From protein networks to biological systems

Transcription

1 FEBS FEBS Letters 579 (2005) Minireview From protein networks to biological systems Peter Uetz a,1, Russell L. Finley Jr. b, * a Research Center Karlsruhe, Institute of Genetics, P.O. Box 3640, D Karlsruhe, Germany b Center for Molecular Medicine & Genetics, Wayne State University School of Medicine, 540 East Canfield Avenue, Detroit, MI 48201, USA Accepted 31 January 2005 Available online 7 February 2005 Edited by Robert Russell and Giulio Superti-Furga Abstract A system-level understanding of any biological process requires a map of the relationships among the various molecules involved. Technologies to detect and predict protein interactions have begun to produce very large maps of protein interactions, some including most of an organismõs proteins. These maps can be used to study how proteins work together to form molecular machines and regulatory pathways. They also provide a framework for constructing predictive models of how information and energy flow through biological networks. In many respects, protein interaction maps are an entrée into systems biology. Ó 2005 Federation of European Biochemical Societies. Published by Elsevier B.V. All rights reserved. Keywords: Interaction map; Protein network; Pathway; System 1. Introduction Systems Biology should give us the tools to model how genes, gene products, and other molecules work together to mediate biological processes. Use of such tools, and indeed their very development, requires, for each biological process, lists of the molecules involved and their interconnections. The genes and proteins predicted from genome sequences have provided a long list of parts (genes and gene products), and new technologies have begun to define lists of other molecules not directly encoded by the genome that are present in cells and tissues at particular times. New computational and experimental technologies have begun to produce enormous datasets representing interactions between the parts. For the moment, most of the interaction data comes from technologies to detect physical or functional interactions between genes and proteins. Here, we will review some of the sources of these data and consider how the quantity and quality of the available interaction data may impact systems-level studies. 2. Protein protein interactions The prominent role that protein protein interactions play in most biological processes, combined with the fact that we * Corresponding author. Fax: addresses: peter.uetz@itg.fzk.de (P. Uetz), rfinley@wayne.edu (R.L. Finley Jr.). 1 Fax: know so little about the functions of most proteins, has inspired efforts to map interactions on a proteome-wide scale (e.g., for all of the proteins encoded by a genome) [1]. To date, most of the interactions that have been detected experimentally have relied on one of two technologies, the yeast two-hybrid system [2] and mass spectrometry (MS) identification of proteins that co-affinity purify (co-ap) with a bait protein [3]. The two technologies detect complementary types of interactions. Co-AP/MS identifies the constituents of multi-protein complexes but does not reveal the individual binary contacts that make up each complex. Without data on the constituent binary contacts, the possible paths of energy or information flow through the complex and its relationship to other cellular components may not be apparent. Yeast two-hybrid data, on the other hand, identifies likely binary interactions that may suggest possible paths through a pathway or complex, but cannot reveal the constituents of multiprotein complexes. Thus, both types of data will be important for understanding protein and pathway function, and ideally both approaches would be performed on a proteome-wide scale. Yeast two-hybrid screens aiming to cover entire proteomes, or at least very large numbers of proteins, have detected thousands of interactions for a few eukaryotic model organisms (Table 1), bacteria and phage [4,5] and viruses [6]. By contrast, proteome-wide co-ap/ms screens have been conducted only in yeast (Table 1), where most of the proteome could be easily affinity tagged through the use of homologous recombination. Co-AP/MS data for other organisms is only just beginning to emerge through the use of high throughput cloning [7] and the expression of large sets of tagged proteins in tissue culture cells (e.g. [8,9]). Thus, it is likely that we will begin to see protein complex data for humans and other metazoans in similar quantities as the yeast studies have produced. 3. How complete are current protein interaction datasets? Despite the volumes of interaction data produced, several independent analyses have shown that the data from the large two-hybrid and co-ap/ms screens is far from complete. Various authors have estimated that the roughly 6000 yeast proteins are connected by interactions [10 12], yet the high throughput screens have detected only a small fraction of those numbers (Table 1). Another clue comes from the lack of overlap among the different datasets for a particular proteome. For example, in Table 1, the overlap among the large two-hybrid screens for yeast was only 6 interactions /$30.00 Ó 2005 Federation of European Biochemical Societies. Published by Elsevier B.V. All rights reserved. doi: /j.febslet

