Human Synthetic LethalInference as Potential Anti-cancer Target Gene DetectionNuria Conde-PueyoAndreea MunteanuRicard V. SoléCarlos Rodriguez-Caso

SFI WORKING PAPER: 2009-03-006

SFI Working Papers contain accounts of scientific work of the author(s) and do not necessarily represent theviews of the Santa Fe Institute. We accept papers intended for publication in peer-reviewed journals or proceedings volumes, but not papers that have already appeared in print. Except for papers by our externalfaculty, papers must be based on work done at SFI, inspired by an invited visit to or collaboration at SFI, orfunded by an SFI grant.©NOTICE: This working paper is included by permission of the contributing author(s) as a means to ensuretimely distribution of the scholarly and technical work on a non-commercial basis. Copyright and all rightstherein are maintained by the author(s). It is understood that all persons copying this information willadhere to the terms and constraints invoked by each author's copyright. These works may be reposted onlywith the explicit permission of the copyright holder.www.santafe.edu

SANTA FE INSTITUTE

Human synthetic lethal inference as potential anti-cancer target gene detection

Nuria Conde-Pueyoa, Andreea Munteanua, Ricard V. Solea,b, Carlos Rodrıguez-Casoa,∗

aICREA-Complex Systems Lab, Universitat Pompeu Fabra (PRBB-GRIB), Dr Aiguader 88, E-08003 Barcelona, SpainbSanta Fe Institute, 1399 Hyde Park Rd., Santa Fe, NM 87501, USA

Abstract

Background:. Two genes are called synthetic lethal (SL) if mutation of either alone is not lethal, but mutation ofboth leads to death or a significant decrease in the organism’s fitness. The detection of SL gene pairs constitutes apromising alternative for anti-cancer therapy. As cancer cells exhibit a large number of mutations, the determinationof these mutated genes’ SL partners may provide specific anti-cancer drug candidates, with minor perturbations to thehealthy cells. Since existent SL data is mainly restricted to yeast screenings, the road towards human SL candidatesis limited to inference methods.

Results:. In the present work, we use phylogenetic analysis and database manipulation (BioGRID for interactions,Ensembl and NCBI for homology, Gene Ontology for GO attributes) in order to reconstruct the phylogenetically-inferred SL gene network for humans. In addition, availabledata on cancer mutated genes (COSMIC and GeneCensusdatabases) as well as on existent approved drugs (DrugBank database) provides supporting evidence of our selectionof cancer-therapy candidates.

Conclusions:.Our work provides a complementary alternative to the current methods for drug discovering and genetarget identification in anti-cancer research.

Key words: Synthetic lethality, cancer therapy, high-throughput analysis

1. Background

High-throughput analyseshave provided a tremen-dous boost to massive drug screening Pereira &Williams (2007). However, these improved techniquesare still blind to biological or structural knowledge. Inthis sense,chemogenomicsprovides a complementarystrategy for a rational screening that includes structuralinformation of chemical compounds for gene targetsBajorath (2002); Rognan (2007). Computational ap-proaches in this so-called virtual screening allow thematching of compounds to their specific gene-producttargets, completing the experimental screening Mestreset al. (2008). However, the computational approachis still limited by the huge combinatorics representedby the chemical space of possibilities associated to thecompounds and their possible targets. As a conse-quence, all these experimental and computational ap-proaches require the use of the cumulative biological

∗Corresponding author: carlos.rodriguez@upf.edu

knowledge. For this purpose, database integration intoan ontological organization of the current biologicalknowledge has been suggested as a way to reduce thecombinatorics either in virtual or experimental screen-ings Agarwal & Searls (2008).

The work presented here belongs to this last frame-work, intended as a tool for identifying potential tar-gets for anti-cancer therapy. Cancer is a heterogeneousdisease with numerous causes and typologies Weinberg(2007). One of the essential traits of cancer progres-sion is the underlying high mutational capacity of tumorcells Loeb (2001); Loeb et al. (2008); Sole & Deisboeck(2004), having as a consequence the rapid adaptative ca-pacity of the disease. It has been suggested that theseingredients define cancer progression as a Darwinianmicro-evolutionary process Merlo et al. (2006). As aconsequence, cancer cells mutated in essential genes areeliminated from the tumor population. Therefore, it isexpected that essential genes are conserved in cancer.Under this perspective, targeting essential genes in anti-cancer therapy could kill malignant cells, but might re-

Preprint submitted to Preprint March 11, 2009

Figure 1: The rationale of synthetic lethality applied to the design ofnovel anti-cancer therapies. Two linked nodes (blue circles) representa SL interaction. (A) In cancer disease, one of the SL partners wouldappear mutated (red triangle) contrasting to healthy cells where nomutation is accounted for. (B) The attack by drugs to the SL partnerof the cancer-related mutated gene might cause a selective damage totumor cells. In this case, the inactivation ofboth SL partners onlyhappens in tumor cells.

sult to be lethal for healthy cells too. This is the caseof the anti-proliferative drugs that also damage highturnover tissues, such as epithelium.

The problems reported from the failure of mostsingle-target drug treatments Fernandez et al. (2009)suggest that a new perspective is needed. In this context,a different conceptual framework related with syntheticlethality has been suggested for anti-cancer researchHartwell et al. (1997); Whitehurst et al. (2007); Meur& Gentleman (2008). Two genes are called syntheticlethal (SL) if mutation of either alone is not lethal, butmutation of both leads to death or a significant decreasein organism’s fitness. The rationale of synthetic lethal-ity offers new insights on selective anti-cancer therapydesign by exploiting the existence of SL partners ofmutated (cancer-related) genes Hartwell et al. (1997);Kamb (2003); Kaelin (2005). Accordingly, given a mu-tated gene causing function deletion in a cancer cell, anattack using specific drugs to block the activity of oneof its SL partners would cause a lethal condition in suchtumor cells. Meanwhile, only minor damage in healthycells would be expected, constituting thus a selectiveanti-cancer therapy (see Figure 1). And thus, this ap-proximation can help to overcome a dramatic limitationin drug design.

Another relevant aspect in drug screening is that onedrug is tested only for a specific disease and relatedpathologies. Given a SL pair of genes as describedabove, one cancer mutated and the other non-mutated,conceptually it is possible that an already approved andeven commercialized drug might block the activity ofthe non-mutated gene product. Therefore, SL-partner

Figure 2: Schematic representation of our methodology. Potentialanti-cancer gene-targets are proposed for future experimental valida-tion.

screening has a special interest for gene-target identifi-cation but also for drug repositioning, i.e, the discover-ing of novel uses for old drugs Ashburn & Thor (2004).

Unfortunately, large-scale screenings of SL genepairs have been performed only in yeast Forsburg(2001); Pan et al. (2004); Ooi et al. (2006); Boone et al.(2007) and, in a significantly smaller degree, inC. ele-gansJorgensen & Mango (2002); Baugh et al. (2005);Lehner et al. (2006) and in other model organisms. Toovercome this limitation, we propose the use of the phy-logenetic inference of SL genes from yeast to human forpharmacological purposes. We emphasize though thatthe aim of this work is not to provide a general infer-ence method of SL network from one organism to an-other. Instead, it adds to the rationale of drug designby supplying a candidate-list of human gene pairs, po-tentially SL, that could constitute the basis for futurepharmacological testing.

