Integrative Genomics Analysis Identifies Candidate Drivers at 3Q26-29 Amplicon in Squamous Cell Carcinoma of the Lung

Published OnlineFirst August 1, 2013; DOI: 10.1158/1078-0432.CCR-13-0594

Clinical Cancer Human Cancer Biology Research

Integrative Genomics Analysis Identiﬁes Candidate Drivers at 3q26-29 Amplicon in Squamous Cell Carcinoma of the Lung

Jing Wang1, Jun Qian2, Megan D. Hoeksema2, Yong Zou2, Allan V. Espinosa2, S.M. Jamshedur Rahman2, Bing Zhang1, and Pierre P. Massion2,3

Abstract Purpose: Chromosome 3q26-29 is a critical region of genomic amplification in lung squamous cell carcinomas (SCC). Identification of candidate drivers in this region could help uncover new mechanisms in the pathogenesis and potentially new targets in SCC of the lung. Experimental Design: We conducted a meta-analysis of seven independent datasets containing a total of 593 human primary SCC samples to identify consensus candidate drivers in 3q26-29 amplicon. Through integrating protein–protein interaction network information, we further filtered for candidates that may function together in a network. Computationally predicted candidates were validated using RNA interfer- ence (RNAi) knockdown and cell viability assays. Clinical relevance of the experimentally supported drivers was evaluated in an independent cohort of 52 lung SCC patients using survival analysis. Results: The meta-analysis identified 20 consensus candidates, among which four (SENP2, DCUN1D1, DVL3, and UBXN7) are involved in a small protein–protein interaction network. Knocking down any of the four proteins led to cell growth inhibition of the 3q26-29–amplified SCC. Moreover, knocking down of SENP2 resulted in the most significant cell growth inhibition and downregulation of DCUN1D1 and DVL3. Importantly, a gene expression signature composed of SENP2, DCUN1D1, and DVL3 stratified patients into subgroups with different response to adjuvant chemotherapy. Conclusion: Together, our findings show that SENP2, DCUN1D1, and DVL3 are candidate driver genes in the 3q26-29 amplicon of SCC, providing novel insights into the molecular mechanisms of disease progression and may have significant implication in the management of SCC of the lung. Clin Cancer Res; 1–11. 2013 AACR.

Introduction with stage II NSCLC will die within 5 years, mainly due to Lung cancer is the leading cause of cancer-related death the development of distant metastases (2). among men and women in the United States. In 2011 alone, In SCC of the lung, chromosome 3 aberrations are a total of 221,130 new cases and 156,940 deaths were prevalent. Recurrent deletion on the 3p arm and gain on reported (1). Eighty-five percent of these lung cancers are the 3q arm are present in the majority of cases (3, 4). non–small cell lung carcinomas (NSCLC), a heterogeneous Chromosome 3q26-29 amplification represents a critical group composed of mostly squamous cell carcinomas region of genomic alteration in lung SCC. It occurs in up to (SCC) and adenocarcinomas. Despite improvements in 70% of lung SCCs and is of significant magnitude (i.e., 2- to response to increasingly sophisticated combination thera- 10-fold increase in copy number; ref. 5). The amplicon py, approximately 40% of patients with stage I and 60% develops early in lung cancer development and persists at the metastatic stage (6). Identification of genes driving the selection for the 3q26-29 amplicon has been a significant Authors' Affiliations: 1Department of Biomedical Informatics, 2Division of research focus during the last decade. Using array compar- Allergy, Pulmonary and Critical Care Medicine, Vanderbilt University ative genomic hybridization (CGH) and mRNA expression School of Medicine; and 3Veterans Affairs, Tennessee Valley Health Care Systems, Nashville, Tennessee array technologies, a number of candidate driver genes in this amplicon have been identified and proposed to con- Note: Supplementary data for this article are available at Clinical Cancer Research Online (http://clincancerres.aacrjournals.org/). tribute to the development of lung cancer. These genes include PIK3CA (7, 8), SOX2 (9), DCUN1D1 (10), TP63 J. Wang and J. Qian contributed equally to this work. (11), EIF4G1 (12), EVI1 (13), THPO (14), TERC (15), ECT2 Corresponding Authors: Pierre P. Massion, Vanderbilt University School (16), PRKCI (17), EPHB3 (18), MASP1 (18), and SST (18). of Medicine, 2220 Pierce Avenue, 640 PRB 6838, Nashville, TN 37203. Phone: 615-936-2256; Fax: 615-936-1790; E-mail: Despite remarkable progress on candidate driver gene [email protected]; and Bing Zhang, identification for this amplicon, no consensus exists as [email protected] which are the drivers and what are their functional implica- doi: 10.1158/1078-0432.CCR-13-0594 tions (9, 10). Moreover, existing studies tend to focus on a 2013 American Association for Cancer Research. single gene in the focal region of amplification. It has been

www.aacrjournals.org OF1

Wang et al.

