As genomic scientists sequence their way to 10 petabytes of DNA data—a quantity The Cancer Genome Atlas project alone is predicted to eventually generate—one thing is becoming increasingly clear: Every cancer tumor is a moving target.
Eric Schadt, director of the Institute for Genomics and Multiscale Biology at Mount Sinai Hospital and a member of the New York Genome Center Executive Committee, says “amazing progress is being made in understanding different cancers—how they evolve, how they signal, et cetera—and new treatments continue to come into the marketplace.” But, he adds, “we continue to learn that there is great heterogeneity in most tumors, and that heterogeneity may increase as patients go on treatment and the tumor figures out ways to get around [it].”
Speaking at the Scripps Research Institute conference on The Future of Genomic Medicine last week, Schadt outlined a new Mount Sinai program that is leveraging next-generation sequencing and predictive network models to drive personalized cancer therapy.
“Genomic data is one way we can uncover what is happening specifically in a patient’s tumor, what key driver genes are mutated, what signaling pathways are altered, and then whether we can better match a given patient’s condition to a more appropriate treatment,” he says.
But genomic data is only part of the picture. Schadt and his colleagues are also performing patient-specific mutational analyses, applying those results to inform screens using patient-specific mutant fly models, among other model systems and in silico approaches. They’re then weaving all of the resulting DNA, RNA, CNV, and mutation information together to generate probabilistic causal networks in the hopes of identifying the perturbations that drive cancer.
The team has taken on seven cases to date, though more are in the pipeline. Schadt says that even in this small test sampling, tumor heterogeneity has been substantial. “In all cases we are finding things that are very interesting,” he says. “With the network informing on what the mutations mean in the patient, we can identify disruptions that will be missed by those doing targeted gene panels.”
For example, a targeted gene panel on a particular patient might show that key signaling genes, like PTEN or HER2, are not themselves mutated. However, Schadt says, “the networks around these genes may be mutated in ways that activate these signaling pathways.” Such insights can help researchers identify which therapies might be most appropriate for that patient.
In an interview following FOGM, Schadt drilled down into specifics, charting the course from initial pathology sampling and diagnosis to network model analysis and therapy predictions.
He says that with cancers for which ample public data exist, the model is more or less established. However, for less-studied or otherwise urgent cases—such as pancreatic cancer—the researchers perform a rapid sequencing run with the Illumina HiSeq 2500. For that, he says, getting from sample prep to sequence data takes around 30 hours. From there, the analysis runs about two to three days. “We are tuning that and think we can [eventually] get all analyses done in a day and ultimately in under an hour,” Schadt adds. Overall, he says, “from receiving DNA to getting analyzed results informed by the network models, it’s around four days.”
Robert West, who is not a part of this work, says predictive modeling for personalized cancer therapy seems a “logical next step” for researchers in the field.
Like many others, West, an associate professor of biochemistry and molecular biology at SUNY Upstate Medical Center, followed FOGM from afar. Discussing Schadt’s talk, West says “one thing is clear: tumor heterogeneity is a very big part of the puzzle.” And that, he adds, means scientists must study tumors longitudinally, taking multiple biopsies over time.
“Because the tumor is changing,” West says, “as you treat with chemotherapies of various sorts, biopsy, and sequence it, you come up with new targets.”