literature tells us” is relevant to patients, says Rollins. Every two quarters Rollins and others review which genes to swap in or out.
To add to the complexity, Project GENIE will have a second layer of information, drawn from the non-genomic health records of the patients whose tumor sequences are in the pool. That information would include the type of cancer, plus the patient’s age, gender, location, other diseases, and general health history—what researchers often lump together into the catch-all term “phenotype.”
Those health records will not be stored centrally, so researchers will have to request data sets for employees at each of the seven institutions to pull together. That part of GENIE will be slower to build, says Sawyers. “It’s not easy to retrieve that data from medical records. It’s a manual process that costs money, so at first it has to be a focused extraction of the most important elements,” he says. “We envision over time there will be software to extract data at scale, but right now that does not exist.”
The consortium is also going to be more protective of that extra layer—rich context for the genomic data—and giving its members priority to pursue questions that Sawyers calls “super compelling.” Sawyers gives the example of the BRAF gene. Mutations to the gene drive certain cancers, and the Roche drug vemurafinib (Velboraf) does quite well treating people with a dire form of skin cancer driven by a particular BRAF mutation called V600e, the most common. But a small portion of BRAF mutations are not V600e, which “we all know about but we don’t know what to do about,” Sawyers says. “We know the answer to one or two variants, but only based on a small sample of patients. The field would benefit from a larger analysis across seven centers.”
When asked what project he would initiate on GENIE if he could start tomorrow, Velculescu had a practical answer. He would expand a recent Johns Hopkins study that explored why the cancer drug cetuximab (Erbitux) fails to help a portion of people with late stage colon cancer. Cetuximab blocks the protein EGFR that, when mutated, puts cancer cells into overdrive.
In that study, researchers sequenced the exome—the small portion of the genome that codes for proteins—of the tumors of more than 100 people, then transplanted and grew those peoples’ tumors in mice. Combining genomic and phenotypic information, the study found new mutations in six genes that could be driving drug resistance. But studying the question in thousands of cases, says Velculescu, could lead to more solid clues for new treatments.
GENIE came together under the roof of the American Association for Cancer Research, which is funding it with $2 million for two years, but who pays for it after that is up in the air, says Sawyers. (Sawyers was recently AACR president, and Velculescu is on the board of directors.) In addition to philanthropic funding, one potential source of future support is drug companies who want custom data sets for their own research, the project’s leaders say. (For-profit users of the system would still have to publish their results.)
But the data will eventually be free. To appease institutes and researchers who demanded some reward for their own work, originators will have six months of exclusive use of their own data before the rest of the consortium gains access. After six more months, the data will be open to all comers—even for-profit groups, which means companies like 23andMe, Helix, Craig Venter’s Human Longevity, and Invitae, all looking to build treasure chests of health data, could conceivably also tap in.
One limitation of the data set at first is that it will include information about mutations that occur in cancer cells, but not inherited genetic data. While plenty of insight about cancer can be gleaned strictly from the changes that take place in cancer cells, genes inherited from previous generations can play a critical role in a person’s cancer risks. Sometimes those correlations are straightforward, which is why women with certain mutations of the BRCA genes might opt for preventive measures—such as the double mastectomy actress Angelina Jolie opted for in 2013—even if they have not yet shown signs of cancer. But researchers say our understanding of those correlations is growing more complex, making the hereditary layer of information increasingly relevant in diagnosis and treatment.
Adding that layer of information to GENIE will have to come later, says Velucescu. “We didn’t want the perfect to be the enemy of the good. We wanted to get this off the ground sooner rather than later.”