2 1822 P. Uetz, R.L. Finley Jr. / FEBS Letters 579 (2005) Table 1 Large protein interaction screens for eukaryotes Organism (genes) Method Interactions a Proteins Reference Yeast (6000) Yeast two-hybrid [63] Yeast two-hybrid [13] Yeast two-hybrid [64,65] Co-AP/MS [66] Co-AP/MS [67] Drosophila (14 000) Yeast two-hybrid [49] Yeast two-hybrid [14] Worm (20 000) Yeast two-hybrid [68] a For two-hybrid screens, the approximate number of unique binary interactions is shown. For co-ap/ms screens, the approximate number of binary interactions that would result if each bait protein contacted every protein that co-purified with it (the hub and spoke model) is shown. Data can be retrieved from one of the databases cited [42 44]. [13] while the overlap between the screens for Drosophila was a measly 28, or less than 2% of the smallest data set [14]. The co- AP/MS data is not much different. For example, when results from the two large-scale studies are compared, the number of interactions common to both datasets is less than 9% of the total in both datasets [15]. Data from the high throughput screens also fails to overlap significantly with published low throughput studies, which are generally considered to be less subject to false positives and false negatives. Such analyses have led some authors to estimate false negative rates as high as 85% in large yeast two-hybrid screens and 50% in co-ap/ MS screens [16,17]. These results suggest that many more interactions could be detected by more exhaustive application of these technologies. In addition, there is a need for improved or new high throughput technologies to identify interactions that may be difficult to detect with two-hybrid or co-ap/ms, such as interactions involving membrane proteins. 4. Physical and functional interactions Comparison of the data from yeast two-hybrid and co-ap/ MS provides an example of an important distinction between two types of interaction data: physical interactions (A touches B) and functional interactions (A functions with B in some biological process). A functional relationship may or may not correspond to a direct physical interaction. Thus, physical and functional interactions are two distinct though partially overlapping types of interactions and the distinction is likely to be important for the development of systems-level models of protein networks and pathways. Yeast two-hybrid is an experimental approach to detect physical interactions. Co-AP/MS detects group of proteins in stable complexes, implying that they function together. Another example of a functional but not necessarily physical interaction is a genetic interaction, in which the combination of alleles of two different genes has specific phenotypic consequences. This is often taken to suggest that the two genes function in the same or parallel pathways affecting a particular biological process. Thus, a genetic interaction is a measured functional interaction that may or may not correspond to a physical interaction, but that could be usefully represented as a connection between the two genes or gene products. Ongoing large-scale screens in yeast have mapped thousands of genetic interactions [18]. Combination of genetic and physical interaction data is a powerful approach to mapping pathways [18,19]. 5. Predicted and experimentally measured interactions The increasing use of computational approaches to predict protein interactions has led to additional large datasets (e.g. [20]) and to another distinction between two types of interaction data: experimentally measured and predicted. Predicted interactions can also be classified as either physical or functional. Gene expression profiles have been used, for example, to infer functional interactions among gene products, based on the assumption that proteins that function together in the same pathway or complex should be frequently expressed together; which is supported by data for stable protein complexes [21,22]. Similarly, genes whose coexpression profiles are conserved through evolution are often functionally related [23,24], as are genes that are co-conserved from species to species [25 27]. The functional links between proteins in each of these cases may be direct or very general; they may suggest roles in the same pathways, or in distinct cellular systems that are concomitant but that have very few direct molecular connections. Genetic interactions have also been predicted based on physical interactions, gene expression, protein localization, and other experimental data [28,29]. Numerous methods for predicting physical protein protein interactions have also been developed [22,30 35]. One very powerful approach takes advantage of the large number of experimentally measured interactions available for organisms like yeast and Drosophila to predict interactions in other organisms [36]. Simply put, the approach predicts that two proteins will interact if their orthologs were shown to interact; such conserved interactions have been referred to as interologs [37 39]. This approach has been used, for example, to predict interactions involving proteins encoded by a third of the human genes [40]. Several other studies have begun to effectively integrate genomic and proteomic data to make increasingly accurate interaction predictions [33,41]. The further development and use of in silico approaches to map interactions seems particularly important in light of the shortcomings of high throughout experimental detection systems. 6. Protein interaction maps The wealth of data from high throughput screening and other studies has begun to be consolidated into centralized, standardized databases. Three of the largest public database repositories for interaction data are BIND, DIP, and IntAct

