Identification of Common Gene Networks Responsive To

Home , ABCD4, ABLIM1, CD226, DIAPH2, Gene nomenclature, NRCAM

ORIGINAL ARTICLE Identiﬁcation of common gene networks responsive to radiotherapy in human cancer cells

D-L Hou1, L Chen2, B Liu1, L-N Song1 and T Fang1

Identification of the genes that are differentially expressed between radiosensitive and radioresistant cancers by global gene analysis may help to elucidate the mechanisms underlying tumor radioresistance and improve the efficacy of radiotherapy. An integrated analysis was conducted using publicly available GEO datasets to detect differentially expressed genes (DEGs) between cancer cells exhibiting radioresistance and cancer cells exhibiting radiosensitivity. Gene Ontology (GO) enrichment analyses, Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis and protein–protein interaction (PPI) networks analysis were also performed. Five GEO datasets including 16 samples of radiosensitive cancers and radioresistant cancers were obtained. A total of 688 DEGs across these studies were identified, of which 374 were upregulated and 314 were downregulated in radioresistant cancer cell. The most significantly enriched GO terms were regulation of transcription, DNA-dependent (GO: 0006355, P = 7.00E-09) for biological processes, while those for molecular functions was protein binding (GO: 0005515, P = 1.01E-28), and those for cellular component was cytoplasm (GO: 0005737, P = 2.81E-26). The most significantly enriched pathway in our KEGG analysis was Pathways in cancer (P = 4.20E-07). PPI network analysis showed that IFIH1 (Degree = 33) was selected as the most significant hub protein. This integrated analysis may help to predict responses to radiotherapy and may also provide insights into the development of individualized therapies and novel therapeutic targets.

Cancer Gene Therapy (2014) 21, 542–548; doi:10.1038/cgt.2014.62; published online 21 November 2014

INTRODUCTION and analytic methods were totally different from that of our study. Radiotherapy is a major modality of cancer therapy, but certain The objective of this study was to identify common genes asso- evidence reveals that radiotherapy (as monotherapy or in com- ciated with radioresistance and relevant biological processes bination with other types of treatments) is effective in 52% of by systematic integration of gene expression data from differ- cancer patients, despite of more accurate tumor localization by ent microarray platforms which is capable of increasing the computed tomography and better radiotherapy techniques.1 The statistical power to understand the mechanism of cancer cell major reason for such low effectiveness of radiotherapy may be radioresistance. due to intrinsic tumor cell radioresistance.2 Radioresistance of tumor cells is a multifactorial characteristic, mainly depending on the repair capacity of radiation-induced DNA lesions3 and other MATERIALS AND METHODS factors including hypoxia,4 differential gene expression,5 growth Identification of eligible radioresistance gene expression datasets factor receptors, mutations in proto-oncogenes and tumor sup- Radioresistance expression profiling studies were identified by searching pressor genes,6 expression of receptor tyrosine kinases7 and the Gene Expression Omnibus database (GEO, http://www.ncbi.nlm.nih. adhesion of cells to extracellular matrix molecules.8 Although such gov/geo).15 The following key words and their combinations were used: discoveries have significantly advanced the understanding of ‘radioresistance, cancer, gene expression, microarray’. Those datasets were molecular mechanisms responsible for radiosensitivity of tumor obtained from original experimental articles, which make a comparison cells which are essential for developing the personalized therapy between radiosensitive and radioresistant cancer cells on the gene expression profiling. Non-human studies, review articles and integrated analysis and improving the patient prognosis, the entire process remains of expression profiles were excluded. to be uncovered. Hence, it is urgently needed to discover new insights into the radioresistance mechanisms. As a high-throughput technology, microarray analysis allows Data preprocessing the simultaneous analysis of thousands of genes at the transcript Normalization is important for an accurate comparison of microarray expression level in a single experiment. The microarray have been datasets from multiple platforms. The heterogeneity of different microarray datasets is caused by different platforms, different gene nomenclature and widely used to detect the gene expression difference between fi radiosensitive and radioresistant tumors9 to assess genes involved different cancer type, so it is dif cult to compare the microarray datasets directly. Inappropriate normalization may lead to skewing comparison in radioresistance in a number of cancer cell types, including cervical, 10–13 results, and would reduce the credibility of measurements of changes in pancreatic, oral, lung, esophageal, head and neck cancers. the expression of individual genes. In this study, the expression values for 14 Kim et al. performed an integration of four radiosensitivity each gene were transformed to the z-score for global normalization. We profiling data applied to NCI-60, of which the selected datasets investigated the global shifts of gene expression between radiosensitive

