Using differential gene expression analysis & drug repurposing techniques to find compounds with similar gene expression profiles to the HSP90 inhibitor Geldanamycin
Finding a broad-spectrum antiviral compound could greatly help in combating viral diseases, drug resistance and pandemics. One candidate for broad-spectrum antivirals are HSP90 inhibitors such as Geldanamycin, however, there may be compounds with similar antiviral properties that are better candidates. Broad-spectrum antiviral drug discovery seems to be a particularly neglected area of research perhaps because broad-spectrum antivirals may be less useful in developing countries which do not have rapid healthcare/drug access. In this project, bioinformatics methods in gene expression, drug discovery and repurposing were used to find compounds with similar gene expression profiles, and therefore potentially antiviral activity, to the HSP90 inhibitor Geldanamycin. Using drug repurposing techniques of this kind to find broad-spectrum antiviral candidates may be a good strategy to counteract the biggest problem for these compounds, their toxicity. While very preliminary, the results were promising because numerous compounds where found to have similar gene expression profiles, such as kinase inhibitors which may provide an alternative host-directed antiviral.
Overall it was deemed that this project was unlikely to be an information hazard and so has been made publically available.
While broad-spectrum antibiotics are one of the greatest inventions of all times due to their ability to treat a wide array of bacteria, they are ineffective against viruses. Due to viruses being more difficult to disinfect, for this reason, as well as them mutating more rapidly and being more transmissible it's likely the viruses pose a greater risk of a pandemic than bacteria. Therefore, a broad-spectrum antiviral would be a useful therapeutic and countermeasure to safeguard from this. Especially because a broad-spectrum antiviral would be able to administered rapidly regardless of the viral pathogen responsible for the disease/pandemic.
One such candidate for a broad-spectrum antiviral are HSP90 inhibitors (HSP90is), some of which are in clinical trials as an anti-cancer drug. HSP90is are a class of molecules which inhibit heat shock proteins. Heat shock proteins are chaperone proteins which play an important role in the synthesis and folding of proteins in humans. It is these mechanisms which are often hijacked by a wide variety of viruses and used to make viral proteins. Therefore, by inhibiting these proteins with HSP90is it is possible to prevent viruses from replicating. While they may not be suitable as a long-term treatment due to their side effects/toxicity they may provide an effective countermeasure in a pandemic by buying more time for the immune system to fight of the infection and for vaccine development (Somerville and Youngs, 2018).
One of the biggest challenges of developing HSP90is as a broad-spectrum antiviral is their toxicity (Wang et al., 2017). Therefore, by using drug repurposing methods it may be possible to find other compounds which posses the same antiviral activity as HSP90is, however, are already approved drugs and are therefore likely to be much safer. In general, repurposed drugs are approved sooner (3-12 years), at a redeuced cost (50-60%) and are more likely to be approved (Hernandez et al., 2017).
The aim of this project was to find compounds with similar gene expression profiles to the HSP90 inhibitor Geldanamycin, especially compounds which are already approved. This is because such compounds may have similar antiviral activity as HSP90is while having lower toxicity and greater chance of approval.
Scripts were written in Nextflow which is a workflow manager allowing highly scalable and parallelised analysis. Docker containers were used to bundle software dependencies for reproducibility. The first three steps were run on Lifebit's Deploit platform which allows versioning and sharing of the analysis.
A data-donwloader pipeline was written & executed to download publicly available bulk RNASeq data from EMBL-EBI. In total 96 FastQ files (94 GB) of raw gzipped paired-end data from the study PRJNA480665 were used. The data came from eight patients and contained airway epithelial cells treated with and without the HSP90 inhibitor Geldanamycin (Pezzulo et al., 2019).
For full execution details see Deploit job
The downloaded reads were then analysed using the nf-core/rnaseq pipeline, which is developed by the open source bioinformatics community. The reads were first trimmed using TrimGalore to remove adapter sequences/contamination from the sequencing and to remove low quality regions. The reads could then be aligned to the GRCh37 reference genome using STAR. After the alignment it was then possible to determine the gene counts using FeatureCounts and merge this data for all of the samples.
