- Email: [email protected]
Integrating a functional proteomic approach into the target discovery process
Biochimie 86 (2004) 625–632 www.elsevier.com/locate/biochi
Integrating a functional proteomic approach into the target discovery process Frédéric Colland, Laurent Daviet * Hybrigenics SA, 3-5, Impasse Reille, 75014 Paris, France Received 15 June 2004; accepted 30 September 2004 Available online 22 October 2004
Abstract Functional proteomics is a promising technique for the rational identification of novel therapeutic targets by elucidation of the function of newly identified proteins in disease-relevant cellular pathways. Of the recently described high-throughput approaches for analyzing protein–protein interactions, the yeast two-hybrid (Y2H) system has turned out to be one of the most suitable for genome-wide analysis. However, this system presents a challenging technical problem: the high prevalence of false positives and false negatives in datasets due to intrinsic limitations of the technology and the use of a high-throughput, genetic assay. We discuss here the different experimental strategies applied to Y2H assays, their general limitations and advantages. We also address the issue of the contribution of protein interaction mapping to functional biology, especially when combined with complementary genomic and proteomic analyses. Finally, we illustrate how the combination of protein interaction maps with relevant functional assays can provide biological support to large-scale protein interaction datasets and contribute to the identification and validation of potential therapeutic targets. © 2004 Elsevier SAS. All rights reserved. Keywords: Yeast two-hybrid; Protein interaction map; Functional proteomics; Signaling pathway; Target validation
1. Introduction The availability of an increasing number of complete genome sequences has raised the exciting possibility of functional interpretation for large amounts of genomic information. Several technologies, group together under the term “proteomics”, have emerged with the common objective of studying protein function at the scale of an entire pathway, a whole cell or even a whole organism. Proteomic analyses encompass large-scale studies of protein–protein interactions or complexes to establish comprehensive protein interaction maps (“interactome”), the global examination of protein expression profiles and, more recently, of protein posttranslational modifications. The study of protein–protein interactions has benefited from the development of highthroughput technologies such as the yeast two-hybrid (Y2H) system, protein chips, phage display or systematic analysis of protein complexes by tandem affinity purification (TAP) and * Corresponding author. Tel.: +33-1-58-10-38-00; fax: +33-1-58-10-38-49. E-mail address: [email protected] (L. Daviet). 0300-9084/$ - see front matter © 2004 Elsevier SAS. All rights reserved. doi:10.1016/j.biochi.2004.09.014
mass spectrometry. However, the information generated should be interpreted with caution due to the intrinsic limitations of these assays. Biologists now also face the major challenge of meaningfully exploiting the very large amounts of data generated by these technologies to formulate valid functional hypotheses. The quality of the functional annotations produced could be substantially improved by integrating data from various genomic and proteomic analyses. These tasks require dedicated tools to help the biologists making this transition. In the context of a target discovery strategy, the identification and selection of potential therapeutic targets also depends on the formulation of an accurate hypothesis concerning protein function. Subsequent functional validation of this hypothesis (i.e. demonstration of the functional role of the potential target in the disease phenotype) requires the use of disease-relevant assays. Novel technologies, such as RNA interference, have made it possible to develop medium- to high-throughput cell assays for evaluating the effects of modulating selected targets on disease-relevant pathways and/or cellular phenotypes.
626
F. Colland, L. Daviet / Biochimie 86 (2004) 625–632
We discuss these issues here in the context of a streamlined process running from raw protein–protein interaction data to target identification and validation. 2. The yeast two-hybrid system (Y2H) as a functional proteomic tool In the last 10 years, Y2H technology has turned out to be one of the methods of choice for comprehensive functional annotation by interaction mapping in the post-genomic era. The success of this technique is based on its relative simplicity, low cost and the possibility of scaling it up to highthroughput levels (potential for industrialization and automation). This technology, originally developed by Fields and Song, can detect interactions between two known proteins or polypeptides and can also search for unknown partners (preys) of a given protein (bait) [1]. Nevertheless, due to its intrinsic properties (detecting interaction between two heterologous, chimeric proteins in a yeast nucleus environment), this method is not applicable to all classes of protein–protein interaction, resulting in a significant proportion of falsenegative and false-positive results. The intrinsic limitations of conventional Y2H technology concern the possible misfolding and subsequent instability of the hybrid polypeptides, their inappropriate subcellular localization, the absence of certain post-translational modifications (such as tyrosine phosphorylation or complex glycosylation), and the lack of physiological context (the absence of physiological spatial and temporal regulation of protein interaction). Other properties of the assay may also lead to the selection of false-positive. They include the “auto-activation” properties of certain bait proteins, which themselves activate reporter gene transcription, and the “stickiness” of some prey proteins or fragments, which bind nonspecifically to a wide variety of baits. These are critical issues that must be carefully addressed when defining an experimental strategy for generating largescale protein interaction maps. 3. One Y2H technology, different experimental strategies Recently, several global studies for the large-scale investigation of protein–protein interactions in viruses (hepatitis C and vaccinia viruses, T7 bacteriophage) [2–4], prokaryotic (Helicobacter pylori) [5] and unicellular eukaryotic organisms (Saccharomyces cerevisiae) [6,7] have been described. More recently, extended interactome maps have been generated in two multicellular model organisms: the fruit fly and the nematode [8,9]. More focused studies have been carried out to analyze specific signaling or metabolic pathways [10,11]. These studies were conducted using various experimental strategies and gave quite different results, particular as concerned the rates of false positives and false negatives [12]. What have we learned from these studies?
Two main screening formats have been used: the matrix and library formats. The matrix approach uses a collection of predefined open-reading frame products (ORFs), usually full-length proteins, as bait and prey for interaction assays. Combinations of bait and prey are then assessed either individually or in pools. One of the major limitations of the matrix format is that it tests only canonical, full-length proteins, which are in many cases, inappropriate for interaction screening. Prime examples include proteins bearing transmembrane domains, membrane-anchoring motifs, transcriptional activation domains, or signal peptides. In the prey fragment library format, extensive collections of randomly generated prey protein fragments are screened against a given bait polypeptide [5,13,14]. An optimized version of this approach combines the screening of short fragmented prey libraries with the rational design of a bait polypeptide (i.e. selection of the most appropriate constructs for interaction screening). This experimental approach dramatically improves the quality of the resulting interaction map by minimizing the rate of false negatives when analyzing, for example, ‘difficult’ bait proteins such as membrane receptors [3,15–17]. The analysis of protein–protein interactions using multiple truncated forms of bait and prey polypeptides maximizes the chances of obtaining properly folded, stable domains and, therefore, successful interactions. This finding has been substantiated in yeast, for which screening a domain library rather than full-length proteins has considerably enriched the yeast protein interaction database [18]. In addition, interacting domains are, in many full-length proteins, masked by intramolecular interactions and are, therefore, not accessible to their ligands unless exposed in response to a specific activation signal. Conversely, an interaction may be missed if the interaction domain is large and discontinuous. Such interactions are more likely to be detected with fulllength proteins. A unique feature of the screening of randomly generated protein fragments resides in the mapping of the interacting domains. The sequence common to the overlapping prey fragments defines the smallest selected binding site for the bait protein (Fig. 1). These domains may facilitate the discovery of novel structural/functional domain signatures [16] and may be used in functional validation assays as dominantnegative alleles and in the development of assays for the modulation of interaction. Finally, an adaptation of the selection procedure (modification of the composition of selective media) to take into
Fig. 1. Mapping interacting domain. The prey fragment library format identifies the domain of each prey protein interacting with a given bait.
