Walter Jessen

Pharma ranks biomarker identification, diagnostic test dev highest scientific challenge for personalized medicine http://csdd.tufts.edu/news/complete_story/pr_ir_may_june_2015

Walter Jessen

Personal Genome Diagnostics Study Highlights Limitations of Tumor-only Sequencing for Cancer http://www.biomarkercommons.org/biomarker-news/personal-genome-diagnostics-study-highlights-limitations-of-tumor-only-sequencing-for-cancer/

Walter Jessen

Identification of Candidate Biomarkers for PD0325901 to Treat NF1

3 min read

In June, I was invited to give a presentation at a systems biology symposium called "Applications of Systems Biology Approaches in Drug Discovery & Development.";

My presentation (slides available here) is titled "Network Modeling of Protein-protein Interactions to Identify and Prioritize Candidate Biomarkers." Since all the client work I do at Covance is confidential, for the presentation I created a pseudo-client project. The goal of the project was to identify and prioritize pharmacodynamic (PD) biomarkers for an experimental cancer drug.

The project is based off of research I published last year in the Journal of Clinical Investigation on preclinical trials with a MEK inhibitor to treat Neurofibromatosis (NF1). Since this is real data and could be potentially useful in the development of patient therapeutics, I wanted to highlight it here.

I used a linear model to identify biomarkers from MEK1/2 (the target of PD0325901), through NF1-associated genes (to contextualize the model), to SPRY4 and DUSP6 (the markers used in the JCI paper).

Since this is a shortest-path problem, I used the MetaCore algorithm “shortest-path,” which is based on Dijkstra's algorithm. The maximum number of steps in the path (e.g. the maximum number of genes/proteins that can be added to the network between seeds) was set to three. To contextualize the model, I used GATACA at Cincinnati Children's Hospital to identify 33 NF1-associated genes.

The model was further contextualized using interactions specific to those in peripheral nerves. Since I'm looking for PD candidates, biomarkers closer to the drug target should have greater sensitivity to network perturbation than markers further away.

The model (below) is comprised of 404 nodes and 1,087 interactions. I identified 16 first-degree proteins that were 1-step downstream of MEK1/2, and 115 second-degree proteins that were 2-steps downstream. Between these two sets, there were 15 biomarker candidates. Those candidate biomarkers that are second-degree proteins downstream of MEK1/2 can be further prioritized by the number of interactions upstream to first-degree proteins.

The complete list of candidate biomarkers is shown below. CLS assay indicates that an established, validated assay exists for the analyte at Covance. In addition to biomarkers for which Covance offers an assay, I also identified secreted proteins that could be potential blood-based markers.

The top biomarker candidate is APOE -- not only are there three interactions back to first-degree proteins downstream of MEK1/2 (thus making it potentially very sensitive to network perturbation), but APOE is a secreted protein and a Covance assay is already established. APOE was also one of 13 genes linked to altered transcriptional regulation of Raf/MEK/ERK signaling in the JCI paper (heatmap below).