The molecular properties of CD8+ T cells that respond to SARS-CoV-2 infection are not fully known. Here, we report on the single-cell transcriptomes of >80,000 virus-reactive CD8+ T cells, obtained using a modified Antigen-Reactive T cell Enrichment (ARTE) assay, from 39 COVID-19 patients and 10 healthy subjects. COVID-19 patients segregated into two groups based on whether the dominant CD8+ T cell response to SARS-CoV-2 was ‘exhausted’ or not. SARS-CoV-2-reactive cells in the exhausted subset were increased in frequency and displayed lesser cytotoxicity and inflammatory features in COVID-19 patients with mild compared to severe illness. In contrast, SARS-CoV-2-reactive cells in the dominant non-exhausted subset from patients with severe disease showed enrichment of transcripts linked to co-stimulation, pro-survival NF-κB signaling, and anti-apoptotic pathways, suggesting the generation of robust CD8+ T cell memory responses in patients with severe COVID-19 illness. CD8+ T cells reactive to influenza and respiratory syncytial virus from healthy subjects displayed polyfunctional features and enhanced glycolysis. Cells with such features were largely absent in SARS-CoV-2-reactive cells from both COVID-19 patients and healthy controls non-exposed to SARS-CoV-2. Overall, our single-cell analysis revealed substantial diversity in the nature of CD8+ T cells responding to SARS-CoV-2.
Recent studies in COVID-19 patients have verified the presence of CD8+ T cells that are reactive to SARS-CoV-2. However, the nature and types of CD8+ T cell subsets that respond to SARS-CoV-2 and whether they play an essential role in driving protective or pathogenic immune responses remain elusive. Here, we report on single-cell transcriptome and TCR sequence analyses of >87,000 in vitro activated virus-reactive CD8+ T cells and >20,000 CD8+ T cells expressing activation markers ex vivo, from a total of 39 COVID-19 patients and 10 healthy, pre-pandemic donors. To compare the molecular properties of antigen-specific SARS-CoV-2-reactive CD8+ T cells to other common respiratory virus-reactive CD8+ T cells, we also isolated virus-reactive CD8+ memory T cells from healthy control subjects and analyzed their single-cell transcriptomes.
Across all the virus-reactive CD8+ T cells studied, we delineated eight distinct clusters with distinct transcriptomic features. TCR sharing between clusters identified a high degree of plasticity among virus-reactive CD8+ T cells. We find that in vitro peptide stimulation does not significantly alter the baseline ex vivo state of the viral-reactive CD8+ memory T cells and these cells can acquire a wide range of transcriptional programs following different viral infections. For example, in healthy subjects, CD8+ T cells with polyfunctional features, linked to protective anti-viral immunity, are abundant among CD8+ memory T cells reactive to FLU and RSV. In contrast, these cells were mostly absent in SARS-CoV-2 responsive cells from both COVID-19 patients and healthy non-exposed subjects. Notably, cells in this polyfunctional cluster were also significantly enriched for genes related to aerobic glycolysis, which is considered to enhance effector functions of CD8+ memory T cells. The absence of such polyfunctional memory CD8+ T cells in SARS-CoV-2 infection may be due to the short interval between symptom onset and blood collection for analysis (median 20 days), a time period when effector responses as opposed to long-term memory responses are likely to be captured. Long-term follow up studies will be required to clarify if SARS-CoV-2 infection generates such polyfunctional long-term memory T cells.
A large fraction of SARS-CoV-2 reactive cells (43% and 37%) from healthy non-exposed subjects (pre-pandemic), presumed to be human CoV-reactive cells that cross-react with SARS-CoV-2 peptide pools, were present in clusters 1 and 0, respectively. These clusters also had similar representation of SARS-CoV-2 reactive cells from patients with COVID-19 illness. Cells in cluster 1 showed significant positive enrichment for type I interferon signaling, CD4 T-‘unhelped’ and ‘exhaustion’ signatures, reminiscent of the ‘exhausted’ CD8+ T cells reported in murine LCMV infection models. Cluster 0, in contrast, was non-exhausted and showed significant negative enrichment of exhaustion and interferon signatures. SARS-CoV-2 reactive cells from COVID-19 patients also contributed the large majority of cells in cluster 2, which was characterized by enriched expression of cell cycle-related genes. Like cells in cluster 0, cluster 2 cells also showed negative enrichment of interferon genes and relatively lower exhaustion signature scores. Thus, we find that the nature of CD8+ T cells reactive to Coronaviridae differed substantially from those responding to FLU or RSV.
