A substantial proportion of the genes in any genome has no known function. In yeast, this 'function gap' is being tackled using several genome-wide approaches, and Ito et al. have now published the results of a systematic two-hybrid screen, which provides a large collection of new data on protein–protein interactions. A similar screen was published by Uetz et al. last year, and a comparison of the two data sets allows Ito et al. to conclude that this is a valuable approach for identifying novel gene functions, but that the individual studies are far from complete.

The two-hydrid screen involves two plasmid constructs that each express a fusion protein — between test protein X and the yeast Gal4 DNA-binding domain, and between test protein Y and the Gal4 activation domain. A reporter gene that is regulated by Gal4 will be activated if proteins X and Y interact. This provides the basis for a simple genetic screen for interactions between any two gene products in the yeast genome. Ito et al. pooled sets of constructs, such that each pool expressed either 96 DNA-binding fusion proteins or 96 activation-domain fusion proteins. By using all pairwise combinations of the pools, they screened the entire set of ∼6,000 yeast open reading frames (ORFs) for protein–protein interactions.

Two-hybrid screens are subject to false-positive errors, so to ensure that the results of the screens reflect real protein interactions, stringent criteria are applied to the data. For example, Ito et al. scored an interaction as positive only if it was identified on at least three independent occasions. Overall, they found such evidence for ∼800 interactions and, when compared to the ∼700 interactions identified by Uetz et al., there were only 141 interactions in common between the two screens. The clear implication is that neither study has identified all yeast protein interactions, perhaps because of the different strategies used to reduce the false-positive error rate.

There is also little doubt that these studies provide valuable functional information about genes of unknown function, potential protein complexes and novel links between different cellular functions. Ultimately, this information will be summarized in a comprehensive protein interaction network for the cell. The relatively small overlap between the first two systematic protein-interaction studies and the new functional insights that these data provide should encourage similar efforts using variations on this and other proteomic techniques.