Additionally, network thinking has provided an ex-cellent framework for the study of very large geneticsystems. In particular, gene-disease Goh et al. (2007);Platzer et al. (2007), gene-target Yildirim et al. (2007)

2

Figure 4: About here.

Figure 5: The 933-giant component extracted from the 1002-nodesnetwork of inferred SL human genes (Supplementary material S6andS6a). The human orthologs from yeast are related through a yeastSL-relation.

and synthetic-lethal Boone et al. (2007); Chipman &Singh (2009); Paladugu et al. (2008) network represen-tations have contributed to the understanding of thesesystems as a whole. Following this approach, we inte-grate the information in a network framework from sev-eral databases in order to provide supporting evidencefor candidates’ reliability.

2. Results

2.1. The iHSLN

Yeast SL network is a graph of 2383 nodes and 12707links. This network is an extension of the SL networkconstructed by Tong et al. (2004), who remarked that SLinteractions yield a giant component with a non-randomtopology of small-world characteristics.

The collection of inferred human genes from yeastSL genes revealed that 52% of the 2383 yeast genes,that is 1253 genes, are found in Ensemble (see Fig-ure 10) and thus have human orthologs. Previous anal-ysis of global genomic homology between species es-timated that the coverage betweenS. cerevisiaeandH.sapiensis about 20% Tatusov et al. (2003). Interest-ingly, the higher percentage observed from the set of

the yeast SL genes compared with the global coveragesuggests a possible conservation for SL proteins. Fromthe above-mentioned 1253 genes, only 1078 genes havean SL partner, that is those genes among the 1253 thathave SL partners without any human ortholog are elim-inated (Supplementary material S0). The final humanSL network presented a set of 1002 nodes and 2847links (Figure 5), where almost all the nodes formed agiant component (933 nodes) with a non-random topol-ogy. Considering yeast-to-human inference relations,the nodes in the human network resulted from 766one-to-onehomology cases, 64 human nodes are composedof 138 genes due to then-to-onecases, and 170 nodesare formed of 386 genes due toone-to-mrelations (moreprecisely, 148 nodes include gene of similar biologicalfunction, while 21 nodes contain genes with heteroge-neous functionality). Only 2 nodes came from 2n-to-m situations. We mention that 1.5% (43 nodes) of thelinks in the human network were autolinks (see Figure9 B for explanation). This is a result of then-to-onecases, as among thesen yeast genes some may be SLpartners. This is an artifact of the human SL network.These autolinks in the iHSLN are not eliminated fromour network, as the discussion concerning these nodesas candidates needs to be addressed with caution. Westress once more that in our subsequent analysis we re-fer to nodes rather than to simple genes, as a node maycontain several genes.

2.2. Obtaining potential cancer-related SL targets2.2.1. Cancer-related database approach

In order to evaluate potential candidates, we firstidentify those human genes in our inference list that areknown to be involved in cancer mutations, as detailed inMaterial and methods. Given this information, the SLpartners of these genes are then potential candidates foranti-cancer therapy. In the inset of Figure 6 we repre-sent the 124-nodes sub-network containing those nodesin the human SL network that have been found in theCOSMIC and GeneCensus databases. In this network,we observe the SL interactions between cancer-relatedgenes. However, our objective is to obtain possible can-didates, that is the SL partners of these 124 that are notdescribed as cancer-related genes. Therefore, we extractfrom the SL human network also the first neighbors ofthese 124 nodes. We illustrate this network in Figure 6where triangles represent cancer-related nodes and cir-cles indicate their neighbors that are not known to becancer-related in the used databases. According to ourrationale, cancer-cancer SL interaction should not occurin a same tumor cell. In our data, cancer-related infor-mation is a compendium obtained from many samples

3

Figure 3: The orthology relation in the process of yeast-to-human inference method. Theone-to-onecase of one yeast gene having one humanhomolog (left). Then-to-onecase of several human genes having the same yeast homolog (middle) . Theone-to-mcase can be defined analogously.Then-to-mis a very rare case where more than one human gene are homologs of more than one yeast gene (right).

and cancer-types. As a consequence, if a cancer-cancerlink exists in our network, it indicates that the two mu-tations do not occur in the same sample. Therefore, thecancer-cancer SL interactions can provide useful infor-mation for our purpose.

One of the hubs observed in Figure 6 correspondswith CDC73. It has been observed that mutations inthis gene (also known as HRPT2) are associated withmalignancy in sporadic parathyroid tumors and hered-itary hyperparathyroidism-jaw tumor syndrome Howellet al. (2003). Parafibromin is a tumor suppressor proteinencoded by HRPT2 that binds to RNA polymerase II aspart of a PAF1 transcriptional regulatory complex. Ex-pression inhibition by HRPT2 RNA interference stim-ulates cell proliferation and increases the levels of thec-myc proto-oncogene product Lin et al. (2008). Ac-cording to Figure 6, the CDC73 hub establishes a largenumber of potential SL candidates. We must stress that,due to the long phylogenetic distance between humanand yeast, most of them are likely false positives. Nev-ertheless, biological arguments in favor of their rele-vance can be sought in order to strengthen their can-didacy. This is the case of the SL link between CDC73and genes related with the RNA polymerase machin-ery such as RPAP2 Jeronimo et al. (2007), and the tran-scription elongation factor TCEA3 Labhart & Morgan(1998) that capture the common function in RNA pro-cessing of these partners.

In addition, CDC73 has as SL partner in Figure 6 theJUP (junction plakoglobin orγ-catenin) protein, a com-ponent of the catenin complex. JUP is involved in celladhesion Knudsen & Wheelock (1992), an apparentlyvery distant function regarding RNA processing. How-ever,γ-catenin strongly activates the proto-oncogene c-myc Kolligs et al. (2000). According to this evidence,it seems reasonable to consider that a biological path-

Figure 7: SL-inferred human orthologs organized by GO numbers(Supplementary material S8). Genes’ color denotes their biologicalfunction (GO number): in blue, cell-cycle related genes, in red, DNAdamage related genes and in white, genes related to both processes.

way relates JUP and CDC73 (depicted as SL partners inour study) with c-myc, but with opposite effects. In thiscase, our methodology reveals a potential anti-cancerstrategy: inhibitingγ-catenin expression would com-pensate the activation of c-myc by the HTRP2 cancermutation.

2.2.2. GOA filtering approachIn a second approach, we added a new layer of infor-

mation by using the cancer-related functions extractedfrom GOA. In so doing, we searched for candidates thatmight not be included in cancer-related databases, but

4

Figure 6: Cancer-related genes and their first neighbors from the SL human network (Supplementary material S7). The inset illustrates thesub-network of cancer-related nodes alone (59 nodes). Nodes without SL partners (65 of 124) were eliminated from this picture. In the largenetwork, both cancer-related (triangles) and their neighbors (circles) in the SL human network. Only the links connecting a cancer-related geneand its neighbor (411 nodes and 694 links) are illustrated.

their functions are essential for cancer progression. As abench test for our approach, we focused the analysis ontwo biological functions: DNA damage and cell cycle.It is well established that mutations in genes involved inDNA damage repair or in cell cycle checkpoints are as-sociated with tumor progression Massague (2004). Forexample, synthetic lethal relationships between DNA-replication genes (such as certain DNA polymerases)and DNA-repair genes (such as mismatch-repair genes)are well documented in model organisms Yuen et al.(2007); Tarailo, Kitagawa & Rose (2007). Moreover,it seems likely that the efficiency of many anti-cancerdrugs that interfere with DNA synthesis is in some casesdue to the presence of tumor-associated mutations af-

fecting DNA repair or the response to DNA damageKaelin (2005).

