A Comparative and Integrative Approach Identifies Atpase Family, AAA Domain Containing 2 As a Likely Driver of Cell Proliferation in Lung Adenocarcinoma

Published OnlineFirst August 22, 2012; DOI: 10.1158/1078-0432.CCR-12-0505

Clinical Cancer Human Cancer Biology Research

A Comparative and Integrative Approach Identiﬁes ATPase Family, AAA Domain Containing 2 as a Likely Driver of Cell Proliferation in Lung Adenocarcinoma

Robert Fouret1, Julien Laffaire2, Paul Hofman6, Michele Beau-Faller7, Julien Mazieres8, Pierre Validire3, Philippe Girard3, Sophie Camilleri-Broet€ 4, Fabien Vaylet9, Francois¸ Leroy-Ladurie10, Jean-Charles Soria11,13, and Pierre Fouret5,12

Abstract Purpose: To identify genetic changes that could drive cancer pathogenesis in never and ever smokers with lung adenocarcinoma. Experimental Design: We analyzed the copy number and gene expression profiles of lung adenocarcinomas in 165 patients and related the alterations to smoking status. Having found differences in the tumor profiles, we integrated copy number and gene expression data from 80 paired samples. Results: Amplifications at 8q24.12 overlapping MYC and ATAD2 were more frequent in ever smokers. Unsupervised analysis of gene expression revealed two groups: in the group with mainly never smokers, the tumors expressed genes common to normal lung; in the group with more ever smokers, the tumors expressed "proliferative" and "invasive" gene clusters. Integration of copy number and gene expression data identified one module enriched in mitotic genes and MYC targets. Its main associated modulator was ATAD2, a cofactor of MYC. A strong dose–response relationship between ATAD2 and proliferation-related gene expression was noted in both never and ever smokers, which was verified in two independent cohorts. Both ATAD2 and MYC expression correlated with 8q24.12 amplification and were higher in ever smokers. However, only ATAD2, and not MYC, overexpression explained the behavior of proliferation-related genes and predicted a worse prognosis independently of disease stage in a large validation cohort. Conclusions: The likely driving force behind MYC contribution to uncontrolled cell proliferation in lung adenocarcinoma is ATAD2. Deregulation of ATAD2 is mainly related to gene amplification and is more frequent in ever smokers. Clin Cancer Res; 18(20); 1–11. 2012 AACR.

Introduction smokers accounted in France for 17% cancer deaths in The majority of lung cancers are caused by tobacco women and 4% in men (2). smoking. However, even in people who have never smoked, For 4 major genes involved in the pathogenesis of lung lung cancer would rank as the seventh most common cause cancers, ALK, EGFR, KRAS, and TP53, striking differences in of cancer death worldwide (1). In 2000, lung cancer in never the molecular alterations of these genes have been found in lung cancers in never and ever smokers (3, 4). Molecular alterations include translocations for ALK or point mutations for EGFR, KRAS, and TP53 (5, 6). In addition, copy Authors' Affiliations: 1DCom, Tel ecom ParisTech; 2Programme Carte d'Identite des Tumeurs, Ligue Nationale Contre le Cancer; 3Institut Mutua- number changes contribute through associated gene dereg- liste Montsouris; 4Hopital^ Europeen George Pompidou; 5Universite Pierre ulation to the malignant phenotype. For instance, MYC is et Marie Curie, Paris; 6CHU Nice, Nice; 7CHU Strasbourg, Strasbourg; frequently amplified and overexpressed in lung cancer (7). 8CHU Toulouse, Toulouse; 9Hopital^ d'instruction des armees Percy, Cla- mart; 10Centre Chirurgical Marie-Lannelongue, Le Plessis-Robinson; No study has reported definitive associations between 11Institut Gustave-Roussy; 12INSERM Gen etique des tumeurs, Villejuif, amplifications or deletions and smoking status (8). and 13Université Paris XI, Le Kremlin-Bicêtre, France We analyzed the copy number and gene expression Note: Supplementary data for this article are available at Clinical Cancer profiles of lung adenocarcinomas and related the alterations Research Online (http://clincancerres.aacrjournals.org/). to smoking status. Having found differences in the tumor R. Fouret and J. Laffaire are first co-authors. profiles, we integrated copy number and gene expression data to identify genetic changes that could drive cancer Corresponding Author: Pierre Fouret, INSERM Gen etique des tumeurs U985, Institut Gustave-Roussy, 114 rue E. Vaillant, 94805 Villejuif Cedex, pathogenesis. The present study differed from previous France. Phone: 33-0-1-42-177782; Fax: 33-0-1-42-177777; E-mail: studies on 2 aspects. First, the number of tumors from never [email protected] smokers was greater in our study than in previous studies (8, doi: 10.1158/1078-0432.CCR-12-0505 9). Second, to control for potential bias, the ever smoker 2012 American Association for Cancer Research. group was constructed by matching ever smokers to never

www.aacrjournals.org OF1

Fouret et al.

