This study used the “host response signature network” identified in the human cells infected with SARS-CoV-2 in our previous study (5). Briefly, transcriptome data of SARS-CoV-2 infected normal human bronchial epithelial (NHBE) cells and controlled NHBE cells were used to identify the differentially expressed host genes (DEHGs) (19). A network was created by integrating DEHGs with human protein-protein interaction data. Furthermore, analysis of the network revealed an important sub-network of highly inter-connected 31 proteins (IL6, TNF, CXCL8, CXCL3, CXCL5, IRF9, SAA1, OAS3, CSF2, IFI6, OAS2, CSF3, IRF7, ICAM1, CXCL2, MX1, OAS1, MMP9, IL1A, IL1B, C3, TLR2, IFI27, CXCL6, CXCL1, CCL20, XAF1, IFI44L, MX2, BST2, IFITM1) predominantly involved in cytokine storm and under the regulation of transcription factors STAT1, STAT2, STAT3, POU2F2, and NFkB1. The sub-network of 31 proteins and their five master regulators are called as “host response signature network”. In order to understand the biological pathways associated with the “host response signature network”, KEGG (Kyoto Encyclopedia of Genes and Genomes) pathway enrichment analysis was performed using DAVID 6.8 (https://david.ncifcrf.gov/home.jsp).

The “host response signature network” was analyzed to find potential drug molecules using the “Gene Association” module of Network Data Exchange (NDEx version 2.4.5) (https://ndexbio.org/) (20, 21). NDEx is an open-source platform for biological network analysis and knowledge discovery. The output of “Gene Association” analysis gives networks of query gene/protein and its direct or indirect interaction with chemical and drug molecules. All output networks of drug-target interaction were merged to make a non-redundant network and visualized with Cytoscape software version 3.7. Subsequently, we analyzed the potential drug molecules targeting the “host response signature network”. For further study, we selected a drug molecule, andrographolide, naturally present in a medicinal plant used to treat common cold and inflammation (22).

The three-dimensional coordinates of andrographolide were retrieved from the PubChem database with PubChem CID, 5318517. The crystal structure of human NFkB1, p50 subunit, with bound native DNA was selected and retrieved from the PDB database with PDB Id, 2V2T. The crystal structure of human TNF as a dimer with bound native inhibitor was selected and retrieved from the PDB database with PDB Id, 5MU8.

The drug‐likeness and pharmacokinetic properties, including absorption, distribution, metabolism, excretion, and toxicity, were analyzed using the online tool “pkCSM‐pharmacokinetics” (http://biosig.unimelb.edu.au/pkcsm/). This machine learning-based method uses the various graph-based signature of chemical compounds having sets of distance patterns between atoms (23).

It was reported that andrographolide is a covalent inhibitor of NFkB1 and makes a covalent bond with Cys-62 of NFkB1 (24, 25). Autodock Vina (26) was used for covalent docking of andrographolide to NFkB1. To perform covalent docking, the reported atom of the andrographolide was linked to the sulfide atom of Cys-62 through a covalent bond (25). The adduct Cys-andrographolide was made flexible for rotation through various bonds using Autodock Tools (27). Then the docking of a very small molecule (e.g., water molecule) was performed, leading to the relaxed conformation of the flexible adduct Cys-andrographolide. Finally, this relaxed conformation of the adduct Cys-andrographolide was reported as the final covalent docking of andrographolide. Autodock Tools was used to prepare the protein, the ligand, and the grid box required for docking by Autodock Vina.

For normal molecular docking, we used Dock version 6.5 (28). For TNF, the homodimer (Chains A and B) was used for molecular docking, and the native inhibitor bound within the binding site formed by both monomers (Chains A and B) was used as a probe for the TNF homodimer binding site. The protein and ligand preparation required for molecular docking and visualization at different stages of docking was performed using Chimera version 1.6.2 (29). The illustrations for binding poses were generated using Pymol version 2.4.0 (Schrödinger, LLC) and the protein-ligand interaction plots were prepared using Ligplot+ version 2.1 (30). The binding energy and dissociation constant scores were predicted using XScore version 1.2.11 (31).

The “host response signature network” was analyzed with the KEGG pathway, which identified several significantly associated biological pathways. We found that TNF signaling pathway is significantly associated with IL6, CSF2, CCL20, IL1B, CXCL1, CXCL3, TNF, CXCL2, MMP9, NFkB1, and ICAM1 of the “host response signature network” and responsible for the induction of cytokine storm ( Figure 1 ). The complete list of all KEGG pathways is provided in the Supplementary Table S1 .

