Systematic functional interrogation of SARS-CoV-2 host factors using Perturb-seq

Abstract

Genomic and proteomic screens have identified numerous host factors of SARS-CoV-2, but efficient delineation of their molecular roles during infection remains a challenge. Here we use Perturb-seq, combining genetic perturbations with a single-cell readout, to investigate how inactivation of host factors changes the course of SARS-CoV-2 infection and the host response in human lung epithelial cells. Our high-dimensional data resolve complex phenotypes such as shifts in the stages of infection and modulations of the interferon response. However, only a small percentage of host factors showed such phenotypes upon perturbation. We further identified the NF-κB inhibitor IκBα (NFKBIA), as well as the translation factors EIF4E2 and EIF4H as strong host dependency factors acting early in infection. Overall, our study provides massively parallel functional characterization of host factors of SARS-CoV-2 and quantitatively defines their roles both in virus-infected and bystander cells.

Fig. 1. Perturb-seq for single-cell transcriptional analysis and functional validation of SARS-CoV-2 host factors.

A Experimental design for the Perturb-seq experiment in Calu-3 cells engineered to express CRISPRi machinery. We perturbed 183 different host factors (individually or in combination) using a lentivirally-delivered library, infected the cells with SARS-CoV-2 for 24 hours, and performed droplet-based single-cell RNA sequencing, reading out host and viral transcripts as well as the sgRNA, indicating the perturbed host factor. B Single-cell transcriptomes were projected into UMAP space and colored by viral RNA fraction per cell. C, D Density of cells identified as either uninfected/bystander (C) or infected (D) by our classifier, overlaid onto all cells in gray. E Cells color-coded by their cell cycle phase. F Fraction of bystander and infected cells assigned to each cell cycle phase. G Cells color-coded by the number of detected UMIs per cell.

Fig. 2. Transcriptional heterogeneity in SARS-CoV-2 infection.

A Single-cell transcriptomes were projected in UMAP space and colored by Leiden cluster. Leiden clusters were subsequently characterized by the mean viral fraction, the number of cells, and the cell cycle composition per cluster. Cluster T are all cells that could not be assigned an unambiguous infection state. B Differential expression of Leiden clusters revealed transcriptionally distinct subclusters of bystander cells, infected cells, and a small subset of interferon-producing cells. The color of each dot is pseudobulk gene expression of each gene per cluster, and the size of each dot is the expression normalized to the cluster with maximum expression of that gene. C–E Host transcriptional analysis revealed heterogeneity in infected and bystander populations, including differential gene expression in UMAP space of: C ISG15; D IFNL1; and E NFKBIA.

Fig. 3. Host perturbations alter SARS-CoV-2 infection dynamics.

A The effect of how each CRISPRi perturbation altered viral load was displayed as the change in mean viral load by KS p-value (Kolmogorov–Smirnov test, two-sided) of viral load distribution change. Color code indicates CRISPRi targets, non-targeting controls, and targets in which knockdown significantly altered viral loads. B To orthogonally validate CRISPRi targets, we transduced Huh7.5.1 cells overexpressing ACE2 and TMPRSS2 with lentivirus targeting control and test genes. Cells were subsequently infected with SARS-CoV-2, and percent infection was calculated (C) by immunofluorescence and microscopy (D), counting the fraction of cells positive for SARS-CoV-2 nucleoprotein. Original images were taken on an inverted fluorescence microscope at 4× (>2000 cells). Representative, zoomed-in fields of view are shown and the scale bar represents 100 µm. In the bar plot, each bar represents the mean percent infection for a given cell line and points represent the individual data points. Additionally, we quantified infectious virion production using the TCID50 assay and calculated fold change for each cell line relative to the mean TCID50/mL of non-targeting control cells. E In the bar plot, each bar represents the mean TCID50/mL for a given cell line, and the points represent the individual data points. The ACE2 knockout positive control was measured once for the TCID50 assay; two biological replicates were performed for all other infection conditions (C–E). Source data for orthogonal validation are provided as a Source Data file.

Fig. 4. Host perturbations alter localization of cells in UMAP space and Leiden cluster membership.

A Library elements for non-targeting controls, factors that alter SARS-CoV-2 infection, and interferon signaling are highlighted in UMAP space. B Library element representation by cluster was calculated, normalized, and visualized on a heatmap. C, D Subsequent dimensionality reduction of this odds-ratio was projected into UMAP space and revealed subclusters by biological function.

Fig. 5. Host perturbations alter interferon signaling in bystander-activated cells.

A We scored bystander cells based on their ability to respond to interferon (ISG score) and tested which perturbations significantly altered the ISG score distribution by perturbation. This is represented by the mean change in ISG score when compared to non-targeting controls by the KS p value (Kolmogorov–Smirnov test, two-sided) per perturbation. B Expression heatmap of select targeting and non-targeting library elements showing the mean z-score for a subset of interferon-stimulated genes.