F. Colland, L. Daviet / Biochimie 86 (2004) 625–632
account the intrinsic ability of each bait to auto-activate reporter genes, can be used to facilitate the analysis of a larger proportion of bait proteins and considerably decreases the frequency of false-positive interactions [5,14,16]. However, the library approach has some limitations, including (i) the need to prepare highly complex prey fragment libraries to ensure comprehensive coverage of the transcriptome of interest (for rare mRNA molecules in particular) and to take into account the fact that only a fraction of the genomic or cDNA fragments are likely to encode bona fide polypeptides (due to the location of the fragment, its orientation or reading frame); (ii) the technical difficulty and cost of interaction screens required to cover such complex libraries exhaustively. The comprehensiveness of the screening procedure is, however, critical to search for rare transcripts and for sufficient reproducibility, an important parameter in assessing interaction reliability (see below). In our hands, these ‘rules’ define an acceptable compromise between throughput and quality (i.e. reproducibility of the screens, density of the map, interaction scoring). 4. Assessing interaction reliability As explained above, the prevalence of non-relevant interactions in datasets resulting from the use of a highthroughput method and the intrinsic limitations of the technology constitutes a challenging technical problem in Y2H systems. It is, therefore, essential to be able to score every single interaction pair for reliability with respect to the technology. The high level of reproducibility of the library screening strategy originally described by Rain et al. [5] and systematic identification of all the positive clones obtained in each individual screen made it possible to develop a systematic statistical approach for assigning a confidence score to each interaction. In particular, prey fragments selected with unrelated bait proteins (i.e. probable “sticky”, nonspecific partners) were labeled and discarded from subsequent analyses [5]. This score, termed PBS®, has been shown in several studies to correlate with the biological significance of interactions [5,14,16,19]. Other attempts at confidence scoring have been described but used only the redundancy in prey fragments [6] or identified the parameters used in PBS calculation as global predictors in a trained linear model [9]. The latter may be poorly suited to confidence assignment because it is based on training sets, which are by definition biased by the state of the art in biology. Ideally, the establishment of experimental standards for the uniform evaluation of false positives or, alternatively, access to primary data would be required to compare or to integrate data from different sources. 5. Exploring and analyzing interaction maps for the prediction of protein functions: towards target identification and selection Biologists are the leading users of proteomic data and they need tools for exploring and analyzing these large sets of
627
data, to formulate hypotheses concerning protein function and to design experiments for validating the functional assignment. In the context of a target discovery strategy, this constitutes the first step towards the identification of proteins with a therapeutic potential. The PIMRider® web-based graphical interface was developed to facilitate this transition from raw protein interaction networks to pathway characterization by visualizing, exploring and analyzing protein interaction maps [5,14,16]. This software platform includes four different viewers: (i) the ProteinViewer™, which displays textual annotations for a given protein, its sequence and the list of interacting partners; (ii) the PIMViewer™, which displays a dynamic graphical view of protein interaction networks and facilitates the filtering of interactions according to their reliability score (Fig. 2A); (iii) the InteractionViewer™, which provides access to raw experimental data on prey, bait and SID sequences (Fig. 2B); and (iv) the DomainViewer™, which displays the domains and motifs extracted from both experimental (bait and SID) and calculated (transmembrane segments, signal peptides, and InterPro domains) analyses for any protein and all its partners in the map (Fig. 2C). The most recent versions of this tool make it possible to visualize multiple PIMs generated, for example, around orthologous genes from different species. A demonstration version of this tool is freely available upon registration at http://www.hybrigenics.com. A community standard data model was recently developed by the Proteomics Standards Initiative, to integrate and to represent protein interaction data originating from different sources [20]. These data can be visualized with the PIMWalker tool (http://www.pim.hybrigenics.com/ pimwalker/), which we recently developed to display interaction networks described in this format graphically.
6. The future: matching protein interaction maps with other data sets Beyond interactome representation, analysis and visualization tools, the integration of multiple, independent sets of genomic and proteomic data, as proposed for model organisms such as yeast and C. elegans [8,21] has proved to be even more powerful in assigning unannotated proteins to biological pathways. Other large-scale genomic analyses include expression profiling, genetic data, phenotypic analyses and protein localization. The exploitation of these large amounts of heterogeneous data requires meaningful integration and useful presentation tools. In recent years, an increasing number of tools have been proposed for the integration of data and methods, and for analysis through dedicated platforms [22]. One such tool, GenoLink, developed within the Genostar Consortium [22], allows the integration of several different data types and their exploration using a graph pattern query engine. As shown in Fig. 3, by integrating protein interaction and orthology data, this tool makes it possible to
628
F. Colland, L. Daviet / Biochimie 86 (2004) 625–632
Fig. 2. PIMRider screen shots. (A) Protein interaction map surrounding the Drosophila Ras protein as displayed by the PIMViewer. Links between proteins identify physical connections with their color-coded PBS® score. (B) The InteractionViewer™ exhibits, for each interaction, the prey fragments found in the screen, their respective sequence and the resulting SID®. (C) DomainViewer™ for Ras and all its partners displaying, for each protein, the domains and motifs extracted from both experimental (bait and SID) and calculated (transmembrane segments, signal peptides, and functional InterPro domains) analyses.