1Department of Radiation Oncology, Beijing Shijitan Hospital, Capital Medical University, Beijing, China and 2Department of Oncology, Teng Zhou Central People’s Hospital of Shandong Province, Tengzhou, China. Correspondence: Dr T Fang, Department of Radiation Oncology, Beijing Shijitan Hospital, Capital Medical University, Beijing 100038, China. E-mail: [email protected] Received 3 September 2014; revised 18 October 2014; accepted 23 October 2014; published online 21 November 2014 Radiotherapy-responsive gene networks in cancer D-L Hou et al 543 and radioresistant cancer cell by using the assembled expression chemistry pathways.18 In this work, the KEGG database was applied to compendium. Z-scores were calculated for each probe as following: investigate the enrichment analysis of the DEGs to find the biochemistry - pathways which might be involved in the tumor radioresistance. ¼ xi x Z score δ PPI network construction where xi indicates raw intensity data for each gene; x indicates the average intensity of gene within a single experiment and δ indicates standard The protein–protein interactions (PPIs) research is of pivotal role to reveal deviation (s.d.) of all measured intensities. the functions of proteins at the molecular level and discover the rules underlying cellular activities including growth, development, metabolism, 19 fi Statistical analysis differentiation and apoptosis. The identi cation of protein interactions in a genome-wide scale is important to interpret the cellular control mech- fi The Signi cance Analysis of Microarray software was then used to identify anisms.20 In this study, we attempted to construct a PPI network based on differentially expressed genes (DEGs) between radiosensitive and radio- Biological General Repository for Interaction Datasets (BioGRID) (http:// resistant cancer cell by carrying out gene-specific t-tests, with a ‘relative thebiogrid.org/) and the data in the constructed PPI network on the top 10 difference’ score for each gene. The definition of D value was the average up- and downregulated DEGs were visualized with Cytoscape software gene expression change from different expression levels to the standard 21 deviation of measurements for that gene. Genes with a fold-change 41.5 (http://cytoscape.org/). and false discovery rate ⩽ 0.05 were considered as DEGs.16 The heat map of the top 50 DEGs was constructed by R statistical software. RESULTS Functional classification of DEGs Short overview of the studies included The biological functions of the DEGs were interpreted by Gene Ontology A total of five expression profiling studies met inclusion criteria 17 (GO) enrichment analysis using the web-based software GENECODIS. GO and were included. Across these studies, 16 samples of radio- provides a common descriptive framework and functional annotation and sensitive and radioresistant cancer cell was obtained. The general classification for analyzing gene sets. GO categories are organized into three groups: biological process, cellular component and molecular func- information of these studies is detailed in Table 1. In these, studies tion. Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway database on several types of cancer were included, such as head and neck is a recognized and comprehensive database including all kinds of bio- cancer, nasopharyngeal carcinoma and lung cancer.

Table 1. Characteristics of the individual studies

GEO ID Platform Tumor type Samples (sensitive: Country Time resistant)

GSE9712 GPL96 [HG-U133A] Affymetrix Human Genome U133A Array Head and neck cancer 3 : 3 USA 2007 GSE9713 GPL96 [HG-U133A] Affymetrix Human Genome U133A Array Head and neck cancer 3 : 3 USA 2007 GSE9714 GPL96 [HG-U133A] Affymetrix Human Genome U133A Array Head and neck cancer 2 : 2 USA 2007 GSE48503 GPL570 [HG-U133_Plus_2] Affymetrix Human Nasopharyngeal carcinoma 2 : 2 China 2014 Genome U133 Plus 2.0 Array GSE20549 GPL6244 [HuGene-1_0-st] Affymetrix Human Lung cancer 6 : 6 Korea 2010 Gene 1.0 ST Array [transcript (gene) version] Abbreviation: GEO, Gene Expression Omnibus database.

