In 2012, Boston Children’s Hospital held the international CLARITY Challenge—an invitation to interpret genomic sequence data from three children with rare diseases and provide a meaningful, actionable report for clinicians and families. (Click for more background on the children, findings and winners.)
The full proceedings, published March 25 in Genome Biology, concluded that while the technical approaches were markedly similar from center to center, the costs, efficiency and scalability were not. Most variable, and most in need of future work, was the quality of the clinical reporting and patient consenting process. The exercise also underscored the need for medical expertise to bring meaning to the genomic data.
That was CLARITY 1. CLARITY 2, focusing on cancer genomics in children, promises to be exponentially more complex.
Led by Boston Children’s and the Center for Biomedical Informatics at Harvard Medical School (HMS—in collaboration with the Dana-Farber/Boston Children’s Cancer and Blood Disorders Center and Massachusetts General Hospital—CLARITY 2 kicked off on March 13 with an academic conference of experts to establish the challenge’s parameters.
Sorting out the genetics of a patient’s cancer won’t be straightforward. To start with, not all cancers are the same, even when they look the same under the microscope: some are slow-growing and some are highly aggressive, as has been seen in brain tumors, for example. This variation is largely driven by genetics.
Secondly, there are at least two genomes to be sequenced—that of the tumor itself (“somatic sequencing”) and that of the child, who may carry one or more genes that influence her susceptibility to cancer (“germline sequencing”). Many cancers involve mutations in both genomes.
A moving target
Another challenge: When and how should the tumor be sampled? At last month’s conference, Gad Getz, PhD, director of the Cancer Genome Computational Analysis group at the Broad Institute, pointed out that a tumor is made up of different populations of cells, each potentially having a different genomic profile. Moreover, tumor genomes aren’t static: they change as the tumor mutates and acquires resistance to chemotherapy.
“You also need to look at the trajectory,” Getz advised. “One subclone may become more predominant over time.” Peter Park, PhD, of the HMS Center for Biomedical Informatics, referred to this kind of information as “metadata,” akin to the information collected by AT&T about patterns of phone calling.
Then there’s the challenge of establishing that a discovered genetic variant is actually involved in the cancer. And if a variant has a very low frequency in the population, you might need thousands of samples to reasonably expect to detect it.
Germline mutations (those in the patient’s own genome) are no less complicated. As Judy Garber, MD, MPH, of the Dana-Farber Cancer Institute pointed out, even relatively well understood mutations like those to BRCA1 and BRCA2 (implicated in breast and ovarian cancer) confer very different levels of risk that remain difficult to gauge. Additional “modifier genes” can factor in the equation. The risk posed by a mutation is unknown for many genes, and a mutation’s effects can vary widely within a single family.
The genetic architecture of cancer
Invited audience members—some of whom may enter this year’s challenge—wanted to know what kinds of data will be provided. Whole genome sequences? Whole exome sequences? RNA sequences (indicating gene expression)? Epigenetic markers (indicating modifications to the chromosomes by environmental factors)? What about mitochondrial DNA?
Stephen Chanock, MD, of the National Cancer Institute, described the “genetic architecture” underlying cancer susceptibility. There are many ways to discover genetic variants, but some will have larger effect sizes than others. Genetic linkage analysis—traditional family pedigree studies—yields genes with the largest effect size. Population-based, genome-wide association studies (GWAS) identify hard-to-spot genetic variants with a low effect size, contributing to susceptibility along with many other genes. Genes found through next-generation genomic sequencing fall somewhere in the middle, Chanock said.
Also to be considered, panelists noted, are mosiacism—mutations occurring in just a portion of a person’s cells—and environmental factors. In bladder cancer, for example, susceptibility mutations have an effect only in smokers. Mosaicism is seen in certain genetic syndromes that predispose children to cancer.
Getting the best information
So, how to design the challenge? The working idea was to look at childhood cancers, specifically solid tumors in children who have relapsed, or that are treatment-resistant. Audience members had many interesting suggestions—for example, why not sample twice, once at the start of treatment and again later? Some cancers are more genetically complex than others, attendees noted.
Next, what should be included in the report? There seemed to be consensus that to be useful to an oncologist, the report should discuss and rank the various therapeutic options suggested by the genetic analysis.
Finally, what sort of time limit should be imposed? Some attendees argued that in the context of a relapsed cancer, you’d want to have results within a week or two to guide further therapy. Others suggested allowing more time—assuming that “industry can figure out ways to speed it up”—or having teams take both a fast approach and a more deliberate one. “Sometimes, we need to pause, rather than do harm by giving indeterminate information,” Chanock noted.
The CLARITY 2 organizers are now digesting all this input. Watch this space for more on the challenge.