The yeast two-hybrid (Y2H) system can pump out vast quantities of data on a genome of any size. Although simple – it measures the ability of two proteins to find each other and "interact" in the milieu of a living yeast cell – the process revealed a dazzling volume of information about the networks in which specific proteins operate. Y2H assigns greater importance to those proteins that interact with many partners, and less to those with fewer partners. This raised the prospect of using Y2H to map the interactome. Researchers showed the approach worked by mapping a phage, and they honed the technique in the organism in which it is assayed, Saccharomyces cerevisiae. Soon after, Y2H was ready for the challenge of a metazoan.

This issue's Hot Papers offer the first interactome maps of Drosophila melanogaster and Caenorhabditis elegans, two of biology's most important model organisms. Marc Vidal and coworkers at Harvard University Medical School used 3,024 C. elegans proteins that they predicted would be present in the nematode to identify 4,027 interactions. Combining their data with previously described interactions and interologs predicted in silico, they produced a map containing 5,534 interactions.[1] This approach, argues Loic Giot at Curagen, eliminated the bias potentially introduced by another unvalidated high-throughput technology.

Jonathan Rothberg and collaborators at CuraGen in New Haven, Conn., along with researchers at Johns Hopkins, Wayne State, and Yale Universities, produced an interactome map of 7,048 Drosophila proteins and 20,405 interactions. They ranked the interactions depending on whether the relationship was already published and generally accepted as being correct (high confidence) or whether the interaction was unlikely in vivo (low confidence), such as an interaction between a nuclear and extracellular protein. This refined the draft into a map containing 4,679 proteins and 4,780 interactions.[2]

These were the first maps of their kind for metazoa. According to Stan Fields, Howard Hughes Medical Institute investigator at the University of Washington, this type of interactome cartography yields insights into basic biological processes, including transcription, signal transduction, and cell-cycle regulation. It also provides information about human orthologs that might illuminate disease pathology and define new therapeutic targets. Furthermore, the maps describe the organisms' global organization and help complete the picture of cellular activity. "Finally, the large-scale functional information about protein interactions can be integrated with transcriptional, phenotypic, and other analyses to yield hypotheses about numerous proteins that could not be generated by single studies," says Fields.

As the name suggests, Y2H uses two hybrid proteins. The "bait" is a protein of interest fused to a DNA-binding domain incapable of activating transcription. The "prey" is another protein fused to a transcription factor's activation domain. When haploid yeasts (one expressing bait, and the other prey) combine, an interaction between the two hybrid proteins reconstitutes a functional transcription factor, activating a reporter gene. The Y2H system allows researchers to detect previously unknown interactions in high throughput. The interactions are then scored using existing information and confirmed independently through such methods as coaffinity purification assays.

Fields and coworkers at State University of New York, Stony Brook, developed Y2H and used it to construct the first interactome map of the Escherichia coli bacteriophage T7 in 1996.[3] The map contained 25 interactions. Two maps of Saccharomyces followed in 2000.[4,5] The maps contained between 957 and 4,549 interactions, but few of the interactions overlapped. These maps proved that the methods worked, but moving to metazoa introduced a new level of complexity. Yeast has around 5,800 genes in roughly 12 million base pairs, and Drosophila has an estimated 14,000 genes in 165 million bases.

These were the first maps of their kind, which posed some challenges. Giot says that technically, the biggest challenge involved in moving from a single cellular organism to a metazoan was cloning 12,000 predicted ORFs from Drosophila cDNA rather than cloning from genomic DNA.

And coverage is scant. Vidal's worm map contains between 5% and 10% of the total number of protein-protein interactions expected to exist in C. elegans. For Drosophila, a recent paper from a group at Hybrigenics, a French biotech, found 2,300 protein-protein interactions, with little overlap with the earlier map.[6] "Like the earliest maps of the world that missed the existence of the Americas, these interactome maps are far from complete," says Barry Causier at the University of Leeds, UK.

These maps are incomplete partially because of Y2H's inherent limitations. For example, yeast genes can mutate during the numerous cell divisions that occur during an assay. Some autoactivating mutations allow the bait to activate transcription, leading to false-positive results. Furthermore, Takashi Ito of the University of Tokyo, who headed one of the teams that drew a Saccharomyces map,[5] notes that Y2H may not detect weak or rapid interactions that are still of "regulatory significance." Indeed, Y2H can reveal interactions that don't exist in vivo while not detecting some that do. "In other words, what fraction of the biological interactome was detected and what is noise?" asks Bertrand Seraphin, from the Center for Molecular Genetics at National Center for Scientific Research (CNRS) in France.

Because of these limitations, Causier says, "both studies made great efforts to ensure the biological significance of their data, using statistical, bioinformatics, and alternative experimental approaches." Franck Martin, from the Institute of Molecular and Cellular Biology in Strasbourg, France, also underscores the importance of confirming these interactions with basic biochemical and cellular bench work. "The real work starts after these papers," he says.

Confirming the interactions means working with dynamic, living cells. "Pathways are extremely fluid," Giot says. "Components move from one complex to another, adding layers of complexity in the overall regulation of intracellular communication." Ito points out that interactome maps provide neither spatiotemporal resolution nor quantitative descriptions. "What fraction of protein X participates in the interaction with protein Y?" he asks. Seraphin suggests that ideally researchers should determine the interactome for a specific cell type at a given time. The current maps show interactions independently of time. "What is the usefulness of an interaction if the two partners are never found simultaneously in the same cell?" he asks.

Vidal, however, draws an analogy with genome maps. "The genome sequence doesn't tell you which genes are expressed and when, but rather offers a framework for researchers," he says. "It's the same with the interactome map; it shows the basic network but not the functional consequences."

Nevertheless, the maps remain an intimidating mesh of sticks and balls. Indeed, there's a pressing need for new visualization tools to make these data more accessible to nonspecialists,[7] so that interactome cartography can open scientists' minds to other biological processes. "Sometimes scientists are so focused on the topic that they forget to explore other pathways," Martin says.

Interactome maps consider proteins as nodes connected by interactions or edges. The probability that a newly evolved protein will connect with a second protein is proportional to the latter's number of existing interactions. In other words, the linkage-rich proteins get richer.[2] This results in a few highly connected hubs and many less well-connected nodes on the periphery.[7] The C. elegans map, for example, contains 2,898 nodes and 5,460 edges.[1] "It's rather like air travel. The nodes are the airports and the edges the routes. The hubs are the main airports like O'Hare or JFK," Vidal says.

This model might explain why biological systems can be remarkably resistant to random failures but at the same time vulnerable to targeted attacks. An air traffic problem in Des Moines might have little impact on the overall network, whereas a problem at O'Hare could throw worldwide travel into disarray. "This helps understand how a biological system can at once be robust and sensitive to change," Vidal says.

The maps could also reveal new drug targets. Around three-quarters of human disease genes show strong matches to Drosophila sequences.[7] For example, in humans the gene BCL6 encodes a transcription factor that contributes to human B-cell non-Hodgkin lymphoma. In Drosophila, the BCL6 ortholog (CG1832) connects to two calcium-dependent calcineurin phosphatases and a calcium-binding protein. This raises the prospect that targeting calcineurin phosphatases could offer a new target for some cancers.[2] Furthermore, highly connected hubs could offer targets for drugs that need to produce widespread effects, an antibiotic for example. In contrast, targeting a peripheral node with few interactions might yield a highly specific agent with few side effects.

More fundamentally, the Hot Papers show that it is possible to map complex interactomes. Vidal is now in the midst of mapping the human interactome, and several initiatives are nearing completion. "The C. elegans study was a stepping stone to our ultimate goal."