Abstract
As part of the virtual BioHackathon 2020, we formed a working group that focused on the analysis of gene expression in the context of COVID-19. More specifically, we performed transcriptome analyses on published datasets in order to better understand the interaction between the human host and the SARS-CoV-2 virus. The ideas proposed during this hackathon were divided into five projects. Projects 1 and 2 aimed to identify human genes that are important in the process of viral infection of human cells. Projects 3 and 4 aimed to take the candidate genes identified in projects 1 and 2, as well as by independent studies, and relate them to clinical information and to possible therapeutic interventions. Finally, Project 5 aimed to package and containerize software and workflows used and generated here in a reusable manner, ultimately providing scalable and reproducible workflows.
Introduction
In the early stages of the COVID-19 pandemic, there was an urgent need to understand how SARS-CoV-2 interacts with human cells at a molecular level. While many datasets were being rapidly produced, there was a gap in integrated analysis that compared SARS-CoV-2 to other respiratory viruses (like Influenza A and MERS) and linked these findings to clinical risk factors and potential drugs. Furthermore, the "reproducibility crisis" in bioinformatics meant that many early findings were difficult for other researchers to verify or extend. This paper presents a comprehensive, multi-pronged bioinformatics analysis conducted during the virtual BioHackathon 2020 to address these challenges by providing a unified analysis framework and open-source, reproducible pipelines.
Methods
The research was structured into five distinct projects utilizing two primary analytical pipelines to ensure both breadth and depth of analysis.
- Pipeline 1 (Global Analysis): Focused on differential gene expression (DGE) using tools like STAR for alignment, StringTie for quantification, and a consensus approach using DESeq2, edgeR, and limma-voom to identify high-confidence differentially expressed genes.
- Pipeline 2 (Comparative Analysis): Used the ARMOR workflow (based on Salmon) to compare SARS-CoV-2 responses against other viruses (IAV, RSV, MERS-CoV, SARS-CoV-1).
- Host-Virus Interaction Modeling: The team scanned the SARS-CoV-2 genome for binding motifs of human RNA-binding proteins (RBPs) using Position Weight Matrices (PWMs) and statistical enrichment tests.
- Reproducibility: All workflows were implemented using Nextflow and Docker, ensuring that the entire analysis can be re-run in a containerized environment.
Datasets
The study utilized several publicly available transcriptomic and proteomic datasets:
- RNA-Seq: GSE147507 (primary human airway epithelial cells - NHBE and A549), GSE122876 (MERS-CoV), GSE56192 (SARS-CoV-1, MERS-CoV), and GSE57148 (COPD clinical samples).
- Proteomics: PXD107710 (SARS-CoV-2 infected CaCo-2 cells).
- Viral Genomes: Reference sequences for SARS-CoV-2 (Wuhan-Hu-1), SARS-CoV-1, and the bat coronavirus RaTG13.
Key Results and Insights
Transcriptional Signatures and Data Structure
Principal Component Analysis (PCA) of the transformed counts revealed clear clustering based on infection status and cell line, validating the robustness of the differential expression approach. The analysis identified 213 upregulated and 87 downregulated genes in primary lung cells (NHBE), with S100A7 emerging as the top upregulated gene (log2FC of 8.34).
Host-Virus Interactions
SARS-CoV-2 elicits a specific response distinct from other viruses, notably the upregulation of "alarmins" like S100A7 and S100A8, and proinflammatory cytokines like IL32 and IL36G. Furthermore, the SARS-CoV-2 genome contains significantly enriched binding sites for specific human RBPs (e.g., HNRNPL, FUS, PABPC1). This suggests the virus might "sponge" these proteins, potentially disrupting normal cellular mRNA processing.
Clinical and Pharmacological Links
- COPD Overlap: Patients with Chronic Obstructive Pulmonary Disease (COPD) share several activated pathways with SARS-CoV-2 infected cells, such as innate immune signaling, explaining their higher clinical risk.
- HLA Susceptibility: In silico modeling identified HLA-B*46:01 as having the lowest predicted affinity for presenting SARS-CoV-2 peptides, while HLA-B*15:03 showed the greatest capacity.
- Pharmacological Leads: Gene-compound association analysis identified several significant candidates, including dexamethasone (which later became a standard of care), and cancer drugs like taxol and patupilone.
Limitations and Future Work
The authors acknowledge that the response in immortalized A549 cells was much milder than in primary NHBE cells, suggesting that cell line choice significantly impacts the relevance of findings. Additionally, the predicted interactions between viral RNA and human RBPs require experimental validation (e.g., through CLIP-seq). Future work should incorporate a broader range of tissues and single-cell RNA-seq data as they become available, alongside expanding pipelines to analyze transposable elements and complex regulatory networks.
Summary updated