Mechanisms of Ageing and Development 128 (2007) 117–124 www.elsevier.com/locate/mechagedev Longevity network: Construction and implications Arie Budovsky a, Amir Abramovich a, Raphael Cohen b, Vered Chalifa-Caspi b, Vadim Fraifeld a,* a Department of Microbiology and Immunology, Faculty of Health Sciences, Center for Multidisciplinary Research in Aging, Ben-Gurion University of the Negev, P.O. Box 653, Beer-Sheva 84105, Israel b Bioinformatics Support Unit, National Institute for Biotechnology in the Negev, Ben-Gurion University of the Negev, P.O. Box 653, Beer-Sheva 84105, Israel Available online 20 November 2006 Abstract The vast majority of studies on longevity have focused on individual genes/proteins, without adequately addressing the possible role of interactions between them. This study is the first attempt towards constructing a ‘‘longevity network’’ via analysis of human protein–protein interactions (PPIs). For this purpose, we (i) compiled a complete list of established longevity genes from different species, including those that most probably affect the longevity in humans, (ii) defined the human orthologs of the longevity genes, and (iii) determined whether the encoded proteins could be organized as a network. The longevity gene-encoded proteins together with their interacting proteins form a continuous network, which fits the criteria for a scale-free network with an extremely high contribution of hubs to the network connectivity. Most of them have never been annotated before in connection with longevity. Remarkably, almost all of the hubs of the ‘‘longevity network’’ were reported to be involved in at least one age-related disease (ARD), with many being involved in several ARDs. This may be one of the ways by which the proteins with multiple interactions affect the longevity. The hubs offer the potential of being primary targets for longevity-promoting interventions. # 2006 Elsevier Ireland Ltd. All rights reserved. Keywords: Longevity genes and proteins; Protein–protein interactions; Longevity network; Age-related diseases 1. Introduction: from longevity genes to a ‘‘longevity that, until now, the vast majority of biogerontological studies network’’ have focused on individual genes/proteins, without considering the possible role of interactions between them. Given that the Genetic manipulations in model organisms (S. cerevisae, LAGs act in a cooperative manner, it seems unrealistic to C .elegans, D. melanogaster, M. musculus) and the studies of examine experimentally the effect of various gene combina- genetic polymorphisms in human populations revealed a tions on aging/longevity. Ironically, the time required for such number of longevity-associated genes (LAGs) and pathways an examination might, to some extent, be comparable to that of (Guarente and Kenyon, 2000; Jazwinski, 2000; Finch and evolution. Thus, a more efficient strategy is required to focus Ruvkun, 2001a; Hekimi and Guarente, 2003; Khalyavkin and the studies on the basic mechanisms of aging and longevity. Yashin, 2003a, b; Atzmon et al., 2005, 2006; de Magalhaes, One of the principles of such a strategy could be based on the 2005; Franceschi et al., 2005; Kenyon, 2005; Vijg and Suh, idea of an interactome or, more specifically, a network. 2005; Warner, 2005; Christensen et al., 2006; Gami and During the last decade, it has become more and more Wolkow, 2006). However, our knowledge of the key obvious that biological systems function as complex networks. determinants of aging and/or longevity is still limited. Indeed, Thus, the properties of a system should not be reduced to the in most cases, the increase in the maximum life span of mutants properties of its components (though they are also important) versus wild type did not exceed 20–40%. One of the reasons is but rather that the network’s topology determines the system’s behavior (Barabasi and Albert, 1999; Albert and Barabasi, 2002). The important point is that the formation and the * Corresponding author. Tel.: +972 8 6477292; fax: +972 8 6477626. properties of various complex networks (biological, technical, E-mail addresses: [email protected], social) are governed by universal principles (Albert and [email protected] (V. Fraifeld). Barabasi, 2002; Barabasi and Oltvai, 2004). 0047-6374/$ – see front matter # 2006 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.mad.2006.11.018 118 A. Budovsky et al. / Mechanisms of Ageing and Development 128 (2007) 117–124 Despite of existing limitations (for example, incomplete and Table 1 partially imperfect data; for recent review see Siegal et al., Distribution of established LAGs according to the species of origin 2007), the network-based approach has been successfully Species Number of LAGs applied for analysis of the protein–protein networks in yeast, Yeast, S. cerevisae 62 (14.5%) worms, and flies (Jeong et al., 2001; Yook et al., 2004; Hahn Worm, C. elegans 252 (58.9%) and Kern, 2005), the protein–protein network for degeneration Fly, D. melanogaster 45 (10.5%) of Purkinje cells and human