Intriguingly, COVID-19 patients broadly segregated into two groups, according to whether the majority of their virus-reactive CD8+ memory T cells were in the ‘non-exhausted’ cluster 0 or the ‘exhausted’ cluster 1. Patients with mild COVID-19 illness had a greater proportion of SARS-CoV-2 reactive cells in the ‘exhausted’ cluster 1 with the ability to maintain their ‘exhausted’ state, even after viral infection resolution. Besides, cells in the ‘exhausted’ subset (cluster 1) from patients with mild COVID-19 illness expressed significantly higher levels of type I interferon response genes, lesser levels of transcripts encoding for cytotoxicity molecules, Fas ligand and proinflammatory cytokines (CCL3, CCL4, CSF-2, TNF, LTA and LTB), and were significantly less clonally expanded. This raises the possibility that the magnitude and quality of the ‘exhausted’ CD8+ T cell response may be clinically important for limiting excessive tissue damage by SARS-CoV-2-reactive CD8+ T cells in COVID-19 illness.
Qualitative differences in the ‘non-exhausted’ clusters 0 and 2 further emerged between patients with mild and severe COVID-19 illness. Transcripts increased in cluster 0 cells from severe relative to mild illness were significantly enriched in multiple co-stimulation pathways (OX40, CD27, CD28, 4-1BB, CD40) that are linked to CD4+ T cell-mediated ‘helped’ features, NF-κB and cell survival pathways thought to be important for IL-2 production, proliferation, and survival. This finding suggested patients with severe disease mount a more effective CD8+ memory T cell response to SARS-CoV-2 infection that could potentially lead to durable protection against re-exposure. Indeed, recent studies highlighted that convalescent COVID-19 patients with history of severe disease mounted more robust CD8+ memory T cell response to SARS-CoV-2 infection that could potentially lead to durable protection against re-exposure. Overall, our findings indicate that SARS-CoV-2-reactive CD8+ T cells from patients with severe COVID-19 displayed multiple features that support the generation of robust CD8+ T cell memory responses with pro-survival properties and a lack of “restrained function” via ‘exhaustion’ features. Whether these cells play a role in disease pathogenesis or provide long-term immunity is not clear, and further longitudinal analysis and function studies in relevant model organisms are required to clarify this.
MATERIALS AND METHODS
Patient recruitment, ethics approval and sample processing
The Ethics Committee of La Jolla Institute (USA) and the Berkshire Research Ethics Committee (UK) 20/SC/0155 provided ethical approval for this study with written consent from all participants. 22 hospitalized patients with reverse transcriptase polymerase chain reaction (RT-PCR) assay confirmed SARS-CoV-2 infection were recruited between April-May 2020. A cohort of 17 non-hospitalized participants were also recruited with RT-PCR assay or serological evidence of SARS-CoV-2 infection. Up to 80 ml of blood was obtained from all subjects for this research. Clinical metadata linked to hospitalized patients such as age, gender, comorbidities, level of clinical care required, radiological findings and laboratory results are provided in table S1. The COVID-19 cohort consisted of 30 (77%) White British/White Other, 4 (10%) Indian, 2 (5%) Black British, 2 Arab (5%) and 1 Chinese (3%) participants. Of the 39 COVID-19 subjects, 22 (56%) had moderate/severe disease requiring hospitalization and 17 (44%) had mild disease, not requiring hospitalization. The median age of the hospitalized patients was 60 (33-82) and 77% were male. The median age of the non-hospitalized participants was 39 (22-50) and 47% were male. To study SARS-CoV-2, FLU and RSV-reactive CD8+ T cells from healthy subjects, we utilized de-identified buffy coat samples from healthy adult subjects who donated blood at the San Diego Blood Bank before 2019, prior to the COVID-19 pandemic. Peripheral blood mononuclear cells (PBMCs) were isolated from blood by density centrifugation over Lymphoprep (Axis-Shield PoC AS, Oslo, Norway) and cryopreserved in 50% human serum, 40% complete RMPI 1640 medium and 10% DMSO.