Analysis of potential drug molecules for “host response signature network” with the “Gene Association” module of online tool NDEx returns a list of drug-target interactions networks ( Table 1 ). The list of drug-target interactions from different networks was merged to create a non-redundant drug-target network ( Figure 2 ). The detailed information of drug molecules is provided in the Supplementary Table S2 . The data was further investigated to find a potential drug for better management of the cytokine storm in severe COVID-19 patients. We have selected andrographolide for further study based upon the following criteria from the list of repurposed drugs.

(a) The drug must target/suppress the highest number of proteins of the “host response signature network”; (b) targeted proteins associate with the same biological pathway; (c) targeted proteins play a crucial role in the cytokine storm; and (d) a medicinal plant could be a source of the drug.

We found that andrographolide can target four different proteins IL1B, NFkB1, TNF, and IL6 ( Table S2A ) of the TNF signaling pathway ( Figure 1 ), which might result in blockage of the cytokine storm in COVID-19. Andrographolide is a labdane diterpenoid isolated from Andrographis paniculata, also known as the “king of bitters”, a medicinal plant that mainly grows in Asian countries ( Figure 3 ). Andrographis paniculata traditionally used in Unani, Ayurvedic, and Chinese herbal medicines with a broad range of therapeutic applications, including treating a common cold, upper respiratory tract infections, and inflammation (22).

As shown in Table 2 , the andrographolide was having a low molecular weight of 350.46 (< 500) and the lipophilicity (LogP) value of 1.96 (<5), H-bond donors 3 (<5) and acceptors 5 (<10), and three rotatable bonds (<10). The values for all five conditions (Lipinski’s Rule of Five) for the andrographolide were well within the desired range of a drug molecule. While most of the pharmacokinetic properties, including absorption, distribution, metabolism, excretion, and toxicity, were acceptable for andrographolide ( Table 3 ). Therefore, this suggests that andrographolide is a potentially safe drug candidate for use in humans.

The PDB structures of IL1B and IL6 are available as a complex with their receptors (for example, PDB Ids: 3O4O and 1P9M, respectively) and with the antibody (for example, PDB Ids: 4G6J and 4CNI, respectively). The information for the binding site of ILIB and IL6 for a chemical compound was not obvious. So, we proceeded with the docking of andrographolide to NFkB1 and TNF only.

The experimental laboratories showed that andrographolide reacted with the sulfhydryl group (-SH) of Cys-62 of NFkB1 and formed a covalent bond (24, 25). One of the studies mentioned above further proceed to a normal docking of andrographolide to NFkB1, and they reported Cys-62 and Arg-57 as the interacting residues forming hydrogen bonds (25). We tried normal docking considering Cys-62 as a clue for the binding site and failed to see a binding site cavity close to Cys-62 residue where this compound may fit. Therefore, we proceeded with the covalent docking of andrographolide to NFkB1. The docking results showed that the andrographolide formed a covalent bond and a hydrogen bond (2.86 Å) with the sulfide atom of Cys-62 ( Figure 4 ). The Cys-62 was also involved in 13 non-bonding contacts with andrographolide ( Table 4 ). Another interacting residue Glu-63 was involved in another 13 non-bonding contacts. Thus, the andrographolide formed an adduct with Cys-62 and became part of the protein. In addition, it formed a hydrogen bond and 26 non-bonding interactions with the protein ( Figure 4 and Table 4 ). Thus, our study suggests andrographolide could block the DNA binding site, interfering with the binding of DNA, and thus the transcription factor NFkB1 would not be able to carry out its function. This is in agreement with previous findings, which revealed that andrographolide binds with NFkB1, which blocks the binding of NFkB1 to DNA and prevents transcriptional activity (24, 25).

The andrographolide docked well within the binding site formed by two monomeric units of TNF ( Figure 5A ) and stabilized by 20 non-bonded contacts and one hydrogen bond through nine interacting residues ( Figure 5B and Table 5 ). The absolute values of binding scores, including dock score (-32.67), binding energy (-7.52 Kcal/mol), and dissociation constant measure pK_d (5.51), were also reasonably high required for the stable protein-ligand complex. The nine interacting residues include four residues from chain A (Tyr-59, Tyr-119, Leu-120, and Gly-121) and five residues from chain B (Leu-94, Tyr-119, Leu-120, Gly-121, and Gly-122). Interestingly, the three identical residues Tyr-119, Leu-120, and Gly-121 from both the monomers, played a role in binding. Of nine interacting residues, Gly-121(B) formed hydrogen bonding interaction with the andrographolide with bond length 3.06 Å and also formed eight (maximum) non-bonded contacts. Therefore, the Gly-121(B) is proposed as the key residue playing a role in binding. While comparing the binding of andrographolide to that of the native inhibitor JNJ525 ( Figures 5B, C ), it was found that the andrographolide was also binding to the same site and sharing common six interacting residues Tyr-59 (A), Tyr-119 (A), Gly-121 (A), Tyr-119 (B), Gly-121 (B), and Gly-122 (B). Thus, the andrographolide was binding in the site where the TNF inhibitor is binding and inhibiting TNF activity.

