Skip to main content

Stanford Spinout NuMedii Seeks Pharma Partners to Use PGx Repositioning Algorithm in Drug Pipelines

Premium

Originally published August 22.

By Turna Ray

Researchers at Stanford University have validated a computational method they hope can help pharmaceutical companies find new, pharmacogenomically derived uses for drugs in their pipeline.

In a paper published last week in Science Translational Medicine, researchers led by Atul Butte, associate professor of systems medicine in pediatrics at Stanford, described an algorithm with which they identified novel drug-disease associations based on gene expression data in the public domain. One of the surprising findings highlighted in the paper was that the over-the-counter anti-ulcer drug cimetidine could potentially be used to treat lung cancer.

In a second paper in the same publication, Butte and colleagues applied the method to predict that the anticonvulsant drug topiramate may be used to treat inflammatory bowel disease, and validated this hypothesis in a rodent model.

Through a Stanford spinoff called NuMedii, Butte and several others are talking to drug developers who want to test this computational model to repurpose drugs in their pipelines.

"The goal for NuMedii is to … enter into partnerships with pharma companies to do this kind of modeling and show what their drugs can be used for," Butte told PGx Reporter. "The idea is to de-risk [the clinical trial] for the pharmaceutical company in some way — not just make the predictions but actually test it in these preclinical models, and maybe even do a small, Phase IIa trial as a proof of concept."

At NuMedii's helm are experts who have extensive experience in genetics, running clinical trials, and finding new purposes for products. These skills can come in handy when translating an academic exercise into a real-world application.

Corey Goodman, formerly president of Pfizer's Biotherapeutics and Bioinnovation Center, is the chair of the company's board. Eric Schadt, a member of NuMedii's scientific advisory board, is the cofounder of Sage Bionetworks, chief scientific officer of Pacific Biosciences, and previously the scientific director of genetics at Merck subsidiary Rosetta Inpharmatics. And during her time heading up the strategic marketing of Affymetrix's Academic Research Business Unit, NuMedii cofounder Gini Deshpande led a team to identify new applications for the company's Gene Atlas system.

Discussions with drug firms are in early stages, and NuMedii hasn't inked any deals yet, but the company is eager to demonstrate that applying pharmacogenomics can add value to pharma's pipelines. We want to "go past that prediction [model] and show that actually does work and take that back to the pharma company," Butte said. "Otherwise it's just a software tool in some ways."

In order to test their algorithm, the Stanford researchers conducted validation studies for two drugs — the anti-seizure drug topiramate and the ulcer drug cimetidine — in animal models. According to Butte, the in vivo and in vitro validation experiments took between two to three months. He was surprised to find that much of the resources for the validation portion of the project weren't that hard to come by.

"We as bioinformaticians are surprised there are lots of contract research organizations, even within academia, that can do these kinds of studies to actually try the drug in the disease model," Butte said. While much of the animal disease modeling was done with collaborators and resources within Stanford, the researchers used an outside CRO to validate the topiramate prediction in in vivo rat models.

"One of the things I've learned is not to be shy about just pulling the trigger and [trying] these things, because more often than not I think we're going to have a lot of data showing that many of these drugs do work," Butte said.

For the studies, the researchers collected high-throughput gene expression data on 100 diseases from the Gene Expression Omnibus, housed at the National Center for Biotechnology Information. They then integrated this with information on 164 drugs from the Broad Institute's Connectivity Map, which contains gene expression data on human cell lines treated with therapeutic compounds.

Butte and colleagues developed an algorithm that searched through all the possible drug-disease pairs and then matched up treatments and diseases with gene expression patterns that cancelled each other out. The computational model thus operates on the hypothesis that "if a disease state is signified by a set of genome-wide expression changes, and if exposure to a particular drug causes the reverse set of changes in a model cell line, then that drug has the potential to have a therapeutic effect on that disease," the researchers wrote in the paper describing their overall methodology.

Using this system, Butte and his colleagues found more than 2,600 statistically significant drug-disease associations for 53 out of the 100 diseases.

"We have hundreds of drug and disease pairings here," Butte noted. The researchers were able to identify at least one disease to which each one of the 164 drugs were significantly associated.

Many of the associations identified known drug indications, providing proof that the computational model was working. For example, in the paper in which researchers validated topiramate as being potentially effective against inflammatory bowel disease, they highlight that the predictive algorithm also identified prednilosone, a drug already indicated as a treatment for the disease.

Oncology indications, such as gastric cancer, melanoma, and translational cell carcinoma, had the most match-ups with drugs in the database. Treatments that spanned the widest range of indications included Merck's HDAC inhibitor Zolinza, used to treat T cell lymphoma; AstraZeneca's non-small cell lung cancer EGFR inhibitor Iressa; and the antifungal trichostatin A, which were associated with 21, 18, and 16 diseases out of 100 in the dataset, respectively.