3 P. Uetz, R.L. Finley Jr. / FEBS Letters 579 (2005) Fig. 1. Interaction maps today and tomorrow. (A) Typical representation of a protein protein interaction map. (B) Proteins are usually shown as nodes (e.g., circles and boxes) and interactions as edges (lines) connecting them. This organizes information so that attributes of both proteins and interactions are easily accessible, for example, by hypertext links, but fails to capture structural information. However, additional information could be made easily accessible through pop-up menus by clicking on the edges (like here) or the nodes. (C) and (D) Ideally, a protein interaction map visualization tool would allow the structures of proteins and interaction interfaces to be expanded and browsed, and would also provide access to more global interaction attributes (e.g., conditions under which a certain set of interaction can be found, experimental conditions, dissociation constants, expression levels, etc.). Based on interactions published by Measday et al. [69], Uetz et al. [63], and reviewed in [70]. [42 44]. These allow researchers online access to browse and download data in a standardized format [45]. Sets of interaction data can be viewed as graphs or maps in which each gene/protein is a node and each interaction is a line connecting two nodes (Fig. 1). The importance of this view has led to use of the term interaction map to refer generically to interaction datasets. The map view provides not only an intuitive interface for biologists to explore the data, but also a formal mathematical framework for computational biologists to explore the properties of interaction networks. However, before interaction maps can be used to represent biological networks, their limitations must be considered. In addition to the problem of false negatives discussed previously, most interaction maps and particularly those from high throughput screens have false positives. Estimates of false positive rates vary widely, in part because of the difficulty in definitively demonstrating that any particular interaction does not have a biological function. Because the false positive rates may be substantial, the maps from high throughput studies might be usefully regarded as the results from a first pass filter, which reduces the possible search space for functionally important interactions. Thus, the question becomes how to identify the more likely true positives. Several studies have confirmed the general principle that interactions detected in multiple screens and by different techniques or in different species are more likely to be true positives than those only found once or twice [17]. Due to the high rates of false negatives in high throughput screens, however, there has been very little overlap between different datasets, thus, limiting the opportunities for such experimental cross-validation. Alternatively, a variety of confidence scoring systems have been developed that calculate the likelihood of an interaction being a true positive, based on various parameters, including attributes of the proteins and the specific assays, whether the interaction was detected by other technologies or screens, and network topology [20,46 50]. However, thus far most of these scoring systems are specific to particular datasets or methodologies, and no universal system has yet been effective. Another limitation of most protein interaction maps is that each node generally represents some generic version of a

4 YDR009W YBR020W YML051W YDR532C YHR079C YGL174W YMR057C YIR005W YMR297W YLR055C YBR081C YMR214W YML043C YBL014C YJL025W YDR498C YFL036W YMR228W YLR294C YCR045C YNL151C YOR348C YDR142C YHR160C YOL105C YPR190C YDR383C YGL153W YLR191W YNR003C YDL078CYMR291W YLR315W YJR083C YJL084C YOR110W YHL042W YPL248C YOL051W YGR252W YJL034W YCL020W YMR232W YLR447C YDR244W YLR102C YGL013C YOR298CA YPL019C YGR238C YOR331C YBR123C YJL114W YCL046W YBL010C YDL214C YJL185C YOL148C YBR108W YKR022C YPR105C YEL062W YHR060W YJR122WYFL002WA YNL257C YKR062W YCR009C YMR092C YOR254C YGR258C YGL115W YHR206W YDR176W YPL082C YDR277C YLR190W YML110C YKL081W YGR246C YDR389W YLR150W YNL042W YOR127W YDR448W YOR138C YHR158C YLR015W YER148W YLR292C YIL143C YOR262W YJR072C YDR392W YDR422C YCR042C YGL208W YDL194W YDL135C YER171W YLR324W YLR424W YGR274C YML042W YDR017C YDR129C YGR032W YBR198C YER059W YCR088W YDL178W YER068W YLR371W YOL012C YJL162C YLL050C YGL126W YDR216W YKL028W YDR099W YLR249W YAL003W YIL038C YBL043W YMR181C YGR268C YLR005W YPR165W YDR309C YDR388W YJL137C YNL138W YHR128W YDR469W YBR118W YPR072W YFR042W YDL138W YML114C YLR273C YJL089W YML109W YGR253C YDR477W YDR329C YGL180W YBL101W-A YGR047C YNL039W YPR159W YPR080W YIR024C YLR342W YDR376W YCR093WYLR243W YOR250C YGR123C YBR171W YDL089W YER177W YBR140C YNL126W YIL163C YJL019W YKL020C YKL129C YLR049C YHL007C YNL298W YGR241C YJL178C YPL161C YIL123W YML038C YER027C YHR071W YGR080W YNL122C YPR049C YHR016C YDL165W YOR062C YJL005W YFL039C YPR185W YGR178C YGR160W YMR316CB YGR167W YKL061W YNL020C YOR047C YIL159W YKR055W YBR170C YCR035C YLR258W YBR196C YAL028W YDR311W YPL219W YHR172W YGL206C YDL179W YBL105C YGL154C YLR212C