Previous studies showed that andrographolide has a potent inhibitor of wide verities of viruses, including influenza virus (37, 38), hepatitis C virus (39), herpes simplex virus type 1 (40), Chikungunya virus (41). It also has the potential to inhibit the proliferation of lung carcinoma (42) and nasopharyngeal carcinoma (43), and tumor metastasis (44). Another report demonstrated that mice treated with andrographolide reduce the inflammation and fibrosis of damaged liver (45). It also prevents liver neutrophil infiltration and reduced TNF and COX-2 signaling (45). A recent study with the molecular docking method suggested andrographolide as a potential inhibitor of SARS-CoV-2 main protease M^pro (47). Another computational-based study found that andrographolide and its derivative, 14-deoxy-11,12-didehydroandrographolide, have a high binding affinity with three target proteins of SARS-CoV-2, i.e., main protease, papain-like protease (PL^pro), and spike protein (48). In addition, the same study used a DIGEP-Pred tool and predicted that both compounds induce the level of several proteins involved in regulating immune systems, including the NFkB signaling pathway (48).

The NFkB1 is a family of transcription factors, which regulate the expression of various genes of pro-inflammatory cytokines, including IL6 and IL8 (CXCL8), mainly responsible for cytokine storm in COVID-19 ( Figure 1 ) (5, 49). NFkB1 inhibitory protein (IkB) binds with NFkB1 dimer in an unstimulated cell, which prevents movement of NFkB1 from the cytoplasm to the nucleus. In response to SARS-CoV-2 infection, the TNF signaling pathway is activated, which causes degradation of IkB, resulting in the release and translocation of NFkB1 to the nucleus for transcription of various cytokine genes (5, 49). Our study found that andrographolide bound well with TNF and NFkB1 and provided structural insights into their binding. This binding of andrographolide blocks the TNF signaling pathways at these two critical steps, thus prevents the expression of various cytokine genes responsible for the influx of cytokine in COVID-19.

In addition, our study identified Dilmapimod as potential compounds to reduce the cytokine storm through targeting/suppressing IL1B, TNF, and IL6 ( Table S2A ). Activation of the p38 MAPK signaling pathway is associated with lung inflammation in asthma and chronic obstructive pulmonary disease (COPD) (50). The p38 MAPK regulated NFkB dependent transcription of multiple inflammatory cytokines (51). Clinical studies showed that Dilmapimod, a novel inhibitor of p38 MAPK, decreases the TNF and IL1B level in the blood of COPD patients (52, 53). Another study also suggested the potential use of Dilmapimod for COVID-19 management (54). Furthermore, our analysis identified that Prednisolone, a corticosteroid, has the potential to suppress the level of IL6 ( Table S2A ); thus, supported previous work indicating that Prednisolone could be useful in treating severe COVID-19 (55).

Previous studies showed the anti-inflammatory and anticancer effects of Dehydroxymethylepoxyquinomicin (DHMEQ), a new inhibitor of NFkB (56, 57). Another study found that DHMEQ reduces allergic airway inflammation and cytokines in a mice model of asthma (58). An investigation conducted in Jurkat T-lymphoblastic leukemia cells found that DHMEQ prevents the TNF-α-induced nuclear translocation of NFkB (46). These findings also raised the potential use of DHMEQ in inhibiting cytokine storms and thus need to be studied in more detail in treating COVID-19.

Our study used a rational computational approach to identify the potential drug molecule for COVID-19. Though it is associated with the following limitations: (i) we used the “host response signature network” identified in the NHBE cells infected with SARS-CoV-2 (5). The work included a limited sample size of transcriptomic data with three infected and three controlled groups (19). (ii) This study lacks in vitro drug screening data on human lung cells infected with SARS-CoV-2. Therefore, it is difficult to draw the effect of andrographolide on the “host response signature network” in SRAS-CoV-2 infected cells. (iii) Furthermore, our study lacks the experimental evaluation of the dose-dependent effect of the drug and its associated risk.