Application of the PHENotype SIMulator for rapid identification of potential candidates in effective COVID-19 drug repurposing

Abstract

The current, rapidly diversifying pandemic has accelerated the need for efficient and effective identification of potential drug candidates for COVID-19. Knowledge on host-immune response to SARS-CoV-2 infection, however, remains limited with few drugs approved to date. Viable strategies and tools are rapidly arising to address this, especially with repurposing of existing drugs offering significant promise. Here we introduce a systems biology tool, the PHENotype SIMulator, which -by leveraging available transcriptomic and proteomic databases-allows modeling of SARS-CoV-2 infection in host cells in silico to i) determine with high sensitivity and specificity (both>96%) the viral effects on cellular host-immune response, resulting in specific cellular SARS-CoV-2 signatures and ii) utilize these cell-specific signatures to identify promising repurposable therapeutics. Powered by this tool, coupled with domain expertise, we identify several potential COVID-19 drugs including methylprednisolone and metformin, and further discern key cellular SARS-CoV-2-affected pathways as potential druggable targets in COVID-19 pathogenesis.

Keywords: 2DG, 2-Deoxy-Glucose; ACE2, Angiotensin-converting enzyme 2; COVID-19; COVID-19, Coronavirus disease 2019; Caco-2, Human colon epithelial carcinoma cell line; Calu-3, Epithelial cell line; Cellular SARS-CoV-2 signatures; Cellular host-immune response; Cellular simulation models; DEGs, Differentially Expressed Genes; DEPs, Differentially expressed proteins; Drug repurposing; HCQ-CQ, (Hydroxy)chloroquine; IFN, Interferon; ISGs, IFN-stimulated genes; MITHrIL, Mirna enrIched paTHway Impact anaLysis; MOI, Multiplicity of infection; MP, Methylprednisolone; NHBE, Normal human bronchial epithelial cells; PHENSIM, PHENotype SIMulator; SARS-CoV-2, Severe acute respiratory syndrome coronavirus 2; Systems biology; TLR, Toll-like Receptor.

Application of the PHENotype SIMulator: By modeling human host-cell infection with a pathogen in silico - in this case SARS-CoV-2 - we can acquire a cell-specific viral signature and formulate multiple drug repurposing hypotheses; (I) logFold Changes (logFCs) of Differentially Expressed Genes (DEGs) arising from transcriptomic genome wide expression analysis of infected vs. baseline uninfected cells are used to represent a virus in the meta-pathway; (II) we run the PHENSIM simulation by upregulating the viral node and collect all perturbation values computed by PHENSIM for pathway endpoints to define a cell-specific pathogen signature. (III) The same process is applied to expression data arising from whole transcriptome-wide expression analysis of drug treated vs. mock-treated cell lines, yielding a cell-specific drug signature. This process is iterated for each drug we wish to test and collected in a database of drug signatures. (IV) Finally, a Pearson correlation analysis between the pathogen and each drug signature is utilized to score repurposing candidates.

Fig. 1

Schematic representation of the PHENSIM Drug repurposing Strategy. Outline for our approach to acquire a cell-specific viral signature in silico and formulate repurposing hypotheses: (a) first, logFold Changes (logFCs) of Differentially Expressed Genes (DEGs) arising from transcriptomic genome wide expression analysis of infected vs. baseline uninfected cells are used to represent a virus in the meta-pathway; (b) then, we run the PHENSIM simulation by upregulating the viral node; (c) therefore, we collect all perturbation values computed by PHENSIM for pathway endpoints to define a cell-specific pathogen signature. (d) The same process is applied to expression data arising from whole transcriptome-wide expression analysis of treated vs. mock-treated cell lines, (e) to perform a PHENSIM simulation of the drug activity, (f) yielding a cell-specific drug signature. (g) Thus, this process is iterated for each drug we wish to test and collected in a database of drug signatures. (h) Finally, a Pearson correlation analysis between the pathogen and each drug signature is used to score repurposing candidates, yielding hypothesis for further laboratory tests. In panels (a) and (d), we report upregulated DEGs in red and downregulated ones in blue. In panels (c), (f), and (g), we report positively perturbated endpoints in green and negatively perturbated endpoints in blue. Finally, in panel (h), negative correlation (reported in green) predicts promising drug candidates that inhibit the pathogen signature and positive correlation (reported in red) suggests exacerbation of the viral signature when introducing the drug.