YNR025C YDR463W YDL065C YNL188W YLR229C YOR257W YPR180W YDR167W YLR337C YPR040W YER040W YGL134W YNR032W YOR181W YHR102W YDR228C YMR032W YOL144W YNL289W YOR122C YNL098C YHR215W YHR161C YDR482C YNL090W YJL085W YOR353C YOL101C YPL211W YLR233C YDL127W YIL050W YNL243W YBL085W YBR274W YOR275C YHR061C YNL086W YJL211C YER165W YBR155W YGL215W YGR152C YPL222W YGR120C YLL016W YMR009W YGL028C YIL045W YBL007C YNL229C YGR162W YMR109W YGL254W YNL271C YIL118W YDL070W YGR048W YGL229C YOR346W YMR019W YNL335W YMR039C YIR037W YMR139W YNL218W YLR432W YPL031C YGR153W YDR356W YDR390C YNL309W YOR276W YOR101W YMR186W YBR200W YMR311C YMR102C YER114C YJR094C YLR305C YNL201C YJL157C YHR200W YOL139C YDR382W YMR055C YKL042W YDR028C YGL073W YOR180C YPL124W YLR277C YGR233C YOR326W YNL225C YGL044C YGL240W YLR071C YOL082W YNL084C YKL072W YLR284C YCL055W YGL192W YOL016C YIR006C YBR050C YGL024W YDR169C YOR373W YOL004W YHR197W YFR036W YLR216C YBL075C YDL226C YAL041W YGR068C YBL102W YHR100C YHR178W YLR423C YPL242C YML057W YOR178C YDR207C YPR115W YBR057C YIL061C YGL019W YNL330C YBL084C YIL035C YMR061W YOR231W YHR166C YER149C YBR244W YMR255W YPR086W YLR147C YJR093C YKR002W YMR180C YGL116W YJR032W YDR214W YLR310C YBR190W YMR053C YML015C YOR212WYDR264C YJL095W YKL178C YNL116W YKL022C YOR027W YGL015C YOR315W YOL130W YKL103CYFL056C YDR122W YJR086W YER029C YOR061W YKL112W YDL161W YOR105W YGR014W YDL020C YDL008W YKL117W YDR146C YBR006W YDR206W YKL008C YJL064W YLR045C YAR014C YKL204W YKL095W YBR045C YDL215C YOL001W YAR018C YBR109C YNR029C YOR329C YJL030W YDL106C YJL065C YGL112C YGL086W YIL007C YJL047C YLL021W YHR140W YER133W YER124C YER143W YER079W YCR011C YMR267W YIL001W YOR039W YER054C YJL013C YPR025C YOR151C YCR005C YIL004C YPR008W YHR185C YCL059C YHR014W YDL012C YPL218W YML103C YLR363C YGR188C YIL047C YDR130C YDL101C YOR026W YHR038W YDR151C YJR109C YFR002W YJR056C YPL140C YJL042W YLR127C YHR005C YKL161C YFR047C YLR117C YPR107C YHR135C YLL046C YAL047C YNL154C YLR319C YPL240C YOR303W YJL200C YOR115C YKL193C YKL023W YLL024C YLR362W YHR198C YMR095C YNL112W YKR030W YGL049C YOR355W YHR032W YDR335W YJL151C YLR291C YIL072W YPL089C YAL030W YLR078C YPR048W YDR103W YLR313C YGL025C YPL192C YDR026C YDL108W YML064C YDL097C YLL019C YOL133W YFL059W YOR292C YLR166C YOL123W YJL141C YPR070W YPL204W YMR096W YLR263W YOR327C YFR034C YIL172C YNL251C YGL212W YLR268W YGL130W YPL070W YNL146W YDR195W YGL201C YDR472W YDR174W YGR003W YHR145C YDR110W YCR087W YMR071C YGR172C YER031C YKR026C YAL034WA YCL054W YDL229W YMR129W YNL027W YKR068C YNL333W YIL113W YDR460W YGL122C YGR009C YHR030C YBR254C YMR322C YDR416W YER032W YNR023W YPL228W YMR183C YBR205W YHL046C YLR433C YPL022W YBR188C YEL009C YDL159W YLL026W YGR046W YIL011W YDR211W YMR269W YJR034W YGL150C YDR246W YGL198W YFL038C YML077W YGR129W YOR260W YOR106W YKL196C YFR052W YJL057C YIL013C YMR236W YDR468C YDL246C YER161C YDL002C YKL190W YFL060C YGL161C YGR083C YDR273W YDR084C YGR117C YLR452C YNL091W YMR211W YLR384C YJR007W YNL064C YMR080C YFR033C YDR061W YER132C YDL140C YNL164C YNL288W YNL263C YHR077C YAL005C YOR036W YGR072W YNL236W YBR236C YHR105W YPL246C YDR172WYER105C YBL016W YER092W YGL043W YML095C YLR182W YPL038W YOR370C YGL095C YBR143C YJL128C YPR187W YPL237W YMR197C YOR359W YOL018C YOR208W YKR034W YBR134W YNL199C YOR284W YDR253C YMR153W YGR040W YER118C YKL075C YPL229W YDR323C YOL149W YOR341W YOL129W YBL050W YCL032W YJL110C YPR041W YER179W YDL056W YER047C YLR269C YJR077C YOR034C YGR155W YJL058C YJR131W YNL047C YJR159W YMR201C YLL036C YHL006C YGL104C YML001W YBR080C YAL040C YNL021W YNL312W YJR050W YOR361C YDL154W YIL144W YBR270C YER075C YLR046C YPR103W YMR309C YER100W YGL175C YHR152W YER112W YDR412W YNL118C YCR086W YBR133C YNL244C YER111C YJL092W YDR078C YOR089C YPL260W YLR113W Y EL1 YEL015W YGL158W YOR128C YMR213W YMR087W YIL150C YPL049C YER012W YGL178W YNL103W YHR039C M YOR157C YML097C YMR012W YMR094W YNL135C YBR079C YLL053C YOR264W YJR133W YDL203C YDR357C YER136W YPL075W YFL003C YPR019W YMR146C YDR106W YDR128W YJL187C YGR058W YOL106W YML032C YLL039C YHR129C YPL059W YGR136W YPL238C YIR017C YOR229W YDR480W YPR101W YOL059W YOL020W YOR167C YPR182W YDR464W YDR054C YDR378C YLR208W YAR007C YER052C YML031W YJL173C YCR067C YOR372C YKL012W YPL174C YIL046W YHL031C YOR230W YLR275W YGL170C YNL233W YBR023C YPL151C YPL256C YBR138C YBR017C YNL147W YER095W YFL005W YDR369C YJL184W YGR179C YIL105C YJR132W YMR068W YDL088C YDL195W YPL213W YDR429C YBL079W YPR011C YBL061C YML046W YBR202W YDR326C YKL145W YGR119C YJR022W YDR070C YPR163C YDR439W YMR117C YOR362C YPR017C YFL017C YER007CA YKL203C YJR060W YGL238W YLR026C YGR140W YMR224C YHR084W YLR264W YPL085W YDR201W YPL018W YDR032C YBL026W YBR094W YEL037C YHR083W YEL032W YMR233W YNL092W YMR168C YJL124C YIL109C YDL043C YOR160W YKL074C YPR119W YNL250W YDL064W YGL172W YDR076W YOR259C YOR191W YKL068W YDR510W YGR057C YNR022C YLR082C YKL049C YOL034WYPR010C YGL048C YNL307C YKL089W YDL160C YML049C YPR181C YPL268W YOR159C YMR199W YOL006C YKR010C YIR009W YLR167W YER102W YAL038W YLR103C YHR170W YER018C YCR004C YDR189W YER162C YLR175W YLR274W YHR165C YER110C YGL197W YNL272C YLR116W YDR318W YGL092W YOR020C YHR089C YBR052C YER146W YOR375C YLR293C YNL216W YLR347C YDR004W YDL071C YAL032C YNL286W YNL004W YDR432W YDR386W YMR308C YLR335W YHR098C YBR211C YKR037C YDL147W YOR098C YKL015W YKL002W Y SM3 YNL273W YNL155W YEL068C YBR135W YDL132W YKL155C YDL111C YPR082C YJR052W YDR171W YBR160W YBR172C YBR055C YDR394W YDR183W YDR148C YPL105C