- Email: [email protected]
A network medicine approach to human disease
FEBS Letters 583 (2009) 1759–1765
journal homepage: www.FEBSLetters.org
Minireview
A network medicine approach to human disease Andreas Zanzoni a, Montserrat Soler-López a, Patrick Aloy a,b,* a b
Institute for Research in Biomedicine (IRB) and Barcelona Supercomputing Center (BSC), c/ Baldiri i Reixac 10-12, 08028 Barcelona, Spain Institució Catalana de Recerca i Estudis Avançats (ICREA) Pg. Lluís Companys 23, 08010 Barcelona, Spain
a r t i c l e
i n f o
Article history: Received 2 February 2009 Accepted 2 March 2009 Available online 6 March 2009 Edited by Miguel De la Rosa Keywords: Network medicine Protein–protein interaction Protein network Drug discovery
a b s t r a c t High-throughput interaction discovery initiatives are providing thousands of novel protein interactions which are unveiling many unexpected links between apparently unrelated biological processes. In particular, analyses of the first draft human interactomes highlight a strong association between protein network connectivity and disease. Indeed, recent exciting studies have exploited the information contained within protein networks to disclose some of the molecular mechanisms underlying complex pathological processes. These findings suggest that both protein–protein interactions and the networks themselves could emerge as a new class of targetable entities, boosting the quest for novel therapeutic strategies. Ó 2009 Federation of European Biochemical Societies. Published by Elsevier B.V. All rights reserved.
1. Introduction For over a decade, genome-sequencing projects have been delivering nearly complete lists of the genes and gene products present in many organisms, including human [1,2]. However, taken individually, these components reveal relatively little about how complex biological systems organize the many discrete functions needed for survival. Virtually every process in a cell is performed by macromolecular complexes, often composed of many protein components, and regulated through the coordination of an intricate network of transient protein–protein interactions. It is thus the relationships between molecules what will ultimately determine the behavior of the system and, consequently, big efforts have been devoted to the large-scale identification of new protein interactions and macromolecular complexes. The most widely used experimental techniques in highthroughput interaction discovery are yeast two-hybrid assays [3] and binding affinity purifications coupled to mass spectrometry analyses [4], in their many different flavors (e.g. [5,6]). A system based on simple genetic manipulation with relatively easy readouts has made these two methodologies readily applicable to whole genomes of some of the major model organisms [7–12] and to a large fraction of human proteome [13,14]. Despite the clear success of large-scale interaction discovery initiatives, which
* Corresponding author. Address: Institute for Research in Biomedicine (IRB) and Barcelona Supercomputing Center (BSC), c/ Baldiri i Reixac 10-12, 08028 Barcelona, Spain. Fax: +34 934039954. E-mail address: [email protected] (P. Aloy).
have revealed thousands of unexpected interactions, the poor overlap observed between different screens run on the same organism [8,15] promoted the conception that their results are too dirty to be applicable. It has taken the community a decade of intense research to retaliate this notion and finally, recent studies have provided a general framework to correctly interpret the outcome of high-throughput interaction discovery experiments [16,17]. Today, protein interaction maps are of enough quality to become very valuable tools to help in the understanding of the underlying mechanisms of life, and we anticipate that they will play a pivotal role in the future of biological and biomedical research. In this manuscript, we explore the relationship between protein interaction networks and human disease and review recent discoveries in the field of network biology that have contributed to gain a deeper knowledge of the molecular bases of pathological processes. Moreover, and perhaps more importantly, we discuss how these findings can be readily translated into novel therapeutic strategies to fight complex diseases. 2. The role of connectivity in human disease Several human genetic diseases are caused by one or more mutations on a single gene, showing a Mendelian inheritance pattern, while the onset and the progression of others depend on the interplay of many causative genes. However, it has been suggested that this categorization might be an oversimplification [18,19]. Many well-established monogenic diseases, such as Phenylketonuria and Cystic Fibrosis, are actually oligogenic, meaning that there
0014-5793/$36.00 Ó 2009 Federation of European Biochemical Societies. Published by Elsevier B.V. All rights reserved. doi:10.1016/j.febslet.2009.03.001
1760
A. Zanzoni et al. / FEBS Letters 583 (2009) 1759–1765
are one or few modifier genes acting on the expression of a disease trait inherited by one gene [20]. There is also convincing evidence indicating that the causative genes, in both oligogenic and complex diseases, do not usually have isolated cellular functions but they rather work together in the same biological module or pathway [20,21], which highlights a strong link between protein connectivity and disease (Table 1). Early structural studies showed that, at least for a small set of disease genes tested, the onset of a disorder phenotype is related to mutations that may perturb the quaternary structure of the disease proteins and therefore disrupt protein interactions [22]. More recently, Schuster-Boeckler and Bateman compiled a list of 199 manually curated mutations in 65 diseases which do alter the interaction ability of the proteins affected [23]. The authors classified these mutations according to whether they suppress or add interactions in the protein connectivity, resulting into a loss of function (LOF) or a gain of function (GOF) (Fig. 1). Most of the considered mutations showed a LOF by mainly disrupting transient protein–protein interactions. For instance, a mutation in BRCA2, a breast cancer related gene, impairs the interaction with the replication protein A complex, essential for DNA repair [24], which leads to an accumulation of carcinogenic DNA damages. Another case of a LOF mutation occurs in the von Hippel–Lindau (VHL) tumor suppressor gene involved in the VHL syndrome, an inherited predisposition to a variety of cancer types [25]. A common VHL mutation disrupts the interaction with the hypoxia-inducible transcription factor, which is then no longer degraded, causing the expression of angiogenic growth factors that promote local proliferation of blood vessels [26]. Fewer than twenty mutations resulted in GOF, which can be divided into pathological aggregation and aberrant recognition mutations. To the first group belong mutations related to amyloid
diseases like Alzheimer’s but also sickle cell anemia, for instance. One case of aberrant recognition mutations is represented by a single amino acid substitution that increases the binding affinity of the glycoprotein GP1BA with the von Willebrand factor, leading to von Willebrand disease, a bleeding disorder [27]. Individual examples are always illustrative but, to extract relevant conclusions, it is necessary to check whether the relationship between protein connectivity and human disease is a general trend or only a curiosity. Recently, the availability of human protein interaction data from large-scale experiments [13,14] and literature mining [28,29], has permitted to study the network properties of disease genes from a global perspective. Several groups suggested that human inherited disease genes tend to encode proteins that have a larger number of interactions than non-disease-related proteins and that they preferentially interact with other disease-causing genes [30–32], representing thus the hubs in the interaction networks [33]. However, this notion has been questioned by Goh and colleagues, who have provided evidence that the ‘hub-ness’ of inherited disease genes may be only apparent [31]. They showed that almost a fifth of inherited disease genes are essential for survival, meaning that mutations in these genes are lethal, and thus they are not maintained at population level. When this set of essential genes are removed from the analysis, the encoding hubs completely disappear and non-essential disease genes tend to be located on the periphery of the interaction network, giving the cell a higher chance of survival [31]. On the other hand, disease genes whose mutations are somatic (i.e. most cancer types) are not necessarily subject to the evolutionary constraints that shape the topology of interaction networks and thus, they are more likely to encode for protein hubs [31,34,35] and to occur within the same sub-network [35].
Table 1 Collection of human diseases directly related to changes in the associated interaction networks. Disease proteins in complexes and pathways (adapted from [21]): the causative gene products of Fanconi anemia and Usher syndrome are part of complexes involved in DNA repair and hair cell differentiation, respectively. Congenital malformation syndromes are caused by mutations in genes involved in the hedgehog pathway, whereas bone-mass related diseases and osteoarthritis genes are part of the Wnt pathway. Disease
Functional relationship
Description
Reference
Fanconi anemia Usher syndrome von Hippel–Lindau syndrome
Protein complex Protein complex Protein complex
[87] [88] [26]
Inherited ataxias
Interaction subnetwork Pathway
Mutated genes are involved in the functioning of DNA repair complex Causative genes form a complex involved in hair cell differentiation Mutation in VHL gene disrupts the interaction with the hypoxia-inducible transcription factor, HIF Causative genes are part of a well connected interaction network Mutated genes are part of the hedgehog signaling pathway
[89]
Disease onset is associated to mutations in a co-receptor of Wnt proteins Causative gene is part of the Wnt signaling pathway
[90] [91]
Congenital malformation syndromes Bone-mass related diseases Osteoarthritis
Pathway Pathway
[49]
Fig. 1. Mutations in disease genes may alter their interaction properties and therefore perturb the network topology. Some of these mutations disrupt the interaction interface leading to loss of function (LOF) effect, whereas other mutations modify the interaction properties allowing the recruitment of novel partners (GOF).
A. Zanzoni et al. / FEBS Letters 583 (2009) 1759–1765
Another relationship between protein connectivity and disease that has been noted is the change in the wiring and information flow that some cell types display when moving from healthy to disease states. Wolf-Yadlin and colleagues have shown this wiring effect on cells over-expressing HER2 (a member of the EGF receptor family), which activate signaling networks controlling cell proliferation and migration that promote the phosphorylation of several downstream effectors that are not activated when HER2 is expressed at normal physiological levels [36]. It has also been reported that several pathways in the liver show a rewiring in the receptor-nucleus downstream routes when comparing normal hepatocytes with HepG2, a human hepatocellular liver carcinoma cell line [37].