ref. 21), Expression Project for Oncology (Expo; GSE2109), Translational Relevance Lee and colleagues (GSE8894; ref. 22), Raponi and collea- The 3q26-29 amplicon is the most common and gues (GSE4573; ref. 23), Wilkerson and colleagues significant chromosomal aberration in the progression (GSE17710; ref. 24), and Roepman and colleagues (25). of squamous cell carcinoma (SCC) of the lung from Gene expression data with survival and adjuvant chemo- severe dysplasia to metastasis. Identification of candi- therapy information was from Zhu and colleagues date driver genes in this 3q26-29 amplicon would (GSE14814 from the JBR.10 trial; ref. 26), which only uncover new mechanisms in the pathogenesis and contained stage I and II patients. Only lung SCC samples potentially new targets in SCC. Using an integrative from these datasets were included in this study. For each of computational analysis followed by experimental vali- the filtered datasets, the gene expression data were subjected dation, we identified three consensus candidate driver to quantile normalization and standardized using a gene- genes (SENP2, DCUN1D1, and DVL3) that could stratify wise Z-score transformation. patients into subgroups with different response to adju- Of note, 299 and 110 paired gene expression and copy vant chemotherapy. This integrated signature could number data for head and neck SCC and cervical SCC were serve as a molecular assay to guide personalized treat- also downloaded from TCGA portal, in which gene expres- ment decisions in SCC of the lung. sion data were from Illumina HiSeq 2000 RNA Sequencing Version 2 analysis and copy number data were from Affy- metrix Genome-Wide Human SNP Array 6.0 array. Human protein interaction data were collected and inte- suggested that different genes in an amplicon may contrib- grated from HPRD, MINT, intact, REACTOME, BioGRID, ute synergistically to tumor progression (19, 20). We there- and DIP in April 2010. Only experimentally determined fore hypothesized that genomic amplification leads to interactions supported by publications were considered to increased expression of several key regulators in the assure the reliability of the network. The consolidated 3q26-29 amplicon that cooperate with one another to dataset comprised 94,146 interactions involving 11,660 promote SCC cell growth. To identify these key regulators, genes. we took an integrative genomics approach as depicted in Fig. 1. Although existing bioinformatics methods usually Calculating the amplification scores for TCGA copy require paired copy number alteration (CNA) and gene number data expression data for the prioritization of candidate genes in On the basis of the TCGA level 2 data downloaded from an amplicon, in this study, we have developed and validated the TCGA data portal, we identified the markers in the 3q26- a simple but effective method for prioritizing candidate 29 regions for each sample and then calculated the average driver genes using expression data alone. This method normalized log2 score for all these markers to get the allowed us to integrate seven independent datasets contain- amplification score for each sample. ing a total of 593 human primary tumor samples to identify 20 consensus candidate drivers in the 3q26-29 amplicon. Concordance index To further filter for genes with potential functional connec- Concordance index (C-index; ref. 27) was defined as the tivity, we used a network-based approach and identified a fraction of all pairs of subjects (e.g., gene expression samples small protein interaction network consisting of SENP2, or genes in the amplicon) whose predicted scores (e.g., DCUN1D1, DVL3, and UBXN7, which are involved in average expression scores or correlation between gene posttranslational modifications such as SUMOylation, ned- expression and average expression score) were correctly dylation, and ubiquitination. Experimental interrogations ordered among all subjects that can actually be ordered in showed a role of the network in promoting cell proliferation the observed scores (e.g., amplification scores or correlation and suggested potential regulatory relationships between between gene expression and amplification score). SENP2, DCUN1D1, and DVL3. Moreover, patient stratification based on gene expression values of this three-gene Order statistics signature provides potential useful information about the We divided the ranks by the total number of ranked genes adjuvant chemotherapy benefit for patients with SCC of the (excluding genes with no rank because of missing values) lung. and then calculated Q statistic for each gene in the 3q26-29 regions according to the following equation: Materials and Methods Xk V i1 ki i Vk ¼ ð1Þ rNkþ Datasets i¼1 i! 1 One hundred and fifty-two paired gene expression and copy number data for lung SCC were downloaded from where Q(r1, r2, ..., rN) ¼ N!VN, V0 ¼ 1, ri is the rank ratio for The Cancer Genome Atlas (TCGA) portal, in which gene data source i, N is the number of data sources used, and r0 ¼ expression data were from Agilent 244 K whole genome 0. Because the Q statistic calculated by the above equation expression array and copy number data were from Agilent are not uniformly distributed under the null hypothesis and CGH 415 K array. Six unpaired gene expression data were thus cannot be used directly as P values (28). Thus, we from the following studies: Bild and colleagues (GSE3141; randomly permutated the ranks of all genes in each of seven

OF2 Clin Cancer Res; 2013 Clinical Cancer Research

3q26-29 Candidate Drivers in Lung Squamous Cell Carcinoma

3q26-29 (164 genes)

Predefined amplicon and matched A CNA and gene expression data Method for prioritizing genes in an amplicon using gene expression data alone

Ranked list based on matched CNA and gene expression data Ranked lists based on individual gene expression datasets

B Order statistics-based data fusion

20 candidate driver genes

C Network-based prioritization Protein—protein interaction network 4 functionally linked candidates

RNAi knock down and cell viability assays 1.00.80.60.40.20.0

D Proproliferation effect 3 experimentally supported candidates

High expression with CTX n = 14 P = 0.031 HR = 0.24 (0.06–0.97) High expression without CTX n = 12 P = 0.24 HR = 1.98 (0.62–6.33) Low expression without CTX n = 13 P = 0.79 Overall survival probability survival Overall Survival analysis Low expression with CTX n = 13 HR = 0.84 (0.22–3.11)

86420 Survival time (y) Clinical data Clinical relevance

Figure 1. Overview of the integrative genomics approach. A, developing a method for estimating sample-specific amplification level and prioritizing candidate driver genes using expression data alone, and applying the method to six independent datasets. B, using the order statistics technique to identify candidate driver genes that are supported by multiple datasets. C, using network analysis to filter for functionally related candidate drivers. D, experimental refinement and clinical evaluation of the identified driver genes. CNA, copy number alteration. datasets and calculated the random Q statistics. We repeated NetWalker evaluates the statistical significance of the scores above process 1,000,000 times and generated a null Q using a global P value and a local P value. A significant statistic distribution for each gene. Comparing the real Q global P value indicates a nonrandom association between statistic with null distribution, we got a P value for each the gene and the input seeds, whereas a significant local gene. Finally, the P values were corrected by the false P value ensures that the significant association is not discovery rate (FDR) of Benjamini–Hochberg procedure simply due to network topology. (29). Under 1% FDR, we identified candidate driver genes in the 3q26-29 regions. Cell cultures Six human lung squamous carcinoma cell lines H520, Random walk analysis HCC95, H2882, HCC15, H157, and SW900 and four NetWalker (http://bioinfo.vanderbilt.edu/netwalker; human lung adenocarcinoma cell lines H1819, H1648, refs. 30, 31) was used to run random walk analysis based A549, and H23, were purchased from the American Type on the integrated human protein–protein interaction net- Culture Collection. Lung cancer cell lines were maintained work described earlier. NetWalker identifies genes of poten- in RPMI with 10% FBS. All cells were grown in 1 mmol/L tial biologic importance based on the assumption that penicillin/streptomycin. mechanistically important genes are likely to form tightly connected groups, whereas other genes tend to be randomly RNAi knockdown assays distributed on the network. Using a set of genes as "seeds," For transient siRNA transfection, DharmaFECT 1 Trans- NetWalker calculated a score for each gene in the network fection Reagent, nontargeting siRNA #1, siGENOME based on its overall proximity to the seed genes, where the SMART pool siRNA against SENP2 (M-006033-01-005), proximity is measured by the random walk similarity (32). DCUN1D1 (M-019139-01-0005), DVL3 (M-004070-01-

www.aacrjournals.org Clin Cancer Res; 2013 OF3

Wang et al.

