June 8, 2011 | SAN DIEGO—In one of three outstanding keynote presentations at CHI’s X-Gen Congress*, Stanford University physician scientist Atul Butte offered some compelling ideas around the medical utility of genome sequencing.
Butte’s research centers around how the increasing prevalence of personal genome data can actually be interpreted and incorporated into routine medical practice. He was one of the lead authors in a landmark study published in the Lancet in 2010 that provided comprehensive clinical annotation of the genome of his Stanford colleague, Steve Quake. A doctor reviewing Quake’s medical history—40-something male with no symptoms or medications but a family history of aortic aneurysm/sudden death—would be faced with 2.8 million single nucleotide polymorphisms (SNPs) and more than 750 copy number variants. “Oh, and by the way, your next patient is here in 15 minutes,” Butte said.
The Stanford team studying Quake’s genome found rare “private” variants in three genes associated with sudden cardiac death—TMEM43, DSP, and MYBPFC3. “We need better tools to figure out what to do with these rare variants,” said Butte. Another gene variant was consistent with a family history of coronary artery disease, but Butte said he didn’t think Quake had started taking the recommended statin. “We still haven’t found the compliance gene,” he joked.
While Butte’s Stanford colleague Russ Altman analyzed pharmacogenetic markers, Butte’s lab was “tasked with all the rest of medicine.” But Butte said that existing SNP-disease databases were inadequate for a personal human genome (see “The State of Mutation Curation,” Bio•IT World Mar 2011). So Butte’s group has built a new SNP-disease database, in collaboration with India’s Optra Systems. After three years, the database has some 115,000 records based on the curation of more than 4,000 papers, and about 25,000 SNPs linked to 1,166 diseases. The database includes information on odds ratio, P values, platform technology, genetic models, and much more. However, Butte has not yet received public funding to support the initiative.
Another barrier to the uptake of genomic data in medicine according to Butte is that physicians are not necessarily trained to think in terms of odds ratios. “Generations of physicians have been taught likelihood ratios,” said Butte. “You can’t pass the Boards without training in likelihood ratios.” The distinction is a subtle one, but there are advantages to presenting data in terms of likelihood ratios, including the fact that it is possible to chain tests together.
As doctors routinely calculate the likelihood probability of disease (from 0-100%), Butte’s idea was to compute the LRs for all publications that mention SNPs. Obtaining figures for the pre-test probability for various diseases is not straightforward, but work by Rong Chen and Alex Morgan produced a Risk-o-Gram. “We start with the prior probability and add back SNPs. This is how doctors talk to each other,” said Butte. Diseases are ranked in order of the post-test risk to the individual.
Butte’s team is also trying to “mine the hell out of Medline.” By screening the subheads (MeSH qualifiers) associated with relevant papers, environmental links and therapeutic options can be identified. Butte stressed the need to examine environmental factors in complex disease—the “human exposome” as he calls it. In a 2010 study published in PLoS ONE, Butte, Chirag Patel and colleagues conducted the first environment-wide association study (EWAS), ingeniously plotting the significance of environmental risk factors in a Manhattan plot, a hallmark in genomewide association studies (Patel C. et al. PLoS ONE 2010).
From Patient to Population
Opening keynote Hugh Rienhoff, who has waged an inspiring search to solve the mystery of his daughter Beatrice’s genetic disorder (see “Hugh Rienhoff’s voyage round his daughter’s DNA,” Bio•IT World Sept 2010), believes he has finally identified the cause of “Bea’s syndrome.” But he declined to identify the gene, other than to say it was involved—as he has long suspected—in TGF-beta signaling.
The latest candidate was identified by exome sequencing (performed by Illumina) on the entire Rienhoff family. One of two new mutations uncovered in this approach lies in a gene in the TGF-beta pathway. Rienhoff added that predictive modeling from the protein crystal structure suggests the mutation would have a big effect.
While confirmatory studies are required, Rienhoff feels confident in saying, “We’re a post-genomic family. For us, the genome is a thing of the past. But what we know about TGF signaling we could put in a thimble.” Rienhoff hopes that ongoing animal model studies could lead to “therapies that are actually rational.”
Rienhoff’s involvement in personal genomics “started with conception” but he says he got lucky by being able to compare his daughter’s syndromes with overlapping phenotypes.
He hopes to find other cases like his daughter, who has become the model patient. “She’d rather have her DNA examined than have a flu shot,” he joked. •