Peptide pools Two peptide pools (Miltenyi Biotec) consist of lyophilized peptides (15-mer sequences) with 11 amino acids overlap. For the S protein the peptides cover Spike glycoprotein of SARS-CoV-2 domains aa 304–338, 421–475, 492–519, 683–707, 741–770, 785–802, and 885–1273 (PepTivator SARS-CoV-2 Prot_S). For the M protein the peptide pools cover the entire 221 sequence membrane glycoprotein (“M”) of SARS-CoV-2 (GenBank MN908947.3, Protein QHD43419.1) (PepTivator SARS-CoV-2 Prot_M). Our analysis (fig. S1E) showed that both M-reactive and S-reactive CD8+ T cells were evenly distributed across all of the clusters of viral-reactive T cells, which indicated relatively similar transcriptional patterns of S- and M-reactive cells. For capturing FLU- and RSV-reactive CD8+ T cells, PepTivator Influenza A (H1N1) and RSV strain B1 peptide pools that covered the entire sequence of Hemagglutinin (HA) and Nucleoprotein (N) of each virus respectively, were obtained from Miltenyi Biotec.
Antigen-reactive T cell enrichment (ARTE) assay Virus-reactive CD8+ memory T cells were isolated using the protocol from Bacher et al. (24) with minor modifications. Thawed PBMC were sorted with FACSAria Fusion Cell Sorter (Becton Dickinson) to retrieve ex vivo pre-activated CD8+ T cells or plated overnight (5% CO2, 37°C) in 1 ml (concentration of 5 × 106 cells/ml) of TexMACS medium (Miltenyi Biotec) on 24-well culture plates. Each of the SARS-CoV-2-specific peptide pools (1 μg/ml) were added separately to the PBMC culture for 24 hours. For subsequent magnetic-based enrichment of CD137+ cells, cells were sequentially stained with human serum IgG (Sigma Aldrich) for FcR block, cell viability dye (eFluor780/APC.Cy7, eBioscience), fluorescence-conjugated antibodies, Cell-hashtag TotalSeq-C antibody (0.5 μg/condition, clone: LNH-94;2M2, Biolegend), and a biotin-conjugated CD137 antibody (clone REA765; Miltenyi Biotec) followed by anti-biotin microbeads (Miltenyi Biotec). The following fluorescence-conjugated antibodies were used: anti-human CD3 (UCHT1, Biolegend), CD4 (OKT4, Biolegend), CD8B (SIDI8BEE, eBioscience), CD137 (4B4-1, Biolegend), CD69 (FN50, Biolegend), CCR7 (3D12, BD Biosciences), CD45RA (HI100, Biolegend), CD38 (HB-7, Biolegend), HLA-DR (G46-6, BD Biosciences), PD-1 (EH12.1, Biolegend). Antibody-tagged cells were added to MS columns (Miltenyi Biotec) to positively select CD137+ cells. After elution, FACSAria Fusion Cell Sorter (Becton Dickinson) was utilized to sort memory CD8+ memory T cells expressing CD137 and CD69. Fig. S1A shows the gating strategy used for sorting. FlowJo software (v10.6.0) was employed for all Flow Cytometry Data analysis. In parallel, virus-reactive CD4+ memory T cells were isolated from the same cultures and analysis of their single-cell transcriptomes reported elsewhere (99).