3. Network-based approaches to understand human disease In the recent years, technologies such as high-throughput gene expression profiling [38] have permitted the characterization of molecular differences between healthy and disease states, leading to the identification of an increasing number of disease-related genes [39–42]. However, a clear limitation of these approaches is that they often deal with data about single players (i.e. changes in the expression of individual genes) and, to be most effective, novel strategies should also integrate systemic information to contextualize the differential expression patterns observed. Lately, protein network-based approaches have proved to be useful in understating relevant disease-related biological processes, such as inflammation [43], and identifying candidate disease genes and biomarkers [44]. Partly because high-throughput interaction discovery initiatives (e.g. [13,45]), there are a few human diseases for which we are able to gather interaction data of enough quality to explain their underlying mechanisms. Chuang et al. [46] successfully applied a network-based approach to show that known breast cancer genes that do not change their expression profile might still play a central role interconnecting genes in the protein network whose transcription is dysregulated in metastatic samples. They identified many discriminative sub-networks that are significantly more reproducible than individual marker genes selected without network information, and that are also better classifiers of metastatic tumors. More recently, Mani et al. [47] combined a weighted B-cell interactome and a large set of microarray expression profiles to identify dysregulated interactions in three different type of lymphomas. By including network information in their study, they could successfully identify both known and candidate oncogenes as well as secondary effectors that could not be highlighted by conventional analysis based on differential expression. A prerequisite to conduct these analyses is the availability of good quality interactome networks which, unfortunately, are still missing for most human diseases, which means that most efforts to unveil the molecular bases of these pathologies should necessarily involve an initial interaction discovery step. To help closing this information gap, Ewing and colleagues [48] recently conducted a large-scale immunoprecipitation analysis followed by mass spectrometry and identified 6400 high-confidence interactions of special relevance in several human diseases. Additionally, different research groups have focused their interest in charting the interaction network of specific disorders to find out putative disease genes or novel modifier genes involved in the control of the disease phenotype. For instance, Lim and colleagues [49] built an interaction network for ataxia-causing and ataxia-associated proteins using yeast two-hybrid assays, and identified more than seven hundred interactions, a subset of which has been also confirmed by co-affinity purifications from mammalian cells. Moreover, they reported that many ataxia-causing proteins cluster in a well-con-
1761
nected sub-network and share many interactors, some of which proved to be ataxia modifying genes in animal models. Kaltenbach et al. [50] applied a similar approach to identify interactors of the Huntingtin protein, whose expansion of the polyQ tail leads to the onset of Huntington’s disease (HD). From a defined high-confidence set of roughly 200 interaction partners, they found that 45% of a randomly chosen subset of 60 interactors turned to act as genetic modifiers of neurodegeneration in a HD fly model. Camargo et al. [51] also used yeast two-hybrid screens to generate an interaction network around DISC1 protein, a well-established risk factor for schizophrenia, which consisted of 158 interactions among 127 proteins, many of which are located in schizophrenia risk loci. Moreover, their findings proved that DISC1 and other schizophrenia-related proteins are involved in synaptic function. Perhaps the most illustrative work corresponds to the strategy followed by Vidal and coworkers [52], who employed a combined approach based on computational modeling and experimental techniques to identify genes that could be potentially associated with higher risk of breast cancer. They constructed a breast cancer network integrating several ‘omic’ datasets (both from human and other organisms) containing 118 genes linked by 866 functional associations. Through yeast two-hybrid and co-immunoprecipitation assays, they validated and extended the breast cancer network pinpointing putative disease players. Their gene association studies identified a genetic link between breast cancer susceptibility and centrosome dysfunction.
4. Targeting networks We have seen that, for many diseases, there are increasing pieces of evidence supporting the fact that their onset and progression arise from the interplay of a number of well-interconnected causative genes. This novel perspective is shifting the gene-centric paradigm of drug discovery towards a network/pathway-centric approach [53]. A detailed interaction map of a given pathology can extend the knowledge of disease mechanism [54] and can suggest potential points for therapeutic intervention (i.e. drug targets). Using interactome maps to select these strategic network nodes will permit to consider the robustness of the system, which is not possible in gene-centered approaches since they mostly disregard the target biological context. Biological systems, such as disease states, are generally resistant to perturbations and they are able to maintain their functions through different mechanisms such as back-up circuits and fail-safe mechanisms (i.e. redundancy and diversity) [55]. Therefore, the selection process of new putative drug targets should also consider the network positioning (i.e. involvement in fewer pathways or on their topological properties), preferring those enclaves that are essential to drive the network traffic and able to avoid back-up circuits that could neutralize the drug effect. Another therapeutic front where we anticipate that network biology approaches will make an impact is the identification of protein–protein interactions suitable to be directly targeted with drug-like compounds [56,57]. Targeting interactions has certain advantages over more traditional targets such as enzyme active sites as it often offers a more subtle, specific form of regulation that can avoid off-target side effects or total ablation of normal enzyme activity. Nutlins, a class of drug cancer candidates, are a wellknown example because they block the interaction between the tumor suppressor p53 an its negative regulator MDM2, which is over-expressed in many cancer types, thus allowing p53 to carry out its task in mediating apoptosis [58]. Many other chemicals have been designed in order to disrupt the interaction among translation initiation factors [59] or to sequestrate cytokines to impede receptor binding [60]. Although it is difficult to identify small
1762
A. Zanzoni et al. / FEBS Letters 583 (2009) 1759–1765
molecules interfering with an interaction, new methodologies (e.g. [61]) and technical improvements will hopefully provide the necessary toolkit to further expand this domain of drug discovery. Besides the basic network wiring, target selection strategies should also consider individual treats, such as the genetic diversity of human population, since they contribute considerably to the variability in drug action [62] affecting both the drug disposition (pharmacokinetics) and the drug effect itself (pharmacodynamics) [63]. Considering this type of information in the target selection process will improve the effectiveness and the applicability of future therapeutic treatments. Nevertheless, given the recent improvements in genome sequencing techniques (e.g. [64,65]) it is likely that decoding an individual genome will be affordable in a few years, making personalized medicine a common practice.
5. Expanding the druggable space The quest for new therapeutic agents often involves many complex processes and, unfortunately, information on the molecular wiring associated to a given disease, although crucial, it is often insufficient to develop a successful drug. Even for those cases where the analysis of the disease network reveals potential points of intervention, these selected nodes need to be modulated by small drug-like molecules (i.e. they need to be druggable). Recent studies have estimated the druggable genome roughly includes 3000 of the 30 000 human genes which means that, on average, only 10–15% of the proteins present in disease-associated interactomes will be amenable for being targeted with the current available chemistry [66]. The good news is that only 400 of these
Fig. 2. Representation of the main drug target classes displayed as threedimensional structures on a sized spiral distribution according to their frequency as FDA-approved drug targets. Kinases: protein kinases represent the 47% of current drugged families. GPCRs: transmembrane G-protein-coupled receptors, the 30%; other receptors: non-GPCR receptors, the 8%; ion channels: the 7%; transporters: the 4%; and others: targets which are not classified under any of the families mentioned above, the 4%.
A. Zanzoni et al. / FEBS Letters 583 (2009) 1759–1765
genes are currently targeted by marketed drugs [67], which leaves plenty of room for development of drugs against new targets. Fig. 2 gives an overview of the main protein classes currently used as drug targets. Disease biology is extremely complex and even when we identify a good drug target and a chemical modulator of its activity with good pharmacological properties, there are things that can go wrong. This has been the case for some of the most successful drugs, like Genentech’s breast cancer treatment Herceptin and Novartis’ leukaemia drug Gleevec, which eventually become inactive in many patients because of gradually increasing resistance of cancer cells [68]. It is thus not surprising that attempts to create more effective treatments for polygenic diseases by developing more selective and potent drugs for single molecular targets (the so-called ‘magic bullets’) have been of limited success [69]. Some specific drugs have been directed towards a target that resulted not essential for the pathophysiology of the disease or designed to interact with a single target that has generated unpredicted effects on off-target biochemical mechanisms [70]. An illustrative example is the case of Rofecoxib [71], a COX2-inhibitor used for mitigating inflammation. It was considered a promising drug because, compared to other anti-inflammatory agents, had a higher selectively for COX2 over COX1, thus avoiding the common adverse gastrointestinal effects. However, it showed a higher level of crosstalk with other pathways in different tissues and displayed a higher risk of adverse cardiovascular events at high doses [72]. Indeed, recent analysis of drug and drug-target networks show a rich pattern of interactions among drugs and their targets, where drugs acting on single targets appear to be the exception. Likewise, many proteins are targeted by more than one drug containing distinct chemical structures [34,73–75]. Consequently, a concept that is increasingly gaining adepts is that of polypharmacology, both from target and drug perspectives. On the one hand, the analysis of the biological networks associated to a given disease can suggest multiple targets to achieve the desired outcome. But on the other hand, we can also envisage new molecules specifically synthesized from building block that should confer them the capacity to bind multiple targets with low specificity (i.e. ‘magic shotguns’ or ‘dirty drugs’) This polypharmacology strategy has already been successfully applied to some CNS disorders [69], Alzheimer’s Disease [76], oncogenic mutations [77], and it looks very promising for the identification of effective antibacterial drugs [78]. Similarly, multi-target antibodies (in forms of diabodies, triabodies, tetrabodies and recombinant polyclonal antibodies) are also increasingly used in cancer therapy to delay the development of resistance [79]. The efficacy of such therapies can be explained by the fact that drugs targeting different proteins in the disease network or pathway could trigger a more-than-additive effect, and that their combination can eliminate compensatory reactions, avoiding thus a drugresistance denouement [80]. Moreover, if we hit the disease from different fronts, it is likely that we would not need to be highly effective in depleting the function of each single target, meaning that we could reduce the dose of each drug and circumvent some dose-related adverse events. Another strategy that is becoming increasingly important in drug development, due to the difficulty in obtaining active druglike molecules with acceptable side effects, consists of finding of new therapeutic uses for already approved drugs, an activity referred to as ‘drug repurposing’. This approach blindly screens existing compounds against a multitude of targets, and identifies either possible therapeutic benefits or side effects in a unbiased manner [81]. For instance, by testing thousands of combinations of off-patent drugs in cell-based assays, it has been found that two drugs with different indications (an antipsycothic and an antiprotozoal) exhibit an unexpected anti-tumoral activity [82]. A similar analysis was recently developed by measuring the side-effect similarity of
1763
marketed drugs to identify potential new targets [83]. However, repositioning implies unique challenges like identification and validation of the candidate from incomplete or outdated data, development of novel clinical trials for indications that have never been pursued before, and overcoming patents that could block commercialization [84]. The analysis of disease networks has also permitted the discovery of novel classes of therapeutic targets not amenable to classical drug-like compounds but reachable through novel bioactive compounds. Indeed, new biotherapeutics already make up the majority of FDA approved biotechnology medicines, including therapeutic proteins, monoclonal antibodies, antibody fragments, stem cells, antisense oligonucleotides and other RNA therapeutics [85]. Biotherapeutics have several advantages over small-molecule drugs as they often act through highly specific and complex functions that cannot be mimicked by simple chemical compounds. Moreover, they are produced in high amounts by the human body and by other organisms, which converts them in safer and less toxic agents. As a drawback, their pharmacology characteristics are generally quite different from those of small-molecule drugs, which may complicate drug development and clinical use [86].
