Drug Repositioning Suggests a Role for the Heat Shock Protein 90 Inhibitor Geldanamycin in Treating COVID-19 Infection

Drug repositioning offers an unmatched opportunity to offer novel therapeutics to treat SARS family of coronaviruses (SARS-FCoVs); an issue that became extremely urgent with the spreading of a novel virus with potential to threaten the lives of millions of people. Hereby, we analyzed a dataset of patients who presented with SARS during the 2003 outbreak. We established a gene signature that defines differential gene expression in patients who were sick with SARS vs. healthy controls and convalescent patients. We used a robust platform to conduct drug repositioning based on clustered gene expression and pathway enrichment to identify best matching drugs. We identified 55 agents of potential benefit. In most of these drugs we were able to establish a link to previous related research, use as antiviral, or at least a hypothetical role in treating SARS-FCoVs. Most notably, the heat shock protein 90 (hsp90) emerged as a major component that enables viruses to hijack infected cells through the process of autophagy. Almost half of the drugs identified could be linked to hsp90. As such, we propose using hsp90 inhibitors, mainly geldanamycin and its derivatives, to treat COVID-19.


Introduction
A new outbreak of Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2), which causes Coronavirus Disease 2019 (COVID- 19), was initially recognized in January 2020. Its rapid spread poses a potential threat to the security, health and economy of the world. [1] While researchers struggle to produce new therapeutics in timely fashion; the rapid spread of the disease calls for immediate interventions to protect exposed individuals and treat those affected. In the best-case scenario, new therapeutics require months to develop and manufacture, a luxury the global health community can ill afford given the urgency of the situation. To date, WHO reports more than 90,000 affected individuals, 3,000 deaths, and 70 affected countries, statistics that increase daily. At the time of writing this manuscript, more than 80 clinical trials are being launched in China related to treatment of COVID-19. These trials include studies involving stem cells, herbal remedies, novel antivirals (e.g. remdisivir, a nucleotide analog), anti-retroviral drugs (e.g. Kaletra, a combination of ritonavir and ipinavir), antimalarial drugs (e.g. chloroquine) and serotherapy (using convalescent plasma). [2][3][4][5] Previous experience with coronaviruses proved how difficult their treatment can be.
Severe Acute Respiratory Syndrome Coronavirus (SARS-CoV) and Middle East Respiratory Syndrome (MERS) viruses had high case fatality rates despite attempts to identify effective therapy. [6] Vaccine development by multiple biopharmaceutical companies is ongoing; with the most promising approach being nucleic acid-based vaccines. [7] Biologically, SARS-CoV-2 shares many similarities to previous viruses. All three viruses are enveloped, positive-sense, single-stranded RNA Beta-coronaviruses. Additionally, the spike glycoprotein of these viruses use the angiotensin-converting enzyme 2 (ACE2) surface protein as a receptor, making this interaction the focus of intensive research in different animal species. [8] Due to these similarities and for simplification, we will refer to the three viruses (SARS-CoV, MERS and SARS-CoV2) as the SARS family of coronaviruses (SARS-FCoVs).
Drug repositioning, defined as the use of approved drugs for new indications, often unrelated to their primary use, can identify already available drugs that may effectively treat emerging diseases like COVID-19 immediately, even while more specific therapies and vaccines are being developed. In-silico testing offers immediate insight into pathways and drugs that interact with these pathways to promote or inhibit their activity. Eleven studies have been published about previous drug-repositioning efforts for SARS-FCoVs [9][10][11][12][13][14][15][16][17][18][19]. These studies were identified using the keywords ("Drug Repurposing" OR "Drug Repositioning") AND (SARS OR MERS OR coronavirus OR COVID-19 OR SARS-CoV-2). With most of these publications being review and insight papers, we were looking for studies that used gene signature generated by comparing sick patients with SARS-FCoVs to controls, in order to identify modifiable host factors that may help to combat the disease.
This model is at the core of drug repositioning/repurposing strategy. [20] This approach was highlighted by the recent study of Amemiya et al for Dengue hemorrhagic fever, which identified 8 candidate drugs through drug repositioning. Notably, 3 of these drugs appeared in our identified drugs (sirolimus, valproic acid and estradiol) as shown in our results below. We felt there is a need for a study that includes sufficient number of patients to perform our analysis. Due to time-constraints, we chose to study previously deposited data. We hope this work will help researches identify new agents of potential use to treat COVID-19 and related illnesses.

