Reversal of Infected Host Gene Expression Identifies Repurposed Drug Candidates for COVID-19

Jing Xing; Rama Shankar; Aleksandra Drelich; Shreya Paithankar; Eugene Chekalin; Thomas Dexheimer; Surender Rajasekaran; Chien-Te Kent Tseng; Bin Chen

To utilize this approach for drug discovery against SARS-CoV-2, we first need to collect virus-related host gene expression profiles, which were not available at the time of writing. Given the high genomic similarity between SARS-CoV, MERS-CoV, and SARS-CoV-2, we reasoned that existing host gene expression profiles of the samples infected by SARS-or MERS-CoV could approximate to those infected by SARS-CoV-2. To verify this assumption, we compiled 331 virus-induced signatures from enrichR and GEO (Table S1 ) and used an established pipeline to score 1740 drugs in our repurposing library regarding their reversal of signature gene expression. Clustering of these signatures based on their drug prediction scores suggests that signatures derived from the same virus or the virus family under the similar experimental model tend to cluster together ( Figure S1 ). An example cluster includes one signature derived from primary human microvascular endothelial cells (MMVE001) after 48h of MERS-CoV infection (study id: GSE79218) and another derived from melanoma cells in mice after seven days of SARS-CoV infection (study id: GSE68820). In addition, Spearman correlation coefficient of the in vitro drug efficacy data (EC50: Half maximal effective concentration) of SARS-CoV and MERS-CoV is up to 0.6 ( Figure 1B) . The clustering and correlation results suggested that drugs predicted based on the signatures related to SARS-CoV and MERS-CoV could also be applied for SARS-CoV-2. Therefore, we developed a pipeline to repurpose existing drugs against MERS-CoV and SARS-CoV, and then experimentally evaluate these drugs in SARS-CoV-2 ( Figure 1C ). database. A good candidate should activate the repressed biological processes and inhibit the upregulated processes. B, Correlation of the published antiviral activities of 30 drugs (pEC50, -log10 transformed EC50 value in mol/L) against MERS-and SARS-CoV. C, Study workflow including creation of disease signatures, prediction of drug candidates, selection of a final drug list, and in vitro validation. One disease signature composed by the differentially expressed genes of each comparison led to one drug prediction list. Only the signature resulting in a prediction list where known positive drugs were enriched on the top was considered as a valid signature. D, Dysregulated pathways after SARS infection at 7h compared with 2h in lungs. E, The enrichment of top six dysregulated pathways in primary human microvascular endothelial (MMVE001) cells through 0h to 48h in MERS-CoV infection (left) and in mock (right). Only one study was selected for D and E, respectively, and the dysregulated pathways and their dynamics for other studies are available in supplementary materials ( Figure S2 and S3, Extended Data 1).