Profiling of immune dysfunction in COVID-19 patients allows early prediction of disease progression

(A) The ratio of CD4 to CD8 cells is dependent on disease state and clinical intervention. (B) The abundance of CD45RA/RO cells in either CD4⁺ or CD8⁺ compartments is dependent on disease state or clinical intervention. (C) Abundance of immune populations changes significantly between disease states. (D) Uniform Manifold Approximation and Projection (UMAP) projection of all cells colored by either surface receptor expression, cluster assignment, or disease severity. (E) Immune phenotype of each cluster (top) and its composition in disease severity (bottom). (F) Expression levels of CD25 and FAS receptors in the UMAP projection. (G) FAS expression across all clusters depending on disease severity (left) and the proportion of cells not expressing FAS for each sample (right). (H) Scatter plot of CD25 and FAS expression for each cell according to disease severity. (I) Abundance of CD4⁺ CXCR5⁺ PD-1⁺ T_FH by disease severity. (J) Immune populations with significantly different amounts of cells expressing immune checkpoint receptors by disease severity. Significance was assessed by Mann–Whitney U tests and corrected for multiple testing with the Benjamini–Hochberg false discovery rate (FDR). **FDR-adjusted P-value < 0.01; *FDR-adjusted P-value 0.01–0.05.

(A) Abundance of myeloid-derived suppressor cells as a percentage of all immune cells according to disease severity. (B) Uniform Manifold Approximation and Projection (UMAP) projection of all cells from all patients colored by the expression levels of surface receptors, derived clusters, or disease severity among all patients. (B, C) Immune profile of each cluster from (B) based on the expression of surface markers (top) and composition in disease severity (bottom). (D) Expression levels of CD15 dependent on disease severity (left) and quantification of cells expressing it (right) according to CD16, CD3, and CD33 expression. (E) Abundance of cells expressing various KIR receptors as a percentage of NK cells according to disease severity. (F) UMAP projection of all cells from all patients colored by the expression of surface receptors, derived clusters, or disease severity. (F, G) Immune profile of each cluster from (F) based on the expression of surface markers (top) and composition in disease severity (bottom). (H) Expression levels of all four measured KIR receptors in each disease state. Significance was assessed using Mann–Whitney U tests and corrected for multiple testing with the Benjamini–Hochberg false discovery rate (FDR). **FDR-adjusted P-value < 0.01; *FDR-adjusted P-value 0.01–0.05.

(A, B) The abundance of total B cells, plasma, and IgG+ and IgG+ cells between disease states. (C) Uniform Manifold Approximation and Projection (UMAP) projection of all cells colored by surface receptor expression, cluster assignment, or disease severity. (D) Immunophenotype of each cluster (top) and its composition by disease severity (bottom). (E) Identification and quantification of five populations of B cells dependent on CD20 and CD19 expression. (E, F) Comparison of the abundance of the populations identified in (E) between disease states. Significance was assessed using Mann–Whitney U tests and corrected for multiple testing with Benjamini–Hochberg false discovery rate (FDR). **FDR-adjusted P-value < 0.01; *FDR-adjusted P-value 0.01–0.05.

(A) Projection of immune profiles into a two-dimensional latent space that reconstructs the hierarchy of disease progression. The x-axis represents disease progression in the pseudo-temporal space. Sample from patients which died from COVID-19 are marked with a diagonal black line. (B) Distribution of samples grouped by disease state along the pseudo-temporal axis derived in (B). (C) Immune populations associated with the pseudo-temporal axis represented by either the absolute change in percentage in their extremes (x-axis) or strength of linear association (y-axis). (D) Clusters of immune populations based on their abundance along the pseudo-temporal axis. (D, E) Examples of immune populations from each cluster in (D).

(A) Directed graph of clinical factors (green) and immune populations (pink). Edges represent the association between factors and immune populations and are colored by the direction and strength of association (blue, negative; red, positive). (B) Abundance of select immune populations with significantly different responses between sexes dependent on outcome. (C) Estimated coefficients of change for severe versus mild disease (left) or tocilizumab treatment (right) for immune populations that change discordantly. (D, E) Abundance of select immune populations with significantly different responses between sexes dependent on tocilizumab treatment (D) or intubation (E). (F) Graphical depiction of the machine-learning framework for predicting disease severity using the earliest available samples per patient and cross validation. (G, H) Performance of classifiers trained with real or randomly shuffled labels and either all immune populations (G) or with selection for the top most predictive eight populations (H). (I) Predicted severity scores over time since symptoms started for immune profiles from patients with at least three longitudinal sampling points. (G, J) Relative expression of CD25, CD45RA, and CD45RO over time in four patients from (G). **False discovery rate-adjusted P-value < 0.01; *false discovery rate-adjusted P-value 0.01–0.05.