YHR035WYDR139C YKL173W YNL023C L YBR275C YDR192C YLR419W YFL018WA YDL030W YGR158C YDL110C YCL066W YNL036W YGR218W YDL013W YDR315C YBL023C YMR047CYDL207W YIL115C YPR145W YFR046C YLR456W YER045C YLR259C YCR002C YLR323C YOR078W YLR392C YDR179C YOL021C YKR060W YER009W YJR090C YOR206W YOR117W YDL155W YLR100W YDR132C YAL021C YNR069C YDR104C YNR052C YIR025W YER116C YBR114W YGR108W YDR328C YMR043W YCR073C YIL063C YJR074W YGR232W YJR091C YOR319W YIL125W YJR117W YGR269W YJL194W YGL249W YKL148C YLL041C YLR210W YLR429W YLR345W YJL203W YGL096W YGL221C YFL029C YLR442C YDR485C YJL041W YHR057C YLR128W YER107C YFL009W YMR025W YOR057W YNL189W YKL130C YPR054W YKL101W YLR224W YML088W YLR352W YKR101W YLR097C YLR399C YDR227W YOL058W YNL031C YJL061W YMR125W YOR038C YML065W YBR221C YGL097W YPL133C YDL029W YBR234C YLL049W YJL218W YDL216C YOR185C YCR039C YLR006C YIL065C YBR009C YNR068C YBR103W YJR065C YBL008W YDR052C YCL067C YLR314C YDR507C YBR010W YJL048C YMR048W YPR062W YDL017W YOR204W YCR084C YMR314W YER065C YDR002W YIL112W YPR020WYDR503C YDR313C YGR063C YML010W YNL030W YCR097W YPL178W YGR116W YKL205W YLR079W YLR321C YFL061W YDL235C YKR048C YPL214C YNR031C YMR001C YMR284W YNR048W YER025W YDR395W YKL039WYGR099W YDL236W YIL147C YPR120C YDL042C YLR368W YCL024W YPL153C YML028W YBR252W YOR156C YFR028C YMR106C YCR050C YPL111W YMR138W YKL144C YDL225W YJL076W YFR057W YHL009C YHR107C YJL088W YMR226C YJR076C YDR255C YLR303W YLR109W YLR176C YDR225W YCR022C YDR085C YDR217C YNL078W YDL117W YDR218C YOR006C YKL135C YBR284W YIL132C YIL126W YPL241C YOL070C YOL127W YLR406C YOR070C YBR112C YDR043C YDL075W YHR075C YEL023C YPR102C YGR085C YHR141C YNR050C YLR322W YMR035W YLR376C YPR093C YDL011C YPR078C YDL051W YHL019C YPL259C YDR087C YFR037C YHR106W YGR024C YMR091C YMR150C YJL036W YDL113C YBL025W YMR270C YOR164C YPL025C YGL035C YGL145W YLR075W YIR012W YHR216W YDR123C YKL017C YOL108C YPL110C YMR317W YNL279W YIL082W YKL090W YPL128C 1824 P. Uetz, R.L. Finley Jr. / FEBS Letters 579 (2005) Galactose Gal2 Gal7 Galactose Gal1-P Glu1-P Glucose-6-P Gal1 Gal10 Gal5 Other Translation Translation processes mrna mrna Cytoplasm mrna RNA-processing + export factors?gal80 P? RNA-processing factors Gal80 Gal80 Gal80 Gal80 Gal80 DNA Gal1 Gal10 Gal7 Gal7 X X Nucleus P P P P Srb10 Srb11 Gal11 RNA pol II holoenzyme General TAFs Gal11 Chromatin TRENDS in Cell Biology Gal1 Gal80 Gal11 Fig. 2. Integrating protein networks and other biological information. (Bottom) The map of interactions among 1200 yeast proteins represents only the tip of the information iceberg. Highlighted in blue are proteins involved in galactose regulation, which in the map are found in a topological cluster (see text). A cluster of cytoskeletal proteins (highlighted in red) is also visible. (Top) Integrating the protein interaction network with the metabolic network (shown as green compounds and grey enzymes), the gene regulatory network (shown as pink genes and blue transcription factors), and the signaling network indicated by the Cyclin-dependent kinase Srb10 and its cyclin Srb11 (red). Yellow boxes indicate functional modules that involve additional protein interactions within complexes and with other proteins (not shown). Modified after Tucker et al. [12] and Rohde et al. [71]. cellular protein, without regard to the various splice variants or post-translationally modified forms that may exist. Isoforms could interact differently from the form that was actually used in the assay, which in many cases is unclear. This is particularly true for assays that use only one or a small number of the possible alternative transcripts from each gene. Thus, many

5 P. Uetz, R.L. Finley Jr. / FEBS Letters 579 (2005) so-called protein interactions maps are actually gene or locus interaction maps, which tell us only that one or more of the proteins encoded by one locus is capable of interacting with one or more of the proteins encoded by another locus. Nevertheless, such maps have proven to be useful as starting points for additional studies, particularly if the caveats are borne in mind. 7. Using the interaction maps for systems biology A complete systems-level understanding of any biological process may require more input data than current technologies can offer. Intuitively, we imagine that we could model a system best only after knowing all of the molecules involved, their concentrations, how they fit together, the effect of each individual part on its neighbors, and dynamic parameters such as how concentrations, interactions, and mechanics change over time. But this seems unrealistic given the fact that the high throughput technologies for measuring many of these parameters are still on the drawing board, if they exist at all. Do we really need to know all of the details of a process to be able to develop a useful systems-wide understanding or to have a predictive model? Analysis of protein interaction maps has suggested that even sparse data can be used to derive initial, rudimentary models of biological networks. Topological analyses, for example, initially of metabolic pathways and subsequently of protein interaction maps, began to reveal some common properties of biological networks [51,52]. These initial studies suggested the exciting possibility that cellular networks may be organized according to some general principles that could be understood without a detailed knowledge of all the constituent proteins and interactions [53,54]. Moreover, analysis of network topology can provide insights into protein and pathway function. For example, protein networks contain highly connected hub proteins, which have been shown to correlate with evolutionarily conserved proteins, and in yeast with proteins encoded by essential genes [51,55,56]. Thus, a proteinõs relative position in a network has implications for its function and importance. Analysis of topology also reveals clusters of highly interconnected proteins that correlate with conserved functional modules (Fig. 2), such as protein complexes or signaling pathways [57 59]. Thus, even the currently available noisy protein interaction maps can be used to explore the hierarchical organization of biological networks and to reveal interconnected modules that control specific biological processes. As these modules are defined and further elaborated, understanding them and their higher order organization will increasingly rely on advances in information technology. a map of interactions [60 62]. These programs allow exploration and ad hoc analyses of interaction data but they rarely incorporate all of the useful available information about the molecules and interactions they represent (e.g., see Figs. 1 and 2). Moreover, they usually fail to capture the essential dynamic properties of biological networks. Animated cartoons, on the other hand, can provide at least a qualitative representation of the dynamics of a process (see, for example, Fig. 3). However, such oversimplification does not capture the details of the system or facilitate quantitative modeling. In a way, visualization of molecular networks is where word processing was in the early 1980s. To help us model biological processes, and to visualize and manipulate those models, we need programs to generate more dynamic and realistic representations of biological events and structures. We need what might be called interactive Cell- TV to visualize and manipulate models of cellular events and behavior. Importantly, icell-tv must operate across several scales of time and space to allow biologists to navigate all available relevant information. Such a system, for example, might allow users to explore the changes in the molecular structure resulting from a post-translational modification, zoom out to witness the subsequent changes in network and pathway dynamics, and then change time scales to observe organelle movement or cell behavior. The number and complexity of the experiments that must be done to test hypotheses 8. Perspectives: icell-tv How can biologists access and integrate the deluge of proteomics data to help them understand biology? While this information should help drive the generation of hypotheses and hypothesis-testing research, we may be generating data faster than we are learning how to use it. Tools for accessing and analyzing molecular interaction data have just begun to emerge over the past few years. Several visualization tools and graphing programs, for example, allow users to construct Fig. 3. Visualizing the dynamics of protein interaction and signaling networks. The pheromone signaling pathway in yeast is a highly dynamic process that involves numerous protein interactions, phosphorylation events, and small-molecule interactions involving ATP and GTP. Typical textbook (i.e., static) representations like this do not reflect the dynamics of this process. A more realistic representation is available through animation, as shown at MAPK/MAPk.htm. Simplified from an animation by Tom Dallman, by permission of the author.

6 1826 P. Uetz, R.L. Finley Jr. / FEBS Letters 579 (2005) coming from network analyses are likely to be costly and inefficient. The next generation of biological information management systems must, therefore, allow us to do biology truly in silico. For this to be possible, they must enable the development and manipulation of quantitative models, which are often initially based on a qualitative understanding. However, it is often the case that about the time we understand a system well enough to be able to model it, it becomes too hard to understand in a qualitative sense. A system for navigating qualitative information based on quantitative data would give users the ability not only to understand the complexity of biological processes but also to manipulate those processes, to construct new models, and to test new hypotheses. Zoom in, change a K d or a V max, then zoom out and watch what happens to the system. This would be systems biology for the rest of us, and would open biological inquiry to a vast resource of creativity. Acknowledgments: We thank Tom Dallman for permission to use his pheromone pathway animation. References [1] Phizicky, E., Bastiaens, P.I., Zhu, H., Snyder, M. and Fields, S. (2003) Protein analysis on a proteomic scale. Nature 422, [2] Fields, S. and Sternglanz, R. (1994) The two-hybrid system: an assay for protein protein interactions. Trends Genet. 10, [3] Pandey, A. and Mann, M. (2000) Proteomics to study genes and genomes. Nature 405, [4] Rain, J.C., et al. (2001) The protein protein interaction map of Helicobacter pylori. Nature 409, [5] Bartel, P.L., Roecklein, J.A., SenGupta, D. and Fields, S. (1996) A protein linkage map of Escherichia coli bacteriophage T7. Nat. Genet. 