the R packages ggseqlogo (49). All other figures were produced using Prism v9 (Graphpad).