The connected sub-network from Fig. 7 was obtainedby extracting the genes annotated for these functions.Interestingly, the analysis of each gene shows that forsix pairs one gene is directly cancer-related, whereas theother is not known to be so (Table 1). These latter genesconstitute preliminary and putative candidates for futureexperimental validation. It is also worth noting that, asshown in Table 1, we also consider as candidates someSL partners with not known relation to cancer. They areSL partners consistent with awithin pathwayapproachBoone et al. (2007) (they act, or are part of the samecomplex in a pathway).

5

Yeast Human Biological processYMR224C MRE11A Single-strand and double-strand-3’-5’ exonuclease activity for DNA repair and

recombination.YER016W MAPRE1 Microtubule assemblingYLR418C CDC73 PAF1 complex. Interacts with RNA pol II. Causes hyperparathyroidism-jaw tumor

syndromeYIL128W MMS19 Interacts with helicase subunits of NER-transcription factorYNL299W PAPD5 Sister chromatid cohesion. Resistance to Campothecin anti-tumor agentYJL115W ASF1 Histone chaperoneYJL194W CDC6 Initiation of DNA replication. Oncogenic activity throughrepression of

INK4/ARF locusYGL087C UBE2V2 Ubiquitin-conjugating enzymes without catalytic activityYJL013C BUB1 Involved in cell cycle checkpoint enforcement. Colorectalcancer

YGL058W UBE2B Central roll in postreplicative DNA repair in eukaryotic cellsYKL113C FEN1 Removes 5’ overhanging flaps in DNA repair and Okazaki fragments. Fast tumor

progressYGL240W ANAPC10 Component of anaphase promoting complex (APC), progression through mitosis

and G1 phase

*YDL132W CUL1 Core component of multiple cullin-RING-based SCF which mediate the ubiquiti-nation of proteins involved in cell cycle progression, signal transduction and tran-scription.

YOL133W RBX1 Component of the SCF.*YBR087W RFC3 form a complex required by DNA polymerase delta and epsilon.YNL102W POLA1 DNA polymerase subunit.*YJR068W RFC4 form a complex recruited by DNA polymeraseYDL102W POLD1 DNA polymerase subunit

Table 1: Yeast SL gene-pairs. Human homologs of the SL gene-pairs and the biological process that they fulfill. A preliminarycandidates list,where a human gene is cancer-related (in bold) and its SL partner is not known to be directly related with cancer. Asterisk symbol indicates SLpartners that belong to the same cellular pathway (within-pathwaymodel: Boone et al. (2007)).

An illustrative case by means of the GOA filteringis again CDC73, also observed in the previous section.In this case, GOA filtering evidences its relation withthe DNA damage related gene MMS19 (see Table 1)and helicase component. Interestingly, helicases havebeen proposed as targets for anti-cancer therapy Litmanet al. (2008) as they are closely related with the requiredgenetic instability for tumor progression.

2.2.3. Drug association to SL human target genes

DrugBank database information was used here to es-tablish possible relations between existing drugs and in-ferred SL human genes. We have found that 130 nodesof our human SL network contain genes associated toone or several drugs in the DrugBank. More specif-ically, 17% of them are anti-cancer drugs. SeeSup-plementary material S9a for pattern of interactions forthis set of nodes. As we shall see in the next section, thecombination of cancer-related and drug-target informa-tion into the SL human network produces a set of genes

that are known drug targets and have cancer-related SLpartners, i.e. the suitable candidates according to ourmethodology illustrated in Figure 1.

2.2.4. Finding SL partners of cancer-related genes as-sociated to drugs

As we previously argued, SL partners of cancer-related genes constitute a set of potential targets for anti-cancer treatment. The knowledge about drugs affectingthese genes provides, on one hand, supporting evidencesof our methodology. This is the case of cancer-unrelatedgenes for which there exists an anti-cancer drug, genesthat are also SL partners of a cancer-related gene. Onother hand, when the cancer-unrelated gene is associ-ated to a cancer-unrelated drug, we have a potential newuse of a drug that initially has not been conceived foranti-cancer treatment. The integration of this informa-tion is included in Figure 8. In this network, we elim-inated the cancer-unrelated nodes that are not associ-ated to any drug. The remaining set contains 47 isolated

6

nodes and a graph of 155 nodes of which 124 are cancer-related nodes and 31 are drug-target nodes. The set of47 nodes matches the number of cancer-related nodeshaving lost all the SL partners during filtering process,and it also indicates that no drug is associated to theirSL neighborhood in the current version of DrugBank.

Figure 8 also singles out (large circles and triangles)three SL pairs that are fundamental to existent anti-cancer treatments consisting in targeting SL partners ofthe cancer-related genes. In the first example, ERCC2(also known as XPD) is a cancer-related gene Kiyohara& Yoshimasu (2007). Studies of its homologous genein drosophyla show that an excess of XPD produces atitration of CAK complex and reduces CDK7 activity,leading to a cell-cycle arrest Chen et al. (2003). More-over, CDK7 is a protein of the CAK complex requiredfor cell-cycle progression Lolli & Johnson (2005). Ithas already been suggested as a potential target for anti-cancer therapies. Pharmacological implications are dis-cussed in Lolli & Johnson (2005).

In the second example, PPP6C is a phosphatase in-volved in cell-cycle progression in human cells throughthe control of cyclin D1 Stefansson & Brautigan (2007).Its SL partner, PSMB2, is a proteasome subunit relatedwith protein degradation processes. Bortezomib, a ther-apeutic proteasome inhibitor is a target of PSMB2 (ac-cording to DrugBank information). It is pharmacologi-cally approved for treating relapsed multiple myelomaand mantle cell lymphoma. In addition, the use ofBortezomib in hepathocarcinoma cells reduces the tran-scriptional levels of cyclin D1, among other effects,leading to cell-cycle arrest. It is postulated that cyclinD1 can act as an oncogene Hinds et al. (1994); Sanchezet al. (2008). In this case, we observe that PSMB2 andPPP6C affect the levels of this cell-cycle related gene.

The third example, BUB1*8 is a prototype member ofa family of genes, some of which encode proteins thatbind to the kinetochore and all of which are requiredfor a normal mitotic delay in response to spindle dis-ruption. Mutations in this gene have been associatedwith aneuploidy and several forms of cancer Cahill et al.(1998). Our results reveal that this gene is associated toTUBB*8 in the SL network. TUBB* represents a num-ber of genes encoding tubulin proteins. Tubulins aretargets of the anti-cancer drug Paclitaxel and the vincaalkaloids Vincristine and Vinblastine that also affect themitotic-spindle assembly process Jordan et al. (1998).

