Integrated Genomic Analysis of the 8q24 Amplification in Endometrial Cancers Identifies ATAD2 as Essential to MYC-Dependent Cancers

Chromosome 8q24 is the most commonly amplified region across multiple cancer types, and the typical length of the amplification suggests that it may target additional genes to MYC. To explore the roles of the genes most frequently included in 8q24 amplifications, we analyzed the relation between copy number alterations and gene expression in three sets of endometrial cancers (N = 252); and in glioblastoma, ovarian, and breast cancers profiled by TCGA. Among the genes neighbouring MYC, expression of the bromodomain-containing gene ATAD2 was the most associated with amplification. Bromodomain-containing genes have been implicated as mediators of MYC transcriptional function, and indeed ATAD2 expression was more closely associated with expression of genes known to be upregulated by MYC than was MYC itself. Amplifications of 8q24, expression of genes downstream from MYC, and overexpression of ATAD2 predicted poor outcome and increased from primary to metastatic lesions. Knockdown of ATAD2 and MYC in seven endometrial and 21 breast cancer cell lines demonstrated that cell lines that were dependent on MYC also depended upon ATAD2. These same cell lines were also the most sensitive to the histone deacetylase (HDAC) inhibitor Trichostatin-A, consistent with prior studies identifying bromodomain-containing proteins as targets of inhibition by HDAC inhibitors. Our data indicate high ATAD2 expression is a marker of aggressive endometrial cancers, and suggest specific inhibitors of ATAD2 may have therapeutic utility in these and other MYC-dependent cancers.


Introduction
Endometrial carcinoma is the most common pelvic gynecologic malignancy, with a lifetime risk among women of 2-3% [1]. Approximately 75% of tumors are confined to the uterine corpus at diagnosis and are resected. However, 15%-20% of these tumors relapse. These tumors, and tumors that are metastatic at presentation, respond poorly to chemotherapy or radiation and are generally fatal [1,2].
There is a need for novel markers to identify patients with high risk of relapse, and to develop new therapies for patients with metastatic disease [3,4]. Unfortunately, research towards these goals is heavily underrepresented in endometrial cancer compared to other cancer types such as breast and ovarian cancers. One approach is to identify genes that, when altered by somatic genetic events, drive tumor progression. These alterations can then serve as markers of aggressive cancers and the genes can serve as potential therapeutic targets.
The most frequent focal amplification in endometrial cancer is on 8q24 [5]. Indeed, 8q24 is the most commonly amplified region across multiple cancer types [6], and this amplification is a negative prognostic marker in several cancers [7]. Although MYC is a likely target [6], the effects of this amplification in endometrial cancer have never been dissected. Indeed, it is possible that it targets multiple genes, as has been shown for amplifications elsewhere in the cancer genome [8]. For example, a neighboring gene, ATAD2, has been found to be a co-regulator of MYC and overexpression of ATAD2 has been associated with poor prognosis in breast, lung, and prostate cancers [9,10,11].
We explore the role of the 8q24 amplification in endometrial cancer through integrative genomic analyses of primary and metastatic endometrial cancers with comprehensive clinical data, and identify ATAD2 as an additional target of the 8q24 amplification in these cancers. We identify copy number gain of ATAD2 as a regulator of ATAD2 expression, present the first data linking ATAD2 overexpression to MYC activation, and provide functional data suggesting ATAD2 as a therapeutic target in MYC-dependent cancers.

Ethics Statement
The collection of endometrial carcinoma primaries and metastases for this study was approved by the Norwegian Data Inspectorate (961478-2), Norwegian Social Sciences Data Services (15501) and the ''Regional Research Ethics.
Committee in Medicine, Western Norway'' (reference 052.01). All the participants gave written informed consent.

Patient Series
Endometrial carcinoma primaries and metastases were collected from patients treated at Haukeland University Hospital, Norway as previously described [5]. Tumors collected for the primary investigation and qPCR validation series were frozen immediately upon resection; tumors collected for FISH were formalin fixed and paraffin embedded. Patients were followed from primary surgery until October 2010 or death. The copy-number profiles of the primary investigation series, and the expression profiles (Agilent 21 k and 22 k oligoarrays) from a subset of 57 tumors, were published previously [5].