Gene expression analysis Translational Relevance Total RNAs were hybridized to Affymetrix Human Our results suggest that the aberrant expression of Genome U133 Plus 2.0 GeneChip (Affymetrix). Unsuper- MYC targets that participate in the program responsible vised hierarchical clustering analysis of tumor samples from for uncontrolled proliferation may be attributed to the LG cohort and normal lung samples from 11 female ATAD2-deregulated expression. This further suggests Asian never smokers (accession number: GSE 19804) was that ATAD2 levels may predict the MYC dependency of conducted on the most variant probe sets. Differences lung adenocarcinoma, which should be exploited for between sample clusters were tested using the c2 test. Hyper- therapeutic purposes. While MYC has been considered geometric enrichment for Gene Ontology sets (GeneOntol- as a frequent and very relevant therapeutic target in lung ogy.org) and MYC target genes (www.myccancergene.org) cancer, specific inhibition of MYC has not been achieved were calculated with false discovery rate (FDR) correction of and no MYC inhibitor is currently in the clinic. ATAD2 P values. Literature Vector Analysis (LitVAn) was used to is worthwhile to investigate as a therapeutic target, infer gene cluster functionality with an evaluation of the which appears feasible given its ATPase activity and its significance of their scores (litvan.bio.columbia.edu). bromodomain. Both genomic and gene expression data were deposited in ArrayExpress database (accession number: E-MTAB-923). ATAD2 relative expression was measured by real-time reverse transcriptase PCR (RT-PCR) using the Hs00204205 smokers, such that the whole cohort was enriched in never TaqMan probe (Applied Biosystems). smokers and the group of ever smokers had clinical characteristics (sex, disease stage) identical to never smokers. Integration of copy number and gene expression data We used Copy Number and Expression In Cancer (CON- Materials and Methods EXIC) to integrate matched copy number (amplifications or Detailed information on patients, samples, and methods deletions) and gene expression data from 80 paired sample used in copy number, gene expression, and survival anal- (13). yses is available as Supplementary Materials and Methods. As described by Akavia and colleagues, CONEXIC is based on the following assumptions: (i) a driver mutation Patients and samples in a "modulator" gene should be associated (correlated) All 165 study patients were treated by surgery for lung with a group of genes that form a "module" and (ii) copy adenocarcinoma without prior chemotherapy. Fifty-eight number aberrations often influence the expression of genes patients received cisplatin-based adjuvant chemotherapy. in the module via changes in expression of the modulator. Never smoking status was defined by a lifetime exposure of The CONEXIC learning algorithm consists of 3 key steps: less than 100 cigarettes. The tumors were classified according to the tumor–node–metastasis (TNM) system in use at 1. Selection of candidate genes that are recurrently the time of diagnosis (10). The pathologic diagnoses were amplified or deleted in tumors. reviewed according to current histologic classification for 2. Single Modulator step that creates an initial lung carcinoma (11, 12). Cases for which a doubt about the association between expression of candidate drivers primary site in the lung remained were excluded. All ade- and expression of genes modules. nocarcinomas were invasive. A bronchiolo-alveolar compo- 3. An iterative Network Learning step to improve the nent was recorded when a noninvasive lepidic growth was initial model. seen adjacent to a component of invasive adenocarcinoma. This study was part of the Lung Genes (LG) project, which During the Single Modulator and the Network learning was approved by the Institut National du Cancer review steps, the search is driven by the optimization of a Bayesian board (Programme National d’Excellence Specialise Pou- scoring function similar to Module Network (14). For each mon). Informed consent was obtained from patients for the node, the driver–split combination that achieves the highest use of their lung surgical samples. score is selected as long as it is verified to be statistically Only cases with an average of tumor cells equal to or significant. Significance is tested using Lee and colleagues above 50% were included. Genomic DNA and RNA were permutation test (15); up to 3 top-scoring modulator genes extracted and assessed for integrity and quantity following are tried, and if none of them pass the permutations stringent quality control criteria (cit.ligue-cancer.net). significance test, no more splits are added to the driver tree. In addition to significance testing, nonparametric bootstrap Genomic DNA analysis serves to eliminate spurious correlations. Genomic DNAs were hybridized on Illumina SNP The output is a driver network that divides the expressed HumanCNV370 chips (Illumina). The GISTIC version genes into modules and associates each module with a 2.0 algorithm (www.broadinstitute.org/cancer/pub/GIS- driver tree. Each node in the tree is associated with a driver TIC2) was used to identify significant regions of amplifica- gene (modulator gene) and a threshold expression level tion or deletion. The frequencies of aberrations contribut- (split value) and divides the expression values of the mod- ing to significant peak regions were compared using c2 tests. ule’s members into samples in which the modulator’s

OF2 Clin Cancer Res; 18(20) October 15, 2012 Clinical Cancer Research

ATAD2 Drives Cell Proliferation in Lung Adenocarcinoma

expression is below the threshold and those in which the multivariate proportional hazard Cox overall survival anal- modulator’s expression is above the threshold. Each side of ysis, ATAD2 expression was studied together with age, sex, the split at the first root of the tree (herein designated as the and disease stage. first-order split) can contain further splits (secondary splits) using other modulator/expression threshold pairs. A detailed description of the selection of candidate genes Results and of the specified parameters for the Single Modulator Different frequency of aberrations in significant peak and the Network Learning steps as well as a discussion of the regions between never and ever smokers significance levels under which modulators were identified A total of 121 high-quality genomic profiles were are available as Supplementary Information (Supplemen- obtained. Frequent aberrations (frequency >25%) included tary Materials and Methods). gains on 1p, 1q, 5p, 5q, 6p, 7p, 7q, 8q, 14q, 16p, 17q, 20p, The modules and their modulators were visualized using and 20q and losses on 1p, 3p, 4q, 5q, 6q, 8p, 9p, 10q, 12p, Genatomy (www.c2b2.columbia.edu/danapeerlab/html/ 13q, 15q, 17p, and 18q. The GISTIC 2.0 algorithm was genatomy.html). applied to identify regions that were significantly amplified To nominate the module associated with smoking status, or deleted (Fig. 1). A total of 59 significant peak regions with we used gene set enrichment analysis (GSEA; ref. 16). a frequency of 13% to 84% were identified, including 22 For validation, we used the publicly available gene expres- regions that were amplified and 37 regions that were deleted. sion data from 68 lung adenocarcinomas (accession num- The frequency of amplifications or deletions in the 59 ber: GSE 12667; ref. 17) and from 391 lung adenocarcino- significant regions was compared between never and ever mas (caarraydb.nci.nih.gov, pId ¼ 1015945236141280; smokers. After adjustment for multiple comparisons using ref. 18). The linear relationship between a modulator and Bonferroni method, only 2 regions were differentially its associated genes was measured using the Pearson corre- amplified or deleted according to smoking status: amplifi- lation coefficient. cations were more frequent in ever smokers (83%) compared with never smokers (52%) at 8q24.12 (q-value ¼ Survival analysis 0.02), whereas deletions were more frequent in ever smo- The univariate overall survival analyses were conducted kers (50%) compared with never smokers (13%) at 4q35.2 using the Kaplan–Meier method and log-rank tests. In the (q-value ¼ 0.0006).

Figure 1. GISTIC 2.0 analysis of copy number changes in 121 lung adenocarcinomas. Plots of the G- scores (top) and q-values (bottom) with respect to ampliﬁcations (A) or deletions (B) over the entire region analyzed. The signiﬁcance level for the q-value is indicated by a vertical dotted line. Chromosome positions are indicated along the y-axis with centromere positions indicated by horizontal dotted lines. The locations of the peak regions are indicated on the right of each panel.

www.aacrjournals.org Clin Cancer Res; 18(20) October 15, 2012 OF3

Fouret et al.