generate a query that addresses a specific biological question, and then to extract and to visualize the results. 7. From pathway mapping to the identification and validation of potential targets As discussed above, protein interaction mapping can be used to predict protein function but, ultimately, experimental
validation of the functional assignment is required. The transition from a large number of interactions to more specific functional analyses with the ultimate goal of validating the predicted biological function for each newly identified interaction seems to be a key limitation. This type of study was initially performed in model organisms such as C. elegans, with the establishment of protein interaction maps for the DNA damage response [11] and the DAF7/TGFb signal
F. Colland, L. Daviet / Biochimie 86 (2004) 625–632
629
Fig. 2. (continued)
transduction pathway [23], and was followed by functional validation. Two recent studies in humans combined protein interaction mapping with functional validation assays. The first analyzed the TNF-a/NF-jB signal transduction pathway [24] whereas the second deciphered the human Smad signaling pathway [14]. In these two studies, protein interaction mapping, achieved with different technologies, generated large datasets that were explored with the help of diverse functional annotations from public databases. Global validation of the network was achieved by the identification of proteins previously known to be involved in these pathways. A set of proteins, as yet poorly annotated or functionally unrelated to these pathways, was chosen for testing in lossof-function and overexpression experiments in cells, to confirm their involvement in these signaling pathways. Colland et al. studied the human Smad signaling pathway. Members of the TGFb superfamily (e.g. TGFb, activins and bone morphogenetic proteins (BMPs)) are secreted signaling molecules that regulate many biological processes such as cell growth, differentiation and morphogenesis. Deregulations of the TGFb pathway have been associated with oncogenic transformation, metastasis, fibrosis and inflammatory disorders. TGFb signal transduction involves Ser/Thr kinase receptors at the cell surface and their substrates, the Smad proteins, which upon activation are translocated to the nucleus where they regulate target gene expression [25]. Protein–protein interactions involving 23 Smad and related proteins were screened using the yeast two-hybrid library strategy described above. A protein interaction map containing 755 interactions and 591 proteins was generated. In this network, 18 known Smad-related proteins and 179 poorly annotated or unannotated proteins were identified. The full data set is freely available and can be visualized using the PIMRider tool at http://www.pim.hybrigenics.com.
On the basis of interaction scores and selected functional domain annotations, 14 proteins were selected for further validation. These proteins were tested for their involvement in the Smad pathway by loss-of-function and overexpression studies in mammalian cells. Experiments using siRNAmediated gene knockdown combined with Smad-specific reporter genes and Smad-dependent endogenous target genes as readout showed that eight of these 14 proteins were involved in Smad signaling. Interestingly, four of these eight proteins (PPP1CA, ZNF8, MAN1 and RNF11) were independently validated as effectors of the Smad pathway by other groups [26–29]. The other four (KIAA1196, HYPA, LMO4 and LAPTm5) corresponds to proteins of unknown function or poorly characterized proteins for which biological functions can be predicted. For example, LAPTm5, a predicted five-transmembrane domain protein is transcriptionally up-regulated by TGFb and acts as a negative regulator of the TGFb pathway, suggesting its involvement in a negative feed-back loop. These findings, combined with interaction data, suggest that LAPTm5 may be a Smurf2 receptor in the lysosomal membrane and may target specific TGFb signaling components for lysosomal degradation [14]. Another interesting example concerns RNF11, a RING finger protein recently shown to be overexpressed in breast and prostate cancer tissues [28]. We originally identified RNF11 as an interacting partner of various members of the Smad pathway, such as SARA and Smurf proteins. A functional link with the TGFb axis was established by demonstrating that the siRNA-mediated knockdown of RNF11 impaired TGFb-dependent transcription [14]. More detailed proteomic analysis of RNF11 revealed an intimate protein interaction network connecting several factors involved in protein degradation, such as ubiquitin-related proteins and E3 ligases of the HECT family (Fig. 4). Based on these
630
F. Colland, L. Daviet / Biochimie 86 (2004) 625–632
Fig. 3. GenoLink screen shots. (A) Query Builder, the graphical user interface dedicated to the construction of queries using a graph-based modeling. Shown is the construction of the query: “what are the protein–protein interactions for which the interacting partners are orthologous proteins”. (B) Result of the query in (A) when applied to protein interaction maps generated around the human and Drosophila Ras oncoproteins. Shown on the right is a snapshot of the results as displayed by the Graph Rider. In the absence of interactive displaying and for clarity reason, the query results are also shown in a “PIMViewer format” with the human (upper part) and Drosophila (lower part) Ras interaction maps. The red vertical lines connect the orthologous human and Drosophila Ras interactors.
interaction data, we suggested that RNF11 might be an E3 ligase playing a central role in the regulation of these HECT enzymes. We further demonstrated that RNF11 displays selfubiquitination activity in vitro, suggesting that it has intrinsic ubiquitin ligase activity. We investigated whether inhibiting RNF11 function was therapeutically relevant in cancer, using siRNA to silence RNF11 gene transcription selectively in a set of human cancer cell lines. Decreasing RNF11 gene expression significantly inhibited prostate cancer cell proliferation by inducing G1 cell-cycle arrest (F. Colland, E. Formstecher, X. Jacq, S. Aresta, A. Calabrese, J.C. Rain, L. Daviet, unpublished results). Together, these data identify RNF11 as
a novel regulator of the TGFb pathway and a potential cancer target. With a view to extending this approach, Hybrigenics and the Institute Curie (Paris) have established a strategic alliance for investigating major cancer-relevant pathways using comparative proteomic analysis in human and Drosophila. The resulting Drosophila Protein Interaction Map, containing 1727 proteins (12% of the proteome), connected by over 2300 protein–protein interactions has been made freely available to the scientific community (http://www.pim. hybrigenics.com).
F. Colland, L. Daviet / Biochimie 86 (2004) 625–632
631
Fig. 4. RNF11 protein interaction map. Part of the interaction map surrounding the RING finger protein RNF11. The interactors of RNF11 can be subdivided in three groups: (i) effectors of the Smad pathway; (ii) ubiquitin and ubiquitin-like proteins; and (iii) HECT-containing E3 ubiquitin ligases.
8. Conclusion The availability of genome sequences for large numbers of organisms provides a major impetus for functional proteomic analyses. Of the methods used, the yeast two-hybrid system combines undeniable advantages for large-scale functional biology studies and recent methodological optimizations of the technology have greatly increased its comprehensiveness. However, appropriate bioinformatics tools for the exploration and analysis of the resulting large sets of data and for integrating heterogeneous genomic and proteomic analysis are becoming crucial for the in-depth analysis of extended protein networks. Finally, combining protein interaction mapping with functional assays should greatly facilitate the experimental validation of interactome-derived, functional assignment. Ultimately, the application of this strategy to disease-related pathways may lead to the identification and validation of novel therapeutic targets.
Acknowledgements We thank all Hybrigenics staff for their contribution.
References [1] [2]
[3]
M. Vidal, P. Legrain, Yeast forward and reverse ‘n’-hybrid systems, Nucleic Acids Res. 27 (1999) 919–929. P.L. Bartel, J.A. Roecklein, D. SenGupta, S. Fields, A protein linkage map of Escherichia coli bacteriophage T7, Nat. Genet. 12 (1996) 72–77. M. Flajolet, G. Rotondo, L. Daviet, F. Bergametti, G. Inchauspe, P. Tiollais, C. Transy, P. Legrain, A genomic approach of the hepatitis C virus generates a protein interaction map, Gene 242 (2000) 369– 379.