RNA Analysis
RNA was extracted and hybridized to Agilent 44K arrays (Cat.no. G4112F) according to manufacturer's instructions and as previously described [5]. Signal intensities were evaluated using BRB-ArrayTools (National Cancer Institute, USA). The arrays were batch median normalized.

Real-time Quantitative PCR
cDNA was synthesized from 1 mg RNA using High capacity RNA to cDNA kits (Applied Biosystems). Expression of ATAD2 and MYC was determined using TaqMan gene expression assays Hs00204205 and Hs00905030 respectively (Applied Biosystems) and all samples were run on microfluidic cards per manufacturer's instructuions, using GAPDH-Hs99999905_m1 as endogenous control. Samples were run in triplicate and analyzed in RQ manager (Applied Biosystems).

FISH
Tissue microarrays (TMAs) representing the highest-grade areas in each tumor were prepared as previously reported [12] and treated at 56Cu overnight before hybridization. FISH was done using the MYC Spectrum Orange FISH probe kit and Chromosome enumeration probe 8 (CEP8) (Vysis) according to manufacturer's instructions, as previously reported [13]. Counting was performed in areas of optimal tissue digestion and no overlapping nuclei. Probe and control signals were counted in 40-60 cells within areas of optimal tissue digestion and no overlapping nuclei. Amplifications were scored when the MYC/CEP8 ratio was .1.0.

TCGA Validation Dataset
We accessed level 3 data from the TCGA data portal in November and December 2010. For breast cancer we obtained gene expression data for 279 tumors and 24 normal controls (Agilent 244K expression arrays), and copy-number from 176 tumors (Affymetrix SNP 6.0 Arrays). For ovarian cancer we obtained gene expression (Agilent 244K expression arrays) and copy-number (Agilent 1M arrays) data from 514 and 489 tumors, respectively. For glioblastoma we obtained gene expression data from 385 tumors and 10 normal controls (Affymetrix U133A arrays) and copy-number data from 261 tumors (Agilent 244K arrays).

Cell Viability
Lentiviral vectors encoding shRNAs specific for ATAD2, MYC, and the controls GFP, LACZ1, and LACZ2 (Table S1) were obtained from The RNAi Consortium. Lentivirus was produced by transfection of 293T cells with vectors encoding each shRNA (5 mg) with packaging plasmids encoding PSPAX2 and PDM2.G using Fugene HD (Roche). Lentivirus-containing supernatant was collected 48 and 72 h after transfection, pooled, and stored at 280 uC. Cells were infected in polybrene-containing media, centrifuged at 1,000 g for 30 min, and selected in puromycin (2.5 mg ml 21 ) starting 24 h after infection.
Cancer cell lines were obtained from ATCC, DSMZ, ECACC and HSRRB, and grown according to supplier's instructions (Table S2). Cell viability after RNAi was measured in 96-well plates. Eight wells seeded with cells were infected using 1:30 dilutions of virus containing each shRNA. Half of the wells underwent puromycin selection, and cell viability was measured using Cell-Titer Glo (Promega) one week later. The values from each quadruplicate were averaged; ''outlier'' wells were excluded if the replicate wells had SD/mean .0.2 and excluding the well improved the variance. The mean ATAD2-and MYC-hairpin values were normalized to the mean values from the GFP control.
To determine Trichostatin-A sensitivity, Trichostatin-A (Sigma) (0.040 to 10 mM) and vehicle (DMSO) control were each added to three wells containing each cell line on 96-well plates. Cell viability was determined after 72 hours using Cell-Titer Glo (Promega).