Figure 2. Unsupervised analysis of gene expression in 103 lung adenocarcinomas from the LG cohort and 11 normal lungs from Asian female never smokers. In the heat map, each cell represents the expression value for a probe in a sample. The largest expression values are in red, and the lowest expression values are in green. The sample clusters shown at the top are colored in red or blue for tumor samples and in green for normal lung samples, wherein the blue-colored tumor samples and the green-colored normal lung samples are clustered together. Each box below the sample clusters represents the value for a discrete clinicopathologic annotation in a sample. A black box denotes presence of a bronchiolo-alveolar component, ever-smoker status, male sex, EGFR mutation, or KRAS mutation. The P values associated with annotations are obtained by comparing the 2 groups of tumor samples using c2 tests. The gene clusters shown on the left of the heat map are labeled A to J.

Two groups of tumors with distinctive gene ative" cluster. Typical genes in the proliferative cluster expression clusters and different clinicopathologic encoded cyclins (CCNA2, CCNB1, and CCNE2), the annotations cyclin-dependent kinase CDK6, E2F transcription factors Unsupervised hierarchical clustering of gene expression (E2F7, E2F8), and 12 proteins involved in mitosis. The in 103 high-quality tumor profiles and 11 normal lung cluster i was enriched for the GO term "extra-cellular samples (GSE19804) revealed 2 groups of tumors (Fig. 2). matrix" (7.0E-14). Typical genes in cluster i encoded mem- The partition was stable as assessed by resampling. The first bers of the disintegrin and metalloproteinase (ADAM12, group of tumors was characterized by the expression of a ADAMDEC1, ADAMTS5) or matrix metalloproteinase gene cluster (cluster c) that was common to normal lung (MMP1, MMP3, MMP11, MMP12, MMP13) families. samples and mainly absent from tumors in the second Using LitVAn, significant terms associated with the pro- group. Two genes clusters (cluster f and cluster i) were liferative cluster were "cyclin" and "mitotic" as well as overexpressed in the second group of tumors, whereas they "spindle," reflecting the enrichment for genes participating were both expressed at low levels in most tumors of the first to the mitotic spindle (BUB1, CENPF, KIF14, KIF15, group and in normal lung samples. NDC80, NEK2, NUF2, SKA1, SPC25, TPX2, TTK; genome. The cluster f was enriched for Gene Ontology (GO) terms ucsc.edu). LitVAn significant terms for cluster i included "cell cycle process" (q-value ¼ 1.7E-17) and "mitotic cell "invasion," favoring its designation as the "invasive" gene cycle" (q-value ¼ 1.8E-11) and designated as the "prolifer- cluster (Supplementary Table S1).

OF4 Clin Cancer Res; 18(20) October 15, 2012 Clinical Cancer Research

ATAD2 Drives Cell Proliferation in Lung Adenocarcinoma

The first group of tumors whose gene expression resem- Table 1. Association of gene expression with bled normal lung comprised 35 never smokers (74%) and smoking status–related copy number 12 ever smokers, whereas the second group comprised 28 never smokers (50%) and 28 ever smokers (P ¼ 0.01). In the alterations in the LG cohort first group, the tumors more frequently presented with a bronchiolo-alveolar component (P ¼ 5E-6) or harbored an Aberration Cytoband Gene name Welch t test (P)a ¼ EGF receptor (EGFR) mutation (P 0.0002), whereas in the Amplification 8q24.12 ATAD2 1.1E-04 second group, they more often harbored a KRAS mutation Amplification 8q24.12 DERL1 1.5E-04 ¼ (P 0.02). Amplification 8q24.12 DSCC1 2.2E-05 Amplification 8q24.12 FAM83A 5.4E-08 ATAD2 as a likely driver of cell proliferation Amplification 8q24.12 FAM91A1 1.5E-09 Eight-hundred and eighteen genes overlapped significant Amplification 8q24.12 KIAA0196 1.3E-05 aberrations less than a third of chromosome length, includ- Amplification 8q24.12 MRPL13 1.6E-05 ing 350 genes overlapping 19 amplifications and 468 genes Amplification 8q24.12 MYC 8.0E-05 overlapping 34 deletions. Among these genes, 109 genes Amplification 8q24.12 NDUFB9 2.3E-04 overlapped the 2 regions that were differentially altered Amplification 8q24.12 NSMCE2 0.002 between never and ever smokers. The expression of 175 Amplification 8q24.12 RNF139 0.01 genes, including 35 that overlapped the 8q24.12 and Amplification 8q24.12 SQLE 1.9E-04 4q35.2 smoking status–related aberrations (Table 1), was Amplification 8q24.12 TATDN1 9.8E-05 < significantly altered (P 0.05) by either their amplification Amplification 8q24.12 TMEM65 8.1E-06 status or deletion status. Deletion 4q35.2 ACSL1 5.5E-06 Using CONEXIC, we found a model comprising 67 Deletion 4q35.2 ANKRD37 0.002 modules that were associated with 31 main modulators Deletion 4q35.2 CCDC111 5.2E-04 (i.e., likely drivers at the first-order split of the regulatory Deletion 4q35.2 CDKN2AIP 2.2E-05 programs) explaining the behavior of 10,001 genes. Deletion 4q35.2 CYP4V2 4.6E-04 We wished to uncover the likely drivers of differentially Deletion 4q35.2 DCTD 8.0E-09 expressed clusters, which were revealed by the unsupervised Deletion 4q35.2 F11 0.02 analysis of gene expression in the whole cohort. We had 2 Deletion 4q35.2 FRG1 4.0E-04 collections of gene sets, one provided by the unsupervised Deletion 4q35.2 GALNT7 0.009 analysis in 103 patients (the a to j gene clusters), the other by Deletion 4q35.2 GPM6A 0.03 CONEXIC in 78 patients with paired genomic and gene Deletion 4q35.2 HMGB2 0.047 expression data (the 67 modules). Both collections were Deletion 4q35.2 HPGD 3.6E-07 determined without the help of clinical or biologic annota- Deletion 4q35.2 ING2 0.04 tions. We thus conducted hypergeometric enrichment anal- Deletion 4q35.2 IRF2 1.1E-04 ysis to identify which of the gene sets overlapped in the 2 Deletion 4q35.2 NEIL3 0.009 datasets and thus indicate likely drivers of the overlapping Deletion 4q35.2 RWDD4A 7.5E-08 clusters in the whole cohort. The proliferative cluster (clus- Deletion 4q35.2 SNX25 1.7E-05 ter f in the unsupervised analysis) intersected very signifi- Deletion 4q35.2 SORBS2 0.01 ¼ cantly (q-value 2.0E–37) with CONEXIC module 62 (Fig. Deletion 4q35.2 STOX2 0.04 3A). Twenty-eight of 46 genes of cluster f were identified as Deletion 4q35.2 TLR3 1.6E-06 module 62 genes. The proliferative cluster did not intersect Deletion 4q35.2 UFSP2 3.5E-04 with any other CONEXIC modules. Module 62 genes were a enriched in the GO term "cell cycle process" (q-value ¼ Comparing amplification versus normal or deletion versus 1.2E-87). The main modulator associated with module 62 normal. was ATPase family, AAA domain containing 2 (ATAD2), a gene located at 8q24.12. (Supplementary Fig. S1A). When ATAD2 was low, however, A linear relationship between the expression of ATAD2 the second-order regulatory programs depending on and the expression of proliferation-related genes in TUBB3 did not classify samples in the Ding cohort as well both never and ever smokers as in the LG cohort, suggesting that this secondary regulator To verify the association between ATAD2 expression and was not optimally chosen by CONEXIC. We replaced genes in its module, the module 62 regulatory programs TUBB3 expression by ATAD2 expression, which improved identified in the LG cohort were applied to gene expression the classification of samples in the Ding cohort (Supple- data from 68 lung adenocarcinomas of the Ding cohort mentary Fig. S1B) without altering substantially the original (17). The profiles of module 62 genes were compared using module 62 profiles in the LG cohort (Supplementary Fig. identical split expression values for the modulators. The S1C). These results suggested that ATAD2 expression alone relationship between high ATAD2 levels and overexpres- could explain the behavior of module 62 genes across sion of module 62 genes was verified in the Ding cohort datasets.