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
S. McCraith, T. Holtzman, B. Moss, S. Fields, Genome-wide analysis of vaccinia virus protein–protein interactions, Proc. Natl. Acad. Sci. USA 97 (2000) 4879–4884. J.C. Rain, L. Selig, H. De Reuse, V. Battaglia, C. Reverdy, S. Simon, G. Lenzen, F. Petel, J. Wojcik, V. Schachter, et al., The protein–protein interaction map of Helicobacter pylori, Nature 409 (2001) 211–215. T. Ito, T. Chiba, R. Ozawa, M. Yoshida, M. Hattori, Y. Sakaki, A comprehensive two-hybrid analysis to explore the yeast protein interactome, Proc. Natl. Acad. Sci. USA 98 (2001) 4569–4574. P. Uetz, L. Giot, G. Cagney, T.A. Mansfield, R.S. Judson, J.R. Knight, D. Lockshon, V. Narayan, M. Srinivasan, P. Pochart, et al., A comprehensive analysis of protein–protein interactions in Saccharomyces cerevisiae, Nature 403 (2000) 623–627. S. Li, C.M. Armstrong, N. Bertin, H. Ge, S. Milstein, M. Boxem, P.O. Vidalain, J.D. Han, A. Chesneau, T. Hao, et al., A map of the interactome network of the metazoan C. elegans, Science 303 (2004) 540–543. L. Giot, J.S. Bader, C. Brouwer, A. Chaudhuri, B. Kuang, Y. Li, Y.L. Hao, C.E. Ooi, B. Godwin, E. Vitols, et al., A protein interaction map of Drosophila melanogaster, Science 302 (2003) 1727–1736. A.J. Walhout, R. Sordella, X. Lu, J.L. Hartley, G.F. Temple, M.A. Brasch, N. Thierry-Mieg, M. Vidal, Protein interaction mapping in C. elegans using proteins involved in vulval development, Science 287 (2000) 116–122. S.J. Boulton, A. Gartner, J. Reboul, P. Vaglio, N. Dyson, D.E. Hill, M. Vidal, Combined functional genomic maps of the C. elegans DNA damage response, Science 295 (2002) 127–131. P. Legrain, J. Wojcik, J.M. Gauthier, Protein–protein interaction maps: a lead towards cellular functions, Trends Genet. 17 (2001) 346–352. P.J. Adam, R. Boyd, K.L. Tyson, G.C. Fletcher, A. Stamps, L. Hudson, H.R. Poyser, N. Redpath, M. Griffiths, G. Steers, et al., Comprehensive proteomic analysis of breast cancer cell membranes reveals unique proteins with potential roles in clinical cancer, J. Biol. Chem. 278 (2003) 6482–6489. F. Colland, X. Jacq, V. Trouplin, C. Mougin, C. Groizeleau, A. Hamburger, et al., Functional proteomic mapping of a human signaling pathway, Genome Res. 14 (2004) 1324–1332. A.D. Strosberg, L. Daviet, C. Borg-Capra, Functional proteomics applied to obesity research, in: G. Medeiros-Neto, A. Halpern, C. Bouchard (Eds.), Progress in Obesity Research, John Libbey Eurotext Ltd, 2003, pp. 229–233.
632
F. Colland, L. Daviet / Biochimie 86 (2004) 625–632
[16] E. Formstecher, S. Aresta, V. Collura, A. Hamburger, A. Meil, A. Trehin, C. Reverdy, V. Betin, S. Maire, C. Brun, et al., High fidelity protein interaction mapping: a Drosophila case study, (submitted for publication). [17] B. Boeda, A. El-Amraoui, A. Bahloul, R. Goodyear, L. Daviet, S. Blanchard, I. Perfettini, K.R. Fath, S. Shorte, J. Reiners, et al., Myosin VIIa, harmonin and cadherin 23, three Usher I gene products that cooperate to shape the sensory hair cell bundle, EMBO J. 21 (2002) 6689–6699. [18] M. Fromont-Racine, J.C. Rain, P. Legrain, Toward a functional analysis of the yeast genome through exhaustive two-hybrid screens, Nat. Genet. 16 (1997) 277–282. [19] J. Wojcik, I.G. Boneca, P. Legrain, Prediction, assessment and validation of protein interaction maps in bacteria, J. Mol. Biol. 323 (2002) 763–770. [20] H. Hermjakob, L. Montecchi-Palazzi, G. Bader, J. Wojcik, L. Salwinski, A. Ceol, S. Moore, S. Orchard, U. Sarkans, C. Von Mering, et al., The HUPO PSI’s molecular interaction format—a community standard for the representation of protein interaction data, Nat. Biotechnol. 22 (2004) 177–183. [21] H. Ge, Z. Liu, G.M. Church, M. Vidal, Correlation between transcriptome and interactome mapping data from Saccharomyces cerevisiae, Nat. Genet. 29 (2001) 482–486. [22] P. Durand, C. Medigue, A. Morgat, Y. Vandenbrouck, A. Viari, F. Rechenmann, Integration of data and methods for genome analysis, Curr. Opin. Drug Discov. Dev. 6 (2003) 346–352.
[23] M. Tewari, P.J. Hu, J.S. Ahn, N. Ayivi-Guedehoussou, P.O. Vidalain, S. Li, S. Milstein, C.M. Armstrong, M. Boxem, M.D. Butler, et al., Systematic interactome mapping and genetic perturbation analysis of a C. elegans TGF-beta signaling network, Mol. Cell 13 (2004) 469– 482. [24] T. Bouwmeester, A. Bauch, H. Ruffner, P.O. Angrand, G. Bergamini, K. Croughton, C. Cruciat, D. Eberhard, J. Gagneur, S. Ghidelli, et al., A physical and functional map of the human TNF-alpha/NF-kappaB signal transduction pathway, Nat. Cell. Biol. 6 (2004) 97–105. [25] J. Massague, TGF-beta signal transduction, Annu. Rev. Biochem. 67 (1998) 753–791. [26] D. Bennett, L. Alphey, PP1 binds Sara and negatively regulates Dpp signaling in Drosophila melanogaster, Nat. Genet. 31 (2002) 419– 423. [27] G.P. Raju, N. Dimova, P.S. Klein, H.C. Huang, SANE, a novel LEM domain protein, regulates bone morphogenetic protein signaling through interaction with Smad1, J. Biol. Chem. 278 (2003) 428–437. [28] V. Subramaniam, H. Li, M. Wong, R. Kitching, L. Attisano, J. Wrana, J. Zubovits, A.M. Burger, A. Seth, The RING-H2 protein RNF11 is overexpressed in breast cancer and is a target of Smurf2 E3 ligase, Br. J. Cancer 89 (2003) 1538–1544. [29] K. Jiao, Y. Zhou, B.L. Hogan, Identification of mZnf8, a Mouse Kruppel-like transcriptional repressor, as a novel nuclear interaction partner of Smad1, Mol. Cell. Biol. 22 (2002) 7633–7644.