12, [6] Uetz, P., Rajagopala, S.V., Dong, Y.A. and Haas, J. (2004) From ORFeomes to protein interaction maps in viruses. Genome Res. 14, [7] Marsischky, G. and LaBaer, J. (2004) Many paths to many clones: a comparative look at high-throughput cloning methods. Genome Res. 14, [8] Bouwmeester, T., et al. (2004) A physical and functional map of the human TNF-alpha/NF-kappa B signal transduction pathway. Nat. Cell Biol. 6, [9] Brajenovic, M., Joberty, G., Kuster, B., Bouwmeester, T. and Drewes, G. (2004) Comprehensive proteomic analysis of human Par protein complexes reveals an interconnected protein network. J. Biol. Chem. 279, [10] Walhout, A.J., Boulton, S.J. and Vidal, M. (2000) Yeast twohybrid systems and protein interaction mapping projects for yeast and worm. Yeast 17, [11] Grigoriev, A. (2003) On the number of protein protein interactions in the yeast proteome. Nucleic Acids Res. 31, [12] Tucker, C.L., Gera, J.F. and Uetz, P. (2001) Towards an understanding of complex protein networks. Trends Cell Biol. 11, [13] Ito, T., Chiba, T., Ozawa, R., Yoshida, M., Hattori, M. and Sakaki, Y. (2001) A comprehensive two-hybrid analysis to explore the yeast protein interactome. Proc. Natl. Acad. Sci. USA 98, [14] Stanyon, C.A., et al. (2004) A Drosophila protein-interaction map centered on cell-cycle regulators. Genome Biol. 5, R96. [15] Cornell, M., Paton, N.W. and Oliver, S.G. (2004) A critical and integrated view of the yeast interactome. Comp. Funct. Genom. 5, [16] Edwards, A.M., Kus, B., Jansen, R., Greenbaum, D., Greenblatt, J. and Gerstein, M. (2002) Bridging structural biology and genomics: assessing protein interaction data with known complexes. Trends Genet. 18, [17] von Mering, C., Krause, R., Snel, B., Cornell, M., Oliver, S.G., Fields, S. and Bork, P. (2002) Comparative assessment of largescale data sets of protein protein interactions. Nature 417, [18] Tong, A.H., et al. (2004) Global mapping of the yeast genetic interaction network. Science 303, [19] Tewari, M., et al. (2004) Systematic interactome mapping and genetic perturbation analysis of a C. elegans TGF-beta signaling network. Mol. Cell 13, [20] von Mering, C., et al. (2005) STRING: known and predicted protein protein associations, integrated and transferred across organisms. Nucleic Acids Res. 33 (Database Issue), D433 D437. [21] Jansen, R., Lan, N., Qian, J. and Gerstein, M. (2002) Integration of genomic datasets to predict protein complexes in yeast. J. Struct. Funct. Genom. 2, [22] Jansen, R., et al. (2003) A Bayesian networks approach for predicting protein protein interactions from genomic data. Science 302, [23] Bergmann, S., Ihmels, J. and Barkai, N. (2004) Similarities and differences in genome-wide expression data of six organisms. PLoS Biol. 2, E9. [24] Stuart, J.M., Segal, E., Koller, D. and Kim, S.K. (2003) A genecoexpression network for global discovery of conserved genetic modules. Science 302, [25] Pellegrini, M., Marcotte, E.M., Thompson, M.J., Eisenberg, D. and Yeates, T.O. (1999) Assigning protein functions by comparative genome analysis: protein phylogenetic profiles. Proc. Natl. Acad. Sci. USA 96, [26] Date, S.V. and Marcotte, E.M. (2003) Discovery of uncharacterized cellular systems by genome-wide analysis of functional linkages. Nat. Biotechnol. 21, [27] Bowers, P.M., Cokus, S.J., Eisenberg, D. and Yeates, T.O. (2004) Use of logic relationships to decipher protein network organization. Science 306, [28] Wong, S.L., et al. (2004) Combining biological networks to predict genetic interactions. Proc. Natl. Acad. Sci. USA 101, [29] Marcotte, E.M., Pellegrini, M., Thompson, M.J., Yeates, T.O. and Eisenberg, D. (1999) A combined algorithm for genome-wide prediction of protein function. Nature 402, [30] Aloy, P., et al. (2004) Structure-based assembly of protein complexes in yeast. Science 303, [31] Aloy, P. and Russell, R.B. (2002) Interrogating protein interaction networks through structural biology. Proc. Natl. Acad. Sci. USA 99, [32] Lu, L., Arakaki, A.K., Lu, H. and Skolnick, J. (2003) Multimeric threading-based prediction of protein protein interactions on a genomic scale: application to the Saccharomyces cerevisiae proteome. Genome Res. 13, [33] Zhang, L.V., Wong, S.L., King, O.D. and Roth, F.P. (2004) Predicting co-complexed protein pairs using genomic and proteomic data integration. BMC Bioinform. 5, 38. [34] Enright, A.J., Iliopoulos, I., Kyrpides, N.C. and Ouzounis, C.A. (1999) Protein interaction maps for complete genomes based on gene fusion events. Nature 402, [35] Reiss, D.J. and Schwikowski, B. (2004) Predicting protein peptide interactions via a network-based motif sampler. Bioinformatics 20 (Suppl 1), I274 I282. [36] Sharon, R., et al. (2005) Conserved patterns of protein interaction in multiple species. Proc. Natl. Acad. Sci. USA 102, [37] Walhout, A.J., Sordella, R., Lu, X., Hartley, J.L., Temple, G.F., Brasch, M.A., Thierry-Mieg, N. and Vidal, M. (2000) Protein interaction mapping in C. elegans using proteins involved in vulval development. Science 287, [38] Matthews, L.R., Vaglio, P., Reboul, J., Ge, H., Davis, B.P., Garrels, J., Vincent, S. and Vidal, M. (2001) Identification of potential interaction networks using sequence-based searches for conserved protein protein interactions or interologs. Genome Res. 11, [39] Yu, H., et al. (2004) Annotation transfer between genomes: protein protein interologs and protein DNA regulogs. Genome Res. 14, [40] Lehner, B. and Fraser, A.G. (2004) A first-draft human proteininteraction map. Genome Biol. 5, R63.