Cell isolation and single-cell RNA-seq assay (10X Genomics platform) To facilitate the integration of single-cell RNA-seq and TCR-seq profiling from the sorted CD8+ T cells, 10x Genomics 5′TAG v1.0 chemistry was utilized. A maximum of 60,000 virus-responsive memory CD8+ T cells from up to 8 donors were sorted into a low retention 1.5mL collection tube, containing 500 μL of a solution of PBS:FBS (1:1) supplemented with RNase inhibitor (1:100). After sorting, ice-cold PBS was added to make up to a volume of 1400 μl. Cells were then spun down (5 min, 600 g, 4°C) and the supernatant was carefully aspirated, leaving 5 to 10 μl. The cell pellet was gently resuspended in 25 μl of resuspension buffer (0.22 μm filtered ice-cold PBS supplemented with ultra-pure bovine serum albumin; 0.04%, Sigma-Aldrich). Following that, 33 μl of the cell suspension was transferred to a PCR-tube and single-cell libraries prepared as per the manufacturer’s instructions (10x Genomics).
Single-cell transcriptome analysis Using 10x Genomics’ Cell Ranger software (v3.1.0) and the GRCh37 reference (v3.0.0) genome, reads from single-cell RNA-seq were aligned and collapsed into Unique Molecular Identifiers (UMI) counts. The Feature Barcoding Analysis pipeline from Cell Ranger was used to generate hashtag UMI counts for each TotalSeq-C antibody-capture library. UMI counts of cell barcodes were first obtained from the raw data output, and barcodes with less than 100 UMI for the most abundant hashtag were filtered out. Donor identities were assigned using Seurat’s (v3.1.5) MULTIseqDemux (autoThresh = TRUE and maxiter = 10) with the UMI counts. Cell barcodes were classified into three categories: donor ID (singlet), Doublet, Negative enrichment. Singlet cells were then stringently re-classified as doublet if the ratio of UMI counts between the top 2 barcodes was less than 3. All cells that were not classified as doublets or negative were used for downstream analyses. Cells from two COVID-19 patients with mild disease (patient 28 and 48) were not identifiable in the downstream analyses due to the lack of cell hashtags.
Single-cell RNA-Seq libraries (N = 4 and N = 15 for the ex vivo (0-hours) and in vitro activated (24-hours) cells, respectively) were aggregated using Cell Ranger’s aggr function (v3.1.0). Analysis of the combined data was carried out in the R statistical environment using the package Seurat (v3.1.5). To filter out doublets and to eliminate cells with low quality transcriptomes, cells were excluded if they were expressing < 800 or > 4400 unique genes, had < 1500 or > 20,000 total UMI content, and > 10% of mitochondrial UMIs. The summary statistics for each single-cell transcriptome library are given in table S3 and show good quality data with no major differences in quality control metrics between batches (fig. S1C). Only transcripts expressed in at least 0.1% of the cells were included for further analysis. Using default settings in Seurat software, the filtered transcriptome data was then normalized (by a factor of 10,000) and log-transformed per cell. The top variable genes with a mean expression greater than 0.01 counts per million (CPM) and explaining 25% and 16% of the total variance (for the 24- and 0-hours datasets, respectively) were selected using the Variance Stabilizing Transformation method. The transcriptomic data was then further scaled by regressing the number of UMIs detected and the percentage of mitochondrial counts per cell. This process was applied independently for the 0-hours and 24-hours datasets. Principal component analysis was performed using the top variable genes and based on the standard deviation of the principal components portrayed as an “elbow plot”, the first 16 principal components (PCs) were selected for the 0-hours dataset and the first 25 PCs were selected for the 24-hours dataset for downstream analyses. Cells were clustered using the FindNeighbors and FindClusters functions in Seurat with a resolution of 0.2 for either dataset. The robustness of clustering was verified by other clustering methods and by modifying the number of PCs and variable genes utilized for clustering. Analysis of clustering patterns of SARS-CoV-2-reactive CD8+ T cells across multiple batches revealed no evidence of strong batch effects (fig. S1D). Plots to visualize normalized UMI data were created using the Seurat package and custom R scripts.
Single-cell differential gene expression analysis MAST package in R (v1.8.2) was used to perform pair-wise single-cell differential gene expression analysis after converting UMI data to log2(CPM+1). For genes to be considered differentially expressed, the following thresholds were used: Benjamini-Hochberg–adjusted P-value < 0.05 and a log2 fold change greater than 0.25. Cluster markers (transcripts enriched in a given cluster) were determined using the function FindAllMarkers from Seurat.
