To appropriate a quote from John Donne, "no protein is an island entire of itself" — or at least, very few proteins are. Most seem to function within complicated cellular pathways, interacting with other proteins either in pairs or as components of larger complexes. A comprehensive understanding of these interactions will be needed before we can appreciate the mechanisms by which cellular pathways function and interlink. On pages 141 and 180 of this issue, Gavin et al.1 and Ho et al.2 describe significant advances towards this goal. Each group has characterized hundreds of distinct multiprotein complexes in the budding yeast Saccharomyces cerevisiae, using approaches in which individual proteins are tagged and used to pull down associated proteins, which are then analysed by mass spectrometry.
These studies1, 2 exemplify an emerging paradigm in protein biology: the systematic analysis of an organism's complete complement of proteins (its 'proteome'). Protein interactions on a proteome-wide scale have already been analysed in several ways. In a pair of landmark papers, Uetz et al.3 and Ito et al.4 adapted the yeast 'two-hybrid' assay — a means of assessing whether two single proteins interact — into a high-throughput method of mapping pair-wise protein interactions on a large scale. The authors collectively identified over 4,000 protein–protein interactions in S. cerevisiae. Our own group5 has developed a microarray technology in which purified, active proteins from almost the entire yeast proteome are printed onto a microscope slide at high density, such that thousands of protein interactions (and other protein functions) can be assayed simultaneously.
Gavin et al.1 and Ho et al.2 take a different approach — one that is particularly effective at identifying protein complexes that contain three or more components. Large-scale efforts to characterize protein complexes are generally rate-limited by the need for a nearly pure preparation of each complex. In the new studies1, 2, protein complexes were purified as follows (Fig. 1). First, the authors attached tags to hundreds of different proteins (to create 'bait' proteins). They then introduced DNA encoding these bait proteins into yeast cells, allowing the modified proteins to be expressed in the cells and to form physiological complexes with other proteins. Then, using the tag, each bait protein was pulled out, often fishing out the entire complex with it (hence the term 'bait'). The proteins extracted with the tagged bait were identified using standard mass-spectrometry methods.
Figure 1: Analysing protein interactions.
In the 'co-precipitation/mass spectrometry' approach used by Gavin et al.1 and Ho et al.2, an 'affinity tag' is first attached to a target protein (the 'bait'; a). b, Bait proteins are systematically precipitated, along with any associated proteins, on an 'affinity column'. c, Purified protein complexes are resolved by one-dimensional SDS–PAGE, a technique that involves running an electric charge through the complexes on a gel, so that proteins become separated according to mass. d, Proteins are excised from the gel, digested with the enzyme trypsin, and analysed by mass spectrometry. Database- search algorithms (bioinformatics) are then used to identify specific proteins from their mass spectra.
High resolution image and legend (85K)Applying this approach on a proteome-wide scale, Gavin et al.1 have identified 1,440 distinct proteins within 232 multiprotein complexes in yeast. As 91% of these complexes contain at least one protein of previously unknown function, the study provides a wealth of new information on 231 previously uncharacterized yeast proteins, and on a further 113 proteins to which the authors ascribe a previously unknown cellular role. Furthermore, Gavin et al. find that most of these complexes have a component in common with at least one other multiprotein assembly, suggesting a means of coordinating cellular functions into a higher-order network of interacting protein complexes.
An understanding of this high-order organization will undoubtedly offer insight into corresponding networks in other organisms, as most yeast complexes have counterparts in more complex species (one reason why researchers are interested in this unicellular organism). Gavin and colleagues illustrate this point by purifying and analysing three equivalent multiprotein complexes from yeast and human cells: the Arp2/3 complex, a component of the cellular 'skeleton'; the Ccr4–Not1 complex, which is found in the nucleus; and the TRAPP complex, which is involved in transport from one intracellular compartment (the endoplasmic reticulum) to another (the Golgi). In each case, the authors retrieved human and yeast complexes that were similar, if not identical, in composition.
Using the same general approach, Ho et al.2 constructed an initial set of 725 yeast bait proteins, from which they identified 3,617 interactions involving 1,578 different proteins. They describe interaction networks assembled around the protein kinase Kss1 — a known component of pathways involved in mating and filamentous growth — and complexes associated with the cyclin-dependent kinase Cdc28 and the gene-transcription factors Fkh1 and Fkh2. In addition, Ho and colleagues used 86 bait proteins that are implicated in the DNA-damage response, allowing them to delineate much of the yeast damage-response network. In particular, they reveal many regulators and targets of the protein kinase Dun1, and a possible role for the DNA-repair protein Rad7 in processes of targeted protein degradation.
The approach taken by Gavin et al. and Ho et al. is clearly powerful, but it does have drawbacks. Both groups find a significant number of false-positive interactions, while failing to identify many known associations. Gavin et al. estimate that 30% of the interactions they detect may be spurious, as inferred from duplicate analyses of 13 purified complexes. Conversely, they failed to detect any interacting partners for Bmh2 (ref. 6), a regulatory protein that has previously been shown to interact with a number of other proteins, including Ste20 (involved, for example, in yeast mating)7, and Msn2 and Msn4 (stress-responsive transcription factors)8. Ho et al., meanwhile, did not detect nucleotide excision repair factor-2, a tight complex9 that contains the well- characterized DNA-repair proteins Rad4 and Rad23. So, as in most large-scale studies, these results are imperfect. It will be essential to integrate data from many different sources to obtain an accurate understanding of protein networks.
Proteomic studies such as these1, 2 have generated a huge volume of exciting data. Yet — setting aside the problem of false positives and negatives — there is much still to be learned before we have a comprehensive knowledge of functional pathways within even a model organism such as yeast. To understand the magnitude of the task, consider the yeast proteome. Assuming that each protein interacts with an average of five partners — a reasonable estimate drawn from experience and preliminary two-hybrid results — the yeast proteome should encompass some 30,000 protein interactions, many of which change during the life cycle of the organism. So far, protein microarray analyses and studies like those of Gavin et al. and Ho et al. have collectively identified, at most, 11,000 different protein associations (and probably fewer, considering the potential overlap between data sets). Although feasible, the characterization of all remaining interactions will almost certainly be labour intensive. But the resulting data will be more than worth the effort.