6. Concluding remarks Current versions of the basic wiring diagrams of eukaryotic cells are still far from complete and prone to many errors. Nevertheless, recent studies have shown that, when combined with other biological data and computational tools, they contain enough information to largely enhance our understanding of the underlying mechanisms of life. Recent successes in the network biology field, such as the ones reviewed here, have attracted the interest of biotech and pharmaceutical companies that see how a strategy change towards more systemic approaches might increase the revenues of the drug discovery process. A deeper knowledge of the cell networks and molecular bases that drive pathological processes will inspire novel therapeutic strategies, ultimately leading to the development of more effective and safer drugs to fight complex diseases. Acknowledgments We would like to acknowledge the financial support received from the Spanish Ministerio de Educación y Ciencia through the Grants PSE-010000-2007-1 and BIO2007-62426. A.Z. is a Juan de la Cierva fellowship recipient. References [1] Lander, E.S. et al. (2001) Initial sequencing and analysis of the human genome. Nature 409, 860–921. [2] Venter, J.C. et al. (2001) The sequence of the human genome. Science 291, 1304–1351. [3] Fields, S. and Song, O. (1989) A novel genetic system to detect protein–protein interactions. Nature 340, 245–246. [4] Rigaut, G., Shevchenko, A., Rutz, B., Wilm, M., Mann, M. and Seraphin, B. (1999) A generic protein purification method for protein complex characterization and proteome exploration. Nat. Biotechnol. 17, 1030–1032. [5] Eyckerman, S., Lemmens, I., Catteeuw, D., Verhee, A., Vandekerckhove, J., Lievens, S. and Tavernier, J. (2005) Reverse MAPPIT: screening for protein– protein interaction modifiers in mammalian cells. Nat. Methods 2, 427–433. [6] Paumi, C.M. et al. (2007) Mapping protein–protein interactions for the yeast ABC transporter Ycf1p by integrated split-ubiquitin membrane yeast twohybrid analysis. Mol. Cell 26, 15–25. [7] Uetz, P. et al. (2000) A comprehensive analysis of protein–protein interactions in Saccharomyces cerevisiae. Nature 403, 623–627. [8] 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, 4569–4574. [9] Gavin, A.C. et al. (2006) Proteome survey reveals modularity of the yeast cell machinery. Nature 440, 631–636.
1764
A. Zanzoni et al. / FEBS Letters 583 (2009) 1759–1765
[10] Krogan, N.J. et al. (2006) Global landscape of protein complexes in the yeast Saccharomyces cerevisiae. Nature 440, 637–643. [11] Giot, L. et al. (2003) A protein interaction map of Drosophila melanogaster. Science 302, 1727–1736. [12] Li, S. et al. (2004) A map of the interactome network of the metazoan C. elegans. Science 303, 540–543. [13] Rual, J.F. et al. (2005) Towards a proteome-scale map of the human protein– protein interaction network. Nature 437, 1173–1178. [14] Stelzl, U. et al. (2005) A human protein–protein interaction network: a resource for annotating the proteome. Cell 122, 957–968. [15] Collins, S.R., Kemmeren, P., Zhao, X.C., Greenblatt, J.F., Spencer, F., Holstege, F.C., Weissman, J.S. and Krogan, N.J. (2007) Toward a comprehensive atlas of the physical interactome of Saccharomyces cerevisiae. Mol. Cell Proteomics 6, 439–450. [16] Yu, H. et al. (2008) High-quality binary protein interaction map of the yeast interactome network. Science. [17] Venkatesan, K. et al. (2009) An empirical framework for binary interactome mapping. Nat. Methods 6, 83–90. [18] Scriver, C.R. and Waters, P.J. (1999) Monogenic traits are not simple: lessons from phenylketonuria. Trends Genet. 15, 267–272. [19] Dipple, K.M. and McCabe, E.R. (2000) Modifier genes convert ‘‘simple” Mendelian disorders to complex traits. Mol. Genet. Metab. 71, 43–50. [20] Badano, J.L. and Katsanis, N. (2002) Beyond mendel: an evolving view of human genetic disease transmission. Nat. Rev. Genet. 3, 779–789. [21] Oti, M. and Brunner, H.G. (2007) The modular nature of genetic diseases. Clin. Genet. 71, 1–11. [22] Ferrer-Costa, C., Orozco, M. and de la Cruz, X. (2002) Characterization of disease-associated single amino acid polymorphisms in terms of sequence and structure properties. J. Mol. Biol. 315, 771–786. [23] Schuster-Böckler, B. and Bateman, A. (2008) Protein interactions in human genetic diseases. Genome Biol. 9, R9. [24] Wong, J.M.S., Ionescu, D. and Ingles, C.J. (2003) Interaction between BRCA2 and replication protein A is compromised by a cancer-predisposing mutation in BRCA2. Oncogene 22, 28–33. [25] Ivan, M. and Kaelin, W.G. (2001) The von Hippel–Lindau tumor suppressor protein. Curr. Opin. Genet. Dev. 11, 27–34. [26] Ohh, M. et al. (2000) Ubiquitination of hypoxia-inducible factor requires direct binding to the beta-domain of the von Hippel–Lindau protein. Nat. Cell Biol. 2, 423–427. [27] Miller, J.L., Cunningham, D., Lyle, V.A. and Finch, C.N. (1991) Mutation in the gene encoding the alpha chain of platelet glycoprotein Ib in platelet-type von Willebrand disease. Proc. Natl. Acad. Sci. USA 88, 4761–4765. [28] Chatr-aryamontri, A., Ceol, A., Palazzi, L.M., Nardelli, G., Schneider, M.V., Castagnoli, L. and Cesareni, G. (2007) MINT: the Molecular INTeraction database. Nucleic Acids Res. 35, D572–D574. [29] Kerrien, S. et al. (2007) IntAct – open source resource for molecular interaction data. Nucleic Acids Res. 35, D561–D565. [30] Gandhi, T.K.B. et al. (2006) Analysis of the human protein interactome and comparison with yeast, worm and fly interaction datasets. Nat. Genet. 38, 285–293. [31] Goh, K.-I., Cusick, M.E., Valle, D., Childs, B., Vidal, M. and Barabási, A.-L.s. (2007) The human disease network. Proc. Natl. Acad. Sci. USA 104, 8685–8690. [32] Feldman, I., Rzhetsky, A. and Vitkup, D. (2008) Network properties of genes harboring inherited disease mutations. Proc. Natl. Acad. Sci. USA 105, 4323– 4328. [33] Xu, J. and Li, Y. (2006) Discovering disease-genes by topological features in human protein–protein interaction network. Bioinformatics 22, 2800–2805. [34] Yildirim, M.A., Goh, K.-I., Cusick, M.E., Barabási, A.-L. and Vidal, M. (2007) Drugtarget network. Nat. Biotechnol. 25, 1119–1126. [35] Rambaldi, D., Giorgi, F.M., Capuani, F., Ciliberto, A. and Ciccarelli, F.D. (2008) Low duplicability and network fragility of cancer genes. Trends Genet. 24, 427–430. [36] Wolf-Yadlin, A. et al. (2006) Effects of HER2 overexpression on cell signaling networks governing proliferation and migration. Mol. Syst. Biol. 2, 54. [37] Xu, J.J., Henstock, P.V., Dunn, M.C., Smith, A.R., Chabot, J.R. and de Graaf, D. (2008) Cellular imaging predictions of clinical drug-induced liver injury. Toxicol. Sci. 105, 97–105. [38] Schena, M., Shalon, D., Davis, R.W. and Brown, P.O. (1995) Quantitative monitoring of gene expression patterns with a complementary DNA microarray. Science 270, 467–470. [39] Golub, T.R. et al. (1999) Molecular classification of cancer: class discovery and class prediction by gene expression monitoring. Science 286, 531–537. [40] Chen, X. et al. (2002) Gene expression patterns in human liver cancers. Mol. Biol. Cell 13, 1929–1939. [41] van’t Veer, L.J. et al. (2002) Gene expression profiling predicts clinical outcome of breast cancer. Nature 415, 530–536. [42] Wang, Y. et al. (2005) Gene-expression profiles to predict distant metastasis of lymphnode-negative primary breast cancer. Lancet 365, 671–679. [43] Calvano, S.E. et al. (2005) A network-based analysis of systemic inflammation in humans. Nature 437, 1032–1037. [44] Ideker, T. and Sharan, R. (2008) Protein networks in disease. Genome Res. 18, 644–652. [45] Stelzl, U. et al. (2005) A human protein–protein interaction network: a resource for annotating the proteome. Cell 122, 957–968. [46] Chuang, H.-Y., Lee, E., Liu, Y.-T., Lee, D. and Ideker, T. (2007) Network-based classification of breast cancer metastasis. Mol. Syst. Biol. 3, 140.