0005), and UBXN7 (M-023533-01-005) were purchased fication directly contribute to increased mRNA expression from Thermo Scientific. Opti-MEM I media was purchased in 152 lung SCC samples. We first estimated a 3q26-29 from Life Technologies. siRNA were suspended in water at a amplification score for each sample based on the CNA concentration of 20 nmol/L. The transfections were carried data (Supplementary Table S1) and then covisualized out according to the manufacturer’s instructions or as CNA and gene expression data in the Integrative Genomic previously described (33). Briefly, for Western blot analysis, Viewer (IGV; http://www.broadinstitute.org/igv/home), 1 105 cells were seeded into 6-well plates with medium with the samples in both datasets ordered on the basis overnight. For each well, 5 or 10 mL of siRNA were mixed of the 3q26-29 amplification score. As shown in Fig. 2A with 185 mL Opti-MEM I and then combined with another and B, although 3q26-29 amplification is prevalent, dif- mixture prepared using 4 mL DharmaFECT 1 transfection ferent samples showed different levels of amplification. reagent and 15 mL Opti-MEM I. The final concentration of Samples with higher level of amplification also showed the siRNA was 25 to 100 nmol/L. For cell viability assay in relatively higher expression for genes in this amplicon. 96-well plates, 50 or 100 nmol/L final concentration of We also found that most genes in this amplicon showed siRNA was used. similar expression pattern, suggesting that genomic amplification contributes directly to an increase in mRNA Cell viability assay gene expression. Next, we calculated the Spearman cor- Lung cancer cells were transiently transfected in 96-well relation coefficient between amplification scores and plates with target siRNA or negative control siRNA as expression values for each gene in the expression data. described earlier. The CellTiter 96 AQueous One Solution Genes in the 3q26-29 regions were significantly better Cell Proliferation Assay (Promega) was then conducted correlated with the amplification score compared with following the manufacturer’s instructions or as previously genes outside the region (P < 2.2E16; Fig. 2C). There- described (34) on the wells posttransfection at 3 or 6 days. A fore, we reasoned that the amplification level of a chro- representative viability experiment is shown in fold change mosomal region could be inferred on the basis of the that was normalized to negative siRNA control group with expression of genes within the region. average SD. Using TCGA gene expression data, we calculated an average gene expression score for each sample based on all Immunoblot analysis genes in the 3q26-29 amplicon (Supplementary Table S1). Rabbit anti-SENP2 antibody (HPA029248) was pur- The average expression score was significantly correlated chased from Sigma-Aldrich. Rabbit anti-UBXD7 (AB10037) with the amplification score derived from the copy number was from Millipore. Rabbit Dishevelled 3 antibody (GTX- data (Spearman correlation r ¼ 0.94 and P < 2.2E16) and 102509) and rabbit DCUN1D1 antibody (GTX116558) the concordance index between the two scores was 89.4% were from Genetex. Western blotting was carried out using (Fig. 2D), showing the validity of using the average expres- standard procedures as described in our previous study sion score to predict the regional amplification level. We (33), with detection using the enhanced chemilumines- also ordered the samples by average expression scores and cence system (Thermo Scientific). Antibody dilutions for visualized the copy number data in the IGV. As shown in immunoblotting were 1:1,000. The blots were reprobed Supplementary Fig. S1, the magnitude of 3q26-29 ampli- with an anti-b-actin antibody (Sigma-Aldrich) to correct for fication clearly declined with the decreasing of the average protein loading differences. Anti-rabbit secondary antibody expression score. was purchased from Promega. Finally, we compared the correlation between gene expression and amplification level (cor_exp_amp) with the Results correlation between gene expression and average expression Predicting 3q26-29 gene amplification level using gene score (cor_exp_ave) for each gene in the 3q26-29 amplicon expression data (Supplementary Table S2). We found high consistency Correlation between gene expression and regional ampli- between the two types of correlations (Spearman correla- fication level has been widely used to prioritize candidate tion r ¼ 0.95 and P < 2.2E16; concordance index ¼ driver genes in a predefined amplicon because mRNA over- 90.6%; Fig. 2E), providing clear support for using the expression can better translate the effect of elevated copy correlation between gene expression and average expression number to tumor progression and offer biologic insights score to prioritize candidate driver genes. Meanwhile, we that are closer to levels of protein expression and function found that previously reported candidate driver genes had a (35). This approach requires paired CNA and gene expres- wide spread along the diagonal (Fig. 2E), suggesting that sion data, but many large-scale lung SCC studies, including many of them are not supported by the TCGA data, and a those with available clinical endpoints, comprise only gene consensus conclusion on the driver genes at the 3q26-29 expression data. Therefore, we first investigated the possi- amplicon requires integrative analysis of data from multiple bility of using gene expression data alone to predict 3q26-29 patient cohorts. gene amplification level and prioritize candidate driver genes in the region. Integrative identification of candidate driver genes Using the TCGA dataset with paired CNA and gene The above results opened the possibility to estimate expression data (36), we examined whether gene ampli- sample-specific amplification level and prioritize candidate

OF4 Clin Cancer Res; 2013 Clinical Cancer Research

3q26-29 Candidate Drivers in Lung Squamous Cell Carcinoma

p26.2 p25.2 p24.3 p24.1 p22.2 p21.31 p14.3 p14.1 p12.3 p12.1 q11.2 q12.3 q13.2 q13.33 q22.1 q23 q25.1 q25.33 q26.2 q26.32 q27.2 q29 A Gene expression data

B Copy number alteration data

C E Genes in 3q26-29 DCUN1D1 Spearman correlation r = 0.99 Other genes ECT2 0.8 0.8 1.0 P < 2.2E–16 SOX2

0.60.40.20.0 TP63 PIK3CA

0.6 EIF4G1 PRKCI EPHB3 Cumulative probability Kolmogorov–Smirnov test P < 2.2E–16 0.4

1.00.5 0.0 –0.5 SST

Spearman correlation coefficient 0.2 D THPO Spearman correlation r = 0.94

P < 2.2E–16 Gene's corr_exp_amp EVI1

MASP1 –0.2 Concordance index=96.2%

0.0 0.5 1.0 1.5 2.0 Concordance index=89.4% –0.4 0.0 Sample's amplification score –0.2 0.0 0.2 0.4 0.6 0.8 –1.0 –0.5 0.0 0.5 Gene's corr_exp_ave Sample's average expression score

Figure 2. Genomic amplification can be inferred from gene expression data. The IGV plot of 152 paired TCGA gene expression data (A) and copy number data (B), x-axis represents chromosome 3 and y-axis represents TCGA samples that are in descending order according to the amplification scores. Different color shades represent the expression or amplification values of samples from blue (the minimum value) to red (the maximum value). C, cumulative probability distribution of spearman correlation coefficient between amplification scores of samples and expression values of genes in 3q26-29 amplicon (red) and other genes in the expression data (blue) based on 152 TCGA paired copy number and gene expression samples. D, the scatter plot of amplification scores against average expression scores for each of the 152 TCGA SCC samples. E, the scatter plot of correlation between gene expression and amplification score (corr_exp_amp) against correlation between gene expression and average expression score (corr_exp_ave) for 164 genes in the 3q26-29 amplicon. Previously published candidate driver genes are labeled in the scatter plot. The solid lines in (D) and (E) are fitted lines based on the data in the plots. Because the data points in (D) and (E) are distributed along the diagonals, the data series represented by the x- and y-axis in each figure have a high concordance index (89.4% and 96.2%). driver genes using gene expression data alone. Therefore, we ples as candidate drivers of this amplicon. We first ranked applied the method to another six independent lung SCC genes based on the correlation between gene expression gene expression datasets available in the public domain and sample amplification score for the TCGA dataset or (Supplementary Table S3) and estimated sample-specific based on the correlation between gene expression and amplification levels (Supplementary Tables S4–S9). We average expression score for other six datasets, respectively. integrated all seven datasets to identify genes that were Then, using the N-dimensional order statistic, we combined consistently overexpressed in the 3q26-29 amplified sam- the seven ranked lists to calculate a Q statistic and

www.aacrjournals.org Clin Cancer Res; 2013 OF5

Wang et al.

