CHICAGO – Bioinformaticians and virologists from Icahn School of Medicine at Mount Sinai in New York have developed and validated a computational method to identify drugs that could be repurposed to combat COVID-19. Unlike other repurposing research that has looked at post-infection treatment, this effort focused on restoring the transcriptional response triggered by SARS-CoV-2 infection in hopes of inhibiting viral uptake in the first place.
In a preprint paper posted to BioRxiv, the researchers explored viral sequences using PCR analysis, RNA sequencing, and bioinformatics. They were able to identify four compounds that were able to block replication of the novel coronavirus, namely amlodipine, loperamide, terfenadine, and berbamine. They then validated these findings in multiple assays using primate Vero cells infected with SARS-CoV-2, A549 cells, and in human organoids.
"These compounds were found to potently reduce viral load despite having no impact on viral entry or modulation of the host antiviral response in the absence of virus," according to the paper.
"In the absence of a pan-specific coronavirus drug, or a SARS-CoV-2 vaccine, the next most useful tool would be the effective repurposing of FDA-approved drugs," the preprint said. "There is underutilized opportunity to understand the global molecular changes these drugs induce by using available sequencing data from diverse models and cell types."
The researchers, led by Avi Ma'ayan, director of the Mount Sinai Center for Bioinformatics and principal investigator with the academic health system's LymeMIND team, extracted samples from infected cells and then sequenced the RNA, finding that inhibition of SARS-CoV-2 uptake happened on the cholesterol biosynthesis pathway.
This team were looking for compounds that might prevent infection.
"You have a bunch of drugs that are blocking the virus in cell culture," Ma'ayan said of other research into other potential COVID-19 treatments. "But this particular paper is showing a lot of details about why and which drug and … is beginning to understand the molecular mechanism."
"We looked for those drugs that are really matching that signature. There are more drugs that fall into that cloud," Ma'ayan said.
The Mount Sinai researchers used the LINCS L1000 assay, a collection of gene expression profiles from the US National Institutes of Health's Library of Integrated Network-based Cellular Signatures (LINCS) database. The Icahn School of Medicine at Mount Sinai hosts the data coordination and integration center for LINCS, and has been backed by the NIH Common Fund since 2014 to develop computational methods to process LINCS data.
These researchers studied expression patterns of infected cells to identify a region of gene expression that they said was both unique to viral infections and inversely proportional to cell transcripts in response to compounds indexed in the LINCS database. "This resource serves as a promising computational method to find compounds which may counter or mimic the gene expression changes induced by a given perturbation," according to the prepress article.
For this, they used a visualization app they developed called the L1000 Fireworks Display (L1000FWD), which Ma'ayan and colleagues described in a 2019 article in Bioinformatics. In the L1000FWD expression space, genes downregulated by cells infected with the novel coronavirus are upregulated by certain drugs and small molecules profiled on the L1000 platform.
LINCS had previously been applied to identify drugs that attenuate the Ebola virus. With SARS-CoV-2, the Mount Sinai team was able to spot transcriptional irregularities by comparing changes in gene expression before and after infection or drug treatment.
In this new work, the Mount Sinai team studied 50 genes that were downregulated by the virus or 50 upregulated by certain drugs. They also looked at the 100 genes most commonly coexpressed by ACE2, known to be the receptor of SARS-CoV-2. This methodology also led the researchers to quercetin, an effector of these ACE2-coexpressed genes.
The research on ACE2 expression was more of a negative control, according to Ma'ayan. "We did look for drugs that can reduce the level of the expression of the receptor and we found a drug that had a high score, but it actually made things worse," he said. "When you test it with the virus, you actually get more viral replication and more virus when you use that drug."
Ma'ayan's complex, multifaceted study included several bioinformatics analyses. The team aligned RNA-Seq reads to version hg19 of the human reference genome on Illumina's BaseSpace informatics platform. They took the 2,000 gene counts with the highest variance and normalized them with Z-scores, then performed principal component and differential expression analysis on the normalized values.
Next, differentially expressed genes were analyzed with L1000FWD and Ma'ayan's own enrichment analysis tool, called Enrichr, to identify the top 50 upregulated or downregulated genes that would serve as the gene signature for each drug analyzed.
Earlier this year, the Ma'ayan Lab produced a publicly available informatics resource called the COVID-19 Gene Set and Drug Library. This research relies on several self-developed tools, but the work was not fully automated.
Manual examination of drugs revealed that terfenadine, loperamide, berbamine, trifluoperazine, amlodipine, RS-504393, and chlorpromazine regularly targeted this expression space. Notably, the Mount Sinai scientists said, recent research from Institut Pasteur Korea found that loperamide, berbamine, and trifluoperazine inhibit SARS-CoV-2 uptake kidney cells of African green monkeys.
The antidiarrheal loperamide is widely availableover the counter. Terfenadrine is an antihistamine, sold as Seldane in the US, that was pulled from the market in the 1990s after it was linked to cardiac arrhythmia.
While SARS-CoV-2 appears to prevent antiviral response by masking aberrant RNA, replication of the virus still produces a unique transcriptional footprint, they said, citing a May paper in Cell from Daniel Blanco-Melo, a postdoctoral researcher in Mount Sinai's tenOever Laboratory.
