Immunophenotyping of COVID-19 and influenza highlights the role of type I interferons in development of severe COVID-19

Single-cell transcriptomes of PBMCs from patients with COVID-19 and influenza

PBMCs were collected from healthy donors (n=4), hospitalized patients with severe influenza (n=5), and patients with COVID-19 of varying clinical severity, including severe, mild, and asymptomatic (n=8). PBMCs were obtained twice from three (the subject C3, C6, and C7) of the eight COVID-19 patients at different time points during hospitalization. PBMC specimens from COVID-19 patients were assigned to severe or mild COVID-19 groups according to the National Early Warning Score (NEWS; mild 5, severe ≥ 5) evaluated on the day of whole blood sampling (16). In NEWS scoring, respiratory rate, oxygen saturation, oxygen supplement, body temperature, systolic blood pressure, heart rate, and consciousness were evaluated (16). Severe influenza was defined when hospitalization was required irrespective of NEWS score. Patients with severe influenza were enrolled from December 2015 to April 2016, prior to the emergence of COVID-19. The severe COVID-19 group was characterized by significantly lower lymphocyte count and higher serum level of C-reactive protein than the mild COVID-19 group on the day of blood sampling (Fig. S1A). Multiplex real-time PCR for N, RdRP, and E genes of SARS-CoV-2 was performed, and there was no statistical difference in Ct values for all three genes between two groups (Fig. S1B). Demographic information is provided with experimental batch of scRNA-seq in Table S1 and clinical data in Table S2 and S3.

Employing the 10X Genomics scRNA-seq platform, we analyzed a total of 59,572 cells in all patients after filtering the data with stringent high quality, yielding a mean of 6,900 UMIs per cell and detecting 1,900 genes per cell on average (Table S4). The transcriptome profiles of biological replicates (PBMC specimens in the same group) were highly reproducible (Fig. S1C), ensuring the high quality of the scRNA-seq data generated in this study.

To examine the host immune responses in a cell type-specific manner, we subjected 59,572 cells to t-distributed stochastic neighbor embedding (tSNE) based on highly variable genes using the Seurat package (17) and identified 22 different clusters unbiased by patients or experimental batches of scRNA-seq (Fig. 1A, Fig. S1D). These clusters were assigned to 13 different cell types based on well-known marker genes and two uncategorized clusters (Fig. 1B and C, and Table S5). In downstream analysis, we only focused on 11 different immune cell types, including IgG B cell, IgG+ B cell, effector memory (EM)-like CD4+ T cell, non-EM-like CD4+ T cell, EM-like CD8+ T cell, non-EM-like CD8+ T cell, natural killer (NK) cell, classical monocyte, intermediate monocyte, non-classical monocyte, and dendritic cell (DC) after excluding platelets, red blood cells (RBCs), and two uncategorized clusters. The subject C8 (asymptomatic case) was also excluded due to a lack of replicates. In hierarchical clustering, most transcriptome profiles from the same cell type tended to cluster together, followed by disease groups, suggesting that both immune cell type and disease biology, rather than technical artifacts, are the main drivers of the variable immune transcriptome (Fig. S1E).

Fig. 1 Single cell transcriptomes of PBMCs from COVID-19 and influenza patients.

(A) tSNE projections of 59,572 PBMCs from healthy donors (HDs) (4 samples, 17,590 cells), severe influenza (FLU) patients (5 samples, 10,519 cells), COVID-19 patients (asymptomatic: 1 sample, 4,425 cells; mild COVID-19: 4 samples, 16,742 cells; severe COVID-19: 6 samples, 10,296 cells) colored by group information. (B) Normalized expression of known marker genes on a tSNE plot. (C) tSNE plot colored by annotated cell types. EM: effector memory, NK cell: natural killer cell, DC: dendritic cell, RBC: red blood cell. (D) Proportion of cell types in each group excluding ‘Uncategorized 1’, ‘Uncategorized 2’, ‘RBC’, and ‘Platelet’. The colors indicate cell type information. (E) Boxplots showing the fold enrichment in cell type proportions from mild COVID-19 (n=4), severe COVID-19 (n=6), and FLU (n=5) patients compared to the HD group (mild COVID-19 vs. HD: n=16, severe COVID-19 vs. HD: n=24, FLU vs. HD: n=20). For the boxplots, the box represents the interquartile range (IQR) and the whiskers correspond to the highest and lowest points within 1.5 × IQR. ‘Uncategorized 1’ (relatively high UMIs per cells and presence of multiple marker genes), ‘Uncategorized 2’ (B cell-like and high expression of ribosomal protein genes), ‘RBC’, and ‘Platelet’ were excluded. Two-sided Kolmogorov–Smirnov (KS) tests were conducted for each cell type between the disease and HD group. *p0.05, **p0.01, and ***p0.001.

As a feature of immunological changes, we investigated the relative proportions of immune cells among PBMCs in the disease groups compared to the healthy donor group (Fig. 1D and E, and Fig. S1F). Unlike the limited changes in mild COVID-19, significant changes were observed in both influenza and severe COVID-19 across multiple cell types among PBMCs. In severe COVID-19, the proportion of classical monocytes significantly increased whereas those of DCs, non-classical monocytes, intermediate monocytes, NK cells, EM-like CD8+ T cells, and EM-like CD4+ T cells significantly decreased (Fig. 1E). In severe influenza, the proportion of classical monocytes significantly increased whereas those of DCs, non-EM-like CD4+ T cells, EM-like CD4+ T cells, IgG+ B cells, and IgG B cells significantly decreased. We validated the proportions of immune cell subsets from scRNA-seq by flow cytometry analysis. The relative proportions of total lymphocytes, B cells, CD4+ T cells, CD8+ T cells, NK cells, and total monocytes from scRNA-seq significantly correlated with those from flow cytometry analysis (Fig. S1G).