Ifenprodil and Flavopiridol Identified by Genomewide RNA Interference Screening as Effective Drugs To Ameliorate Murine Acute Lung Injury after Influenza A H5N1 Virus Infection

Drug repurposing is a quick and economical strategy for developing new therapies with approved drugs. H5N1 is a highly pathogenic avian influenza virus subtype that can cause severe acute lung injury (ALI) and a high mortality rate due to limited treatments. The use of RNA interference (RNAi) is a reliable approach to identify essential genes in diseases. In most genomewide RNAi screenings, virus replication is the readout of interference. Since H5N1 virus infection could induce significant cell death and the percentage of cell death is associated with virus lethality, we designed a genomewide RNAi screening method to identify repurposable drugs against H5N1 virus with cell death as the readout. We discovered that the neurological drug ifenprodil and the anticancer drug flavopiridol could effectively ameliorate murine ALI after influenza A H5N1 virus infection, suggesting that they might be novel remedies for H5N1 virus-induced ALI in addition to the traditional indications.

I nfluenza A virus is the most common type among the viruses that cause influenza epidemics, and influenza A viruses can be divided into various subtypes based on the hemagglutinin and neuraminidase proteins on the viral surface (1). H5N1 is a highly pathogenic avian influenza virus subtype that can cause severe acute lung injury (ALI) in human subjects (2). Infection with H5N1 virus is associated with a 52.79% mortality rate (https://www.who.int/influenza/human_animal_interface/2019_02_12_tableH5N1 .pdf?uaϭ1). However, effective treatments are limited. The use of corticoids is controversial, and extracorporeal membrane oxygenation (ECMO), artificial liver support system (ALSS), and mechanical ventilation are often expensive (3). Thus, there is an urgent and continuous need for novel treatments.
Drug repurposing is a quick and economical strategy to identify new therapeutic uses of approved drugs outside their original scope and is estimated to be the source of approximately 30% of newly approved drugs every year (4). An example of successful drug repurposing is the use of the chemotherapeutic agent azidothymidine as a therapy for HIV in the 1980s (4). Thus, the exploration of an efficient method for drug repurposing is necessary.
RNA interference (RNAi) represents a reliable approach to identify genes important in diseases such as acute myeloid leukemia, epithelial ovarian cancer, and schistosomiasis (5)(6)(7). Genomewide RNAi screening has been used to identify host genes involved in the replication of the moderately pathogenic influenza H1N1 virus (8)(9)(10)(11)(12)(13)(14)(15). Data mining using public databases such as the DrugBank database and the Therapeutic Target Database (TTD) has been increasingly used to identify candidate drugs that could potentially interact with known validated target genes (16,17). We combined RNAi screening and DrugBank and TTD target gene searches to develop an effective method for identifying repurposable drugs.
In the current study, we used genomewide RNAi screening to identify host genes that influence host cell viability following H5N1 virus infection. The results showed Ͻ10% overlap in the candidate host gene targets between H5N1 and the less lethal H1N1 strain. Next, we screened 1,137 critical host genes in RNAi screens using the DrugBank database and identified 372 approved or clinically investigated drugs (with 146 target genes). Validation experiments performed with 104 commercially available drugs (and 65 target genes) identified 28 candidate drugs that significantly increased the viability of H5N1-infected A549 cells. The two most effective drugs among these candidates in a mouse model of H5N1-induced ALI were ifenprodil, an agent with neurological indications, and flavopiridol, an antineoplastic agent.  S2). Among the top 10 functional enrichment pathways, 4 were related to the immune response and 3 to apoptosis (Fig. 1D). Our results obtained through the RNAi platform show that cell viability may be influenced by over a thousand host genes and that those genes might represent potential drug targets against H5N1 virus. Selection of candidate drugs and screening in H5N1-infected A549 cells. By searching the DrugBank database, we identified 146 drug target genes (372 drug candidates) among 1,137 host genes that altered cell viability by more than 20% in the RNAi screen after H5N1 virus infection. We further evaluated the prophylactic and therapeutic effects of 104 commercially available drugs (with 65 validated target genes) by examining their ameliorative effects in a cell viability assay in H5N1-infected A549 cells ( Fig. S3A and B and Table S2) and found 28 drugs that significantly enhanced cell viability with low cytotoxicity ( Fig. 2A and Fig. S3C; Table S3).