inherited ataxias (Lim et al., 2006), Mouse, M. musculus 49 (11.4%) the neurogenic network in Drosophila (Meir et al., 2002), and Human, H. sapiens 19 (4.7%) Other, P. anserine 1 (0.2%) the metabolic network in E. coli (Ravasz et al., 2002). Most biological networks examined thus far (protein–protein net- Percentage of total number of LAGs (n = 428) is presented in parentheses. works, in particular) are scale-free, which follow a power-law distribution of connectivity: P(k) kÀg, where P(k) is the number of C. elegans LAGs were recently identified by means probability that a selected node has exactly k connections of RNAi screens on the genome wide scale (Lee et al., 2003; (degrees) with other nodes (e.g., proteins); g is the degree Hamilton et al., 2005; Hansen et al., 2005). This powerful exponent, a characteristic value for a given network which approach has not as yet been used widely for identification of determines many properties of the system. The smaller the g LAGs in other model organisms. value, the more important is the role of the nodes with a high During the last decade more than 120 human genes were connectivity in the network (Barabasi and Oltvai, 2004). studied in relation to exceptional longevity (Atzmon et al., Possible relationships between LAGs or their products 2005; Franceschi et al., 2005). However, only a few of them (LAPs, longevity-associated proteins) are only beginning to be were confirmed by independent studies in different human defined. Thus far, the interactions between them have mainly populations (Christensen et al., 2006). been considered in view of their connectivity within entire or With this in mind, we selected 13 genes with the most limited (e.g., nucleus, cytoplasm) protein interactomes of S. probable longevity-predisposing polymorphisms. Also, six cerevisae (Promislow, 2004; Ferrarini et al., 2005), C. elegans, others were included in the present analysis since they were and D. melanogaster (Ferrarini et al., 2005). In general, an shown to be involved in human progeroid syndromes. importance for considering the role of interactions in aging, in Next, we determined the human orthologs for the LAGs particular between proteins, has been proposed (Kirkwood and established in model organisms. These data were mostly Kowald, 1997; Promislow and Pletcher, 2002). extracted from the ‘‘InParanoid’’ database – Eukaryotic The present study is the first attempt towards constructing a Ortholog Groups (http://inparanoid.cgb.ki.se, O’Brien et al., ‘‘longevity network’’ via analysis of human protein–protein 2005) and partially from NCBI HomoloGene (http:// interactions. For this purpose, we (i) compiled a full list of www.ncbi.nlm.nih.gov/entrez/query.fcgi?db=homologene) established LAGs, including those that most probably affect the and Wormbase (http://www.wormbase.org). A common longevity in humans, (ii) defined the human orthologs of the method for the determination of orthologs is described LAGs established in model organisms, and (iii) determined elsewhere (O’Brien et al., 2005) and is based on pair-wise whether the encoded proteins could be organized as a network. similarity scores which are by default calculated with the NCBI The rationale behind the construction of the ‘‘longevity BLAST program (best-best hits between sequences from two network’’ on the basis of human protein–protein interactions different species). (PPIs) was grounded on the following: (a) many LAGs and The majority of LAGs reported for the model organisms had longevity-associated pathways are evolutionary conserved, human orthologs (Fig. 1), indicating their high evolutionary from yeast to humans (reviewed by Warner, 2005); (b) the large number of annotated human PPIs are available from the BioGRID database; (c) the possibility to connect the data to age-related diseases that have been studied extensively in humans. 2. General characterization of longevity-associated genes and proteins Initially, we compiled a full list of LAGs established thus far. A comprehensive analysis of scientific literature revealed 428 LAGs, 409 in model organisms and 19 in humans (Table 1). Their partial or full loss-of-function mutations, RNAi-induced gene silencing, over-expression, or genetic polymorphisms were reported to promote longevity or cause premature aging. Fig. 1. Proportion of human orthologs for LAGs and for all genes of the model organisms, extracted from the ‘‘InParanoid’’ database (http://inparanoid.cgb.- More than a half of the LAGs identified originate from the ki.se). The Fisher’s exact test (one-sided) was highly significant for each pairs studies in C. elegans, the most investigated model organism in (confidence level of 0.95): p = 6.1 Â 10À5, S. cerevisae; p < 2.2 Â 10À16, the genetics of aging and longevity. Of note, a considerable C. elegans; p = 0.00031, D. melanogaster; p = 4.6 Â 10À7, M. musculus. A. Budovsky et al. / Mechanisms of Ageing and Development 128 (2007) 117–124 119 Fig. 2. Distribution of longevity-associated genes/proteins by function and cell location.