Table 1. Twenty candidate driver genes identiﬁed by the order statistic method with an FDR threshold of 1%

Rank

TCGA GSE2109 GSE17710 GSE3141 GSE4573 GSE8894 Roepman Symbol Band (164)a (173) (160) (173) (124) (173) (114) Q statistic FDR ABCC5 q27.1 6b 1 7 15 12 26 7 8.35E08 <0.001 ACTL6A q26.33 2 43 1 22 24 35 26 3.54E06 <0.001 DCUN1D1 q26.33 3 53 10 1 2 12 NAc 3.66E07 <0.001 DLG1 q29 25 44 9 13 30 11 22 1.78E05 <0.001 MFN1 q26.32 20 7 22 10 14 44 25 1.28E05 <0.001 NCBP2 q29 28 23 15 4 15 23 16 1.38E06 <0.001 PRKCI q26.2 43 36 20 6 13 3 18 6.17E06 <0.001 PSMD2 q27.1 29 3 35 33 4 5 20 1.34E06 <0.001 SENP2 q27.2 11 32 3 3 5 7 2 1.27E09 <0.001 SENP5 q29 42 19 12 24 23 1 5 2.12E06 <0.001 UBXN7 q29 NA 39 NA 8 22 18 4 1.05E04 <0.001 ZNF639 q26.32 1 34 11 38 95 21 6 2.68E05 <0.001 ATP11B q26.33 5 48 26 19 21 9 1 2.46E06 0.0091 DVL3 q27.1 31 16 18 27 27 32 10 1.45E05 0.0091 EIF2B5 q27.1 22 11 23 32 16 14 NA 2.83E05 0.0091 FXR1 q26.33 19 12 38 37 9 8 13 3.60E06 0.0091 LSG1 q29 32 30 2 34 38 43 NA 1.11E04 0.0091 NDUFB5 q26.33 27 33 14 2 8 61 NA 6.47E05 0.0091 PIGX q29 23 25 24 14 NA 24 3 6.86E06 0.0091 PIK3CA q26.32 8 41 27 25 3 34 NA 1.61E05 0.0091

aThe number of genes in both dataset and 3q26-29. bGene rank in a dataset according to the corr_exp_amp (TCGA) or corr_exp_ave (other datasets) from largest to smallest. cNA, the gene is not in the corresponding dataset.

corresponding P value (see details in Materials and Meth- "seeds," we applied NetWalker (30) to identify genes ods), which represents the random chance of getting the that might function together in a coordinated manner same or better observed rank combinations for each gene (see details in Materials and Methods). Controlling both (Supplementary Table S10). Under an FDR of 1%, 20 genes local and global P values at the 0.01 level, we identified were identified as candidate drivers (Table 1). Because four of 20 candidate genes (SENP2, DCUN1D1, DVL3, 3q26-29 amplicon is also present in SCC from other sites and UBXN7) that are involved in a small, connected such as head and neck (37) and cervical (38), we tested network with 13 significant nodes (Fig. 3). Many genes whether these 20 genes were also candidate drivers in other in the network are involved in SUMOylation, neddyla- SCC. On the basis of the paired gene expression and copy tion, and ubiquitin pathways. number data in TCGA (see Materials and Methods), we calculated the correlation between gene expression and 3q26-29 amplification level for these 20 genes in head and neck SCC and cervical SCC. As shown in Supplemen- tary Tables S11 and S12, gene expression for all 20 genes were significantly correlated with 3q26-29 amplification level (for head and neck SCC, Spearman correlation P < 1.27E24; for cervical SCC, Spearman correlation P < 7.31E07), implying the validity of these 20 genes as candidate drivers in SCC. Colocalization of candidate driver genes in a focal region of amplification suggests that coordinate amplification and overexpression of some of these genes might

be required to alter speciﬁc mechanisms and eventually Figure 3. Graphical representation of an inferred SCC related candidate provide a growth advantage to cells that overexpress them driver network centered around four candidate driver genes in 3q26-29. (19, 20). Therefore, using the 20 candidate driver genes as Four candidate driver genes are shown in red.

OF6 Clin Cancer Res; 2013 Clinical Cancer Research

3q26-29 Candidate Drivers in Lung Squamous Cell Carcinoma

Functional implications of four candidate drivers in We next assessed the effect on lung cancer cell growth by SCC knocking down the four proteins expression. We measured Prior work has established the oncogenic role of the proliferation of H520 cells with siRNAs against four DCUN1D1 (defective in cullin neddylation 1 domain con- genes respectively or scramble control siRNAs over 3 and 6 taining 1, also known as SCCRO or SCC-related oncogene) days (Fig. 4B). The silencing of the four proteins led to the being an important component of the neddylation E3 cell growth inhibition at varied degrees (20%–40% inhibi- complex in SCC (10, 39). In contrast, the specific function tion), among which, knocking down SENP2 resulted in of SENP2, DLV3, or UBXN7 in SCCs of the lung remains most significant inhibition (40%). Interestingly, we found unknown. To determine the relative contribution of over- that cells with SENP2 knockdown induced the down- expression of four candidate driver genes in lung SCC, we regulation of DCUN1D1 and DVL3, but not UBXN7 (Fig. first sought to examine the expression status of four proteins 4C). In contrast, knockdown of DCUN1D1, DVL3, or in a panel of NSCLC cell lines, including five cell lines that UBXN7 did not lead to downregulation of SENP2 protein harbor 3q amplicon including H520, HCC95, H1648, expression (Fig. 4D). The same findings were confirmed H1819, and H2882 and five cell lines that do not, includ- in another 3q amplified cell line HCC95 but not in A549, ing H157, HCC15, SW900, A549, and H23. As shown which does not harbor 3q amplicon (Supplementary in Fig. 4A, overall expression of the four proteins is higher Fig. S2). in the cell lines harboring 3q amplicon. Using Cancer Cell Line Encyclopedia database (http://www.broadinstitute. Three-gene signature to predict adjuvant org/ccle/home), we further confirmed that four genes are chemotherapy sensitivity in SCC patients amplified and have higher mRNA expression in HCC95, To evaluate the potential clinical significance of the H1648, and H520 compared with the other four cell lines computationally predicted network and experimentally that do not harbor 3q amplification (Supplementary supported driver genes, we tested whether a gene expression Table S13). signature with the three genes (SENP2, DCUN1D1, and