7 P. Uetz, R.L. Finley Jr. / FEBS Letters 579 (2005) [41] Lee, I., Date, S.V., Adai, A.T. and Marcotte, E.M. (2004) A probabilistic functional network of yeast genes. Science 306, [42] Alfarano, C., et al. (2005) The biomolecular interaction network database and related tools 2005 update. Nucleic Acids Res. 33 (Database Issue), D418 D424. [43] Salwinski, L., Miller, C.S., Smith, A.J., Pettit, F.K., Bowie, J.U. and Eisenberg, D. (2004) The database of interacting proteins: 2004 update. Nucleic Acids Res. 32 (Database issue), D449 D451. [44] Hermjakob, H., et al. (2004) IntAct: an open source molecular interaction database. Nucleic Acids Res. 32 (Database issue), D452 D455. [45] Hermjakob, H., et al. (2004) The HUPO PSIÕs molecular interaction format a community standard for the representation of protein interaction data. Nat. Biotechnol. 22, [46] Saito, R., Suzuki, H. and Hayashizaki, Y. (2003) Construction of reliable protein protein interaction networks with a new interaction generality measure. Bioinformatics 19, [47] Goldberg, D.S. and Roth, F.P. (2003) Assessing experimentally derived interactions in a small world. Proc. Natl. Acad. Sci. USA 100, [48] Deane, C.M., Salwinski, L., Xenarios, I. and Eisenberg, D. (2002) Protein interactions: two methods for assessment of the reliability of high throughput observations. Mol. Cell Proteomics 1, [49] Giot, L., et al. (2003) A protein interaction map of Drosophila melanogaster. Science 302, [50] Bader, J.S., Chaudhuri, A., Rothberg, J.M. and Chant, J. (2004) Gaining confidence in high-throughput protein interaction networks. Nat. Biotechnol. 22, [51] Jeong, H., Mason, S.P., Barabasi, A.L. and Oltvai, Z.N. (2001) Lethality and centrality in protein networks. Nature 411, [52] Jeong, H., Tombor, B., Albert, R., Oltvai, Z.N. and Barabasi, A.L. (2000) The large-scale organization of metabolic networks. Nature 407, [53] Strogatz, S.H. (2001) Exploring complex networks. Nature 410, [54] Barabasi, A.L. and Oltvai, Z.N. (2004) Network biology: understanding the cellõs functional organization. Nat. Rev. Genet. 5, [55] Said, M.R., Begley, T.J., Oppenheim, A.V., Lauffenburger, D.A. and Samson, L.D. (2004) Global network analysis of phenotypic effects: protein networks and toxicity modulation in Saccharomyces cerevisiae. Proc. Natl. Acad. Sci. USA 101, [56] Han, J.D., et al. (2004) Evidence for dynamically organized modularity in the yeast protein protein interaction network. Nature 430, [57] Poyatos, J.F. and Hurst, L.D. (2004) How biologically relevant are interaction-based modules in protein networks. Genome Biol. 5, R93. [58] Spirin, V. and Mirny, L.A. (2003) Protein complexes and functional modules in molecular networks. Proc. Natl. Acad. Sci. USA 100, [59] von Mering, C., Zdobnov, E.M., Tsoka, S., Ciccarelli, F.D., Pereira-Leal, J.B., Ouzounis, C.A. and Bork, P. (2003) Genome evolution reveals biochemical networks and functional modules. Proc. Natl. Acad. Sci. USA 100, [60] Shannon, P., et al. (2003) Cytoscape: a software environment for integrated models of biomolecular interaction networks. Genome Res. 13, [61] Breitkreutz, B.J., Stark, C. and Tyers, M. (2003) Osprey: a network visualization system. Genome Biol. 4, R22. [62] Mrowka, R. (2001) A Java applet for visualizing protein protein interaction. Bioinformatics 17, [63] Uetz, P., et al. (2000) A comprehensive analysis of protein protein interactions in Saccharomyces cerevisiae. Nature 403, [64] Drees, B.L., et al. (2001) A protein interaction map for cell polarity development. J. Cell Biol. 154, [65] Fromont-Racine, M., et al. (2000) Genome-wide protein interaction screens reveal functional networks involving Sm-like proteins. Yeast 17, [66] Gavin, A.C., et al. (2002) Functional organization of the yeast proteome by systematic analysis of protein complexes. Nature 415, [67] Ho, Y., et al. (2002) Systematic identification of protein complexes in Saccharomyces cerevisiae by mass spectrometry. Nature 415, [68] Li, S., et al. (2004) A map of the interactome network of the metazoan C. elegans. Science 303, [69] Measday, V., Moore, L., Retnakaran, R., Lee, J., Donoviel, M., Neiman, A.M. and Andrews, B. (1997) A family of cyclin-like proteins that interact with the Pho85 cyclin-dependent kinase. Mol. Cell. Biol. 17, [70] Carroll, A.S. and OÕShea, E.K. (2002) Pho85 and signaling environmental conditions. Trends Biochem. Sci. 27, [71] Rohde, J.R., Trinh, J. and Sadowski, I. (2000) Multiple signals regulate GAL transcription in yeast. Mol. Cell. Biol. 20,