Toxicity is a notorious killer. It kills more "promising" compounds than almost anything else, and, even worse, usually does so late in the development process. It kills or harms many patients in the clinic because not enough is known about how different patients respond to different drugs. 
Iconix Pharmaceuticals' VP of chemogenomics, Kurt Jarnagin |
Iconix Pharmaceuticals won the 2004 Best Practices Grand Prize in Drug Discovery and Development from Bio·IT World for its promising work in helping curb this killer. Founded in 1988, Iconix has built DrugMatrix, a database of expression data consisting of more than 13,000 microarray experiments in which rats were treated with more than 400 FDA-approved drugs, 60 drugs approved in Europe and Japan, 25 withdrawn drugs, and 100 toxicants. These compounds have been profiled in up to seven different rat tissues (3,200 different dose-time-tissue combinations). Its scientists then developed a battery of sophisticated mining and classification techniques to construct a library of 211 genomic biomarkers called Drug Signatures.
"The size of the project was a challenge — to get enough information to start making sense out of it was a huge undertaking — and it has taken us several years," says Alan Roter, vice president of informatics, who has been with the company since its beginning.
Roter says technology development was an early challenge because no one had attempted such a project before. Laboratory information management systems (LIMS) for gene-expression labs were not widely available. Creating consistent protocols and automating the lab were high priorities. Data-mining tools required "novel algorithm development with support vector machines for identifying sets of biomarkers that indicate or predict a biological property," Roter says. "[We] put a huge effort into data processing and data mining. Some of them we are publishing."
Iconix validated its work with 29 blinded studies on marketed drugs or novel compounds. The only data available to Iconix scientists in these blinded experiments were the results of microarray experiments. Of the 29 studies, 19 were conducted with six major global pharmaceutical companies, including Abbott Laboratories and Bristol-Myers Squibb. In every one of the 19 studies, Iconix analysts successfully predicted the specific toxicities associated with the compounds. In all but one case, the mechanism of action was also correctly identified. (For the 19th compound, the mechanism of therapeutic action could not be elucidated.)
Predicting Toxicity
The company is providing QA/QC training for FDA personnel and assisting with efforts to establish standards for microarray use. Iconix says it has "devised methods to control the wide range of factors that can affect microarray data quality, including any conditions that may lead to unwanted changes in gene expression before RNA extraction, such as variations in the environment of the animals or cell cultures, and in post-RNA extraction steps such as sample processing protocols." "The key issue is building biomarkers that can predict toxicity and efficacy," Roter agrees. Most recently, Iconix concluded a study of renal tubular nephrosis, which can occur as a consequence of repeated exposure to a number of diverse drugs, and which is poorly predicted by current histopathological, clinical, and molecular markers. With a 35-gene biomarker for the disease, scientists used gene-expression techniques to predict subchronic renal tubule injury weeks earlier than typically required for the histologic appearance.
Currently, the primary application of Drug Signatures is the selection of the best candidates for advancement to the clinic in late stages of discovery. Iconix is now constructing an in vitro database using rat hepatocytes to move this technology upstream to the initial stages of lead generation, when Drug Signatures can be used to screen large numbers of compounds and compound classes. The earlier that discovery scientists can detect safety liabilities, the more productively they can focus their resources on those compounds with the greatest chance of success.