www.aacrjournals.org Clin Cancer Res; 18(20) October 15, 2012 OF5

Fouret et al.

Figure 3. Integrated analysis of paired copy number and gene expression data in 80 lung adenocarcinomas belonging to the LG cohort. A, CONEXIC analysis. Genatomy module network view of CONEXIC module 62. Each row of the heat map corresponds to the expression of a gene across the 80 samples. Gene names indicated at the right of the heat map are sorted by alphabetical order. Sample names are indicated above the heat map. Samples are ordered according to the regulatory programs found by CONEXIC and shown above sample names. The modulators include ATAD2, TUBB3, SLC25A21,andKCNMB4,whereinATAD2 increased expression at the first-order split and, at the right, second-order split is associated with increased expression of genes in the module. Yellow dotted lines partition the samples according to split values of the modulators. B to D, analysis of gene expression of the proliferative cluster (cluster f) in the LG cohort according to ATAD2 expression and according to smoking status, 8q24.12 amplification, or MYC expression. Proliferative cluster genes are those included in the final set of genes after processing of LG gene expression data during the CONEXIC procedure. Gene names indicated at the right of the heat map are sorted by alphabetical order. B, samples are sorted in ever (blue) or never smokers (white) and then sorted within each smoking status category into 4 groups of increasing ATAD2 expression levels (from whitetored).The4ATAD2 groupsaresortedusingthesplitvaluesfoundbyCONEXICintheanalysisoftheLGcohort.C,samplesaresorted according to 8q24.12 amplification, comparing amplification (blue) versus no amplification (white), then within each category into 4 groups of increasing ATAD2 expression levels as above. D, samples are sorted according to increasing MYC expression levels using quartile values as split values (from light to dark blue), then within each category into 4 groups of increasing ATAD2 expression levels as above.

To enable the conﬁrmation of a linear relationship 391 patient (18). There was a strong dose–response rela- between ATAD2 and the proliferative cluster (cluster f), tionship between ATAD2, and the proliferative cluster in Pearson correlation coefﬁcients were calculated in the LG every cohort as the expression of genes belonging to the cohort, the Ding cohort, and a third independent cohort of proliferative cluster increased with higher ATAD2 levels in

OF6 Clin Cancer Res; 18(20) October 15, 2012 Clinical Cancer Research

ATAD2 Drives Cell Proliferation in Lung Adenocarcinoma

Table 2. Correlation of ATAD2 or MYC expression with the expression of genes of the proliferative cluster in the LG, Ding, and Shedden cohorts

ATAD2 MYC

Pearson Pearson correlation correlation Cohort variable coefficient P coefficient P LG All 0.85 <0.0001 0.39 0.0003 Smoking status Never smoker 0.83 <0.0001 0.28 0.06 Ever smoker 0.87 <0.0001 0.36 0.04 Sex Female 0.86 <0.0001 0.37 0.002 Male 0.80 0.003 0.45 0.16 Disease stage Early (I or II) 0.87 <0.0001 0.33 0.01 Late (III) 0.79 <0.0001 0.58 0.004 Bronchiolo-alveolar component Yes 0.77 <0.0001 0.28 0.09 No 0.86 <0.0001 0.40 0.008 EGFR Mutated 0.77 <0.0001 0.49 0.001 Wild-type 0.88 <0.0001 0.26 0.11 KRAS Mutated 0.89 <0.0001 0.67 0.009 Wild-type 0.85 <0.0001 0.26 0.04 8q24.12 amplification Yes 0.82 <0.0001 0.37 0.006 No 0.86 <0.0001 0.02 0.92 Ding All 0.77 <0.0001 0.28 0.02 Smoking status Never smoker 0.7 0.05 0.1 0.81 Ever smoker 0.78 <0.0001 0.25 0.12 Unknown 0.67 0.001 0.29 0.21 Shedden All 0.75 <0.0001 0.28 <0.0001 Smoking status Never smoker 0.7 <0.0001 0.1 0.54 Ever smoker 0.76 <0.0001 0.26 <0.0001 Unknown 0.72 <0.0001 0.2 0.06

the LG (correlation coefficient 0.85), Ding (correlation in every subgroup defined by sex, disease stage, bronchiolo- coefficient 0.77), and Shedden (correlation coefficient alveolar component, EGFR,orKRAS status (Table 2 and 0.75) cohorts. Supplementary Fig. S2). Although high ATAD2 was less In the LG cohort, the expression of the proliferative frequent among tumors without 8q24.12 amplification, cluster increased with higher ATAD2 levels in never (cor- ATAD2 was differentially expressed and strongly correlated relation coefficient 0.83) and in ever smokers (correlation with expression of the proliferative cluster in tumors with or coefficient 0.87; Fig. 3B). A similarly strong linear relation- without the amplification (Fig. 3C and Table 2). ship was noted in both the Ding and the Shedden cohorts for never and ever smokers and patients with unknown ATAD2 and module 62 relationships with 8q14.12 smoking status (Table 2). amplification and smoking status In the LG cohort, the expression of ATAD2 strongly On the basis of the array data, ATAD2 expression was correlated with the expression of the proliferative cluster associated with 8q24.12 amplification status (P ¼ 1.1E-

www.aacrjournals.org Clin Cancer Res; 18(20) October 15, 2012 OF7

Fouret et al.