Table 2 contains a representative list of the assign-ment of a potential anti-cancer use for existent cancer-

8Here the∗ symbol implies that more than one phylogenetically-related proteins is contained in this node.

unrelated drugs. As examples of them, two genes (andtheir drugs) are commented. The first one is PRDX2(also known as peroxiredoxin-2), a gene encoding amember of the peroxiredoxin family of antioxidant en-zymes, which reduce hydrogen peroxide and alkyl hy-droperoxides Schafer & Werner (2008). It is interestingto mention that PRDX2 has been associated to cell pro-liferation and cell migration by regulation of the PDGFsignalling Choi et al. (2005). N-carbamoyl alanine and3-sulfinoalanine compounds were found to be inhibitorsof PRDX2 activity and it has not been yet associatedto any anti-cancer treatment. Two SL cancer-relatedpartners of PRDX2, RAD50 and MRE11A are part ofMRN protein-complex involved in DNA double-strandbreak repair, cell-cycle checkpoint activation, telomeremaintenance and meiotic recombination Williams et al.(2007).

The second example is the SL link between CDC42and the previously described PPP6C. CDC42 is a geneparticipating in the rearrangement of actin cytoskeleton,membrane trafficking and cell-cycle progression, and itappears to be involved in cardiovascular diseases, dia-betes and neuronal degenerative diseases Sinha & Yang(2008). This is a target-gene of two other compounds(aminophosphonic acid-guanylate ester and guanosine5’diphosphate) with no relation to cancer treatment inthe DrugBank. However, it has been described thatthis gene is also involved in tumorigenesis and tumorprogression, and the aberrant expression of CDC42has been associated to colorectal tumors Pulgar et al.(2008); Sinha & Yang (2008). It is worth mention-ing that it has been observed that CDC42 controls thecell growth of anaplastic large cell lymphoma throughits activation. Pharmacologic inhibition of CDC42 ac-tivity by secramine results in a cell-cycle arrest andapoptosis of these cells Pelish et al. (2006); Ambro-gio et al. (2008). We suggest thus that CDC42 inhi-bition by secramine constitutes a potential anti-cancertreatment, but unfortunately neither CDC42 nor se-cramine appeared in their respective databases used inthis study. This example emphasizes once more thatdatabase information is an useful starting point for se-lecting new candidates. The use of state-of-the-art in-formation would of course contribute to perform a morereliable identification of candidates.

Discussion

By the present study we propose a methodology forproviding liable candidates for future experimental val-idation as drug targets for anti-cancer therapy. The

7

Figure 8: Drug-cancer SL network representing the set of cancer-related nodes and their SL drug-associated partners (Supplementary materialS9). Triangles represent cancer-related nodes and circles, SL drug-associated partners. Red color indicates that this node has associated a genewhich is target of anti-cancer drug. Green color indicates anon-cancer drug associated to the node. Blue color only appears in triangles representingthat there is no drug in the DrugBank database associated to these cancer-related genes. Large nodes are examples of existent anti-cancer drugtarget with a cancer-related SL pair (PPP6C-PSMB2, BUB1*-TUBB*, ERCC2-DCK7), and also our suggested examples of novel use for existentcancer-unrelated drugs that target a SL partner of a cancer-related gene (RAD50-PRDX2, MRE11A-PRDX2 AND PPP6C-CDC42). The BUB1*node represents BUB1 and BUB1B genes, whereas TUBB* node includes TUBB2B,TUBB3,TUBB2A and TUBB6 tubulin family members.

8

Drug-target name Drug (Description)CDC25B Beta-Mercaptoethanol (Glutathione S-Transferase inhibitor); Cysteinesulfonic

Acid(Peroxiredoxin inhibitor); Double Oxidized Cysteine(Peptide Deformylase Pdf1inhibitor)

CDC42 Aminophosphonic Acid-Guanylate Ester (G25K GTP-Binding Protein inhibitor)CDK2 4-(2,4-Dimethyl-Thiazol-5-Yl)-Pyrimidin-2-Ylamine (Cell Division Protein Kinase 2 in-

hibitor)DDX6 D-tartaric acid (D-tartaric acid)EIF4E 7-Methyl-Gpppa (Eukaryotic Translation Initiation Factor 4E inhibitor); 7n-Methyl-8-

Hydroguanosine-5’-Diphosphate (Vp39 inhibitor)GDI1 Geran-8-Yl Geran(Rab GDP Disossociation Inhibitor Alpha inhibitor)GRB2 4-[(10s,14s,18s)-18-(2-Amino-2-Oxoethyl)-14-(1-Naphthylmethyl)-8,17,20-Trioxo-

7,16,19-Triazaspiro[5.14]Icos-11-En-10-Yl]Benzylphosphonic Acid (Growth FactorReceptor-Bound Protein 2 inhibitor)

HSP90AB1 9-Butyl-8-(3,4,5-Trimethoxybenzyl)-9h-Purin-6-Amine(Heat Shock Protein Hsp 90-Beta in-hibitor)

HSPA5 antihemophilic factor (Coagulation factor VIII precursor)LSS Dihydrofolic Acid (Dihydrofolate Reductase inhibitor)

PPP1CC 9,10-Deepithio-9,10-Didehydroacanthifolicin (Serine/Threonine Protein Phosphatase Pp1-Gam inhibitor)

PPP2CA Vitamin E (Dietary supplement)PPP3R2 Cyclosporine (Investigational Immunomodulatory Agents;Immunosuppressive Agents; An-

tifungal Agents; Dermatologic Agents; Enzyme Inhibitors;Antirheumatic Agents For treat-ment of transplant rejection, rheumatoid arthritis, severe psoriasis)

PRDX2 3-Sulfinoalanine (3-Hydroxy-3-Methylglutaryl-Coa Synthase inhibitor)SEC14L2 Palmitic Acid (Enzyme Inhibitors)SNAP25 Botulinum Toxin Type A (Anti-Wrinkle Agents; Antidystonic Agents; Neuromuscular

Blocking Agents)SOD1 S-Oxy Cysteine (Prolyl Oligopeptidase inhibitor)

SQ33LE Naftifine (Anti-Inflammatory Agents, Non-Steroidal; Antifungal Agents)

Table 2: A selection of cancer-unrelated SL partners of cancer-related genes and their associated drugs. These drugs are not oriented to anti-cancertreatment. In this list we excluded general metabolites described as dietary supplements in DrugBank. A short descriptionof drug activities areprovided in parenthesis.

methodology is based on the existence of synthetic-lethal relation between pairs of genes: two genes aresynthetically lethal if their simultaneous mutation leadsto nonviable organism, while their separate mutationhas no substantial effect on the organism’s fitness. Asconceptualized by previous works Kamb (2003); Kaelin(2005), we used here the extensive experimental dataon yeast in order to extend the knowledge to the humangenome, and more precisely to anti-cancer therapy.

The rationale behind such approach is that, assum-ing that there are specific mutations for cancer cells,the identification and artificial mutation (drug action)of their SL partners would result in the death of cancer-ous cells alone. The mutations would affect also healthycells, but would not drastically injure them. The com-bination of different biological databases provides po-tential filters for reducing the number of false positives.