Statistics
Molecular data was related to clinical phenotype using Pearson's x 2 or two-sided Student's t test as appropriate. We used multivariate linear regression analysis for the prediction of ATAD2 expression levels. Univariate and multivariate survival analyses were performed by log rank and Mantel-Cox methods, respectively. ''MYC signaling strength'' and gene expression levels were presented as Z-scores. across multiple cancer types [6]. However, the importance of MYC activation in endometrial cancer is essentially unknown.
We performed an integrated analysis of copy-number and expression data to look for evidence that MYC is a target of 8q24 amplification in endometrial cancer. We evaluated expression data from a series of 82 endometrial cancers obtained in a single county in Norway, with corresponding genome-wide copy-number data from 70 tumors (the ''primary investigation series''). Sixteen of these tumors (23%) had 8q24 amplification. Most of these amplifications were low-level, ranging up to a copy-number of 4.7. We validated our results in four additional datasets. Two of these represent samples with genome-wide expression profiling: an ''internal validation series'' for which we generated data from 40 primary and 19 metastatic endometrial cancers recruited from the same region in Norway, and an ''external validation series'' representing previously published expression profiles from 111 tumors [5]. The other two validation sets represent samples analyzed with focused assays: a ''qPCR series'' of 162 samples and a ''FISH series'' of 399 samples. Patient characteristics and histopathological variables for all of our internal datasets are shown in Table S3.
We found that both MYC and genes upregulated by MYC were overexpressed in endometrial cancers with 8q24 amplification relative to endometrial cancers without it (p = 0.047 and p = 0.0078, respectively) (Figures 1a-b). We used a previously published list of 68 genes found to be upregulated by MYC across multiple contexts and assays (Table S4; www.myccancergene.org) [15] and scored their overexpression (''MYC signaling strength'') using GSEA [16]. We also tested five additional MYC activation signatures obtained from the ''Gene Set Enrichment Database'', reflecting the activation of MYC in different contexts. Four of these were expressed at higher levels in endometrial cancers with 8q24 amplification ( Figure S1a).
Moreover, 8q24 amplifications are longer than the typical distribution of amplification sizes in endometrial cancer (p = 0.0021) (Figure 1c), and usually involve multiple genes. We therefore hypothesized that 8q24 amplifications may target additional genes, some of which may function through increasing MYC signaling. To identify these, we evaluated all 26 genes in the peak region of amplification on 8q24 for which we had expression data, to identify genes whose expression correlated most strongly with amplification.