Integrating a functional proteomic approach into the target discovery process Frédéric Colland, Laurent Daviet * Hybrigenics SA, 3-5, Impasse Reille, 75014 Paris, France Received 15 June 2004; accepted 30 September 2004 Available online 22 October 2004
Abstract Functional proteomics is a promising technique for the rational identification of novel therapeutic targets by elucidation of the function of newly identified proteins in disease-relevant cellular pathways. Of the recently described high-throughput approaches for analyzing protein–protein interactions, the yeast two-hybrid (Y2H) system has turned out to be one of the most suitable for genome-wide analysis. However, this system presents a challenging technical problem: the high prevalence of false positives and false negatives in datasets due to intrinsic limitations of the technology and the use of a high-throughput, genetic assay. We discuss here the different experimental strategies applied to Y2H assays, their general limitations and advantages. We also address the issue of the contribution of protein interaction mapping to functional biology, especially when combined with complementary genomic and proteomic analyses. Finally, we illustrate how the combination of protein interaction maps with relevant functional assays can provide biological support to large-scale protein interaction datasets and contribute to the identification and validation of potential therapeutic targets. © 2004 Elsevier SAS. All rights reserved. Keywords: Yeast two-hybrid; Protein interaction map; Functional proteomics; Signaling pathway; Target validation
1. Introduction The availability of an increasing number of complete genome sequences has raised the exciting possibility of functional interpretation for large amounts of genomic information. Several technologies, group together under the term “proteomics”, have emerged with the common objective of studying protein function at the scale of an entire pathway, a whole cell or even a whole organism. Proteomic analyses encompass large-scale studies of protein–protein interactions or complexes to establish comprehensive protein interaction maps (“interactome”), the global examination of protein expression profiles and, more recently, of protein posttranslational modifications. The study of protein–protein interactions has benefited from the development of highthroughput technologies such as the yeast two-hybrid (Y2H) system, protein chips, phage display or systematic analysis of protein complexes by tandem affinity purification (TAP) and * Corresponding author. Tel.: +33-1-58-10-38-00; fax: +33-1-58-10-38-49. E-mail address: [email protected] (L. Daviet). 0300-9084/$ - see front matter © 2004 Elsevier SAS. All rights reserved. doi:10.1016/j.biochi.2004.09.014
mass spectrometry. However, the information generated should be interpreted with caution due to the intrinsic limitations of these assays. Biologists now also face the major challenge of meaningfully exploiting the very large amounts of data generated by these technologies to formulate valid functional hypotheses. The quality of the functional annotations produced could be substantially improved by integrating data from various genomic and proteomic analyses. These tasks require dedicated tools to help the biologists making this transition. In the context of a target discovery strategy, the identification and selection of potential therapeutic targets also depends on the formulation of an accurate hypothesis concerning protein function. Subsequent functional validation of this hypothesis (i.e. demonstration of the functional role of the potential target in the disease phenotype) requires the use of disease-relevant assays. Novel technologies, such as RNA interference, have made it possible to develop medium- to high-throughput cell assays for evaluating the effects of modulating selected targets on disease-relevant pathways and/or cellular phenotypes.
626
F. Colland, L. Daviet / Biochimie 86 (2004) 625–632
We discuss these issues here in the context of a streamlined process running from raw protein–protein interaction data to target identification and validation. 2. The yeast two-hybrid system (Y2H) as a functional proteomic tool In the last 10 years, Y2H technology has turned out to be one of the methods of choice for comprehensive functional annotation by interaction mapping in the post-genomic era. The success of this technique is based on its relative simplicity, low cost and the possibility of scaling it up to highthroughput levels (potential for industrialization and automation). This technology, originally developed by Fields and Song, can detect interactions between two known proteins or polypeptides and can also search for unknown partners (preys) of a given protein (bait) [1]. Nevertheless, due to its intrinsic properties (detecting interaction between two heterologous, chimeric proteins in a yeast nucleus environment), this method is not applicable to all classes of protein–protein interaction, resulting in a significant proportion of falsenegative and false-positive results. The intrinsic limitations of conventional Y2H technology concern the possible misfolding and subsequent instability of the hybrid polypeptides, their inappropriate subcellular localization, the absence of certain post-translational modifications (such as tyrosine phosphorylation or complex glycosylation), and the lack of physiological context (the absence of physiological spatial and temporal regulation of protein interaction). Other properties of the assay may also lead to the selection of false-positive. They include the “auto-activation” properties of certain bait proteins, which themselves activate reporter gene transcription, and the “stickiness” of some prey proteins or fragments, which bind nonspecifically to a wide variety of baits. These are critical issues that must be carefully addressed when defining an experimental strategy for generating largescale protein interaction maps. 3. One Y2H technology, different experimental strategies Recently, several global studies for the large-scale investigation of protein–protein interactions in viruses (hepatitis C and vaccinia viruses, T7 bacteriophage) [2–4], prokaryotic (Helicobacter pylori) [5] and unicellular eukaryotic organisms (Saccharomyces cerevisiae) [6,7] have been described. More recently, extended interactome maps have been generated in two multicellular model organisms: the fruit fly and the nematode [8,9]. More focused studies have been carried out to analyze specific signaling or metabolic pathways [10,11]. These studies were conducted using various experimental strategies and gave quite different results, particular as concerned the rates of false positives and false negatives [12]. What have we learned from these studies?
Two main screening formats have been used: the matrix and library formats. The matrix approach uses a collection of predefined open-reading frame products (ORFs), usually full-length proteins, as bait and prey for interaction assays. Combinations of bait and prey are then assessed either individually or in pools. One of the major limitations of the matrix format is that it tests only canonical, full-length proteins, which are in many cases, inappropriate for interaction screening. Prime examples include proteins bearing transmembrane domains, membrane-anchoring motifs, transcriptional activation domains, or signal peptides. In the prey fragment library format, extensive collections of randomly generated prey protein fragments are screened against a given bait polypeptide [5,13,14]. An optimized version of this approach combines the screening of short fragmented prey libraries with the rational design of a bait polypeptide (i.e. selection of the most appropriate constructs for interaction screening). This experimental approach dramatically improves the quality of the resulting interaction map by minimizing the rate of false negatives when analyzing, for example, ‘difficult’ bait proteins such as membrane receptors [3,15–17]. The analysis of protein–protein interactions using multiple truncated forms of bait and prey polypeptides maximizes the chances of obtaining properly folded, stable domains and, therefore, successful interactions. This finding has been substantiated in yeast, for which screening a domain library rather than full-length proteins has considerably enriched the yeast protein interaction database [18]. In addition, interacting domains are, in many full-length proteins, masked by intramolecular interactions and are, therefore, not accessible to their ligands unless exposed in response to a specific activation signal. Conversely, an interaction may be missed if the interaction domain is large and discontinuous. Such interactions are more likely to be detected with fulllength proteins. A unique feature of the screening of randomly generated protein fragments resides in the mapping of the interacting domains. The sequence common to the overlapping prey fragments defines the smallest selected binding site for the bait protein (Fig. 1). These domains may facilitate the discovery of novel structural/functional domain signatures [16] and may be used in functional validation assays as dominantnegative alleles and in the development of assays for the modulation of interaction. Finally, an adaptation of the selection procedure (modification of the composition of selective media) to take into
Fig. 1. Mapping interacting domain. The prey fragment library format identifies the domain of each prey protein interacting with a given bait.
