Interactome Yields Data, But Is It Significant?
An investment of $100 million should be enough to correlate the genome with function, and identify new basic research and drug targets
The first human interactome maps, published just last year, offer literally thousands of starting points for further research.1,2 Yet subsequent analyses also demonstrate just how difficult it can be to convert interaction data into reliable hypotheses.
Both Marc Vidal and Erich Wanker, who published his map first, used their data to identify novel interactions. For example, Wanker identified new partners for the protein emerin, a mutation that leads to Emery-Dreifuss muscular dystrophy, as well as novel modulators of the Wnt signaling pathway. Vidal identified 424 interaction pairs in which at least one member had previously been implicated in human disease, including a new interaction between RTN4, a neurite outgrowth inhibitor, and SPG21 (spastic paraplegia 21), a mutation that leads to Mast syndrome.
In January Peter Uetz, Yu-An Dong, and colleagues reported that they had used the two interactomes to chart the interaction of herpesviral proteins with their human hosts.3 The team first mapped the internal protein-protein interactions for two herpesviruses, Kaposi sarcoma-associated herpesvirus (KSHV) and varicella-zoster virus (VZV), identifying 123 and 173 interactions, respectively.
They then mapped those interactions onto a "prototypical" human protein interaction network created by expanding the human interactome network with orthologous interactions from Caenorhabditis elegans, Saccharomyces cerevisiae, and Drosophila melanogaster, predicting 20 KSHV/host and 28 VZV/host interactions. The team independently verified 14 of 19 tested KSHV/host interactions but offered no data to support the VZV predictions.
Roselyn Eisenberg, a herpesvirus researcher at the University of Pennsylvania School of Veterinary Medicine, says observed virus-virus and host-virus interactions "would give me a starting point to looking at [a given protein's] function," but adds, "It's important to verify that this is actually happening. Not just can it happen, but does it, and is it significant?" Jay Nelson, who studies herpesviruses at Oregon Health and Sciences University, is even less sanguine: "To find a drug target using this data would be a miracle."
The published datasets also suggest there is room for improvement in the methodology. Akhilesh Pandey's ongoing analysis of the two datasets - Wanker's 3,186 interactions between 1,705 proteins from a human fetal brain expression library, and Vidal's 2,754 interactions between 1,549 proteins from the Mammalian Gene Collection - reveals that they share fewer than 20 interactions, with just 278 proteins in common. Among those 278 proteins, Vidal's group picked up 792 interactions that Wanker missed, while Wanker identified 1,186 that Vidal missed-this despite the fact that both used the same experimental method. Though he stresses that this is preliminary data, Pandey, who directs the Human Protein Reference Database (HPRD) says, "When A and B are used by two groups, and one group found A and B to interact and the other didn't, now you wonder."
Joel Bader, an interactome researcher at Johns Hopkins University, and colleagues at Curagen have an interactome dataset of their own that could help plug some of these gaps. With more than 30,000 interactions, at least 5,000 of which are high confidence, Bader's dataset exceeds Vidal and Wanker's combined.
Pandey's recently published analysis comparing the yeast, fly, worm, and human literature interactomes suggests other interactome assumptions could be incorrect, too, including the assertion that highly connected proteins are necessarily essential.4 But some conventional wisdom does stand up to scrutiny, including the fact that interacting proteins frequently share function.
The number of "gold standard" interactions - that is, those that are identified piecemeal in the literature, as opposed to high-throughput efforts - stands at around 33,700, according to the HPRD. That's just a tiny fraction of the total predicted interactome, which could run into the hundreds of millions of interactions. "I think it's so far from completion that it's hard to estimate even what fraction we know," says Bader.
1. U. Stelzl et al., "A human protein-protein interaction network: A resource for annotating the proteome," Cell, 122:957-68, Sept. 23, 2005.
2. J.-F. Rual et al., "Towards a proteome-scale map of the human protein-protein interaction network," Nature, 437:1173-8, Oct. 20, 2005.
3. P. Uetz et al., "Herpesviral protein networks and their interaction with the human proteome," Science, 311:239-42, Jan. 13, 2006.