Scientists have developed what could be described as an "opposites attract" computer algorithm to identify new therapeutic applications for existing drugs, by comparing disease expression profiles in a public database with those resulting from drug administration. Funded by the NIH, the Stanford University Medical School-led team built a compendium of disease-drug relationship predictions based on matching genome-wide signatures of disease pathophysiology with drug effect. Importantly, instead of examining a single drug-disease pair or looking at reactions of a large set of drugs on a single disease, the search focused on the discovery of connections between 100 diseases and 164 drugs across all the gene measurements.
Reporting their approach and findings in two papers in Science Translational Medicine, Atul J. Butte, M.D., and colleagues say the results highlighted many known drug and disease relationships, but also predicted many new indications for the 164 drugs. One of these was use of the anticonvulsant drug topiramate, for the treatment of inflammatory bowel disease (IBD). Another suggested the ulcer therapy cimetidine could potentially treat lung cancer. The two publications describe in vitro and in vivo studies supporting the findings for both drugs.
The repositioning of drugs already approved for human use mitigates the costs and risks associated with early stages of drug development and offers shorter routes to approval for therapeutic indications, Dr. Butte et al. note in their paper titled “Discovery and Preclinical Validation of Drug Indications Using Compendia of Public Gene Expression Data." Indeed, successful examples of drug repositioning include the indication of sildenafil for erectile dysfunction and pulmonary hypertension, thalidomide for severe erythema nodosum leprosum, and retinoic acid for acute promyelocytic leukemia.
The prevailing approach to drug repositioning is based on established tools and techniques developed for screening libraries of lead compounds against biological targets of interest in early-stage drug discovery, although computational approaches to the discovery of novel biological targets for approved drugs have been developed as a means to reducing costs and other logistical limitations associated with experimental HTS approaches.
The Stanford researchers’ approach took a different tack to existing computational approaches. Rather than look at individual drugs or individual diseases, they cross referenced data from publicly available databases to identify drug-related gene expression profiles that effectively canceled out gene-expression profiles resulting from disease. The working hypothesis was simple, Dr. Butte notes. “If a drug exerts a change on gene-activity pattern that is opposite to that exerted by a disease, then that drug may have a therapeutic effect on that disease.”
Practically, the researchers identified and combined data from publicly available microarray datasets from National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (GEO) representing 100 diseases, with gene expression data from several human cancer cell lines treated with 164 drugs or small molecules, in order to predict previously undescribed therapeutic drug-disease relationships. They defined a disease signature for each of the 100 diseases as a set of genes that are significantly up- and down-regulated for that disease compared with normal values using significance analysis of microarrays (SAM). The next stage was to statistically compare each of the disease signatures to each of the reference drug expression signatures from the Connectivity Map, a collection of genome-wide transcriptional expression data from cultured human cells treated with a broad range of FDA-approved bioactive small molecules, which was first described back in 2006.
The overall results provided significant candidate therapeutic drug-disease relationships for 53 of the 100 diseases. Each of the 164 drugs was significantly associated with at least 1 of these 53 diseases. Many cancers showed the highest number of significant matches to therapies, and the histone deacetylase (HDAC) inhibitor vorinostat represented the drug predicted to have associations with the largest number of diseases: 21 out of the 100 in the dataset. Other HDAC inhibitors, such as trichostatin A and Helminthosporium carbonum (HC) toxin, as well as the anticancer EGFR inhibitor gefitinib, had more than 20 significant predicted therapeutic indications.
Supporting the basic validity of the approach, the results suggested that drugs with similar mechanisms of action clustered together in terms of the diseases they would potentially treat, while diseases were similarly clustered on the basis of their computed therapeutic scores across the panel of drugs: “We found a large cluster of cancers, including lung, stomach, and other cancers, pointing to potential commonality of predicted therapeutic response between these conditions, the authors write, while “the inflammatory bowel diseases, ulcerative colitis (UC), and Crohn disease (CD) appeared together as part of a larger cluster of diseases for which corticosteroids and other immunosuppressive drugs are broadly indicated.” Interestingly, IBD was clustered near infection by Yersinia enterocolitica, which can present clinical symptoms similar to those of IBD, they add.
As would be expected, the results predicted that the known anticancer drugs, HDAC inhibitors (trichostatin A, valproic acid, vorinostat, and HC toxin) have therapeutic effects in treatment of different types of brain tumors (astrocytoma, glioblastoma, and oligodendroglioma), as well as other cancers (esophagus, lung, and colon). In addition, one of the strongest therapeutic predictions for CD and UC was the corticosteroid prednisolone, a well known treatment for these conditions.
However, the results also threw up a number of surprises, the authors report. One in particular was the finding that topiramate, an anticonvulsant drug currently used to treat epilepsy, had a stronger therapeutic score for CD than prednisolone, and was also one of the strongest predicted therapies for UC. It was this association that was further evaluated in the second Science Translational Medicine paper titled “Computational Repositioning of the Anticonvulsant Topiramate for Inflammatory Bowel Disease.”
The computational results separately turned up a moderately strong prediction for cimetidine as a treatment for lung adenocarcinoma (LA). Testing this drug-disease pairing in vitro, the researchers evaluated the growth, proliferation and apoptisis of LA cells after exposure to cimetidine. LA cells exposed to increasing concentrations of the drug exhibited a dose-dependent reduction in growth and proliferation compared with controls, and cimetidine administration also led to significant levels of apoptosis. To test the findings in vivo, three doses of cimetidine were then administered to mice implanted with a human LA cell line, and growth of tumors monitored for 12 days. The results showed that cimetidine therapy significantly blunted tumor growth in a dose-dependent manner, and at the highest dose the drug was almost as effective as doxorubicin, which was used as a positive control therapy, the researchers state. In contrast, cimetidine had no effect on the growth of renal tumors in mice, supporting the original computational data that showed specificity for cimetidine against lung cancer.
Moving on to further test the prediction that topiramate could treat IBD, the researchers treated IBD-induced experimental rats with either topiramate or prednisolone. Both drugs resulted in animals displaying reduced IBD symptoms including diarrhea, and tissue analyses showed reduced gross pathological inflammation, ulceration, and damage to the colon mucosal layer in both the topiramate and prednisolone-treated groups.
The team admits that their initial findings for both cimetidine and topiramate will need further preclinical and clinical evaluation, but claim the results at least go some way toward validating the concept of using computational analysis of public gene-expression databases as an approach to uncovering additional uses for approved drugs.
The observation that drugs that cluster on the basis of their gene signature similarity across diseases exhibited share mechanisms of action (such as HDAC inhibition) or shared physiological processes (for example, immunomodulation), may also help provide clues about how drugs with hitherto unknown mechanisms of action may work, or even suggest new biology, they note.
“This work is still at an early stage, but it is a promising proof of principle for a creative, fast, and affordable approach to discovering new uses for drugs we already have in our therapeutic arsenal,” adds Rochelle M. Long, Ph.D., who directs the NIH Pharmacogenomics Research Network. “Bringing a new drug to market typically takes about $1 billion, and many years of R&D. If we can find ways to repurpose drugs that are already approved, we could improve treatments and save both time and money.”