[47] Mani, K.M., Lefebvre, C., Wang, K., Lim, W.K., Basso, K., Dalla-Favera, R. and Califano, A. (2008) A systems biology approach to prediction of oncogenes and molecular perturbation targets in B-cell lymphomas. Mol. Syst. Biol. 4, 169. [48] Ewing, R.M. et al. (2007) Large-scale mapping of human protein–protein interactions by mass spectrometry. Mol. Syst. Biol. 3, 89. [49] Lim, J. et al. (2006) A protein–protein interaction network for human inherited ataxias and disorders of Purkinje cell degeneration. Cell 125, 801–814. [50] Kaltenbach, L.S. et al. (2007) Huntingtin interacting proteins are genetic modifiers of neurodegeneration. PLoS Genet. 3, e82. [51] Camargo, L.M. et al. (2007) Disrupted in Schizophrenia 1 interactome: evidence for the close connectivity of risk genes and a potential synaptic basis for schizophrenia. Mol. Psychiatry 12, 74–86. [52] Pujana, M.A. et al. (2007) Network modeling links breast cancer susceptibility and centrosome dysfunction. Nat. Genet. 39, 1338–1349. [53] Russell, R.B. and Aloy, P. (2008) Targeting and tinkering with interaction networks. Nat. Chem. Biol. 4, 666–673. [54] Pache, R.A., Zanzoni, A., Naval, J., Mas, J.M. and Aloy, P. (2008) Towards a molecular characterisation of pathological pathways. FEBS Lett. 582, 1259– 1265. [55] Kitano, H. (2004) Biological robustness. Nat. Rev. Genet. 5, 826–837. [56] Neduva, V. and Russell, R.B. (2006) Peptides mediating interaction networks: new leads at last. Curr. Opin. Biotechnol. 17, 465–471. [57] Wells, J.A. and McClendon, C.L. (2007) Reaching for high-hanging fruit in drug discovery at protein–protein interfaces. Nature 450, 1001–1009. [58] Vassilev, L.T. et al. (2004) In vivo activation of the p53 pathway by smallmolecule antagonists of MDM2. Science 303, 844–848. [59] Moerke, N.J. et al. (2007) Small-molecule inhibition of the interaction between the translation initiation factors eIF4E and eIF4G. Cell 128, 257–267. [60] He, M.M. et al. (2005) Small-molecule inhibition of TNF-alpha. Science 310, 1022–1025. [61] Parthasarathi, L., Casey, F., Stein, A., Aloy, P. and Shields, D.C. (2008) Approved drug mimics of short peptide ligands from protein interaction motifs. J. Chem. Inf. Model. 48, 1943–1948. [62] Roden, D.M. and George, A.L. (2002) The genetic basis of variability in drug responses. Nat. Rev. Drug Discov 1, 37–44. [63] Roden, D.M. et al. (2006) Pharmacogenomics: challenges and opportunities. Ann. Intern. Med. 145, 749–757. [64] Bentley, D.R. et al. (2008) Accurate whole human genome sequencing using reversible terminator chemistry. Nature 456, 53–59. [65] Wang, J. et al. (2008) The diploid genome sequence of an Asian individual. Nature 456, 60–65. [66] Hopkins, A.L. (2008) Network pharmacology: the next paradigm in drug discovery. Nat. Chem. Biol. 4, 682–690. [67] Overington, J.P., Al-Lazikani, B. and Hopkins, A.L. (2006) How many drug targets are there? Nat. Rev. Drug Discov. 5, 993–996. [68] Schrattenholz, A. and Soskic´, V. (2008) What does systems biology mean for drug development? Curr. Med. Chem. 15, 1520–1528. [69] Roth, B.L., Sheffler, D.J. and Kroeze, W.K. (2004) Magic shotguns versus magic bullets: selectively non-selective drugs for mood disorders and schizophrenia. Nat. Rev. Drug Discov. 3, 353–359. [70] van der Greef, J. and McBurney, R.N. (2005) Innovation: rescuing drug discovery: in vivo systems pathology and systems pharmacology. Nat. Rev. Drug Discov. 4, 961–967. [71] FitzGerald, G.A. and Patrono, C. (2001) The coxibs, selective inhibitors of cyclooxygenase-2. N Engl. J. Med. 345, 433–442. [72] Zarraga, I.G.E. and Schwarz, E.R. (2007) Coxibs and heart disease: what we have learned and what else we need to know. J. Am. Coll Cardiol. 49, 1–14. [73] Paolini, G.V., Shapland, R.H.B., van Hoorn, W.P., Mason, J.S. and Hopkins, A.L. (2006) Global mapping of pharmacological space. Nat. Biotechnol. 24, 805– 815. [74] Gregori-Puigjané, E. and Mestres, J. (2008) A ligand-based approach to mining the chemogenomic space of drugs. Comb. Chem. High Throughput Screen. 11, 669–676. [75] Nacher, J.C. and Schwartz, J.-M. (2008) A global view of drug-therapy interactions. BMC Pharmacol. 8, 5. [76] Bolognesi, M.L., Cavalli, A. and Melchiorre, C. (2009) Memoquin: a multitargetdirected ligand as an innovative therapeutic opportunity for Alzheimer’s disease. Neurotherapeutics 6, 152–162. [77] Apsel, B. et al. (2008) Targeted polypharmacology: discovery of dual inhibitors of tyrosine and phosphoinositide kinases. Nat. Chem. Biol. 4, 691–699. [78] Rasko, D.A. et al. (2008) Targeting QseC signaling and virulence for antibiotic development. Science 321, 1078–1080. [79] Todorovska, A., Roovers, R.C., Dolezal, O., Kortt, A.A., Hoogenboom, H.R. and Hudson, P.J. (2001) Design and application of diabodies, triabodies and tetrabodies for cancer targeting. J. Immunol. Methods 248, 47–66. [80] Kitano, H. (2007) A robustness-based approach to systems-oriented drug design. Nat. Rev. Drug Discov. 6, 202–210. [81] O’Connor, K.A. and Roth, B.L. (2005) Finding new tricks for old drugs: an efficient route for public-sector drug discovery. Nat. Rev. Drug Discov. 4, 1005–1014. [82] Lee, M.S. et al. (2007) The novel combination of chlorpromazine and pentamidine exerts synergistic antiproliferative effects through dual mitotic action. Cancer Res. 67, 11359–11367. [83] Campillos, M., Kuhn, M., Gavin, A.-C., Jensen, L.J. and Bork, P. (2008) Drug target identification using side-effect similarity. Science 321, 263–266.
A. Zanzoni et al. / FEBS Letters 583 (2009) 1759–1765 [84] Ashburn, T.T. and Thor, K.B. (2004) Drug repositioning: identifying and developing new uses for existing drugs. Nat. Rev. Drug Discov. 3, 673– 683. [85] Lalonde, R.L. and Honig, P. (2008) Clinical pharmacology in the era of biotherapeutics. Clin. Pharmacol. Ther. 84, 533–536. [86] Leader, B., Baca, Q.J. and Golan, D.E. (2008) Protein therapeutics: a summary and pharmacological classification. Nat. Rev. Drug Discov. 7, 21–39. [87] Garcia-Higuera, I., Taniguchi, T., Ganesan, S., Meyn, M.S., Timmers, C., Hejna, J., Grompe, M. and D’Andrea, A.D. (2001) Interaction of the Fanconi anemia proteins and BRCA1 in a common pathway. Mol. Cell 7, 249–262.
