Probing Protein Interactions

The challenge of proteomics is personified in the Greek god, Proteus. The keeper of all knowledge, past, present and future, Proteus would not give up any information easily; even while held down, he would struggle and assume different forms before giving anything up. Remarkably, proteomics, and proteins for that matter, were not named after Proteus, but the imagery could not be more fitting. It's still anyone's guess what the final gene count will be in the human genome, let alone the total number of proteins. But one thing is for sure: The proteome is astoundingly more complex than the genome. And it is not just the numbers, though they are staggering: 30,000 genes could well translate into one million proteins. Where genes are easily manipulated and synthesized, allowing scientists to anticipate how they will behave chemically, researchers have no such luck with the proteome: The proteome's truly exasperating feature is its dynamic nature. Proteins are made, processed, turned over, modified, shuffled from one compartment to another, and bundled up into complexes with other proteins. "We have to remember," says Marc Vidal, assistant professor of genetics, Dana Farber Cancer Institute in Boston, "that the proteome is vast. It's like terra incognita.... We have a few settlers, we have to explore a huge amount of space." Courtesy of Marc VidalMarc Vidal New Toys in the Sandbox Finding the tools for examining these myriad aspects is the challenge facing researchers in the field. As recently as five years ago, a single dominant technology existed—two-dimensional gel electrophoresis. This technique may still be the best for displaying the proteome's intricacies, and mass spectrometry, used as a way of analyzing the spots seen on 2-D gels, is taking 2-D gel sensitivity to new heights. But other technologies, like two-hybrid analyses and protein microarrays, are rivaling 2-D gels for large-scale efforts at cataloging proteins and their interactions. Studying protein interactions has been a particularly active field, as researchers look for clues to the function of uncharacterized proteins by examining the proteins with which they assemble—guilt by association. Classically, biochemical or immunological techniques have been used to detect protein partners to see what else comes along when a particular protein is pulled, and some still consider these techniques to be the gold standard for proving an interaction. But only recently have these techniques become amenable to scaling up. As sequencing projects began identifying large numbers of protein sequences, researchers began looking for tools that could deal with genome-sized numbers of proteins. Enter two-hybrid systems. Developed in 1989 by Stanley Fields, Howard Hughes Medical Institute investigator and professor of genome sciences and medicine, University of Washington, and researcher Ok-kyu Song, Pohang Institute of Science and Technology, Pohang, Korea,1 these systems exploit the bi-partitite nature of certain transcriptional activators, which have DNA-binding and activation domains. In nature, these two domains frequently co-habit the same protein, but it also turns out that they can work equally well apart, as long as they are near one another. The two domains, then, can be separately fused to proteins of interest, creating what is called 'bait and prey.' When bait and prey interact in a yeast cell, carrying both partners, they unite the two transcriptional domains of the transcriptional activator, which then turns on a reporter gene in the host. Originally, two-hybrid screens were performed with protein pairs or with a single bait protein screened against a library of prey proteins. But, in the past few years, several groups have turned two-hybrid systems loose on the entire proteome of several simple organisms. This is accomplished several ways. Fields, for example, created arrays of clones expressing a single yeast open reading frame (ORF), and devised an automated system for forcing individual yeast proteins to react with the arrays.2 A Japanese group has cloned all 6,000 yeast ORFs into two-hybrid vectors and mated pools of several dozen clones in all possible combinations.3 Researchers, using this technique, have discovered thousands of interactions between proteins in yeast, worms, and humans, and have constructed elaborate networks; in one case involving fully one quarter of all the predicted proteins from the yeast genome.4 But now it has become clear that the information coming from two-hybrid systems is not enough. For one thing, only a fraction of the total number of interactions has been identified this way. A protein may average five partners (the number ranges from 1 to 24). In yeast, with 6,000 gene products, this means there would be minimally 30,000 interactions, and that is considering each gene product only once, which ignores the fact that a single gene can produce a variety of versions of its protein through alternate splicing or various modifications. At best then, 10% of the total number of interactions has been uncovered with even such a simple organism. Not only do two-hybrid screens miss many interactions (false negtives), they are notorious for producing false positives—spurious inter-actions that don't exist in nature. Various reasons exist for this. Proteins called auto-activators turn on transcription by themselves, raising the background of positives, or in the screen, two proteins might come into contact that are located in different subcellular locations and hence would never interact in vivo. There are ways of controlling for false positives, but as researchers develop larger, genome-wide screens, this background can constitute a significant fraction of the data, and cannot be eliminated with the methods that researchers use when working on a single protein. Combining different experimental approaches is one way to overcome false positives. Networks and Fragments Fields and coworkers reported in a paper published last year on a method that combined two "orthogonal," or fundamentally different techniques, to improve interaction-data quality.5 They used phage-display, an in vitro method for finding interacting partners, followed by a computational method to construct a network of proteins that interact with a specific protein domain or motif. They created a second network using yeast two-hybrid data of all the possible binding partners for each motif. Using this approach on the SH3 yeast domain, the team generated a phage-display network containing 394 interactions and a two-hybrid network containing 233 interactions, with 59 common to both. Researchers assessed these common interactions as being more than three-fold likely to be physiologically relevant than those of the overall two-hybrid