A B

5 5 0 1 9 2 9 8 1 0 0 8 1 4 7 9 9 C 2 8 8 6 5 C 4 3 C 5 2 1 1 1 C W 5 2 H H H H H H H S A H 0.8 SENP2

DCUN1D1 3 Days 0.6 6 Days DVL3

0.4 UBXN7 Fold change(%)

β -Actin 0.2

3q Amplified 3q Nonamplified 0 l tr 2 1 3 7 P D L N C N 1 V X E N D B S U U C D C D 48 h 72 h DVL3 DCUN1D1 UBXN7 NC SC 50 100 50 100 siRNA-SENP2 nmol/L NC SC 25 50 25 50 25 50 siRNA nmol/L SENP2 SENP2

DCUN1D1 DCUN1D1

DVL3 DVL3

UBXN7 UBXN7

β-Actin β-Actin

Figure 4. Loss of SENP2 overexperssion leads to the downregulation of DCUN1D1 and DVL3. A, overexpression of SENP2, DCUN1D1, DVL3, and UBXN7 in 3q ampliﬁed SCC cell lines. B, knockdown of SENP2, DCUN1D1, DVL3, and UBXN7 results in decreased cell viability on H520 cells. C, knockdown of SENP2 expression leads to downregulation of DCUN1D1 and DVL3 but not UBXN7 in H520 cells. D, knockdown DCUN1D1, DVL3, or UBXN7 fails to induce downregulation of SENP2 expression.

www.aacrjournals.org Clin Cancer Res; 2013 OF7

Wang et al.

DVL3) could inform patient prognosis and treatment predict potential benefit from adjuvant chemotherapy in response. On the basis of the average of standardized the management of patients with SCC of the lung. Similar expression levels of the three genes, we classified 52 patients results were obtained when the analyses using three-gene from an independent dataset (GSE14814) into two groups: signature were conducted for stage I and II patients sepa- a "high-expression" group with above median expression of rately (Supplementary Fig. S3). We also conducted survival the genes and a "low-expression" group with below median analysis by stratifying patients based on each of the three expression of the genes. For each group, we further sepa- genes individually. As shown in Supplementary Fig. S4, rated patients into two subgroups based on whether they stratification based on all three genes outperformed those received adjuvant chemotherapy. based on individual genes, suggesting that cooperation of As a baseline comparison (i.e., for patients who did not the three genes may play an important role in cancer received adjuvant chemotherapy), the high-expression therapy. group had a relatively shorter survival outcome compared with the low-expression group [Fig. 5, red solid line vs. green Discussion solid line; HR, 1.98; 95% confidence interval (CI), 0.62– In this report, we investigated the functional implications 6.33; P ¼ 0.24]. Interestingly, adjuvant chemotherapy was of the chromosome 3q amplicon in lung SCC. We show that significantly associated with improved survival outcome for (i) a simple but effective method can be used to estimate the high-expression group (Fig. 5, red dashed line vs. red sample-specific amplification level and prioritize candidate solid line; HR, 0.243; 95% CI, 0.061–0.967; P ¼ 0.031], driver genes using expression data alone; (ii) data fusion with a 85.7% overall survival rate at 5 years for patients who and network analysis techniques enable data integration received chemotherapy compared with 41.7% for patients across different studies and among different genes, allowing who did not receive adjuvant chemotherapy. In contrast, identification of driver genes that are supported by multiple adjuvant chemotherapy had limited effect on patients in the studies and involved in a common biologic theme; (iii) all low-expression group (Fig. 5, green dashed line vs. green four computationally inferred driver genes including solid line; HR, 0.835; 95% CI, 0.224–3.11; P ¼ 0.788), with SENP2, DCUN1D1, DVL3, and UBXN7 were experimental- a 76.9% overall survival rate at 5 years for patients who ly proven to regulate cellular proliferation in lung squa- received adjuvant chemotherapy as compared with 69.2% mous cancer cells that harbor the amplicon and three of for patients who did not receive chemotherapy. Thus, them (SENP2, DCUN1D1, and DVL3) might be function- patient stratification based on the three-gene signature may ally related to one another; and (iv) the three-gene signature may be predictive of response to conventional chemotherapy in SCC. Although analyzing gene expression in conjunction with

1.0 DNA-level changes such as CNA holds great promise in the identiﬁcation of candidate driver genes of lung SCC, many

0.8 ﬁndings from this type of analysis are not reproducible in a new patient cohort. For example, many previously reported candidate driver genes are not supported by the TCGA data 0.6 (Fig. 2E). One potential solution is to integrate paired CNA and gene expression datasets from multiple cohorts. How-

0.4 ever, paired lung SCC molecular datasets are still limited High expression with CTX n = 14 P = 0.031 and even fewer with clinical annotation. Because there are HR = 0.24 (0.06–0.97) High expression without CTX n = 12 sufﬁcient numbers of published lung SCC gene expression Overall survivalprobability P = 0.24 0.2 HR = 1.98 (0.62–6.33) Low expression without CTX n = 13 datasets, in this study, we developed a method to predict P = 0.79 3q26-29 ampliﬁcation level for individual tumor samples Low expression with CTX n = 13 HR = 0.84 (0.22–3.11)

0.0 from a patient cohort based on gene expression data alone. 0 2 4 6 8 This allowed us to integrate data from one cohort with Survival time (y) paired CNA and gene expression measurements and another six cohorts with only gene expression measurements to identify consensus candidate driver genes with consistent Figure 5. Three-gene (SENP2, DCUN1D1, and DVL3) expression overexpression in amplified tumors from multiple patient signature predicts response to adjuvant chemotherapy (CTX) in early- cohorts. For many cancer types, genomic studies have – stage NSCLC. Red solid and red dashed lines represent Kaplan Meier identified amplicons without well-defined driver genes. survival curves for patients with high expression of the three-gene signature, without and with adjuvant chemotherapy, respectively; green Although paired copy number and gene expression datasets solid and green dashed lines represent Kaplan–Meier survival curves for remain very limited across all cancer types, gene expression patients with low expression of the three-gene signature, without and datasets are relatively abundant in public databases such as with adjuvant chemotherapy, respectively. "n" represents the number of the Gene Expression Omnibus (GEO). Thus, our method patients in the corresponding group; "P" and "HR" represents log-rank test P value and hazard ratio for a pair of survival curves, respectively. could have a general application in predicting amplification Numbers in the parentheses indicate the 95% CI of HR. The analysis was level and candidate driver genes in predefined amplicons based on GSE14814 from the JBR.10 trial. for other cancer types.