Fig. 2

In silico PHENSIM prediction of host transcriptional response to SARS-Cov-2.In vitro results from Blanco-Melo et al. (left column; checkered boxes) are compared to in silico PHENSIM predictions (right; solid) for all evaluated respiratory related cells assessed; NHBE, Calu-3, A549 cells at low (0.2) and high (2.0) MOI, ± ACE2 transduction respectively. A) Heatmap depicting the perturbation of a select subset of anti-viral, ISGs and inflammatory genes. B) Heatmaps depicting unbiased analysis of the top-10 upregulated (red) and top-10 downregulated (blue) DEGs from Blanco-Melo et al. (left) with side-by-side PHENSIM predictions (right). For A&B, legend shows denoted perturbations for PHENSIM prediction and Blanco-Melo et al. See legend box for DEG annotation. C) Heatmap depicts whole genome pathway analysis as predicted by PHENSIM for a select set of signaling pathways of interest in all assessed cell types. Pathway selection was based on highlighted pathways affected by SARS-CoV-2 infection. Color gradient depicts the average pathway perturbation as predicted in our PHENSIM in silico experiments. D&E) MITHrIL pathway analysis was used to assess top meta-pathways for D) A549-ACE2 MOI 0.2 (low viral load) and E) A549-ACE2 MOI 2.0 (high viral load), according to impact (circle size) and significance (color-gradient for adjusted p-value) for the top 12 up- (+accumulator) and down-regulated pathways. The accumulator is the accumulation/sum of all perturbations computed for that particular pathway. NHBE; Normal Human Bronchial Epithelial cells, Calu-3; Cultured human airway epithelial cells, A549; Transformed lung alveolar cells, ACE2; angiotensin-converting enzyme, MOI; multiplicity of infection. DEGs; Differentially expressed genes, ISGs; IFN-stimulated genes.

Fig. 3

PHENSIM proteomic pathway analysis in SARS-CoV-2-infected human host cells. PHENSIM pathway analysis of the Caco-2 cell experiment was simulated in silico to reproduce in vitro results presented by Bojkova et al. at the 24 h time-point post SARS-CoV-2 infection A) Schematic representation depicting the experimental design as described by Bojkova et al. in vitro: the human colon epithelial carcinoma cell line, Caco-2 cells, were infected and monitored for 24hrs post SARS-CoV-2 infection. Naturally occurring heavy isotype SILAC labelling was used to quantify translational changes, as this method does not affect cellular behavior allowing for unbiased pathway analysis. Quantitative translation and whole cell proteomics by LC-MS/MS was performed [9]. B&C) Heatmaps depicting a representative subset of the 30 top differentially expressed proteins (FDR<0.05) involved in viral infection after 24hr SARS-CoV-2 infection B) as predicted by PHENSIM in silico (right column, solid squares), compared to expression results as determined by Bojkova et al. (left column, checkered squares) and C) as described by Bojkova et al. (left column, checkered) with side-by-side PHENSIM expression prediction for that protein (right column, solid). D) The heatmap shows the perturbation, as computed by PHENSIM, for a selection signaling pathways described as significant by Bojkova et al. in their analysis. E) The heatmap depicts the Top 50 pathways (25 Up- and 25 Down-regulated) significantly affected at 24h post infection, hence with p-value ≤ 0.05, according to PHENSIM prediction. In this case the Activity Score was taken into account. Color gradient reflects PHENSIM activity; the value of the activity score attributed to each pathway from blue (downregulation) to red (maximum upregulation). Caco-2; the human colon epithelial carcinoma cell line, SILAC; Stable Isotype Labeling by Amino Acids in Cell culture, LC-MS/MS; Liquid chromatography mass spectrometry, DEPs; Differentially expressed proteins, Max; maximum.

Fig. 4

Drug repositioning candidates for COVID-19. We leverage our PHENSIM drug strategy approach to test candidate drugs for potential repurposing for COVID-19 treatment. Once a cell-specific viral signature is defined, it can be exploited to search for possible repositioning candidates by building a drug signature database. A Pearson correlation p (x,y) between the viral and drug signatures gives rise to a correlation score. Drug candidates having a positive effect on ameliorating SARS-CoV-2 infection have a negative correlation score (green) between viral and drug signature, whereas candidate drugs worsening disease correlate positively (red). Here we show distinct candidate drugs having a variable effect depending on the multiplicity of infection (MOI) of virus infection in A459-ACE2 expressing cells in A) low MOI 0.2 and B) high MOI 2.0. This analysis shows the modeling viral load dynamics and discerning what candidate could work best in low vs higher viral load. Resulted top pathways significantly affected by Methylprednisolone treatment are depicted for C) low MOI 0.2 and D) high MOI 2.0. Drug candidates represented here: Methylprednisolone, Metformin (mTOR-inhibitor), (Hydroxy)chloroquine (HCQ-CQ), Acalabrutinib (BTK-inhibitor), Dexamethasone, 2-Deoxy-Glucose (2DG) and Everolimus (mTOR-inhibitor). ACE2; angiotensin-converting enzyme, MOI; multiplicity of infection.

Fig. 5

Methylprednisolone inhibits key inflammatory and viral signaling pathways in host lung and airway cells after SARS-CoV-2 infection. Heatmap depicts the effects of Methylprednisolone in silico in SARS-CoV-2 infection on select signaling pathways of interest as recently identified to be of importance by A) Catanzaro et al. 2020 and B) Draghici et al. 2020. From left to right for each cell-line depicted, column A: pathway analysis results of SARS-CoV-2 infection in vitro as performed using the MITHrIL algorithm; column B: PHENSIM results of SARS-CoV-2 infection in silico; column C: PHENSIM simulation results of Methylprednisolone on SARS-CoV-2 infected cells in silico. Color gradient depicts the average pathway perturbation as predicted in our PHENSIM in silico experiments for column B&C. NHBE; Normal Human Bronchial Epithelial cells, Calu-3; Cultured human airway epithelial cells, A549; Transformed lung alveolar cells, ACE2; angiotensin-converting enzyme, MOI; multiplicity of infection.