Table 2. The top 10 most signiﬁcantly up- or downregulated DEGs

Gene ID Gene symbol Ofﬁcial full name P-value Fold change

Upregulated genes 151188 ARL6IP6 ADP-ribosylation-like factor 6 interacting protein 6 1.37E-07 2.0233 7673 ZNF222 Ring finger protein 182 1.64E-06 2.3815 84925 DIRC2 GATS, stromal antigen 3 opposite strand 5.28E-06 1.8515 221687 RNF182 OCIA domain containing 1 1.11E-05 6.6735 84314 TMEM107 Interferon induced with helicase C domain 1 1.28E-05 1.5662 79986 ZNF702P Zinc finger protein 222 1.77E-05 2.6197 352954 GATS Zinc finger protein 702 pseudogene 1.99E-05 2.0265 64135 IFIH1 Transmembrane protein 107 2.51E-05 2.0085 91975 ZNF300 Disrupted in renal carcinoma 2 3.16E-05 7.0214 54940 OCIAD1 Zinc finger protein 300 3.52E-05 2.0728

Downregulated genes 8797 TNFRSF10A TRAF3 interacting protein 2 2.25E-06 − 1.7028 13 AADAC Arylacetamide deacetylase (esterase) 3.91E-06 − 5.2005 29126 CD274 ATP-binding cassette, sub-family B (MDR/TAP), member 10 1.23E-05 − 3.963 57532 NUFIP2 CD274 molecule 1.32E-05 − 1.9252 10758 TRAF3IP2 Serpin peptidase inhibitor, clade B (ovalbumin), member 2 1.55E-05 − 2.7456 9615 GDA Nuclear fragile X mental retardation protein interacting protein 2 2.06E-05 − 10.19 23456 ABCB10 Calcium/calmodulin-dependent protein kinase kinase 1, alpha 2.23E-05 − 2.0969 84254 CAMKK1 Tumor necrosis factor receptor superfamily, member 10a 2.41E-05 − 3.4501 5055 SERPINB2 Arrestin domain containing 4 2.54E-05 − 10.079 91947 ARRDC4 Guanine deaminase 2.70E-05 − 3.2741

Figure 1. Heat map visualization of the patterns of expression change for the top 50 most signiﬁcantly DEGs across different datasets.

DEGs associated with tumor radioresistance Presented in Table 4 are pathways involving the identified DEGs A total of 688 DEGs were found according to the cutoff criteria of that were statistically significant when a threshold P-value of less fold-change 41.5 and false discovery rate ⩽ 0.05. Among those than 0.05 was used in the hypergeometric test. The most signi- DEGs, 374 were upregulated and 314 genes were downregulated ficant pathways in our KEGG analysis were pathways in cancer in radioresistant cancer cell. A list of the top 10 most significantly (P = 4.20E-07). In addition, focal adhesion (P = 1.59E-06) and cell up- or downregulated genes is presented in Table 2. The expres- adhesion molecules (CAMs) (P = 1.82E-06) are also highly enriched sion pattern of expression change for top 50 most significantly (Table 4, Figure 2b). DEGs is displayed in Figure 1. The upregulated gene with the lowest P-value was ARL6IP6, PPI network whose function is currently unknown. TNFRSF10A, a member of The constructed PPI network consisted of 77 nodes and 76 edges. the TNF-receptor superfamily, which transduces cell death signal In PPI networks, degrees are defined to measure how many and induces cell apoptosis by activating tumor necrosis factor- neighbors a node directly connect to, and the nodes with a high related apoptosis inducing ligand (TNFSF10/TRAIL), was the degree are defined as hub protein, and as illustrated in Figure 3, downregulated gene with the lowest P-value. The full list of these the significant hub proteins contained IFIH1 (Degree = 33), GDA genes is provided in the Supplementary Table 1. (Degree = 27) and TNFRSF10A (Degree = 25).

Functional annotation of DEGs Genes with a nominal significance level of Po0.01 were selected DISCUSSION and tested against the background set of all genes with GO Radiotherapy is the primary treatment modality for cancer, but annotations. Regulation of transcription, DNA-dependent (GO: radioresistance on account of repeated treatments remains a 0006355, P = 7.00E-09) and multicellular organismal development serious obstacle to a satisfactory effect in many cases. To identify (GO: 0007275, P = 1.24E-08) were significantly enriched for bio- genes involved in radioresistance, we compared the expression logical processes (Figure 2a). The other radioresistance-related profiles of 16 samples of radiosensitive and radioresistant cancer functions that were enriched included cell adhesion, signal trans- cell by integrating five microarray datasets. We also performed duction, cell junction assembly and response to drug. While for functional annotation including GO enrichment analysis, KEGG molecular functions, protein binding (GO: 0005515, P = 1.01E-28) pathway analysis and PPI networks analysis to highlight the and metal ion binding (GO: 0046872, P = 2.15E-20) were signifi- biological role for the DEGs. cantly enriched, for cellular component, cytoplasm (GO: 0005737, A total of 688 genes were found to show altered expression in P = 2.81E-26) and nucleus (GO: 0005634, P = 9.16E-22) were signifi- radiosensitive versus radioresistant cancer cell (374 upregulated cantly enriched (Table 3). and 314 downregulated). The upregulated gene with the lowest