These false positives result from the long evolutionarydistance and different architectures between yeast andhuman genomes. However, potential candidate pairs areprovided in these study that need to be subsequently val-idated through experimental approach. Thus the aim ofthis work is to present a methodology for providing po-tential anti-cancer targets based on evolutionary traits.At this point, the evolutionary conservation of SL pairsis controversial Tischler et al. (2008); Tarailo, Tarailo& Rose (2007); Yuen et al. (2007); Dixon et al. (2008),even though a recent study inclines the balance towardsa significant conservation of synthetic lethal interactionsbetween eukaryotes Dixon et al. (2008). In addition, theexistence of conserved SL associated to particular func-tions Yuen et al. (2007); Measday et al. (2005); Tarailo,Tarailo & Rose (2007) is promising evidence for the in-ference methodology presented here. Even though not

9

all yeast SL pairs are expected to be conserved in distantorganisms, those associated to essential functions havea higher conservation probability. The identification ofonly a few of such partners could constitute alone aninvaluable information in the strategy of drug design.

In this context, the analysis of some of these genesrevealed interesting biological features suggesting po-tential candidates for the development of anti-cancertherapies. In this work a gene is considered cancer-related if it belongs to COSMIC or Cancer Gene Censusdatabases. This implies certain limitations to our studysince it depends on the accuracy and completeness ofthese databases. In this context, the better annotation ofthe used database, the more reliable results can be ob-tained. An example of that is the case of CDC42-PPP6CSL pair.

As Loeb stated with its mutator-phenotype scenariofor cancer evolution Loeb (2001); Cahill et al. (1999);Sole & Deisboeck (2004), some genetic instability butnot too much is required for cancer progression. Anillustrative example, the BUB1-TUBB* SL pair, isclosely related with the strategy of forcing instability inorder to kill cancer cells. It is reasonable to assume thatan attack to tubulins by drugs in tumors where BUB1appears mutated may drift the tumor population towardsextinction by exceeding the limits of mutation tolerance.In this particular case, we speculate that treatments withvinca alcaloids should be more efficient in those cancerswhere BUB1 is mutated. Analogously, our results re-veal the suitability of an attack to MMS19 helicase com-ponent when CDC73 is mutated. However, other anti-cancer strategies such as the attack to protein degrada-tion function by blocking PSMB2 proteasome compo-nent has been also uncovered by the presented method-ology. In this case, our result suggests that this therapyshould be more efficient in tumors where PPP6C is mu-tated.

Conclusions

We have proposed by the present study a tool forphylogenetical inference of candidates for future exper-imental validation as drug targets in anti-cancer ther-apy. Once more, we stress that we do not argue in favorof a methodology of SL-genes inference across distantspecies, as it has been already discussed in the literatureto be a controversial step Dixon et al. (2008); Tischleret al. (2008). Rather our study has a pharmacologicalutility and constitutes an alternative for massive drugscreenings. In addition, the arguments brought forwardin favor of the proposed candidates above are strong

enough to justify their consideration for future experi-mental validation. Furthermore, we provide supplemen-tary material on the results discussed above in order tofoster the bioinformatics and pharmacological commu-nities towards further analysis of this methodology.

Methods

Our identification of potential anti-cancer gene tar-gets proceeds through the integration of the biologicalinformation originating from different databases. Fig-ure 2 illustrates the methodology for the data-collectionprocess and for the selection of potential anti-cancergene targets. The yeast SL network was constructedfrom the yeast SL interaction list available from Bi-oGRID database Stark et al. (2006). In this network,nodes represent genes and the link between them indi-cates a SL interaction, i.e. when both are simultane-ously mutated, a lethal condition is satisfied. The phy-logenetic inference from yeast to human genes was ob-tained from the Ensembl database1. The yeast-geneslist belonging to SL network was contrasted with thisdatabase. The inferred human SL network (in shortiHSLN) is then obtained by introducing the yeast SLinteractions on their human phylogenetically-conservedcounterparts, that is their orthologs (see also Sonnham-mer & Koonin (2002)). Subsequently, as we detail be-low, the resulting network was filtered by different bi-ological databases for public use: Gene ontology (GOand GOA), COSMIC and Cancer census gene, as wellas DrugBank database.

2.3. Obtaining the yeast synthetic lethal network

The collected data on yeast genetic screens includ-ing synthetic lethality is available at BioGRID database(version 2.0.38) Stark et al. (2006). From this data, weretain only those genes having a systematic name as itappears inwww.yeastgenome.org. The resultant compi-lation is derived from 1233 articles, with half of the totalof 12707 SL relations being contained in five main arti-cles: Tong et al. (2004); Pan et al. (2006); Krogan et al.(2003); Davierwala et al. (2005); Tong et al. (2001) .

2.4. Obtaining the iHSLN

Homologs betweenH. Sapiensand S. cerevisiaewere obtained by similarity measures from the whole-genome multiple alignments by the Ensembl compara-tive genomics team. Human genes in this study were

1Ensembl:f tp.ensembl.org/pub/current em f/ensemblcompara/homologies, March 3rd, 2008

10

Figure 9: Gene-node convention for the iHSLN construction.Ev-ery node in yeast SL network (YSLN) has associated gene. (A) Forn-to-one, the corresponding node in the iHSLN contains the humanparalogs of the yeast homologous counterpart. (B) In the casethattwo SL yeast partners are phylogenetically related to a single humangene, an autolink appears in the human SL network. This is and arti-fact of the method. However, we have kept this information to providea more detailed picture in case such a node is a suitable candidate.

named according HUGO gene nomenclature commit-tee (HGNC)2. Each entry of the homology data filecontains gene’s evolutionary history corresponding to agene tree that diverged from a common ancestor (seeFigures 3 and 9). In our study, gene conservation inyeast and human can be summarized in three types ofrelations. The simplest case corresponds toone-to-onerelation between yeast and human genes (orthologousrelation). However, duplication events during evolutionled to two alternative cases (see middle panel of Fig-ure 3), where one yeast gene has more than one humanhomolog (one-to-mrelation), and vice versa (n-to-onerelation) Sonnhammer & Koonin (2002). The case ofn-to-mwas rarely encountered (Figure 3, right panel).

A yeast-human inference relation is straightforwardfor the one-to-onecases, and the human gene inheritsthe SL neighbors of its yeast homolog. On the contrary,the paralog cases require the introduction of approxi-mate relations. More precisely, we manually curated allthe cases where the inference is notone-to-one. For theone-to-mcases, we collapsed the multiple human genesinto a single node that inherits the SL links of the yeastortholog (Figure 9 A). This situation does not affect thepattern of interactions derived from yeast SL network.In addition, by checking the biological function of thesem human genes, we classify them into subsets of sim-ilar functions, and collapse these subsets into separatenodes. Thus a yeastgenehavingmorthologs might cor-

2http://www.genenames.org/

Figure 10: The statistics of the genes vs nodes the from gene-inferenceprocess.

respond to more than onenode in the human SL net-work. Again, we emphasize the distinction between thesingle-gene nodes in the yeast SL network and the po-tentially multiple-gene nodes in the iHSLN. The statis-tics of the genes vs nodes the from gene-inference pro-cess is illustrated in Figure 10.

The n-to-onecases merge the SL information fromthe multiple yeast nodes, and therefore a single humangene inherits the SL interactions from more than oneyeast gene. The extreme case is the third one,n-to-m.

2.5. Filtering the iHSLNBiological filtering through the use of databases led

to the functional classification of the nodes forming theiHSLN into: (1) cancer-related genes, (2) genes relatedto the DNA repair mechanism and cell cycle (two rel-evant functions altered in most cancers) and (3) drug-target genes.