1765
[88] El-Amraoui, A. and Petit, C. (2005) Usher I syndrome: unravelling the mechanisms that underlie the cohesion of the growing hair bundle in inner ear sensory cells. J. Cell Sci. 118, 4593–4603. [89] Nieuwenhuis, E. and Hui, C.c. (2005) Hedgehog signaling and congenital malformations. Clin. Genet. 67, 193–208. [90] Wesenbeeck, L.V. et al. (2003) Six novel missense mutations in the LDL receptorrelated protein 5 (LRP5) gene in different conditions with an increased bone density. Am. J. Hum. Genet. 72, 763–771. [91] Corr, M. (2008) Wnt-beta-catenin signaling in the pathogenesis of osteoarthritis. Nat. Clin. Pract. Rheumatol. 4, 550–556.
journal homepage: www.FEBSLetters.org
Minireview
A network medicine approach to human disease Andreas Zanzoni a, Montserrat Soler-López a, Patrick Aloy a,b,* a b
Institute for Research in Biomedicine (IRB) and Barcelona Supercomputing Center (BSC), c/ Baldiri i Reixac 10-12, 08028 Barcelona, Spain Institució Catalana de Recerca i Estudis Avançats (ICREA) Pg. Lluís Companys 23, 08010 Barcelona, Spain
a r t i c l e
i n f o
Article history: Received 2 February 2009 Accepted 2 March 2009 Available online 6 March 2009 Edited by Miguel De la Rosa Keywords: Network medicine Protein–protein interaction Protein network Drug discovery
a b s t r a c t High-throughput interaction discovery initiatives are providing thousands of novel protein interactions which are unveiling many unexpected links between apparently unrelated biological processes. In particular, analyses of the first draft human interactomes highlight a strong association between protein network connectivity and disease. Indeed, recent exciting studies have exploited the information contained within protein networks to disclose some of the molecular mechanisms underlying complex pathological processes. These findings suggest that both protein–protein interactions and the networks themselves could emerge as a new class of targetable entities, boosting the quest for novel therapeutic strategies. Ó 2009 Federation of European Biochemical Societies. Published by Elsevier B.V. All rights reserved.
1. Introduction For over a decade, genome-sequencing projects have been delivering nearly complete lists of the genes and gene products present in many organisms, including human [1,2]. However, taken individually, these components reveal relatively little about how complex biological systems organize the many discrete functions needed for survival. Virtually every process in a cell is performed by macromolecular complexes, often composed of many protein components, and regulated through the coordination of an intricate network of transient protein–protein interactions. It is thus the relationships between molecules what will ultimately determine the behavior of the system and, consequently, big efforts have been devoted to the large-scale identification of new protein interactions and macromolecular complexes. The most widely used experimental techniques in highthroughput interaction discovery are yeast two-hybrid assays [3] and binding affinity purifications coupled to mass spectrometry analyses [4], in their many different flavors (e.g. [5,6]). A system based on simple genetic manipulation with relatively easy readouts has made these two methodologies readily applicable to whole genomes of some of the major model organisms [7–12] and to a large fraction of human proteome [13,14]. Despite the clear success of large-scale interaction discovery initiatives, which
* Corresponding author. Address: Institute for Research in Biomedicine (IRB) and Barcelona Supercomputing Center (BSC), c/ Baldiri i Reixac 10-12, 08028 Barcelona, Spain. Fax: +34 934039954. E-mail address: [email protected] (P. Aloy).
have revealed thousands of unexpected interactions, the poor overlap observed between different screens run on the same organism [8,15] promoted the conception that their results are too dirty to be applicable. It has taken the community a decade of intense research to retaliate this notion and finally, recent studies have provided a general framework to correctly interpret the outcome of high-throughput interaction discovery experiments [16,17]. Today, protein interaction maps are of enough quality to become very valuable tools to help in the understanding of the underlying mechanisms of life, and we anticipate that they will play a pivotal role in the future of biological and biomedical research. In this manuscript, we explore the relationship between protein interaction networks and human disease and review recent discoveries in the field of network biology that have contributed to gain a deeper knowledge of the molecular bases of pathological processes. Moreover, and perhaps more importantly, we discuss how these findings can be readily translated into novel therapeutic strategies to fight complex diseases. 2. The role of connectivity in human disease Several human genetic diseases are caused by one or more mutations on a single gene, showing a Mendelian inheritance pattern, while the onset and the progression of others depend on the interplay of many causative genes. However, it has been suggested that this categorization might be an oversimplification [18,19]. Many well-established monogenic diseases, such as Phenylketonuria and Cystic Fibrosis, are actually oligogenic, meaning that there
0014-5793/$36.00 Ó 2009 Federation of European Biochemical Societies. Published by Elsevier B.V. All rights reserved. doi:10.1016/j.febslet.2009.03.001
1760
A. Zanzoni et al. / FEBS Letters 583 (2009) 1759–1765
are one or few modifier genes acting on the expression of a disease trait inherited by one gene [20]. There is also convincing evidence indicating that the causative genes, in both oligogenic and complex diseases, do not usually have isolated cellular functions but they rather work together in the same biological module or pathway [20,21], which highlights a strong link between protein connectivity and disease (Table 1). Early structural studies showed that, at least for a small set of disease genes tested, the onset of a disorder phenotype is related to mutations that may perturb the quaternary structure of the disease proteins and therefore disrupt protein interactions [22]. More recently, Schuster-Boeckler and Bateman compiled a list of 199 manually curated mutations in 65 diseases which do alter the interaction ability of the proteins affected [23]. The authors classified these mutations according to whether they suppress or add interactions in the protein connectivity, resulting into a loss of function (LOF) or a gain of function (GOF) (Fig. 1). Most of the considered mutations showed a LOF by mainly disrupting transient protein–protein interactions. For instance, a mutation in BRCA2, a breast cancer related gene, impairs the interaction with the replication protein A complex, essential for DNA repair [24], which leads to an accumulation of carcinogenic DNA damages. Another case of a LOF mutation occurs in the von Hippel–Lindau (VHL) tumor suppressor gene involved in the VHL syndrome, an inherited predisposition to a variety of cancer types [25]. A common VHL mutation disrupts the interaction with the hypoxia-inducible transcription factor, which is then no longer degraded, causing the expression of angiogenic growth factors that promote local proliferation of blood vessels [26]. Fewer than twenty mutations resulted in GOF, which can be divided into pathological aggregation and aberrant recognition mutations. To the first group belong mutations related to amyloid
diseases like Alzheimer’s but also sickle cell anemia, for instance. One case of aberrant recognition mutations is represented by a single amino acid substitution that increases the binding affinity of the glycoprotein GP1BA with the von Willebrand factor, leading to von Willebrand disease, a bleeding disorder [27]. Individual examples are always illustrative but, to extract relevant conclusions, it is necessary to check whether the relationship between protein connectivity and human disease is a general trend or only a curiosity. Recently, the availability of human protein interaction data from large-scale experiments [13,14] and literature mining [28,29], has permitted to study the network properties of disease genes from a global perspective. Several groups suggested that human inherited disease genes tend to encode proteins that have a larger number of interactions than non-disease-related proteins and that they preferentially interact with other disease-causing genes [30–32], representing thus the hubs in the interaction networks [33]. However, this notion has been questioned by Goh and colleagues, who have provided evidence that the ‘hub-ness’ of inherited disease genes may be only apparent [31]. They showed that almost a fifth of inherited disease genes are essential for survival, meaning that mutations in these genes are lethal, and thus they are not maintained at population level. When this set of essential genes are removed from the analysis, the encoding hubs completely disappear and non-essential disease genes tend to be located on the periphery of the interaction network, giving the cell a higher chance of survival [31]. On the other hand, disease genes whose mutations are somatic (i.e. most cancer types) are not necessarily subject to the evolutionary constraints that shape the topology of interaction networks and thus, they are more likely to encode for protein hubs [31,34,35] and to occur within the same sub-network [35].
Table 1 Collection of human diseases directly related to changes in the associated interaction networks. Disease proteins in complexes and pathways (adapted from [21]): the causative gene products of Fanconi anemia and Usher syndrome are part of complexes involved in DNA repair and hair cell differentiation, respectively. Congenital malformation syndromes are caused by mutations in genes involved in the hedgehog pathway, whereas bone-mass related diseases and osteoarthritis genes are part of the Wnt pathway. Disease
Functional relationship
Description
Reference
Fanconi anemia Usher syndrome von Hippel–Lindau syndrome
Protein complex Protein complex Protein complex
[87] [88] [26]
Inherited ataxias
Interaction subnetwork Pathway
Mutated genes are involved in the functioning of DNA repair complex Causative genes form a complex involved in hair cell differentiation Mutation in VHL gene disrupts the interaction with the hypoxia-inducible transcription factor, HIF Causative genes are part of a well connected interaction network Mutated genes are part of the hedgehog signaling pathway
[89]
Disease onset is associated to mutations in a co-receptor of Wnt proteins Causative gene is part of the Wnt signaling pathway
[90] [91]
Congenital malformation syndromes Bone-mass related diseases Osteoarthritis
Pathway Pathway
[49]
Fig. 1. Mutations in disease genes may alter their interaction properties and therefore perturb the network topology. Some of these mutations disrupt the interaction interface leading to loss of function (LOF) effect, whereas other mutations modify the interaction properties allowing the recruitment of novel partners (GOF).
A. Zanzoni et al. / FEBS Letters 583 (2009) 1759–1765
Another relationship between protein connectivity and disease that has been noted is the change in the wiring and information flow that some cell types display when moving from healthy to disease states. Wolf-Yadlin and colleagues have shown this wiring effect on cells over-expressing HER2 (a member of the EGF receptor family), which activate signaling networks controlling cell proliferation and migration that promote the phosphorylation of several downstream effectors that are not activated when HER2 is expressed at normal physiological levels [36]. It has also been reported that several pathways in the liver show a rewiring in the receptor-nucleus downstream routes when comparing normal hepatocytes with HepG2, a human hepatocellular liver carcinoma cell line [37].