Figure 2. The signiﬁcantly enriched functional annotation of differentially expressed genes. (a) The top 10 enriched GO categories for biological process; (b) The top 10 enriched KEGG pathways.

P-value was ARL6IP6, whose function has been unknown. important role of TNFRSF10A in radioresistance, which may be TNFRSF10A, which induces cell apoptosis upon binding to TRAIL, meaningful for further targeted therapy for cancer. was the downregulated gene with the lowest P-value. TRAIL may Common points have been noted between radioresistance and be a promising candidate for cancer molecular-targeted agents chemoresistance, which involves defects in apoptosis signaling because of its remarkable feature of selectively inducing apoptosis or an increased ability to repair DNA.23 Hence, we searched the in tumors without causing toxicity to normal cells. In accordance reports of tumor radioresistance or chemoresistance referred to with previous studies, a recent study detected epigenetic silence the top 10 most signiﬁcantly up- or downregulated DEGs. In of TNFRSF10A in radioresistant laryngeal cancer, which lead to docetaxel-resistant castration-resistant prostate cancer cell lines, resistance to TRAIL-induced apoptosis.22 This ﬁnding suggests an IFIH1 showed noteworthy down-expression by whole-genome

Table 3. The top 10 enriched GO categories of DEGs

GO ID GO term No. of genes FDR

Biological process GO:0006355 Regulation of transcription, DNA-dependent 76 7.00E-09 GO:0007275 Multicellular organismal development 53 1.24E-08 GO:0007155 Cell adhesion 39 1.50E-08 GO:0007165 Signal transduction 55 2.85E-06 GO:0034329 Cell junction assembly 12 9.09E-05 GO:0042493 Response to drug 21 0.0002045 GO:0045892 Negative regulation of transcription, DNA-dependent 25 0.0002083 GO:0030154 Cell differentiation 29 0.0002314 GO:0007010 Cytoskeleton organization 12 0.0002733 GO:0030168 Platelet activation 18 0.0002735

Cellular component GO:0005737 Cytoplasm 221 2.81E-26 GO:0005634 Nucleus 213 9.16E-22 GO:0005886 Plasma membrane 161 1.52E-21 GO:0005622 Intracellular 96 3.00E-14 GO:0016021 Integral to membrane 152 2.99E-10 GO:0005887 Integral to plasma membrane 53 1.36E-08 GO:0005576 Extracellular region 80 1.53E-08 GO:0016020 Membrane 136 2.94E-08 GO:0005829 Cytosol 84 9.51E-08 GO:0031012 Extracellular matrix 17 1.89E-07

Molecular function GO:0005515 Protein binding 204 1.01E-28 GO:0046872 Metal ion binding 139 2.15E-20 GO:0008270 Zinc ion binding 104 6.46E-18 GO:0003677 DNA binding 86 6.24E-12 GO:0005509 Calcium ion binding 44 5.89E-10 GO:0016787 Hydrolase activity 50 1.43E-07 GO:0003700 Sequence-speciﬁc DNA binding transcription factor activity 41 9.04E-05 GO:0008081 Phosphoric diester hydrolase activity 6 0.0005 GO:0003779 Actin binding 19 0.000528 GO:0000166 Nucleotide binding 72 0.0005455 Abbreviations: FDR, false discovery rate; GO, Gene Ontology.

