By Kevin Davies
April 28, 2009 | BOSTON – Eric Schadt, the Rosetta/Merck biomathematician who is in the process of setting up a new non-profit organization called Sage, gave a powerful testament in his keynote address to the need to focus on non-linear biological networks, rather than linear pathways, as the future of disease research and drug discovery.
Schadt, dressed in his standard white polo shirt and khaki shorts, opened by addressing the data tidal wave posed by the surge in sequencing data. The scales of storage needed are vast, he said. In 2000, only four genome centers in the US could generate terabytes of sequence data. Today, virtually any lab can generate that volume in a few days. Data storage is relatively easy, but moving the data around is a bigger problem. “Network speeds have not scaled to Moore’s law,” said Schadt.
In 2007, genome-wide association studies (GWAS) were “all the rage,” rapidly producing more than 200 replicated loci for a host of complex diseases. But Schadt’s team took one of those candidate genes – ERBB3 in type 1 diabetes (T1D) – identified by major GWAS in 2007, and asked: what was the functional support for these gene identifications?
“Is [the rmarker] a surrogate or is it the causal mutation? Does the variation activate or inactivate the gene? Can the gene be inferred from these studies? This can take years to figure out,” said Schadt, pointing to the example of Apolipoprotein E and Alzheimer’s disease.
To dissect these diseases, Schadt said, “we need to leverage all the biological data… Our brains appear to be wired for storytelling, not statistical uncertainty. The truth is, we have little idea on the underlying causes of disease.”
Pathways Out, Networks In.
Schadt argued that there is no such thing as a tidy linear pathway, only networks. That puts a premium on generating large-scale expression data, filtering GWAS data on expressed single-nucleotide polymorphisms (eSNPs) and causal networks to identify candidate genes and produce predictive network models. In this way, Schadt’s team identified a gene called RPS26 as the T1D susceptibility gene first mapped in the 2007 GWAS. The expression of RPS26 showed the greatest association of genes in the region with the variant, and KEGG analysis showed the major pathway enriched was indeed type 1 diabetes.
Schadt praised Microsoft’s Amalga Life Sciences as one of the significant efforts tackling this issue, and that Gene Network Sciences (Schadt is an advisor) is a good example of integrating data and building models.
Gene expression patterns do a good job of refining clinical phenotypes, as Schadt showed himself several years ago in collaborating with the Netherlands Cancer Institute to develop gene signatures for signatures for breast cancer recurrence risk. “Can we go beyond correlation to find causal genes?” he said. He expressed incredulity that, a few years ago, “drug companies made $100-million bets based [solely] on gene expression correlations.”
Schadt’s major emphasis of late has been modeling causal relationships between and within different tissues. Ongoing work looking at gene expression data from three mouse tissues – hypothalamus, liver and adipose tissue – is used to build networks reflecting interactions between those tissues. In this way, Schadt’s team has found that highly interconnected sub-networks are connected between tissues. Cross-tissue networks are specific to those interactions, and would be invisible if one was focusing on any single tissue. “You’re missing those networks that are facilitating connection between tissues and disease processes,” said Schadt. “We want to get away from the single gene view of diseases, and view these sub-networks.”
Having identified DNA variants on chromosome 17, Schadt used RNA interference to target those sub-networks. “We punch genes in the liver network and see if that induces changes in the islet network,” he said. From there, his colleagues can design drug candidates targeting the network node. Ongoing work with a specific drug candidate renders mice immune to becoming obese despite being on a high-fat diet. That drug is now moving into the clinic.
Schadt is transitioning out of Merck this summer to build a non-profit group called Sage, where much of this data-intensive focus on biological networks – the layer between DNA and disease -- will continue to grow. “It’s the network that’s sensing the variation and driving disease,” said Schadt.