In order to identify cancer-related candidates we usethe COSMIC database3 that stores the current knowl-edge on somatic mutations and related details on humancancers. The information was completed by the use ofthe Cancer census genes4 that is a project to cataloguethose genes for which mutations have been causally im-plicated in cancer.

Gene Ontology (GO)5 database provides the biologi-cal description of gene products for a number of prede-fined functions. To relate one gene with its GO annota-tion, we used the gene ontology annotation (GO num-bers) from GOA6. We also used the available relation-ships between GO terms The Gene Ontology Consor-tium (2000) to create differentfilters for different bio-logical processes, in particular for Cell cycle and DNAdamage.

3COSMIC: www.sanger.ac.uk/genetics/CGP/cosmic: 37th version4GeneCensus: www.sanger.ac.uk/genetics/CGP/Census5GO: www.geneontology.org, April 6th, 20086GOA: www.ebi.ac.uk/GOA/ 8th,2008

11

In addition to these databases, we have also used theinformation extracted from the DrugBank7. It providesthe available studies from the US Food and Drud Ad-ministration that relate tested drugs and their gene tar-gets Hopkins (2007); Yildirim et al. (2007); Wishartet al. (2008). In this way, we inquire whether clinicalstudies of drugs acting on our proposed candidates havebeen previously described in literature.

All the representations of the resultant networks wereperformed with Cytoscape Shannon et al. (2003). Elec-tronic files are provided as Supplementary material.

3. Abbreviations

SL: Synthetic Lethal; iHSLN: inferred Human Syn-thetic Lethal Network; YSLN: Yeast Synthetic LethalNetwork; GO: Gene Ontology; HGNC: HUGO genenomenclature commitee.

4. Authors contributions

NCP performed database manipulation, graph con-struction and drafted the manuscript. AM and CRCwrote the manuscript, conceived and designed the in-vestigation. AM contributed to computational analysisand CRC performed the biological interpretation. RVScontributed to manuscript prepraration and promotedthe work. All authors have read and approved the finalmanuscript.

5. Acknowledgements

This work was supported by the EU 6th Frameworkprojects SYNLET (NEST-043312, AM and NCP) andComplexDis (NEST-043241, CRC), NIH (CA 113004,RVS) and Santa Fe Institute (RVS). We are grateful toBernd Mayer for inspiring this work, and to the SYN-LET consortium for useful discussions. We also thankNuria Lopez Bigas, Jordi Mestres Lopez and MiguelAngel Medina Torres for providing expertise and adviceon this subject.

6. Additional Files

7. Supplementary material

Supplementary material provided upon request.

7DrugBank: www.drugbank.ca, release 2.0

7.1. Supplementary FileS0: List of 1078 human geneshomologous to the 2383 SL yeast genes from theBioGRID database (Figure 10).

7.2. Supplementary FileS6a: List of SL relations be-tween the inferred human SL genes (iHSLN).

7.3. Supplementary FileS6: The network of SL rela-tions between the inferred human SL genes (iH-SLN) (Figure 5)- Cytoscape file.

7.4. Supplementary FileS7: Cancer-related genes andtheir first neighbors from the SL human network(Figure 6)- Cytoscape file.

7.5. Supplementary FileS8: An extract of SL-inferredhuman orthologs organized by GO numbers andrelated with DNA damage and cell-cycle (Figure 7)- Cytoscape file.

7.6. Supplementary FileS9a: Sub-network of SL nodesthat are also drug targets, and their first SL neigh-bors - Cytoscape file.

7.7. Supplementary FileS9: Drug-cancer SL networkrepresenting the set of cancer-related nodes andtheir SL drug-associated partners (Figure 8) - Cy-toscape file.

References

Agarwal, P. & Searls, D. B. 2008, Literature mining in supportof drugdiscovery.,Brief Bioinform.

Ambrogio, C., Voena, C., Manazza, A. D., Martinengo, C., Costa,C., Kirchhausen, T., Hirsch, E., Inghirami, G. & Chiarle, R. 2008,The anaplastic lymphoma kinase controls cell shape and growthofanaplastic large cell lymphoma through Cdc42 activation.,CancerRes68(21), 8899–8907.

Ashburn, T. T. & Thor, K. B. 2004, Drug repositioning: identifyingand developing new uses for existing drugs.,Nat Rev Drug Discov3(8), 673–683.

Bajorath, J. 2002, Integration of virtual and high-throughput screen-ing.,Nat Rev Drug Discov1(11), 882–894.

Baugh, L. R., Wen, J. C., Hill, A. A., Slonim, D. K., Brown, E. L.& Hunter, C. P. 2005, Synthetic lethal analysis of Caenorhabditiselegans posterior embryonic patterning genes identifies conservedgenetic interactions.,Genome Biol6(5), R45.

Boone, C., Bussey, H. & Andrews, B. J. 2007, Exploring genetic in-teractions and networks with yeast.,Nat Rev Genet8(6), 437–449.

Cahill, D. P., Kinzler, K. W., Vogelstein, B. & Lengauer, C. 1999, Ge-netic instability and darwinian selection in tumours.,Trends CellBiol 9(12), M57–M60.

Cahill, D. P., Lengauer, C., Yu, J., Riggins, G. J., Willson,J. K.,Markowitz, S. D., Kinzler, K. W. & Vogelstein, B. 1998, Mu-tations of mitotic checkpoint genes in human cancers.,Nature392(6673), 300–303.

Chen, J., Larochelle, S., Li, X. & Suter, B. 2003, Xpd/Ercc2 regulatesCAK activity and mitotic progression.,Nature 424(6945), 228–232.

Chipman, K. & Singh, A. 2009, Predicting genetic interactionswith random walks on biological networks.,BMC Bioinformatics10(1), 17.

12

Choi, M. H., Lee, I. K., Kim, G. W., Kim, B. U., Han, Y.-H., Yu,D.-Y., Park, H. S., Kim, K. Y., Lee, J. S., Choi, C., Bae, Y. S.,Lee, B. I., Rhee, S. G. & Kang, S. W. 2005, Regulation of PDGFsignalling and vascular remodelling by peroxiredoxin II.,Nature435(7040), 347–353.

Davierwala, A. P., Haynes, J., Li, Z., Brost, R. L., Robinson, M. D.,Yu, L., Mnaimneh, S., Ding, H., Zhu, H., Chen, Y., Cheng, X.,Brown, G. W., Boone, C., Andrews, B. J. & Hughes, T. R. 2005,The synthetic genetic interaction spectrum of essential genes.,NatGenet37(10), 1147–1152. BIOGRID.

Dixon, S. J., Fedyshyn, Y., Koh, J. L. Y., Prasad, T. S. K., Chahwan,C., Chua, G., Toufighi, K., Baryshnikova, A., Hayles, J., Hoe, K.-L., Kim, D.-U., Park, H.-O., Myers, C. L., Pandey, A., Durocher,D., Andrews, B. J. & Boone, C. 2008, Significant conservationof synthetic lethal genetic interaction networks between distantlyrelated eukaryotes.,Proc Natl Acad Sci U S A105(43), 16653–16658.

Fernandez, A., Crespo, A. & Tiwari, A. 2009, Is there a case for se-lectively promiscuous anticancer drugs?,Drug Discov Today14(1-2), 1–5.