The Ma'ayan work builds on this by attempting to identify drugs that might invert transcriptional signatures to inhibit replication. The tenOever Lab assisted on this new experiment, and Benjamin tenOever, director of Mount Sinai's Virus Engineering Center for Therapeutics and Research (VECToR), is listed as an author on the preprint.
The findings, or "predictions," as Ma'ayan called them, were validated with multiple assays and multiple cell lines. Ma'ayan's lab built the computational model, but tenOever's lab tested the hypotheses about drugs that the model predicted. Ma'ayan said that some predictions were based on earlier work by tenOever's team.
According to Ma'ayan, seven of the eight drugs initially tested completely blocked transmission in monkey Vero cells. Some also worked in human cells.
Ma'ayan said that he communicated the predictions to tenOever in March, when the New York area was temporarily becoming the global epicenter for the COVID-19 pandemic.
The prepress paper only mentioned the genetic signatures from the Cell paper, but Ma'ayan said that he and his colleagues looked at signatures from other viruses and compounds, including the controversial hydroxycholoroquine, in an effort to determine if that might inhibit SARS-CoV-2 replication.
Ma'ayan said that hydroxychloroquine does work similarly to the drugs his Mount Sinai tested, but requires a much higher concentration than the others. It also has more side effects, so he said that loperamide and amlodipine are likely to be more effective than hydroxychloroquine.
After the prepress article was posted, the researchers found several other studies published on the Gene Expression Omnibus that have data from cells infected with the coronavirus, so they now are trying to see if there is a consensus on the drug and pathway predictions. "That we think is critical. We are creating a more comprehensive across-lab analysis," Ma'ayan said.
They are using machine learning to look for common themes among published "hits," according to Ma'ayan. "A lot of people are publishing drugs that are working in cells, but we're trying to synthesize all that information and try to make sense of it, and also explain the mechanisms behind those observations," he said.
Ma'Ayan said that his goal is to get these compounds into human clinical trials, whether at Mount Sinai or elsewhere. He also wants to study combination therapies.
Critical assessment
Liudmila Mainzer, a technical program manager of NCSA Genomics in the National Center for Supercomputing Applications at the University of Illinois at Urbana-Champaign, said that she found this paper "quite intriguing" and unlike anything she had seen before. No NCSA employees were involved in this study.
"Their objective was not the standard objective that would have been behind normal RNA-seq analysis," namely which genes are significantly upregulated or downregulated in response to a signal, Mainzer said. Instead, the Mount Sinai team sought drugs that can counteract the up- or downregulation based on patterns of gene expression specifically related to COVID-19, an exercise that required looking at multiple genes.
Then they tried to match gene expression to opposite patterns of up- and down-regulation in response to drugs. "They're matching pattern to pattern, not gene to drug, and they're matching pattern to an inverse pattern. I thought that was very neat," Mainzer said.
Mainzer did note that hg19 is an older version of the human reference genome, though she said that it is not critical for this exercise because the reads to not need to be exactly aligned to figure out whether a gene is up- or downregulated.
According to Mainzer, the Ma'ayan method of differential expression analysis is nonstandard in RNA-seq, though it fits here because it was developed specifically to measure drug perturbation responses.
Weihao Ge, an NCSA Genomics research scientist, said that the decision to use the top 50 differentially expressed genes seemed "quite arbitrary" and perhaps was not a large enough assay for analysis with the Gene Ontology (GO) resource.
Christina Fliege technical research lead at NCSA Genomics, said it was important not to "overinterpret" the results because the work involved Gene Ontology analysis. "Gene ontology ... analysis, in general just gives a hint of further experiments," Fliege said.
"That's a problem because the number of genes they're using for Gene Ontology as an input is quite small, which means that whatever you miss will have a big impact on the statistical results," Mainzer added.
Fliege was impressed that the experiment went beyond RNA-seq data to include RT-PCR and Western blot. "It definitely got into the trenches there to give some real molecular biology that looked pretty straightforward," she said.
Ge, however, had several unanswered questions. She said that the researchers did not specify what version of the Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) they were using. For example, she said that she has experienced missed annotations using STRING version 10.
To her, the preprint also was not clear on how the Mount Sinai researchers calculated whether a gene is an essential element of a specific drug's signature. "I feel they do not really explain how the Z-scores are calculated," Gu said. "I still don't know how they score whether the gene is really essential or not."
The researchers ranked genes based on "probable essentiality," but Mainzer thought that this term or methodology was not adequately explained. "That's something reviewers will need to pick up on when they send this to a journal," she said.
Mainzer said that reviewers and others who might consider adopting the methodology or running trials with the identified compounds will want to pay attention to the inputs in this experiment in order to validate the findings.
"Right now, it's very difficult for us to make that interpretation because we don't know the versions of the software they used for annotation and the number of genes they used was very small, so there can be some room for possible misinterpretation," Mainzer cautioned.
However, Mainzer said that the researchers accomplished their main objective of matching gene expression patterns in response to SARS-CoV-2 with the inverse of patterns related to drug response. "As they can make that match, ultimately, it doesn't matter exactly what the Gene Ontology database says because you're not diving into the biological underpinnings of these processes," she said.
"The ultimate validation is, if you apply this drug, does it work? Does it prevent COVID from reproducing?" Mainzer said. "They did the match, they found the drugs, and then they applied them to the cells."