Data Analysis for Predictive Markers
Data analysis must be carefully designed, Dr. Raghavan observed, in order to come up with results that are reproducible and real. Scientists in J&JPRD’s mechanistic toxicology group looked for a small set of genes that can predict NGTC in compounds using 24-hour experiments. The experimental protocol called for high doses of known NGTC and non-NGTC drug compounds to be administered to rats. Subsequent extraction of mRNA from their liver tissue and analysis for gene expression changes were used as predictive markers of changes that might presage NGTC down the road.
The methodology—ensemble classification algorithm, based on the well-established algorithm, linear discriminant analysis—was used to generate gene-signatures to predict which class (NGTC or non-NGTC) a compound will fall into.
“Ensemble algorithms such as the well-known random forest,” Dr. Raghavan said, “were used to look for partitions where compounds from each class fall on one side or another.”
The ensemble approach is based on identifying several such partitions and aggregating the results, “which gives you a much more robust classifier,” she noted. Another main thrust of this methodology is the elimination of noisy genes, and a number of strategies developed in-house are used to select genes with high differential expression.
Regulatory agencies are interested in incorporating these screens into their regulations, Dr. Raghavan continued. “This is just a starting point, but it’s a good starting point. We need a lot of validation with vigilance about overfitting at every stage.”
Multiplex Biomarker Assays
Michael Pisano, Ph.D., president and CEO of NextGen Sciences, pointed out that NGS has developed a suite of biomarker mass spectrometry-based services that utilize proprietary methods to significantly decrease timelines and increase success rates traditionally associated with the various stages of biomarker development.
“We can move very rapidly through discovery to assay and into validation without spending lots of time and money,” Dr. Pisano stated.
NGS offers mass spectrometry-based biomarker services that utilize proprietary methods to significantly decrease timelines and increase the success rates traditionally associated with biomarker development. The suite of services, called biomarkerexpress™, includes discovery of protein biomarkers, development of protein biomarker assays, and testing of biological samples to validate putative protein biomarkers and/or determine levels in preclinical and clinical samples. NGS’ emphasis is on protein biomarker assay development and testing of levels in biological and clinical samples.
The company’s peptide multiple reaction monitoring (pMRM) assay is very specific for targeted proteins, Dr. Pisano notes, including post-translational modifications, isoforms and fragments, and provides absolute or relative quantification based on mass spectrometry. The technology directly measures the protein, not the reaction of antigen and antibody, as done in ELISA and immunoassays. pMRM requires small sample sizes and is applicable to biofluids, tissues, cells, and formalin fixed paraffin embedded tissue.
The biomarkerexpress suite includes discoveryexpress™, assayexpress™, testingexpress™, and biomarker library™. The mass spec-based assay platform reduces development time. The multiplexing capability of the pMRM assay easily processes 30–50 proteins simultaneously and can go as high as 100 proteins.
Dr. Pisano commented, “We have disease specific and biofluid specific panels and assays available and continue to develop new panels and assays. The clients also have the ability to customize a panel or assay for their needs.”
He also presented a case study relevant to the discovery and assay development for putative biomarkers of lung cancer progression. The study performed quantitative protein profiling of the conditioned media obtained from A549 lung adenocarcinoma cells undergoing TGFalpha induced epithelial-mesenchymal transition (EMT), which results in changed cell morphology and acquisition of a migratory and invasive phenotype. Cells undergoing EMT would mimic circulating tumor cells or cells in the process of metastasis. Therefore, NGS researchers theorized that secreted proteins from these cells may represent protein secreted by tumor cells in the early stages of metastasis.
pMRM was used to measure all of the proteins on the panel in both conditioned media from A549 cells and subsequently plasma samples from patients with various stages of lung cancer.
A “relative quantitative assay” was developed and tested in a four-week timeframe. Preliminary data from a small patient population shows great potential for this technology, Dr. Pisano reported, with comparable results “from model to patients (sectretome to plasma)”.
Several proteins that were demonstrated to increase in the conditioned media also increased in plasma with progression of disease. Further work is going to verify and validate the panel for use in drug screening in vivo. Dr. Pisano explained that concurrently, testing across a larger population of patients will be performed.
Zhenhao Qi, Ph.D., senior principal scientist, translational sciences, Boehringer Ingelheim Pharmaceuticals, noted an increasing demand for mechanism of action (MOA) biomarkers in Phase I trials to demonstrate that the compound hits the target in vivo and allows better decision making at this early phase.
Biomarker discovery cannot be easily achieved by a single scientist within a single discipline, Dr. Qi observed. “For this program, we have involved scientists from genomics, statistics, molecular biology, bioinformatics/computational biology, therapeutic areas like immunology and inflammation, medical and clinical, and scientists from wet labs to dry labs.
“This core team worked together using a streamlined, iterative process: hypothesis/ experimental design, wet lab experiments, and statistical analysis followed by team discussion, conclusions and recommendations. Then we continued onto the next round of iteration.”
While a genome-wide search may yield ample number of genes that show significant response to stimuli of a biological pathway, this is not sufficient for those genes to be qualified as MOA biomarkers, Dr. Qi noted. To achieve this, the genes need to have functional relevance to the NFKB-IKKbeta pathway.
“Specifically, we need to have literature or publication support showing that those genes are regulated by NFKB (through transcription, gene expression etc.),” he said. “This is where literature/pathway knowledge-base and tools come into play. In addition, those genes’ response to stimuli should be inhibited by IKKbeta inhibitor.”