Forsburg, S. L. 2001, The art and design of genetic screens: yeast.,Nat Rev Genet2(9), 659–668.

Goh, K.-I., Cusick, M. E., Valle, D., Childs, B., Vidal, M. & Barabasi,A.-L. 2007, The human disease network.,Proc Natl Acad Sci U SA 104(21), 8685–8690.

Hartwell, L. H., Szankasi, P., Roberts, C. J., Murray, A. W. &Friend,S. H. 1997, Integrating genetic approaches into the discovery ofanticancer drugs.,Science278(5340), 1064–1068.

Hinds, P. W., Dowdy, S. F., Eaton, E. N., Arnold, A. & Weinberg,R. A. 1994, Function of a human cyclin gene as an oncogene.,Proc Natl Acad Sci U S A91(2), 709–713.

Hopkins, A. L. 2007, Network pharmacology.,Nat Biotechnol25(10), 1110–1111.

Howell, V. M., Haven, C. J., Kahnoski, K., Khoo, S. K., Petillo,D., Chen, J., Fleuren, G. J., Robinson, B. G., Delbridge, L. W.,Philips, J., Nelson, A. E., Krause, U., Hammje, K., Dralle, H.,Hoang-Vu, C., Gimm, O., Marsh, D. J., Morreau, H. & Teh, B. T.2003, HRPT2 mutations are associated with malignancy in spo-radic parathyroid tumours.,J Med Genet40(9), 657–663.

Jeronimo, C., Forget, D., Bouchard, A., Li, Q., Chua, G., Poitras, C.,Therien, C., Bergeron, D., Bourassa, S., Greenblatt, J., Chabot,B., Poirier, G. G., Hughes, T. R., Blanchette, M., Price, D. H. &Coulombe, B. 2007, Systematic analysis of the protein interactionnetwork for the human transcription machinery reveals the identityof the 7SK capping enzyme.,Mol Cell 27(2), 262–274.

Jordan, A., Hadfield, J. A., Lawrence, N. J. & McGown, A. T. 1998,Tubulin as a target for anticancer drugs: agents which interact withthe mitotic spindle.,Med Res Rev18(4), 259–296.

Jorgensen, E. M. & Mango, S. E. 2002, The art and design of geneticscreens: caenorhabditis elegans.,Nat Rev Genet3(5), 356–369.

Kaelin, W. G. 2005, The concept of synthetic lethality in thecontextof anticancer therapy.,Nat Rev Cancer5(9), 689–698.

Kamb, A. 2003, Mutation load, functional overlap, and syntheticlethality in the evolution and treatment of cancer.,J Theor Biol223(2), 205–213.

Kiyohara, C. & Yoshimasu, K. 2007, Genetic polymorphisms in thenucleotide excision repair pathway and lung cancer risk: a meta-analysis.,Int J Med Sci4(2), 59–71.

Knudsen, K. A. & Wheelock, M. J. 1992, Plakoglobin, or an 83-kDhomologue distinct from beta-catenin, interacts with E-cadherinand N-cadherin.,J Cell Biol 118(3), 671–679.

Kolligs, F. T., Kolligs, B., Hajra, K. M., Hu, G., Tani, M., Cho, K. R.& Fearon, E. R. 2000, Gamma-catenin is regulated by the APCtumor suppressor and its oncogenic activity is distinct fromthat ofbeta-catenin.,Genes Dev14(11), 1319–1331.

Krogan, N. J., Keogh, M.-C., Datta, N., Sawa, C., Ryan, O. W.,Ding,H., Haw, R. A., Pootoolal, J., Tong, A., Canadien, V., Richards,D. P., Wu, X., Emili, A., Hughes, T. R., Buratowski, S. & Green-blatt, J. F. 2003, A Snf2 family ATPase complex required for re-cruitment of the histone H2A variant Htz1.,Mol Cell 12(6), 1565–1576.

Labhart, P. & Morgan, G. T. 1998, Identification of novel genes encod-ing transcription elongation factor TFIIS (TCEA) in vertebrates:conservation of three distinct TFIIS isoforms in frog, mouse,andhuman.,Genomics52(3), 278–288.

Lehner, B., Crombie, C., Tischler, J., Fortunato, A. & Fraser, A. G.2006, Systematic mapping of genetic interactions in Caenorhabdi-tis elegans identifies common modifiers of diverse signaling path-ways.,Nat Genet38(8), 896–903.

Lin, L., Zhang, J.-H., Panicker, L. M. & Simonds, W. F. 2008, Theparafibromin tumor suppressor protein inhibits cell proliferationby repression of the c-myc proto-oncogene.,Proc Natl Acad Sci US A105(45), 17420–17425.

Litman, R., Gupta, R., Brosh, R. M. & Cantor, S. B. 2008, BRCA-FApathway as a target for anti-tumor drugs.,Anticancer Agents MedChem8(4), 426–430.

Loeb, L. A. 2001, A mutator phenotype in cancer.,Cancer Res61(8), 3230–3239.

Loeb, L. A., Bielas, J. H. & Beckman, R. A. 2008, Cancers ex-hibit a mutator phenotype: clinical implications.,Cancer Res68(10), 3551–7; discussion 3557.

Lolli, G. & Johnson, L. N. 2005, CAK-Cyclin-dependent ActivatingKinase: a key kinase in cell cycle control and a target for drugs?,Cell Cycle4(4), 572–577.

Massague, J. 2004, G1 cell-cycle control and cancer.,Nature432(7015), 298–306.

Measday, V., Baetz, K., Guzzo, J., Yuen, K., Kwok, T., Sheikh,B., Ding, H., Ueta, R., Hoac, T., Cheng, B., Pot, I., Tong,A., Yamaguchi-Iwai, Y., Boone, C., Hieter, P. & Andrews, B.2005, Systematic yeast synthetic lethal and synthetic dosage lethalscreens identify genes required for chromosome segregation., ProcNatl Acad Sci U S A102(39), 13956–13961.

Merlo, L. M. F., Pepper, J. W., Reid, B. J. & Maley, C. C. 2006,Cancer as an evolutionary and ecological process.,Nat Rev Cancer6(12), 924–935.

Mestres, J., Gregori-Puigjane, E., Valverde, S. & Sole, R. V. 2008,Data completeness–the achilles heel of drug-target networks.,NatBiotechnol26(9), 983–984.

Meur, N. L. & Gentleman, R. 2008, Modeling synthetic lethality.,Genome Biol9(9), R135.

Ooi, S. L., Pan, X., Peyser, B. D., Ye, P., Meluh, P. B., Yuan, D. S.,Irizarry, R. A., Bader, J. S., Spencer, F. A. & Boeke, J. D. 2006,Global synthetic-lethality analysis and yeast functionalprofiling.,Trends Genet22(1), 56–63.

Paladugu, S. R., Zhao, S., Ray, A. & Raval, A. 2008, Mining proteinnetworks for synthetic genetic interactions.,BMC Bioinformatics9, 426.

Pan, X., Ye, P., Yuan, D. S., Wang, X., Bader, J. S. & Boeke, J. D.2006, A DNA integrity network in the yeast Saccharomyces cere-visiae.,Cell 124(5), 1069–1081.