Table 4. The top 15 enriched KEGG pathway of DEGs

KEGG ID KEGG term No. of genes FDR Gene list

hsa05200 Pathways in cancer 26 4.20E-07 FGF5,IL6,RARB,ITGA6,COL4A6,FZD6,EPAS1,PIK3CA,COL4A2, ITGA3,AR,BIRC3,CDKN1B,FGF13,EGF,PML,MECOM,LAMA1, WNT5A,PTK2,CCNA1,CDKN2A,TP53,TRAF1,EGLN2,MMP1 hsa04510 Focal adhesion 19 1.59E-06 SHC4,ITGA6,COL4A6,PIK3CA,COL4A2,PDGFD,ITGA3,THBS1,BIRC3, PXN,FLNC,COL5A2,EGF,LAMA1,VASP,PTK2,PAK3,FYN,ITGB8 has04514 Cell adhesion molecules (CAMs) 15 1.82E-06 ITGA6,PTPRM,PVRL2,ALCAM,CD226,MPZ,NCAM1, NRCAM,CDH2,NCAM2,PVRL3,ITGB8,CD274,VCAN,PTPRF hsa05222 Small cell lung cancer 12 4.65E-06 RARB,ITGA6,COL4A6,PIK3CA,COL4A2,ITGA3, BIRC3,CDKN1B,LAMA1,PTK2,TP53,TRAF1 hsa04810 Regulation of actin cytoskeleton 16 1.79E-04 FGF5,ITGA6,PIK3CA,F2R,PDGFD,ITGA3,PXN,GSN,FGF13, DIAPH2,EGF,IQGAP2,PTK2,PAK3,ITGB8,CYFIP2 hsa02010 ABC transporters 6 4.96E-03 ABCC1,ABCD2,ABCC9,ABCG2,ABCC2,ABCB10 hsa05146 Amoebiasis 9 5.05E-03 IL6,PLCB1,COL4A6,PIK3CA,COL4A2,SERPINB2,COL5A2,LAMA1,PTK2 hsa05162 Measles 10 5.48E-03 IL6,STAT2,TNFRSF10C,PIK3CA,TNFAIP3,CDKN1B, FYN,TP53,TNFRSF10A,IFIH1 hsa04512 ECM-receptor interaction 8 5.86E-03 ITGA6,COL4A6,COL4A2,ITGA3,THBS1,COL5A2,LAMA1,ITGB8 hsa05218 Melanoma 7 6.95E-03 FGF5,PIK3CA,PDGFD,FGF13,EGF,CDKN2A,TP53 hsa05100 Bacterial invasion of epithelial cells 7 6.95E-03 SHC4,HCLS1,PIK3CA,ELMO1,PXN,MAD2L2,PTK2 hsa00980 Metabolism of xenobiotics 7 6.98E-03 GSTM3,AKR1C3,EPHX1,MGST1,GSTM1,AKR1C1,ALDH3A1 by cytochrome P450 hsa05020 Prion diseases 5 7.00E-03 IL6,NOTCH1,NCAM1,FYN,NCAM2 hsa04360 Axon guidance 9 1.33E-02 SEMA3C,SEMA5A,DPYSL2,ABLIM1,PLXNA2,PTK2,PAK3,FYN,EPHA4 hsa04910 Insulin signaling pathway 9 1.44E-02 SHC4,PIK3CA,PDE3A,SOCS4,PPP1R3B,PRKCI,PPP1R3C,PTPRF,IRS2 Abbreviations: FDR, false discovery rate; KEGG, Kyoto Encyclopedia of Genes and Genomes.