network. Pierre Legrain, vice president of science and technology at Paris-based Hybrigenics, agrees that the current technologies, which detect only 5%-10% of the interactions, are not adequate for constructing a complete protein interaction map. His company has taken a different approach in its "industrial platform." Rather than screening whole proteins, the Hybrigenics technology screens protein fragments, tens of millions of them, from which they select families of fragments that interact with particular proteins. The trick here is that by using fragments, researchers eliminate the parts of the protein that do not interact, which helps reduce the background. In addition, the fragments map specifically to interacting regions of proteins, thereby providing direct links to function. Company scientists used this method to conduct a large-scale screen of the human gastric pathogen Helicobacter pylori, and created a network involving almost half of the H. pylori proteome.6 These fragments, Legrain points out, have further utility, as they can be used to create mutant phenotypes in functional assays. When the fragments are introduced into cells, they behave as dominant negative mutants, competing with and knocking out the function of the protein's normal version in the cell. With randomly selected fragments, this experiment would have a very small chance of success, but having selected the fragments through an interaction screen, Legrain knows that the protein folds correctly and can interact. "So that gives some kind of validation of the interaction, and even more it gives you an assay for the protein," he says. More is Better But the problems with screening methods, like two-hybrid, are not simply technical; it is not just a matter of eliminating false-positives. There are conceptual limitations to the information that comes from a single screen, according to Vidal. A single dot on an array or a list of protein partners, while informative, does not reveal how the proteins function. To determine protein function information from other sources has to be integrated; something Vidal has started to do. In a recent paper, Vidal describes a "transcriptome-interactome correlation mapping procedure," which he uses to show that clusters of genes with similar expression profiles correlated with interacting protein clusters.7 A further advantage of this technique is that it is not restricted to protein pairs; research-ers can analyze triplets and possibly even larger protein clusters. Vidal also has integrated protein-interaction data with high throughput phenotype screens, and has used information on conserved protein interactions in one species, which he calls interologs, to predict interactions in another species.8 Vidal is integrating three in-dependent data sets in his lab: transcriptome, interactome and phenome (phenotype map), all reflecting protein interactions in the nematode worm. Vidal says, "If you know what the binding partners [of a protein] were, and when it and neighboring genes were expressed, and if in addition you know the phenotype of the loss of function of these genes, then you can start building models of what those genes might do and also how they do it." © 2001 Cell Press, Reprinted with Permission from Elsevier ScienceIntegrate, Consolidate, Elucidate: This depiction of correlated functional maps for Caenorhabditis elegans suggests that by integrating several large-scale functional genomic projects, it should be possible to formulate increasingly meaningful hypotheses. Screening for Masses Mass spectrometry (MS) is playing an important role in the proteomics effort.9,10 Groups from the Germany company Cellzome and the Toronto company MDS-Proteomics have used MS in large-scale protein complex screens. The Cellzome scientists first engineered affinity tags onto hundreds of yeast proteins, which they used to pull the protein and its associated partners out of cells. Researchers then analyzed the complexes by conventional MS, which revealed hundreds of multi-protein complexes, more than 90% of which contained at least one previously uncharacterized protein. Furthermore, they found that most of the complexes had a protein in common with another complex, enabling them to draw higher-order networks of interaction.11 According to Fields, MS may be just the ticket for getting at the proteome's dynamic aspects. "MS has gotten so good and the throughput so much better, that you can address dynamic questions at the protein level. There's not much out there that can do it." Vidal warns about being too quick to throw out the false positives that inevitably come from large-scale screening efforts. While in the past, he has recommended care in interpreting data, he now wonders whether important or novel interactions might be overlooked. "We must be careful about the judgments that we make, we might miss some very interesting biology. Why do you want to know an interaction takes place—you want to know when it's surprising, when it makes you think differently." Laura DeFrancesco (defrancesco1@earthlink.net) is a freelance science writer in Pasadena, Calif. References 1. S. Fields and O. Song, "A novel genetic system to detect protein-protein interactions," Nature, 340:245-6, 1989. 2. P. Uetz et al., "A comprehensive analysis of protein-protein interactions in Saccharomyces cerevisiae," Nature, 403:623-7, Feb. 10, 2000. 3. T. Ito et al., "A comprehensive two-hybrid analysis to explore the yeast protein interactome," Proceedings of the National Academy of Sciences, 98:4569-74, April 10, 2001. 4. B. Schwikowski et al, "A network of protein-protein interactions in yeast," Nature Biotechnology, 18:1257-61, 2000. 5. A. Tong et al., "A combined experimental and computational strategy to define protein interaction networks for peptide recognition modules," Science, 295:321-4, Jan. 11, 2002. 6. J.C. Rain et al., "The protein-protein interaction map of Helicobacter pylori," Nature, 409:211-5, Jan. 10, 2001. 7. H. Ge et al., "Correlation between transcriptome and interactome mapping data from Saccharomyces cerevisiae, Nature Genetics, 29:482-6, December 2001. 8. L.R. Matthews et al., "Identification of potential interaction networks using sequence-based searches for conserved protein-protein interactions or 'interologs,'" Genome Research, 11:2120-6, December 2001. 9. A. Gavin et al, "Functional organization of the yeast proteome by systematic analysis of protein complexes," Nature, 415:141-7, Jan. 10, 2002. 10. Y. Ho et al, "Systematic identification of protein complexes in Saccharomyces cerevisiae by mass spectrometry," Nature, 415:180-3, Jan.10, 2002. 11. J. Roberts, "TAPping the power of proteomics," The Scientist, 16[5]:32, March 4, 2002.

Probing Protein Interactions

Interested in reading more?

Become a Member of