1.00.80.60.40.20.0 1.00.80.60.40.20.0 ATAD2 low ATAD2 low Figure 4. Kaplan–Meier curves of overall survival rates according to ATAD2 levels in the LG cohort, ATAD2 high comparing high versus low ATAD2 levels. Shown is the log-rank P Survival rate ATAD2 high Survival rate value. A, LG cohort. B, Shedden cohort.

P value 0.14 P value 0.002

0 1,000 2,000 3,000 4,000 0 102030405060 Time (d) Time (mo)

4; Table 1) and was higher in ever smokers (P ¼ 0.0004). It spindle genes correlated weakly with MYC, whereas they was neither associated with CDKN2A nor RB1 overlapping correlated strongly with ATAD2 (Supplementary Table S3). deletions. Overexpression of the proliferative cluster occurred in In a subset of 76 patients with available RNA for real-time tumors with low MYC and it correlated with ATAD2 RT-PCR analysis, ATAD2 expression was increased in ever (Fig. 3D). The modest correlation of the proliferative cluster smokers compared with never smokers (fold change ¼ 1.75, with MYC was verified in both Ding and Shedden cohorts P ¼ 0.02) and in patients with 8q24.12 amplification (Table 2). compared with no amplification (fold change ¼ 2.27, P ¼ 0.001). Survival of patients As the number of ever smokers was higher in the group of In the 78 patients from the LG cohort with paired copy tumors expressing the proliferative cluster f, we tested the number and gene expression array data, the survival rates association of smoking status with each CONEXIC module were 0.75 [95% confidence interval (CI), 0.65-0.88] at 3 using GSEA. The profiles in ever smokers as compared with years and 0.61 (95% CI, 0.48–0.77) at 4 years for the low never smokers were significantly enriched in the module 62 ATAD2 group and 0.59 at 3 years (95% CI, 0.43–0.83) and gene set to which the highest enrichment score was given 0.44 (95% CI, 0.28–0.71) at 4 years for the high ATAD2 (0.86 for the enrichment score, 1.67 for the normalized group (Fig. 4A). Survival was not significantly different enrichment score, 0.02 for the nominal P, and 0.22 for the according to ATAD2 expression (P ¼ 0.14). Late disease FDR). stage was associated with a shorter survival (P ¼ 0.01). None of the other clinical or biologic variables, including 8q24.12 Relationships between ATAD2, MYC, and proliferation- amplification (P ¼ 0.58) and MYC expression (P ¼ 0.77), related genes was associated with survival. Module 62 genes were enriched (q-value ¼ 3.2E-4) for In the 75 patients who were studied using PCR to measure genes of the MYC target database (www.myccancergene. ATAD2 expression, neither the PCR data (P ¼ 0.41) nor the org). Three CONEXIC modules other than module 62 were ATAD2 array data (P ¼ 0.43) were associated with survival. associated with ATAD2 as their main modulator, one of In the 349 patients from the Shedden cohort, the survival which was also enriched for MYC targets (q-value ¼ 0.005). rates were 0.74 (95% CI, 0.68–0.81) at 3 years and 0.65 Enrichments for the proliferative cluster genes, GO terms (95% CI, 0.58–0.73) at 4 years for the low ATAD2 group and containing "cell cycle process" and MYC targets were 0.58 (95% CI, 0.51–0.66) at 3 years and 0.5 (95% CI, 0.43– aligned only for module 62 (Supplementary Table S2). 0.58) at 4 years for the high ATAD2 group (Fig. 4B). Survival Amplifications of 8q24.12 included both ATAD2 and time was longer in the low ATAD2 group than in the high MYC in every sample save one. Like ATAD2, MYC expres- ATAD2 group (P ¼ 0.002). Late disease stage was strongly sion was associated with 8q24.12 amplification (P ¼ 8.0E- associated with shorter survival (P ¼ 1E-16). 5; Table 1) and was higher in ever smokers than in never Multivariate proportional hazard Cox models were tested smokers (P ¼ 0.002). to investigate the association of ATAD2 array data with The correlation of the proliferative cluster with MYC survival and to adjust for age, sex, and disease stage in the (correlation coefficient 0.39) was less strong, however, than 78 patients from the LG cohort and in the 349 patients from with ATAD2 (correlation coefficient 0.85). Remarkably, the the Shedden cohort. In the LG cohort, the best model correlation of MYC targets in the proliferative cluster was (likelihood P ¼ 0.008) included ATAD2, age, and disease less strong with MYC (correlation coefficient 0.33) than stage. An older age (HR, 2.18; 95% CI, 1.04–4.55; P ¼ 0.04) with ATAD2 (correlation coefficient 0.83). Likewise, mitotic and late disease stage (HR, 2.32; 95% CI, 1.17-4.6; P ¼ 0.02)

