If two proteins interact with one another, they usually participate in the same, or related, cellular functions. Yeast geneticists have a clever way of seeing whether two proteins can physically associate. They attach each of them to separate fragments of a third protein, called a transcriptional activator, which has the ability to switch on genes. If the two proteins interact, then the two fragments of the activator are reunited and a gene is switched on that produces an easily monitored colour change in the yeast cells. Because it is two hybrid proteins that are actually interacting, this method is called the two-hybrid system.
The availability of the complete genome sequence of the yeast Saccharomyces cerevisiae1 raises the possibility of exploiting the two-hybrid approach to identify all possible pairwise interactions between yeast's 6,000 or so proteins. Although such an exhaustive search has been much talked about2, so far only individual yeast protein complexes have been analysed in this way3, 4. Now, a collaborative group from the University of Washington, Seattle, and CuraGen, a biotechnology company, has taken on the daunting task of making a comprehensive two-hybrid analysis of protein interactions in yeast. The group concerned, Uetz et al., report their results on page 623 of this issue5, and readers may manipulate the data themselves at a public web-site6.
Uncovering the functions of gene products predicted by the DNA sequences of entire genomes has become a sort of 'range on the 'omes'. Functional genomics involves a number of levels of investigation that have been named the genome, transcriptome, proteome and metabolome7 ( Fig. 1). What distinguishes the last three levels from that of the genome is that they are context-dependent. The entire complement of messenger RNA molecules, proteins or metabolites in a tissue, organ or organism varies with its physiological, pathological or developmental condition. Transcriptome analysis, using DNA microarray techniques to screen large numbers of genes for messenger RNA abundance8, is yielding huge amounts of data about gene function. But it is an indirect approach — messenger RNAs are transmitters of genetic information, not functional cellular entities.
Figure 1: Levels of gene-function analysis.
The work of Uetz et al.5, discussed here, applies two-hybrid analysis to the yeast proteome. Details of the methods used are to be found in the references cited; SAGE is serial analysis of gene expression.
High resolution image and legend (54K)This is not true of proteins and metabolites, but comprehensive analyses of the proteome and metabolome are more technically demanding. Moreover, 'classical' proteomics, involving the techniques of two-dimensional gel electrophoresis (for protein separation) and peptide mass fingerprinting (for protein identification)9, 10, discards information of immense value in making functional assignments. This is because extracts are immediately placed in denaturing conditions, thereby destroying all protein–protein interactions. Discovering that a protein of unknown function interacts with one of known function provides a valuable clue to the role of the novel gene product, a concept that has been termed guilt-by-association. So a parallel approach to proteome analysis is to examine not the relative abundance of proteins, but their potential interactions with one another.
Two-hybrid analysis (Fig. 2, overleaf) works by separating
the coding sequences for the DNA-binding and activation domains of a transcriptional
activator and cloning them into separate vector molecules. The coding sequence
of a candidate protein whose partners are sought (known as a 'bait')
is then fused with the DNA-binding domain. A library of coding sequences for
proteins that might interact with the 'bait' (called 'prey')
is made in fusion with the activation domain. Yeast (like most sensible organisms)
has two sexes, called a and 
. Therefore, 'baits'
and 'prey' can easily be introduced into the same yeast cell by
mating. If they physically interact, the DNA-binding and activation domains
are closely juxtaposed and the reconstituted transcriptional activator can
mediate the switching-on of the gene that effects the colour change.
Figure 2: Detection of protein associations in yeast using the two-hybrid system.
a, Fusion of the 'bait' protein and the DNA-binding domain of the transcriptional activator cannot turn on the reporter gene. b, Likewise, fusion of the 'prey' protein and the activating region of the transcriptional activator is also insufficient to switch on the reporter gene. c, When 'bait' and 'prey' associate, however, the DNA-binding domain and activator region are brought close enough together to switch on the reporter gene. The result is gene transcription and a colour change that can be monitored.
High resolution image and legend (44K)Uetz et al.5 have used two different strategies in a comprehensive analysis of yeast protein–protein interactions. In array screening, 6,000 yeast colonies, each expressing a different 'prey' molecule, were distributed into microtitre trays. Strains expressing different 'bait' molecules (192 in all) were mated to each member of the array and the positive interactions were identified. One advantage of the array approach is that it rapidly becomes clear which locations produce false-positive interactions, providing reassurance that the system is working properly. In contrast, the library screening method does not keep cells expressing the 6,000 different 'prey' molecules separate. Instead, it pools them, and each of the 6,000 strains expressing a different 'bait' is mated with the pool. Hybrid cells are selected and then screened for positive interactions.
The two strategies yield different results. The array method is more efficient, which is only partly attributable to the judicious choice of 'baits'. The library approach, while benefiting from much higher throughput, has the disadvantage that cells in the 'prey' pool compete with one another during growth and mating, so selecting against cells expressing fusion products that retard either process. Thus, of the 12 'baits' that gave positive interactions with both screens, 48 possible partners were identified by the array approach, against only 14 in the library screen.
Uetz and colleagues' analysis5 has yielded discoveries of three types. First, interactions between proteins of known and unknown function have indeed allowed the role of the latter to be inferred. For instance, two proteins have been linked to the metabolism of arginine (the principal storage amino acid in yeast) because they can interact with ornithine aminotransferase (an enzyme required for arginine biosynthesis).
Second, hitherto unrecognized interactions have been identified between proteins involved in the same biological process. Yeast is an important model system for the study of the cell cycle in eukaryotes (organisms, including ourselves, whose cells have a nucleus; bacterial cells, for instance, don't). Uetz et al. have uncovered further interactions between cell-cycle regulators, thus extending our view of the subtlety and complexity of the control of cell growth and division. Specifically, it turns out that the cyclin-dependent kinase Cks1 interacts with each of three different B-type cyclins, implying that Cks1 may allow them to regulate the activity of the Cdc28 kinase that controls the start of yeast's cell cycle.
Finally, and perhaps most notably, the screen has provided clues for seeing how individual biological events are integrated into larger cellular processes. Meiotic recombination is the process that produces new gene combinations during the production of sperm and eggs, and ensures, for instance, that each of us is a unique individual. It involves the exchange of DNA segments between chromosomes inherited from each parent in a process called crossing-over. This takes place in a structure known as the synaptonemal complex and involves the breakage and rejoining of DNA molecules. The subsequent separation of each chromosome pair (segregation) requires the action of the filaments of a structure termed the meiotic spindle.
Uetz et al. found that Msh5 (a yeast protein required to resolve DNA cross-overs) is able to interact with two proteins implicated, respectively, in the formation of double-strand DNA breaks and the two ends of the meiotic spindle. One of these proteins, in turn, interacts with another that is required for generation of the synaptonemal complex. Thus a network of interactions is revealed that provides an overview of the mechanisms involved in meiotic recombination.
Although comprehensive two-hybrid analysis has the integrative power of all good functional genomics techniques, it lacks the context dependency of classical proteome analysis. It reveals potential protein interactions, but not the biological context in which they happen. Some may occur only when yeast is in a particular physiological state; others may never occur because, in real life, the proteins are located in separate cellular compartments.
New approaches to classical proteomics are required, which exploit separation techniques that do not destroy protein–protein interactions, and which involve methods of mass spectrometry11, 12 and bioinformatics that allow the unambiguous identification of all members of a protein mixture. Such a biochemical approach will complement comprehensive two-hybrid analysis. Together, the two will be a powerful way to discover the functions of newly identified proteins and integrate them into a comprehensive view of the workings of a living cell.