3. Network-based approaches to understand human disease In the recent years, technologies such as high-throughput gene expression profiling [38] have permitted the characterization of molecular differences between healthy and disease states, leading to the identification of an increasing number of disease-related genes [39–42]. However, a clear limitation of these approaches is that they often deal with data about single players (i.e. changes in the expression of individual genes) and, to be most effective, novel strategies should also integrate systemic information to contextualize the differential expression patterns observed. Lately, protein network-based approaches have proved to be useful in understating relevant disease-related biological processes, such as inflammation [43], and identifying candidate disease genes and biomarkers [44]. Partly because high-throughput interaction discovery initiatives (e.g. [13,45]), there are a few human diseases for which we are able to gather interaction data of enough quality to explain their underlying mechanisms. Chuang et al. [46] successfully applied a network-based approach to show that known breast cancer genes that do not change their expression profile might still play a central role interconnecting genes in the protein network whose transcription is dysregulated in metastatic samples. They identified many discriminative sub-networks that are significantly more reproducible than individual marker genes selected without network information, and that are also better classifiers of metastatic tumors. More recently, Mani et al. [47] combined a weighted B-cell interactome and a large set of microarray expression profiles to identify dysregulated interactions in three different type of lymphomas. By including network information in their study, they could successfully identify both known and candidate oncogenes as well as secondary effectors that could not be highlighted by conventional analysis based on differential expression. A prerequisite to conduct these analyses is the availability of good quality interactome networks which, unfortunately, are still missing for most human diseases, which means that most efforts to unveil the molecular bases of these pathologies should necessarily involve an initial interaction discovery step. To help closing this information gap, Ewing and colleagues [48] recently conducted a large-scale immunoprecipitation analysis followed by mass spectrometry and identified 6400 high-confidence interactions of special relevance in several human diseases. Additionally, different research groups have focused their interest in charting the interaction network of specific disorders to find out putative disease genes or novel modifier genes involved in the control of the disease phenotype. For instance, Lim and colleagues [49] built an interaction network for ataxia-causing and ataxia-associated proteins using yeast two-hybrid assays, and identified more than seven hundred interactions, a subset of which has been also confirmed by co-affinity purifications from mammalian cells. Moreover, they reported that many ataxia-causing proteins cluster in a well-con-
1761
nected sub-network and share many interactors, some of which proved to be ataxia modifying genes in animal models. Kaltenbach et al. [50] applied a similar approach to identify interactors of the Huntingtin protein, whose expansion of the polyQ tail leads to the onset of Huntington’s disease (HD). From a defined high-confidence set of roughly 200 interaction partners, they found that 45% of a randomly chosen subset of 60 interactors turned to act as genetic modifiers of neurodegeneration in a HD fly model. Camargo et al. [51] also used yeast two-hybrid screens to generate an interaction network around DISC1 protein, a well-established risk factor for schizophrenia, which consisted of 158 interactions among 127 proteins, many of which are located in schizophrenia risk loci. Moreover, their findings proved that DISC1 and other schizophrenia-related proteins are involved in synaptic function. Perhaps the most illustrative work corresponds to the strategy followed by Vidal and coworkers [52], who employed a combined approach based on computational modeling and experimental techniques to identify genes that could be potentially associated with higher risk of breast cancer. They constructed a breast cancer network integrating several ‘omic’ datasets (both from human and other organisms) containing 118 genes linked by 866 functional associations. Through yeast two-hybrid and co-immunoprecipitation assays, they validated and extended the breast cancer network pinpointing putative disease players. Their gene association studies identified a genetic link between breast cancer susceptibility and centrosome dysfunction.
4. Targeting networks We have seen that, for many diseases, there are increasing pieces of evidence supporting the fact that their onset and progression arise from the interplay of a number of well-interconnected causative genes. This novel perspective is shifting the gene-centric paradigm of drug discovery towards a network/pathway-centric approach [53]. A detailed interaction map of a given pathology can extend the knowledge of disease mechanism [54] and can suggest potential points for therapeutic intervention (i.e. drug targets). Using interactome maps to select these strategic network nodes will permit to consider the robustness of the system, which is not possible in gene-centered approaches since they mostly disregard the target biological context. Biological systems, such as disease states, are generally resistant to perturbations and they are able to maintain their functions through different mechanisms such as back-up circuits and fail-safe mechanisms (i.e. redundancy and diversity) [55]. Therefore, the selection process of new putative drug targets should also consider the network positioning (i.e. involvement in fewer pathways or on their topological properties), preferring those enclaves that are essential to drive the network traffic and able to avoid back-up circuits that could neutralize the drug effect. Another therapeutic front where we anticipate that network biology approaches will make an impact is the identification of protein–protein interactions suitable to be directly targeted with drug-like compounds [56,57]. Targeting interactions has certain advantages over more traditional targets such as enzyme active sites as it often offers a more subtle, specific form of regulation that can avoid off-target side effects or total ablation of normal enzyme activity. Nutlins, a class of drug cancer candidates, are a wellknown example because they block the interaction between the tumor suppressor p53 an its negative regulator MDM2, which is over-expressed in many cancer types, thus allowing p53 to carry out its task in mediating apoptosis [58]. Many other chemicals have been designed in order to disrupt the interaction among translation initiation factors [59] or to sequestrate cytokines to impede receptor binding [60]. Although it is difficult to identify small
1762
A. Zanzoni et al. / FEBS Letters 583 (2009) 1759–1765
molecules interfering with an interaction, new methodologies (e.g. [61]) and technical improvements will hopefully provide the necessary toolkit to further expand this domain of drug discovery. Besides the basic network wiring, target selection strategies should also consider individual treats, such as the genetic diversity of human population, since they contribute considerably to the variability in drug action [62] affecting both the drug disposition (pharmacokinetics) and the drug effect itself (pharmacodynamics) [63]. Considering this type of information in the target selection process will improve the effectiveness and the applicability of future therapeutic treatments. Nevertheless, given the recent improvements in genome sequencing techniques (e.g. [64,65]) it is likely that decoding an individual genome will be affordable in a few years, making personalized medicine a common practice.
5. Expanding the druggable space The quest for new therapeutic agents often involves many complex processes and, unfortunately, information on the molecular wiring associated to a given disease, although crucial, it is often insufficient to develop a successful drug. Even for those cases where the analysis of the disease network reveals potential points of intervention, these selected nodes need to be modulated by small drug-like molecules (i.e. they need to be druggable). Recent studies have estimated the druggable genome roughly includes 3000 of the 30 000 human genes which means that, on average, only 10–15% of the proteins present in disease-associated interactomes will be amenable for being targeted with the current available chemistry [66]. The good news is that only 400 of these
Fig. 2. Representation of the main drug target classes displayed as threedimensional structures on a sized spiral distribution according to their frequency as FDA-approved drug targets. Kinases: protein kinases represent the 47% of current drugged families. GPCRs: transmembrane G-protein-coupled receptors, the 30%; other receptors: non-GPCR receptors, the 8%; ion channels: the 7%; transporters: the 4%; and others: targets which are not classified under any of the families mentioned above, the 4%.
A. Zanzoni et al. / FEBS Letters 583 (2009) 1759–1765
genes are currently targeted by marketed drugs [67], which leaves plenty of room for development of drugs against new targets. Fig. 2 gives an overview of the main protein classes currently used as drug targets. Disease biology is extremely complex and even when we identify a good drug target and a chemical modulator of its activity with good pharmacological properties, there are things that can go wrong. This has been the case for some of the most successful drugs, like Genentech’s breast cancer treatment Herceptin and Novartis’ leukaemia drug Gleevec, which eventually become inactive in many patients because of gradually increasing resistance of cancer cells [68]. It is thus not surprising that attempts to create more effective treatments for polygenic diseases by developing more selective and potent drugs for single molecular targets (the so-called ‘magic bullets’) have been of limited success [69]. Some specific drugs have been directed towards a target that resulted not essential for the pathophysiology of the disease or designed to interact with a single target that has generated unpredicted effects on off-target biochemical mechanisms [70]. An illustrative example is the case of Rofecoxib [71], a COX2-inhibitor used for mitigating inflammation. It was considered a promising drug because, compared to other anti-inflammatory agents, had a higher selectively for COX2 over COX1, thus avoiding the common adverse gastrointestinal effects. However, it showed a higher level of crosstalk with other pathways in different tissues and displayed a higher risk of adverse cardiovascular events at high doses [72]. Indeed, recent analysis of drug and drug-target networks show a rich pattern of interactions among drugs and their targets, where drugs acting on single targets appear to be the exception. Likewise, many proteins are targeted by more than one drug containing distinct chemical structures [34,73–75]. Consequently, a concept that is increasingly gaining adepts is that of polypharmacology, both from target and drug perspectives. On the one hand, the analysis of the biological networks associated to a given disease can suggest multiple targets to achieve the desired outcome. But on the other hand, we can also envisage new molecules specifically synthesized from building block that should confer them the capacity to bind multiple targets with low specificity (i.e. ‘magic shotguns’ or ‘dirty drugs’) This polypharmacology strategy has already been successfully applied to some CNS disorders [69], Alzheimer’s Disease [76], oncogenic mutations [77], and it looks very promising for the identification of effective antibacterial drugs [78]. Similarly, multi-target antibodies (in forms of diabodies, triabodies, tetrabodies and recombinant polyclonal antibodies) are also increasingly used in cancer therapy to delay the development of resistance [79]. The efficacy of such therapies can be explained by the fact that drugs targeting different proteins in the disease network or pathway could trigger a more-than-additive effect, and that their combination can eliminate compensatory reactions, avoiding thus a drugresistance denouement [80]. Moreover, if we hit the disease from different fronts, it is likely that we would not need to be highly effective in depleting the function of each single target, meaning that we could reduce the dose of each drug and circumvent some dose-related adverse events. Another strategy that is becoming increasingly important in drug development, due to the difficulty in obtaining active druglike molecules with acceptable side effects, consists of finding of new therapeutic uses for already approved drugs, an activity referred to as ‘drug repurposing’. This approach blindly screens existing compounds against a multitude of targets, and identifies either possible therapeutic benefits or side effects in a unbiased manner [81]. For instance, by testing thousands of combinations of off-patent drugs in cell-based assays, it has been found that two drugs with different indications (an antipsycothic and an antiprotozoal) exhibit an unexpected anti-tumoral activity [82]. A similar analysis was recently developed by measuring the side-effect similarity of
1763
marketed drugs to identify potential new targets [83]. However, repositioning implies unique challenges like identification and validation of the candidate from incomplete or outdated data, development of novel clinical trials for indications that have never been pursued before, and overcoming patents that could block commercialization [84]. The analysis of disease networks has also permitted the discovery of novel classes of therapeutic targets not amenable to classical drug-like compounds but reachable through novel bioactive compounds. Indeed, new biotherapeutics already make up the majority of FDA approved biotechnology medicines, including therapeutic proteins, monoclonal antibodies, antibody fragments, stem cells, antisense oligonucleotides and other RNA therapeutics [85]. Biotherapeutics have several advantages over small-molecule drugs as they often act through highly specific and complex functions that cannot be mimicked by simple chemical compounds. Moreover, they are produced in high amounts by the human body and by other organisms, which converts them in safer and less toxic agents. As a drawback, their pharmacology characteristics are generally quite different from those of small-molecule drugs, which may complicate drug development and clinical use [86].
