One of the main reasons that promising drugs fail to live up to their potential for ameliorating disease is unforeseen toxicity. The ability to monitor, and diagnostically predict which compounds may engender toxicities from a large lead compound portfolio, will have a dramatic impact on pharmaceutical development and the FDA's ability to monitor and ensure the safety of new drugs or new combination therapies. In the past, two-dimensional polyacrylamide gel electrophoresis (2D-PAGE) has been the major technology utilized by experimentalists to separate and analyze proteomes. This technology, although a powerful method for characterizing thousands of proteins from any given sample, is low-throughput, insensitive, and has limitations with separating low molecular weight and highly basic proteins. Our laboratory accepted the challenge of developing novel methods and tools to utilize proteomic information in the discovery of protein biomarkers of toxicity. Of critical importance, our laboratory invented (patent-pending) and developed, through a collaboration with Correlogic Systems, Inc., a heuristic pattern recognition algorithm which can import the complex data streams produced by SELDI and rapidly identify (in up to 20th-dimensional space) patterns of proteins that are diagnostic for human cancer in serum or cell lysates. A component of our research efforts focuses on the applicability of proteomic technology to identification of changes in protein patterns that could predict drug toxicity. Through a collaboration with Dr. Frank Sistare's group at the center for Drug Evaluation and Research (CDER), Dr. Petricoin obtained sera from animals treated with different drugs, with known toxicities, as well as sera from control animals. These sera were analyzed by SELDI using 1 ul of serum and each sample was analyzed in quintuplicates. Protein profiles were generated in less than 2 weeks. The SELDI profiles were then normalized and analyzed by the novel heuristic cluster algorithm, developed by the Petricoin laboratory, to identify protein patterns that were diagnostic for toxicity. The results of the analysis were remarkably clear and intriguing. Importantly, this tool was able to identify drug-induced toxicity over the vehicle alone background 100% of the time in the blinded test set. Critically, the proteomic profile proved to be more sensitive than histopathology or ELISA analysis in detecting toxicity at lower doses and at consistently earlier time points. Each of these profiles appears to be drug-specific, although the data are currently being tested to assess whether a global toxicity pattern can be uncovered.