Gene Set Enrichment Analysis and Signature Module Scores Signature lists were extracted from published data sets and databases. Gene names from murine datasets were converted to human gene names using the biomaRt R package. Gene lists were then filtered to exclude genes that were expressed (CPM > 0) in < 2% of the cells. Exhaustion consensus signature list was derived by considering genes that were present in > 3 exhaustion signature datasets. Genes that were present in cytotoxicity signatures or viral activation signatures were excluded from the consensus list (table S5). The R package fgsea was used to calculate the GSEA scores with the signal-to-noise ratio as a metric. Default parameters other than minSize = 3 and maxSize = 500 were used. Enrichment scores for each gene signature are presented as enrichment plots. A gene signature set list is considered to be significantly enriched if adjusted P-value is < 0.05.
Signature scores were estimated with the Seurat’s AddModuleScore function, using default settings. Briefly, the signature score is defined for each cell by the mean of the gene list of interest minus the mean expression of aggregate control gene lists. Control gene lists were sampled (size equal to the signature list) from bins defined based on the level of expression of the genes in the signature list. Signature gene lists used for analysis are provided
Single-cell trajectory analysis Monocle 3 (v0.2.1) was used to calculate the “branched” trajectory, settings included the number of UMI and percentage of mitochondrial UMI as the model formula, and taking the highly variable genes from Seurat for consistency. After using the PCA output from Seurat and allocating a single partition for all cells, the cell-trajectory was outlined on the UMAP generated from Seurat as well. The ‘root’ was selected using the get_earliest_principal_node function given in the package’s tutorial.
T cell receptor (TCR) sequence analysis Single-cell libraries enriched for V(D)J TCR sequences were processed to get clonotype information for each independent sample with the vdj pipeline from Cell Ranger (v3.1.0 and human annotations reference GRCh38, v3.1.0, as recommended). Joint analysis of single-cell transcriptomes and TCR repertoires was performed by aggregating independent libraries through custom scripts. For this purpose, cell barcodes were matched between corresponding libraries from each type. Then, every unique clonotype, a set of productive Complementarity-Determining Region 3 (CDR3) sequences as defined by 10x Genomics, was identified across all library annotations files. Finally, clone statistics, mainly clonotypes’ frequencies and proportions, were recalculated for the whole aggregation so that previously-identified good quality cells were annotated with a specific clonotype ID and such clone statistics. Clone size was calculated as the number of cells expressing a given clonotype ID, and a clonotype was called as clonally expanded if this value was greater than 1 (clone size ≥ 2). These steps were followed either 1) independently for the 0- and 24-hours datasets (describing ex vivo and in vitro activated cells, respectively) or 2) for the whole set of cells in order to assess clonal sharing between the two conditions (tables S6 and S7). Clone size was depicted on UMAP (per cell) or in violin plots (per group, where color indicated clone size median of each group) using custom scripts and clonotype sharing was presented using the UpSetR package. For the comparison between ex vivo and in vitro activated cells, only SARS-CoV-2-reactive CD8+ cells specifically isolated from matched patients between datasets were considered for the 24-hours data.
Ingenuity Pathway Analysis (IPA) IPA was performed using default setting (v01-16) on transcripts that were significantly increased in expression in cluster 0 cells from patients with severe COVID-19 illness compared to mild illness. The canonical pathway analysis was performed to elucidate the enriched pathways in this data set and to visualize and highlight the gene overlap between the given data set with a particular enriched pathway. The upstream regulator network analysis was used to identify and visualize the interactions between differentially expressed downstream target genes on a given dataset with a particular upstream regulator.
Statistical Analysis GraphPad Prism 8.4.3 software was utilized for relevant data statistical analysis. Detailed information regarding statistical analysis, including test types and number of batches or samples is provided in the figure legends. P values are specified in the text or the figure legends. The data normality tests were performed and for data that fell within Gaussian distribution, appropriate parametric statistical tests were performed and for those that did not conform to the equal variance-Gaussian distribution, appropriate non-parametric statistical tests were used.
Reference & Source information: https://immunology.sciencemag.org/
Read More on: