The majority of protein complexes in higher eukaryotes contain one or more large proteins; they usually contain multiple protein–protein interaction domains and, thus, serve as scaffolds for the formation of multiprotein complexes. Owing to the problems inherent in isolation and purification of long full-length cDNAs, it has been difficult to undertake a systematic yeast two-hybrid analysis of large proteins. This has resulted in a relative paucity of data on protein–protein interaction in larger proteins.Kazusa DNA Research Institute (http://www.kazusa.or.jp/huge) has generated almost 2000 large (>4kb) human cDNA clones, which are available as a public resource – these are referred to as KIAA clones. Now, Nakayama et al. have used the cytoplasmic domains of 71 large transmembrane proteins, which have been encoded by KIAA clones, to systematically identify their interaction partners using a yeast two-hybrid screen [1xProtein–protein interactions between large proteins: two-hybrid screening using a functionally classified library composed of long cDNAs. Nakayama, M. et al. Genome Res. 2002; 12: 1773–1784CrossRef | Scopus (67)See all References][1]. Their ‘prey’ library was constructed from 1087 KIAA cDNA clones and specifically excluded nuclear and other transmembrane proteins because they were unlikely to interact with the chosen ‘bait’. Random fragments were then generated from the library and cloned into a prey vector using the Gateway™ system. A total of 84 protein-pairs were identified from this screen, and included several novel interacting proteins. The authors observed that the proteins that interacted with cytoplasmic domains of large transmembrane proteins often contained distinctive domains that are known to be localized to the submembrane region, such as PDZ, SH3, spectrin or Pleckstrin homology domains. The interactions discovered by this study are probably involved in organization of signaling complexes, receptor and channel clustering, and other similar cellular functions. A high false-positive rate, common to yeast two-hybrid studies, was avoided by eliminating combinations with very low probabilities of interactions, based on improper localization and function.Although the interactions reported by this study will have to be confirmed by other tests, such as in vitro binding assays, coimmunoprecipitation and colocalization, it is evident from the strategy that a large fraction are likely to represent real interactions. Studies such as this are necessary for filling the gap in our knowledge about protein pathways for the less studied protein classes. Nakayama and colleagues have nicely demonstrated how a high-throughput method, such as the yeast two-hybrid technique, can be used to specifically address the complexities of a subproteome.