OF8 Clin Cancer Res; 18(20) October 15, 2012 Clinical Cancer Research

ATAD2 Drives Cell Proliferation in Lung Adenocarcinoma

were associated with a shorter survival. High ATAD2 was not lapping differentially altered regions explain the differences significantly associated with survival (HR, 2.04; 95% CI, in expression profiles between tumors. 0.98-4.27; P ¼ 0.06). Removing ATAD2 reduced slightly the To identify driving mutations and the processes they model likelihood (likelihood P ¼ 0.02). influence, we integrated copy number and gene expression In the Shedden cohort, the best model (likelihood P ¼ 1E- data using the recently developed algorithm CONEXIC 13) included ATAD2 and stage. High ATAD2 (HR, 1.68; (13). With this approach, the starting list of candidate 95% CI, 1.22-2.32; P ¼ 0.002) and late disease stage (HR, drivers includes only the genes within or near-significant 3.86; 95% CI, 2.74-5.42; P ¼ 8E-15) were significantly regions of copy number changes. As a result, CONEXIC associated with survival. Removing ATAD2 reduced slightly would not detect drivers that are typically associated with the model likelihood (likelihood P ¼ 2E-12). point mutations. We identify ATAD2 as a likely driver whose expression explains the behavior of differentially expressed prolifer- Discussion ation-related genes. Indeed, an ATAD2-associated module Our study reveals that ATAD2 is a likely driver of cell outputted by CONEXIC contained a majority of genes of proliferation in lung adenocarcinoma. ATAD2 overexpres- the proliferative cluster identified in the unsupervised sion both explains the behavior of cell cycle genes and, most analysis of gene expression, an enrichment very unlikely likely, results primarily from amplification, thereby con- caused by chance. A strong dose–response relationship necting proliferation of lung cancer cells to a unique genetic between ATAD2 levels and those of genes belonging to aberration. Furthermore, our results suggest that the onco- the proliferative cluster is shown in the LG cohort and is gene MYC, which is located 4.3 Mb distal to ATAD2 (http:// verified in 2 independent validation cohort (17, 18). The genome.ucsc.edu/), is involved in that pathway as the relationship between ATAD2 and proliferation-related ATAD2 associated proliferative signature includes MYC genes is neither affected by smoking status nor smoking targets involved in cell cycle. Before our study, it was known status–associated characteristics including KRAS or EGFR that ATAD2 is upstream of MYC and that it can exert a role in mutation. the proliferation of normal and cancer cells, strongly sup- ATAD2 is correctly associated by CONEXIC with genes porting our conclusion (19–21). The present study is the that it is known to regulate. ATAD2 has been identified as a first to provide evidence suggesting that amplified ATAD2 is cofactor for MYC-dependent transcription by Ciro and the main driving force behind MYC contribution to uncon- colleague (19). Here, ATAD2-associated genes were signif- trolled cell proliferation in lung adenocarcinoma. The cru- icantly enriched in MYC targets. Kalashnikova and collea- cial driving function shown here for ATAD2 may have gues using chromatin immunoprecipitation assays showed therapeutic implications. While MYC has been considered that ATAD2 occupies the proximal promoter regions of as a frequent and very relevant therapeutic target in lung several key cell-cycle regulators (BUB1, CCNA2, KIF15, cancer, specific inhibition of MYC has not been achieved MCM10, and TOP2A), which we show linearly related to and no MYC inhibitor is currently in the clinic. ATAD2 is ATAD2 expression in lung adenocarcinoma (27). worthwhile to investigate as a therapeutic target, which ATAD2 overlaps the 8q24.12 region which was more appears feasible given its ATPase activity and its bromodo- frequently amplified in ever smokers. ATAD2 expression main (22). Moreover, ATAD2 expression predicts the was associated with 8q24.12 amplification, suggesting that expression of mitotic spindle genes, whose products par- ATAD2 deregulation occurs primarily through copy num- ticipate to a network vulnerable to inhibition of SUMOyla- ber changes. Nevertheless, it was ATAD2 expression and not tion (23–25). 8q24.12 amplification that correlated with the expression of A key factor in uncovering the contrasted phenotypes that proliferation-related genes. Discrepancies between 8q24.12 are summarized by gene clusters in the unsupervised anal- amplification and ATAD2 overexpression point to addition- ysis of gene expression is the comparison of normal lung al genetic or epigenetic events contributing to ATAD2 and tumors, many of which were from never smokers. Lung expression. It has been suggested ATAD2 deregulation may adenocarcinomas in never smokers present typically with a be the consequence of the loss of retinoblastoma (RB)- bronchiolo-alveolar component with well-differentiated mediated control in a subset of highly aggressive breast tumor cells, whereas in ever smokers growth is usually cancer (27). We checked that ATAD2 expression was not "fully invasive," that is, consists exclusively of invasive associated with deletions targeting the RB pathway. component (11, 26). Consistent with histology, gene MYC and ATAD2 are frequently co-amplified in cancers expression in the group where never smokers were numer- (www.broadinstitute.org/tumorscape), a consistent finding ous resembled that of normal lung. In contrast, the group in this cohort. Co-amplification may be selected in tumors with more ever smokers expressed proliferative or invasive as a way to concomitantly overexpress not too far apart gene clusters. cooperating genes. Like ATAD2, MYC was overexpressed in Most of the frequent aberrations shown in this cohort ever smokers, and MYC overexpression was associated with have been previously reported (8). However, the frequen- 8q24.12 amplification. Increased expression of MYC targets cies of aberrations in 2 significant regions of amplification appears necessary to the association of ATAD2 with cell or deletion differ according to smoking status. These results proliferation as other ATAD2-associated modules that raised the question whether the deregulation of genes over- were not enriched in MYC targets were not enriched in

www.aacrjournals.org Clin Cancer Res; 18(20) October 15, 2012 OF9

Fouret et al.