Results
GSE5972 provides logarithmically transformed gene expression values for a group of patients who presented to Toronto area hospitals. They had had positive PCR and/or seroconversion as evidence of their SARS-CoV infection. [21] In total, there were 70 sets; 10 for healthy controls (median age, 28.5 years; Females, 50%), 5 patients had samples during convalescence (6 samples total, median age, 42 years, Females 100%, median days since onset, 27 day), and 35 patients with samples labelled as pre-O2-nadir (26 samples) and post-O2-nadir (28 samples). We selected samples labelled as "pre-O2-nadir" and labelled them as "sick". There were 19 individuals in this group (26 Samples, median age, 49.9 years, Females, 79%, median days since onset, 4 days).
Heat maps with 5 gene clusters were generated using cogena functions. These clusters were subjected to KEGG pathway enrichment. As expected, sets with lower number of genes were less likely to show significant enrichment. Gene enrichment results for the 4 analysis sets are shown in table 1. Subsequent drug repositioning search focused on all up-regulated genes (analysis set 1), cluster 1 (analysis set 2), all down-regulated genes (analysis set 3) and clusters 1 (analysis set 4). Identified drugs were then used for Pubmed search and the number of hits was recorded. Drugs with identified hits were then searched manually and synopsis of their role as antiviral medications is mentioned in the discussion below. Identifying and repositioning existing drugs to treat COVID-19 infections may provide a rapid, inexpensive option with known toxicities. [10,12,14,17,22] The purpose of our study was to evaluate the levels of expression of genes in patients while they are sick and compare them to healthy controls and convalescent patients.
Genes with different levels of expression reflect pathways that are used by the host in response to viral infection (sick vs. Convalescent) or pathways that are dysregulated by the infection (sick vs. Healthy). [22] Similar to all bioinformatics pipelines, parameter and algorithm selection impact results, and may not be optimal even after tuning, which makes use of a black-box approach for drug repositioning less appealing. [23] However, despite limitations, this approach to identify drugs that interact with differentially expressed genes has proven useful in many cases. [20] The algorithm we used in this paper, cogena, overcomes several limitations of this approach. It clusters genes and enriches the clusters using established ontologic pathways, and thus is more likely to find clinically-relevant drugs. In pursuing this analysis, two questions must be addressed: First, which patient group is more suitable for comparison with sick patients, healthy or convalescent patients? One may argue that healthy controls were not exposed to the virus and identifying genes with different levels of expression in this group does not reflect the physiologic pathways that lead to recovery, which are of great importance in this current problem. The second question concerns the selection of the number of differentially expressed genes for drug repositioning. Obviously, including more genes will yield more pathway enrichment. However, increasing the number of used genes will add more confusion to the model. Many differentially expressed genes reflect pathways that are related to other physiologic processes (e.g. improved tissue nutrition, oxygenation and perfusion). To cover all these possibilities we decided to include 4 analysis sets of gene expression.
It is very impressive that different drug repositioning approaches lead to similar results.
For example, our study did not show chloroquine, which was suggested before to be active Omeprazole, an H+K+ proton pump inhibitor inhibits v-ATPase at higher concentration.
[24] Additionally, chloroquine is known to increase the pH of endosomes. [25] Colchicine and vinca alkaloids interact with cellular microtubules, which cause inhibition of lysosomal formation. [26] Non-steroidal anti-inflammatory drugs (e.g.Nabumetone) exhibit their antiviral/antitumor effect by via inhibition of endoplasmic-reticulum-resident protein kinase (PERK) mediated phosphorylation of the alpha subunit of eukaryotic translation initiation factor 2A (eIF2a). This leads to protein synthesis inhibition and cellular death.
[27] Etacrynic acid, a drug identified in analysis set 1, was shown before to be a suitable inhibitor of the main coronavirus protease (M pro) due to its electrophilic binding of its active domain. This binding inhibits the function of this vital protease which cleaves viral proteins at 9 specific sites in preparation for viral assembly. [28] Amiloride derivatives interacts with protein ion channel (protein E of coronavirus) which prevents viral formation and stops replication, at least in "in-vitro" models.
[29] Figure 1 shows a simplified representation of some of the drugs we identified and their mode of action.
One drug of great interest that was identified in our study was captopril. This commonly used antihypertensive medication from the Angiotensin-converting enzyme (ACE) inhibitor class works by blocking the active domain of this enzyme, leading to decreased conversion of angiotensin I to angiotensin II. Another rarely used ACE inhibitor appeared in analysis set 3, rescinnamine; which is a vinca alkaloid derivative. While the SARS-FCoVs get access to cells via the ACE-II protein, which carries similarity to ACE-I protein as more than 60% of their sequences are identical. We wonder why captopril was never studied invitro or in-vivo to treat coronaviruses. Using this drug in particular, despite proposed benefit, should be only attempted in non-human models first. Prolonged use of captopril is associated with increasing levels of ACE-II. Whether this could possibly lead to enhanced entry of the virus is a question that can only be answered through careful preclinical tests.