Drug candidates that ameliorated ALI and improved survival rates in H5N1infected mice. Among the 28 drugs with which we observed a significant recovery of cell viability following H5N1 infection, 8 were glucocorticoids, which, at low doses, are used in conventional treatments against influenza virus-induced ALI (18,19). Therefore, we focused on the other 20 nonhormonal drug candidates and tested the top 10 nonhormonal drug candidates that enhanced the viability of H5N1-infected A549 cells against ALI in C57BL/6 mice infected with H5N1 ( Fig. S4A; see also Table S3). The We measured the impact of these 2 drugs on major subsets of leukocytes during the response to H5N1 virus infection and found that ifenprodil or flavopiridol significantly decreased the levels of neutrophils, natural killer (NK) cells, and T cells in H5N1 virus-infected mouse lung tissues (Fig. S5). Notably, both ifenprodil treatment and flavopiridol treatment significantly increased the survival rate of H5N1 virus-infected mice (Fig. 2G). The weight loss of mice administered ifenprodil or flavopiridol was recovered at the second week after H5N1 virus infection (Fig. S4E). Taken together, these results indicate that 8 of the top 10 nonhormonal drugs can ameliorate ALI in H5N1-infected mice, with ifenprodil and flavopiridol being the most effective.
Ifenprodil and flavopiridol attenuated H5N1-induced cytokine storm in mice. RNA sequencing (RNA-Seq) analysis of the lung tissue showed that ifenprodil and flavopiridol alleviated cytokine storm in H5N1-infected mice ( Fig. 3A and B; see also Fig. S6A and B). We obtained blood samples from H5N1-infected mice and performed multiplex analyses to measure the levels of 23 cytokines/chemokines in plasma from mice administered ifenprodil or flavopiridol. Overall, ifenprodil treatment and flavopiridol treatment significantly decreased the measured levels of 8 and 12 cytokines/ chemokines, respectively, in the plasma of H5N1-infected mice (Fig. 3C). The other six drugs decreased the mRNA levels of several cytokines and cytokine receptors in mouse blood ( Fig. S6C to H). Our results suggest that these drugs, especially ifenprodil and flavopiridol, can attenuate the excessive inflammation caused by H5N1 virus.
Ifenprodil-targeted NMDA receptor genes and flavopiridol-targeted cyclindependent kinase 4 (CDK4) genes are linked to lung injury. Ifenprodil is a selective N-methyl-D-aspartate (NMDA) receptor antagonist (20) and has been clinically used to treat neuronal injury and neurological disorders induced by overstimulation of the NMDA receptor (21). Our RNAi screening data revealed that knockdown of the NMDA receptor gene (GRIN2B) could alter the viability of H5N1-infected cells (Table S1). We obtained RNA sequencing data from lung tissues of H5N1-infected mice treated with ifenprodil and analyzed biological processes and pathways influenced by ifenprodil compared to vehicle control mice. We found that ifenprodil influenced the immune response and neurophysiological processes in the mouse lung (Fig. 4A). Nine of the top 10 pathways of the ifenprodil treatment group were linked to the immune response (Fig. 4C). Dozens of genes in most of the top 10 pathways were reported to be related to lung diseases as well as to traditional neuropathic indications of ifenprodil ( Fig. 4C; see also Table S4 in the supplemental material). Although previous reports indicated that NMDA receptors are expressed in lung tissues, that NMDA receptor signaling is linked to inflammation, and that overstimulation of the NMDA receptor can trigger lung injury (22,23), our study revealed for the first time that ifenprodil is effective in avian influenza A H5N1 virus infection and lung injury.
Flavopiridol is an inhibitor of cyclin-dependent kinases (CDKs) (including CDK1, CDK2, CDK4, CDK5, CDK6, CDK7, CDK8, and CDK9) and has been investigated for the treatment of several types of cancers, including leukemia, liver cancer, and renal cancer, in clinical phase 2 trials (24)(25)(26). Previous research found that CDK4 inhibition resulted in the improvement of lung injury by blocking leukocyte adhesion and migration and that flavopiridol inhibited avian influenza H7N9 virus replication in A549 cells but that its efficiency in vivo could not be confirmed (27). Our RNAi screening results showed that the viability of H5N1 virus-infected cells was increased due to CDK4 gene knockdown (Table S1). Using RNA sequencing data for biological process and pathway analysis, we found that flavopiridol treatment could alter the immune response as well as cell development and differentiation in H5N1 virus-infected mouse lung tissues (Fig. 4B). Among the top 10 pathways of the flavopiridol treatment group, 5 are related to the immune response of lung disease and 2 are involved in cancer (Fig. 4D). We found that dozens of genes in most of the top 10 pathways were linked to lung diseases as well as to the known antineoplastic function of flavopiridol ( Fig. 4D; see also Table S4). Our study revealed for the first time that flavopiridol is effective in lung injury.