proliferation-related genes. As compared with ATAD2, the program responsible for uncontrolled proliferation may expression of MYC only weakly correlated, however, with be attributed to ATAD2-deregulated expression. This further expression of the proliferative cluster, including MYC target suggests that ATAD2 levels may predict a MYC dependency genes and mitotic spindle genes. These results suggest that of lung adenocarcinoma, which should be exploited as a ATAD2 levels through MYC activity are more important priority target for therapeutic purposes. than MYC levels to drive cell proliferation in lung adenocarcinoma. Assuming that MYC protein levels are roughly Disclosure of Potential Conflicts of Interest equivalent to mRNA levels, it may seem surprising that MYC No potential conflicts of interest were disclosed. activity is not directly related to MYC expression. However, using a novel in vivo model of Myc-induced tumorigenesis, Authors' Contributions Conception and design: J.-C. Soria, P. Fouret Murphy and colleagues reported that low levels of deregu- Development of methodology: R. Fouret, J. Laffaire, P. Fouret lated Myc are competent to drive ectopic proliferation of Acquisition of data (provided animals, acquired and managed patients, somatic cells and lung oncogenesis (28). provided facilities, etc.): P. Hofman, M. Beau-Faller, J. Mazieres, P. Vali- dire, P. Girard, S. Camilleri-Broet,€ F. Vaylet, F. Leroy-Ladurie, P. Fouret High ATAD2 is associated with poor survival of patients Analysis and interpretation of data (e.g., statistical analysis, biosta- with breast cancer (19, 27). Caron and colleagues reported tistics, computational analysis): R. Fouret, J. Laffaire, P. Hofman, M. Beau- Faller, F. Vaylet, P. Fouret that high ATAD2 (E. and C. Brambilla, unpublished data) Writing, review, and/or revision of the manuscript: J. Laffaire, P. Hof- predicts a shorter survival of patients with lung cancer (29). man, J. Mazieres, P. Girard, F. Vaylet, J.-C. Soria, P. Fouret In the LG cohort, ATAD2 is not significantly associated with Administrative, technical, or material support (i.e., reporting or orga- nizing data, constructing databases): M. Beau-Faller, P. Validire survival, although there is a trend when the array data are Study supervision: P. Fouret adjusted for disease stage and age. There is much uncertainty in the results in small groups. In the larger Shedden cohort, Acknowledgments high ATAD2 predicts a shorter survival, which is consistent The following investigators participated in the Lung Genes (LG) project: Centre Chirurgical Marie-Lannelongue, Le Plessis-Robinson: P. Dartevelle, E. with the reported prognostic value of a cluster of cell Dulmet, F. Leroy-Ladurie, and V. de Montpreville; Centre Hospitalier Inter- proliferation–related gene (18). The prognostic value of communal Creteil: I. Monnet; Centre Hospitalo-Universitaire Dijon, Paris: ATAD2 is independent of disease stage in that cohort. A. Bernard and F. Piard; Centre Hospitalo-Universitaire Hotel-Dieu,^ Paris: M. Alifano, S. Camilleri-Bro€et, D. Damotte, and J.F. Regnard; Centre Hospitalo- ATAD2 is a co-activator, which can control MYC-depen- Universitaire Nice: P. Hofman, V. Hofman, and J. Mouroux; Centre Hospi- dent transcription (19, 27). Through MYC and E2F tran- talo-Universitaire Saint-Louis, Paris: J. Tredaniel; Centre Hospitalo-Univer- scription factors, ATAD2 increases the expression of prolif- sitaire Strasbourg: M. Beau-Faller, G. Massard, and A. Neuville; Centre Hospitalo-Universitaire Tenon, Paris: M. Antoine and J. Cadranel; Centre eration-related and anti-apoptotic genes in many different Hospitalo-Universitaire Toulouse: L. Brouchet, J. Mazieres, and I. Rouquette; types of cancer, including hormone-dependent prostate or DCom, Telecom ParisTech, Paris: R. Fouret; Hopital^ d’instruction des armees Percy, Clamart: P. Saint-Blancard and F. Vaylet; Institut Gustave-Roussy, breast carcinomas, estrogen receptor–negative breast carci- Villejuif: A. Berhneim, P. Dessen, F. Dufour, N. Dorvault, P. Fouret, B. Job, L. noma, cervical carcinoma, glioblastoma, osteosarcoma, Lacroix, V. Lazar, C. Richon, V. Roux, P. Saulnier, J.C. Soria, E. Taranchon, S. and non–small cell lung carcinoma (19, 20, 21, 27, 29, Toujani, and A. Valent; Institut Mutualiste Montsouris, Paris: P. Girard, D. Gossot, and P. Validire; and Ligue Nationale Contre le Cancer: J. Laffaire and 30). Although these data strongly support that ATAD2 may A. de Reynes. The authors thank D. Simon (Laboratoire Probabilites et drive cell proliferation in various cancers, more experiments Modeles Aleatoires, Universite Pierre et Marie Curie, Paris, France) for help are needed to investigate the mechanisms by which ATAD2 in conducting the bootstrap. likely influences the biologic consequences of MYC deregulation in the context of lung cancer cells. Grant Support The study was supported by Institut National du Cancer (Programme In summary, ATAD2 is identified by a comparative and National d’Excellence Specialise Poumon); Ligue Nationale Contre le Cancer integrative approach as a likely driver of cell proliferation in (Programme Carte d’Identite des Tumeurs); and Association pour la lung adenocarcinoma. MYC is co-amplified with ATAD2 Recherche sur le Cancer (grant number SFI20101201740) grants to P. Fouret. The costs of publication of this article were defrayed in part by the and, like ATAD2, is overexpressed in ever smokers. How- payment of page charges. This article must therefore be hereby marked ever, it is ATAD2 and not MYC expression that is strongly advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate related to the expression of proliferation-related genes, this fact. especially mitotic spindle genes. These results suggest that Received February 15, 2012; revised August 1, 2012; accepted August 5, the aberrant expression of MYC targets that participate in 2012; published OnlineFirst August 22, 2012.

References 1. Subramanian J, Govindan R. Lung cancer in never smokers: a review. J small-cell lung cancer who harbor EML4-ALK. J Clin Oncol 2009;27: Clin Oncol 2007;25:561–70. 4247–53. 2. International Agency for Research on Cancer. Water, air, soil 5. Soda M, Choi YL, Enomoto M, Takada S, Yamashita Y, Ishikawa S, and food pollutants. In: Attributable causes of cancer in France et al. Identiﬁcation of the transforming EML4-ALK fusion gene in non- in the year 2000. Geneva, Switzerland: WHO Press; 2007. p. 97– small-cell lung cancer. Nature 2007;448:561–6. 102. 6. Herbst RS, Heymach JV, Lippman SM. Lung cancer. N Engl J Med 3. Sun S, Schiller JH, Gazdar AF. Lung cancer in never smokers–a 2008;359:1367–80. different disease. Nat Rev Cancer 2007;7:778–90. 7. Little CD, Nau MM, Carney DN, Gazdar AF, Minna JD. Ampliﬁcation 4. Shaw AT, Yeap BY, Mino-Kenudson M, Digumarthy SR, Costa DB, and expression of the c-myc oncogene in human lung cancer cell lines. Heist RS, et al. Clinical features and outcome of patients with non- Nature 1983;306:194–6.