Expression of ATAD2 Correlates Strongly with 8q24 Amplification and MYC Signaling
Expression of ATAD2 was more strongly associated with amplification of 8q24 than was expression of any other gene in the peak region of the amplification (p-value = 2.77E-06) ( Figure 1d). Four other genes, NDUFB9, DERL1, FAM91A1, and WDR67, were significantly upregulated by 8q24 amplification, though less strongly than ATAD2.
Expression of ATAD2 also correlated with MYC signaling strength more strongly than did expression of any other gene in the 8q24 peak region (R2 = 0.48, p,0.001; Figure 1e, Figure S1b). Indeed, the association between ATAD2 expression and MYC signaling strength was observed even among samples without 8q24 amplification (R2 = 0.48, p,0.001). The correlation between ATAD2 expression and MYC signaling strength was more than twice as strong as the next most significantly associated gene (NDUFB9) and stronger than for MYC itself (R2 = 0.05; Figure 1c). ATAD2 is not one of the 68 genes in the MYC activation signature, and to our knowledge MYC has not been found to modulate expression of ATAD2 [15]. However, ATAD2 was previously found to bind to MYC and to the E-box region of several MYC target genes, and ATAD2 levels were found to be limiting for MYC-dependent transcription [9].
Both genome-wide validation series also exhibited the correlation between MYC signaling and ATAD2 expression (R2 = 0.54, p,0.001 and R2 = 0.45, p,0.001 in the internal and external validation series, respectively) and the relative lack of correlation with MYC expression (R2 = 0.00, p = 0.33 and R2 = 0.06, p = 0.012; Figures 1f and S1b). Expression of ATAD2 also correlated with four of the five additional signatures of MYC activation, and correlated more strongly with these signatures than did expression of MYC itself ( Table 1). The last signature showed no association with ATAD2 or MYC expression.
Amplification of 8q24 and Expression of ATAD2, but not MYC, are Associated with Disease Progression Among the 70 endometrial cancers for which we had genomewide SNP array data, 8q24 amplification was associated with reduced progression-free survival (p = 0.024) and increased risk for disease-specific death (p = 0.043). Amplification of 8q24 was most frequent in non-endometrioid (p = 2.98E-05) and high-grade tumors (p = 2.90E-08) (Table S5), features also associated with aggressive cancers [17].
High expression of ATAD2 and MYC signaling were also associated with increased risk of cancer progression (p = 0.003 and p = 0.015, respectively), cancer-specific death (p = 0.004 and p = 0.001) (Figures 2c-d), and other poor-prognosis features. ATAD2 expression was higher in non-endometrioid, high-grade and ER negative tumors (p,0.001, p,0.001, and p = 0.02 respectively; Table S6); high MYC signaling was associated with poorly differentiated (p = 0.0016) and non-endometrioid (p,0.001) cancers. Expression of ATAD2 was also negatively associated with expression of ESR1 (R2 = 0.10, p = 0.005). Similarly, prior studies have found that ATAD2 expression is higher in triple negative breast cancer tumors [18] and is downregulated by estrogen in cell culture [19].
We confirmed these results by assessing ATAD2 expression by quantitative PCR and ER status by immunohistochemistry in our qPCR validation series of 162 additional tumors. Among these, ER-negative tumors with upper-quartile ATAD2 expression were associated with even worse outcomes than in the primary series (HR = 6.8, p,0.001) (Figure S1c). High ATAD2 expression also remained associated with increased risk of disease-specific death (p = 0.0043) and shorter progression-free survival (p = 0.016). After adjusting for ER status, ATAD2 expression continued to predict disease-specific death (HR = 1.86, p = 0.018) but only trended towards an association with disease progression (HR = 1.31, p = 0.11). Expression of ATAD2 was also higher in high-grade (p,0.001), non-endometrioid (p = 0.005), and ER-negative tumors (p = 0.02) (Table S6).
In contrast, MYC expression was not associated with progression or risk of disease-specific death in either our primary investigation series (p = 0.07 and p = 0.68 respectively) or the qPCR validation series (p = 0.53 and p = 0.28 respectively). High expression of MYC was associated with high grade in both series (p,0.001 and p = 0.02, respectively), and with non-endometrioid histology in the primary investigation series (p = 0.03) (Table S6).
Metastases also exhibit more 8q24 amplification, ATAD2 expression, and MYC signaling strength than primary tumors. Relative to primary tumors, metastases exhibited 2.46higher rates of focal 8q24 amplification by FISH (14.3%; p,0.007). Among the 399 patients in the FISH series, 49 had paired primary and metastatic tumors. Five of these samples (10%) did not exhibit 8q24 amplification in the primary but acquired it in the metastasis. Only one sample (2%) exhibited the opposite pattern. To examine ATAD2 expression and MYC signaling, we also compared the 42 primary tumors with the 19 metastases in our internal validation series. Both ATAD2 expression and MYC signaling strength were higher in the metastases (p = 0.002 and 0.004 respectively;

Extension to Other Cancer Types and Normal Tissue
We also investigated whether 8q24 amplification is associated with increased ATAD2 expression in other cancer types. Specifically, we used data from 514 ovarian cancers, 279 breast cancers, and 385 glioblastomas from The Cancer Genome Atlas [20,21]. These included expression data from adjacent normal tissue for 24 breast cancers and 10 glioblastomas. 8q24 amplifications were observed in 72% (N = 126) of the breast cancers, 74% (N = 364) of the ovarian cancers, and 10% (N = 27) of the glioblastomas. ATAD2 was co-amplified to the same level as MYC in nearly all tumors (as it was among our endometrial cancers; Figure S1d-g).
Expression of ATAD2 correlated with 8q24 amplification among all three cancer types (R2 = 0.47, 0.11, and 0.36 for breast cancers, glioblastomas, and ovarian cancers respectively; p,0.001 in all cases; Table S7), and was 2.66 and 2.56 higher among the cancers relative to normal tissue in breast cancer and glioblastoma, respectively (p,0.001 in both cases). MYC expression correlated less strongly with 8q24 amplification in all three cancer types (R2 = 0.12, 0.07, and 0.10 for breast cancers, glioblastomas, and ovarian cancers respectively; p,0.001 in all cases). MYC expression in breast cancers was surprisingly half that of normal tissue (p,0.001); in glioblastoma it was higher by a factor of 3.5 (p,0.001).