Geldanamycin and its derivatives (tanespimycin and alvespimycin) were identified in analysis sets 3 and 4. Geldanamycin is an antibiotic that is believed to be targeting the ADP/ATP binding site of the Heat-shock protein (HSP90). It was shown to be targeting many viruses, including SARS coronavirus, where it inhibits SARS replication with a selectively index of >300 and an EC50 of 0.91uM; for comparison the EC50 for chloroquine is 8.8uM. The drug was submitted for a Chinese patent (application 03146591, July 2003).
In a recent study that evaluated 87 drugs effective in MCF7 breast cancer epithelial cells, 17-allyloamino-geldanamycin, LY294002, valproic acid, trichostatin A and sirolimus (in addition to wortmannin), were evaluated using pathway gene enrichment. Two pathways (ERBB_signaling and Insulin_signaling) were the 2 most combined pathways. [31] Wortmannin, another PI3K inhibitor, similar to LY294002, is known to inhibit RSV replication by inhibiting autophagy, a process that favors viral replication by inhibiting apoptosis. [32] This finding is also supported by wortmannin inhibition of coxsackievirus replication in infected mice treated with a histone-deacetylase which induced replication by inhibiting autophagy and inducing apoptosis. [33] Geldanamycin is a benzoquinone with known antimalrial activity. It belongs to a group of drugs called ansamycins. One of the potent antiviral ansamycins is rifampin. This commonly used antibiotic has a known antiviral function due to its inhibition of DNAdependent RNA polymerase. [34] This selective action makes rifampin of no value in treating RNA viruses, like coronavirus. To the contrary, geldanamycin is a potent HSP90 inhibitor and works on a broad spectrum of viruses through modifying host response. [30] While effective in treating resistant multiple cancer types, hepatotoxicity of geldanamycin limited its use and lead to the development of derivatives with less toxicity and improved oral bioavailability, namely tanespimycin and alvespimycin. [35] Autophagy, a process where the internal cellular components degenerate, with or without integration with lysosomes, resulting in autophagosomes, which are hijacked by some viruses and used as safe sanctuary. This is opposed by programmed cell death, or apoptosis, which programmatically destroys the cell and calls for its clearance. Several studies showed the ability of SARS-CoV to induce autophagy. [36,37] Notably, autophagy plays a role in cancer resistance to chemotherapy and modulation of its components may restore drug sensitivity. [38] Chloroquine and hydroxychloroquine show promising results in restoring chemotherapy sensitivity by modulating autophagy. [39] It is suggested that chloroquine function as antimalarial and antitumor drug can be linked to Hsp90 inhibition with secondary modulation of autophagy, among other pathways. [40] An estrogen (alpha-estradiol), an anti-androgenic drug (nilutamide) and an aromatase inhibitor (exemestane) appeared on the analysis set 1 and set 2. This is noteworthy, particularly as the SARS-FCoVs are known to cause more severe infections in males and are associated with higher mortality; a phenomenon that proved true in male mice and female mice treated with estrogen receptor antagonist. [41] Complex interaction between estrogens, Hsp90 and lysine methyltransferase (SMYD2) point to the interaction of sex hormones with the function of this molecular chaperone in modulating autophagy. [42] Antidiabetic drugs (metformin, pioglitazone and possibly acetohexamide) restrict viral replication by activating AMP-activated protein kinase which modulates autophagy. [43,44] Multiple sulfonamides were identified, including antibiotics (sulfaphenazole, sulfaguanidine, sulfadiazine). In addition, diphemanil metisulfate and copper sulfate are sulfur containing drugs. Of note, the family of sulfonamides includes multiple drugs that are potent antivirals. [45] Many sulfonamides are known to inhibit the ATP-binding pocket of Hsp90. [46] Sirlomus, also known as rapamycin, is a macrolide mTOR inhibitor and was reported in analysis sets 1 and 2. Sirolimus is known to have antiviral effect on multiple viruses, including CMV, HIV, EBV and others. [47][48][49] Interestingly, the mTOR pathway functions in parallel with Hsp90 in modulating autophagy. [50] Inhibition of both pathways is synergistic in infected cells. The suggested link between these pathways involves viralinduced activation of the Hsp90 client, Akt, which in turn phosphorylates mTOR to facilitate translation of viral mRNA. [51] All of these identified drugs highlight Hsp90 as a druggable target that deserves more attention (Fig 2).
Based on our findings, we suggest adding more drugs to the experimental arsenal deployed against SARS-FCoVs. We suggest using nontoxic drugs that are readily available for wide scale prophylaxis or the treatment of mild cases of COVID-19. These drugs would include omeprazole, nonsteroidal anti-inflammatory drugs, colchicine, sulfoanamide antibiotics, antimalarial drugs and clozapine/chlorpromazine. For patients with serious COVID-19 illnesses, the above drugs can be combined with immune modulators like sirolimus, and direct Hsp90 inhibitors under well-designed clinical trials. Given the high pathogenicity of the current ongoing epidemic and its potential to become a pandemic, [52] it seems prudent to attempt to use multiple drugs in combination if low risk for toxicity and drug-drug interaction is established. Choosing a combination of drugs identified in this study or other drug-repositioning studies and careful reporting of the efficacy of these experiences can result in rapid accumulation of knowledge.
In conclusion, we showed that a drug repositioning framework using simple and unambiguous methodology can yield a plethora of drugs that are active for SARS-FCoVs.
We believe that a combination of some of these relatively nontoxic drugs can be worth trying (Fig 3). Serious efforts should be shifted to Hsp90 and autophagy pathways.
Multiple candidate drugs can be of great value; most notably, geldanamycin and its derivatives. Hsp90 inhibitors are being studied for cancer patients. It seems there is an urgent need to focus on this important pathway to overcome the ongoing outbreak.

