Genome-wide CRISPR screens in Vero E6 cells identify host factors required for HCoV infection
In order to identify host factors that promote HCoV infection, we performed genome-wide loss of function CRISPR screens for pathogenic SARS-CoV-2 and common cold-causing OC43 in two susceptible cell lines. Due to the highly cytopathic nature of HCoV infection in the Vero E6 cells derived from African Green Monkey (AGM; Chlorocebus sabaeus), we carried out genome-wide screens using a custom Vervet CRISPR knockout library (see supplemental methods) (Fig. 1A). Vero E6 cells were transduced with the Vervet CRISPR library and infected with SARS-CoV-2 or OC43 at a multiplicity of infection (MOI) 0.01. We observed ~ 60% visible cytopathic effect (CPE) for SARS-CoV-2 and ~ 85% CPE for OC43. Resistant clones were expanded, reinfected with the corresponding virus at MOI 0.1, and re-expanded. Genomic DNA was extracted from surviving cells, sgRNAs amplified, and sequenced. We carried out MAGeCK analysis to identify genes targeted by significantly enriched sgRNAs which are labeled in the volcano plots in Fig. 2A, B. Full data sets are available in Additional files 4 and 5. To facilitate the access and reusability of data sets generated from genome-scale CRISPR screens in HCoV-infected cells, we have also hosted a website (sarscrisprscreens.epi.ufl.edu) with the complete set of MAGeCK results for each of the screens described in this study and data from previously published screens reanalyzed herein [8, 12,13,14,15]. The website was designed to facilitate integration of upcoming screens and we hope for contributions to drive this as a community project.
We identified multiple candidate host factors previously demonstrated to play a functional role in SARS-CoV-2 and OC43 infections. For example, ACE2 was identified in the SARS-CoV-2 screen . Furthermore, TMEM41B was a top-scoring gene in the OC43 screen, supporting recent work by Schneider et al. demonstrating that TMEM41B is a pan-HCoV host factor . We also identified interferon (IFN)-induced transmembrane (IFITM) proteins that have been reported to regulate HCoV infection [35,36,37]. Genes targeted by significantly enriched sgRNAs were next segregated into functional categories listed in tables in Fig. 2C, D. Of note, CDK4, a master regulator of the cell cycle, was identified as a key host factor for both viruses. Disruption of additional genes encoding regulators of cell cycle progression, including CDK1NA, DYRK1A, HRK, and P53, similarly increased cellular resistance to SARS-CoV-2 infection.
During the completion of our studies, Wei et al. reported similar SARS-CoV-2 screens in Vero E6 cells performed with an independent sgRNA library based on an earlier C. sabaeus genome assembly . In order to compare our data sets to those of Wei et al., we downloaded raw data from their study and analyzed them using MAGeCK-VISPR [13, 33] (Additional file 1: Supplementary methods and Additional file 3). There were 6 targeted genes identified in common between studies: ACE2, DPF2, DYRK1A, RAD54L2, SMARCA4, and TP53.
Genome-wide CRISPR screens in HEK293T-hACE2 cells identify host factors required for HCoV infection
We similarly performed CRISPR screens in human HEK293T cells ectopically expressing the human ACE2 receptor (HEK293T-hACE2) transduced with the Brunello sgRNA library  (Fig. 1B). Transduced cells were infected with SARS-CoV-2 or OC43 at MOI 0.01. SARS-CoV-2-infected cultures developed ~ 40% CPE and OC43-infected cultures developed > 85% CPE. Resistant cell populations propagated to confluence were reinfected with the corresponding virus at either MOI 0.01 or MOI 0.1 and re-expanded. Genomic DNA was extracted and sgRNAs from both the initial and secondary infections were sequenced.
The genes targeted by the most highly enriched sgRNAs in each of the SARS-CoV-2 infections are indicated in Fig. 3A–C. The full data set is available in Additional file 6. EDC4, a gene encoding a scaffold protein that functions in programmed mRNA decay, was the overall top-scoring gene. Interestingly, XRN1 encodes another key player in this pathway and was also top-scoring. We further categorized the genes encoding candidate host factors (FDR < 0.1) into functional categories depicted in heat maps in Fig. 3D. Consistent with other published screens, we identified multiple components of the endocytic pathway including CCZ1, DNM2, and WASL. Other functional categories in which multiple genes were identified include cell adhesion, cell cycle, integrator complex, lysosome, mTOR regulation, and ubiquitination/proteolysis. We carried out an independent SARS-CoV-2 screen using the higher MOI of 0.3 for initial infection which resulted in ~ 80% CPE, and MOI 0.03 for secondary infection. Genes targeted by the most significantly enriched sgRNAs in this study are presented in Fig. 3E, F and are segregated into functional categories depicted in heat maps in Fig. 3G. The full data set is available in Additional file 7. ACE2 was a top-scoring gene in this screen. Functional categories with multiple targeted genes include amphisome, autophagy, endosome, exocytosis, lysosome, peroxisome, transcription/transcriptional regulation, and ion transporters. C18orf8, CCZ1, CDH2, and TMEM251 were identified in both the low- and high-MOI SARS-CoV-2 screens.
