Skip to main content
Premium Trial:

Request an Annual Quote

Weill Cornell Scientists Combine Datasets to Find Drugs for Oncogenic Transcription Factors

Premium

NEW YORK (GenomeWeb) – Transcription factors associated with cancer have been elusive drug targets. While there have been some serendipitous successes, the dynamics of DNA binding make them difficult to disrupt.

However, a new study suggests that an informatics-driven approach can find existing drugs that also block these proteins gone awry. And for anyone with access to several large data sets and the expertise to sift through them, more could be found, according to the study authors.

Led by first author Kaitlyn Gayvert and co-senior authors Olivier Elemento and David Rickman of Weill Cornell Medical College, a group of New York City-area scientists have reported a method for finding drugs to target transcription factors by combining ChIP-Seq data sets with drug-induced gene expression profiles to find candidates, a network analysis to rank candidates, and a retrospective electronic medical record (EMR) search to validate the drug's effectiveness in preventing or delaying oncogenesis.

"We took a computational repositioning approach for targeting transcription factors and automated the process," Elemento told GenomeWeb. "We can screen thousands of drugs this way."

Starting with the transcription factor ERG, which is mutated in approximately half of all prostate cancer cases, the study strung together several datasets, including the Broad Institute's Connectivity Map, a curated drug-protein and protein-protein interaction network, to find drugs that might inhibit it.

"Because ERG is a transcription factor, it was not thought to be druggable," Elemento said. However, the study found several drugs that might inhibit ERG though, with the network analysis suggesting the corticosteroid dexamethasone was the most promising. Using EMRs from Columbia University Medical Center, the researchers showed that the drug afforded a protective effect against prostate cancer to those who had taken the drug for other reasons.

"It's a trend, not a fully protective effect. But it's a very clear trend," Elemento said.

The researchers published their results yesterday in Cell Reports.  

Rickman's lab at Weill Cornell has specialized in looking at transcription factors involved in prostate cancer for years, and Elemento uses computational approaches to studying cancer. ERG is well known to be overexpressed in as many as 50 percent of prostate cancer patients, however, "There are not drugs to target this class of prostate cancers," Elemento said. 

That could be due to the fact that there previously weren't any drugs to target ERG. In general, drugging transcription factors is difficult. "The contact between transcription factors and DNA occurs on a very broad surface" and at hundreds of sites in the genome, Elemento said. "It's very hard to block this interaction."

But the researchers wondered if there were existing drugs that might be able to do so. Elemento was familiar with the Broad Institute's Connectivity Map, which provides information on how drugs can alter gene expression profiles. He reasoned that if he could find a drug that lowered expression of most of the genes a particular transcription factor interacted with, it might be an effective inhibitor.

Elemento's lab created a systematic way to do this called CRAFTT, short for computational drug-repositioning approach for targeting transcription factors. In a proof-of-concept study, they recapitulated the finding that the experimental molecule JQ1, a bromodomain inhibitor discovered by Jay Bradner's lab at Dana-Farber Cancer Institute, inhibited MYC.

They then turned their attention to ERG. To do so, the researchers used ChIP-seq data from a previous study to perform a genome-wide analysis of ERG targets. This was the only wet-lab work performed for the study, but in the study the researchers successfully used ChIP-seq data from the ENCODE consortium to find interactions between drugs and transcription factors. Everything else happened with the help of computers.

The next step was finding candidate drugs that appeared to inhibit the genes ERG interacted with. In this step, the detail that matters most is the sheer number of genes that match in both data sets. "It doesn't' matter too much to us which mechanisms the drugs are using to block expression," Elemento said. "We can do experiments later to find out what the mechanisms are."

What's important is that it provides a list of drugs that can then be subjected to network analysis based on drug and protein interactions. Elemento said they constructed a network based on data from BioGrid.

"We looked for short paths between drugs and proteins," he said. "Shorter paths mean more direct interactions."

The network analysis revealed several drugs that might inhibit ERG, including naproxen, acemetacin, ondansetron, and epitiostanol; however, dexamethasone ranked the highest. But the drug had never been linked to ERG, so the researchers looked for a way to validate that it could affect prostate cancer in some way.

Cell invasion and migration assays showed that dexamethasone "significantly decreased cell invasion and migration in DU145 prostate cancer cells over expressing ERG but not in isogenic control cells," the authors wrote.

To go even further, they conceived a retrospective study to look at patients who had received the drug for other indications and whether they developed prostate cancer at slower rates than controls. Partnering with Columbia University Medical Center, they found an age-adjusted cohort of males 40 to 80 years old who had been diagnosed with prostate cancer.

There was a clear negative correlation between being on the drug and dying of prostate cancer, Elemento said. "The longer they were on the drug, the higher their survival probability was, compared to control drugs or commonly prescribed drugs." Even drugs in the same class as dexamethasone, such as prednisone, didn't seem to offer the same protection, he said, suggesting the interaction with ERG was very specific.

Elemento acknowledged that the EMR part was merely a correlation, but he said it was further proof that the CRAFFT approach works.

"It's a clever way to use the data from Broad to specifically target the function of transcription factors," he said.