ATAD2 Expression is Correlated to E2F Gene Expression and ATAD2 Copy Number in an Additive Manner
We also explored the relative contributions of E2F, estrogen, and copy-number on ATAD2 expression. The ATAD2 promoter region contains binding sites for several E2F proteins and previous functional data have shown that E2F increases ATAD2 expression in cell culture [9,22]; ATAD2 has also been induced by estrogen [23]. In our data, expression of every E2F transcription factor was associated with ATAD2 expression, but only the inclusion of E2F1, E2F2 and E2F8 improved the overall fit of a model predicting ATAD2 expression from ATAD2 copy-number and ESR1 expression. The expression levels of these three genes were highly correlated, and we focused on E2F1.
We found that ATAD2 copy-number, ESR1 expression and E2F1 expression explained 77% of the variation in ATAD2 expression in endometrial cancer, and each of the predictor variables remained significantly associated with ATAD2 expression in the adjusted model (Figure 3a and Table S7). We also found that ATAD2 copy-number and E2F expression independently predicted ATAD2 expression in breast cancer, ovarian cancer and glioblastoma (Figure 3b-d and Table S7). ESR1, which was less strongly associated with ATAD2 expression, was significant in the adjusted model only in endometrial cancer (p = 0.016) and glioblastoma (p,0.001), not in ovarian or breast cancer. These data suggest that the copy-number of ATAD2 is an important determinant of ATAD2 expression even in the context of other cellular regulatory mechanisms.

Dependency on MYC Predicts Dependency on ATAD2 and Response to HDAC Inhibitors in Endometrial-and Breast Cancer Cells
The results above led us to hypothesize that ATAD2 expression promotes MYC signaling and that endometrial cancer cells that are dependent upon MYC would also be dependent upon ATAD2. We measured the effect on viability of shRNA knockdowns of ATAD2 and MYC in seven endometrial cancer cell lines. We used two shRNAs against each gene, selecting those that exhibited the greatest reduction of protein expression among six and three shRNAs screened against ATAD2 and MYC respectively (Figure 4a,  b).
Knockdown of either ATAD2 or MYC resulted in highly correlated decreases in viability across the seven cell endometrial cancer lines (R2 = 0.70, p = 0.020; Figure 4c). In two cases, we observed over 75% reductions in viability. We found no  association between expression of ATAD2 or MYC or 8q24 copy number and sensitivity to ATAD2 or MYC knockdown. These results suggested that MYC-dependent cancers of other types might also be dependent on ATAD2. No decrease in proliferation had previously been seen with ATAD2 knockdown in TIG3-T or U2OS cells [9]. When we tested a larger panel of 21 breast cancer lines, however, we confirmed the strong correlation between decrease in viability after knockdown of ATAD2 or MYC (R2 = 0.61, p,0.001; Figure 4d).
The association between dependency on MYC and ATAD2 suggests ATAD2 as a therapeutic target in MYC-dependent cancers. Whereas MYC has long been a known oncogene, clinical approaches to block MYC signaling have not yet been successful. Histone deacetylase (HDAC) inhibitors, however, have been shown to indirectly inhibit bromodomain-containing proteins such as ATAD2 [24].
We used the Connectivity Map [25] to identify compounds whose signatures anticorrelated with the MYC signaling signature. Among the 1309 small molecules represented by the Connectivity Map, the signature of the HDAC inhibitor Trichostatin-A was most anticorrelated with the MYC signaling signature. (pvalue,0.00001; Table S8). We also generated a signature of aggressive disease from the primary investigation series, using the 50 most over-and under-expressed genes in patients with metastatic disease compared to patients without metastatic disease. We found the Trichostatin-A signature was also the most anticorrelated with this signature of aggressive disease, tied with signatures of four other molecules (p,0.00001; Table S8).
To functionally confirm the relation between Trichostatin-A and MYC dependency, we tested all endometrial cancer and   (Table S9) than were the effects of MYC or ATAD2 knockdown.