OF8 Clin Cancer Res; 2013 Clinical Cancer Research

3q26-29 Candidate Drivers in Lung Squamous Cell Carcinoma

In our study, we used the average expression score of all tional modifications may differ (44). Notably, NEDD8 is genes in the 3q26-29 amplicon to predict 3q26-29 ampli- conjugated to three lysines within the C-terminus of p53 fication level. Although chromosomal amplification direct- through the E3 ligase activity of MDM2 (45), adding anoth- ly contributes to an increased mRNA expression level of er mechanism by which MDM2 inhibits p53 activity. genes in the amplified region (see Fig. 2A), only 110 of 164 DCUN1D1 is novel identified NEDD8 E3 complex com- genes in this region showed significant correlation with the ponent and has been proposed be to an oncogene in lung amplification score (Spearman P < 0.01). Moreover, even SCC (10, 39). Mechanistically, DCUN1D1 binds to the genes with significant correlations also showed different components of the neddylation pathway consisting of levels of correlation (Spearman correlation r ranging from Cullin-ROC1, Ubc12, and CAND1 to promote cullinned- 0.87 to 0.21). Thus, it is possible that expression of some dylation via releasing the inhibitory effects of CAND1 (39). genes may better represent genomic amplification than It is unclear how neddylation is regulated and how many others. To test this possibility, we extracted the first principal oncogenic neddylated targets exist in cancer cell. Our data component from gene expression data for all genes in the showed that knocking down SENP2 using siRNA leads to amplicon and used the first principal component as an the downregulation of DCU1ND1, prompting us to pro- alternative approach to estimate the level of amplification. pose that DCU1ND1 might be a potential SUMO-modified However, the results based on first principal component target regulated by SENP2 in lung SCC, a mechanism were very similar to that based on all genes (data not similar to SENP2-MDM2 pathway. shown), suggesting the validity of using average expression Disheveled (DVL) is an essential scaffold protein that score of all genes in an amplicon as an easy method for transduces both the canonical (b-catenin–dependent) and estimating amplification level. the noncanonical (b-catenin–independent) Wntignaling It is generally believed that regions of recurrent genome pathways (46). Activated DVL3 promotes Wnt/b-catenin CNA harbor genes that are essential to cancer progression, activation and undergoes KLHL12–Cul3–mediated ubiqui- nevertheless many functional studies focus on identifying a tination for degradation (47). Wnt-b-catenin pathway is single candidate driver gene in the region of CNA respon- disregulated in lung cancer (48) and may lead to new sible for cancer progression. It has been suggested that treatment strategies (49). Dvl3 overexpression has also been multiple genes from an amplicon may cooperatively pro- found in NSCLC, and once silenced, inhibition of cellular mote tumorigenesis (20, 40). Thus, in our study, we used proliferation is observed (50). SENP2 negatively modulates network-based analysis to identify genes that might func- b-catenin via its deSUMOylation activity in hepatocellular tion in a coordinated manner. Through integrated compu- carcinoma cell (51) and colon cancer cells (52). It is sug- tational network analysis and experimental interrogation, gested that noncanonical Wnt/PCP pathway but not canon- we identified three genes that may function together to ical Wnt/b-catenin was enhanced in lung SCCs (48). Our promote tumor cell growth. These genes are involved in data showed that SENP2 knockdown leads to DVL3 down- SUMOylation, neddylation, or ubiquitination pathway. regulation. Therefore, we hypothesize that SENP2 may Recently, the functional cross-talk between these similar regulates b-catenin through DVL3 dependent or indepen- pathways has emerged (41). Nevertheless, how three path- dent mechanisms in lung SCC cells and this will be tested in ways contribute to lung SCC development remains to be the context of future studies. Thus, unlike previous studies investigated. that identify individual genes from the 3q26-29 amplicon, Our experimental data suggest that SENP2 may regulate the three genes identified in this study may represent a the expression of DCUN1D1 and DVL3. SENP2 belongs to common biologic mechanism that provides growth advan- Sentrin/SUMO-specific proteases family and has diverse tage to tumors harboring the amplicon. function via SUMOylation or deSUMOylation in human Our survival analysis further supports potential impor- cells. In cancer cell, components of the SUMOylation or tance of these three genes in lung SCC patient prognosis, deSUMOylation machinery are enhanced or reduced (42). and particularly, response to adjuvant chemotherapy. In lung SCC cells, SENP2 has highest prevalence of ampli- Despite the small sample size of the patient cohort used fication compared with other SENP family members (Sup- for survival analysis, we observed distinct response to adju- plementary Fig. S5), further supporting its key role in lung vant chemotherapy for patients with high- and low-level SCC development. Whether and how SENP gene expression expression of the three genes, respectively and indepen- levels disturb SUMO homeostasis and contribute to lung dently of the disease stage. Unfortunately, no publicly SCC cancer development and progression are yet to be available datasets in SCC with annotated clinical outcomes determined. Many SUMOmodified proteins function in were identified to confirm our preliminary data. Upon regulation of transcription, chromatin structure, mainte- further validation in independent cohorts of patients with nance of the genome, and signal transduction (41). MDM2- NSCLC, the three-gene signature may provide significant p53 is most well-studied SUMOmodified proteins. SENP2 value in personalizing the management of patients with promotes the deSUMOylation of Mdm2, contributing to SCC of the lung. the p53-dependent regulations (43). Interestingly, the In summary, using an integrative genomics approach, tumour suppressor p53 has been shown to be modified at we identified three functionally linked candidate driver its C-terminus with ubiquitin, SUMO and NEDD8, genes at 3q26-29 that provide novel mechanistic and clin- although the consequence undergoing these posttransla- ical insights into the pathobiology of lung squamous

www.aacrjournals.org Clin Cancer Res; 2013 OF9

Wang et al.

carcinoma. Our results show the power of an integrative Writing, review, and/or revision of the manuscript: J. Wang, J. Qian, A.V. Espinosa, S.M.J. Rahman, B. Zhang, P.P. Massion systems biology approach in effectively identifying candi- Administrative, technical, or material support (i.e., reporting or orga- date driver genes from a predefined tumor amplicon. nizing data, constructing databases): J. Wang, J. Qian, M.D. Hoeksema, Y. Zou, P.P. Massion Study supervision: B. Zhang, P.P. Massion Disclosure of Potential Conflicts of Interest No potential conflicts of interest were disclosed. Grant Support This work was funded in part by the NIH CA102353 (to P.P. Massion), Authors' Contributions GM088822 (to B. Zhang), and the Vanderbilt SPORE in lung CA CA90949. Conception and design: J. Wang, J. Qian, S.M.J. Rahman, B. Zhang, P.P. The costs of publication of this article were defrayed in part by the Massion payment of page charges. This article must therefore be hereby marked Development of methodology: J. Wang, Y. Zou, B. Zhang, P.P. Massion advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate Acquisition of data (provided animals, acquired and managed patients, this fact. provided facilities, etc.): J. Wang, M.D. Hoeksema, Y. Zou, P.P. Massion Analysis and interpretation of data (e.g., statistical analysis, biosta- tistics, computational analysis): J. Wang, J. Qian, Y. Zou, A.V. Espinosa, Received March 1, 2013; revised June 28, 2013; accepted July 17, 2013; S.M.J. Rahman, B. Zhang, P.P. Massion published OnlineFirst August 1, 2013.