The repositioned drugs in the dataset have a diverse marketing profile — with some still sold as branded drugs under active patents and others available in generic form or over the counter — a fact that presents a range of value propositions for drug companies interested in using NuMedii's computational model to inform their pipeline decisions.

Topiramate, the drug that the algorithm predicted may work as an inflammatory bowel disease treatment, is an anticonvulsant that was developed by J&J subsidiaries Noramco and Ortho-McNeil Neurologics under the brand name Topamax. The drug has lost patent protection in the US. Generic versions of the drug are marketed by various firms.

In a somewhat similar story, cimetidine, a drug that researchers validated in in vitro and in vivo mouse xenograft models as a possible treatment for lung cancer, is an over-the-counter drug for treating ulcers and gastroesophageal reflux disease. Once marketed under the brand name Tagamet by GlaxoSmithKline, there are companies that market generic prescription-strength versions of the drug.

"The fact that these drugs are off patents, and cimetidine also happens to be over the counter, to [academic researchers] that's kind of a cool story," Butte said. Although cimetidine and topiramate make interesting drug repositioning stories, Butte acknowledged that actually finding a drug company that would be willing to fund human trials to confirm these leads in drugs that have lost IP protection may take some creative thinking.

When considering what to do with repositioning leads for treatments that have limited revenue potential, pharmaceutical companies may find a way to breathe new value into the drug by figuring out a new delivery method (ie. pill, liquid form, etc.), or develop an entirely new second-generation compound. The computational method "might give us an idea of the science and let us drive around that," Butte noted. "When pharmaceutical companies take a new drug forward, they still have to predict where that drug is useful before they decide to do that first clinical trial. So, we think it might be useful to use this [computational] approach way upstream, before the drug is even launched or the first trial is started."

Although drugmakers have molecular profiles for many of the drugs in their pipelines, most companies aren't using genomic strategies to find value in failed drugs or abandoned projects.

"GlaxoSmithKline has some folks doing this kind of work and Pfizer has an indications discovery unit," Butte reflected. "But, most pharmaceutical companies, if they do any repositioning, it's sometimes finding really obvious uses, such as if one drug is good for one cancer, then it may also be good in a different stage of that same cancer."

Indeed, those cases in which drug firms are applying genomic strategies to rescue failed drugs have focused on identifying patient subpopulations likely to respond to the drug for its initial indication, not entirely novel indications for the compound. For example, Pfizer is using gene markers to identify a subpopulation of melanoma patients that will respond to tremelimumab. If successful, Pfizer will have resuscitated the melanoma drug it previously stopped developing in 2008 due to unimpressive efficacy (PGx Reporter 04/28/2010).

Another instance in this vein is Novartis' lumiracoxib. Previously known as Prexige, the US Food and Drug Administration refused to approve the COX-2 inhibiting painkiller due to hepatic safety concerns. However, Novartis announced last year that it was investigating how to use pharmacogenomics to develop a companion test that would exclude osteoarthritis patients who might be more likely to experience adverse events (PGx Reporter 02/10/2010).

Butte observed that "out-of-the-box" disease-drug predictions "are really tricky," and that "there is really no standard way that these [predictions] are pursued today."

Since the computational model developed by Butte and his colleagues found many potentially new disease indications for Iressa, that example might provide a proof of concept for such an out-of-the-box drug repurposing effort.

Due to a large study showing Iressa to have low efficacy in the general NSCLC population in 2004, the FDA restricted any new patients from receiving the drug in the US, and AstraZeneca withdrew its application for the drug from European markets. Since then, AstraZeneca has been able to relaunch the drug outside the US with a companion test in EGFR-mutated NSCLC patients. However, instead of attempting a PGx strategy or trying to meet its post-marketing requirements – a necessary step if companies want to continue selling drugs that have received accelerated approval from the FDA – AstraZeneca earlier this year entirely gave up trying to gain full approval for the drug in the US (PGx Reporter 02/09/2011).

Iressa, a drug still marketed in European and Asian countries with an active patent, may be just the kind of case where NuMedii can offer some compelling options in terms of repurposing. "If there is a drug that has failed for one reason or another, not necessarily for toxicity, then we would absolutely agree there is some value and some incentive to use an approach like this and find another use for the drug," Butte said. "Otherwise, [lots of money] has gone wasted."


Have topics you'd like to see covered in Pharmacogenomics Reporter? Contact the editor at tray [at] genomeweb [.] com.

The Scan

Gap in COVAX Doses

BBC News reports that COVAX is experiencing a vaccine shortfall, as the Serum Institute of India has paused exports.

Sanofi, GSK Report Promising Results

The Wall Street Journal reports that the candidate SARS-CoV-2 vaccine from Sanofi and GlaxoSmithKline has had encouraging early results.

Influence of Luck

The New York Times examines how the US avoided variant-fueled increases in COVID-19 cases.

PLOS Papers on Retina GWAS, Hantaan Virus, COVID-19 Phenome-Wide Association Study

In PLOS this week: genome-wide association study of retinal morphology, analysis of hantaan virus found in a mouse, and more.