Figure 3. The constructed protein–protein interaction networks of the top 10 up- and downregulated DEGs. Nodes represent proteins, edges represent interactions between two proteins. Red- and green-color nodes represent products of up- and downregulated DEGs, respectively. Blue nodes denote products of genes predicted to interact with the DEGs. arrays and real-time quantitative reverse transcriptase PCR KEGG pathway enrichment analysis was performed to further validation compared with docetaxel-sensitive castration-resistant evaluate the pathways involved in tumor radioresistance. Path- prostate cancer cell lines,24 but there is no related reports about ways in cancer, focal adhesion and cell adhesion molecules its role in tumor radioresistance. ABCB10 and other 12 members of (CAMs) are found to be highly enriched. Many DEGs were involved ATP-binding cassette (ABC) including ABCA3, ABCB1 (MDR1), in pathways in cancer, including FGF5, IL6, RARB, ITGA6, COL4A6, ABCB6, ABCB8, ABCB11, ABCC1 (MRP1), ABCC4, ABCC9, ABCD3, FZD6 and so on, functioning in proliferation and anti-apoptosis. ABCD4, ABCE1 and ABCF2 transporter, were amplified among 19 Cellular adhesion involved in radioresistance is proposed to of the 23 chemoresistant cell lines examined by subtractive produce anti-apoptotic signals when integrin-mediated adhesion comparative genomic hybridization. ABCC9 and ABCC1 were interacts with the extracellular matrix.28 One integrative meta- detected in our study.25 Huang et al.26 determined that SERPINB2, analysis of four microarray datasets between definitely radio- a known inhibitor of extracellular serine proteinase urokinase- sensitive and radioresistant NCI-60 cancer cells also determined type plasminogen activator, was significantly downregulated in the important role of cellular adhesion in tumor radioresistance.14 cisplatin-resistant head and neck squamous cell carcinoma The present study has two limitations. First, heterogeneity subclones compared with their isogenic drug-sensitive parental arising from various cancer types may have distorted the analysis. lines, suggesting its key role in chemoresistance in head and neck We combined five microarray datasets on response to radio- squamous cell carcinoma cell lines. Zhang et al.27 found that GDA, therapy of cancer cell lines to identify common radioresistant which was downregulated by DNA methylation, induced cisplatin genes regardless of cancer type. Defining common radioresistant resistance in non-small cell lung cancer by high-throughput mechanisms not affected by cancer type is meaningful, but the microarrays. Moreover, the important role of IFIH1, GDA and actual cellular response in biological validation might differ among TNFRSF10A in radioresistance could be identified in the PPI cancer types. Second, although we conducted global normal- analysis. ization for different data sets, the heterogeneity from different GO categories enrichment analysis was performed to under- microarray platforms in each study cannot be removed. stand the biological function of the DEGs. For biological processes, the significantly enriched GO term was transcription, DNA- dependent, while that for molecular functions was protein binding, CONCLUSIONS and for cellular component was cytoplasm. Some of the enriched In conclusion, this integrative analysis of five microarray datasets GO terms may be radioresistance-related including cell adhesion, on response to radiotherapy of cancer cell lines showed the signal transduction, cell junction assembly and response to drug. underlying molecular differences between radiosensitive and

