A Path-Based Analysis of Infected Cell Line and COVID-19 Patient Transcriptome Reveals Novel Potential Targets and Drugs Against SARS-CoV-2

Abstract

Most transcriptomic studies of SARS-CoV-2 infection have focused on differentially expressed genes, which do not necessarily reveal the genes mediating the transcriptomic changes. In contrast, exploiting curated biological network, our PathExt tool identifies central genes from the differentially active paths mediating global transcriptomic response. Here we apply PathExt to multiple cell line infection models of SARS-CoV-2 and other viruses, as well as to COVID-19 patient-derived PBMCs. The central genes mediating SARS-CoV-2 response in cell lines were uniquely enriched for ATP metabolic process, G1/S transition, leukocyte activation and migration. In contrast, PBMC response reveals dysregulated cell-cycle processes. In PBMC, the most frequently central genes are associated with COVID-19 severity. Importantly, relative to differential genes, PathExt-identified genes show greater concordance with several benchmark anti-COVID-19 target gene sets. We propose six novel anti-SARS-CoV-2 targets ADCY2, ADSL, OCRL, TIAM1, PBK, and BUB1, and potential drugs targeting these genes, such as Bemcentinib, Phthalocyanine, and Conivaptan.

Keywords: DEGs (Differentially Expressed Genes); PBMCs (Peripheral Blood Mononuclear Cells); SARS-C0V-2; cell lines; network analysis; transcriptome.

Figure 1

Study Workflow. (A) Our PathExt tool accepts as input a curated gene network and gene expression data, to output two weighted sub-networks - an activated and a repressed TopNet comprising activated and repressed paths respectively, and a list of central genes in each TopNet. Activated genes are shown here in shades of red, and repressed genes are in shades of blue. PathExt integrates the inputs such that edges connecting genes with substantial change in expression are preferentially traversed by a shortest paths algorithm (Methods), illustrated here using wider arrows. Shortest paths which are statistically significant (permutation based) now represent differentially active (or repressed) paths and make up the TopNets in which PathExt identifies central genes based on ripple centrality. (B) We apply PathExt to analyze RNa-seq data from SARS-CoV-2 infection in both cell lines and patient PBMC data. PathExt outputs from the cell line data are used to compare cross-cell-line variation in SC2 infection response, and within-cell-line variation in response to other viruses. In patient PBMC data, we identify associations between PathExt results and demographics. We then validate all the results against benchmarks and use the TopNets and central genes to propose novel drug targets, as well as novel drugs for known targets.

Figure 2

SC2 infection across cell lines. We analyzed the transcriptional response across various cell lines post SC2 infection. We obtained the top100 central genes from each cell line for both TopNets (activated & repressed) and compared the gene commonality across cell lines as shown in upset plots for activated TopNets (A) and repressed TopNets (B). Next, we performed gene enrichment analysis and compared the top 10 parent GO terms enriched in various cell lines for activated TopNets (C) and repressed TopNets (D). Here we show only those GO terms that were significant in at least 2 cell lines. Complete list of enriched GO terms is provided in the supplementary tables. Heatmap is created by converting the FDR corrected p-value of each GO terms to -log10 scale. Significance of the terms is shown in various color ranges (0-1.3, 1.3-2.0, 2-5 and >5-30). All the non-significant processes are shown in black (value <1.3).

Figure 3

SC2 infection comparison with other viruses. We analyzed the transcriptional response to different viruses including SC2 in various cell lines. We obtained the top 100 central genes from each cell line for activated & repressed TopNets and compared the gene commonality across various viruses in different cell lines as shown in gene enrichment plot (Observed/Expected overlap) for activated TopNets (A) and repressed TopNets (B). Semantic similarity among the enriched biological processes observed in different viruses activated and repressed networks (C, D respectively) shows higher similarity.

Figure 4

Comparison of PathExt central genes with DEGs. We compare the PathExt result with the results obtained using traditional DEG approach. (A) LogFC comparison of the top 100 genes between PathExt and DEGs obtained from the SC2-infected Calu3 cell line. Differential expression is estimated in infected relative to uninfected cells. (B, C) Venn diagram shows the gene overlap among the top100 central genes & DEGs present in activated (B) network and repressed (C) network across different cell lines. No repressed TopNet was seen in NHBE. (D) Semantic similarity among the enriched PathExt and DEGs biological processes for activated TopNets.

Figure 5

Functions enriched among central TopNet genes in patient PBMCs. Most frequent top 100 central genes were obtained from the activated and repressed networks across patient PBMC data. Enriched biological processes were obtained by performing Gene Ontology study followed by parent child relationship, shown in the form of circular visualization plot for activated TopNet (A) and repressed TopNet (B).

Figure 6

Demographic features analysis. Mann Whitney Test was performed to check statistical significance between top 100 central genes and various demographic features (age, sex and severity). In case of PathExt identified top genes, “Severity” was found to be the only statistically group among activated TopNet genes (A). However, no group was statistically significant in case of repressed TopNet (B).

Figure 7

Overlap of PathExt-identified genes and DEGs with previously published datasets. (A) the overlap of the PathExt identified top 100 activated & repressed genes of various cell lines and PBMC datasets with various previously published drug validation datasets (CRISPR-Cas, Host-Virus PPI networks, Drug-target studies). (B) the overlap of the top 100 upregulated & downregulated DEGs of various cell lines and PBMC datasets with the same drug validation datasets.

Figure 8

Drug-target association. Based on our virtual screening study, we identified, Bemcentinib as a potential inhibitor for PBK and ADSL. Bemcentinib is a well-known drug against AXL, and we saw that this gene (AXL) is connected to our proposed target PBK (A) and ADSL (B) suggesting that the drug effect may be mediated by multiple genes within a closely linked gene module.