DISCUSSION
In this study, we established an efficient whole-genome RNAi-based platform for repurposing drugs for H5N1 infection in A549 cells. Compared to primary epithelial cells, A549 cells are widely used in large-scale experiments for the study of influenza A virus based on convenience and economic considerations (11,13,15). We identified potential gene targets via a genomewide RNAi screen, in which we searched for drugs that would maximize the viability of H5N1-infected cells. Most genomewide RNAi screens use replication of the moderately pathogenic influenza H1N1 virus as the readout of interference (8)(9)(10)(11)(12)(13)(14)(15). Since the highly pathogenic avian influenza H5N1 virus can cause significant cell death (28) and since cell viability is highly correlated to virus lethality, we designed our genomewide RNAi screening platform to use cell death as the readout in order to identify repurposable drugs for treatment of H5N1 virus infection. Of the 28 effective drugs identified in vitro, 8 were glucocorticoids, which are already used to treat ALI in patients (18,19). We showed that 8 of the top 10 nonhormone drugs effectively ameliorated ALI in H5N1-infected mice. We identified the neurological drug ifenprodil and the anticancer drug flavopiridol as potential novel effective remedies for avian influenza A H5N1 virus-infected lung injury.
The mechanisms by which ifenprodil and flavopiridol ameliorate lung injury and improve the survival of H5N1-infected mice may operate via an influence on the host immune response. Previous studies reported that limiting the infiltration of leukocytes  Table S3. such as NK cells and neutrophils could alleviate influenza A virus-induced ALI (29,30). We found that both ifenprodil and flavopiridol significantly decreased the number of infiltrating cells in H5N1-infected mouse lung tissues. In addition, hypercytokinemia was reported in the blood samples of patients with lethal avian influenza A virus infection, and plasma cytokine and chemokine levels were linked to disease fatal outcomes (31,32). In our study, both ifenprodil and flavopiridol alleviated cytokine storm, as they significantly decreased the RNA and protein levels of some cytokines and chemokines in lung tissues and blood samples from mice with H5N1 virus infection. In addition, we obtained H5N1-infected mouse lung tissues for RNA sequencing and conducted bioinformatic analysis, which showed that the genes influenced by ifenprodil or flavopiridol administration were highly clustered in GO functional terms and pathways related to the immune response. From these collective results, we speculated that the impact of ifenprodil and flavopiridol on the immune response contributes to the amelioration of H5N1-induced ALI and to improvement in mouse survival. However, the efficacy of ifenprodil and flavopiridol for the treatment of H5N1-induced ALI also needs clinical evaluation.
In our study, we evaluated the efficiency in H5N1-infected A549 cells of only 104 commercially available drugs from 372 drug candidates identified using our genomewide RNAi screening method; the efficiencies of the rest of the 268 drugs need to be further tested. Furthermore, we examined the efficiencies of only 10 of the 28 in vitro-effective drugs in an animal model; the rest of the drug candidates also need to be tested in vivo.
Our report provides an economical, quick, and highly effective method (compared with traditional drug development strategies) for identifying novel remedies against avian influenza virus-induced lung injury. This approach could be generalized for identifying other contexts in which drugs can be repurposed. where siRNA H5N1 represents the H5N1-infected siRNA group, siRNA AF represents the AF-treated siRNA group, NC H5N1 represents the H5N1-infected NC group, and NC AF represents the AF-treated NC group. A two-tailed P value of Ͻ0.05 was considered statistically significant. Genes associated with normalized cell viabilities that were increased or decreased by Ͼ10% were taken into the second round of screening. Detailed information about the host genes identified in the genomewide RNAi screen is provided in Table S1 in the supplemental material. Gene enrichment and network analyses. The top hits in the siRNA screen were functionally grouped according to Gene Ontology (GO) terms and functional pathways as clustered by Metacore (Clarivate Analytics) software. The criteria used to identify "disease-related genes" followed the standards  Table S4. of Metacore (Clarivate Analytics) software. Cytoscape (version 3.6.1) with the Enrichment Map application (version 3.1.0) was used for visualization. Interactions among the genes were assessed using the STRING database (https://string-db.org/). Drugs. A total of 1,137 genes were searched as drug targets in the DrugBank database (www .DrugBank.ca/), and 104 drugs were ultimately selected as candidates. All drugs were purchased from Selleck Chemicals (Houston, TX, USA), and detailed information about the drugs is provided in Table S2.