© 2014 Nature America, Inc. Cancer Gene Therapy (2014), 542 – 548 Radiotherapy-responsive gene networks in cancer D-L Hou et al 548 radioresistant cancer cells, and identified DEGs and biological 14 Kim HS, Kim SC, Kim SJ, Park CH, Jeung HC, Kim YB et al. Identification of a function to develop effective targeted therapies and understand radiosensitivity signature using integrative metaanalysis of published microarray common radioresistance mechanisms. Further functional studies data for NCI-60 cancer cells. BMC Genomics 2012; 13: 348. may provide additional insights into the role of the differentially 15 Barrett T, Wilhite SE, Ledoux P, Evangelista C, Kim IF, Tomashevsky M et al. NCBI regulated genes in the tumor radioresistance. GEO: archive for functional genomics data sets—update. Nucleic Acids Res 2013; 41(D1): D991–D995. 16 Tusher VG, Tibshirani R, Chu G. Significance analysis of microarrays applied to the – CONFLICT OF INTEREST ionizing radiation response. Proc Natl Acad Sci USA 2001; 98: 5116 5121. 17 Tabas-Madrid D, Nogales-Cadenas R, Pascual-Montano A. GeneCodis3: a non- fl The authors declare no con ict of interest. redundant and modular enrichment analysis tool for functional genomics. Nucleic Acids Res 2012; 40(Web Server issue): W478–W483. REFERENCES 18 Altermann E, Klaenhammer TR. PathwayVoyager: pathway mapping using the Kyoto Encyclopedia of Genes and Genomes (KEGG) database. BMC Genomics 1 Delaney G, Jacob S, Featherstone C, Barton M. The role of radiotherapy in cancer 2005; 6: 60. treatment: estimating optimal utilization from a review of evidence-based clinical 19 Giot L, Bader JS, Brouwer C, Chaudhuri A, Kuang B, Li Y et al. A protein interaction guidelines. Cancer 2005; 104: 1129–1137. map of Drosophila melanogaster. Science 2003; 302: 1727–1736. 2 Kristensen CA, Kjaer-Kristoffersen F, Sapru W, Berthelsen AK, Loft A, Specht L. 20 Li S, Armstrong CM, Bertin N, Ge H, Milstein S, Boxem M et al. A map of the Nasopharyngeal carcinoma. Treatment planning with IMRT and 3D conformal interactome network of the metazoan C. elegans. Science 2004; 303: 540–543. radiotherapy. Acta Oncol 2007; 46:214–220. 21 Shannon P, Markiel A, Ozier O, Baliga NS, Wang JT, Ramage D et al. Cytoscape: a 3 Jeggo P, Lobrich M. Radiation-induced DNA damage responses. Radiat Prot software environment for integrated models of biomolecular interaction net- Dosimetry 2006; 122:124–127. – 4 Vaupel P. Tumor microenvironmental physiology and its implications for radiation works. Genome Res 2003; 13: 2498 2504. oncology. Semin Radiat Oncol 2004; 14: 198–206. 22 Lee JC, Lee WH, Min YJ, Cha HJ, Han MW, Chang HW et al. Development of TRAIL 5 Ogawa K, Murayama S, Mori M. Predicting the tumor response to radiotherapy resistance by radiation-induced hypermethylation of DR4 CpG island in recurrent using microarray analysis (Review). Oncol Rep 2007; 18: 1243–1248. laryngeal squamous cell carcinoma. Int J Radiat Oncol Biol Phys 2014; 88: – 6 Hanahan D, Weinberg RA. The hallmarks of cancer. Cell 2000; 100:57–70. 1203 1211. 7 Rodemann HP, Dittmann K, Toulany M. Radiation-induced EGFR-signaling and 23 Bergman PJ, Harris D. Radioresistance, chemoresistance, and apoptosis resistance. – control of DNA-damage repair. Int J Radiat Biol 2007; 83:781–791. The past, present, and future. Vet Clin North Am Small Anim Pract 1997; 27:47 57. 8 Hehlgans S, Cordes N. Caveolin-1: an essential modulator of cancer cell radio-and 24 Marin-Aguilera M, Codony-Servat J, Kalko SG, Fernandez PL, Bermudo R, Buxo E fi chemoresistance. Am J Cancer Res 2011; 1: 521–530. et al. Identi cation of docetaxel resistance genes in castration-resistant – 9 Lee NH, Saeed AI. Microarrays: an overview. Methods Mol Biol 2007; 353:265–300. prostate cancer. Mol Cancer Ther 2012; 11: 329 339. 10 Wong YF, Sahota DS, Cheung TH, Lo KW, Yim SF, Chung TK et al. Gene expression 25 Yasui K, Mihara S, Zhao C, Okamoto H, Saito-Ohara F, Tomida A et al. Alteration in pattern associated with radiotherapy sensitivity in cervical cancer. Cancer J 2006; copy numbers of genes as a mechanism for acquired drug resistance. Cancer Res 12: 189–193. 2004; 64: 1403–1410. 11 Ogawa K, Utsunomiya T, Mimori K, Tanaka F, Haraguchi N, Inoue H et al. 26 Huang Z, Li H, Huang Q, Chen D, Han J, Wang L et al. SERPINB2 down-regulation Differential gene expression profiles of radioresistant pancreatic cancer cell lines contributes to chemoresistance in head and neck cancer. Mol Carcinog 2013; 53: established by fractionated irradiation. Int J Oncol 2006; 28: 705–713. 777–786. 12 Higo M, Uzawa K, Kouzu Y, Bukawa H, Nimura Y, Seki N et al. Identification of 27 Zhang YW, Zheng Y, Wang JZ, Lu XX, Wang Z, Chen LB et al. Integrated analysis of candidate radioresistant genes in human squamous cell carcinoma cells through DNA methylation and mRNA expression profiling reveals candidate genes asso- gene expression analysis using DNA microarrays. Oncol Rep 2005; 14: 1293–1298. ciated with cisplatin resistance in non-small cell lung cancer. Epigenetics 2014; 9: 13 Guo WF, Lin RX, Huang J, Zhou Z, Yang J, Guo GZ et al. Identification of differ- 896–909. entially expressed genes contributing to radioresistance in lung cancer cells using 28 Sandfort V, Koch U, Cordes N. Cell adhesion-mediated radioresistance revisited. microarray analysis. Radiat Res 2005; 164:27–35. Int J Radiat Biol 2007; 83:727–732.

Supplementary Information accompanies the paper on Cancer Gene Therapy website (http://www.nature.com/cgt)