6. Concluding remarks Current versions of the basic wiring diagrams of eukaryotic cells are still far from complete and prone to many errors. Nevertheless, recent studies have shown that, when combined with other biological data and computational tools, they contain enough information to largely enhance our understanding of the underlying mechanisms of life. Recent successes in the network biology field, such as the ones reviewed here, have attracted the interest of biotech and pharmaceutical companies that see how a strategy change towards more systemic approaches might increase the revenues of the drug discovery process. A deeper knowledge of the cell networks and molecular bases that drive pathological processes will inspire novel therapeutic strategies, ultimately leading to the development of more effective and safer drugs to fight complex diseases. Acknowledgments We would like to acknowledge the financial support received from the Spanish Ministerio de Educación y Ciencia through the Grants PSE-010000-2007-1 and BIO2007-62426. A.Z. is a Juan de la Cierva fellowship recipient. References [1] Lander, E.S. et al. (2001) Initial sequencing and analysis of the human genome. Nature 409, 860–921. [2] Venter, J.C. et al. (2001) The sequence of the human genome. Science 291, 1304–1351. [3] Fields, S. and Song, O. (1989) A novel genetic system to detect protein–protein interactions. Nature 340, 245–246. [4] Rigaut, G., Shevchenko, A., Rutz, B., Wilm, M., Mann, M. and Seraphin, B. (1999) A generic protein purification method for protein complex characterization and proteome exploration. Nat. Biotechnol. 17, 1030–1032. [5] Eyckerman, S., Lemmens, I., Catteeuw, D., Verhee, A., Vandekerckhove, J., Lievens, S. and Tavernier, J. (2005) Reverse MAPPIT: screening for protein– protein interaction modifiers in mammalian cells. Nat. Methods 2, 427–433. [6] Paumi, C.M. et al. (2007) Mapping protein–protein interactions for the yeast ABC transporter Ycf1p by integrated split-ubiquitin membrane yeast twohybrid analysis. Mol. Cell 26, 15–25. [7] Uetz, P. et al. (2000) A comprehensive analysis of protein–protein interactions in Saccharomyces cerevisiae. Nature 403, 623–627. [8] 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, 4569–4574. [9] Gavin, A.C. et al. (2006) Proteome survey reveals modularity of the yeast cell machinery. Nature 440, 631–636.
1764
A. Zanzoni et al. / FEBS Letters 583 (2009) 1759–1765
[10] Krogan, N.J. et al. (2006) Global landscape of protein complexes in the yeast Saccharomyces cerevisiae. Nature 440, 637–643. [11] Giot, L. et al. (2003) A protein interaction map of Drosophila melanogaster. Science 302, 1727–1736. [12] Li, S. et al. (2004) A map of the interactome network of the metazoan C. elegans. Science 303, 540–543. [13] Rual, J.F. et al. (2005) Towards a proteome-scale map of the human protein– protein interaction network. Nature 437, 1173–1178. [14] Stelzl, U. et al. (2005) A human protein–protein interaction network: a resource for annotating the proteome. Cell 122, 957–968. [15] Collins, S.R., Kemmeren, P., Zhao, X.C., Greenblatt, J.F., Spencer, F., Holstege, F.C., Weissman, J.S. and Krogan, N.J. (2007) Toward a comprehensive atlas of the physical interactome of Saccharomyces cerevisiae. Mol. Cell Proteomics 6, 439–450. [16] Yu, H. et al. (2008) High-quality binary protein interaction map of the yeast interactome network. Science. [17] Venkatesan, K. et al. (2009) An empirical framework for binary interactome mapping. Nat. Methods 6, 83–90. [18] Scriver, C.R. and Waters, P.J. (1999) Monogenic traits are not simple: lessons from phenylketonuria. Trends Genet. 15, 267–272. [19] Dipple, K.M. and McCabe, E.R. (2000) Modifier genes convert ‘‘simple” Mendelian disorders to complex traits. Mol. Genet. Metab. 71, 43–50. [20] Badano, J.L. and Katsanis, N. (2002) Beyond mendel: an evolving view of human genetic disease transmission. Nat. Rev. Genet. 3, 779–789. [21] Oti, M. and Brunner, H.G. (2007) The modular nature of genetic diseases. Clin. Genet. 71, 1–11. [22] Ferrer-Costa, C., Orozco, M. and de la Cruz, X. (2002) Characterization of disease-associated single amino acid polymorphisms in terms of sequence and structure properties. J. Mol. Biol. 315, 771–786. [23] Schuster-Böckler, B. and Bateman, A. (2008) Protein interactions in human genetic diseases. Genome Biol. 9, R9. [24] Wong, J.M.S., Ionescu, D. and Ingles, C.J. (2003) Interaction between BRCA2 and replication protein A is compromised by a cancer-predisposing mutation in BRCA2. Oncogene 22, 28–33. [25] Ivan, M. and Kaelin, W.G. (2001) The von Hippel–Lindau tumor suppressor protein. Curr. Opin. Genet. Dev. 11, 27–34. [26] Ohh, M. et al. (2000) Ubiquitination of hypoxia-inducible factor requires direct binding to the beta-domain of the von Hippel–Lindau protein. Nat. Cell Biol. 2, 423–427. [27] Miller, J.L., Cunningham, D., Lyle, V.A. and Finch, C.N. (1991) Mutation in the gene encoding the alpha chain of platelet glycoprotein Ib in platelet-type von Willebrand disease. Proc. Natl. Acad. Sci. USA 88, 4761–4765. [28] Chatr-aryamontri, A., Ceol, A., Palazzi, L.M., Nardelli, G., Schneider, M.V., Castagnoli, L. and Cesareni, G. (2007) MINT: the Molecular INTeraction database. Nucleic Acids Res. 35, D572–D574. [29] Kerrien, S. et al. (2007) IntAct – open source resource for molecular interaction data. Nucleic Acids Res. 35, D561–D565. [30] Gandhi, T.K.B. et al. (2006) Analysis of the human protein interactome and comparison with yeast, worm and fly interaction datasets. Nat. Genet. 38, 285–293. [31] Goh, K.-I., Cusick, M.E., Valle, D., Childs, B., Vidal, M. and Barabási, A.-L.s. (2007) The human disease network. Proc. Natl. Acad. Sci. USA 104, 8685–8690. [32] Feldman, I., Rzhetsky, A. and Vitkup, D. (2008) Network properties of genes harboring inherited disease mutations. Proc. Natl. Acad. Sci. USA 105, 4323– 4328. [33] Xu, J. and Li, Y. (2006) Discovering disease-genes by topological features in human protein–protein interaction network. Bioinformatics 22, 2800–2805. [34] Yildirim, M.A., Goh, K.-I., Cusick, M.E., Barabási, A.-L. and Vidal, M. (2007) Drugtarget network. Nat. Biotechnol. 25, 1119–1126. [35] Rambaldi, D., Giorgi, F.M., Capuani, F., Ciliberto, A. and Ciccarelli, F.D. (2008) Low duplicability and network fragility of cancer genes. Trends Genet. 24, 427–430. [36] Wolf-Yadlin, A. et al. (2006) Effects of HER2 overexpression on cell signaling networks governing proliferation and migration. Mol. Syst. Biol. 2, 54. [37] Xu, J.J., Henstock, P.V., Dunn, M.C., Smith, A.R., Chabot, J.R. and de Graaf, D. (2008) Cellular imaging predictions of clinical drug-induced liver injury. Toxicol. Sci. 105, 97–105. [38] Schena, M., Shalon, D., Davis, R.W. and Brown, P.O. (1995) Quantitative monitoring of gene expression patterns with a complementary DNA microarray. Science 270, 467–470. [39] Golub, T.R. et al. (1999) Molecular classification of cancer: class discovery and class prediction by gene expression monitoring. Science 286, 531–537. [40] Chen, X. et al. (2002) Gene expression patterns in human liver cancers. Mol. Biol. Cell 13, 1929–1939. [41] van’t Veer, L.J. et al. (2002) Gene expression profiling predicts clinical outcome of breast cancer. Nature 415, 530–536. [42] Wang, Y. et al. (2005) Gene-expression profiles to predict distant metastasis of lymphnode-negative primary breast cancer. Lancet 365, 671–679. [43] Calvano, S.E. et al. (2005) A network-based analysis of systemic inflammation in humans. Nature 437, 1032–1037. [44] Ideker, T. and Sharan, R. (2008) Protein networks in disease. Genome Res. 18, 644–652. [45] Stelzl, U. et al. (2005) A human protein–protein interaction network: a resource for annotating the proteome. Cell 122, 957–968. [46] Chuang, H.-Y., Lee, E., Liu, Y.-T., Lee, D. and Ideker, T. (2007) Network-based classification of breast cancer metastasis. Mol. Syst. Biol. 3, 140.