The genes targeted by the most highly enriched sgRNAs in the OC43 HEK293T-hACE2 screens are indicated in Fig. 4A–C and segregated into functional categories in Fig. 4D. The full data set is available in Additional file 8. As expected, based on prior work, genes encoding IFITM proteins were identified as proviral factors for OC43 . TMEM41B was a top-scoring gene along with the functionally related VMP1, as were CCZ1, CCZ1B, SLC35B2, and WDR81 which have all been reported in other recent OC43 genome-wide screens [11, 12]. When comparing the SARS-CoV-2 and OC43 HEK293T-hACE2 datasets, there were 6 genes in common targeted by significantly enriched sgRNAs (C18orf8, CCZ1, CCZ1B, RAB7A, WDR81, and WDR91). Notably, all of the corresponding gene products function in vesicle-mediated transport.
During the completion of our studies, similar SARS-CoV-2 screens in human Huh-7.5 [11, 12, 15] and A549 [10, 14] cells were published. In order to compare our data sets to those of other groups, we downloaded raw data from four published studies [11, 12, 14, 15], analyzed them using a common analysis framework (MAGeCK) and stringency (FDR < 0.25), and compared the results to our data sets (Additional files 9 and 10). Using this stringency, no genes were identified in all five studies, 1 gene was identified in four studies (ACE2), 6 genes were identified in three studies (VPS35, CTSL, DNM2, CCZIB, TMEM106B, and VAC14), and 25 genes were identified in two studies (ALG5, ARVCF, ATP6V1A, ATP6V1G1, B3GAT3, CNOT4, EPT1, EXOC2, EXT1, EXTL3, GDI2, LUC7L2, MBTPS2, PIK3C3, RAB7A, RNH1, SCAF4, SCAP, SLC30A1, SLC33A1, SNX27, TMEM41B, TMEM251, WDR81, and WDR91) (Fig. 5A, B). It should be noted that these genes were top-scoring across studies performed in different human cell lines, suggesting they are broadly important in SARS-CoV-2 replication. Shared pathways include vesicle-mediated transport (CCZ1B, DNM2, EXOC2, GDI2, PIK3C3, RAB7A, SNX27, VAC14, VPS35, WDR81, WDR91), vacuolar ATPases important in organelle acidification (ATP6V1A, ATP6V1E, ATP6V1G1), and heparan sulfate biosynthesis genes (EXT1, EXTL3, B3GAT3). We identified 53 genes targeted by enriched sgRNAs in our study that were not identified in published studies (Fig. 5C), including EDC4 and XRN1.
Validation of a subset of gene candidates that promote HCoV replication
To confirm that unique genes identified in our screens promote HCoV replication, HEK293T-hACE2 cells were engineered to stably express gene-specific shRNAs targeting CCZ1 or EDC4. CTSL knockdown was tested as a positive control for SARS-CoV-2 . Efficiency of gene knockdown assessed by western blotting was robust for all three genes (Fig. 6A and Additional file 1: Fig. S3A). Knockdown cells were then infected with SARS-CoV-2 or OC43 and viral genome copy number determined at 2 days post-infection (dpi). All three genes were required for optimal SARS-CoV-2 infection while CCZ1 and EDC4, but not CTSL, promoted OC43 infection (Fig. 6B). Because EDC4 was unique to our screens, we next tested whether it plays a role in promoting HCoV replication in respiratory cells. The small airway epithelial cell (SAEC) line was engineered to express human ACE2 (hACE2) and then EDC4 was targeted for knockout using a CRISPR/Cas9-based approach. As mentioned above, Xrn1 functions in the same mRNA decay pathway as Edc4 and was also identified in our screens so we engineered Xrn1−/− SAEC-hACE2 expressing cells in parallel. After verifying gene knockout (Fig. 6C and Additional file 1: Fig. S3B), cells were infected with either SARS-CoV-2 or OC43 and virus titers measured at various time points by TCID50 assay. EDC4 and XRN1 were necessary for efficient replication of both viruses (Fig. 6D).