Methods
We searched for a well annotated dataset and identified an efficient algorithm to generate a list of drugs that are potentially useful to treat this disease. [23] Our methodology is based on classifying patients into "sick", "convalescent" and "healthy controls" categories; followed by identifying gene signatures that are amenable for drugrepositioning analysis.
After searching the Gene Expression Omnibus (GEO) for the transcriptomic data related to SARS-FCoVs, GSE5972 was identified as a suitable candidate. Gene expression of 40 individuals infected with SARS and 10 healthy controls were included in our analysis.
Studied RNA was extracted from peripheral blood mononuclear cells. Expression levels were analyzed using the noncommercial UHNMAC Homo sapiens 19K Hu19Kv8 microarray platform. Applied microarray procedures are mentioned in the study paper. [21] Phenotypic data provided with the dataset included a unique patient identifier, gender, age, dayssince-onset of symptoms, and status. This latter variable indicated if the sample was collected at time of active infection (pre-O2 nadir and post-O2 nadir), or convalescence.
Data was submitted to NCBI on October 5th, 2006 and was made available on July 31st, 2007. We downloaded this file on February 20th, 2020.
For the purpose of differential gene expression, samples in the dataset were classified into sick (pre-O2 status, N = 26), convalescent (N = 6 samples) and healthy (N = 10). The Limma package and R code provided by the GEO analyzer were used to perform differential gene expression (DGE) of sick vs. convalescent patients and sick vs. healthy. No competing interests for any of the authors.

Data availability:
Data used for this study is already available through the NCBI-GEO website.  Figure 1 A simplified cartoon showing suggested pathways that some of the drugs indentified in our study exploit. Some drugs work on more than one pathway.

Figure 2
Interactions between the Heat Shock Protein (Hsp90) and different drugs, placing  Categories of drugs used to treat SARS-FCoVs according to our analysis.

Supplementary Files
This is a list of supplementary files associated with this preprint. Click to download.
Supplementary information.pdf