For full execution details & Multiqc report see Deploit job
The merged gene counts were then used to generate a list of differentially expressed genes with lifebit-ai/dean pipeline and DESeq2. This experiment file was also used as input data to assign each of reads the relevant experimental group.
For full execution details & R Markdown report see Deploit job
From the list of differentially expressed genes, a R Script was used to extract the top underexpressed and overexpressed genes based on the adjusted p-value. This data was then uploaded to the Connectivity Map to find compounds with similar gene expression profiles.
A R Shiny web app was built to visualise the drug repurposing results. This allowed the identification of mechanism of actions, associated conditions, targets and pathways that were common across multiple compounds which have similar gene expression profiles. Data across all cell lines and the cell line A549 were used. Cell line A549 was chosen because the samples were from airway epithelial cells which may have a more similar gene expression profile to alveolar basal epithelial cells compared to other cell types.
By plotting the data from the RNASeq edgeR sample correlation it seemed as though the differences in gene expression between the patients were greater than those between the experimental groups of Geldanamycin treatment compared with the control. Therefore, the design formula (∼ patient + treatment) for the DESeq2 analysis was used to try and control for the effect of patient.
A sample distance matrix was performed to calculate the distance between each of sample using the read counts. Here the green & orange represent the conditions. Not all of the treated samples cluster together and so the Geldanamycin treatment may not be enough to separate the samples.
In total 25 compounds were found to have a similarity score over 95 for cell line A549.
| Score | Name | Description |
|---|---|---|
| 99.98 | AG-490 | EGFR inhibitor |
| 99.83 | nor-binaltorphimine | Opioid receptor antagonist |
| 99.8 | corynanthine | Adrenergic receptor antagonist |
| 99.56 | tetrahydrobiopterin | Nitric oxide stimulant |
| 99.56 | 16,16-dimethylprostaglandin-e2 | Prostanoid receptor agonist |
| 99.53 | velnacrine | cholinesterase inhibitor |
A relatively high number of compounds had similar gene expression profiles to Geldanamycin. This is especially promising given that the Connectivity Map database largely consists of FDA-approved drugs and so it is likely that there may be other compounds with similar gene expression profiles, which may be already approved drugs. Some of these compounds may confer similar antiviral activity to HSP90is and be able to gain market approval sooner than HSP90is as a broad-spectrum antiviral.
For example, the top compound from the drug repurposing, AG490, has been shown to have some antiviral properties against the herpes simplex virus (Yura et al., 1997) and down-regulate HSP70 expression (another chaperone protein) (Xu et al., 2018). AG490 is a kinase inhibitor which inhibits EGFR, a host factor broadly required by viruses. Kinase inhibitors ranked very highly in the durg repurposing analysis and are a broad-spectrum antiviral candidate (Schor and Einav, 2018). What's more is that kinase inhibitors such as Erlotinib are already approved and so may provide the fastest to getting an approved broad-spectrum antiviral. Other compounds of interest include Vidarabine which is already approved as an antiviral. As well as HDAC inhibitors which have a shred mechanism of action as HSP90 inhibitors and there are currently four approved (anti-cancer) drugs for (Pezzulo et al., 2019). The method used in this project could be applied to more publicly available datasets and different host-directed antivirals. Further research could also be done using pathway analysis to find suitable a mechanism of action, target and pathway such as the methods used in Prussia et al., 2011.
However, some limitations of the project were that the gene expression profile used here may not capture the treatment of Geldanamycin well. This can be shown by the sample distance matrix where, despite adjusting for the patients, the Geldanamycin treated samples did not cluster together. This may mean that there was another confounding variable which was contributing to the gene expression profile. It seems like the gene expression profile does not accurately represent Geldanamycin treatment because Geldanamycin was ranked very low in terms of similarity score when it should have been the highest. The highest HSP inhibitor was dihydro-7-desacetyldeoxygedunin which had a similarity score of 88.56. Therefore, further work needs to be done to produce a more representative gene expression profile of a HSP90i before finding compounds with a similar gene expression profile to this.