F. Colland, L. Daviet / Biochimie 86 (2004) 625–632
account the intrinsic ability of each bait to auto-activate reporter genes, can be used to facilitate the analysis of a larger proportion of bait proteins and considerably decreases the frequency of false-positive interactions [5,14,16]. However, the library approach has some limitations, including (i) the need to prepare highly complex prey fragment libraries to ensure comprehensive coverage of the transcriptome of interest (for rare mRNA molecules in particular) and to take into account the fact that only a fraction of the genomic or cDNA fragments are likely to encode bona fide polypeptides (due to the location of the fragment, its orientation or reading frame); (ii) the technical difficulty and cost of interaction screens required to cover such complex libraries exhaustively. The comprehensiveness of the screening procedure is, however, critical to search for rare transcripts and for sufficient reproducibility, an important parameter in assessing interaction reliability (see below). In our hands, these ‘rules’ define an acceptable compromise between throughput and quality (i.e. reproducibility of the screens, density of the map, interaction scoring). 4. Assessing interaction reliability As explained above, the prevalence of non-relevant interactions in datasets resulting from the use of a highthroughput method and the intrinsic limitations of the technology constitutes a challenging technical problem in Y2H systems. It is, therefore, essential to be able to score every single interaction pair for reliability with respect to the technology. The high level of reproducibility of the library screening strategy originally described by Rain et al. [5] and systematic identification of all the positive clones obtained in each individual screen made it possible to develop a systematic statistical approach for assigning a confidence score to each interaction. In particular, prey fragments selected with unrelated bait proteins (i.e. probable “sticky”, nonspecific partners) were labeled and discarded from subsequent analyses [5]. This score, termed PBS®, has been shown in several studies to correlate with the biological significance of interactions [5,14,16,19]. Other attempts at confidence scoring have been described but used only the redundancy in prey fragments [6] or identified the parameters used in PBS calculation as global predictors in a trained linear model [9]. The latter may be poorly suited to confidence assignment because it is based on training sets, which are by definition biased by the state of the art in biology. Ideally, the establishment of experimental standards for the uniform evaluation of false positives or, alternatively, access to primary data would be required to compare or to integrate data from different sources. 5. Exploring and analyzing interaction maps for the prediction of protein functions: towards target identification and selection Biologists are the leading users of proteomic data and they need tools for exploring and analyzing these large sets of
627
data, to formulate hypotheses concerning protein function and to design experiments for validating the functional assignment. In the context of a target discovery strategy, this constitutes the first step towards the identification of proteins with a therapeutic potential. The PIMRider® web-based graphical interface was developed to facilitate this transition from raw protein interaction networks to pathway characterization by visualizing, exploring and analyzing protein interaction maps [5,14,16]. This software platform includes four different viewers: (i) the ProteinViewer™, which displays textual annotations for a given protein, its sequence and the list of interacting partners; (ii) the PIMViewer™, which displays a dynamic graphical view of protein interaction networks and facilitates the filtering of interactions according to their reliability score (Fig. 2A); (iii) the InteractionViewer™, which provides access to raw experimental data on prey, bait and SID sequences (Fig. 2B); and (iv) the DomainViewer™, which displays the domains and motifs extracted from both experimental (bait and SID) and calculated (transmembrane segments, signal peptides, and InterPro domains) analyses for any protein and all its partners in the map (Fig. 2C). The most recent versions of this tool make it possible to visualize multiple PIMs generated, for example, around orthologous genes from different species. A demonstration version of this tool is freely available upon registration at http://www.hybrigenics.com. A community standard data model was recently developed by the Proteomics Standards Initiative, to integrate and to represent protein interaction data originating from different sources [20]. These data can be visualized with the PIMWalker tool (http://www.pim.hybrigenics.com/ pimwalker/), which we recently developed to display interaction networks described in this format graphically.
6. The future: matching protein interaction maps with other data sets Beyond interactome representation, analysis and visualization tools, the integration of multiple, independent sets of genomic and proteomic data, as proposed for model organisms such as yeast and C. elegans [8,21] has proved to be even more powerful in assigning unannotated proteins to biological pathways. Other large-scale genomic analyses include expression profiling, genetic data, phenotypic analyses and protein localization. The exploitation of these large amounts of heterogeneous data requires meaningful integration and useful presentation tools. In recent years, an increasing number of tools have been proposed for the integration of data and methods, and for analysis through dedicated platforms [22]. One such tool, GenoLink, developed within the Genostar Consortium [22], allows the integration of several different data types and their exploration using a graph pattern query engine. As shown in Fig. 3, by integrating protein interaction and orthology data, this tool makes it possible to
628
F. Colland, L. Daviet / Biochimie 86 (2004) 625–632
Fig. 2. PIMRider screen shots. (A) Protein interaction map surrounding the Drosophila Ras protein as displayed by the PIMViewer. Links between proteins identify physical connections with their color-coded PBS® score. (B) The InteractionViewer™ exhibits, for each interaction, the prey fragments found in the screen, their respective sequence and the resulting SID®. (C) DomainViewer™ for Ras and all its partners displaying, for each protein, the domains and motifs extracted from both experimental (bait and SID) and calculated (transmembrane segments, signal peptides, and functional InterPro domains) analyses.
