A preprint by Muus et al. reports a large collaborative meta-analysis of more than hundred single-cell and single-nucleus RNA-seq studies to characterize key proteins involved in host-pathogen interactions of the SARS-CoV-2 virus that causes the COVID-19 pandemic. Here are my learning notes.

Major findings

The key proteins in focus include angiotensin-converting enzyme 2 (ACE2, GeneID 59272), the main mediator of entry into human cells by the SARS-Cov-2 virus, and two proteases, transmembrane protease serine 2 (TMPRSS2, GeneID 7113), and cathepsin L (CTSL, GeneID 1514), both of which are proteases, enzymes that break down other proteins, and both of which are believed to prime and assist the cellular entry of the virus.

The expression patterns of ACE2 mRNA and co-expression patterns between ACE2 and the proteases show concordance with tissues and organs associated with pathology induced by SARS-CoV-2, including gut enterocytes, corneal epithelial cells (the outermost layer of the cornea in the eye, see LifeMap), cardiomyocytes (heart muscle cells), heart pericytes, olfactory sustentacular cells (a type of supporting cells in the olfactory epithelium, serving as metabolic and physical support for the olfactory epithelium), and renal epithelial cells. The mRNA expression profiles of three genes are positively associated with increasing age, male gender, and smoking, which are epidemiologically linked to COVID-19 susceptibility and outcomes.

Another interesting finding is that some cells that show ACE2+TMPRSS2+ expression patterns in nasal, gut, and lung also express genes associated with viral entry, immune function, and epithelial-macrophage cross-talk. Among them are interleukin 6 (IL6), its receptor and co-receptor, IL1R, TNF response pathways, and complement genes. The cell type specificity in the lung and airways and smoking effects were conserved in mice.

Methodologies

Here are some quick notes about the methodologies. They are not comprehensive and may even be inaccurate. Please look up the original paper if needed.

  • The authors searched Gene Expression Omnibus, and compiled expression profiles of more than two million cells from 469 samples from 64 distinct datasets. They also included unpublished studies.
  • They used two levels of annotation, and used a PyTorch impelementation of the Harmony algorithm (Korsunsky et al., Nature Methods, 2019), available at GitHub lilab-bcb/harmoney-pytorch, to generate cell clusters.
  • Single-cell and single-nucleus RNA-sequencing data analysis with done with Scanpy.
  • Co-expression of TMPRSS2 and CTSL with ACE2 was assessed with a logistic (e.g. binarized) linear mixed effect model, with ACE2 mRNA abundance as an independent variable and individuals as a random effect.
  • Association with gender, age, and smoking status was assessed with a generalized effect model, using statsmodels and a Wald test from Diffxpy.
  • Single-cell ATAC-sequencing data was analysed with timoast/signac. I introduced ATAC-sequencing in a previous blog post as an experimental approach to profile accessible chromatin regions. In this study, scATAC-sequencing data were used to support immune-like molecular phentoypes of epithelial cells that express ACE2.
  • Random forest models were built with sklearn to distinguish double-positive from double-negative cells, and genes that build the trees are used as distinguishing features between the two types of cells.
  • Networks were illustrated for visual purposes with networkx.
  • The code is available at terra.bio.
  • The data and interactive visualizations are available at The Human Cell Atlas.

Conclusions and reflections

Overall, the study suggests that cell type-specific expression of genes mediating SARS-CoV-2 viral entry is consistent with disease pathology. Their expression profiles and potentially other genes involved in immune response may explain some aspects of COVID-19 epidemiology and clinical course.

Currently, we have no data of mRNA-protein abundance correlation of these genes in the human body, in particular in patients affected by the virus. In addition, the meta-analysis cannot provide an estimate of the variance of clinical symptoms explained by the expression profile. Nevertheless, it is interesting to observe the concordance between cell-type specific gene expression and anatomical sites of disease manifest.

Furthermore, it can be envisioned that such analysis can empower further, similar analysis that spans multiple tissues and organs which focus on genes that interact to perform biological processes.

I thank Klas Hatje who discovered this preprint and shared it with me.