CRISPR screening reveals novel antiviral drugs displaying in vitro efficacy
We next determined whether gene products and pathways identified in our screens could be targeted with commercially available inhibitors to block HCoV infection. Numerous genes involved in cell cycle regulation were identified in our screens. The following inhibitors targeting this class of host factors were tested: abemaciclib (ABE; Cdk4, Cdk6 inhibitor), UC2288 (UC2; CDKN1A/p21 inhibitor), harmine (HAR) and INDY (Dyrk1A/B inhibitors), AZ1 (Usp25/28 inhibitor), olaparib (OPB; ARID1A inhibitors were not available so inhibition of PARP-mediated DNA repair was investigated to see if DNA damage repair was involved in SARS-CoV-2 replication), and nintedanib (NIN; FGFR1/2/3, VEGFR1/2/3, and PEGFRα/β inhibitor). Host factors involved in endocytosis have been widely reported to regulate HCoV replication and were identified in our and others’ CRISPR screens  so we also tested several drugs targeting this process including CID1067700 (CID; Rab7a inhibitor), chlorpromazine (CPZ) and promethazine (PMZ) both suppress clathrin function in cells through signaling receptor inhibition. Finally, we tested amlexanox (AMX) which inhibits TANK binding kinase 1 (TBK1) and its adaptor protein TBK-binding protein 1 (TBKBP1) which has been reported to variously regulate Rab7a activity  or induction of IFN response genes . The heat map in Fig. 7A shows the fold enrichment of sgRNAs targeting the genes of interest across the screens performed in this study.
In an initial experiment of the entire panel of small molecules, inhibitors were added to culture supernatants at the initiation of SARS-CoV-2 infection and evaluated for their capacity to inhibit virus-induced CPE at 3 dpi in Vero E6 cells. The concentrations of inhibitors used, based on available toxicity data, were generally nontoxic in Vero E6 cells (Fig. 7B, white bars). ABE, AMX, HAR, NIN, OPB, PMZ, and UC2 significantly inhibited virus-induced cytotoxicity while AZ1, CPZ, INDY, and CID did not (Fig. 7B, gray bars). Our CID results are consistent with prior work which showed reduced CoV egress, but no effect on cell viability or viral replication, in response to CID treatment . Although INDY and HAR both target Dyrk1A, only HAR displayed activity in this assay, potentially due to the lower enzymatic EC50 of HAR for Dyrk1A (0.24 μM for INDY vs. 0.08 μM for HAR). As a complementary approach to measure antiviral activity of these compounds, we quantified viral genome copies by RT-qPCR in cells treated with each compound at 2 dpi. ABE, AMX, HAR, PMZ, and UC2 significantly decreased viral genome copy number (Fig. 7C), consistent with their ability to protect from virus-induced cytotoxicity. On the other hand, NIN and OPB had no effect on viral genome copy number despite their moderate inhibition of SARS-CoV-2-induced cytotoxicity. Conversely, AZ1 completely inhibited viral genome replication in spite of having no significant effect on cytotoxicity.
For compounds displaying activity in one or both of these assays, we next determined their EC50 and CC50 against SARS-CoV-2 infection by cytotoxicity measurements in the presence or absence of virus across a series of inhibitor dilutions (Fig. 7D–J). Several of the inhibitors had EC50 values below 20 μM (10.86 μM for ABE, 14.1 μM for NIN, and 2.16 μM for UC2), with the p21 inhibitor UC2 being the most potent. AMX is typically used as a topical treatment and had a high EC50 at 342.96 μM. AZ1 (37.49 μM), HAR (61.44 μM), and PMZ (88.41 μM) showed intermediate EC50 levels. The selectivity indices (SI; ratios of CC50 to EC50) of the investigated compounds from highest to lowest are AMX > 53.23, ABE > 5.68, UC2 > 7.40 with the SI of AZ1, HAR, NIN, and PMZ falling below 2. For comparison, the SI of clinically relevant antiviral drugs are as follows: remdesivir is > 129.87, nafamostat > 4.44, and ribavirin > 3.65 . Generally, determining EC50 with cytotoxicity measurements results in overestimation of EC50, leading to a conservative estimate of SI.
A subset of inhibitors displaying efficacy in Vero E6 cells was further assessed for their capacity to inhibit SARS-CoV-2 and OC43 replication in SAEC-hACE2 or SAEC, respectively, using TCID50 assay as a readout of infectious virus titers. ABE, AMX, AZ1, HAR, NIN, and PMZ inhibited OC43 replication (Fig. 8B) and all compounds except for AZ1 inhibited SARS-CoV-2 replication (Fig. 8A). Phospholipidosis of cell membranes by drug treatment has been implicated as a confounding issue during in vitro viral inhibition screens . While others have disputed this claim , we decided to test our compounds for phospholipidosis induction. In Fig. 8C–H, phospholipidosis was measured in SAEC-hACE2 cells treated with each inhibitor. Compared to the positive phospholipidosis control amiodarone (AMD), induction of phospholipidosis by PMZ was strongest followed by HAR. The phospholipidosis curve for HAR was biphasic, indicating a potential therapeutic window between 2.5 and 20 μg/ml. NIN induced minimal levels of phospholipidosis while ABE and AMX did not induce phospholipidosis. Overall, these findings reveal novel candidates for anti-HCoV drug development.