generate a query that addresses a specific biological question, and then to extract and to visualize the results. 7. From pathway mapping to the identification and validation of potential targets As discussed above, protein interaction mapping can be used to predict protein function but, ultimately, experimental
validation of the functional assignment is required. The transition from a large number of interactions to more specific functional analyses with the ultimate goal of validating the predicted biological function for each newly identified interaction seems to be a key limitation. This type of study was initially performed in model organisms such as C. elegans, with the establishment of protein interaction maps for the DNA damage response [11] and the DAF7/TGFb signal
F. Colland, L. Daviet / Biochimie 86 (2004) 625–632
629
Fig. 2. (continued)
transduction pathway [23], and was followed by functional validation. Two recent studies in humans combined protein interaction mapping with functional validation assays. The first analyzed the TNF-a/NF-jB signal transduction pathway [24] whereas the second deciphered the human Smad signaling pathway [14]. In these two studies, protein interaction mapping, achieved with different technologies, generated large datasets that were explored with the help of diverse functional annotations from public databases. Global validation of the network was achieved by the identification of proteins previously known to be involved in these pathways. A set of proteins, as yet poorly annotated or functionally unrelated to these pathways, was chosen for testing in lossof-function and overexpression experiments in cells, to confirm their involvement in these signaling pathways. Colland et al. studied the human Smad signaling pathway. Members of the TGFb superfamily (e.g. TGFb, activins and bone morphogenetic proteins (BMPs)) are secreted signaling molecules that regulate many biological processes such as cell growth, differentiation and morphogenesis. Deregulations of the TGFb pathway have been associated with oncogenic transformation, metastasis, fibrosis and inflammatory disorders. TGFb signal transduction involves Ser/Thr kinase receptors at the cell surface and their substrates, the Smad proteins, which upon activation are translocated to the nucleus where they regulate target gene expression [25]. Protein–protein interactions involving 23 Smad and related proteins were screened using the yeast two-hybrid library strategy described above. A protein interaction map containing 755 interactions and 591 proteins was generated. In this network, 18 known Smad-related proteins and 179 poorly annotated or unannotated proteins were identified. The full data set is freely available and can be visualized using the PIMRider tool at http://www.pim.hybrigenics.com.
On the basis of interaction scores and selected functional domain annotations, 14 proteins were selected for further validation. These proteins were tested for their involvement in the Smad pathway by loss-of-function and overexpression studies in mammalian cells. Experiments using siRNAmediated gene knockdown combined with Smad-specific reporter genes and Smad-dependent endogenous target genes as readout showed that eight of these 14 proteins were involved in Smad signaling. Interestingly, four of these eight proteins (PPP1CA, ZNF8, MAN1 and RNF11) were independently validated as effectors of the Smad pathway by other groups [26–29]. The other four (KIAA1196, HYPA, LMO4 and LAPTm5) corresponds to proteins of unknown function or poorly characterized proteins for which biological functions can be predicted. For example, LAPTm5, a predicted five-transmembrane domain protein is transcriptionally up-regulated by TGFb and acts as a negative regulator of the TGFb pathway, suggesting its involvement in a negative feed-back loop. These findings, combined with interaction data, suggest that LAPTm5 may be a Smurf2 receptor in the lysosomal membrane and may target specific TGFb signaling components for lysosomal degradation [14]. Another interesting example concerns RNF11, a RING finger protein recently shown to be overexpressed in breast and prostate cancer tissues [28]. We originally identified RNF11 as an interacting partner of various members of the Smad pathway, such as SARA and Smurf proteins. A functional link with the TGFb axis was established by demonstrating that the siRNA-mediated knockdown of RNF11 impaired TGFb-dependent transcription [14]. More detailed proteomic analysis of RNF11 revealed an intimate protein interaction network connecting several factors involved in protein degradation, such as ubiquitin-related proteins and E3 ligases of the HECT family (Fig. 4). Based on these
630
F. Colland, L. Daviet / Biochimie 86 (2004) 625–632
Fig. 3. GenoLink screen shots. (A) Query Builder, the graphical user interface dedicated to the construction of queries using a graph-based modeling. Shown is the construction of the query: “what are the protein–protein interactions for which the interacting partners are orthologous proteins”. (B) Result of the query in (A) when applied to protein interaction maps generated around the human and Drosophila Ras oncoproteins. Shown on the right is a snapshot of the results as displayed by the Graph Rider. In the absence of interactive displaying and for clarity reason, the query results are also shown in a “PIMViewer format” with the human (upper part) and Drosophila (lower part) Ras interaction maps. The red vertical lines connect the orthologous human and Drosophila Ras interactors.
interaction data, we suggested that RNF11 might be an E3 ligase playing a central role in the regulation of these HECT enzymes. We further demonstrated that RNF11 displays selfubiquitination activity in vitro, suggesting that it has intrinsic ubiquitin ligase activity. We investigated whether inhibiting RNF11 function was therapeutically relevant in cancer, using siRNA to silence RNF11 gene transcription selectively in a set of human cancer cell lines. Decreasing RNF11 gene expression significantly inhibited prostate cancer cell proliferation by inducing G1 cell-cycle arrest (F. Colland, E. Formstecher, X. Jacq, S. Aresta, A. Calabrese, J.C. Rain, L. Daviet, unpublished results). Together, these data identify RNF11 as
a novel regulator of the TGFb pathway and a potential cancer target. With a view to extending this approach, Hybrigenics and the Institute Curie (Paris) have established a strategic alliance for investigating major cancer-relevant pathways using comparative proteomic analysis in human and Drosophila. The resulting Drosophila Protein Interaction Map, containing 1727 proteins (12% of the proteome), connected by over 2300 protein–protein interactions has been made freely available to the scientific community (http://www.pim. hybrigenics.com).
F. Colland, L. Daviet / Biochimie 86 (2004) 625–632
631
Fig. 4. RNF11 protein interaction map. Part of the interaction map surrounding the RING finger protein RNF11. The interactors of RNF11 can be subdivided in three groups: (i) effectors of the Smad pathway; (ii) ubiquitin and ubiquitin-like proteins; and (iii) HECT-containing E3 ubiquitin ligases.
8. Conclusion The availability of genome sequences for large numbers of organisms provides a major impetus for functional proteomic analyses. Of the methods used, the yeast two-hybrid system combines undeniable advantages for large-scale functional biology studies and recent methodological optimizations of the technology have greatly increased its comprehensiveness. However, appropriate bioinformatics tools for the exploration and analysis of the resulting large sets of data and for integrating heterogeneous genomic and proteomic analysis are becoming crucial for the in-depth analysis of extended protein networks. Finally, combining protein interaction mapping with functional assays should greatly facilitate the experimental validation of interactome-derived, functional assignment. Ultimately, the application of this strategy to disease-related pathways may lead to the identification and validation of novel therapeutic targets.
Acknowledgements We thank all Hybrigenics staff for their contribution.
References [1] [2]
[3]
M. Vidal, P. Legrain, Yeast forward and reverse ‘n’-hybrid systems, Nucleic Acids Res. 27 (1999) 919–929. P.L. Bartel, J.A. Roecklein, D. SenGupta, S. Fields, A protein linkage map of Escherichia coli bacteriophage T7, Nat. Genet. 12 (1996) 72–77. M. Flajolet, G. Rotondo, L. Daviet, F. Bergametti, G. Inchauspe, P. Tiollais, C. Transy, P. Legrain, A genomic approach of the hepatitis C virus generates a protein interaction map, Gene 242 (2000) 369– 379.