Pan, X., Yuan, D. S., Xiang, D., Wang, X., Sookhai-Mahadeo, S.,Bader, J. S., Hieter, P., Spencer, F. & Boeke, J. D. 2004, A ro-bust toolkit for functional profiling of the yeast genome.,Mol Cell16(3), 487–496.

Pelish, H. E., Peterson, J. R., Salvarezza, S. B., Rodriguez-Boulan,E., Chen, J.-L., Stamnes, M., Macia, E., Feng, Y., Shair, M. D.& Kirchhausen, T. 2006, Secramine inhibits cdc42-dependentfunctions in cells and cdc42 activation in vitro.,Nat Chem Biol2(1), 39–46.

Pereira, D. A. & Williams, J. A. 2007, Origin and evolution of high

13

throughput screening.,Br J Pharmacol152(1), 53–61.Platzer, A., Perco, P., Lukas, A. & Mayer, B. 2007, Characteriza-

tion of protein-interaction networks in tumors.,BMC Bioinformat-ics 8, 224.

Pulgar, T. G. D., Valdes-Mora, F., Bandres, E., Perez-Palacios, R., Es-pina, C., Cejas, P., Garcıa-Cabezas, M. A., Nistal, M., Casado,E., Gonzalez-Baron, M., Garcıa-Foncillas, J. & Lacal, J. C.2008, Cdc42 is highly expressed in colorectal adenocarcinoma anddownregulates ID4 through an epigenetic mechanism.,Int J Oncol33(1), 185–193.

Rognan, D. 2007, Chemogenomic approaches to rational drug design.,Br J Pharmacol152(1), 38–52.

Sanchez, G., Delattre, O., Auboeuf, D. & Dutertre, M. 2008, Cou-pled alteration of transcription and splicing by a single oncogene:boosting the effect on cyclin D1 activity.,Cell Cycle7(15), 2299–2305.

Schafer, M. & Werner, S. 2008, Oxidative stress in normal and im-paired wound repair.,Pharmacol Res58(2), 165–171.

Shannon, P., Markiel, A., Ozier, O., Baliga, N. S., Wang, J. T., Ram-age, D., Amin, N., Schwikowski, B. & Ideker, T. 2003, Cytoscape:a software environment for integrated models of biomolecular in-teraction networks.,Genome Res13(11), 2498–2504.

Sinha, S. & Yang, W. 2008, Cellular signaling for activationof RhoGTPase Cdc42.,Cell Signal20(11), 1927–1934.

Sole, R. V. & Deisboeck, T. S. 2004, An error catastrophe in cancer?,J Theor Biol228(1), 47–54.

Sonnhammer, E. L. L. & Koonin, E. V. 2002, Orthology, paralogyand proposed classification for paralog subtypes.,Trends Genet18(12), 619–620.

Stark, C., Breitkreutz, B.-J., Reguly, T., Boucher, L., Breitkreutz, A.& Tyers, M. 2006, BioGRID: a general repository for interactiondatasets.,Nucleic Acids Res34(Database issue), D535–D539.

Stefansson, B. & Brautigan, D. L. 2007, Protein phosphatasePP6 Nterminal domain restricts G1 to S phase progression in human can-cer cells.,Cell Cycle6(11), 1386–1392.

Tarailo, M., Kitagawa, R. & Rose, A. M. 2007, Suppressors of spindlecheckpoint defect (such) mutants identify new mdf-1/mad1 inter-actors in caenorhabditis elegans.,Genetics175(4), 1665–1679.

Tarailo, M., Tarailo, S. & Rose, A. M. 2007, Synthetic lethalinter-actions identify phenotypic ”interologs” of the spindle assemblycheckpoint components.,Genetics177(4), 2525–2530.

Tatusov, R. L., Fedorova, N. D., Jackson, J. D., Jacobs, A. R.,Kiryutin, B., Koonin, E. V., Krylov, D. M., Mazumder, R., Mekhe-dov, S. L., Nikolskaya, A. N., Rao, B. S., Smirnov, S., Sverdlov,A. V., Vasudevan, S., Wolf, Y. I., Yin, J. J. & Natale, D. A. 2003,The COG database: an updated version includes eukaryotes.,BMCBioinformatics4, 41.

The Gene Ontology Consortium 2000, Gene Ontology: tool for theunification of biology,Nature Genetics25, 25–29.

Tischler, J., Lehner, B. & Fraser, A. G. 2008, Evolutionary plasticityof genetic interaction networks.,Nat Genet40(4), 390–391.

Tong, A. H., Evangelista, M., Parsons, A. B., Xu, H., Bader, G. D.,Page, N., Robinson, M., Raghibizadeh, S., Hogue, C. W., Bussey,H., Andrews, B., Tyers, M. & Boone, C. 2001, Systematic geneticanalysis with ordered arrays of yeast deletion mutants.,Science294(5550), 2364–2368.

Tong, A. H. Y., Lesage, G., Bader, G. D., Ding, H., Xu, H., Xin,X.,Young, J., Berriz, G. F., Brost, R. L., Chang, M., Chen, Y., Cheng,X., Chua, G., Friesen, H., Goldberg, D. S., Haynes, J., Humphries,C., He, G., Hussein, S., Ke, L., Krogan, N., Li, Z., Levinson,J. N., Lu, H., Menard, P., Munyana, C., Parsons, A. B., Ryan,O., Tonikian, R., Roberts, T., Sdicu, A.-M., Shapiro, J., Sheikh, B.,Suter, B., Wong, S. L., Zhang, L. V., Zhu, H., Burd, C. G., Munro,S., Sander, C., Rine, J., Greenblatt, J., Peter, M., Bretscher, A.,Bell, G., Roth, F. P., Brown, G. W., Andrews, B., Bussey, H. &

Boone, C. 2004, Global mapping of the yeast genetic interactionnetwork.,Science303(5659), 808–813.

Weinberg, R. 2007,The biology of Cancer, Garland. New York.Whitehurst, A. W., Bodemann, B. O., Cardenas, J., Ferguson, D.,

Girard, L., Peyton, M., Minna, J. D., Michnoff, C., Hao, W.,Roth, M. G., Xie, X.-J. & White, M. A. 2007, Synthetic lethalscreen identification of chemosensitizer loci in cancer cells., Na-ture 446(7137), 815–819.

Williams, R. S., Williams, J. S. & Tainer, J. A. 2007, Mre11-Rad50-Nbs1 is a keystone complex connecting DNA repair machin-ery, double-strand break signaling, and the chromatin template.,Biochem Cell Biol85(4), 509–520.

Wishart, D. S., Knox, C., Guo, A. C., Cheng, D., Shrivastava,S.,Tzur, D., Gautam, B. & Hassanali, M. 2008, Drugbank: a knowl-edgebase for drugs, drug actions and drug targets.,Nucleic AcidsRes36(Database issue), D901–D906.

Yildirim, M. A., Goh, K.-I., Cusick, M. E., Barabasi, A.-L. & Vi-dal, M. 2007, Drug-target network.,Nat Biotechnol25(10), 1119–1126.

Yuen, K. W. Y., Warren, C. D., Chen, O., Kwok, T., Hieter, P. &Spencer, F. A. 2007, Systematic genome instability screens inyeastand their potential relevance to cancer.,Proc Natl Acad Sci U S A104(10), 3925–3930.

14