[47] Mani, K.M., Lefebvre, C., Wang, K., Lim, W.K., Basso, K., Dalla-Favera, R. and Califano, A. (2008) A systems biology approach to prediction of oncogenes and molecular perturbation targets in B-cell lymphomas. Mol. Syst. Biol. 4, 169. [48] Ewing, R.M. et al. (2007) Large-scale mapping of human protein–protein interactions by mass spectrometry. Mol. Syst. Biol. 3, 89. [49] Lim, J. et al. (2006) A protein–protein interaction network for human inherited ataxias and disorders of Purkinje cell degeneration. Cell 125, 801–814. [50] Kaltenbach, L.S. et al. (2007) Huntingtin interacting proteins are genetic modifiers of neurodegeneration. PLoS Genet. 3, e82. [51] Camargo, L.M. et al. (2007) Disrupted in Schizophrenia 1 interactome: evidence for the close connectivity of risk genes and a potential synaptic basis for schizophrenia. Mol. Psychiatry 12, 74–86. [52] Pujana, M.A. et al. (2007) Network modeling links breast cancer susceptibility and centrosome dysfunction. Nat. Genet. 39, 1338–1349. [53] Russell, R.B. and Aloy, P. (2008) Targeting and tinkering with interaction networks. Nat. Chem. Biol. 4, 666–673. [54] Pache, R.A., Zanzoni, A., Naval, J., Mas, J.M. and Aloy, P. (2008) Towards a molecular characterisation of pathological pathways. FEBS Lett. 582, 1259– 1265. [55] Kitano, H. (2004) Biological robustness. Nat. Rev. Genet. 5, 826–837. [56] Neduva, V. and Russell, R.B. (2006) Peptides mediating interaction networks: new leads at last. Curr. Opin. Biotechnol. 17, 465–471. [57] Wells, J.A. and McClendon, C.L. (2007) Reaching for high-hanging fruit in drug discovery at protein–protein interfaces. Nature 450, 1001–1009. [58] Vassilev, L.T. et al. (2004) In vivo activation of the p53 pathway by smallmolecule antagonists of MDM2. Science 303, 844–848. [59] Moerke, N.J. et al. (2007) Small-molecule inhibition of the interaction between the translation initiation factors eIF4E and eIF4G. Cell 128, 257–267. [60] He, M.M. et al. (2005) Small-molecule inhibition of TNF-alpha. Science 310, 1022–1025. [61] Parthasarathi, L., Casey, F., Stein, A., Aloy, P. and Shields, D.C. (2008) Approved drug mimics of short peptide ligands from protein interaction motifs. J. Chem. Inf. Model. 48, 1943–1948. [62] Roden, D.M. and George, A.L. (2002) The genetic basis of variability in drug responses. Nat. Rev. Drug Discov 1, 37–44. [63] Roden, D.M. et al. (2006) Pharmacogenomics: challenges and opportunities. Ann. Intern. Med. 145, 749–757. [64] Bentley, D.R. et al. (2008) Accurate whole human genome sequencing using reversible terminator chemistry. Nature 456, 53–59. [65] Wang, J. et al. (2008) The diploid genome sequence of an Asian individual. Nature 456, 60–65. [66] Hopkins, A.L. (2008) Network pharmacology: the next paradigm in drug discovery. Nat. Chem. Biol. 4, 682–690. [67] Overington, J.P., Al-Lazikani, B. and Hopkins, A.L. (2006) How many drug targets are there? Nat. Rev. Drug Discov. 5, 993–996. [68] Schrattenholz, A. and Soskic´, V. (2008) What does systems biology mean for drug development? Curr. Med. Chem. 15, 1520–1528. [69] Roth, B.L., Sheffler, D.J. and Kroeze, W.K. (2004) Magic shotguns versus magic bullets: selectively non-selective drugs for mood disorders and schizophrenia. Nat. Rev. Drug Discov. 3, 353–359. [70] van der Greef, J. and McBurney, R.N. (2005) Innovation: rescuing drug discovery: in vivo systems pathology and systems pharmacology. Nat. Rev. Drug Discov. 4, 961–967. [71] FitzGerald, G.A. and Patrono, C. (2001) The coxibs, selective inhibitors of cyclooxygenase-2. N Engl. J. Med. 345, 433–442. [72] Zarraga, I.G.E. and Schwarz, E.R. (2007) Coxibs and heart disease: what we have learned and what else we need to know. J. Am. Coll Cardiol. 49, 1–14. [73] Paolini, G.V., Shapland, R.H.B., van Hoorn, W.P., Mason, J.S. and Hopkins, A.L. (2006) Global mapping of pharmacological space. Nat. Biotechnol. 24, 805– 815. [74] Gregori-Puigjané, E. and Mestres, J. (2008) A ligand-based approach to mining the chemogenomic space of drugs. Comb. Chem. High Throughput Screen. 11, 669–676. [75] Nacher, J.C. and Schwartz, J.-M. (2008) A global view of drug-therapy interactions. BMC Pharmacol. 8, 5. [76] Bolognesi, M.L., Cavalli, A. and Melchiorre, C. (2009) Memoquin: a multitargetdirected ligand as an innovative therapeutic opportunity for Alzheimer’s disease. Neurotherapeutics 6, 152–162. [77] Apsel, B. et al. (2008) Targeted polypharmacology: discovery of dual inhibitors of tyrosine and phosphoinositide kinases. Nat. Chem. Biol. 4, 691–699. [78] Rasko, D.A. et al. (2008) Targeting QseC signaling and virulence for antibiotic development. Science 321, 1078–1080. [79] Todorovska, A., Roovers, R.C., Dolezal, O., Kortt, A.A., Hoogenboom, H.R. and Hudson, P.J. (2001) Design and application of diabodies, triabodies and tetrabodies for cancer targeting. J. Immunol. Methods 248, 47–66. [80] Kitano, H. (2007) A robustness-based approach to systems-oriented drug design. Nat. Rev. Drug Discov. 6, 202–210. [81] O’Connor, K.A. and Roth, B.L. (2005) Finding new tricks for old drugs: an efficient route for public-sector drug discovery. Nat. Rev. Drug Discov. 4, 1005–1014. [82] Lee, M.S. et al. (2007) The novel combination of chlorpromazine and pentamidine exerts synergistic antiproliferative effects through dual mitotic action. Cancer Res. 67, 11359–11367. [83] Campillos, M., Kuhn, M., Gavin, A.-C., Jensen, L.J. and Bork, P. (2008) Drug target identification using side-effect similarity. Science 321, 263–266.
A. Zanzoni et al. / FEBS Letters 583 (2009) 1759–1765 [84] Ashburn, T.T. and Thor, K.B. (2004) Drug repositioning: identifying and developing new uses for existing drugs. Nat. Rev. Drug Discov. 3, 673– 683. [85] Lalonde, R.L. and Honig, P. (2008) Clinical pharmacology in the era of biotherapeutics. Clin. Pharmacol. Ther. 84, 533–536. [86] Leader, B., Baca, Q.J. and Golan, D.E. (2008) Protein therapeutics: a summary and pharmacological classification. Nat. Rev. Drug Discov. 7, 21–39. [87] Garcia-Higuera, I., Taniguchi, T., Ganesan, S., Meyn, M.S., Timmers, C., Hejna, J., Grompe, M. and D’Andrea, A.D. (2001) Interaction of the Fanconi anemia proteins and BRCA1 in a common pathway. Mol. Cell 7, 249–262.
1765
[88] El-Amraoui, A. and Petit, C. (2005) Usher I syndrome: unravelling the mechanisms that underlie the cohesion of the growing hair bundle in inner ear sensory cells. J. Cell Sci. 118, 4593–4603. [89] Nieuwenhuis, E. and Hui, C.c. (2005) Hedgehog signaling and congenital malformations. Clin. Genet. 67, 193–208. [90] Wesenbeeck, L.V. et al. (2003) Six novel missense mutations in the LDL receptorrelated protein 5 (LRP5) gene in different conditions with an increased bone density. Am. J. Hum. Genet. 72, 763–771. [91] Corr, M. (2008) Wnt-beta-catenin signaling in the pathogenesis of osteoarthritis. Nat. Clin. Pract. Rheumatol. 4, 550–556.