OF10 Clin Cancer Res; 18(20) October 15, 2012 Clinical Cancer Research

ATAD2 Drives Cell Proliferation in Lung Adenocarcinoma

8. Weir BA, Woo MS, Getz G, Perner S, Ding L, Beroukhim R, et al. 20. Zou JX, Revenko AS, Li LB, Gemo AT, Chen H-W. ANCCA, an Characterizing the cancer genome in lung adenocarcinoma. Nature estrogen-regulated AAA þATPase coactivator for ERalpha, is required 2007;450:893–8. for coregulator occupancy and chromatin modification. Proc Natl Acad 9. Wong MP, Fung L-F, Wang E, Chow W-S, Chiu S-W, Lam W-K, et al. Sci U S A 2007;104:18067–72. Chromosomal aberrations of primary lung adenocarcinomas in non- 21. ZouJX,GuoL,RevenkoAS,TepperCG,GemoAT,KungH-J,etal. smokers. Cancer 2003;97:1263–70. Androgen-induced coactivator ANCCA mediates specificandrogen 10. Hermanek P, Hutter R, Sobin L, Wagner G, Wittekind C, editors. TNM receptor signaling in prostate cancer. Cancer Res 2009;69: Atlas. Guide illustre de la classification TNM/pTNM des tumeurs .3339–46 malignes. 4th ed. Paris, France: Springer-Verlag; 1998. 22. Hsia EY, Goodson ML, Zou JX, Privalsky ML, Chen H-W. Nuclear 11. Travis WD, Brambilla E, Muller-Hermelink€ HK, Harris CC. International receptor coregulators as a new paradigm for therapeutic targeting. Adv Agency for Research on Cancer. Pathology and genetics of tumours of Drug Deliv Rev 2010;62:1227–37. the lung, pleura, thymus and heart. Lyon, France: IARC; 2004. 23. Joukov V, Groen AC, Prokhorova T, Gerson R, White E, Rodriguez 12. Travis WD, Brambilla E, Noguchi M, Nicholson AG, Geisinger KR, A, Walter JC, Livingston DM. The BRCA1/BARD1 heterodimer Yatabe Y, et al. International association for the study of lung cancer/ modulates ran-dependent mitotic spindle assembly. Cell 2006;127: American thoracic society/European respiratory society international 539–52. multidisciplinary classification of lung adenocarcinoma. J Thorac 24. Kiyomitsu T, Obuse C, Yanagida M. Human Blinkin/AF15q14 is required Oncol 2011;6:244–85. for chromosome alignment and the mitotic checkpoint through direct 13. Akavia UD, Litvin O, Kim J, Sanchez-Garcia F, Kotliar D, Causton HC, interaction with Bub1 and BubR1. Dev Cell 2007;13:663–76. et al. An integrated approach to uncover drivers of cancer. Cell 25. Kessler JD, Kahle KT, Sun T, Meerbrey KL, Schlabach MR, Schmitt 2010;143:1005–17. EM, et al. A SUMOylation-dependent transcriptional subprogram is 14. Segal E, Shapira M, Regev A, Pe'er D, Botstein D, Koller D, et al. required for Myc-driven tumorigenesis [Internet]. Science 2012;335: Module networks: identifying regulatory modules and their condition- 348–53. specific regulators from gene expression data. Nat Genet 2003;34: 26. Miller VA, Kris MG, Shah N, Patel J, Azzoli C, Gomez J, et al. Bronch- 166–76. ioloalveolar pathologic subtype and smoking history predict sensitivity 15. Lee S-I, Pe'er D, Dudley AM, Church GM, Koller D. Identifying to gefitinib in advanced non-small-cell lung cancer. J Clin Oncol regulatory mechanisms using individual variation reveals key role 2004;22:1103–9. for chromatin modification. Proc Natl Acad Sci U S A 2006;103: 27. Kalashnikova EV, Revenko AS, Gemo AT, Andrews NP, Tepper CG, 14062–7. Zou JX, et al. ANCCA/ATAD2 overexpression identifies breast cancer 16. Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette patients with poor prognosis, acting to drive proliferation and survival MA, et al. Gene set enrichment analysis: a knowledge-based approach of triple-negative cells through control of B-Myb and EZH2. Cancer for interpreting genome-wide expression profiles. Proc Natl Acad Sci Res 2010;70:9402–2. U S A 2005;102:15545–50. 28. Murphy DJ, Junttila MR, Pouyet L, Karnezis A, Shchors K, Bui DA, et al. 17. Ding L, Getz G, Wheeler DA, Mardis ER, McLellan MD, Cibulskis K, Distinct thresholds govern Myc's biological output in vivo. Cancer Cell et al. Somatic mutations affect key pathways in lung adenocarcinoma. 2008;14:447–57. Nature 2008;455:1069–75. 29. Caron C, Lestrat C, Marsal S, Escoffier E, Curtet S, Virolle V, et al. 18. Shedden K, Taylor JMG, Enkemann SA, Tsao M-S, Yeatman TJ, Functional characterization of ATAD2 as a new cancer/testis factor Gerald WL, et al. Gene expression-based survival prediction in lung and a predictor of poor prognosis in breast and lung cancers. Onco- adenocarcinoma: a multi-site, blinded validation study. Nat Med gene 2010;29:5171–81. 2008;14:822–7. 30. Revenko AS, Kalashnikova EV, Gemo AT, Zou JX, Chen H-W. Chro- 19. Ciro M, Prosperini E, Quarto M, Grazini U, Walfridsson J, McBlane F, matin loading of E2F-MLL complex by cancer-associated coregulator et al. ATAD2 is a novel cofactor for MYC, overexpressed and amplified ANCCA via reading a specific histone mark. Mol Cell Biol 2010;30: in aggressive tumors. Cancer Res 2009;69:8491–8. 5260–72.

www.aacrjournals.org Clin Cancer Res; 18(20) October 15, 2012 OF11

A Comparative and Integrative Approach Identifies ATPase Family, AAA Domain Containing 2 as a Likely Driver of Cell Proliferation in Lung Adenocarcinoma

Robert Fouret, Julien Laffaire, Paul Hofman, et al.

Clin Cancer Res Published OnlineFirst August 22, 2012.

Updated version Access the most recent version of this article at: doi:10.1158/1078-0432.CCR-12-0505

Supplementary Access the most recent supplemental material at: Material http://clincancerres.aacrjournals.org/content/suppl/2012/08/23/1078-0432.CCR-12-0505.DC1

E-mail alerts Sign up to receive free email-alerts related to this article or journal.

Reprints and To order reprints of this article or to subscribe to the journal, contact the AACR Publications Subscriptions Department at [email protected].

Permissions To request permission to re-use all or part of this article, use this link http://clincancerres.aacrjournals.org/content/early/2012/10/06/1078-0432.CCR-12-0505. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC) Rightslink site.