Other limitations are that just because compounds have a similar expression profile does not necessarily mean that they will confer similar antiviral activity. This could be improved upon with better experimental design. By using pre-existing public data the number of possible experimental deisgns was far more restricted. Instead screening could be performed whereby cells infected with viruses could be treated with the different compounds and the number of live dead cells recorded. This would be beneficial in that compounds with very different expression profiles to HSP90is which are still effective antivirals would not be discriminated against. Single-cell RNA sequencing of these samples could be used and differential expression analysis performed to get higher resolution detail of the genes/pathways which are affected. In vivo toxicity could also be investigated as this seems to be one of the key bottlenecks for the current clinical trials for HSP90is as an anti-cancer drug (Wang et al., 2017).
Another limitation is that the cell line which was used for the gene expression profile was different from that used to comapre against drugs in the Connectivity Map database. While both were from epithelial cells, this is a confounding variable and so a cell line used in the Connectivity Map database should have been used to measure the gene expression profile. Also, it may be possible to replicate and improve upon the results generating by the Connectivity Map by training and testing custom machine learning algorithms such as GANs to find compounds with similar gene expression profiles.
Overall, many compounds were found with similar gene expression profiles, including kinase inhibitors which seem like a promosing broad-spectrum antiviral candidate. However, it is unclear if the gene expression profile captures the effect of Geldanamycin or other confounding variables. Therefore, more work needs to be done to confirm this with improved experimental design. Other methods could also be used alongside drug repurposing such as pathway analysis and single-cell RNASeq. While there may be an incredibly small chance of finding a broad-spectrum antiviral with a sufficiently strong mechanism of action and low toxicity, this chance can be increased by applying methods such as drug repurposing and large scale screening. Even if the chance is very low it seems clear that investigating broad-spectrum host-directed antivirals is important enough that it warrants investigation regardless.
- Somerville, C. and Youngs, H. (2018). Research and Development to Decrease Biosecurity Risks from Viral Pathogens. [online] Open Philanthropy Project. Available at: https://www.openphilanthropy.org/research/cause-reports/research-and-development-decrease-biosecurity-risks-viral-pathogens [Accessed 22 Sep. 2019].
- Wang, Y., Jin, F., Wang, R., Li, F., Wu, Y., Kitazato, K. and Wang, Y. (2017). HSP90: a promising broad-spectrum antiviral drug target. Archives of Virology, 162(11), pp.3269-3282.
- Pezzulo, A., Tudas, R., Stewart, C., Buonfiglio, L., Lindsay, B., Taft, P., Gansemer, N. and Zabner, J. (2019). HSP90 inhibitor geldanamycin reverts IL-13– and IL-17–induced airway goblet cell metaplasia. Journal of Clinical Investigation, 129(2), pp.744-758.
- Hernandez, J., Pryszlak, M., Smith, L., Yanchus, C., Kurji, N., Shahani, V. and Molinski, S. (2017). Giving Drugs a Second Chance: Overcoming Regulatory and Financial Hurdles in Repurposing Approved Drugs As Cancer Therapeutics. Frontiers in Oncology, 7.
- Yura, Y., Kusaka, J., Tsujimoto, H., Yoshioka, Y., Yoshida, H. and Sato, M. (1997). Effects of Protein Tyrosine Kinase Inhibitors on the Replication of Herpes Simplex Virus and the Phosphorylation of Viral Proteins. Intervirology, 40(1), pp.7-14.
- Xu, N., Chen, Y., Liu, W., Chen, Y., Fan, Z., Liu, M. and Li, L. (2018). Inhibition of JAK2/STAT3 Signaling Pathway Suppresses Proliferation of Burkitt’s Lymphoma Raji Cells via Cell Cycle Progression, Apoptosis, and Oxidative Stress by Modulating HSP70. Medical Science Monitor, 24, pp.6255-6263.
- Schor, S. and Einav, S. (2018). Repurposing of Kinase Inhibitors as Broad-Spectrum Antiviral Drugs. DNA and Cell Biology, 37(2), pp.63-69.
- Prussia, A., Thepchatri, P., Snyder, J. and Plemper, R. (2011). Systematic Approaches towards the Development of Host-Directed Antiviral Therapeutics. International Journal of Molecular Sciences, 12(6), pp.4027-4052.