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
S. McCraith, T. Holtzman, B. Moss, S. Fields, Genome-wide analysis of vaccinia virus protein–protein interactions, Proc. Natl. Acad. Sci. USA 97 (2000) 4879–4884. J.C. Rain, L. Selig, H. De Reuse, V. Battaglia, C. Reverdy, S. Simon, G. Lenzen, F. Petel, J. Wojcik, V. Schachter, et al., The protein–protein interaction map of Helicobacter pylori, Nature 409 (2001) 211–215. T. Ito, T. Chiba, R. Ozawa, M. Yoshida, M. Hattori, Y. Sakaki, A comprehensive two-hybrid analysis to explore the yeast protein interactome, Proc. Natl. Acad. Sci. USA 98 (2001) 4569–4574. P. Uetz, L. Giot, G. Cagney, T.A. Mansfield, R.S. Judson, J.R. Knight, D. Lockshon, V. Narayan, M. Srinivasan, P. Pochart, et al., A comprehensive analysis of protein–protein interactions in Saccharomyces cerevisiae, Nature 403 (2000) 623–627. S. Li, C.M. Armstrong, N. Bertin, H. Ge, S. Milstein, M. Boxem, P.O. Vidalain, J.D. Han, A. Chesneau, T. Hao, et al., A map of the interactome network of the metazoan C. elegans, Science 303 (2004) 540–543. L. Giot, J.S. Bader, C. Brouwer, A. Chaudhuri, B. Kuang, Y. Li, Y.L. Hao, C.E. Ooi, B. Godwin, E. Vitols, et al., A protein interaction map of Drosophila melanogaster, Science 302 (2003) 1727–1736. A.J. Walhout, R. Sordella, X. Lu, J.L. Hartley, G.F. Temple, M.A. Brasch, N. Thierry-Mieg, M. Vidal, Protein interaction mapping in C. elegans using proteins involved in vulval development, Science 287 (2000) 116–122. S.J. Boulton, A. Gartner, J. Reboul, P. Vaglio, N. Dyson, D.E. Hill, M. Vidal, Combined functional genomic maps of the C. elegans DNA damage response, Science 295 (2002) 127–131. P. Legrain, J. Wojcik, J.M. Gauthier, Protein–protein interaction maps: a lead towards cellular functions, Trends Genet. 17 (2001) 346–352. P.J. Adam, R. Boyd, K.L. Tyson, G.C. Fletcher, A. Stamps, L. Hudson, H.R. Poyser, N. Redpath, M. Griffiths, G. Steers, et al., Comprehensive proteomic analysis of breast cancer cell membranes reveals unique proteins with potential roles in clinical cancer, J. Biol. Chem. 278 (2003) 6482–6489. F. Colland, X. Jacq, V. Trouplin, C. Mougin, C. Groizeleau, A. Hamburger, et al., Functional proteomic mapping of a human signaling pathway, Genome Res. 14 (2004) 1324–1332. A.D. Strosberg, L. Daviet, C. Borg-Capra, Functional proteomics applied to obesity research, in: G. Medeiros-Neto, A. Halpern, C. Bouchard (Eds.), Progress in Obesity Research, John Libbey Eurotext Ltd, 2003, pp. 229–233.
632
F. Colland, L. Daviet / Biochimie 86 (2004) 625–632
[16] E. Formstecher, S. Aresta, V. Collura, A. Hamburger, A. Meil, A. Trehin, C. Reverdy, V. Betin, S. Maire, C. Brun, et al., High fidelity protein interaction mapping: a Drosophila case study, (submitted for publication). [17] B. Boeda, A. El-Amraoui, A. Bahloul, R. Goodyear, L. Daviet, S. Blanchard, I. Perfettini, K.R. Fath, S. Shorte, J. Reiners, et al., Myosin VIIa, harmonin and cadherin 23, three Usher I gene products that cooperate to shape the sensory hair cell bundle, EMBO J. 21 (2002) 6689–6699. [18] M. Fromont-Racine, J.C. Rain, P. Legrain, Toward a functional analysis of the yeast genome through exhaustive two-hybrid screens, Nat. Genet. 16 (1997) 277–282. [19] J. Wojcik, I.G. Boneca, P. Legrain, Prediction, assessment and validation of protein interaction maps in bacteria, J. Mol. Biol. 323 (2002) 763–770. [20] H. Hermjakob, L. Montecchi-Palazzi, G. Bader, J. Wojcik, L. Salwinski, A. Ceol, S. Moore, S. Orchard, U. Sarkans, C. Von Mering, et al., The HUPO PSI’s molecular interaction format—a community standard for the representation of protein interaction data, Nat. Biotechnol. 22 (2004) 177–183. [21] H. Ge, Z. Liu, G.M. Church, M. Vidal, Correlation between transcriptome and interactome mapping data from Saccharomyces cerevisiae, Nat. Genet. 29 (2001) 482–486. [22] P. Durand, C. Medigue, A. Morgat, Y. Vandenbrouck, A. Viari, F. Rechenmann, Integration of data and methods for genome analysis, Curr. Opin. Drug Discov. Dev. 6 (2003) 346–352.
[23] M. Tewari, P.J. Hu, J.S. Ahn, N. Ayivi-Guedehoussou, P.O. Vidalain, S. Li, S. Milstein, C.M. Armstrong, M. Boxem, M.D. Butler, et al., Systematic interactome mapping and genetic perturbation analysis of a C. elegans TGF-beta signaling network, Mol. Cell 13 (2004) 469– 482. [24] T. Bouwmeester, A. Bauch, H. Ruffner, P.O. Angrand, G. Bergamini, K. Croughton, C. Cruciat, D. Eberhard, J. Gagneur, S. Ghidelli, et al., A physical and functional map of the human TNF-alpha/NF-kappaB signal transduction pathway, Nat. Cell. Biol. 6 (2004) 97–105. [25] J. Massague, TGF-beta signal transduction, Annu. Rev. Biochem. 67 (1998) 753–791. [26] D. Bennett, L. Alphey, PP1 binds Sara and negatively regulates Dpp signaling in Drosophila melanogaster, Nat. Genet. 31 (2002) 419– 423. [27] G.P. Raju, N. Dimova, P.S. Klein, H.C. Huang, SANE, a novel LEM domain protein, regulates bone morphogenetic protein signaling through interaction with Smad1, J. Biol. Chem. 278 (2003) 428–437. [28] V. Subramaniam, H. Li, M. Wong, R. Kitching, L. Attisano, J. Wrana, J. Zubovits, A.M. Burger, A. Seth, The RING-H2 protein RNF11 is overexpressed in breast cancer and is a target of Smurf2 E3 ligase, Br. J. Cancer 89 (2003) 1538–1544. [29] K. Jiao, Y. Zhou, B.L. Hogan, Identification of mZnf8, a Mouse Kruppel-like transcriptional repressor, as a novel nuclear interaction partner of Smad1, Mol. Cell. Biol. 22 (2002) 7633–7644.