Cells
In vitro experiments. Drugs were added at 1, 10, and 100 M prophylactically (3 h before infection) or therapeutically (3 h after infection) to A549 cells infected with H5N1 influenza virus, and cell viabilities were measured by MTS assay at 48 h after infection. We performed two rounds of screens to confirm drug efficacy. Drug cytotoxicity was measured by assessing the viability of drug-treated A549 cells that were not infected with H5N1. Negative-control cultures were treated with dimethyl sulfoxide (DMSO). Drugs that significantly improved the viability of H5N1-infected A549 cells and that showed low cytotoxicity in the two screens were considered effective for treating H5N1 infection in vitro. Drug information and the effective concentrations are listed in Table S3.
In vivo experiments. Wild-type C57BL/6 mice (6 to 8 weeks old; catalog no. 5653791, RRID MGI: 5653791) were purchased from Vital River (Beijing, China) and divided into two groups (5 mice per group) based on whether drug was administered by intraperitoneal (i.p.) injection or by gavage. A control group was administered vehicle. Mice were intratracheally instilled with live H5N1 virus (10 6 50% tissue culture infectious doses [TCID 50 ]), and they received drug or vehicle at 4 time points: 24 h and 3 h before and 24 h and 48 h after infection. Drug administration methods and concentrations are listed in Table S3. Mice were sacrificed at 3 days after viral infection, and bilateral lungs were collected to assess lung injury and pulmonary edema (33). The animal experiments in this work were approved by the Ethics Committee of the Institute of Basic Medical Sciences, Chinese Academy of Medical Sciences (ACUC-A02-2017-014) and adhered to the Chinese National Guidelines for the Care of Laboratory Animals and the Institutional Animal Care.
Measurement of cytokines and chemokines. Three groups of five C57BL/6 mice (6 to 8 weeks old) were used in this study. Mice were infected with live H5N1 virus (10 6 TCID 50 ) and administered vehicle, ifenprodil (20 mg/kg/body weight, intraperitoneally), or flavopiridol (5 mg/kg, intraperitoneally) at 4 time points: 24 h and 3 h before and 24 h and 48 h after infection. Mouse plasma samples were collected 72 h after infection. The levels of cytokines and chemokines in the plasma were measured with a Bio-Plex mouse cytokine 23-plex panel in a Bio-Plex protein array system (Bio-Rad Laboratories).
Measurement of the survival rate and body weight changes. Wild-type C57BL/6 mice were divided into three groups (9 to 10 mice per group). Mice were administered vehicle, ifenprodil (20 mg/kg, intraperitoneally), or flavopiridol (5 mg/kg, intraperitoneally) at 24 h and 3 h before and 24 h and 48 h after infection (10 6 TCID 50 of H5N1 virus). The survival and body weight changes of the mice in each group were assessed daily for 14 days. The log rank test was used to compare the Kaplan-Meier survival curves.
RNA sequencing and data analysis. Lung tissue and blood from mice treated with drug or vehicle were obtained at 3 days after H5N1 infection. Total RNA was isolated from lung tissue and blood using TRIzol (Invitrogen, CA, USA). High-throughput strand-specific RNA sequencing was performed using an Illumina HiSeq 2500 platform (Berry Genomics, Beijing, China). Following quality control of the RNA sequencing reads by FastQC (version 0.11.2), the reads were mapped to the mouse genome (version mmc10; http://hgdownload.cse.ucsc.edu/downloads.html) with Bowtie2 (version 2.1.0) and TopHat2 (version 2.0.11). We then assembled the transcription units to calculate the number of mapped fragments per kilobase per million (FPKM) and analyzed the differentially expressed genes (DEGs) between samples using Cufflinks (version 2.2.1), Cuffmerge (version 2.2.1), and Cuffdiff (version 2.2.1). The functional pathways of DEGs between animals administered drug or vehicle were analyzed using Metacore.
Statistical analysis. Differences between two groups were assessed for significance using unpaired t tests or the Mann-Whitney U test. Differences among three or more groups were assessed using analysis of variance (ANOVA). The log rank test was used for Kaplan-Meier survival analysis. Statistical analysis was performed using GraphPad Prism 7.0. Differences associated with a two-tailed P value of Ͻ0.05 and a false-discovery-rate value (q value) of Ͻ0.05 were considered statistically significant.