References 1. Siegel R, Ward E, Brawley O, Jemal A. Cancer statistics, 2011: the 15. Yokoi S, Yasui K, Iizasa T, Imoto I, Fujisawa T, Inazawa J. TERC impact of eliminating socioeconomic and racial disparities on prema- identified as a probable target within the 3q26 amplicon that is ture cancer deaths. CA Cancer J Clin 2011;61:212–36. detected frequently in non–small cell lung cancers. Clin Cancer Res 2. Strauss GM. Adjuvant chemotherapy of lung cancer: methodologic 2003;9:4705–13. issues and therapeutic advances. Hematol Oncol Clin North Am 16. Fields AP, Justilien V. The guanine nucleotide exchange factor (GEF) 2005;19:263–81. Ect2 is an oncogene in human cancer. Adv Enzyme Regul 2010;50: 3. Yakut T, Schulten HJ, Demir A, Frank D, Danner B, Egeli U, et al. 190–200. Assessment of molecular events in squamous and non-squamous cell 17. Regala RP, Weems C, Jamieson L, Khoor A, Edell ES, Lohse CM, et al. lung carcinoma. Lung Cancer 2006;54:293–301. Atypical protein kinase C iota is an oncogene in human non–small cell 4. Purdie KJ, Harwood CA, Gulati A, Chaplin T, Lambert SR, Cerio R, et al. lung cancer. Cancer Res 2005;65:8905–11. Single nucleotide polymorphism array analysis defines a specific 18. KangJU,KooSH,KwonKC,ParkJW,KimJM.Identification of genetic fingerprint for well-differentiated cutaneous SCCs. J Invest novel candidate target genes, including EPHB3, MASP1 and SST at Dermatol 2009;129:1562–8. 3q26.2-q29 in squamous cell carcinoma of the lung. BMC Cancer 5. Qian J, Massion PP. Role of chromosome 3q amplification in lung 2009;9:237. cancer. Journal of Thoracic Oncology: official publication of the Inter- 19. Xu C, Liu Y, Wang P, Fan W, Rue TC, Upton MP, et al. Integrative national Association for the Study of Lung Cancer. 2008;3:212–5. analysis of DNA copy number and gene expression in metastatic oral 6. Levin NA, Brzoska PM, Warnock ML, Gray JW, Christman MF. Iden- squamous cell carcinoma identifies genes associated with poor sur- tification of novel regions of altered DNA copy number in small cell lung vival. Mol Cancer 2010;9:143. tumors. Genes Chromosomes Cancer 1995;13:175–85. 20. Zender L, Spector MS, Xue W, Flemming P, Cordon-Cardo C, Silke J, 7. Massion PP, Taflan PM, Shyr Y, Rahman SM, Yildiz P, Shakthour B, et al. Identification and validation of oncogenes in liver cancer using an et al. Early involvement of the phosphatidylinositol 3-kinase/Akt path- integrative oncogenomic approach. Cell 2006;125:1253–67. way in lung cancer progression. Am J Respir Crit Care Med 2004;170: 21. Bild AH, Yao G, Chang JT, Wang Q, Potti A, Chasse D, et al. Oncogenic 1088–94. pathway signatures in human cancers as a guide to targeted therapies. 8. Massion PP, Kuo WL, Stokoe D, Olshen AB, Treseler PA, Chin K, et al. Nature 2006;439:353–7. Genomic copy number analysis of non–small cell lung cancer using 22. Lee ES, Son DS, Kim SH, Lee J, Jo J, Han J, et al. Prediction of array comparative genomic hybridization: implications of the phos- recurrence-free survival in postoperative non–small cell lung cancer phatidylinositol 3-kinase pathway. Cancer Res 2002;62:3636–40. patients by using an integrated model of clinical information and gene 9. Bass AJ, Watanabe H, Mermel CH, Yu S, Perner S, Verhaak RG, et al. expression. Clin Cancer Res 2008;14:7397–404. SOX2 is an amplified lineage-survival oncogene in lung and esoph- 23. Raponi M, Zhang Y, Yu J, Chen G, Lee G, Taylor JM, et al. Gene ageal squamous cell carcinomas. Nat Genet 2009;41:1238–42. expression signatures for predicting prognosis of squamous cell and 10. Sarkaria I, O-charoenrat P, Talbot SG, Reddy PG, Ngai I, Maghami E, adenocarcinomas of the lung. Cancer Res 2006;66:7466–72. et al. Squamous cell carcinoma related oncogene/DCUN1D1 is highly 24. Wilkerson MD, Yin X, Hoadley KA, Liu Y, Hayward MC, Cabanski CR, conserved and activated by amplification in squamous cell carcino- et al. Lung squamous cell carcinoma mRNA expression subtypes are mas. Cancer Res 2006;66:9437–44. reproducible, clinically important, and correspond to normal cell types. 11. Massion PP, Taflan PM, Jamshedur Rahman SM, Yildiz P, Shyr Y, Clin Cancer Res 2010;16:4864–75. Edgerton ME, et al. Significance of p63 amplification and overexpres- 25. Roepman P, Jassem J, Smit EF, Muley T, Niklinski J, van de Velde T, sion in lung cancer development and prognosis. Cancer Res 2003;63: et al. An immune response enriched 72-gene prognostic profile for 7113–21. early-stage non–small-cell lung cancer. Clin Cancer Res 2009;15: 12. Brass N, Heckel D, Sahin U, Pfreundschuh M, Sybrecht GW, Meese E. 284–90. Translation initiation factor eIF-4gamma is encoded by an amplified 26. Zhu CQ, Ding K, Strumpf D, Weir BA, Meyerson M, Pennell N, et al. gene and induces an immune response in squamous cell lung carci- Prognostic and predictive gene signature for adjuvant chemother- noma. Hum Mol Genet 1997;6:33–9. apy in resected non–small-cell lung cancer. J Clin Oncol 2010;28: 13. Choi YW, Choi JS, Zheng LT, Lim YJ, Yoon HK, Kim YH, et al. 4417–24. Comparative genomic hybridization array analysis and real time PCR 27. Harrell FE Jr, Lee KL, Mark DB. Multivariable prognostic models: issues reveals genomic alterations in squamous cell carcinomas of the lung. in developing models, evaluating assumptions and adequacy, and Lung Cancer 2007;55:43–51. measuring and reducing errors. Stat Med 1996;15:361–87. 14. Racz A, Brass N, Heckel D, Pahl S, Remberger K, Meese E. Expression 28. Aerts S, Lambrechts D, Maity S, Van Loo P, Coessens B, De Smet F, analysis of genes at 3q26-q27 involved in frequent amplification in et al. Gene prioritization through genomic data fusion. Nat Biotechnol squamous cell lung carcinoma. Eur J Cancer 1999;35:641–6. 2006;24:537–44.