Discussion
Our data suggest that ATAD2 overexpression in human endometrial cancers is a consequence of 8q24 amplification and associated with MYC pathway activation. We also find that ATAD2 overexpression is associated with E2F activation and poor prognosis. Analyses of TCGA data suggest similar relationships between ATAD2, 8q24 amplification, and MYC pathway activation in glioblastoma, breast, and ovarian cancers. We also find that endometrial and breast cancer cell lines that are dependent upon MYC expression also depend upon expression of ATAD2.
High expression of ATAD2 has previously been found to be associated with an unfavorable prognosis in breast, lung, and prostate cancers and it has been suggested that ATAD2 contributes to the development of aggressive cancer through linking of the E2F and MYC pathways [9,10,11]. We demonstrate an association between high ATAD2 expression and negative outcome in endometrial cancer, using clinically well-characterized test and validation datasets. We also find that progression from primary to metastatic endometrial cancer is associated with a further increase of MYC signaling and ATAD2 expression.
Ciro et al [9] previously showed that ATAD2 interacts with MYC in breast cancer cell lines and is overexpressed in 8q24 amplified breast cancers. Our results indicate that, in endometrial cancers, expression of ATAD2 is more highly correlated with 8q24 amplification than is expression of its neighbors (including MYC), and that ATAD2 amplification and overexpression are strongly associated with multiple measures of MYC pathway activation in human tumors.
The finding of cooperative effects between MYC and coamplified genes on 8q24 is not entirely surprising. Indeed, the concept of oncogene cooperation was established through the study of positive interactions between MYC and other oncogenes such as BCL2 [26]. Moreover, clustered genes are often functionally related [27]. The relevance of this phenomenon in cancer has been shown for the genes MMP13, Birch2, and Birch3, which are functionally related oncogenes contained on the same amplification in osteosarcoma [28], and for BIRC2 and YAP1, cooperating oncogenes in an amplification in hepatocellular carcinomas [8].
Such a mechanistic association between ATAD2 and MYC, and the finding that MYC-dependent cells are sensitive to ATAD2 knockdown, suggest ATAD2 as a therapeutic target in MYCdependent cancers. Although MYC has long been known as an oncogene [14] and is a promising drug target, it has not been successfully targeted therapeutically. Small molecule inhibitors have, however, been generated against other bromodomaincontaining proteins [29]. Indeed, inhibition of the bromodomaincontaining protein BRD4 has recently been suggested as an alternative approach to targeting MYC [30]. HDAC inhibitors also indirectly inhibit bromodomain-containing proteins by inducing histone hyperacetylation, thus probably diverting the specific bromodomain proteins from their targets [24]. This may account for some of the effectiveness of HDAC inhibitors as cancer therapeutics [30], and we found cell lines that were sensitive to knockdown of MYC or ATAD2 were also sensitive to the HDAC inhibitor Trichostatin-A. However, the reduction in viability after application of Trichostatin-A was smaller than the reduction in viability after MYC or ATAD2 knockdown. It is possible that a more direct inhibitor of ATAD2 would be more effective in these cells.
Major obstacles to treatment of patients with endometrial cancer include a lack of targeted therapeutics and of prognostic indicators. Indeed, endometrial cancer remains understudied relative to other cancer types. We find that ATAD2 amplification and expression is a prognostic marker in endometrial cancer and our findings suggest that development of specific ATAD2 inhibitors is a promising approach to treatment of endometrial and other MYC driven cancers. Figure S1 a) Five additional MYC activation genes sets obtained from the literature and their relative expression in 8q24 amplified versus unamplified samples. b) The correlation between MYC signaling strength and MYC-(left) and ATAD2-(right) gene expression in the primary investigation series and in the internal and external validation series. c) Estrogen receptor negative (ER2) tumors with ATAD2 expression in the top quartile were associated with a high risk of disease-specific death; the risk was much lower among estrogen receptor positive (ER+) tumors with ATAD2 expression in the bottom quartile. d-g) Results from the qPCR validation series. The copy-number of MYC and ATAD2 genes are highly correlated in endometrial cancer (d), glioblastoma (e), ovarian cancer (f) and breast cancer (g).

(EPS)
Table S1 Details about the shRNA used in the study. (DOCX)