OF10 Clin Cancer Res; 2013 Clinical Cancer Research

3q26-29 Candidate Drivers in Lung Squamous Cell Carcinoma

29. Benjamini Y, Hochberg Y. Controlling the false discovery rate: a 40. Kim H, Huang W, Jiang X, Pennicooke B, Park PJ, Johnson MD. practical and powerful approach to multiple testing. J R Stat Soc B Integrative genome analysis reveals an oncomir/oncogene cluster (Methodological) 1995;57:289–300. regulating glioblastoma survivorship. Proc Natl Acad Sci U S A 30. Zhang B, Shi Z, Duncan DT, Prodduturi N, Marnett LJ, Liebler DC. 2010;107:2183–8. Relating protein adduction to gene expression changes: a systems 41. Gill G. SUMO and ubiquitin in the nucleus: different functions, similar approach. Mol Biosyst 2011;7:2118–27. mechanisms? Genes Dev 2004;18:2046–59. 31. Shi M, Beauchamp RD, Zhang B. A network-based gene expression 42. Bawa-Khalfe T, Yeh ET. SUMO losing balance: SUMO proteases signature informs prognosis and treatment for colorectal cancer disrupt SUMO homeostasis to facilitate cancer development and patients. PLoS ONE 2012;7:e41292. progression. Genes Cancer 2010;1:748–52. 32. Kohler S, Bauer S, Horn D, Robinson PN. Walking the interactome for 43. Jiang M, Chiu SY, Hsu W. SUMO-speciﬁc protease 2 in Mdm2- prioritization of candidate disease genes. Am J Hum Genet 2008;82: mediated regulation of p53. Cell Death Differ 2011;18:1005–15. 949–58. 44. Carter S, Vousden KH. p53-Ubl fusions as models of ubiquitination, 33. Qian J, Zou Y, Rahman JS, Lu B, Massion PP. Synergy between sumoylation and neddylation of p53. Cell Cycle 2008;7:2519–28. phosphatidylinositol 3-kinase/Akt pathway and Bcl-xL in the control 45. Xirodimas DP, Saville MK, Bourdon JC, Hay RT, Lane DP. Mdm2- of apoptosis in adenocarcinoma cells of the lung. Mol Cancer Ther mediated NEDD8 conjugation of p53 inhibits its transcriptional activity. 2009;8:101–9. Cell 2004;118:83–97. 34. Hassanein M, Hoeksema MD, Shiota M, Qian J, Harris BK, Chen H, 46. Wallingford JB, Habas R. The developmental biology of Dishevelled: et al. SLC1A5 mediates glutamine transport required for lung an enigmatic protein governing cell fate and cell polarity. Development cancer cell growth and survival. Clin Cancer Res 2013;19: 2005;132:4421–36. 560–70. 47. Angers S, Thorpe CJ, Biechele TL, Goldenberg SJ, Zheng N, MacCoss 35. Chitale D, Gong Y, Taylor BS, Broderick S, Brennan C, Somwar R, et al. MJ, et al. The KLHL12-Cullin-3 ubiquitin ligase negatively regulates the An integrated genomic analysis of lung cancer reveals loss of DUSP4 in Wnt-beta-catenin pathway by targeting Dishevelled for degradation. EGFR-mutant tumors. Oncogene 2009;28:2773–83. Nat Cell Biol 2006;8:348–57. 36. Hammerman PS, Hayes DN, Wilkerson MD, Schultz N, Bose R, Chu A, 48. Shi I, Hashemi Sadraei N, Duan ZH, Shi T. Aberrant signaling pathways et al. Comprehensive genomic characterization of squamous cell lung in squamous cell lung carcinoma. Cancer Inform 2011;10:273–85. cancers. Nature 2012;489:519–25. 49. Lee EH, Chari R, Lam A, Ng RT, Yee J, English J, et al. Disruption of the 37. Redon R, Muller D, Caulee K, Wanherdrick K, Abecassis J, du Manoir S. non-canonical WNT pathway in lung squamous cell carcinoma. Clin A simple speciﬁc pattern of chromosomal aberrations at early stages Med Oncol 2008;2008:169–79. of head and neck squamous cell carcinomas: PIK3CA but not p63 50. Uematsu K, He B, You L, Xu Z, McCormick F, Jablons DM. Activation of gene as a likely target of 3q26-qter gains. Cancer Res 2001;61: the Wnt pathway in non small cell lung cancer: evidence of dishevelled 4122–9. overexpression. Oncogene 2003;22:7218–21. 38. Umayahara K, Numa F, Suehiro Y, Sakata A, Nawata S, Ogata H, et al. 51. Shen HJ, Zhu HY, Yang C, Ji F. SENP2 regulates hepatocellular Comparative genomic hybridization detects genetic alterations during carcinoma cell growth by modulating the stability of beta-catenin. early stages of cervical cancer progression. Genes Chromosomes Asian Pac J Cancer Prev 2012;13:3583–7. Cancer 2002;33:98–102. 52. Kadoya T, Yamamoto H, Suzuki T, Yukita A, Fukui A, Michiue T, et al. 39. Kim AY, Bommelje CC, Lee BE, Yonekawa Y, Choi L, Morris LG, et al. Desumoylation activity of Axam, a novel Axin-binding protein, is SCCRO (DCUN1D1) is an essential component of the E3 complex for involved in downregulation of beta-catenin. Mol Cell Biol 2002;22: neddylation. J Biol Chem 2008;283:33211–20. 3803–19.

www.aacrjournals.org Clin Cancer Res; 2013 OF11

Integrative Genomics Analysis Identifies Candidate Drivers at 3q26-29 Amplicon in Squamous Cell Carcinoma of the Lung

Jing Wang, Jun Qian, Megan D. Hoeksema, et al.

Clin Cancer Res Published OnlineFirst August 1, 2013.

Updated version Access the most recent version of this article at: doi:10.1158/1078-0432.CCR-13-0594

Supplementary Access the most recent supplemental material at: Material http://clincancerres.aacrjournals.org/content/suppl/2013/08/01/1078-0432.CCR-13-0594.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/2013/09/11/1078-0432.CCR-13-0594. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC) Rightslink site.